targets/esp32: add .rxring and .wifibss sections to DRAM1

Place espradio's RX ring buffer and WiFi-only BSS (ISR tables, timer slots, ISR ring) in SRAM1 pool 7/6 ahead of the arena. Frees ~14 KB of SRAM2 for the Go GC heap. The arena assertion still guarantees ≥32 KB remains for WiFi DMA buffers. Signed-off-by: deadprogram <ron@hybridgroup.com>
runtime: prevent timer starvation in cores scheduler
2026-07-26 06:38:42 +00:00 · 2026-07-25 17:42:01 +02:00 · 2026-07-25 10:02:09 +02:00 · 2026-07-25 07:14:00 +02:00 · 2026-07-25 07:14:00 +02:00 · 2026-07-24 20:50:26 +02:00
478 changed files with 22305 additions and 4183 deletions
@@ -1,119 +0,0 @@
-version: 2.1
-
-commands:
-  submodules:
-    steps:
-      - run:
-          name: "Pull submodules"
-          command: git submodule update --init
-  llvm-source-linux:
-    steps:
-      - restore_cache:
-          keys:
-            - llvm-source-19-v1
-      - run:
-          name: "Fetch LLVM source"
-          command: make llvm-source
-      - save_cache:
-          key: llvm-source-19-v1
-          paths:
-            - llvm-project/clang/lib/Headers
-            - llvm-project/clang/include
-            - llvm-project/compiler-rt
-            - llvm-project/lld/include
-            - llvm-project/llvm/include
-  hack-ninja-jobs:
-    steps:
-      - run:
-          name: "Hack Ninja to use less jobs"
-          command: |
-            echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
-            chmod +x /go/bin/ninja
-  build-binaryen-linux:
-    steps:
-      - restore_cache:
-          keys:
-            - binaryen-linux-v3
-      - run:
-          name: "Build Binaryen"
-          command: |
-            make binaryen
-      - save_cache:
-          key: binaryen-linux-v3
-          paths:
-            - build/wasm-opt
-  test-linux:
-    parameters:
-      llvm:
-        type: string
-      fmt-check:
-        type: boolean
-        default: true
-    steps:
-      - checkout
-      - submodules
-      - run:
-          name: "Install apt dependencies"
-          command: |
-            echo 'deb https://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-<<parameters.llvm>> main' > /etc/apt/sources.list.d/llvm.list
-            wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | apt-key add -
-            apt-get update
-            apt-get install --no-install-recommends -y \
-                llvm-<<parameters.llvm>>-dev \
-                clang-<<parameters.llvm>> \
-                libclang-<<parameters.llvm>>-dev \
-                lld-<<parameters.llvm>> \
-                cmake \
-                ninja-build
-      - hack-ninja-jobs
-      - build-binaryen-linux
-      - restore_cache:
-          keys:
-            - go-cache-v4-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
-            - go-cache-v4-{{ checksum "go.mod" }}
-      - llvm-source-linux
-      - run: go install -tags=llvm<<parameters.llvm>> .
-      - when:
-          condition: <<parameters.fmt-check>>
-          steps:
-          - run:
-              # Do this before gen-device so that it doesn't check the
-              # formatting of generated files.
-              name: Check Go code formatting
-              command: make fmt-check lint
-      - run: make gen-device -j4
-      # TODO: change this to -skip='TestErrors|TestWasm' with Go 1.20
-      - run: go test -tags=llvm<<parameters.llvm>> -short -run='TestBuild|TestTest|TestGetList|TestTraceback'
-      - run: make smoketest XTENSA=0
-      - save_cache:
-          key: go-cache-v4-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
-          paths:
-            - ~/.cache/go-build
-            - /go/pkg/mod
-
-jobs:
-  test-oldest:
-    # This tests our lowest supported versions of Go and LLVM, to make sure at
-    # least the smoke tests still pass.
-    docker:
-      - image: golang:1.23-bullseye
-    steps:
-      - test-linux:
-          llvm: "15"
-    resource_class: large
-  test-newest:
-    # This tests the latest supported LLVM version when linking against system
-    # libraries.
-    docker:
-      - image: golang:1.26-bookworm
-    steps:
-      - test-linux:
-          llvm: "20"
-    resource_class: large
-
-workflows:
-  test-all:
-    jobs:
-      - test-oldest
-      # disable this test, since CircleCI seems unable to download due to rate-limits on Dockerhub.
-      # - test-newest
@@ -1,5 +1,4 @@
 build/
 llvm-*/
 .github
-.circleci

@@ -40,13 +40,13 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Restore LLVM source cache
        uses: actions/cache/restore@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-${{ matrix.os }}-v1
+          key: llvm-source-22-${{ matrix.os }}-v1
          path: |
            llvm-project/clang/lib/Headers
            llvm-project/clang/include
@@ -71,7 +71,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-build
        with:
-          key: llvm-build-20-${{ matrix.os }}-v2
+          key: llvm-build-22-${{ matrix.os }}-v1
          path: llvm-build
      - name: Build LLVM
        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -131,7 +131,7 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Build TinyGo (LLVM ${{ matrix.version }})
        run: go install -tags=llvm${{ matrix.version }}
@@ -0,0 +1,72 @@
+# This CI job checks whether at least the smoke tests pass for the oldest
+# Go/LLVM version we claim to support.
+
+name: Version compatibility test
+
+on:
+  pull_request:
+  push:
+    branches:
+      - dev
+      - release
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-compat:
+    runs-on: ubuntu-22.04 # this must be a specific version for the apt install below
+    env:
+      # Oldest versions currently supported by TinyGo
+      LLVM: "15"
+      Go: "1.25" # when updating this, also update minorMin in builder/config.go
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Install Go
+        uses: actions/setup-go@v6
+        with:
+          go-version: ${{ env.Go }}
+          cache: true
+      - name: Install LLVM
+        run: |
+          echo 'deb https://apt.llvm.org/jammy/ llvm-toolchain-jammy-${{ env.LLVM }} main' | sudo tee /etc/apt/sources.list.d/llvm.list
+          wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
+          sudo apt-get update
+          sudo apt-get install --no-install-recommends -y \
+              llvm-${{ env.LLVM }}-dev \
+              clang-${{ env.LLVM }} \
+              libclang-${{ env.LLVM }}-dev \
+              lld-${{ env.LLVM }} \
+              binaryen
+      - name: Restore LLVM source cache
+        uses: actions/cache/restore@v5
+        id: cache-llvm-source
+        with:
+          key: llvm-source-22-linux-compat
+          path: llvm-project/compiler-rt
+      - name: Download LLVM source
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        run: make llvm-source
+      - name: Save LLVM source cache
+        uses: actions/cache/save@v5
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        with:
+          key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
+          path: llvm-project/compiler-rt
+      - name: Go cache
+        uses: actions/cache@v5
+        with:
+          path: |
+            ~/.cache/go-build
+            ~/go/pkg/mod
+          key: go-build-${{ env.Go }}-llvm${{ env.LLVM }}-${{ hashFiles('go.sum') }}
+      - name: Build TinyGo
+        run: go install -tags=llvm${{ env.LLVM }}
+      - run: tinygo version
+      - run: make gen-device -j4
+      - run: go test -tags=llvm${{ env.LLVM }} -short -skip=TestErrors
+      - run: make smoketest XTENSA=0
@@ -12,13 +12,31 @@ concurrency:
  cancel-in-progress: true

 jobs:
+  go-mod-tidy:
+    # Check that go.sum is up to date.
+    runs-on: ubuntu-slim
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Install Go
+        uses: actions/setup-go@v6
+        with:
+          go-version: '1.27.0-rc.2'
+          cache: true
+      - name: Run go mod tidy
+        run: go mod tidy
+      - name: Check go.mod and go.sum are up to date
+        run: git diff --exit-code
+
  build-linux:
    # Build Linux binaries, ready for release.
    # This runs inside an Alpine Linux container so we can more easily create a
    # statically linked binary.
    runs-on: ubuntu-latest
    container:
-      image: golang:1.26-alpine
+      image: golang:1.27rc2-alpine
    outputs:
      version: ${{ steps.version.outputs.version }}
    steps:
@@ -48,7 +66,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-linux-alpine-v2
+          key: llvm-source-22-linux-alpine-v1
          path: |
            llvm-project/clang/lib/Headers
            llvm-project/clang/include
@@ -73,7 +91,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-build
        with:
-          key: llvm-build-20-linux-alpine-v2
+          key: llvm-build-22-linux-alpine-v1
          path: llvm-build
      - name: Build LLVM
        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -97,7 +115,7 @@ jobs:
        uses: actions/cache@v5
        id: cache-binaryen
        with:
-          key: binaryen-linux-alpine-v2
+          key: binaryen-linux-alpine-v3
          path: build/wasm-opt
      - name: Build Binaryen
        if: steps.cache-binaryen.outputs.cache-hit != 'true'
@@ -142,7 +160,7 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Install wasmtime
        uses: bytecodealliance/actions/wasmtime/setup@v1
@@ -186,12 +204,12 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Install Node.js
        uses: actions/setup-node@v6
        with:
-          node-version: '18'
+          node-version: '22'
      - name: Install wasmtime
        uses: bytecodealliance/actions/wasmtime/setup@v1
        with:
@@ -202,7 +220,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-linux-asserts-v1
+          key: llvm-source-22-linux-asserts-v1
          path: |
            llvm-project/clang/lib/Headers
            llvm-project/clang/include
@@ -227,7 +245,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-build
        with:
-          key: llvm-build-20-linux-asserts-v1
+          key: llvm-build-22-linux-asserts-v1
          path: llvm-build
      - name: Build LLVM
        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -266,6 +284,8 @@ jobs:
      - run: make smoketest
      - run: make wasmtest
      - run: make tinygo-test-baremetal
+      - name: Check Go code formatting
+        run: make fmt-check lint
  build-linux-cross:
    # Build ARM Linux binaries, ready for release.
    # This intentionally uses an older Linux image, so that we compile against
@@ -303,13 +323,13 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Restore LLVM source cache
        uses: actions/cache/restore@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-linux-v1
+          key: llvm-source-22-linux-v1
          path: |
            llvm-project/clang/lib/Headers
            llvm-project/clang/include
@@ -334,7 +354,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-build
        with:
-          key: llvm-build-20-linux-${{ matrix.goarch }}-v1
+          key: llvm-build-22-linux-${{ matrix.goarch }}-v1
          path: llvm-build
      - name: Build LLVM
        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -358,7 +378,7 @@ jobs:
        uses: actions/cache@v5
        id: cache-binaryen
        with:
-          key: binaryen-linux-${{ matrix.goarch }}-v4
+          key: binaryen-linux-${{ matrix.goarch }}-v5
          path: build/wasm-opt
      - name: Build Binaryen
        if: steps.cache-binaryen.outputs.cache-hit != 'true'
@@ -29,7 +29,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-linux-nix-v1
+          key: llvm-source-22-linux-nix-v1
          path: |
            llvm-project/compiler-rt
      - name: Download LLVM source
@@ -19,6 +19,10 @@ jobs:
      - name: Add GOBIN to $PATH
        run: |
          echo "$HOME/go/bin" >> $GITHUB_PATH
+      - name: Install Go
+        uses: actions/setup-go@v6
+        with:
+          go-version: '~1.25'
      - name: Checkout
        uses: actions/checkout@v6
        with:
@@ -30,7 +34,7 @@ jobs:
        uses: actions/cache@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-sizediff-v1
+          key: llvm-source-22-sizediff-v1
          path: |
            llvm-project/compiler-rt
      - name: Download LLVM source
@@ -17,12 +17,6 @@ jobs:
    outputs:
      version: ${{ steps.version.outputs.version }}
    steps:
-      - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.4
-        with:
-          minimum-size: 8GB
-          maximum-size: 24GB
-          disk-root: "C:"
      - uses: MinoruSekine/setup-scoop@v4
      - name: Install Dependencies
        shell: bash
@@ -40,13 +34,13 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Restore cached LLVM source
        uses: actions/cache/restore@v5
        id: cache-llvm-source
        with:
-          key: llvm-source-20-windows-v1
+          key: llvm-source-22-windows-v3
          path: |
            llvm-project/clang/lib/Headers
            llvm-project/clang/include
@@ -71,7 +65,7 @@ jobs:
        uses: actions/cache/restore@v5
        id: cache-llvm-build
        with:
-          key: llvm-build-20-windows-v2
+          key: llvm-build-22-windows-v1
          path: llvm-build
      - name: Build LLVM
        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -93,7 +87,7 @@ jobs:
      - name: Cache Go cache
        uses: actions/cache@v5
        with:
-          key: go-cache-windows-v2-${{ hashFiles('go.mod') }}
+          key: go-cache-windows-v3-${{ hashFiles('go.mod') }}
          path: |
            C:/Users/runneradmin/AppData/Local/go-build
            C:/Users/runneradmin/go/pkg/mod
@@ -124,12 +118,6 @@ jobs:
    runs-on: windows-2022
    needs: build-windows
    steps:
-      - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.4
-        with:
-          minimum-size: 8GB
-          maximum-size: 24GB
-          disk-root: "C:"
      - uses: MinoruSekine/setup-scoop@v4
      - name: Install Dependencies
        shell: bash
@@ -141,7 +129,7 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Download TinyGo build
        uses: actions/download-artifact@v8
@@ -157,18 +145,12 @@ jobs:
    runs-on: windows-2022
    needs: build-windows
    steps:
-      - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.4
-        with:
-          minimum-size: 8GB
-          maximum-size: 24GB
-          disk-root: "C:"
      - name: Checkout
        uses: actions/checkout@v6
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Download TinyGo build
        uses: actions/download-artifact@v8
@@ -183,12 +165,6 @@ jobs:
    runs-on: windows-2022
    needs: build-windows
    steps:
-      - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.4
-        with:
-          minimum-size: 8GB
-          maximum-size: 24GB
-          disk-root: "C:"
      - uses: MinoruSekine/setup-scoop@v4
      - name: Install Dependencies
        shell: bash
@@ -200,7 +176,7 @@ jobs:
      - name: Install Go
        uses: actions/setup-go@v6
        with:
-          go-version: '1.26.2'
+          go-version: '1.27.0-rc.2'
          cache: true
      - name: Download TinyGo build
        uses: actions/download-artifact@v8
@@ -1,3 +1,12 @@
+0.41.1
+---
+* **machine**
+  - esp32c3: correct pin interrupt setup call that was overlooked from #5320
+* **runtime**
+  - esp32s3: wait for TIMG0 update register to clear before reading timer registers
+* **net**
+  - update net module to a version that is backwards compatible with Go 1.25.x to fix #5332
+
 0.41.0
 ---
 * **general**
@@ -1,5 +1,5 @@
 # tinygo-llvm stage obtains the llvm source for TinyGo
-FROM golang:1.26 AS tinygo-llvm
+FROM golang:1.27rc2 AS tinygo-llvm

 RUN apt-get update && \
    apt-get install -y apt-utils make cmake clang-17 ninja-build && \
@@ -33,7 +33,7 @@ RUN cd /tinygo/ && \

 # tinygo-compiler copies the compiler build over to a base Go container (without
 # all the build tools etc).
-FROM golang:1.26 AS tinygo-compiler
+FROM golang:1.27rc2 AS tinygo-compiler

 # Copy tinygo build.
 COPY --from=tinygo-compiler-build /tinygo/build/release/tinygo /tinygo
@@ -133,7 +133,7 @@ endif

 .PHONY: all tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-nxp gen-device-avr gen-device-rp

-LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf debuginfopdb executionengine frontenddriver frontendhlsl frontendopenmp instrumentation interpreter ipo irreader libdriver linker lto mc mcjit objcarcopts option profiledata scalaropts support target windowsdriver windowsmanifest
+LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf debuginfopdb dtlto executionengine frontenddriver frontendhlsl frontendopenmp instrumentation interpreter ipo irreader libdriver linker lto mc mcjit objcarcopts option profiledata scalaropts support target windowsdriver windowsmanifest

 ifeq ($(OS),Windows_NT)
    EXE = .exe
@@ -142,6 +142,9 @@ ifeq ($(OS),Windows_NT)

    # PIC needs to be disabled for libclang to work.
    LLVM_OPTION += -DLLVM_ENABLE_PIC=OFF
+    # Statically link the C++ and GCC runtime into LLVM tools so they don't
+    # depend on MinGW DLLs that may not be on PATH when executed during the build.
+    LLVM_OPTION += '-DCMAKE_EXE_LINKER_FLAGS=-static-libgcc -static-libstdc++'

    CGO_CPPFLAGS += -DCINDEX_NO_EXPORTS
    CGO_LDFLAGS += -static -static-libgcc -static-libstdc++
@@ -169,7 +172,7 @@ endif
 MD5SUM ?= md5sum

 # Libraries that should be linked in for the statically linked Clang.
-CLANG_LIB_NAMES = clangAnalysis clangAPINotes clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangExtractAPI clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangInstallAPI clangLex clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangSupport clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
+CLANG_LIB_NAMES = clangAnalysis clangAnalysisLifetimeSafety clangAPINotes clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangExtractAPI clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangInstallAPI clangLex clangOptions clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangSupport clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
 CLANG_LIBS = $(START_GROUP) $(addprefix -l,$(CLANG_LIB_NAMES)) $(END_GROUP) -lstdc++

 # Libraries that should be linked in for the statically linked LLD.
@@ -177,7 +180,7 @@ LLD_LIB_NAMES = lldCOFF lldCommon lldELF lldMachO lldMinGW lldWasm
 LLD_LIBS = $(START_GROUP) $(addprefix -l,$(LLD_LIB_NAMES)) $(END_GROUP)

 # Other libraries that are needed to link TinyGo.
-EXTRA_LIB_NAMES = LLVMInterpreter LLVMMCA LLVMRISCVTargetMCA LLVMX86TargetMCA
+EXTRA_LIB_NAMES = LLVMDTLTO LLVMInterpreter LLVMMCA LLVMRISCVTargetMCA LLVMX86TargetMCA

 # All libraries to be built and linked with the tinygo binary (lib/lib*.a).
 LIB_NAMES = clang $(CLANG_LIB_NAMES) $(LLD_LIB_NAMES) $(EXTRA_LIB_NAMES)
@@ -194,6 +197,12 @@ NINJA_BUILD_TARGETS = clang llvm-config llvm-ar llvm-nm lld $(addprefix lib/lib,
 ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config*)","")
    CGO_CPPFLAGS+=$(shell $(LLVM_CONFIG_PREFIX) $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(LLVM_BUILDDIR))/tools/clang/include -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
    CGO_CXXFLAGS=-std=c++17
+ifneq ($(uname),Windows_NT)
+    # Disable GCC DWARF compression: lld built without zlib cannot link
+    # object files with ELFCOMPRESS_ZLIB debug sections.
+    CGO_CFLAGS+=-gz=none
+    CGO_CXXFLAGS+=-gz=none
+endif
    CGO_LDFLAGS+=-L$(abspath $(LLVM_BUILDDIR)/lib) -lclang $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_CONFIG_PREFIX) $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS)) -lstdc++ $(CGO_LDFLAGS_EXTRA)
 endif

@@ -263,7 +272,7 @@ gen-device-renesas: build/gen-device-svd
 	GO111MODULE=off $(GO) fmt ./src/device/renesas

 $(LLVM_PROJECTDIR)/llvm:
-	git clone -b tinygo_20.x --depth=1 https://github.com/tinygo-org/llvm-project $(LLVM_PROJECTDIR)
+	git clone -b tinygo_22.x --depth=1 https://github.com/tinygo-org/llvm-project $(LLVM_PROJECTDIR)
 llvm-source: $(LLVM_PROJECTDIR)/llvm ## Get LLVM sources

 # Configure LLVM.
@@ -298,7 +307,7 @@ wasi-cm:
 	rsync -rv --delete --exclude go.mod --exclude '*_test.go' --exclude '*_json.go' --exclude '*.md' --exclude LICENSE $(shell go list -m -f {{.Dir}} $(WASM_TOOLS_MODULE)/cm)/ ./src/internal/cm

 # Check for Node.js used during WASM tests.
-MIN_NODEJS_VERSION=18
+MIN_NODEJS_VERSION=22

 .PHONY: check-nodejs-version
 check-nodejs-version:
@@ -310,9 +319,9 @@ check-nodejs-version:

 tinygo: ## Build the TinyGo compiler
 	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  $(MAKE) llvm-source"; echo "  $(MAKE) $(LLVM_BUILDDIR)"; exit 1; fi
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm osusergo" .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CFLAGS="$(CGO_CFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm llvm22 osusergo" .
 test: check-nodejs-version
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm osusergo" $(GOTESTPKGS)
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm llvm22 osusergo" $(GOTESTPKGS)

 # Standard library packages that pass tests on darwin, linux, wasi, and windows, but take over a minute in wasi
 TEST_PACKAGES_SLOW = \
@@ -379,22 +388,22 @@ TEST_PACKAGES_FAST = \
 # archive/zip requires os.ReadAt, which is not yet supported on windows
 # bytes requires mmap
 # compress/flate appears to hang on wasi
-# crypto/aes fails on wasi, needs panic()/recover()
+# crypto/aes needs reflect.Type.Method(), not yet implemented
 # crypto/des fails on wasi, needs panic()/recover()
 # crypto/hmac fails on wasi, it exits with a "slice out of range" panic
 # debug/plan9obj requires os.ReadAt, which is not yet supported on windows
 # encoding/xml takes a minute on linux and gives a stack overflow on wasi
-# image requires recover(), which is not yet supported on wasi
+# image fails on wasi, needs panic()/recover()
 # io/ioutil requires os.ReadDir, which is not yet supported on windows or wasi
-# mime: fail on wasi; neds panic()/recover()
+# mime: fails on wasi, needs panic()/recover()
 # mime/multipart: needs wasip1 syscall.FDFLAG_NONBLOCK
 # mime/quotedprintable requires syscall.Faccessat
 # net/mail: needs wasip1  syscall.FDFLAG_NONBLOCK
 # net/ntextproto: needs wasip1 syscall.FDFLAG_NONBLOCK
-# regexp/syntax: fails on wasip1; needs panic()/recover()
-# strconv requires recover() which is not yet supported on wasi
-# text/tabwriter requires recover(), which is not yet supported on wasi
-# text/template/parse requires recover(), which is not yet supported on wasi
+# regexp/syntax: fails on wasip1, needs panic()/recover()
+# strconv: fails on wasi, needs panic()/recover()
+# text/tabwriter: fails on wasi, needs panic()/recover()
+# text/template/parse: fails on wasi, needs panic()/recover()
 # testing/fstest requires os.ReadDir, which is not yet supported on windows or wasi

 # Additional standard library packages that pass tests on individual platforms
@@ -404,6 +413,7 @@ TEST_PACKAGES_LINUX := \
 	context \
 	crypto/aes \
 	crypto/des \
+	crypto/ecdh \
 	crypto/hmac \
 	debug/dwarf \
 	debug/plan9obj \
@@ -419,6 +429,7 @@ TEST_PACKAGES_LINUX := \
 	os/user \
 	regexp/syntax \
 	strconv \
+	testing/fstest \
 	text/tabwriter \
 	text/template/parse

@@ -429,7 +440,11 @@ TEST_PACKAGES_WINDOWS := \
 	compress/flate \
 	crypto/des \
 	crypto/hmac \
+	image \
+	mime \
+	regexp/syntax \
 	strconv \
+	text/tabwriter \
 	text/template/parse \
 	$(nil)

@@ -464,6 +479,11 @@ TEST_PACKAGES_NONBAREMETAL = \
 	math \
 	$(nil)

+TEST_PACKAGES_FAST_WASI = $(filter-out $(TEST_PACKAGES_NOWASI), $(TEST_PACKAGES_FAST))
+TEST_PACKAGES_NOWASI = \
+	crypto/ecdsa \
+	$(nil)
+
 # Report platforms on which each standard library package is known to pass tests
 report-stdlib-tests-pass:
 	$(eval jointmp := $(shell echo /tmp/join.$$$$))
@@ -478,10 +498,12 @@ report-stdlib-tests-pass:
 ifeq ($(uname),Darwin)
 TEST_PACKAGES_HOST := $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_DARWIN)
 TEST_IOFS := true
+TEST_ENCODING_XML := true
 endif
 ifeq ($(uname),Linux)
 TEST_PACKAGES_HOST := $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_LINUX)
 TEST_IOFS := true
+TEST_ENCODING_XML := true
 endif
 ifeq ($(OS),Windows_NT)
 TEST_PACKAGES_HOST := $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_WINDOWS)
@@ -496,8 +518,11 @@ TEST_ADDITIONAL_FLAGS ?=
 .PHONY: tinygo-test
 tinygo-test:
 	@# TestExtraMethods: used by many crypto packages and uses reflect.Type.Method which is not implemented.
-	@# TestParseAndBytesRoundTrip/P256/Generic: relies on t.Skip() which is not implemented
-	$(TINYGO) test $(TEST_ADDITIONAL_FLAGS) $(TEST_SKIP_FLAG) $(TEST_PACKAGES_HOST) $(TEST_PACKAGES_SLOW)
+	@# TestParseAndBytesRoundTrip/P256/Generic: needs Goexit to run defers on wasm.
+	$(TINYGO) test $(TEST_ADDITIONAL_FLAGS) $(TEST_SKIP_FLAG) $(filter-out encoding/xml,$(TEST_PACKAGES_HOST)) $(TEST_PACKAGES_SLOW)
+ifeq ($(TEST_ENCODING_XML),true)
+	$(TINYGO) test $(TEST_ADDITIONAL_FLAGS) $(TEST_SKIP_FLAG) -stack-size=16MB encoding/xml
+endif
 	@# io/fs requires os.ReadDir, not yet supported on windows or wasi. It also
 	@# requires a large stack-size. Hence, io/fs is only run conditionally.
 	@# For more details, see the comments on issue #3143.
@@ -519,12 +544,12 @@ tinygo-test-wasi:
 tinygo-test-wasip1:
 	GOOS=wasip1 GOARCH=wasm $(TINYGO) test $(TEST_SKIP_FLAG) $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW) ./tests/runtime_wasi
 tinygo-test-wasip1-fast:
-	$(TINYGO) test -target=wasip1 $(TEST_SKIP_FLAG) $(TEST_PACKAGES_FAST) ./tests/runtime_wasi
+	$(TINYGO) test -target=wasip1 $(TEST_SKIP_FLAG) $(TEST_PACKAGES_FAST_WASI) ./tests/runtime_wasi

 tinygo-test-wasip2-slow:
 	$(TINYGO) test -target=wasip2 $(TEST_SKIP_FLAG) $(TEST_PACKAGES_SLOW)
 tinygo-test-wasip2-fast:
-	$(TINYGO) test -target=wasip2 $(TEST_SKIP_FLAG) $(TEST_PACKAGES_FAST) ./tests/runtime_wasi
+	$(TINYGO) test -target=wasip2 $(TEST_SKIP_FLAG) $(TEST_PACKAGES_FAST_WASI) ./tests/runtime_wasi

 tinygo-test-wasip2-sum-slow:
 	TINYGO=$(TINYGO) \
@@ -554,13 +579,13 @@ tinygo-test-baremetal:

 # Test external packages in a large corpus.
 test-corpus:
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm llvm22" -run TestCorpus . -corpus=testdata/corpus.yaml
 test-corpus-fast:
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus -short . -corpus=testdata/corpus.yaml
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm llvm22" -run TestCorpus -short . -corpus=testdata/corpus.yaml
 test-corpus-wasi:
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip1
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm llvm22" -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip1
 test-corpus-wasip2:
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip2
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm llvm22" -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip2

 .PHONY: testchdir
 testchdir:
@@ -632,6 +657,8 @@ smoketest: testchdir
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pico2-ice           examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build             -o test.efi -target=uefi-amd64          examples/test
+	@$(MD5SUM) test.efi
 	# test simulated boards on play.tinygo.org
 ifneq ($(WASM), 0)
 	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=arduino_uno          examples/blinky1
@@ -742,6 +769,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=xiao                examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=xiao-ble-plus       examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=rak4631             examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/dac
@@ -778,6 +807,10 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pico                examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico -gc=leaking    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico-w              examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nano-33-ble         examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nano-rp2040         examples/blinky1
@@ -816,8 +849,14 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pico2               examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico2-w             examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=tiny2350            examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=badger2350          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=blinky2350          examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pico-plus2          examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=metro-rp2350        examples/blinky1
@@ -859,6 +898,8 @@ ifneq ($(STM32), 0)
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nucleo-f722ze       examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-h753zi       examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nucleo-l031k6       examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nucleo-l432kc       examples/blinky1
@@ -887,6 +928,8 @@ ifneq ($(STM32), 0)
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=stm32l0x1           examples/serial
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32u031           examples/empty
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=arduino-uno-q       examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=arduino-uno-q       examples/serial
@@ -941,12 +984,19 @@ ifneq ($(XTENSA), 0)
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target m5stack             examples/machinetest
 	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target m5stamp-s3a         examples/machinetest
+	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target m5stick-c           examples/machinetest
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target m5paper             examples/machinetest
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target mch2022             examples/machinetest
 	@$(MD5SUM) test.bin
+	# xiao-esp32c6
+	$(TINYGO) build -size short -o test.bin -target=xiao-esp32c6      	examples/blinky1
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=xiao-esp32c6   		examples/blinkm
+	@$(MD5SUM) test.bin
 	# xiao-esp32s3
 	$(TINYGO) build -size short -o test.bin -target=xiao-esp32s3   		examples/blinky1
 	@$(MD5SUM) test.bin
@@ -967,6 +1017,8 @@ ifneq ($(XTENSA), 0)
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target=esp32s3-supermini   	examples/adc
 	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=esp32s3-box-3       examples/blinky1
+	@$(MD5SUM) test.bin
 endif
    # esp32c3-supermini
 	$(TINYGO) build -size short -o test.bin -target=esp32c3-supermini	    examples/blinky1
@@ -1163,8 +1215,8 @@ endif
 	@cp -rp lib/wasi-libc/libc-top-half/musl/src/unistd             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
 	@cp -rp lib/wasi-libc/libc-top-half/sources                     build/release/tinygo/lib/wasi-libc/libc-top-half
 	@cp -rp lib/wasi-cli/wit                                        build/release/tinygo/lib/wasi-cli/wit
-	@cp -rp llvm-project/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt-builtins
-	@cp -rp llvm-project/compiler-rt/LICENSE.TXT  build/release/tinygo/lib/compiler-rt-builtins
+	@cp -rp ${LLVM_PROJECTDIR}/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt-builtins
+	@cp -rp ${LLVM_PROJECTDIR}/compiler-rt/LICENSE.TXT  build/release/tinygo/lib/compiler-rt-builtins
 	@cp -rp src                          build/release/tinygo/src
 	@cp -rp targets                      build/release/tinygo/targets

@@ -1,6 +1,6 @@
 # TinyGo - Go compiler for small places

-[![Linux](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml) [![macOS](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml) [![Windows](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml) [![Docker](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml) [![Nix](https://github.com/tinygo-org/tinygo/actions/workflows/nix.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/nix.yml) [![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev)
+[![Linux](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml) [![macOS](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml) [![Windows](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml) [![Docker](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml) [![Nix](https://github.com/tinygo-org/tinygo/actions/workflows/nix.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/nix.yml)

 TinyGo is a Go compiler intended for use in small places such as microcontrollers, WebAssembly (wasm/wasi), and command-line tools.

@@ -14,6 +14,7 @@ import (
 	"fmt"
 	"go/types"
 	"hash/crc32"
+	"maps"
 	"math/bits"
 	"os"
 	"os/exec"
@@ -137,9 +138,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		if _, ok := globalValues[pkgPath]; !ok {
 			globalValues[pkgPath] = map[string]string{}
 		}
-		for k, v := range vals {
-			globalValues[pkgPath][k] = v
-		}
+		maps.Copy(globalValues[pkgPath], vals)
 	}

 	// Check for a libc dependency.
@@ -255,6 +254,18 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		result.PackagePathMap[pkg.OriginalDir()] = pkg.Pkg.Path()
 	}

+	// Strip default initializers for -X globals from the type info before
+	// building SSA. This prevents go/ssa from emitting init stores for them,
+	// so that makeGlobalsModule can supply the correct values at final link
+	// time without any runtime init overwriting them. The -X values themselves
+	// are kept out of the per-package build cache; only the variable names
+	// appear in the cache key.
+	for _, pkg := range lprogram.Sorted() {
+		for name := range globalValues[pkg.Pkg.Path()] {
+			pkg.StripVarInitializer(name)
+		}
+	}
+
 	// Create the *ssa.Program. This does not yet build the entire SSA of the
 	// program so it's pretty fast and doesn't need to be parallelized.
 	program := lprogram.LoadSSA()
@@ -266,7 +277,6 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe

 	var embedFileObjects []*compileJob
 	for _, pkg := range lprogram.Sorted() {
-		pkg := pkg // necessary to avoid a race condition

 		var undefinedGlobals []string
 		for name := range globalValues[pkg.Pkg.Path()] {
@@ -477,7 +487,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 				if pkgInit.IsNil() {
 					panic("init not found for " + pkg.Pkg.Path())
 				}
-				err := interp.RunFunc(pkgInit, config.Options.InterpTimeout, config.DumpSSA())
+				err := interp.RunFunc(pkgInit, config.Options.InterpTimeout, config.Options.InterpMaxLoopIterations, config.DumpSSA())
 				if err != nil {
 					return err
 				}
@@ -763,7 +773,6 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	// TODO: do this as part of building the package to be able to link the
 	// bitcode files together.
 	for _, pkg := range lprogram.Sorted() {
-		pkg := pkg
 		for _, filename := range pkg.CFiles {
 			abspath := filepath.Join(pkg.OriginalDir(), filename)
 			job := &compileJob{
@@ -830,22 +839,25 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			}
 			ldflags = append(ldflags, "-mllvm", "-mcpu="+config.CPU())
 			ldflags = append(ldflags, "-mllvm", "-mattr="+config.Features()) // needed for MIPS softfloat
-			if config.GOOS() == "windows" {
-				// Options for the MinGW wrapper for the lld COFF linker.
+			switch config.LinkerFlavor() {
+			case "coff":
+				// Options for driving ld.lld in PE/COFF mode.
 				ldflags = append(ldflags,
 					"-Xlink=/opt:lldlto="+strconv.Itoa(speedLevel),
 					"--thinlto-cache-dir="+filepath.Join(cacheDir, "thinlto"))
-			} else if config.GOOS() == "darwin" {
+			case "darwin":
 				// Options for the ld64-compatible lld linker.
 				ldflags = append(ldflags,
 					"--lto-O"+strconv.Itoa(speedLevel),
 					"-cache_path_lto", filepath.Join(cacheDir, "thinlto"))
-			} else {
+			case "gnu":
 				// Options for the ELF linker.
 				ldflags = append(ldflags,
 					"--lto-O"+strconv.Itoa(speedLevel),
 					"--thinlto-cache-dir="+filepath.Join(cacheDir, "thinlto"),
 				)
+			default:
+				return fmt.Errorf("unknown linker flavor: %s", config.LinkerFlavor())
 			}
 			if config.CodeModel() != "default" {
 				ldflags = append(ldflags,
@@ -902,7 +914,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			}

 			// Run wasm-opt for wasm binaries
-			if arch := strings.Split(config.Triple(), "-")[0]; arch == "wasm32" {
+			if arch, _, _ := strings.Cut(config.Triple(), "-"); arch == "wasm32" {
 				optLevel, _, _ := config.OptLevel()
 				opt := "-" + optLevel

@@ -1064,7 +1076,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		if err != nil {
 			return result, err
 		}
-	case "esp32", "esp32-img", "esp32c3", "esp32s3", "esp8266":
+	case "esp32", "esp32-img", "esp32c3", "esp32s3", "esp32c6", "esp8266":
 		// Special format for the ESP family of chips (parsed by the ROM
 		// bootloader).
 		result.Binary = filepath.Join(tmpdir, "main"+outext)
@@ -1196,7 +1208,7 @@ func createEmbedObjectFile(data, hexSum, sourceFile, sourceDir, tmpdir string, c
 // needed to convert a program to its final form. Some transformations are not
 // optional and must be run as the compiler expects them to run.
 func optimizeProgram(mod llvm.Module, config *compileopts.Config) error {
-	err := interp.Run(mod, config.Options.InterpTimeout, config.DumpSSA())
+	err := interp.Run(mod, config.Options.InterpTimeout, config.Options.InterpMaxLoopIterations, config.DumpSSA())
 	if err != nil {
 		return err
 	}
@@ -1340,6 +1352,14 @@ func determineStackSizes(mod llvm.Module, executable string) ([]string, map[stri
 	}
 	baseStackSize, baseStackSizeType, baseStackSizeFailedAt := functions["tinygo_startTask"][0].StackSize()

+	// Account for the bytes that tinygo_swapTask pushes onto the goroutine stack
+	// on every context switch. The static analysis correctly traces Go calls,
+	// but it cannot see into the assembly-level register push.
+	var contextSwitchOverhead uint64
+	if swapFuncs, ok := functions["tinygo_swapTask"]; ok && len(swapFuncs) == 1 {
+		contextSwitchOverhead = swapFuncs[0].FrameSize
+	}
+
 	sizes := make(map[string]functionStackSize)

 	// Add the reset handler function, for convenience. The reset handler runs
@@ -1388,6 +1408,12 @@ func determineStackSizes(mod llvm.Module, executable string) ([]string, map[stri
 			// overflow will occur even before the goroutine is started.
 			stackSize = baseStackSize
 		}
+		if stackSizeType == stacksize.Bounded {
+			// Add the overhead of context switching. This is needed because the
+			// context switch (tinygo_swapTask) pushes callee-saved registers
+			// onto the current stack, which is not seen by the static analysis.
+			stackSize += contextSwitchOverhead
+		}
 		sizes[name] = functionStackSize{
 			stackSize:        stackSize,
 			stackSizeType:    stackSizeType,
@@ -5,6 +5,7 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+	"strings"
 	"testing"

 	"github.com/tinygo-org/tinygo/compileopts"
@@ -28,6 +29,7 @@ func TestClangAttributes(t *testing.T) {
 		"cortex-m4",
 		"cortex-m7",
 		"esp32c3",
+		"esp32c6",
 		"esp32s3",
 		"fe310",
 		"gameboy-advance",
@@ -35,6 +37,7 @@ func TestClangAttributes(t *testing.T) {
 		"nintendoswitch",
 		"riscv-qemu",
 		"tkey",
+		"uefi-amd64",
 		"wasip1",
 		"wasip2",
 		"wasm",
@@ -46,7 +49,6 @@ func TestClangAttributes(t *testing.T) {
 		targetNames = append(targetNames, "esp32", "esp8266")
 	}
 	for _, targetName := range targetNames {
-		targetName := targetName
 		t.Run(targetName, func(t *testing.T) {
 			testClangAttributes(t, &compileopts.Options{Target: targetName})
 		})
@@ -128,8 +130,10 @@ func testClangAttributes(t *testing.T, options *compileopts.Options) {
 	defer mod.Dispose()

 	// Check whether the LLVM target matches.
-	if mod.Target() != config.Triple() {
-		t.Errorf("target has LLVM triple %#v but Clang makes it LLVM triple %#v", config.Triple(), mod.Target())
+	// Use ClangTriple since LLVM 22 normalizes wasm32-unknown-wasi to wasip1.
+	expectedTriple := compileopts.ClangTriple(config.Triple())
+	if mod.Target() != expectedTriple {
+		t.Errorf("target has LLVM triple %#v but Clang makes it LLVM triple %#v", expectedTriple, mod.Target())
 	}

 	// Check the "target-cpu" and "target-features" string attribute of the add
@@ -153,11 +157,65 @@ func testClangAttributes(t *testing.T, options *compileopts.Options) {
 			// The reason is that Debian has patched Clang in a way that
 			// modifies the LLVM features string, changing lots of FPU/float
 			// related flags. We want to test vanilla Clang, not Debian Clang.
-			t.Errorf("target has LLVM features\n\t%#v\nbut Clang makes it\n\t%#v", config.Features(), features)
+			//
+			// Rather than requiring an exact match (which breaks across LLVM
+			// versions as new features are added), check that every feature
+			// TinyGo specifies is consistent with what Clang produces:
+			//  - A "+feature" in TinyGo must appear in Clang's output.
+			//  - A "-feature" in TinyGo must not be "+feature" in Clang's output.
+			checkFeatureFlags(t, config.Features(), features)
 		}
 	}
 }

+// checkFeatureFlags verifies that all features specified by TinyGo's target
+// configuration are consistent with Clang's output. This allows Clang to add
+// new features across LLVM versions without breaking the test.
+func checkFeatureFlags(t *testing.T, targetFeatures, clangFeatures string) {
+	t.Helper()
+
+	// Build a set of Clang's features for fast lookup.
+	clangSet := make(map[string]bool) // feature name -> enabled
+	for _, f := range strings.Split(clangFeatures, ",") {
+		f = strings.TrimSpace(f)
+		if len(f) < 2 {
+			continue
+		}
+		enabled := f[0] == '+'
+		name := f[1:]
+		clangSet[name] = enabled
+	}
+
+	// Check each feature that TinyGo specifies.
+	var missing, conflicts []string
+	for _, f := range strings.Split(targetFeatures, ",") {
+		f = strings.TrimSpace(f)
+		if len(f) < 2 {
+			continue
+		}
+		wantEnabled := f[0] == '+'
+		name := f[1:]
+
+		clangEnabled, inClang := clangSet[name]
+		if wantEnabled && (!inClang || !clangEnabled) {
+			// TinyGo requires +feature but Clang doesn't enable it.
+			missing = append(missing, f)
+		} else if !wantEnabled && inClang && clangEnabled {
+			// TinyGo requires -feature but Clang enables it.
+			conflicts = append(conflicts, fmt.Sprintf("target has %q but Clang has %q", f, "+"+name))
+		}
+	}
+
+	if len(missing) > 0 {
+		t.Errorf("target specifies features not present in Clang output: %s\n\ttarget features: %s\n\tclang features:  %s",
+			strings.Join(missing, ", "), targetFeatures, clangFeatures)
+	}
+	if len(conflicts) > 0 {
+		t.Errorf("target disables features that Clang enables: %s",
+			strings.Join(conflicts, "; "))
+	}
+}
+
 // This TestMain is necessary because TinyGo may also be invoked to run certain
 // LLVM tools in a separate process. Not capturing these invocations would lead
 // to recursive tests.
@@ -204,7 +204,18 @@ var avrBuiltins = []string{

 // Builtins needed specifically for windows/386.
 var windowsI386Builtins = []string{
-	"i386/chkstk.S", // also _alloca
+	"i386/chkstk.S",  // __chkstk_ms
+	"i386/chkstk2.S", // _alloca (__alloca)
+}
+
+// Builtins needed specifically for windows/amd64.
+var windowsAMD64Builtins = []string{
+	"x86_64/chkstk.S",
+}
+
+// Builtins needed specifically for windows/arm64.
+var windowsARM64Builtins = []string{
+	"aarch64/chkstk.S",
 }

 // libCompilerRT is a library with symbols required by programs compiled with
@@ -233,13 +244,28 @@ var libCompilerRT = Library{
 			builtins = append(builtins, aeabiBuiltins...)
 		case "avr":
 			builtins = append(builtins, avrBuiltins...)
-		case "x86_64", "aarch64", "riscv64": // any 64-bit arch
+		case "x86_64":
+			builtins = append(builtins, genericBuiltins128...)
+			if isWindowsTriple(target) {
+				builtins = append(builtins, windowsAMD64Builtins...)
+			}
+		case "aarch64":
+			builtins = append(builtins, genericBuiltins128...)
+			if isWindowsTriple(target) {
+				builtins = append(builtins, windowsARM64Builtins...)
+			}
+		case "riscv64":
 			builtins = append(builtins, genericBuiltins128...)
 		case "i386":
-			if strings.Split(target, "-")[2] == "windows" {
+			if isWindowsTriple(target) {
 				builtins = append(builtins, windowsI386Builtins...)
 			}
 		}
 		return builtins, nil
 	},
 }
+
+func isWindowsTriple(target string) bool {
+	parts := strings.Split(target, "-")
+	return len(parts) > 2 && parts[2] == "windows"
+}
@@ -281,7 +281,7 @@ func parseDepFile(s string) ([]string, error) {
 	s = strings.ReplaceAll(s, "\\\n", " ")

 	// Only use the first line, which is expected to begin with "deps:".
-	line := strings.SplitN(s, "\n", 2)[0]
+	line, _, _ := strings.Cut(s, "\n")
 	if !strings.HasPrefix(line, "deps:") {
 		return nil, errors.New("readDepFile: expected 'deps:' prefix")
 	}
@@ -23,7 +23,7 @@
 #include "clang/Basic/Diagnostic.h"
 #include "clang/Basic/DiagnosticOptions.h"
 #include "clang/Driver/DriverDiagnostic.h"
-#include "clang/Driver/Options.h"
+#include "clang/Options/Options.h"
 #include "clang/Frontend/FrontendDiagnostic.h"
 #include "clang/Frontend/TextDiagnosticPrinter.h"
 #include "clang/Frontend/Utils.h"
@@ -35,6 +35,7 @@
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInstPrinter.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCObjectFileInfo.h"
 #include "llvm/MC/MCObjectWriter.h"
@@ -67,8 +68,7 @@
 #include <optional>
 #include <system_error>
 using namespace clang;
-using namespace clang::driver;
-using namespace clang::driver::options;
+using namespace clang::options;
 using namespace llvm;
 using namespace llvm::opt;

@@ -109,7 +109,7 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
  // Construct the invocation.

  // Target Options
-  Opts.Triple = llvm::Triple::normalize(Args.getLastArgValue(OPT_triple));
+  Opts.Triple = llvm::Triple(llvm::Triple::normalize(Args.getLastArgValue(OPT_triple)));
  if (Arg *A = Args.getLastArg(options::OPT_darwin_target_variant_triple))
    Opts.DarwinTargetVariantTriple = llvm::Triple(A->getValue());
  if (Arg *A = Args.getLastArg(OPT_darwin_target_variant_sdk_version_EQ)) {
@@ -125,8 +125,8 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
  Opts.Features = Args.getAllArgValues(OPT_target_feature);

  // Use the default target triple if unspecified.
-  if (Opts.Triple.empty())
-    Opts.Triple = llvm::sys::getDefaultTargetTriple();
+  if (Opts.Triple.getTriple().empty())
+    Opts.Triple = llvm::Triple(llvm::sys::getDefaultTargetTriple());

  // Language Options
  Opts.IncludePaths = Args.getAllArgValues(OPT_I);
@@ -267,7 +267,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
  std::string Error;
  const Target *TheTarget = TargetRegistry::lookupTarget(Opts.Triple, Error);
  if (!TheTarget)
-    return Diags.Report(diag::err_target_unknown_triple) << Opts.Triple;
+    return Diags.Report(diag::err_target_unknown_triple) << Opts.Triple.str();

  ErrorOr<std::unique_ptr<MemoryBuffer>> Buffer =
      MemoryBuffer::getFileOrSTDIN(Opts.InputFile, /*IsText=*/true);
@@ -327,7 +327,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
      TheTarget->createMCSubtargetInfo(Opts.Triple, Opts.CPU, FS));
  assert(STI && "Unable to create subtarget info!");

-  MCContext Ctx(Triple(Opts.Triple), MAI.get(), MRI.get(), STI.get(), &SrcMgr,
+  MCContext Ctx(Opts.Triple, MAI.get(), MRI.get(), STI.get(), &SrcMgr,
                &MCOptions);

  bool PIC = false;
@@ -390,8 +390,8 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,

  // FIXME: There is a bit of code duplication with addPassesToEmitFile.
  if (Opts.OutputType == AssemblerInvocation::FT_Asm) {
-    MCInstPrinter *IP = TheTarget->createMCInstPrinter(
-        llvm::Triple(Opts.Triple), Opts.OutputAsmVariant, *MAI, *MCII, *MRI);
+    std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
+        Opts.Triple, Opts.OutputAsmVariant, *MAI, *MCII, *MRI));

    std::unique_ptr<MCCodeEmitter> CE;
    if (Opts.ShowEncoding)
@@ -400,7 +400,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
        TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));

    auto FOut = std::make_unique<formatted_raw_ostream>(*Out);
-    Str.reset(TheTarget->createAsmStreamer(Ctx, std::move(FOut), IP,
+    Str.reset(TheTarget->createAsmStreamer(Ctx, std::move(FOut), std::move(IP),
                                           std::move(CE), std::move(MAB)));
  } else if (Opts.OutputType == AssemblerInvocation::FT_Null) {
    Str.reset(createNullStreamer(Ctx));
@@ -422,7 +422,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
        DwoOS ? MAB->createDwoObjectWriter(*Out, *DwoOS)
              : MAB->createObjectWriter(*Out);

-    Triple T(Opts.Triple);
+    Triple T = Opts.Triple;
    Str.reset(TheTarget->createMCObjectStreamer(
        T, Ctx, std::move(MAB), std::move(OW), std::move(CE), *STI));
    Str.get()->initSections(Opts.NoExecStack, *STI);
@@ -453,7 +453,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
  std::unique_ptr<MCTargetAsmParser> TAP(
      TheTarget->createMCAsmParser(*STI, *Parser, *MCII, MCOptions));
  if (!TAP)
-    Failed = Diags.Report(diag::err_target_unknown_triple) << Opts.Triple;
+    Failed = Diags.Report(diag::err_target_unknown_triple) << Opts.Triple.str();

  // Set values for symbols, if any.
  for (auto &S : Opts.SymbolDefs) {
@@ -506,12 +506,12 @@ int cc1as_main(ArrayRef<const char *> Argv, const char *Argv0, void *MainAddr) {
  InitializeAllAsmParsers();

  // Construct our diagnostic client.
-  IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions();
+  DiagnosticOptions DiagOpts;
  TextDiagnosticPrinter *DiagClient
-    = new TextDiagnosticPrinter(errs(), &*DiagOpts);
+    = new TextDiagnosticPrinter(errs(), DiagOpts);
  DiagClient->setPrefix("clang -cc1as");
  IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
-  DiagnosticsEngine Diags(DiagID, &*DiagOpts, DiagClient);
+  DiagnosticsEngine Diags(DiagID, DiagOpts, DiagClient);

  // Set an error handler, so that any LLVM backend diagnostics go through our
  // error handler.
@@ -528,7 +528,7 @@ int cc1as_main(ArrayRef<const char *> Argv, const char *Argv0, void *MainAddr) {
        llvm::outs(), "clang -cc1as [options] file...",
        "Clang Integrated Assembler", /*ShowHidden=*/false,
        /*ShowAllAliases=*/false,
-        llvm::opt::Visibility(driver::options::CC1AsOption));
+        llvm::opt::Visibility(clang::options::CC1AsOption));

    return 0;
  }
@@ -20,7 +20,7 @@ struct AssemblerInvocation {
  /// @{

  /// The name of the target triple to assemble for.
-  std::string Triple;
+  llvm::Triple Triple;

  /// If given, the name of the target CPU to determine which instructions
  /// are legal.
@@ -142,7 +142,7 @@ struct AssemblerInvocation {

 public:
  AssemblerInvocation() {
-    Triple = "";
+    Triple = llvm::Triple();
    NoInitialTextSection = 0;
    InputFile = "-";
    OutputPath = "-";
@@ -27,9 +27,9 @@ bool tinygo_clang_driver(int argc, char **argv) {
 	std::vector<const char*> args(argv, argv + argc);

 	// The compiler invocation needs a DiagnosticsEngine so it can report problems
-	llvm::IntrusiveRefCntPtr<clang::DiagnosticOptions> DiagOpts = new clang::DiagnosticOptions();
-	clang::TextDiagnosticPrinter DiagnosticPrinter(llvm::errs(), &*DiagOpts);
-	clang::DiagnosticsEngine Diags(llvm::IntrusiveRefCntPtr<clang::DiagnosticIDs>(new clang::DiagnosticIDs()), &*DiagOpts, &DiagnosticPrinter, false);
+	clang::DiagnosticOptions DiagOpts;
+	clang::TextDiagnosticPrinter DiagnosticPrinter(llvm::errs(), DiagOpts);
+	clang::DiagnosticsEngine Diags(llvm::IntrusiveRefCntPtr<clang::DiagnosticIDs>(new clang::DiagnosticIDs()), DiagOpts, &DiagnosticPrinter, false);

 	// Create the clang driver
 	clang::driver::Driver TheDriver(args[0], llvm::sys::getDefaultTargetTriple(), Diags);
@@ -60,7 +60,7 @@ bool tinygo_clang_driver(int argc, char **argv) {
 			}

 			// Create the actual diagnostics engine.
-			Clang->createDiagnostics(*llvm::vfs::getRealFileSystem());
+			Clang->createDiagnostics();
 			if (!Clang->hasDiagnostics()) {
 				return false;
 			}
@@ -17,7 +17,7 @@ import (
 var commands = map[string][]string{}

 func init() {
-	llvmMajor := strings.Split(llvm.Version, ".")[0]
+	llvmMajor, _, _ := strings.Cut(llvm.Version, ".")
 	commands["clang"] = []string{"clang-" + llvmMajor}
 	commands["ld.lld"] = []string{"ld.lld-" + llvmMajor, "ld.lld"}
 	commands["wasm-ld"] = []string{"wasm-ld-" + llvmMajor, "wasm-ld"}
@@ -25,8 +25,8 @@ func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
 	}

 	// Version range supported by TinyGo.
-	const minorMin = 19
-	const minorMax = 26
+	const minorMin = 25 // when updating the min version, also update .github/workflows/compat.yml
+	const minorMax = 27

 	// Check that we support this Go toolchain version.
 	gorootMajor, gorootMinor, err := goenv.GetGorootVersion()
@@ -15,7 +15,7 @@ func makeDarwinLibSystemJob(config *compileopts.Config, tmpdir string) *compileJ
 	return &compileJob{
 		description: "compile Darwin libSystem.dylib",
 		run: func(job *compileJob) (err error) {
-			arch := strings.Split(config.Triple(), "-")[0]
+			arch, _, _ := strings.Cut(config.Triple(), "-")
 			job.result = filepath.Join(tmpdir, "libSystem.dylib")
 			objpath := filepath.Join(tmpdir, "libSystem.o")
 			inpath := filepath.Join(goenv.Get("TINYGOROOT"), "lib/macos-minimal-sdk/src", arch, "libSystem.s")
@@ -65,15 +65,6 @@ func makeESPFirmwareImage(infile, outfile, format string) error {
 	// Sort the segments by address. This is what esptool does too.
 	sort.SliceStable(segments, func(i, j int) bool { return segments[i].addr < segments[j].addr })

-	// Calculate checksum over the segment data. This is used in the image
-	// footer.
-	checksum := uint8(0xef)
-	for _, segment := range segments {
-		for _, b := range segment.data {
-			checksum ^= b
-		}
-	}
-
 	// Write first to an in-memory buffer, primarily so that we can easily
 	// calculate a hash over the entire image.
 	// An added benefit is that we don't need to check for errors all the time.
@@ -88,6 +79,86 @@ func makeESPFirmwareImage(infile, outfile, format string) error {
 		chip = format[:len(format)-len("-img")]
 	}

+	// For ESP32 (original): separate RAM segments (loadable by ROM bootloader)
+	// from flash-mapped segments (DROM/IROM, require MMU setup by startup code).
+	// The ROM bootloader on ESP32 does NOT handle flash-mapped segments —
+	// it tries to memcpy to the virtual address, which crashes.
+	var flashSegments []*espImageSegment
+	if chip == "esp32" {
+		var ramSegments []*espImageSegment
+		for _, seg := range segments {
+			if (seg.addr >= 0x3F400000 && seg.addr < 0x3F800000) || // DROM
+				(seg.addr >= 0x400D0000 && seg.addr < 0x40400000) { // IROM
+				flashSegments = append(flashSegments, seg)
+			} else {
+				ramSegments = append(ramSegments, seg)
+			}
+		}
+		segments = ramSegments
+	}
+
+	// ESP32 flash XIP: compute where the DROM segment will be placed in flash
+	// (page-aligned, right after the RAM segments) and patch the
+	// _drom_flash_addr variable so the startup code can program the cache MMU.
+	// This must happen before the checksum/hash are computed so the patched
+	// value is covered by both.
+	const esp32FlashBase = 0x1000 // esptool flashes the image at 0x1000
+	// The ESP32 flash cache MMU supports configurable page sizes down to 256 B. 64 KiB is the reset/default size.
+	// If the startup code ever changes the MMU page size, this constant must change too.
+	const esp32PageSize = 0x10000 // 64KB MMU pages
+	var esp32DromFlashAddr uint32
+	if chip == "esp32" && len(flashSegments) > 0 {
+		// Compute the size of the RAM portion of the image (everything the ROM
+		// bootloader loads, up to and including the appended SHA256 hash).
+		ramImageSize := 0
+		if makeImage {
+			ramImageSize += 4096
+		}
+		ramImageSize += 24 // image header (8) + trailer fields (16)
+		for _, seg := range segments {
+			ramImageSize += 8 + len(seg.data) // segment header + data (4-aligned)
+		}
+		ramImageSize += 16 - ramImageSize%16 // footer padding + checksum byte
+		ramImageSize += 32                   // appended SHA256 hash
+
+		// DROM flash address must be 64KB page-aligned.
+		esp32DromFlashAddr = uint32(esp32FlashBase+ramImageSize+esp32PageSize-1) &^ (esp32PageSize - 1)
+
+		// Patch _drom_flash_addr in whichever RAM segment contains it.
+		syms, _ := inf.Symbols()
+		var dromSymAddr uint64
+		for _, s := range syms {
+			if s.Name == "_drom_flash_addr" {
+				dromSymAddr = s.Value
+				break
+			}
+		}
+		if dromSymAddr == 0 {
+			return fmt.Errorf("ESP32: _drom_flash_addr symbol not found")
+		}
+		patched := false
+		for _, seg := range segments {
+			if dromSymAddr >= uint64(seg.addr) && dromSymAddr+4 <= uint64(seg.addr)+uint64(len(seg.data)) {
+				off := int(dromSymAddr - uint64(seg.addr))
+				binary.LittleEndian.PutUint32(seg.data[off:], esp32DromFlashAddr)
+				patched = true
+				break
+			}
+		}
+		if !patched {
+			return fmt.Errorf("ESP32: _drom_flash_addr (0x%x) not in any RAM segment", dromSymAddr)
+		}
+	}
+
+	// Calculate checksum over the segment data. This is used in the image
+	// footer.
+	checksum := uint8(0xef)
+	for _, segment := range segments {
+		for _, b := range segment.data {
+			checksum ^= b
+		}
+	}
+
 	if makeImage {
 		// The bootloader starts at 0x1000, or 4096.
 		// TinyGo doesn't use a separate bootloader and runs the entire
@@ -100,12 +171,24 @@ func makeESPFirmwareImage(infile, outfile, format string) error {
 	chip_id := map[string]uint16{
 		"esp32":   0x0000,
 		"esp32c3": 0x0005,
+		"esp32c6": 0x000d,
 		"esp32s3": 0x0009,
 	}[chip]

+	// SPI flash speed/size byte (byte 3 of header):
+	//   Upper nibble = flash size, lower nibble = flash frequency.
+	//   The espflasher auto-detects and patches the flash size (upper nibble),
+	//   but the frequency (lower nibble) must be correct per chip.
+	spiSpeedSize := map[string]uint8{
+		"esp32":   0x1f, // 80MHz=0x0F, 2MB=0x10
+		"esp32c3": 0x1f, // 80MHz=0x0F, 2MB=0x10
+		"esp32c6": 0x10, // 80MHz=0x00, 2MB=0x10 (C6 uses different freq encoding)
+		"esp32s3": 0x1f, // 80MHz=0x0F, 2MB=0x10
+	}[chip]
+
 	// Image header.
 	switch chip {
-	case "esp32", "esp32c3", "esp32s3":
+	case "esp32", "esp32c3", "esp32s3", "esp32c6":
 		// Header format:
 		// https://github.com/espressif/esp-idf/blob/v4.3/components/bootloader_support/include/esp_app_format.h#L71
 		// Note: not adding a SHA256 hash as the binary is modified by
@@ -126,8 +209,8 @@ func makeESPFirmwareImage(infile, outfile, format string) error {
 		}{
 			magic:          0xE9,
 			segment_count:  byte(len(segments)),
-			spi_mode:       2,    // ESP_IMAGE_SPI_MODE_DIO
-			spi_speed_size: 0x1f, // ESP_IMAGE_SPI_SPEED_80M, ESP_IMAGE_FLASH_SIZE_2MB
+			spi_mode:       2, // ESP_IMAGE_SPI_MODE_DIO
+			spi_speed_size: spiSpeedSize,
 			entry_addr:     uint32(inf.Entry),
 			wp_pin:         0xEE, // disable WP pin
 			chip_id:        chip_id,
@@ -179,6 +262,58 @@ func makeESPFirmwareImage(infile, outfile, format string) error {
 		outf.Write(hash[:])
 	}

+	// For ESP32: append flash-mapped segments (DROM/IROM) at page-aligned flash
+	// offsets after the RAM portion. The startup code maps them via the flash
+	// cache MMU (DROM at esp32DromFlashAddr, patched into _drom_flash_addr).
+	if len(flashSegments) > 0 {
+		const flashBase = esp32FlashBase
+		const pageSize = esp32PageSize
+		dromFlashAddr := esp32DromFlashAddr
+
+		// Separate DROM and IROM segments.
+		var dromSegs, iromSegs []*espImageSegment
+		for _, seg := range flashSegments {
+			if seg.addr >= 0x3F400000 && seg.addr < 0x3F800000 {
+				dromSegs = append(dromSegs, seg)
+			} else {
+				iromSegs = append(iromSegs, seg)
+			}
+		}
+
+		// Write DROM segments at the computed page-aligned flash offset.
+		dromSize := 0
+		if len(dromSegs) > 0 {
+			targetImageOffset := int(dromFlashAddr - flashBase)
+			if outf.Len() > targetImageOffset {
+				return fmt.Errorf("ESP32: RAM segments too large (%d bytes), overlap DROM at flash 0x%x", outf.Len(), dromFlashAddr)
+			}
+			outf.Write(make([]byte, targetImageOffset-outf.Len()))
+			for _, seg := range dromSegs {
+				outf.Write(seg.data)
+				dromSize += len(seg.data)
+			}
+		}
+
+		// Write IROM segments immediately after DROM, at the next page boundary.
+		// IROM flash addr = dromFlashAddr + ceil(dromSize/pageSize)*pageSize
+		// (must match the computation in the startup assembly).
+		if len(iromSegs) > 0 {
+			dromPages := (dromSize + pageSize - 1) / pageSize
+			if dromPages == 0 {
+				dromPages = 1
+			}
+			iromFlashAddr := dromFlashAddr + uint32(dromPages)*pageSize
+			targetImageOffset := int(iromFlashAddr - flashBase)
+			if outf.Len() > targetImageOffset {
+				return fmt.Errorf("ESP32: DROM too large, overlaps IROM at flash 0x%x", iromFlashAddr)
+			}
+			outf.Write(make([]byte, targetImageOffset-outf.Len()))
+			for _, seg := range iromSegs {
+				outf.Write(seg.data)
+			}
+		}
+	}
+
 	// QEMU (or more precisely, qemu-system-xtensa from Espressif) expects the
 	// image to be a certain size.
 	if makeImage {
@@ -195,11 +195,11 @@ type intHeap struct {
 	sort.IntSlice
 }

-func (h *intHeap) Push(x interface{}) {
+func (h *intHeap) Push(x any) {
 	h.IntSlice = append(h.IntSlice, x.(int))
 }

-func (h *intHeap) Pop() interface{} {
+func (h *intHeap) Pop() any {
 	x := h.IntSlice[len(h.IntSlice)-1]
 	h.IntSlice = h.IntSlice[:len(h.IntSlice)-1]
 	return x
@@ -133,7 +133,7 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 	// Note: -fdebug-prefix-map is necessary to make the output archive
 	// reproducible. Otherwise the temporary directory is stored in the archive
 	// itself, which varies each run.
-	args := append(l.cflags(target, headerPath), "-c", "-Oz", "-gdwarf-4", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir)
+	args := append(l.cflags(target, headerPath), "-c", "-Oz", "-gdwarf-4", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target="+compileopts.ClangTriple(target), "-fdebug-prefix-map="+dir+"="+remapDir)
 	resourceDir := goenv.ClangResourceDir(false)
 	if resourceDir != "" {
 		args = append(args, "-resource-dir="+resourceDir)
@@ -218,7 +218,7 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 				return err
 			}
 			// Store this archive in the cache.
-			return os.Rename(f.Name(), archiveFilePath)
+			return robustRename(f.Name(), archiveFilePath)
 		},
 	}

@@ -232,7 +232,6 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 	}
 	for _, path := range paths {
 		// Strip leading "../" parts off the path.
-		path := path
 		cleanpath := path
 		for strings.HasPrefix(cleanpath, "../") {
 			cleanpath = cleanpath[3:]
@@ -28,8 +28,8 @@ func buildMuslAllTypes(arch, muslDir, outputBitsDir string) error {
 		if err != nil {
 			return err
 		}
-		lines := strings.Split(string(data), "\n")
-		for _, line := range lines {
+		lines := strings.SplitSeq(string(data), "\n")
+		for line := range lines {
 			if strings.HasPrefix(line, "TYPEDEF ") {
 				matches := regexp.MustCompile(`TYPEDEF (.*) ([^ ]*);`).FindStringSubmatch(line)
 				value := matches[1]
@@ -0,0 +1,9 @@
+//go:build !windows
+
+package builder
+
+import "os"
+
+func robustRename(oldpath, newpath string) error {
+	return os.Rename(oldpath, newpath)
+}
@@ -0,0 +1,44 @@
+package builder
+
+import (
+	"errors"
+	"math/rand"
+	"os"
+	"syscall"
+	"time"
+)
+
+const robustRenameTimeout = 2 * time.Second
+
+func robustRename(oldpath, newpath string) error {
+	var bestErr error
+	start := time.Now()
+	nextSleep := time.Millisecond
+	for {
+		err := os.Rename(oldpath, newpath)
+		if err == nil || !isEphemeralRenameError(err) {
+			return err
+		}
+		if bestErr == nil {
+			bestErr = err
+		}
+		if d := time.Since(start) + nextSleep; d >= robustRenameTimeout {
+			return bestErr
+		}
+		time.Sleep(nextSleep)
+		nextSleep += time.Duration(rand.Int63n(int64(nextSleep)))
+	}
+}
+
+func isEphemeralRenameError(err error) bool {
+	var errno syscall.Errno
+	if errors.As(err, &errno) {
+		switch errno {
+		case syscall.Errno(2), // ERROR_FILE_NOT_FOUND
+			syscall.Errno(5),  // ERROR_ACCESS_DENIED
+			syscall.Errno(32): // ERROR_SHARING_VIOLATION
+			return true
+		}
+	}
+	return false
+}
@@ -501,8 +501,9 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz
 						Align:   section.Addralign,
 						Type:    memoryStack,
 					})
-				} else {
-					// Regular .bss section.
+				} else if section.Flags&elf.SHF_WRITE != 0 {
+					// Regular .bss section. Zero-initialized RAM is always
+					// writable, so require SHF_WRITE here.
 					sections = append(sections, memorySection{
 						Address: section.Addr,
 						Size:    section.Size,
@@ -510,6 +511,11 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz
 						Type:    memoryBSS,
 					})
 				}
+				// Other (non-writable) SHT_NOBITS sections are address-space
+				// placeholders that occupy no RAM, such as the ESP linker
+				// script's .irom_dummy / .rodata_dummy sections which reserve
+				// the flash-mapped XIP virtual address ranges. They must not be
+				// counted as bss/RAM usage.
 			} else if section.Type == elf.SHT_PROGBITS && section.Flags&elf.SHF_EXECINSTR != 0 {
 				// .text
 				sections = append(sections, memorySection{
@@ -42,16 +42,15 @@ func TestBinarySize(t *testing.T) {
 	// This is a small number of very diverse targets that we want to test.
 	tests := []sizeTest{
 		// microcontrollers
-		{"hifive1b", "examples/echo", 3680, 280, 0, 2252},
-		{"microbit", "examples/serial", 2694, 342, 8, 2248},
-		{"wioterminal", "examples/pininterrupt", 7074, 1510, 120, 7248},
+		{"hifive1b", "examples/echo", 4313, 323, 0, 2260},
+		{"microbit", "examples/serial", 2838, 382, 8, 2256},
+		{"wioterminal", "examples/pininterrupt", 8027, 1665, 132, 7488},

 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the
 		// output varies by binaryen version.
 	}
 	for _, tc := range tests {
-		tc := tc
 		t.Run(tc.target+"/"+tc.path, func(t *testing.T) {
 			t.Parallel()

@@ -85,7 +84,6 @@ func TestSizeFull(t *testing.T) {
 	pkgMatch := regexp.MustCompile(`^[a-z/]+$`)    // example: "internal/task"

 	for _, target := range tests {
-		target := target
 		t.Run(target, func(t *testing.T) {
 			t.Parallel()

@@ -99,7 +97,7 @@ func TestSizeFull(t *testing.T) {
 				t.Fatal("could not read program size:", err)
 			}
 			for _, pkg := range sizes.sortedPackageNames() {
-				if pkg == "(padding)" || pkg == "(unknown)" {
+				if pkg == "(padding)" || pkg == "(unknown)" || pkg == "Go types" {
 					// TODO: correctly attribute all unknown binary size.
 					continue
 				}
@@ -28,7 +28,7 @@ func RunTool(tool string, args ...string) error {
 	var cflag *C.char
 	buf := C.calloc(C.size_t(len(args)), C.size_t(unsafe.Sizeof(cflag)))
 	defer C.free(buf)
-	cflags := (*[1 << 10]*C.char)(unsafe.Pointer(buf))[:len(args):len(args)]
+	cflags := unsafe.Slice((**C.char)(buf), len(args))
 	for i, flag := range args {
 		cflag := C.CString(flag)
 		cflags[i] = cflag
@@ -116,18 +116,24 @@ func parseLLDErrors(text string) error {
 		// This can happen in some cases like with CGo and //go:linkname tricker.
 		if matches := regexp.MustCompile(`^ld.lld(-[0-9]+)?: error: undefined symbol: (.*)\n`).FindStringSubmatch(message); matches != nil {
 			symbolName := matches[2]
-			for _, line := range strings.Split(message, "\n") {
+			for line := range strings.SplitSeq(message, "\n") {
 				matches := regexp.MustCompile(`referenced by .* \(((.*):([0-9]+))\)`).FindStringSubmatch(line)
 				if matches != nil {
 					parsedError = true
 					line, _ := strconv.Atoi(matches[3])
-					// TODO: detect common mistakes like -gc=none?
+					msg := "linker could not find symbol " + symbolName
+					switch symbolName {
+					case "runtime.alloc":
+						msg = "object allocated on the heap with -gc=none"
+					case "runtime.alloc_noheap":
+						msg = "object allocated on the heap in //go:noheap function"
+					}
 					linkErrors = append(linkErrors, scanner.Error{
 						Pos: token.Position{
 							Filename: matches[2],
 							Line:     line,
 						},
-						Msg: "linker could not find symbol " + symbolName,
+						Msg: msg,
 					})
 				}
 			}
@@ -40,7 +40,7 @@ func convertBinToUF2(input []byte, targetAddr uint32, uf2FamilyID string) ([]byt
 	}
 	bl.SetNumBlocks(len(blocks))

-	for i := 0; i < len(blocks); i++ {
+	for i := range blocks {
 		bl.SetBlockNo(i)
 		bl.SetData(blocks[i])

@@ -26,10 +26,6 @@ import (
 	"golang.org/x/tools/go/ast/astutil"
 )

-// Function that's only defined in Go 1.22.
-var setASTFileFields = func(f *ast.File, start, end token.Pos) {
-}
-
 // cgoPackage holds all CGo-related information of a package.
 type cgoPackage struct {
 	generated       *ast.File
@@ -44,7 +40,7 @@ type cgoPackage struct {
 	tokenFiles      map[string]*token.File
 	definedGlobally map[string]ast.Node
 	noescapingFuncs map[string]*noescapingFunc // #cgo noescape lines
-	anonDecls       map[interface{}]string
+	anonDecls       map[any]string
 	cflags          []string // CFlags from #cgo lines
 	ldflags         []string // LDFlags from #cgo lines
 	visitedFiles    map[string][]byte
@@ -263,7 +259,7 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 		tokenFiles:      map[string]*token.File{},
 		definedGlobally: map[string]ast.Node{},
 		noescapingFuncs: map[string]*noescapingFunc{},
-		anonDecls:       map[interface{}]string{},
+		anonDecls:       map[any]string{},
 		visitedFiles:    map[string][]byte{},
 	}

@@ -306,7 +302,7 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 	// Find `import "C"` C fragments in the file.
 	p.cgoHeaders = make([]string, len(files)) // combined CGo header fragment for each file
 	for i, f := range files {
-		var cgoHeader string
+		var cgoHeader strings.Builder
 		for i := 0; i < len(f.Decls); i++ {
 			decl := f.Decls[i]
 			genDecl, ok := decl.(*ast.GenDecl)
@@ -341,7 +337,8 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 			// Iterate through all parts of the CGo header. Note that every //
 			// line is a new comment.
 			position := fset.Position(genDecl.Doc.Pos())
-			fragment := fmt.Sprintf("# %d %#v\n", position.Line, position.Filename)
+			var fragment strings.Builder
+			fragment.WriteString(fmt.Sprintf("# %d %#v\n", position.Line, position.Filename))
 			for _, comment := range genDecl.Doc.List {
 				// Find all #cgo lines, extract and use their contents, and
 				// replace the lines with spaces (to preserve locations).
@@ -358,12 +355,13 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 				} else { // comment
 					c = "  " + c[2:len(c)-2]
 				}
-				fragment += c + "\n"
+				fragment.WriteString(c)
+				fragment.WriteByte('\n')
 			}
-			cgoHeader += fragment
+			cgoHeader.WriteString(fragment.String())
 		}

-		p.cgoHeaders[i] = cgoHeader
+		p.cgoHeaders[i] = cgoHeader.String()
 	}

 	// Define CFlags that will be used while parsing the package.
@@ -654,7 +652,6 @@ func (p *cgoPackage) createUnionAccessor(field *ast.Field, typeName string) {
 		X: &ast.Ident{
 			NamePos: pos,
 			Name:    "union",
-			Obj:     nil,
 		},
 		Sel: &ast.Ident{
 			NamePos: pos,
@@ -708,7 +705,6 @@ func (p *cgoPackage) createUnionAccessor(field *ast.Field, typeName string) {
 						X: &ast.Ident{
 							NamePos: pos,
 							Name:    typeName,
-							Obj:     nil,
 						},
 					},
 				},
@@ -764,7 +760,6 @@ func (p *cgoPackage) createBitfieldGetter(bitfield bitfieldInfo, typeName string
 		X: &ast.Ident{
 			NamePos: bitfield.pos,
 			Name:    "s",
-			Obj:     nil,
 		},
 		Sel: &ast.Ident{
 			NamePos: bitfield.pos,
@@ -811,11 +806,6 @@ func (p *cgoPackage) createBitfieldGetter(bitfield bitfieldInfo, typeName string
 						{
 							NamePos: bitfield.pos,
 							Name:    "s",
-							Obj: &ast.Object{
-								Kind: ast.Var,
-								Name: "s",
-								Decl: nil,
-							},
 						},
 					},
 					Type: &ast.StarExpr{
@@ -823,7 +813,6 @@ func (p *cgoPackage) createBitfieldGetter(bitfield bitfieldInfo, typeName string
 						X: &ast.Ident{
 							NamePos: bitfield.pos,
 							Name:    typeName,
-							Obj:     nil,
 						},
 					},
 				},
@@ -881,7 +870,6 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 		X: &ast.Ident{
 			NamePos: bitfield.pos,
 			Name:    "s",
-			Obj:     nil,
 		},
 		Sel: &ast.Ident{
 			NamePos: bitfield.pos,
@@ -964,11 +952,6 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 						{
 							NamePos: bitfield.pos,
 							Name:    "s",
-							Obj: &ast.Object{
-								Kind: ast.Var,
-								Name: "s",
-								Decl: nil,
-							},
 						},
 					},
 					Type: &ast.StarExpr{
@@ -976,7 +959,6 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 						X: &ast.Ident{
 							NamePos: bitfield.pos,
 							Name:    typeName,
-							Obj:     nil,
 						},
 					},
 				},
@@ -997,7 +979,6 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 							{
 								NamePos: bitfield.pos,
 								Name:    "value",
-								Obj:     nil,
 							},
 						},
 						Type: bitfield.field.Type,
@@ -1015,7 +996,6 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 							X: &ast.Ident{
 								NamePos: bitfield.pos,
 								Name:    "s",
-								Obj:     nil,
 							},
 							Sel: &ast.Ident{
 								NamePos: bitfield.pos,
@@ -1239,7 +1219,7 @@ func getPos(node ast.Node) token.Pos {
 // getUnnamedDeclName creates a name (with the given prefix) for the given C
 // declaration. This is used for structs, unions, and enums that are often
 // defined without a name and used in a typedef.
-func (p *cgoPackage) getUnnamedDeclName(prefix string, itf interface{}) string {
+func (p *cgoPackage) getUnnamedDeclName(prefix string, itf any) string {
 	if name, ok := p.anonDecls[itf]; ok {
 		return name
 	}
@@ -1,17 +0,0 @@
-//go:build go1.22
-
-package cgo
-
-// Code specifically for Go 1.22.
-
-import (
-	"go/ast"
-	"go/token"
-)
-
-func init() {
-	setASTFileFields = func(f *ast.File, start, end token.Pos) {
-		f.FileStart = start
-		f.FileEnd = end
-	}
-}
@@ -45,7 +45,6 @@ func TestCGo(t *testing.T) {
 		"flags",
 		"const",
 	} {
-		name := name // avoid a race condition
 		t.Run(name, func(t *testing.T) {
 			// Read the AST in memory.
 			path := filepath.Join("testdata", name+".go")
@@ -130,7 +130,7 @@ func (f *cgoFile) readNames(fragment string, cflags []string, filename string, c
 	// convert Go slice of strings to C array of strings.
 	cmdargsC := C.malloc(C.size_t(len(cflags)) * C.size_t(unsafe.Sizeof(uintptr(0))))
 	defer C.free(cmdargsC)
-	cmdargs := (*[1 << 16]*C.char)(cmdargsC)
+	cmdargs := unsafe.Slice((**C.char)(cmdargsC), len(cflags))
 	for i, cflag := range cflags {
 		s := C.CString(cflag)
 		cmdargs[i] = s
@@ -160,7 +160,7 @@ func (f *cgoFile) readNames(fragment string, cflags []string, filename string, c
 			pos := f.getClangLocationPosition(location, unit)
 			f.addError(pos, severity+": "+spelling)
 		}
-		for i := 0; i < numDiagnostics; i++ {
+		for i := range numDiagnostics {
 			diagnostic := C.clang_getDiagnostic(unit, C.uint(i))
 			addDiagnostic(diagnostic)

@@ -190,7 +190,7 @@ func (f *cgoFile) readNames(fragment string, cflags []string, filename string, c
 			// Sanity check. This should (hopefully) never trigger.
 			panic("libclang: file contents was not loaded")
 		}
-		data := (*[1 << 24]byte)(unsafe.Pointer(rawData))[:size]
+		data := unsafe.Slice((*byte)(unsafe.Pointer(rawData)), size)

 		// Hash the contents if it isn't hashed yet.
 		if _, ok := f.visitedFiles[path]; !ok {
@@ -217,10 +217,6 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 	case C.CXCursor_FunctionDecl:
 		cursorType := C.tinygo_clang_getCursorType(c)
 		numArgs := int(C.tinygo_clang_Cursor_getNumArguments(c))
-		obj := &ast.Object{
-			Kind: ast.Fun,
-			Name: "_Cgo_" + name,
-		}
 		exportName := name
 		localName := name
 		var stringSignature string
@@ -258,7 +254,6 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 			Name: &ast.Ident{
 				NamePos: pos,
 				Name:    "_Cgo_" + localName,
-				Obj:     obj,
 			},
 			Type: &ast.FuncType{
 				Func: pos,
@@ -283,7 +278,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 				Text:  strings.Join(doc, "\n"),
 			})
 		}
-		for i := 0; i < numArgs; i++ {
+		for i := range numArgs {
 			arg := C.tinygo_clang_Cursor_getArgument(c, C.uint(i))
 			argName := getString(C.tinygo_clang_getCursorSpelling(arg))
 			argType := C.clang_getArgType(cursorType, C.uint(i))
@@ -295,11 +290,6 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 					{
 						NamePos: pos,
 						Name:    argName,
-						Obj: &ast.Object{
-							Kind: ast.Var,
-							Name: argName,
-							Decl: decl,
-						},
 					},
 				},
 				Type: f.makeDecayingASTType(argType, pos),
@@ -315,7 +305,6 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 				},
 			}
 		}
-		obj.Decl = decl
 		return decl, stringSignature
 	case C.CXCursor_StructDecl, C.CXCursor_UnionDecl:
 		typ := f.makeASTRecordType(c, pos)
@@ -325,39 +314,25 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 			// Convert to a single-field struct type.
 			typeExpr = f.makeUnionField(typ)
 		}
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: typeName,
-		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
 				NamePos: typ.pos,
 				Name:    typeName,
-				Obj:     obj,
 			},
 			Type: typeExpr,
 		}
-		obj.Decl = typeSpec
 		return typeSpec, typ
 	case C.CXCursor_TypedefDecl:
 		typeName := "_Cgo_" + name
 		underlyingType := C.tinygo_clang_getTypedefDeclUnderlyingType(c)
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: typeName,
-		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
 				NamePos: pos,
 				Name:    typeName,
-				Obj:     obj,
 			},
-			Type: f.makeASTType(underlyingType, pos),
+			Assign: pos,
+			Type:   f.makeASTType(underlyingType, pos),
 		}
-		if underlyingType.kind != C.CXType_Enum {
-			typeSpec.Assign = pos
-		}
-		obj.Decl = typeSpec
 		return typeSpec, nil
 	case C.CXCursor_VarDecl:
 		cursorType := C.tinygo_clang_getCursorType(c)
@@ -376,19 +351,13 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 				},
 			},
 		}
-		obj := &ast.Object{
-			Kind: ast.Var,
-			Name: "_Cgo_" + name,
-		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{{
 				NamePos: pos,
 				Name:    "_Cgo_" + name,
-				Obj:     obj,
 			}},
 			Type: typeExpr,
 		}
-		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
 		return gen, nil
 	case C.CXCursor_MacroDefinition:
@@ -405,26 +374,16 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 			Lparen: token.NoPos,
 			Rparen: token.NoPos,
 		}
-		obj := &ast.Object{
-			Kind: ast.Con,
-			Name: "_Cgo_" + name,
-		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{{
 				NamePos: pos,
 				Name:    "_Cgo_" + name,
-				Obj:     obj,
 			}},
 			Values: []ast.Expr{expr},
 		}
-		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
 		return gen, nil
 	case C.CXCursor_EnumDecl:
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: "_Cgo_" + name,
-		}
 		underlying := C.tinygo_clang_getEnumDeclIntegerType(c)
 		// TODO: gc's CGo implementation uses types such as `uint32` for enums
 		// instead of types such as C.int, which are used here.
@@ -432,12 +391,10 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 			Name: &ast.Ident{
 				NamePos: pos,
 				Name:    "_Cgo_" + name,
-				Obj:     obj,
 			},
 			Assign: pos,
 			Type:   f.makeASTType(underlying, pos),
 		}
-		obj.Decl = typeSpec
 		return typeSpec, nil
 	case C.CXCursor_EnumConstantDecl:
 		value := C.tinygo_clang_getEnumConstantDeclValue(c)
@@ -452,19 +409,13 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 			Lparen: token.NoPos,
 			Rparen: token.NoPos,
 		}
-		obj := &ast.Object{
-			Kind: ast.Con,
-			Name: "_Cgo_" + name,
-		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{{
 				NamePos: pos,
 				Name:    "_Cgo_" + name,
-				Obj:     obj,
 			}},
 			Values: []ast.Expr{expr},
 		}
-		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
 		return gen, nil
 	default:
@@ -581,7 +532,7 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient

 // Get the precise location in the source code. Used for uniquely identifying
 // source locations.
-func (f *cgoFile) getUniqueLocationID(pos token.Pos, cursor C.GoCXCursor) interface{} {
+func (f *cgoFile) getUniqueLocationID(pos token.Pos, cursor C.GoCXCursor) any {
 	clangLocation := C.tinygo_clang_getCursorLocation(cursor)
 	var file C.CXFile
 	var line C.unsigned
@@ -624,7 +575,7 @@ func (p *cgoPackage) getClangLocationPosition(location C.CXSourceLocation, tu C.
 		// now by reading the file from libclang.
 		var size C.size_t
 		sourcePtr := C.clang_getFileContents(tu, file, &size)
-		source := ((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[:size:size]
+		source := unsafe.Slice((*byte)(unsafe.Pointer(sourcePtr)), size)
 		lines := []int{0}
 		for i := 0; i < len(source)-1; i++ {
 			if source[i] == '\n' {
@@ -639,7 +590,8 @@ func (p *cgoPackage) getClangLocationPosition(location C.CXSourceLocation, tu C.
 			Package: f.Pos(0),
 			Name:    ast.NewIdent(p.packageName),
 		}
-		setASTFileFields(astFile, f.Pos(0), f.Pos(int(size)))
+		astFile.FileStart = f.Pos(0)
+		astFile.FileEnd = f.Pos(int(size))
 		p.cgoFiles = append(p.cgoFiles, astFile)
 	}
 	positionFile := p.tokenFiles[filename]
@@ -852,6 +804,14 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 			f.addError(pos, fmt.Sprintf("unknown elaborated type (libclang type kind %s)", typeKindSpelling))
 			typeName = "<unknown>"
 		}
+	case C.CXType_Unexposed:
+		// LLVM 22+ may report certain builtin type aliases (e.g. __size_t)
+		// as Unexposed. Resolve via the canonical type.
+		canonical := C.clang_getCanonicalType(typ)
+		if canonical.kind != C.CXType_Unexposed && canonical.kind != C.CXType_Invalid {
+			return f.makeASTType(canonical, pos)
+		}
+		// If still unexposed, fall through to the error below.
 	case C.CXType_Record:
 		cursor := C.tinygo_clang_getTypeDeclaration(typ)
 		name := getString(C.tinygo_clang_getCursorSpelling(cursor))
@@ -956,22 +916,16 @@ func (p *cgoPackage) getIntegerType(name string, cursor clangCursor) *ast.TypeSp
 	}

 	// Construct an *ast.TypeSpec for this type.
-	obj := &ast.Object{
-		Kind: ast.Typ,
-		Name: name,
-	}
 	spec := &ast.TypeSpec{
 		Name: &ast.Ident{
 			NamePos: pos,
 			Name:    name,
-			Obj:     obj,
 		},
 		Type: &ast.Ident{
 			NamePos: pos,
 			Name:    goName,
 		},
 	}
-	obj.Decl = spec
 	return spec
 }

@@ -1104,7 +1058,6 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 				pos:      prevField.Names[0].NamePos,
 			})
 			prevField.Names[0].Name = bitfieldName
-			prevField.Names[0].Obj.Name = bitfieldName
 		}
 		prevBitfield := &(*bitfieldList)[len(*bitfieldList)-1]
 		prevBitfield.endBit = bitfieldOffset
@@ -1121,11 +1074,6 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 		{
 			NamePos: pos,
 			Name:    name,
-			Obj: &ast.Object{
-				Kind: ast.Var,
-				Name: name,
-				Decl: field,
-			},
 		},
 	}
 	fieldList.List = append(fieldList.List, field)
@@ -3,7 +3,7 @@
 package cgo

 /*
-#cgo linux        CFLAGS:  -I/usr/include/llvm-19 -I/usr/include/llvm-c-19 -I/usr/lib/llvm-19/include
+#cgo linux        CFLAGS:  -I/usr/include/llvm-19 -I/usr/include/llvm-c-19 -I/usr/lib/llvm-19/include -I/usr/lib64/llvm19/include
 #cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@19/include
 #cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@19/include
 #cgo freebsd      CFLAGS:  -I/usr/local/llvm19/include
@@ -1,9 +1,9 @@
-//go:build !byollvm && !llvm15 && !llvm16 && !llvm17 && !llvm18 && !llvm19
+//go:build !byollvm && !llvm15 && !llvm16 && !llvm17 && !llvm18 && !llvm19 && !llvm21 && !llvm22

 package cgo

 /*
-#cgo linux        CFLAGS:  -I/usr/include/llvm-20 -I/usr/include/llvm-c-20 -I/usr/lib/llvm-20/include
+#cgo linux        CFLAGS:  -I/usr/include/llvm-20 -I/usr/include/llvm-c-20 -I/usr/lib/llvm-20/include -I/usr/lib64/llvm20/include
 #cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@20/include
 #cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@20/include
 #cgo freebsd      CFLAGS:  -I/usr/local/llvm20/include
@@ -0,0 +1,15 @@
+//go:build !byollvm && llvm21
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/include/llvm-21 -I/usr/include/llvm-c-21 -I/usr/lib/llvm-21/include -I/usr/lib64/llvm21/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@21/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@21/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm21/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-21/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@21/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@21/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm21/lib -lclang
+*/
+import "C"
@@ -0,0 +1,15 @@
+//go:build !byollvm && llvm22
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/include/llvm-22 -I/usr/include/llvm-c-22 -I/usr/lib/llvm-22/include -I/usr/lib64/llvm22/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@22/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@22/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm22/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-22/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@22/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@22/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm22/lib -lclang
+*/
+import "C"
@@ -12,17 +12,17 @@ import "C"
 // C. It is useful if an API uses function pointers and you cannot pass a Go
 // pointer but only a C pointer.
 type refMap struct {
-	refs map[unsafe.Pointer]interface{}
+	refs map[unsafe.Pointer]any
 	lock sync.Mutex
 }

 // Put stores a value in the map. It can later be retrieved using Get. It must
 // be removed using Remove to avoid memory leaks.
-func (m *refMap) Put(v interface{}) unsafe.Pointer {
+func (m *refMap) Put(v any) unsafe.Pointer {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	if m.refs == nil {
-		m.refs = make(map[unsafe.Pointer]interface{}, 1)
+		m.refs = make(map[unsafe.Pointer]any, 1)
 	}
 	ref := C.malloc(1)
 	m.refs[ref] = v
@@ -31,7 +31,7 @@ func (m *refMap) Put(v interface{}) unsafe.Pointer {

 // Get returns a stored value previously inserted with Put. Use the same
 // reference as you got from Put.
-func (m *refMap) Get(ref unsafe.Pointer) interface{} {
+func (m *refMap) Get(ref unsafe.Pointer) any {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	return m.refs[ref]
@@ -8,6 +8,7 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"slices"
 	"strconv"
 	"strings"

@@ -60,13 +61,15 @@ func (c *Config) BuildMode() string {
 // RISC-V processor, that could be "+a,+c,+m". For many targets, an empty list
 // will be returned.
 func (c *Config) Features() string {
+	var features string
 	if c.Target.Features == "" {
-		return c.Options.LLVMFeatures
+		features = c.Options.LLVMFeatures
+	} else if c.Options.LLVMFeatures == "" {
+		features = c.Target.Features
+	} else {
+		features = c.Target.Features + "," + c.Options.LLVMFeatures
 	}
-	if c.Options.LLVMFeatures == "" {
-		return c.Target.Features
-	}
-	return c.Target.Features + "," + c.Options.LLVMFeatures
+	return patchFeatures(features)
 }

 // ABI returns the -mabi= flag for this target (like -mabi=lp64). A zero-length
@@ -140,13 +143,7 @@ func (c *Config) GC() string {
 func (c *Config) NeedsStackObjects() bool {
 	switch c.GC() {
 	case "conservative", "custom", "precise", "boehm":
-		for _, tag := range c.BuildTags() {
-			if tag == "tinygo.wasm" {
-				return true
-			}
-		}
-
-		return false
+		return slices.Contains(c.BuildTags(), "tinygo.wasm")
 	default:
 		return false
 	}
@@ -245,7 +242,7 @@ func (c *Config) RP2040BootPatch() bool {
 // Return a canonicalized architecture name, so we don't have to deal with arm*
 // vs thumb* vs arm64.
 func CanonicalArchName(triple string) string {
-	arch := strings.Split(triple, "-")[0]
+	arch, _, _ := strings.Cut(triple, "-")
 	if arch == "arm64" {
 		return "aarch64"
 	}
@@ -352,7 +349,7 @@ func (c *Config) CFlags(libclang bool) []string {
 	// Use the same optimization level as TinyGo.
 	cflags = append(cflags, "-O"+c.Options.Opt)
 	// Set the LLVM target triple.
-	cflags = append(cflags, "--target="+c.Triple())
+	cflags = append(cflags, "--target="+ClangTriple(c.Triple()))
 	// Set the -mcpu (or similar) flag.
 	if c.Target.CPU != "" {
 		if c.GOARCH() == "amd64" || c.GOARCH() == "386" {
@@ -466,6 +463,22 @@ func (c *Config) LDFlags() []string {
 	return ldflags
 }

+// LinkerFlavor returns how the configured linker should be driven.
+// Usually this is derived from GOOS, but targets may override it explicitly.
+func (c *Config) LinkerFlavor() string {
+	if c.Target.LinkerFlavor != "" {
+		return c.Target.LinkerFlavor
+	}
+	switch c.GOOS() {
+	case "windows":
+		return "coff"
+	case "darwin":
+		return "darwin"
+	default:
+		return "gnu"
+	}
+}
+
 // ExtraFiles returns the list of extra files to be built and linked with the
 // executable. This can include extra C and assembly files.
 func (c *Config) ExtraFiles() []string {
@@ -539,7 +552,7 @@ func (c *Config) Programmer() (method, openocdInterface string) {
 	case "openocd", "msd", "command", "adb":
 		// The -programmer flag only specifies the flash method.
 		return c.Options.Programmer, c.Target.OpenOCDInterface
-	case "bmp":
+	case "bmp", "probe-rs":
 		// The -programmer flag only specifies the flash method.
 		return c.Options.Programmer, ""
 	default:
@@ -0,0 +1,9 @@
+//go:build !llvm22 && !llvm14 && !llvm15 && !llvm16 && !llvm17 && !llvm18 && !llvm19
+
+package compileopts
+
+// patchFeatures applies LLVM-version-specific feature name mappings.
+// For LLVM 20/21, features in the target JSON files are already correct.
+func patchFeatures(features string) string {
+	return features
+}
@@ -0,0 +1,26 @@
+//go:build llvm22
+
+package compileopts
+
+import "strings"
+
+// patchFeatures applies LLVM-version-specific feature name mappings.
+// LLVM 22 renamed several Xtensa target features.
+func patchFeatures(features string) string {
+	// Xtensa feature renames in LLVM 22:
+	//   atomctl → (removed, no direct replacement)
+	//   memctl  → (removed, no direct replacement)
+	//   esp32s3 → esp32s3ops
+	//   timerint → timers3 (for esp32/esp32s3) or timers1 (for esp8266)
+	// Since we can't distinguish which timer variant at this level,
+	// just remove the obsolete features. The CPU definition already
+	// implies the correct features in LLVM 22.
+	replacer := strings.NewReplacer(
+		"+atomctl,", "",
+		"+memctl,", "",
+		"+esp32s3,", "+esp32s3ops,",
+		"+timerint,", "",
+		",+timerint", "",
+	)
+	return replacer.Replace(features)
+}
@@ -0,0 +1,16 @@
+//go:build llvm14 || llvm15 || llvm16 || llvm17 || llvm18 || llvm19
+
+package compileopts
+
+import "strings"
+
+// patchFeatures applies LLVM-version-specific feature name mappings.
+// LLVM 19 and earlier do not have +bulk-memory-opt or
+// +call-indirect-overlong for WebAssembly (added in LLVM 20).
+func patchFeatures(features string) string {
+	features = strings.ReplaceAll(features, ",+bulk-memory-opt", "")
+	features = strings.ReplaceAll(features, "+bulk-memory-opt,", "")
+	features = strings.ReplaceAll(features, ",+call-indirect-overlong", "")
+	features = strings.ReplaceAll(features, "+call-indirect-overlong,", "")
+	return features
+}
@@ -3,6 +3,7 @@ package compileopts
 import (
 	"fmt"
 	"regexp"
+	"slices"
 	"strings"
 	"time"
 )
@@ -21,51 +22,53 @@ var (
 // usually passed from the command line, but can also be passed in environment
 // variables for example.
 type Options struct {
-	GOOS            string // environment variable
-	GOARCH          string // environment variable
-	GOARM           string // environment variable (only used with GOARCH=arm)
-	GOMIPS          string // environment variable (only used with GOARCH=mips and GOARCH=mipsle)
-	Directory       string // working dir, leave it unset to use the current working dir
-	Target          string
-	BuildMode       string // -buildmode flag
-	Opt             string
-	GC              string
-	PanicStrategy   string
-	Scheduler       string
-	StackSize       uint64 // goroutine stack size (if none could be automatically determined)
-	Serial          string
-	Work            bool // -work flag to print temporary build directory
-	InterpTimeout   time.Duration
-	PrintIR         bool
-	DumpSSA         bool
-	VerifyIR        bool
-	SkipDWARF       bool
-	PrintCommands   func(cmd string, args ...string) `json:"-"`
-	Semaphore       chan struct{}                    `json:"-"` // -p flag controls cap
-	Debug           bool
-	Nobounds        bool
-	PrintSizes      string
-	PrintAllocs     *regexp.Regexp // regexp string
-	PrintStacks     bool
-	Tags            []string
-	GlobalValues    map[string]map[string]string // map[pkgpath]map[varname]value
-	TestConfig      TestConfig
-	Programmer      string
-	OpenOCDCommands []string
-	LLVMFeatures    string
-	Monitor         bool
-	BaudRate        int
-	Timeout         time.Duration
-	WITPackage      string // pass through to wasm-tools component embed invocation
-	WITWorld        string // pass through to wasm-tools component embed -w option
-	ExtLDFlags      []string
-	GoCompatibility bool // enable to check for Go version compatibility
+	GOOS                    string // environment variable
+	GOARCH                  string // environment variable
+	GOARM                   string // environment variable (only used with GOARCH=arm)
+	GOMIPS                  string // environment variable (only used with GOARCH=mips and GOARCH=mipsle)
+	Directory               string // working dir, leave it unset to use the current working dir
+	Target                  string
+	BuildMode               string // -buildmode flag
+	Opt                     string
+	GC                      string
+	PanicStrategy           string
+	Scheduler               string
+	StackSize               uint64 // goroutine stack size (if none could be automatically determined)
+	Serial                  string
+	Work                    bool // -work flag to print temporary build directory
+	InterpTimeout           time.Duration
+	InterpMaxLoopIterations int
+	PrintIR                 bool
+	DumpSSA                 bool
+	VerifyIR                bool
+	SkipDWARF               bool
+	PrintCommands           func(cmd string, args ...string) `json:"-"`
+	Semaphore               chan struct{}                    `json:"-"` // -p flag controls cap
+	Debug                   bool
+	Nobounds                bool
+	PrintSizes              string
+	PrintAllocs             *regexp.Regexp // regexp string
+	PrintAllocsCover        bool           // emit allocs in go coverage tool format
+	PrintStacks             bool
+	Tags                    []string
+	GlobalValues            map[string]map[string]string // map[pkgpath]map[varname]value
+	TestConfig              TestConfig
+	Programmer              string
+	OpenOCDCommands         []string
+	LLVMFeatures            string
+	Monitor                 bool
+	BaudRate                int
+	Timeout                 time.Duration
+	WITPackage              string // pass through to wasm-tools component embed invocation
+	WITWorld                string // pass through to wasm-tools component embed -w option
+	ExtLDFlags              []string
+	GoCompatibility         bool // enable to check for Go version compatibility
 }

 // Verify performs a validation on the given options, raising an error if options are not valid.
 func (o *Options) Verify() error {
 	if o.BuildMode != "" {
-		valid := isInArray(validBuildModeOptions, o.BuildMode)
+		valid := slices.Contains(validBuildModeOptions, o.BuildMode)
 		if !valid {
 			return fmt.Errorf(`invalid buildmode option '%s': valid values are %s`,
 				o.BuildMode,
@@ -73,7 +76,7 @@ func (o *Options) Verify() error {
 		}
 	}
 	if o.GC != "" {
-		valid := isInArray(validGCOptions, o.GC)
+		valid := slices.Contains(validGCOptions, o.GC)
 		if !valid {
 			return fmt.Errorf(`invalid gc option '%s': valid values are %s`,
 				o.GC,
@@ -82,7 +85,7 @@ func (o *Options) Verify() error {
 	}

 	if o.Scheduler != "" {
-		valid := isInArray(validSchedulerOptions, o.Scheduler)
+		valid := slices.Contains(validSchedulerOptions, o.Scheduler)
 		if !valid {
 			return fmt.Errorf(`invalid scheduler option '%s': valid values are %s`,
 				o.Scheduler,
@@ -91,7 +94,7 @@ func (o *Options) Verify() error {
 	}

 	if o.Serial != "" {
-		valid := isInArray(validSerialOptions, o.Serial)
+		valid := slices.Contains(validSerialOptions, o.Serial)
 		if !valid {
 			return fmt.Errorf(`invalid serial option '%s': valid values are %s`,
 				o.Serial,
@@ -100,7 +103,7 @@ func (o *Options) Verify() error {
 	}

 	if o.PrintSizes != "" {
-		valid := isInArray(validPrintSizeOptions, o.PrintSizes)
+		valid := slices.Contains(validPrintSizeOptions, o.PrintSizes)
 		if !valid {
 			return fmt.Errorf(`invalid size option '%s': valid values are %s`,
 				o.PrintSizes,
@@ -109,7 +112,7 @@ func (o *Options) Verify() error {
 	}

 	if o.PanicStrategy != "" {
-		valid := isInArray(validPanicStrategyOptions, o.PanicStrategy)
+		valid := slices.Contains(validPanicStrategyOptions, o.PanicStrategy)
 		if !valid {
 			return fmt.Errorf(`invalid panic option '%s': valid values are %s`,
 				o.PanicStrategy,
@@ -118,19 +121,10 @@ func (o *Options) Verify() error {
 	}

 	if o.Opt != "" {
-		if !isInArray(validOptOptions, o.Opt) {
+		if !slices.Contains(validOptOptions, o.Opt) {
 			return fmt.Errorf("invalid -opt=%s: valid values are %s", o.Opt, strings.Join(validOptOptions, ", "))
 		}
 	}

 	return nil
 }
-
-func isInArray(arr []string, item string) bool {
-	for _, i := range arr {
-		if i == item {
-			return true
-		}
-	}
-	return false
-}
@@ -38,7 +38,8 @@ type TargetSpec struct {
 	Scheduler        string   `json:"scheduler,omitempty"`
 	Serial           string   `json:"serial,omitempty"` // which serial output to use (uart, usb, none)
 	Linker           string   `json:"linker,omitempty"`
-	RTLib            string   `json:"rtlib,omitempty"` // compiler runtime library (libgcc, compiler-rt)
+	LinkerFlavor     string   `json:"linker-flavor,omitempty"` // how to drive the configured linker (for example: gnu, coff, darwin)
+	RTLib            string   `json:"rtlib,omitempty"`         // compiler runtime library (libgcc, compiler-rt)
 	Libc             string   `json:"libc,omitempty"`
 	AutoStackSize    *bool    `json:"automatic-stack-size,omitempty"` // Determine stack size automatically at compile time.
 	DefaultStackSize uint64   `json:"default-stack-size,omitempty"`   // Default stack size if the size couldn't be determined at compile time.
@@ -67,6 +68,7 @@ type TargetSpec struct {
 	ADBPreCommands   []string `json:"adb-pre-commands,omitempty"`
 	ADBPushRemote    string   `json:"adb-push-remote,omitempty"`
 	ADBPostCommands  []string `json:"adb-post-commands,omitempty"`
+	ProbeRSChip      string   `json:"probe-rs-chip,omitempty"`
 	CodeModel        string   `json:"code-model,omitempty"`
 	RelocationModel  string   `json:"relocation-model,omitempty"`
 	WITPackage       string   `json:"wit-package,omitempty"`
@@ -473,10 +475,12 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 				"-m", "i386pep",
 				"--image-base", "0x400000",
 			)
+			spec.RTLib = "compiler-rt"
 		case "arm64":
 			spec.LDFlags = append(spec.LDFlags,
 				"-m", "arm64pe",
 			)
+			spec.RTLib = "compiler-rt"
 		}
 		spec.LDFlags = append(spec.LDFlags,
 			"-Bdynamic",
@@ -484,6 +488,8 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 			"--no-insert-timestamp",
 			"--no-dynamicbase",
 		)
+		spec.ExtraFiles = append(spec.ExtraFiles,
+			"src/runtime/runtime_windows.c")
 	case "wasm", "wasip1", "wasip2":
 		return nil, fmt.Errorf("GOOS=%s but GOARCH is unset. Please set GOARCH to wasm", options.GOOS)
 	default:
@@ -112,3 +112,52 @@ func TestOverrideProperties(t *testing.T) {
 	}

 }
+
+func TestConfigLinkerFlavor(t *testing.T) {
+	tests := []struct {
+		name   string
+		target *TargetSpec
+		goos   string
+		want   string
+	}{
+		{
+			name:   "default gnu",
+			target: &TargetSpec{},
+			goos:   "linux",
+			want:   "gnu",
+		},
+		{
+			name:   "default coff",
+			target: &TargetSpec{},
+			goos:   "windows",
+			want:   "coff",
+		},
+		{
+			name:   "default darwin",
+			target: &TargetSpec{},
+			goos:   "darwin",
+			want:   "darwin",
+		},
+		{
+			name: "target override",
+			target: &TargetSpec{
+				LinkerFlavor: "coff",
+			},
+			goos: "linux",
+			want: "coff",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			tc.target.GOOS = tc.goos
+			config := &Config{
+				Options: &Options{},
+				Target:  tc.target,
+			}
+			if got := config.LinkerFlavor(); got != tc.want {
+				t.Fatalf("LinkerFlavor() = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
@@ -0,0 +1,12 @@
+//go:build llvm22
+
+package compileopts
+
+import "strings"
+
+// ClangTriple returns the target triple to pass to Clang's --target flag.
+// LLVM 22 deprecated the "wasm32-unknown-wasi" triple in favor of
+// "wasm32-unknown-wasip1", so we substitute it here to avoid warnings.
+func ClangTriple(triple string) string {
+	return strings.Replace(triple, "wasm32-unknown-wasi", "wasm32-unknown-wasip1", 1)
+}
@@ -0,0 +1,9 @@
+//go:build !llvm22
+
+package compileopts
+
+// ClangTriple returns the target triple to pass to Clang's --target flag.
+// For pre-LLVM 22, the triple is used as-is.
+func ClangTriple(triple string) string {
+	return triple
+}
@@ -241,24 +241,37 @@ func (b *builder) createRuntimeAssert(assert llvm.Value, blockPrefix, assertFunc
 		}
 	}

-	// Put the fault block at the end of the function and the next block at the
-	// current insert position.
-	faultBlock := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".throw")
+	faultBlock := b.getRuntimeAssertBlock(blockPrefix, assertFunc)
 	nextBlock := b.insertBasicBlock(blockPrefix + ".next")
 	b.currentBlockInfo.exit = nextBlock // adjust outgoing block for phi nodes

 	// Now branch to the out-of-bounds or the regular block.
 	b.CreateCondBr(assert, faultBlock, nextBlock)

-	// Fail: the assert triggered so panic.
-	b.SetInsertPointAtEnd(faultBlock)
-	b.createRuntimeCall(assertFunc, nil, "")
-	b.CreateUnreachable()
-
 	// Ok: assert didn't trigger so continue normally.
 	b.SetInsertPointAtEnd(nextBlock)
 }

+func (b *builder) getRuntimeAssertBlock(blockPrefix, assertFunc string) llvm.BasicBlock {
+	if b.runtimeAssertBlocks == nil {
+		b.runtimeAssertBlocks = make(map[string]llvm.BasicBlock)
+	}
+	if block := b.runtimeAssertBlocks[assertFunc]; !block.IsNil() {
+		return block
+	}
+	savedBlock := b.GetInsertBlock()
+	block := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".throw")
+	b.runtimeAssertBlocks[assertFunc] = block
+	b.SetInsertPointAtEnd(block)
+	if b.hasDeferFrame() {
+		b.createFaultCheckpoint()
+	}
+	b.createRuntimeCall(assertFunc, nil, "")
+	b.CreateUnreachable()
+	b.SetInsertPointAtEnd(savedBlock)
+	return block
+}
+
 // extendInteger extends the value to at least targetType using a zero or sign
 // extend. The resulting value is not truncated: it may still be bigger than
 // targetType.
@@ -35,6 +35,9 @@ const (

 	// Whether this is a readonly parameter (for example, a string pointer).
 	paramIsReadonly
+
+	// Whether this parameter is passed through backing storage.
+	paramIsIndirect
 )

 // createRuntimeCallCommon creates a runtime call. Use createRuntimeCall or
@@ -102,6 +105,18 @@ func (b *builder) createInvoke(fnType llvm.Type, fn llvm.Value, args []llvm.Valu
 // Expand an argument type to a list that can be used in a function call
 // parameter list.
 func (c *compilerContext) expandFormalParamType(t llvm.Type, name string, goType types.Type) []paramInfo {
+	if c.isIndirectAggregate(t) {
+		return []paramInfo{{
+			llvmType: c.dataPtrType,
+			name:     name,
+			elemSize: c.targetData.TypeAllocSize(t),
+			flags:    paramIsGoParam | paramIsReadonly | paramIsIndirect,
+		}}
+	}
+	return c.expandDirectFormalParamType(t, name, goType)
+}
+
+func (c *compilerContext) expandDirectFormalParamType(t llvm.Type, name string, goType types.Type) []paramInfo {
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
 		fieldInfos := c.flattenAggregateType(t, name, goType)
@@ -115,6 +130,38 @@ func (c *compilerContext) expandFormalParamType(t llvm.Type, name string, goType
 	return []paramInfo{c.getParamInfo(t, name, goType)}
 }

+func (c *compilerContext) storedParamType(t llvm.Type, exported bool) llvm.Type {
+	if c.isIndirectParam(t, exported) {
+		return c.dataPtrType
+	}
+	return t
+}
+
+func (c *compilerContext) isIndirectParam(t llvm.Type, exported bool) bool {
+	return !exported && c.isIndirectAggregate(t)
+}
+
+func (b *builder) appendStoredValueTypes(valueTypes []llvm.Type, values []ssa.Value, exported bool) []llvm.Type {
+	for _, value := range values {
+		valueTypes = append(valueTypes, b.storedParamType(b.getLLVMType(value.Type()), exported))
+	}
+	return valueTypes
+}
+
+func (b *builder) appendStoredParamTypes(valueTypes []llvm.Type, params []*types.Var, exported bool) []llvm.Type {
+	for _, param := range params {
+		valueTypes = append(valueTypes, b.storedParamType(b.getLLVMType(param.Type()), exported))
+	}
+	return valueTypes
+}
+
+func (b *builder) prependIndirectResult(sig *types.Signature, exported bool, params []llvm.Value, name string) []llvm.Value {
+	if resultType, indirect := b.hasIndirectResult(sig); !exported && indirect {
+		return append([]llvm.Value{b.createIndirectStorage(resultType, name)}, params...)
+	}
+	return params
+}
+
 // expandFormalParamOffsets returns a list of offsets from the start of an
 // object of type t after it would have been split up by expandFormalParam. This
 // is useful for debug information, where it is necessary to know the offset
@@ -30,17 +30,15 @@ func (b *builder) createMakeChan(expr *ssa.MakeChan) llvm.Value {
 // actual channel send operation during goroutine lowering.
 func (b *builder) createChanSend(instr *ssa.Send) {
 	ch := b.getValue(instr.Chan, getPos(instr))
-	chanValue := b.getValue(instr.X, getPos(instr))

 	// store value-to-send
 	valueType := b.getLLVMType(instr.X.Type())
 	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
-	var valueAlloca, valueAllocaSize llvm.Value
+	var storage valueStorage
 	if isZeroSize {
-		valueAlloca = llvm.ConstNull(b.dataPtrType)
+		storage.ptr = llvm.ConstNull(b.dataPtrType)
 	} else {
-		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
-		b.CreateStore(chanValue, valueAlloca)
+		storage = b.getValueStorage(instr.X, "chan.value")
 	}

 	// Allocate buffer for the channel operation.
@@ -48,15 +46,13 @@ func (b *builder) createChanSend(instr *ssa.Send) {
 	channelOpAlloca, channelOpAllocaSize := b.createTemporaryAlloca(channelOp, "chan.op")

 	// Do the send.
-	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAlloca, channelOpAlloca}, "")
+	b.createRuntimeInvoke("chanSend", []llvm.Value{ch, storage.ptr, channelOpAlloca}, "")

 	// End the lifetime of the allocas.
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
 	// https://bugs.llvm.org/show_bug.cgi?id=41742
 	b.emitLifetimeEnd(channelOpAlloca, channelOpAllocaSize)
-	if !isZeroSize {
-		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
-	}
+	b.endValueStorage(storage)
 }

 // createChanRecv emits a pseudo chan receive operation. It is lowered to the
@@ -66,42 +62,22 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
 	ch := b.getValue(unop.X, getPos(unop))

 	// Allocate memory to receive into.
-	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
-	var valueAlloca, valueAllocaSize llvm.Value
-	if isZeroSize {
-		valueAlloca = llvm.ConstNull(b.dataPtrType)
-	} else {
-		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
-	}
+	result := b.createRuntimeValueResult(valueType, unop.CommaOk, true, "chan")

 	// Allocate buffer for the channel operation.
 	channelOp := b.getLLVMRuntimeType("channelOp")
 	channelOpAlloca, channelOpAllocaSize := b.createTemporaryAlloca(channelOp, "chan.op")

 	// Do the receive.
-	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAlloca, channelOpAlloca}, "")
-	var received llvm.Value
-	if isZeroSize {
-		received = llvm.ConstNull(valueType)
-	} else {
-		received = b.CreateLoad(valueType, valueAlloca, "chan.received")
-		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
-	}
+	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, result.valuePtr, channelOpAlloca}, "")
+	received := result.finish(b, commaOk, "chan.received")
 	b.emitLifetimeEnd(channelOpAlloca, channelOpAllocaSize)
-
-	if unop.CommaOk {
-		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
-		tuple = b.CreateInsertValue(tuple, received, 0, "")
-		tuple = b.CreateInsertValue(tuple, commaOk, 1, "")
-		return tuple
-	} else {
-		return received
-	}
+	return received
 }

 // createChanClose closes the given channel.
 func (b *builder) createChanClose(ch llvm.Value) {
-	b.createRuntimeCall("chanClose", []llvm.Value{ch}, "")
+	b.createRuntimeInvoke("chanClose", []llvm.Value{ch}, "")
 }

 // createSelect emits all IR necessary for a select statements. That's a
@@ -170,9 +146,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		case types.SendOnly:
 			// Store this value in an alloca and put a pointer to this alloca
 			// in the send state.
-			sendValue := b.getValue(state.Send, state.Pos)
-			alloca := llvmutil.CreateEntryBlockAlloca(b.Builder, sendValue.Type(), "select.send.value")
-			b.CreateStore(sendValue, alloca)
+			alloca := b.getSelectSendStorage(state.Send)
 			selectState = b.CreateInsertValue(selectState, alloca, 1, "")
 		default:
 			panic("unreachable")
@@ -280,7 +254,17 @@ func (b *builder) getChanSelectResult(expr *ssa.Extract) llvm.Value {
 		// receive can proceed at a time) so we'll get that alloca, bitcast
 		// it to the correct type, and dereference it.
 		recvbuf := b.selectRecvBuf[expr.Tuple.(*ssa.Select)]
-		typ := b.getLLVMType(expr.Type())
-		return b.CreateLoad(typ, recvbuf, "")
+		return b.loadFromStorage(recvbuf, expr.Type(), "select.received")
 	}
 }
+
+func (b *builder) getSelectSendStorage(value ssa.Value) llvm.Value {
+	typ := b.getLLVMType(value.Type())
+	if b.isIndirectAggregate(typ) {
+		return b.getValuePointer(value)
+	}
+	llvmValue := b.getValue(value, getPos(value))
+	ptr := llvmutil.CreateEntryBlockAlloca(b.Builder, typ, "select.send.value")
+	b.CreateStore(llvmValue, ptr)
+	return ptr
+}
@@ -90,8 +90,10 @@ type compilerContext struct {
 	astComments      map[string]*ast.CommentGroup
 	embedGlobals     map[string][]*loader.EmbedFile
 	pkg              *types.Package
-	packageDir       string // directory for this package
+	loaderPkg        *loader.Package // current package being compiled (for AST access)
+	packageDir       string          // directory for this package
 	runtimePkg       *types.Package
+	localTypeNames   typeutil.Map // *types.Named (synthetic local from generic instantiation) -> string
 }

 // newCompilerContext returns a new compiler context ready for use, most
@@ -153,6 +155,8 @@ type builder struct {
 	llvmFn            llvm.Value
 	info              functionInfo
 	locals            map[ssa.Value]llvm.Value // local variables
+	indirectValues    map[ssa.Value]llvm.Value
+	indirectReturn    llvm.Value
 	blockInfo         []blockInfo
 	currentBlock      *ssa.BasicBlock
 	currentBlockInfo  *blockInfo
@@ -167,7 +171,7 @@ type builder struct {
 	dilocals          map[*types.Var]llvm.Metadata
 	initInlinedAt     llvm.Metadata            // fake inlinedAt position
 	initPseudoFuncs   map[string]llvm.Metadata // fake "inlined" functions for proper init debug locations
-	allDeferFuncs     []interface{}
+	allDeferFuncs     []any
 	deferFuncs        map[*ssa.Function]int
 	deferInvokeFuncs  map[string]int
 	deferClosureFuncs map[*ssa.Function]int
@@ -176,6 +180,9 @@ type builder struct {
 	deferBuiltinFuncs map[ssa.Value]deferBuiltin
 	runDefersBlock    []llvm.BasicBlock
 	afterDefersBlock  []llvm.BasicBlock
+
+	runtimeAssertBlocks  map[string]llvm.BasicBlock
+	interfaceAssertBlock llvm.BasicBlock
 }

 func newBuilder(c *compilerContext, irbuilder llvm.Builder, f *ssa.Function) *builder {
@@ -188,6 +195,7 @@ func newBuilder(c *compilerContext, irbuilder llvm.Builder, f *ssa.Function) *bu
 		llvmFn:          fn,
 		info:            c.getFunctionInfo(f),
 		locals:          make(map[ssa.Value]llvm.Value),
+		indirectValues:  make(map[ssa.Value]llvm.Value),
 		dilocals:        make(map[*types.Var]llvm.Metadata),
 	}
 }
@@ -294,12 +302,18 @@ func CompilePackage(moduleName string, pkg *loader.Package, ssaPkg *ssa.Package,
 	c.packageDir = pkg.OriginalDir()
 	c.embedGlobals = pkg.EmbedGlobals
 	c.pkg = pkg.Pkg
+	c.loaderPkg = pkg
 	c.runtimePkg = ssaPkg.Prog.ImportedPackage("runtime").Pkg
 	c.program = ssaPkg.Prog

 	// Convert AST to SSA.
 	ssaPkg.Build()

+	// Assign names to function-local named types before compiling the
+	// package, so that types declared in different functions (or in
+	// different instantiations of a generic function) do not collide.
+	c.scanLocalTypes(ssaPkg)
+
 	// Initialize debug information.
 	if c.Debug {
 		c.cu = c.dibuilder.CreateCompileUnit(llvm.DICompileUnit{
@@ -314,9 +328,6 @@ func CompilePackage(moduleName string, pkg *loader.Package, ssaPkg *ssa.Package,
 	// Load comments such as //go:extern on globals.
 	c.loadASTComments(pkg)

-	// Predeclare the runtime.alloc function, which is used by the wordpack
-	// functionality.
-	c.getFunction(c.program.ImportedPackage("runtime").Members["alloc"].(*ssa.Function))
 	if c.NeedsStackObjects {
 		// Predeclare trackPointer, which is used everywhere we use runtime.alloc.
 		c.getFunction(c.program.ImportedPackage("runtime").Members["trackPointer"].(*ssa.Function))
@@ -414,19 +425,19 @@ func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 			return c.ctx.Int8Type()
 		case types.Int16, types.Uint16:
 			return c.ctx.Int16Type()
-		case types.Int32, types.Uint32:
+		case types.Int32, types.Uint32, types.UntypedRune:
 			return c.ctx.Int32Type()
-		case types.Int, types.Uint:
+		case types.Int, types.Uint, types.UntypedInt:
 			return c.intType
 		case types.Int64, types.Uint64:
 			return c.ctx.Int64Type()
 		case types.Float32:
 			return c.ctx.FloatType()
-		case types.Float64:
+		case types.Float64, types.UntypedFloat:
 			return c.ctx.DoubleType()
 		case types.Complex64:
 			return c.ctx.StructType([]llvm.Type{c.ctx.FloatType(), c.ctx.FloatType()}, false)
-		case types.Complex128:
+		case types.Complex128, types.UntypedComplex:
 			return c.ctx.StructType([]llvm.Type{c.ctx.DoubleType(), c.ctx.DoubleType()}, false)
 		case types.String, types.UntypedString:
 			return c.getLLVMRuntimeType("_string")
@@ -862,10 +873,9 @@ func (c *compilerContext) createPackage(irbuilder llvm.Builder, pkg *ssa.Package
 			}
 			// Create the function definition.
 			b := newBuilder(c, irbuilder, member)
-			if _, ok := mathToLLVMMapping[member.RelString(nil)]; ok {
+			if ok := b.defineMathOp(); ok {
 				// The body of this function (if there is one) is ignored and
 				// replaced with a LLVM intrinsic call.
-				b.defineMathOp()
 				continue
 			}
 			if ok := b.defineMathBitsIntrinsic(); ok {
@@ -1275,10 +1285,28 @@ func (b *builder) createFunctionStart(intrinsic bool) {

 	// Load function parameters
 	llvmParamIndex := 0
+	if _, indirectResult := b.hasIndirectResult(b.fn.Signature); indirectResult && !b.info.exported {
+		b.indirectReturn = b.llvmFn.Param(llvmParamIndex)
+		b.indirectReturn.SetName("return")
+		llvmParamIndex++
+	}
 	for _, param := range b.fn.Params {
 		llvmType := b.getLLVMType(param.Type())
+		if b.isIndirectParam(llvmType, b.info.exported) {
+			llvmParam := b.llvmFn.Param(llvmParamIndex)
+			llvmParam.SetName(param.Name())
+			b.indirectValues[param] = llvmParam
+			llvmParamIndex++
+			continue
+		}
+		var paramInfos []paramInfo
+		if b.info.exported {
+			paramInfos = b.expandDirectFormalParamType(llvmType, param.Name(), param.Type())
+		} else {
+			paramInfos = b.expandFormalParamType(llvmType, param.Name(), param.Type())
+		}
 		fields := make([]llvm.Value, 0, 1)
-		for _, info := range b.expandFormalParamType(llvmType, param.Name(), param.Type()) {
+		for _, info := range paramInfos {
 			param := b.llvmFn.Param(llvmParamIndex)
 			param.SetName(info.name)
 			fields = append(fields, param)
@@ -1416,7 +1444,7 @@ func (b *builder) createFunction() {
 	for _, phi := range b.phis {
 		block := phi.ssa.Block()
 		for i, edge := range phi.ssa.Edges {
-			llvmVal := b.getValue(edge, getPos(phi.ssa))
+			llvmVal := b.getCallArgument(edge, false)
 			llvmBlock := b.blockInfo[block.Preds[i].Index].exit
 			phi.llvm.AddIncoming([]llvm.Value{llvmVal}, []llvm.BasicBlock{llvmBlock})
 		}
@@ -1425,6 +1453,9 @@ func (b *builder) createFunction() {
 	if b.NeedsStackObjects {
 		// Track phi nodes.
 		for _, phi := range b.phis {
+			if b.isOversizedAggregate(phi.ssa.Type()) {
+				continue
+			}
 			insertPoint := llvm.NextInstruction(phi.llvm)
 			for !insertPoint.IsAPHINode().IsNil() {
 				insertPoint = llvm.NextInstruction(insertPoint)
@@ -1516,10 +1547,7 @@ func (b *builder) createInstruction(instr ssa.Instruction) {
 			b.diagnostics = append(b.diagnostics, err)
 			b.locals[instr] = llvm.Undef(b.getLLVMType(instr.Type()))
 		} else {
-			b.locals[instr] = value
-			if len(*instr.Referrers()) != 0 && b.NeedsStackObjects {
-				b.trackExpr(instr, value)
-			}
+			b.setValue(instr, value)
 		}
 	case *ssa.DebugRef:
 		// ignore
@@ -1539,10 +1567,8 @@ func (b *builder) createInstruction(instr ssa.Instruction) {
 		b.CreateBr(blockJump)
 	case *ssa.MapUpdate:
 		m := b.getValue(instr.Map, getPos(instr))
-		key := b.getValue(instr.Key, getPos(instr))
-		value := b.getValue(instr.Value, getPos(instr))
 		mapType := instr.Map.Type().Underlying().(*types.Map)
-		b.createMapUpdate(mapType.Key(), m, key, value, instr.Pos())
+		b.createMapUpdate(mapType.Key(), m, instr.Key, instr.Value, instr.Pos())
 	case *ssa.Panic:
 		value := b.getValue(instr.X, getPos(instr))
 		b.createRuntimeInvoke("_panic", []llvm.Value{value}, "")
@@ -1551,19 +1577,7 @@ func (b *builder) createInstruction(instr ssa.Instruction) {
 		if b.hasDeferFrame() {
 			b.createRuntimeCall("destroyDeferFrame", []llvm.Value{b.deferFrame}, "")
 		}
-		if len(instr.Results) == 0 {
-			b.CreateRetVoid()
-		} else if len(instr.Results) == 1 {
-			b.CreateRet(b.getValue(instr.Results[0], getPos(instr)))
-		} else {
-			// Multiple return values. Put them all in a struct.
-			retVal := llvm.ConstNull(b.llvmFn.GlobalValueType().ReturnType())
-			for i, result := range instr.Results {
-				val := b.getValue(result, getPos(instr))
-				retVal = b.CreateInsertValue(retVal, val, i, "")
-			}
-			b.CreateRet(retVal)
-		}
+		b.createReturn(instr.Results, getPos(instr))
 	case *ssa.RunDefers:
 		// Note where we're going to put the rundefers block
 		run := b.insertBasicBlock("rundefers.block")
@@ -1577,18 +1591,246 @@ func (b *builder) createInstruction(instr ssa.Instruction) {
 		b.createChanSend(instr)
 	case *ssa.Store:
 		llvmAddr := b.getValue(instr.Addr, getPos(instr))
-		llvmVal := b.getValue(instr.Val, getPos(instr))
 		b.createNilCheck(instr.Addr, llvmAddr, "store")
-		if b.targetData.TypeAllocSize(llvmVal.Type()) == 0 {
+		llvmType := b.getLLVMType(instr.Val.Type())
+		if b.targetData.TypeAllocSize(llvmType) == 0 {
 			// nothing to store
 			return
 		}
-		b.CreateStore(llvmVal, llvmAddr)
+		b.storeValue(llvmAddr, instr.Val)
 	default:
 		b.addError(instr.Pos(), "unknown instruction: "+instr.String())
 	}
 }

+func (b *builder) setValue(value ssa.Value, llvmValue llvm.Value) {
+	if b.isAggregateValue(value.Type()) && !llvmValue.IsNil() && llvmValue.Type().TypeKind() == llvm.PointerTypeKind {
+		b.indirectValues[value] = llvmValue
+		return
+	}
+	b.locals[value] = llvmValue
+	if len(*value.Referrers()) != 0 && b.NeedsStackObjects {
+		b.trackExpr(value, llvmValue)
+	}
+}
+
+func (b *builder) createReturn(results []ssa.Value, pos token.Pos) {
+	if len(results) == 0 {
+		b.CreateRetVoid()
+	} else if !b.indirectReturn.IsNil() {
+		if len(results) == 1 {
+			b.storeValue(b.indirectReturn, results[0])
+		} else {
+			returnType := b.getLLVMResultType(b.fn.Signature)
+			for i, result := range results {
+				fieldPtr := b.CreateStructGEP(returnType, b.indirectReturn, i, "")
+				b.storeValue(fieldPtr, result)
+			}
+		}
+		b.CreateRetVoid()
+	} else if len(results) == 1 {
+		b.CreateRet(b.getValue(results[0], pos))
+	} else {
+		result := llvm.ConstNull(b.llvmFn.GlobalValueType().ReturnType())
+		for i, value := range results {
+			result = b.CreateInsertValue(result, b.getValue(value, pos), i, "")
+		}
+		b.CreateRet(result)
+	}
+}
+
+func (b *builder) isOversizedAggregate(typ types.Type) bool {
+	if !b.isAggregateValue(typ) {
+		return false
+	}
+	return b.isIndirectAggregate(b.getLLVMType(typ))
+}
+
+func (b *builder) isAggregateValue(typ types.Type) bool {
+	if tuple, ok := typ.(*types.Tuple); ok {
+		for i := 0; i < tuple.Len(); i++ {
+			if !isLLVMValueType(tuple.At(i).Type()) {
+				return false
+			}
+		}
+	} else {
+		switch typ.Underlying().(type) {
+		case *types.Array, *types.Struct:
+		default:
+			return false
+		}
+		if !isLLVMValueType(typ) {
+			return false
+		}
+	}
+	return true
+}
+
+func (b *builder) getValuePointer(value ssa.Value) llvm.Value {
+	if ptr, ok := b.indirectValues[value]; ok {
+		return ptr
+	}
+	llvmType := b.getLLVMType(value.Type())
+	ptr := b.createIndirectStorage(llvmType, value.Name())
+	b.storeValue(ptr, value)
+	return ptr
+}
+
+func (b *builder) getCallArgument(value ssa.Value, exported bool) llvm.Value {
+	paramType := b.getLLVMType(value.Type())
+	if b.isIndirectParam(paramType, exported) {
+		return b.getValuePointer(value)
+	}
+	return b.getValue(value, getPos(value))
+}
+
+func (b *builder) createIndirectStorage(typ llvm.Type, name string) llvm.Value {
+	// Use runtime.alloc here so storage that escapes remains valid. The
+	// allocation optimizer moves bounded non-escaping storage to the stack.
+	size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(typ), false)
+	layout := b.createObjectLayout(typ, b.fn.Pos())
+	ptr := b.createAlloc(size, layout, b.targetData.ABITypeAlignment(typ), name)
+	if b.NeedsStackObjects {
+		b.trackPointer(ptr)
+	}
+	return ptr
+}
+
+func (b *builder) copyIndirectAggregate(dst, src llvm.Value, typ llvm.Type) {
+	size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(typ), false)
+	b.createMemCopy("memcpy", dst, src, size)
+}
+
+func (b *builder) storeValue(dst llvm.Value, value ssa.Value) {
+	typ := b.getLLVMType(value.Type())
+	if src, ok := b.indirectValues[value]; ok {
+		b.copyIndirectAggregate(dst, src, typ)
+	} else {
+		b.CreateStore(b.getValue(value, getPos(value)), dst)
+	}
+}
+
+func (b *builder) copyToIndirectStorage(src llvm.Value, typ llvm.Type, name string) llvm.Value {
+	dst := b.createIndirectStorage(typ, name)
+	b.copyIndirectAggregate(dst, src, typ)
+	return dst
+}
+
+func (b *builder) loadFromStorage(ptr llvm.Value, typ types.Type, name string) llvm.Value {
+	llvmType := b.getLLVMType(typ)
+	if b.isIndirectAggregate(llvmType) {
+		return b.copyToIndirectStorage(ptr, llvmType, name)
+	}
+	return b.CreateLoad(llvmType, ptr, name)
+}
+
+func (b *builder) getValueField(value ssa.Value, index int, resultType types.Type, name string) (llvm.Value, bool) {
+	if !b.isAggregateValue(value.Type()) {
+		return llvm.Value{}, false
+	}
+	valueType := b.getLLVMType(value.Type())
+	if _, indirect := b.indirectValues[value]; !indirect && !b.isIndirectAggregate(valueType) {
+		return llvm.Value{}, false
+	}
+	fieldPtr := b.CreateStructGEP(valueType, b.getValuePointer(value), index, "")
+	return b.loadFromStorage(fieldPtr, resultType, name), true
+}
+
+func (b *builder) zeroIndirectStorage(ptr llvm.Value, typ llvm.Type) {
+	memset := b.getMemsetFunc()
+	b.createCall(memset.GlobalValueType(), memset, []llvm.Value{
+		ptr,
+		llvm.ConstInt(b.ctx.Int8Type(), 0, false),
+		llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(typ), false),
+		llvm.ConstInt(b.ctx.Int1Type(), 0, false),
+	}, "")
+}
+
+type valueStorage struct {
+	ptr, size llvm.Value
+	temporary bool
+}
+
+func (b *builder) getValueStorage(value ssa.Value, name string) valueStorage {
+	typ := b.getLLVMType(value.Type())
+	if b.isIndirectAggregate(typ) {
+		return valueStorage{ptr: b.getValuePointer(value)}
+	}
+	ptr, size := b.createTemporaryAlloca(typ, name)
+	b.storeValue(ptr, value)
+	return valueStorage{ptr: ptr, size: size, temporary: true}
+}
+
+func (b *builder) endValueStorage(storage valueStorage) {
+	if storage.temporary {
+		b.emitLifetimeEnd(storage.ptr, storage.size)
+	}
+}
+
+type runtimeValueResult struct {
+	valueType  llvm.Type
+	resultType llvm.Type
+	result     llvm.Value
+	valuePtr   llvm.Value
+	valueSize  llvm.Value
+	temporary  bool
+	zero       bool
+	commaOk    bool
+}
+
+func (b *builder) createRuntimeValueResult(valueType llvm.Type, commaOk, zeroAsNull bool, name string) runtimeValueResult {
+	result := runtimeValueResult{
+		valueType:  valueType,
+		resultType: valueType,
+		valueSize:  llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(valueType), false),
+		commaOk:    commaOk,
+	}
+	if commaOk {
+		result.resultType = b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false)
+	}
+	if b.isIndirectAggregate(result.resultType) {
+		result.result = b.createIndirectStorage(result.resultType, name+".result")
+		result.valuePtr = result.result
+		if commaOk {
+			result.valuePtr = b.CreateStructGEP(result.resultType, result.result, 0, "")
+		}
+		return result
+	}
+	if zeroAsNull && b.targetData.TypeAllocSize(valueType) == 0 {
+		result.valuePtr = llvm.ConstNull(b.dataPtrType)
+		result.zero = true
+		return result
+	}
+	result.valuePtr, result.valueSize = b.createTemporaryAlloca(valueType, name+".value")
+	result.temporary = true
+	return result
+}
+
+func (r runtimeValueResult) finish(b *builder, commaOk llvm.Value, name string) llvm.Value {
+	if !r.result.IsNil() {
+		if r.commaOk {
+			b.CreateStore(commaOk, b.CreateStructGEP(r.resultType, r.result, 1, ""))
+		}
+		return r.result
+	}
+
+	var value llvm.Value
+	if r.zero {
+		value = llvm.ConstNull(r.valueType)
+	} else {
+		value = b.CreateLoad(r.valueType, r.valuePtr, name)
+	}
+	if r.temporary {
+		b.emitLifetimeEnd(r.valuePtr, r.valueSize)
+	}
+	if !r.commaOk {
+		return value
+	}
+	result := llvm.Undef(r.resultType)
+	result = b.CreateInsertValue(result, value, 0, "")
+	return b.CreateInsertValue(result, commaOk, 1, "")
+}
+
 // createBuiltin lowers a builtin Go function (append, close, delete, etc.) to
 // LLVM IR. It uses runtime calls for some builtins.
 func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, callName string, pos token.Pos) (llvm.Value, error) {
@@ -1888,6 +2130,12 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 			// not of the current function.
 			useParentFrame = 1
 		}
+		// Prevent inlining of functions that call recover(), matching the
+		// Go compiler's behavior. If this function were inlined into a
+		// deferred function, recover() would incorrectly succeed because
+		// the inlined code runs in the deferred function's context.
+		noinline := b.ctx.CreateEnumAttribute(llvm.AttributeKindID("noinline"), 0)
+		b.llvmFn.AddFunctionAttr(noinline)
 		return b.createRuntimeCall("_recover", []llvm.Value{llvm.ConstInt(b.ctx.Int1Type(), useParentFrame, false)}, ""), nil
 	case "ssa:wrapnilchk":
 		// TODO: do an actual nil check?
@@ -1968,7 +2216,7 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 	if fn := instr.StaticCallee(); fn != nil {
 		// Direct function call, either to a named or anonymous (directly
 		// applied) function call. If it is anonymous, it may be a closure.
-		name := fn.RelString(nil)
+		name := b.getFunctionInfo(fn).linkName
 		switch {
 		case name == "device.Asm" || name == "device/arm.Asm" || name == "device/arm64.Asm" || name == "device/avr.Asm" || name == "device/riscv.Asm":
 			return b.createInlineAsm(instr.Args)
@@ -2017,14 +2265,10 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		}
 	}

-	var params []llvm.Value
-	for _, param := range instr.Args {
-		params = append(params, b.getValue(param, getPos(instr)))
-	}
-
 	// Try to call the function directly for trivially static calls.
 	var callee, context llvm.Value
 	var calleeType llvm.Type
+	var invokeTypecode, invokeReceiver llvm.Value
 	exported := false
 	if fn := instr.StaticCallee(); fn != nil {
 		calleeType, callee = b.getFunction(fn)
@@ -2054,19 +2298,18 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		exported = info.exported
 	} else if call, ok := instr.Value.(*ssa.Builtin); ok {
 		// Builtin function (append, close, delete, etc.).)
+		var params []llvm.Value
 		var argTypes []types.Type
 		for _, arg := range instr.Args {
 			argTypes = append(argTypes, arg.Type())
+			params = append(params, b.getValue(arg, getPos(instr)))
 		}
 		return b.createBuiltin(argTypes, params, call.Name(), instr.Pos())
 	} else if instr.IsInvoke() {
 		// Interface method call (aka invoke call).
 		itf := b.getValue(instr.Value, getPos(instr)) // interface value (runtime._interface)
-		typecode := b.CreateExtractValue(itf, 0, "invoke.func.typecode")
-		value := b.CreateExtractValue(itf, 1, "invoke.func.value") // receiver
-		// Prefix the params with receiver value and suffix with typecode.
-		params = append([]llvm.Value{value}, params...)
-		params = append(params, typecode)
+		invokeTypecode = b.CreateExtractValue(itf, 0, "invoke.func.typecode")
+		invokeReceiver = b.CreateExtractValue(itf, 1, "invoke.func.value")
 		callee = b.getInvokeFunction(instr)
 		calleeType = callee.GlobalValueType()
 		context = llvm.Undef(b.dataPtrType)
@@ -2080,7 +2323,23 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		b.createNilCheck(instr.Value, callee, "fpcall")
 	}

+	var params []llvm.Value
+	for _, param := range instr.Args {
+		params = append(params, b.getCallArgument(param, exported))
+	}
+	if instr.IsInvoke() {
+		params = append([]llvm.Value{invokeReceiver}, params...)
+		params = append(params, invokeTypecode)
+	}
+
 	if !exported {
+		if resultType, indirectResult := b.hasIndirectResult(instr.Signature()); indirectResult {
+			result := b.createIndirectStorage(resultType, "call.result")
+			params = append([]llvm.Value{result}, params...)
+			params = append(params, context)
+			b.createInvoke(calleeType, callee, params, "")
+			return result, nil
+		}
 		// This function takes a context parameter.
 		// Add it to the end of the parameter list.
 		params = append(params, context)
@@ -2119,6 +2378,9 @@ func (b *builder) getValue(expr ssa.Value, pos token.Pos) llvm.Value {
 		return value
 	default:
 		// other (local) SSA value
+		if value, ok := b.indirectValues[expr]; ok {
+			return b.CreateLoad(b.getLLVMType(expr.Type()), value, "")
+		}
 		if value, ok := b.locals[expr]; ok {
 			return value
 		} else {
@@ -2143,13 +2405,10 @@ func (c *compilerContext) maxSliceSize(elementType llvm.Type) uint64 {
 	if elementSize == 0 {
 		elementSize = 1
 	}
-	maxSize := maxPointerValue / elementSize
-
-	// len(slice) is an int. Make sure the length remains small enough to fit in
-	// an int.
-	if maxSize > maxIntegerValue {
-		maxSize = maxIntegerValue
-	}
+	maxSize := min(
+		// len(slice) is an int. Make sure the length remains small enough to fit in
+		// an int.
+		maxPointerValue/elementSize, maxIntegerValue)

 	return maxSize
 }
@@ -2176,9 +2435,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			}
 			sizeValue := llvm.ConstInt(b.uintptrType, size, false)
 			layoutValue := b.createObjectLayout(typ, expr.Pos())
-			buf := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, layoutValue}, expr.Comment)
 			align := b.targetData.ABITypeAlignment(typ)
-			buf.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align)))
+			buf := b.createAlloc(sizeValue, layoutValue, align, expr.Comment)
 			return buf, nil
 		} else {
 			buf := llvmutil.CreateEntryBlockAlloca(b.Builder, typ, expr.Comment)
@@ -2205,8 +2463,15 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		// This instruction changes the type, but the underlying value remains
 		// the same. This is often a no-op, but sometimes we have to change the
 		// LLVM type as well.
-		x := b.getValue(expr.X, getPos(expr))
 		llvmType := b.getLLVMType(expr.Type())
+		if b.isIndirectAggregate(llvmType) {
+			sourceType := b.getLLVMType(expr.X.Type())
+			if !b.isIndirectAggregate(sourceType) || b.targetData.TypeAllocSize(sourceType) != b.targetData.TypeAllocSize(llvmType) {
+				return llvm.Value{}, errors.New("todo: indirect aggregate ChangeType with different layout")
+			}
+			return b.getValuePointer(expr.X), nil
+		}
+		x := b.getValue(expr.X, getPos(expr))
 		if x.Type() == llvmType {
 			// Different Go type but same LLVM type (for example, named int).
 			// This is the common case.
@@ -2236,9 +2501,15 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		if _, ok := expr.Tuple.(*ssa.Select); ok {
 			return b.getChanSelectResult(expr), nil
 		}
+		if value, ok := b.getValueField(expr.Tuple, expr.Index, expr.Type(), expr.Name()); ok {
+			return value, nil
+		}
 		value := b.getValue(expr.Tuple, getPos(expr))
 		return b.CreateExtractValue(value, expr.Index, ""), nil
 	case *ssa.Field:
+		if value, ok := b.getValueField(expr.X, expr.Field, expr.Type(), expr.Name()); ok {
+			return value, nil
+		}
 		value := b.getValue(expr.X, getPos(expr))
 		result := b.CreateExtractValue(value, expr.Field, "")
 		return result, nil
@@ -2261,11 +2532,11 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 	case *ssa.Global:
 		panic("global is not an expression")
 	case *ssa.Index:
-		collection := b.getValue(expr.X, getPos(expr))
 		index := b.getValue(expr.Index, getPos(expr))

 		switch xType := expr.X.Type().Underlying().(type) {
 		case *types.Basic: // extract byte from string
+			collection := b.getValue(expr.X, getPos(expr))
 			// Value type must be a string, which is a basic type.
 			if xType.Info()&types.IsString == 0 {
 				panic("lookup on non-string?")
@@ -2299,13 +2570,12 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {

 			// Can't load directly from array (as index is non-constant), so
 			// have to do it using an alloca+gep+load.
-			arrayType := collection.Type()
-			alloca, allocaSize := b.createTemporaryAlloca(arrayType, "index.alloca")
-			b.CreateStore(collection, alloca)
+			arrayType := b.getLLVMType(expr.X.Type())
+			storage := b.getValueStorage(expr.X, "index.alloca")
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
-			ptr := b.CreateInBoundsGEP(arrayType, alloca, []llvm.Value{zero, index}, "index.gep")
-			result := b.CreateLoad(arrayType.ElementType(), ptr, "index.load")
-			b.emitLifetimeEnd(alloca, allocaSize)
+			ptr := b.CreateInBoundsGEP(arrayType, storage.ptr, []llvm.Value{zero, index}, "index.gep")
+			result := b.loadFromStorage(ptr, expr.Type(), "index.load")
+			b.endValueStorage(storage)
 			return result, nil
 		default:
 			panic("unknown *ssa.Index type")
@@ -2364,17 +2634,21 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		}
 	case *ssa.Lookup: // map lookup
 		value := b.getValue(expr.X, getPos(expr))
-		index := b.getValue(expr.Index, getPos(expr))
 		valueType := expr.Type()
 		if expr.CommaOk {
 			valueType = valueType.(*types.Tuple).At(0).Type()
 		}
-		return b.createMapLookup(expr.X.Type().Underlying().(*types.Map).Key(), valueType, value, index, expr.CommaOk, expr.Pos())
+		return b.createMapLookup(expr.X.Type().Underlying().(*types.Map).Key(), valueType, value, expr.Index, expr.CommaOk, expr.Pos())
 	case *ssa.MakeChan:
 		return b.createMakeChan(expr), nil
 	case *ssa.MakeClosure:
 		return b.parseMakeClosure(expr)
 	case *ssa.MakeInterface:
+		if b.isOversizedAggregate(expr.X.Type()) {
+			typ := b.getLLVMType(expr.X.Type())
+			ptr := b.copyToIndirectStorage(b.getValuePointer(expr.X), typ, "interface.value")
+			return b.createMakeInterfaceFromPointer(ptr, expr.X.Type()), nil
+		}
 		val := b.getValue(expr.X, getPos(expr))
 		return b.createMakeInterface(val, expr.X.Type(), expr.Pos()), nil
 	case *ssa.MakeMap:
@@ -2408,8 +2682,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		}
 		sliceSize := b.CreateBinOp(llvm.Mul, elemSizeValue, sliceCapCast, "makeslice.cap")
 		layoutValue := b.createObjectLayout(llvmElemType, expr.Pos())
-		slicePtr := b.createRuntimeCall("alloc", []llvm.Value{sliceSize, layoutValue}, "makeslice.buf")
-		slicePtr.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(elemAlign)))
+		slicePtr := b.createAlloc(sliceSize, layoutValue, elemAlign, "makeslice.buf")

 		// Extend or truncate if necessary. This is safe as we've already done
 		// the bounds check.
@@ -2442,7 +2715,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			return b.createMapIteratorNext(rangeVal, llvmRangeVal, it), nil
 		}
 	case *ssa.Phi:
-		phi := b.CreatePHI(b.getLLVMType(expr.Type()), "")
+		phiType := b.storedParamType(b.getLLVMType(expr.Type()), false)
+		phi := b.CreatePHI(phiType, "")
 		b.phis = append(b.phis, phiNode{expr, phi})
 		return phi, nil
 	case *ssa.Range:
@@ -3445,8 +3719,7 @@ func (b *builder) createUnOp(unop *ssa.UnOp) (llvm.Value, error) {
 			return fn, nil
 		} else {
 			b.createNilCheck(unop.X, x, "deref")
-			load := b.CreateLoad(valueType, x, "")
-			return load, nil
+			return b.loadFromStorage(x, unop.Type(), ""), nil
 		}
 	case token.XOR: // ^x, toggle all bits in integer
 		return b.CreateXor(x, llvm.ConstInt(x.Type(), ^uint64(0), false), ""), nil
@@ -4,6 +4,7 @@ import (
 	"flag"
 	"go/types"
 	"os"
+	"regexp"
 	"strconv"
 	"strings"
 	"testing"
@@ -50,6 +51,8 @@ func TestCompiler(t *testing.T) {
 		{"channel.go", "", ""},
 		{"gc.go", "", ""},
 		{"zeromap.go", "", ""},
+		{"generics.go", "", ""},
+		{"large.go", "", ""},
 	}
 	if goMinor >= 20 {
 		tests = append(tests, testCase{"go1.20.go", "", ""})
@@ -57,6 +60,9 @@ func TestCompiler(t *testing.T) {
 	if goMinor >= 21 {
 		tests = append(tests, testCase{"go1.21.go", "", ""})
 	}
+	if goMinor >= 27 {
+		tests = append(tests, testCase{"go1.27.go", "", ""})
+	}

 	for _, tc := range tests {
 		name := tc.file
@@ -121,30 +127,138 @@ func TestCompiler(t *testing.T) {
 				t.Fatal("failed to read golden file:", err)
 			}

-			if !fuzzyEqualIR(mod.String(), string(expected)) {
-				t.Errorf("output does not match expected output:\n%s", mod.String())
+			if diff := diffIR(string(expected), mod.String()); diff != "" {
+				t.Errorf("output does not match expected output (re-run with -update to regenerate):\n%s", diff)
 			}
 		})
 	}
 }

-// fuzzyEqualIR returns true if the two LLVM IR strings passed in are roughly
-// equal. That means, only relevant lines are compared (excluding comments
-// etc.).
-func fuzzyEqualIR(s1, s2 string) bool {
-	lines1 := filterIrrelevantIRLines(strings.Split(s1, "\n"))
-	lines2 := filterIrrelevantIRLines(strings.Split(s2, "\n"))
-	if len(lines1) != len(lines2) {
-		return false
-	}
-	for i, line1 := range lines1 {
-		line2 := lines2[i]
-		if line1 != line2 {
-			return false
+func TestOptimizedLargeAggregateABI(t *testing.T) {
+	options := &compileopts.Options{Target: "wasm"}
+	mod, errs := testCompilePackage(t, options, "large-optimized.go")
+	if len(errs) != 0 {
+		for _, err := range errs {
+			t.Error(err)
 		}
+		return
+	}
+	defer mod.Dispose()
+
+	passOptions := llvm.NewPassBuilderOptions()
+	defer passOptions.Dispose()
+	if err := mod.RunPasses("default<O2>", llvm.TargetMachine{}, passOptions); err != nil {
+		t.Fatal(err)
 	}

-	return true
+	resultFn := mod.NamedFunction("main.makeLargeOptimizedValue")
+	if resultFn.IsNil() {
+		t.Fatal("missing function main.makeLargeOptimizedValue")
+	}
+	if resultType := resultFn.GlobalValueType().ReturnType(); resultType.TypeKind() != llvm.VoidTypeKind {
+		t.Errorf("large aggregate result was promoted to %s", resultType)
+	}
+
+	for _, name := range []string{
+		"main.makeLargeOptimizedValue",
+		"main.readLargeOptimizedValue",
+		"main.readMixedLargeOptimizedValue",
+	} {
+		fn := mod.NamedFunction(name)
+		if fn.IsNil() {
+			t.Fatalf("missing function %s", name)
+		}
+		if paramType := fn.GlobalValueType().ParamTypes()[0]; paramType.TypeKind() != llvm.PointerTypeKind {
+			t.Errorf("%s aggregate parameter was promoted to %s", name, paramType)
+		}
+	}
+}
+
+// normalizeIR canonicalizes LLVM IR so a single golden file keeps matching
+// across LLVM versions. Golden files are written against LLVM <21; newer LLVM
+// prints some attributes differently.
+func normalizeIR(s string) string {
+	// Golden files are written using the pre-LLVM21 'nocapture' spelling,
+	// which LLVM printed before any co-occurring attribute such as
+	// 'readonly' (e.g. "ptr nocapture readonly"). LLVM 21+ prints the
+	// equivalent 'captures(none)' instead, and after such attributes (e.g.
+	// "ptr readonly captures(none)"). Normalize both name and position back
+	// to the old spelling.
+	s = normalizeCapturesAttr(s)
+
+	// LLVM 21+ also added an explicit 'nocreateundeforpoison' attribute to
+	// certain intrinsic declarations (e.g. llvm.umin) that were implicitly
+	// assumed not to create undef/poison before. It's unrelated to the
+	// behavior under test, so ignore it for comparison.
+	s = strings.ReplaceAll(s, "nocreateundeforpoison ", "")
+
+	// LLVM 22 dropped the (redundant) i64 size argument from
+	// llvm.lifetime.start/end. Normalize away that argument so golden files
+	// written against the two-argument form still match.
+	s = lifetimeSizeArgRe.ReplaceAllString(s, "$1")
+
+	return s
+}
+
+// diffIR compares two LLVM IR strings, ignoring irrelevant lines (comments,
+// empty lines, etc.) and normalizing LLVM-version-specific spellings via
+// normalizeIR. It returns "" when they are equal. Otherwise it returns a
+// compact diff of only the region that differs: the common prefix and suffix
+// are trimmed, then the differing expected lines (prefixed "-") are shown
+// followed by the differing actual lines (prefixed "+").
+func diffIR(expected, actual string) string {
+	exp := filterIrrelevantIRLines(strings.Split(normalizeIR(expected), "\n"))
+	act := filterIrrelevantIRLines(strings.Split(normalizeIR(actual), "\n"))
+
+	// Trim the common prefix.
+	start := 0
+	for start < len(exp) && start < len(act) && exp[start] == act[start] {
+		start++
+	}
+	// Trim the common suffix.
+	e, a := len(exp), len(act)
+	for e > start && a > start && exp[e-1] == act[a-1] {
+		e--
+		a--
+	}
+	if start == e && start == a {
+		return "" // equal
+	}
+
+	var b strings.Builder
+	b.WriteString("first difference at relevant line ")
+	b.WriteString(strconv.Itoa(start + 1))
+	b.WriteString(":\n")
+	for _, line := range exp[start:e] {
+		b.WriteString("- ")
+		b.WriteString(line)
+		b.WriteByte('\n')
+	}
+	for _, line := range act[start:a] {
+		b.WriteString("+ ")
+		b.WriteString(line)
+		b.WriteByte('\n')
+	}
+	return b.String()
+}
+
+// capturesNoneAttrRe matches a co-occurring attribute directly followed by
+// 'captures(none)', which is how LLVM 21+ orders these two attributes when
+// printing IR (the pre-LLVM21 'nocapture' attribute printed the other way
+// around).
+var capturesNoneAttrRe = regexp.MustCompile(`\b(readonly|readnone|writeonly|nonnull)\s+captures\(none\)`)
+
+// lifetimeSizeArgRe matches the i64 size argument of an
+// llvm.lifetime.start/end call or declaration, which LLVM 22 removed.
+var lifetimeSizeArgRe = regexp.MustCompile(`(@llvm\.lifetime\.(?:start|end)\.p0\()i64(?: immarg| \d+), `)
+
+// normalizeCapturesAttr rewrites LLVM 21+'s 'captures(none)' attribute back
+// to the pre-LLVM21 'nocapture' spelling and position, so golden IR files
+// written against LLVM <21 keep matching.
+func normalizeCapturesAttr(s string) string {
+	s = capturesNoneAttrRe.ReplaceAllString(s, "nocapture $1")
+	s = strings.ReplaceAll(s, "captures(none)", "nocapture")
+	return s
 }

 // filterIrrelevantIRLines removes lines from the input slice of strings that
@@ -187,9 +301,9 @@ func TestCompilerErrors(t *testing.T) {
 		t.Error(err)
 	}
 	errorsFileString := strings.ReplaceAll(string(errorsFile), "\r\n", "\n")
-	for _, line := range strings.Split(errorsFileString, "\n") {
-		if strings.HasPrefix(line, "// ERROR: ") {
-			expectedErrors = append(expectedErrors, strings.TrimPrefix(line, "// ERROR: "))
+	for line := range strings.SplitSeq(errorsFileString, "\n") {
+		if after, ok := strings.CutPrefix(line, "// ERROR: "); ok {
+			expectedErrors = append(expectedErrors, after)
 		}
 	}

@@ -213,6 +327,49 @@ func TestCompilerErrors(t *testing.T) {
 	}
 }

+func TestAggregateValueCount(t *testing.T) {
+	t.Parallel()
+
+	ctx := llvm.NewContext()
+	defer ctx.Dispose()
+
+	byteType := ctx.Int8Type()
+	tests := []struct {
+		name     string
+		typ      llvm.Type
+		count    uint64
+		exceeded bool
+	}{
+		{"empty", llvm.ArrayType(byteType, 0), 0, false},
+		{"limit", llvm.ArrayType(byteType, 1024), 1024, false},
+		{"over limit", llvm.ArrayType(byteType, 1025), 0, true},
+		{"combined limit", ctx.StructType([]llvm.Type{
+			llvm.ArrayType(byteType, 512),
+			llvm.ArrayType(byteType, 512),
+		}, false), 1024, false},
+		{"combined over limit", ctx.StructType([]llvm.Type{
+			llvm.ArrayType(byteType, 1000),
+			llvm.ArrayType(byteType, 1000),
+		}, false), 0, true},
+		{"comma-ok over limit", ctx.StructType([]llvm.Type{
+			llvm.ArrayType(byteType, 1024),
+			ctx.Int1Type(),
+		}, false), 0, true},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			count, exceeded := aggregateValueCount(test.typ, 0)
+			if exceeded != test.exceeded {
+				t.Errorf("expected exceeded=%t, got %t", test.exceeded, exceeded)
+			}
+			if !exceeded && count != test.count {
+				t.Errorf("expected count=%d, got %d", test.count, count)
+			}
+		})
+	}
+}
+
 // Build a package given a number of compiler options and a file.
 func testCompilePackage(t *testing.T, options *compileopts.Options, file string) (llvm.Module, []error) {
 	target, err := compileopts.LoadTarget(options)
@@ -32,7 +32,7 @@ func (b *builder) supportsRecover() bool {
 		// proposal of WebAssembly:
 		// https://github.com/WebAssembly/exception-handling
 		return false
-	case "riscv64", "xtensa":
+	case "xtensa":
 		// TODO: add support for these architectures
 		return false
 	default:
@@ -60,10 +60,6 @@ func (b *builder) deferInitFunc() {
 	b.deferExprFuncs = make(map[ssa.Value]int)
 	b.deferBuiltinFuncs = make(map[ssa.Value]deferBuiltin)

-	// Create defer list pointer.
-	b.deferPtr = b.CreateAlloca(b.dataPtrType, "deferPtr")
-	b.CreateStore(llvm.ConstPointerNull(b.dataPtrType), b.deferPtr)
-
 	if b.hasDeferFrame() {
 		// Set up the defer frame with the current stack pointer.
 		// This assumes that the stack pointer doesn't move outside of the
@@ -73,12 +69,22 @@ func (b *builder) deferInitFunc() {
 		// in the setjmp-like inline assembly.
 		deferFrameType := b.getLLVMRuntimeType("deferFrame")
 		b.deferFrame = b.CreateAlloca(deferFrameType, "deferframe.buf")
+		// The field index must match the DeferPtr field in runtime.deferFrame,
+		// defined in src/runtime/panic.go.
+		b.deferPtr = b.CreateInBoundsGEP(deferFrameType, b.deferFrame, []llvm.Value{
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(b.ctx.Int32Type(), 6, false), // DeferPtr field
+		}, "deferPtr")
 		stackPointer := b.readStackPointer()
 		b.createRuntimeCall("setupDeferFrame", []llvm.Value{b.deferFrame, stackPointer}, "")

 		// Create the landing pad block, which is where control transfers after
 		// a panic.
 		b.landingpad = b.ctx.AddBasicBlock(b.llvmFn, "lpad")
+	} else {
+		// Create defer list pointer.
+		b.deferPtr = b.CreateAlloca(b.dataPtrType, "deferPtr")
+		b.CreateStore(llvm.ConstPointerNull(b.dataPtrType), b.deferPtr)
 	}
 }

@@ -110,8 +116,8 @@ func (b *builder) createLandingPad() {
 func (b *builder) createCheckpoint(ptr llvm.Value) llvm.Value {
 	// Construct inline assembly equivalents of setjmp.
 	// The assembly works as follows:
-	//   * All registers (both callee-saved and caller saved) are clobbered
-	//     after the inline assembly returns.
+	//   * Registers are either clobbered or, on 386, saved for longjmp to
+	//     restore if the ABI requires them to survive calls.
 	//   * The assembly stores the address just past the end of the assembly
 	//     into the jump buffer.
 	//   * The return value (eax, rax, r0, etc) is set to zero in the inline
@@ -124,8 +130,12 @@ func (b *builder) createCheckpoint(ptr llvm.Value) llvm.Value {
 		asmString = `
 xorl %eax, %eax
 movl $$1f, 4(%ebx)
+movl %ebx, 8(%ebx)
+movl %esi, 12(%ebx)
+movl %edi, 16(%ebx)
+movl %ebp, 20(%ebx)
 1:`
-		constraints = "={eax},{ebx},~{ebx},~{ecx},~{edx},~{esi},~{edi},~{ebp},~{xmm0},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{fpsr},~{fpcr},~{flags},~{dirflag},~{memory}"
+		constraints = "={eax},{ebx},~{ecx},~{edx},~{xmm0},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{fpsr},~{fpcr},~{flags},~{dirflag},~{memory}"
 		// This doesn't include the floating point stack because TinyGo uses
 		// newer floating point instructions.
 	case "x86_64":
@@ -207,12 +217,20 @@ sw $$ra, 4($$5)
 			// So only add them when using hardfloat.
 			constraints += ",~{$f0},~{$f1},~{$f2},~{$f3},~{$f4},~{$f5},~{$f6},~{$f7},~{$f8},~{$f9},~{$f10},~{$f11},~{$f12},~{$f13},~{$f14},~{$f15},~{$f16},~{$f17},~{$f18},~{$f19},~{$f20},~{$f21},~{$f22},~{$f23},~{$f24},~{$f25},~{$f26},~{$f27},~{$f28},~{$f29},~{$f30},~{$f31}"
 		}
-	case "riscv32":
-		asmString = `
+	case "riscv32", "riscv64":
+		if b.archFamily() == "riscv32" {
+			asmString = `
 la a2, 1f
 sw a2, 4(a1)
 li a0, 0
 1:`
+		} else {
+			asmString = `
+la a2, 1f
+sd a2, 8(a1)
+li a0, 0
+1:`
+		}
 		constraints = "={a0},{a1},~{a1},~{a2},~{a3},~{a4},~{a5},~{a6},~{a7},~{s0},~{s1},~{s2},~{s3},~{s4},~{s5},~{s6},~{s7},~{s8},~{s9},~{s10},~{s11},~{t0},~{t1},~{t2},~{t3},~{t4},~{t5},~{t6},~{ra},~{f0},~{f1},~{f2},~{f3},~{f4},~{f5},~{f6},~{f7},~{f8},~{f9},~{f10},~{f11},~{f12},~{f13},~{f14},~{f15},~{f16},~{f17},~{f18},~{f19},~{f20},~{f21},~{f22},~{f23},~{f24},~{f25},~{f26},~{f27},~{f28},~{f29},~{f30},~{f31},~{memory}"
 	default:
 		// This case should have been handled by b.supportsRecover().
@@ -237,6 +255,17 @@ func (b *builder) createInvokeCheckpoint() {
 	b.currentBlockInfo.exit = continueBB
 }

+// createFaultCheckpoint is like createInvokeCheckpoint but for use in fault
+// blocks (e.g., bounds check failures). Unlike createInvokeCheckpoint, it does
+// not update currentBlockInfo.exit because the fault block is a dead-end that
+// does not participate in phi node resolution.
+func (b *builder) createFaultCheckpoint() {
+	isZero := b.createCheckpoint(b.deferFrame)
+	continueBB := b.insertBasicBlock("")
+	b.CreateCondBr(isZero, continueBB, b.landingpad)
+	b.SetInsertPointAtEnd(continueBB)
+}
+
 // isInLoop checks if there is a path from the current block to itself.
 // Use Tarjan's strongly connected components algorithm to search for cycles.
 // A one-node SCC is a cycle iff there is an edge from the node to itself.
@@ -338,6 +367,44 @@ type tarjanNode struct {
 	cyclic bool
 }

+type llvmValueList struct {
+	values []llvm.Value
+	types  []llvm.Type
+}
+
+func newLLVMValueList(values ...llvm.Value) llvmValueList {
+	var list llvmValueList
+	list.append(values...)
+	return list
+}
+
+func (l *llvmValueList) append(values ...llvm.Value) {
+	for _, value := range values {
+		l.values = append(l.values, value)
+		l.types = append(l.types, value.Type())
+	}
+}
+
+func (l *llvmValueList) appendSSAValues(values []ssa.Value, lower func(ssa.Value) llvm.Value) {
+	for _, value := range values {
+		l.append(lower(value))
+	}
+}
+
+func (b *builder) loadDeferredCallParams(structType llvm.Type, ptr llvm.Value) []llvm.Value {
+	fieldTypes := structType.StructElementTypes()
+	values := make([]llvm.Value, 0, len(fieldTypes)-2)
+	zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+	for i := 2; i < len(fieldTypes); i++ {
+		fieldPtr := b.CreateInBoundsGEP(structType, ptr, []llvm.Value{
+			zero,
+			llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
+		}, "gep")
+		values = append(values, b.CreateLoad(fieldTypes[i], fieldPtr, "param"))
+	}
+	return values
+}
+
 // createDefer emits a single defer instruction, to be run when this function
 // returns.
 func (b *builder) createDefer(instr *ssa.Defer) {
@@ -345,31 +412,28 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 	// make a linked list.
 	next := b.CreateLoad(b.dataPtrType, b.deferPtr, "defer.next")

-	var values []llvm.Value
-	valueTypes := []llvm.Type{b.uintptrType, next.Type()}
+	var values llvmValueList
+	lowerArgument := func(value ssa.Value) llvm.Value {
+		return b.getCallArgument(value, false)
+	}
 	if instr.Call.IsInvoke() {
 		// Method call on an interface.

 		// Get callback type number.
-		methodName := instr.Call.Method.FullName()
-		if _, ok := b.deferInvokeFuncs[methodName]; !ok {
-			b.deferInvokeFuncs[methodName] = len(b.allDeferFuncs)
+		key := b.getInvokeFunctionName(&instr.Call)
+		if _, ok := b.deferInvokeFuncs[key]; !ok {
+			b.deferInvokeFuncs[key] = len(b.allDeferFuncs)
 			b.allDeferFuncs = append(b.allDeferFuncs, &instr.Call)
 		}
-		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferInvokeFuncs[methodName]), false)
+		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferInvokeFuncs[key]), false)

 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by the call parameters).
 		itf := b.getValue(instr.Call.Value, getPos(instr)) // interface
 		typecode := b.CreateExtractValue(itf, 0, "invoke.func.typecode")
 		receiverValue := b.CreateExtractValue(itf, 1, "invoke.func.receiver")
-		values = []llvm.Value{callback, next, typecode, receiverValue}
-		valueTypes = append(valueTypes, b.dataPtrType, b.dataPtrType)
-		for _, arg := range instr.Call.Args {
-			val := b.getValue(arg, getPos(instr))
-			values = append(values, val)
-			valueTypes = append(valueTypes, val.Type())
-		}
+		values = newLLVMValueList(callback, next, typecode, receiverValue)
+		values.appendSSAValues(instr.Call.Args, lowerArgument)

 	} else if callee, ok := instr.Call.Value.(*ssa.Function); ok {
 		// Regular function call.
@@ -381,12 +445,11 @@ func (b *builder) createDefer(instr *ssa.Defer) {

 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields).
-		values = []llvm.Value{callback, next}
-		for _, param := range instr.Call.Args {
-			llvmParam := b.getValue(param, getPos(instr))
-			values = append(values, llvmParam)
-			valueTypes = append(valueTypes, llvmParam.Type())
-		}
+		values = newLLVMValueList(callback, next)
+		exported := b.getFunctionInfo(callee).exported
+		values.appendSSAValues(instr.Call.Args, func(value ssa.Value) llvm.Value {
+			return b.getCallArgument(value, exported)
+		})

 	} else if makeClosure, ok := instr.Call.Value.(*ssa.MakeClosure); ok {
 		// Immediately applied function literal with free variables.
@@ -409,14 +472,9 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by all parameters including the
 		// context pointer).
-		values = []llvm.Value{callback, next}
-		for _, param := range instr.Call.Args {
-			llvmParam := b.getValue(param, getPos(instr))
-			values = append(values, llvmParam)
-			valueTypes = append(valueTypes, llvmParam.Type())
-		}
-		values = append(values, context)
-		valueTypes = append(valueTypes, context.Type())
+		values = newLLVMValueList(callback, next)
+		values.appendSSAValues(instr.Call.Args, lowerArgument)
+		values.append(context)

 	} else if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
 		var argTypes []types.Type
@@ -439,11 +497,8 @@ func (b *builder) createDefer(instr *ssa.Defer) {

 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields).
-		values = []llvm.Value{callback, next}
-		for _, param := range argValues {
-			values = append(values, param)
-			valueTypes = append(valueTypes, param.Type())
-		}
+		values = newLLVMValueList(callback, next)
+		values.append(argValues...)

 	} else {
 		funcValue := b.getValue(instr.Call.Value, getPos(instr))
@@ -458,20 +513,15 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by all parameters including the
 		// context pointer).
-		values = []llvm.Value{callback, next, funcValue}
-		valueTypes = append(valueTypes, funcValue.Type())
-		for _, param := range instr.Call.Args {
-			llvmParam := b.getValue(param, getPos(instr))
-			values = append(values, llvmParam)
-			valueTypes = append(valueTypes, llvmParam.Type())
-		}
+		values = newLLVMValueList(callback, next, funcValue)
+		values.appendSSAValues(instr.Call.Args, lowerArgument)
 	}

 	// Make a struct out of the collected values to put in the deferred call
 	// struct.
-	deferredCallType := b.ctx.StructType(valueTypes, false)
+	deferredCallType := b.ctx.StructType(values.types, false)
 	deferredCall := llvm.ConstNull(deferredCallType)
-	for i, value := range values {
+	for i, value := range values.values {
 		deferredCall = b.CreateInsertValue(deferredCall, value, i, "")
 	}

@@ -487,8 +537,9 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		// This may be hit a variable number of times, so use a heap allocation.
 		size := b.targetData.TypeAllocSize(deferredCallType)
 		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
-		nilPtr := llvm.ConstNull(b.dataPtrType)
-		alloca = b.createRuntimeCall("alloc", []llvm.Value{sizeValue, nilPtr}, "defer.alloc.call")
+		layoutValue := b.createObjectLayout(deferredCallType, instr.Pos())
+		align := b.targetData.ABITypeAlignment(deferredCallType)
+		alloca = b.createAlloc(sizeValue, layoutValue, align, "defer.alloc.call")
 	}
 	if b.NeedsStackObjects {
 		b.trackPointer(alloca)
@@ -572,19 +623,11 @@ func (b *builder) createRunDefers() {
 				valueTypes = append(valueTypes, b.dataPtrType, b.dataPtrType)
 			}

-			for _, arg := range callback.Args {
-				valueTypes = append(valueTypes, b.getLLVMType(arg.Type()))
-			}
+			valueTypes = b.appendStoredValueTypes(valueTypes, callback.Args, false)

 			// Extract the params from the struct (including receiver).
-			forwardParams := []llvm.Value{}
-			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			deferredCallType := b.ctx.StructType(valueTypes, false)
-			for i := 2; i < len(valueTypes); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "gep")
-				forwardParam := b.CreateLoad(valueTypes[i], gep, "param")
-				forwardParams = append(forwardParams, forwardParam)
-			}
+			forwardParams := b.loadDeferredCallParams(deferredCallType, deferData)

 			var fnPtr llvm.Value
 			var fnType llvm.Type
@@ -613,6 +656,7 @@ func (b *builder) createRunDefers() {
 				// with a strict calling convention.
 				forwardParams = append(forwardParams, llvm.Undef(b.dataPtrType))
 			}
+			forwardParams = b.prependIndirectResult(callback.Signature(), false, forwardParams, "defer.result")

 			b.createCall(fnType, fnPtr, forwardParams, "")

@@ -621,27 +665,21 @@ func (b *builder) createRunDefers() {

 			// Get the real defer struct type and cast to it.
 			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
-			for _, param := range getParams(callback.Signature) {
-				valueTypes = append(valueTypes, b.getLLVMType(param.Type()))
-			}
+			exported := b.getFunctionInfo(callback).exported
+			valueTypes = b.appendStoredParamTypes(valueTypes, getParams(callback.Signature), exported)
 			deferredCallType := b.ctx.StructType(valueTypes, false)

 			// Extract the params from the struct.
-			forwardParams := []llvm.Value{}
-			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
-			for i := range getParams(callback.Signature) {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
-				forwardParam := b.CreateLoad(valueTypes[i+2], gep, "param")
-				forwardParams = append(forwardParams, forwardParam)
-			}
+			forwardParams := b.loadDeferredCallParams(deferredCallType, deferData)

 			// Plain TinyGo functions add some extra parameters to implement async functionality and function receivers.
 			// These parameters should not be supplied when calling into an external C/ASM function.
-			if !b.getFunctionInfo(callback).exported {
+			if !exported {
 				// Add the context parameter. We know it is ignored by the receiving
 				// function, but we have to pass one anyway.
 				forwardParams = append(forwardParams, llvm.Undef(b.dataPtrType))
 			}
+			forwardParams = b.prependIndirectResult(callback.Signature, exported, forwardParams, "defer.result")

 			// Call real function.
 			fnType, fn := b.getFunction(callback)
@@ -651,24 +689,16 @@ func (b *builder) createRunDefers() {
 			// Get the real defer struct type and cast to it.
 			fn := callback.Fn.(*ssa.Function)
 			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
-			params := fn.Signature.Params()
-			for i := 0; i < params.Len(); i++ {
-				valueTypes = append(valueTypes, b.getLLVMType(params.At(i).Type()))
-			}
+			valueTypes = b.appendStoredParamTypes(valueTypes, getParams(fn.Signature), false)
 			valueTypes = append(valueTypes, b.dataPtrType) // closure
 			deferredCallType := b.ctx.StructType(valueTypes, false)

 			// Extract the params from the struct.
-			forwardParams := []llvm.Value{}
-			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
-			for i := 2; i < len(valueTypes); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "")
-				forwardParam := b.CreateLoad(valueTypes[i], gep, "param")
-				forwardParams = append(forwardParams, forwardParam)
-			}
+			forwardParams := b.loadDeferredCallParams(deferredCallType, deferData)

 			// Call deferred function.
 			fnType, llvmFn := b.getFunction(fn)
+			forwardParams = b.prependIndirectResult(fn.Signature, false, forwardParams, "defer.result")
 			b.createCall(fnType, llvmFn, forwardParams, "")
 		case *ssa.Builtin:
 			db := b.deferBuiltinFuncs[callback]
@@ -678,20 +708,14 @@ func (b *builder) createRunDefers() {

 			//Get signature from call results
 			params := callback.Type().Underlying().(*types.Signature).Params()
-			for i := 0; i < params.Len(); i++ {
-				valueTypes = append(valueTypes, b.getLLVMType(params.At(i).Type()))
+			for v := range params.Variables() {
+				valueTypes = append(valueTypes, b.getLLVMType(v.Type()))
 			}

 			deferredCallType := b.ctx.StructType(valueTypes, false)

 			// Extract the params from the struct.
-			var argValues []llvm.Value
-			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
-			for i := 0; i < params.Len(); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
-				forwardParam := b.CreateLoad(valueTypes[i+2], gep, "param")
-				argValues = append(argValues, forwardParam)
-			}
+			argValues := b.loadDeferredCallParams(deferredCallType, deferData)

 			_, err := b.createBuiltin(db.argTypes, argValues, db.callName, db.pos)
 			if err != nil {
@@ -10,6 +10,91 @@ import (
 	"tinygo.org/x/go-llvm"
 )

+// LLVM recursively expands each struct field and array element in parameters
+// and results into separate values. It gets very slow with too many values, so
+// pass larger aggregates indirectly before LLVM expands them.
+const maxDirectAggregateValues = 1024
+
+func (c *compilerContext) getLLVMResultType(sig *types.Signature) llvm.Type {
+	switch sig.Results().Len() {
+	case 0:
+		return c.ctx.VoidType()
+	case 1:
+		return c.getLLVMType(sig.Results().At(0).Type())
+	default:
+		results := make([]llvm.Type, sig.Results().Len())
+		for i := range results {
+			results[i] = c.getLLVMType(sig.Results().At(i).Type())
+		}
+		return c.ctx.StructType(results, false)
+	}
+}
+
+func (c *compilerContext) hasIndirectResult(sig *types.Signature) (llvm.Type, bool) {
+	resultType := c.getLLVMResultType(sig)
+	return resultType, c.isIndirectAggregate(resultType)
+}
+
+func (c *compilerContext) isIndirectAggregate(typ llvm.Type) bool {
+	switch typ.TypeKind() {
+	case llvm.ArrayTypeKind, llvm.StructTypeKind:
+		_, exceeded := aggregateValueCount(typ, 0)
+		return exceeded
+	default:
+		return false
+	}
+}
+
+func aggregateValueCount(typ llvm.Type, count uint64) (uint64, bool) {
+	switch typ.TypeKind() {
+	case llvm.ArrayTypeKind:
+		length := uint64(typ.ArrayLength())
+		if length == 0 {
+			return count, false
+		}
+		elementCount, exceeded := aggregateValueCount(typ.ElementType(), 0)
+		if exceeded {
+			return count, true
+		}
+		if elementCount != 0 && length > (maxDirectAggregateValues-count)/elementCount {
+			return count, true
+		}
+		return count + length*elementCount, false
+	case llvm.StructTypeKind:
+		for _, field := range typ.StructElementTypes() {
+			var exceeded bool
+			count, exceeded = aggregateValueCount(field, count)
+			if exceeded {
+				return count, true
+			}
+		}
+		return count, false
+	default:
+		count++
+		return count, count > maxDirectAggregateValues
+	}
+}
+
+func isLLVMValueType(typ types.Type) bool {
+	switch typ := typ.Underlying().(type) {
+	case *types.Basic:
+		return typ.Kind() != types.Invalid
+	case *types.Array:
+		return isLLVMValueType(typ.Elem())
+	case *types.Struct:
+		for field := range typ.Fields() {
+			if !isLLVMValueType(field.Type()) {
+				return false
+			}
+		}
+		return true
+	case *types.Chan, *types.Interface, *types.Map, *types.Pointer, *types.Signature, *types.Slice:
+		return true
+	default:
+		return false
+	}
+}
+
 // createFuncValue creates a function value from a raw function pointer with no
 // context.
 func (b *builder) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
@@ -48,28 +133,18 @@ func (c *compilerContext) getFuncType(typ *types.Signature) llvm.Type {

 // getLLVMFunctionType returns a LLVM function type for a given signature.
 func (c *compilerContext) getLLVMFunctionType(typ *types.Signature) llvm.Type {
-	// Get the return type.
-	var returnType llvm.Type
-	switch typ.Results().Len() {
-	case 0:
-		// No return values.
-		returnType = c.ctx.VoidType()
-	case 1:
-		// Just one return value.
-		returnType = c.getLLVMType(typ.Results().At(0).Type())
-	default:
-		// Multiple return values. Put them together in a struct.
-		// This appears to be the common way to handle multiple return values in
-		// LLVM.
-		members := make([]llvm.Type, typ.Results().Len())
-		for i := 0; i < typ.Results().Len(); i++ {
-			members[i] = c.getLLVMType(typ.Results().At(i).Type())
-		}
-		returnType = c.ctx.StructType(members, false)
-	}
+	returnType, indirectResult := c.hasIndirectResult(typ)

 	// Get the parameter types.
 	var paramTypes []llvm.Type
+	if indirectResult {
+		// LLVM expands aggregate returns into scalar leaves before deciding
+		// whether to pass them indirectly, so a large IR return can exhaust
+		// memory. Returning void avoids that expansion and cannot be demoted
+		// again. Keep the result pointer first so the context remains last.
+		paramTypes = append(paramTypes, c.dataPtrType)
+		returnType = c.ctx.VoidType()
+	}
 	if typ.Recv() != nil {
 		recv := c.getLLVMType(typ.Recv().Type())
 		if recv.StructName() == "runtime._interface" {
@@ -81,8 +156,8 @@ func (c *compilerContext) getLLVMFunctionType(typ *types.Signature) llvm.Type {
 			paramTypes = append(paramTypes, info.llvmType)
 		}
 	}
-	for i := 0; i < typ.Params().Len(); i++ {
-		subType := c.getLLVMType(typ.Params().At(i).Type())
+	for v := range typ.Params().Variables() {
+		subType := c.getLLVMType(v.Type())
 		for _, info := range c.expandFormalParamType(subType, "", nil) {
 			paramTypes = append(paramTypes, info.llvmType)
 		}
@@ -112,7 +187,7 @@ func (b *builder) parseMakeClosure(expr *ssa.MakeClosure) (llvm.Value, error) {

 	// Store the bound variables in a single object, allocating it on the heap
 	// if necessary.
-	context := b.emitPointerPack(boundVars)
+	context := b.emitPointerPack(boundVars, expr.Pos())

 	// Create the closure.
 	_, fn := b.getFunction(f)
@@ -5,11 +5,38 @@ package compiler

 import (
 	"go/token"
+	"slices"

 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )

+// Heap-allocate a buffer of the given size. This will typically call
+// runtime.alloc.
+func (b *builder) createAlloc(sizeValue, layoutValue llvm.Value, align int, comment string) llvm.Value {
+	// Normally allocate using "runtime.alloc", but use "runtime.alloc_noheap"
+	// if the //go:noheap pragma is used.
+	allocFunc := "alloc"
+	if b.info.noheap {
+		allocFunc = "alloc_noheap"
+	}
+
+	// Allocs that don't allocate anything can return an architecture-specific
+	// sentinel value.
+	if !sizeValue.IsAConstantInt().IsNil() && sizeValue.ZExtValue() == 0 {
+		allocFunc = "alloc_zero"
+	}
+
+	// Make the runtime call.
+	call := b.createRuntimeCall(allocFunc, []llvm.Value{sizeValue, layoutValue}, comment)
+	if align == 0 || align&(align-1) != 0 {
+		panic("invalid alignment")
+	}
+	call.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align)))
+
+	return call
+}
+
 // trackExpr inserts pointer tracking intrinsics for the GC if the expression is
 // one of the expressions that need this.
 func (b *builder) trackExpr(expr ssa.Value, value llvm.Value) {
@@ -62,7 +89,7 @@ func (b *builder) trackValue(value llvm.Value) {
 			return
 		}
 		numElements := typ.StructElementTypesCount()
-		for i := 0; i < numElements; i++ {
+		for i := range numElements {
 			subValue := b.CreateExtractValue(value, i, "")
 			b.trackValue(subValue)
 		}
@@ -71,7 +98,7 @@ func (b *builder) trackValue(value llvm.Value) {
 			return
 		}
 		numElements := typ.ArrayLength()
-		for i := 0; i < numElements; i++ {
+		for i := range numElements {
 			subValue := b.CreateExtractValue(value, i, "")
 			b.trackValue(subValue)
 		}
@@ -92,12 +119,7 @@ func typeHasPointers(t llvm.Type) bool {
 	case llvm.PointerTypeKind:
 		return true
 	case llvm.StructTypeKind:
-		for _, subType := range t.StructElementTypes() {
-			if typeHasPointers(subType) {
-				return true
-			}
-		}
-		return false
+		return slices.ContainsFunc(t.StructElementTypes(), typeHasPointers)
 	case llvm.ArrayTypeKind:
 		if t.ArrayLength() == 0 {
 			return false
@@ -47,20 +47,16 @@ func (b *builder) createGo(instr *ssa.Go) {
 		return
 	}

-	// Get all function parameters to pass to the goroutine.
 	var params []llvm.Value
-	for _, param := range instr.Call.Args {
-		params = append(params, b.expandFormalParam(b.getValue(param, getPos(instr)))...)
-	}
-
 	var prefix string
 	var funcPtr llvm.Value
 	var funcType llvm.Type
+	var context llvm.Value
 	hasContext := false
+	exported := false
 	if callee := instr.Call.StaticCallee(); callee != nil {
 		// Static callee is known. This makes it easier to start a new
 		// goroutine.
-		var context llvm.Value
 		switch value := instr.Call.Value.(type) {
 		case *ssa.Function:
 			// Goroutine call is regular function call. No context is necessary.
@@ -73,10 +69,10 @@ func (b *builder) createGo(instr *ssa.Go) {
 			panic("StaticCallee returned an unexpected value")
 		}
 		if !context.IsNil() {
-			params = append(params, context) // context parameter
 			hasContext = true
 		}
 		funcType, funcPtr = b.getFunction(callee)
+		exported = b.getFunctionInfo(callee).exported
 	} else if instr.Call.IsInvoke() {
 		// This is a method call on an interface value.
 		itf := b.getValue(instr.Call.Value, getPos(instr))
@@ -84,23 +80,32 @@ func (b *builder) createGo(instr *ssa.Go) {
 		itfValue := b.CreateExtractValue(itf, 1, "")
 		funcPtr = b.getInvokeFunction(&instr.Call)
 		funcType = funcPtr.GlobalValueType()
-		params = append([]llvm.Value{itfValue}, params...) // start with receiver
-		params = append(params, itfTypeCode)               // end with typecode
+		params = append(params, itfValue)
+		context = itfTypeCode
 	} else {
 		// This is a function pointer.
 		// At the moment, two extra params are passed to the newly started
 		// goroutine:
 		//   * The function context, for closures.
 		//   * The function pointer (for tasks).
-		var context llvm.Value
 		funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value, getPos(instr)))
 		funcType = b.getLLVMFunctionType(instr.Call.Value.Type().Underlying().(*types.Signature))
-		params = append(params, context, funcPtr)
 		hasContext = true
-		prefix = b.fn.RelString(nil)
+		prefix = b.getFunctionInfo(b.fn).linkName
 	}

-	paramBundle := b.emitPointerPack(params)
+	for _, param := range instr.Call.Args {
+		params = append(params, b.getGoroutineCallArgument(param, exported)...)
+	}
+	if !context.IsNil() {
+		params = append(params, context)
+	}
+	if hasContext && instr.Call.StaticCallee() == nil {
+		params = append(params, funcPtr)
+	}
+	params = b.prependIndirectResult(instr.Call.Signature(), exported, params, "go.result")
+
+	paramBundle := b.emitPointerPack(params, instr.Pos())
 	var stackSize llvm.Value
 	callee := b.createGoroutineStartWrapper(funcType, funcPtr, prefix, hasContext, false, instr.Pos())
 	if b.AutomaticStackSize {
@@ -122,6 +127,15 @@ func (b *builder) createGo(instr *ssa.Go) {
 	b.createCall(fnType, start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.dataPtrType)}, "")
 }

+func (b *builder) getGoroutineCallArgument(value ssa.Value, exported bool) []llvm.Value {
+	typ := b.getLLVMType(value.Type())
+	arg := b.getCallArgument(value, exported)
+	if b.isIndirectParam(typ, exported) {
+		return []llvm.Value{b.copyToIndirectStorage(arg, typ, "go.param")}
+	}
+	return b.expandFormalParam(arg)
+}
+
 // Create an exported wrapper function for functions with the //go:wasmexport
 // pragma. This wrapper function is quite complex when the scheduler is enabled:
 // it needs to start a new goroutine each time the exported function is called.
@@ -139,7 +153,7 @@ func (b *builder) createWasmExport() {
 	// Declare the exported function.
 	paramTypes := b.llvmFnType.ParamTypes()
 	exportedFnType := llvm.FunctionType(b.llvmFnType.ReturnType(), paramTypes[:len(paramTypes)-1], false)
-	exportedFn := llvm.AddFunction(b.mod, b.fn.RelString(nil)+suffix, exportedFnType)
+	exportedFn := llvm.AddFunction(b.mod, b.getFunctionInfo(b.fn).linkName+suffix, exportedFnType)
 	b.addStandardAttributes(exportedFn)
 	llvmutil.AppendToGlobal(b.mod, "llvm.used", exportedFn)
 	exportedFn.AddFunctionAttr(b.ctx.CreateStringAttribute("wasm-export-name", b.info.wasmExport))
@@ -414,7 +428,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 			// Extract parameters from the state object, and call the function
 			// that's being wrapped.
 			var callParams []llvm.Value
-			for i := 0; i < numParams; i++ {
+			for i := range numParams {
 				gep := b.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
 					llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 					llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
@@ -146,13 +146,14 @@ func (b *builder) emitSVCall(args []ssa.Value, pos token.Pos) (llvm.Value, error
 	llvmArgs := []llvm.Value{}
 	argTypes := []llvm.Type{}
 	asm := "svc #" + strconv.FormatUint(num, 10)
-	constraints := "={r0}"
+	var constraints strings.Builder
+	constraints.WriteString("={r0}")
 	for i, arg := range args[1:] {
 		arg = arg.(*ssa.MakeInterface).X
 		if i == 0 {
-			constraints += ",0"
+			constraints.WriteString(",0")
 		} else {
-			constraints += ",{r" + strconv.Itoa(i) + "}"
+			constraints.WriteString(",{r" + strconv.Itoa(i) + "}")
 		}
 		llvmValue := b.getValue(arg, pos)
 		llvmArgs = append(llvmArgs, llvmValue)
@@ -161,9 +162,9 @@ func (b *builder) emitSVCall(args []ssa.Value, pos token.Pos) (llvm.Value, error
 	// Implement the ARM calling convention by marking r1-r3 as
 	// clobbered. r0 is used as an output register so doesn't have to be
 	// marked as clobbered.
-	constraints += ",~{r1},~{r2},~{r3}"
+	constraints.WriteString(",~{r1},~{r2},~{r3}")
 	fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
-	target := llvm.InlineAsm(fnType, asm, constraints, true, false, 0, false)
+	target := llvm.InlineAsm(fnType, asm, constraints.String(), true, false, 0, false)
 	return b.CreateCall(fnType, target, llvmArgs, ""), nil
 }

@@ -184,13 +185,14 @@ func (b *builder) emitSV64Call(args []ssa.Value, pos token.Pos) (llvm.Value, err
 	llvmArgs := []llvm.Value{}
 	argTypes := []llvm.Type{}
 	asm := "svc #" + strconv.FormatUint(num, 10)
-	constraints := "={x0}"
+	var constraints strings.Builder
+	constraints.WriteString("={x0}")
 	for i, arg := range args[1:] {
 		arg = arg.(*ssa.MakeInterface).X
 		if i == 0 {
-			constraints += ",0"
+			constraints.WriteString(",0")
 		} else {
-			constraints += ",{x" + strconv.Itoa(i) + "}"
+			constraints.WriteString(",{x" + strconv.Itoa(i) + "}")
 		}
 		llvmValue := b.getValue(arg, pos)
 		llvmArgs = append(llvmArgs, llvmValue)
@@ -199,9 +201,9 @@ func (b *builder) emitSV64Call(args []ssa.Value, pos token.Pos) (llvm.Value, err
 	// Implement the ARM64 calling convention by marking x1-x7 as
 	// clobbered. x0 is used as an output register so doesn't have to be
 	// marked as clobbered.
-	constraints += ",~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7}"
+	constraints.WriteString(",~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7}")
 	fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
-	target := llvm.InlineAsm(fnType, asm, constraints, true, false, 0, false)
+	target := llvm.InlineAsm(fnType, asm, constraints.String(), true, false, 0, false)
 	return b.CreateCall(fnType, target, llvmArgs, ""), nil
 }

@@ -10,6 +10,7 @@ import (
 	"fmt"
 	"go/token"
 	"go/types"
+	"path/filepath"
 	"sort"
 	"strconv"
 	"strings"
@@ -83,7 +84,11 @@ const (
 //
 // An interface value is a {typecode, value} tuple named runtime._interface.
 func (b *builder) createMakeInterface(val llvm.Value, typ types.Type, pos token.Pos) llvm.Value {
-	itfValue := b.emitPointerPack([]llvm.Value{val})
+	itfValue := b.emitPointerPack([]llvm.Value{val}, pos)
+	return b.createMakeInterfaceFromPointer(itfValue, typ)
+}
+
+func (b *builder) createMakeInterfaceFromPointer(itfValue llvm.Value, typ types.Type) llvm.Value {
 	itfType := b.getTypeCode(typ)
 	itf := llvm.Undef(b.getLLVMRuntimeType("_interface"))
 	itf = b.CreateInsertValue(itf, itfType, 0, "")
@@ -97,9 +102,16 @@ func (b *builder) createMakeInterface(val llvm.Value, typ types.Type, pos token.
 // doesn't match the underlying type of the interface.
 func (b *builder) extractValueFromInterface(itf llvm.Value, llvmType llvm.Type) llvm.Value {
 	valuePtr := b.CreateExtractValue(itf, 1, "typeassert.value.ptr")
+	if b.isIndirectAggregate(llvmType) {
+		return valuePtr
+	}
 	return b.emitPointerUnpack(valuePtr, []llvm.Type{llvmType})[0]
 }

+func (b *builder) extractValuePointerFromInterface(itf llvm.Value) llvm.Value {
+	return b.CreateExtractValue(itf, 1, "typeassert.value.ptr")
+}
+
 func (c *compilerContext) pkgPathPtr(pkgpath string) llvm.Value {
 	pkgpathName := "reflect/types.type.pkgpath.empty"
 	if pkgpath != "" {
@@ -124,6 +136,16 @@ func (c *compilerContext) pkgPathPtr(pkgpath string) llvm.Value {
 	return pkgPathPtr
 }

+// isGenericMethod returns true for a method that has its own type parameters
+// (independent of any type parameters on its receiver), e.g. the N method in
+// "func (r *Rand) N[Int intType](n Int) Int { ... }". Like the reflect
+// package, such methods are excluded from runtime method sets: they aren't
+// instantiated, so their signature can't be represented in a type code, and
+// they can never satisfy an interface method anyway.
+func isGenericMethod(fn *types.Func) bool {
+	return fn.Signature().TypeParams().Len() > 0
+}
+
 // getTypeCode returns a reference to a type code.
 // A type code is a pointer to a constant global that describes the type.
 // This function returns a pointer to the 'kind' field (which might not be the
@@ -133,22 +155,26 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 	typ = types.Unalias(typ)

 	ms := c.program.MethodSets.MethodSet(typ)
-	hasMethodSet := ms.Len() != 0
 	_, isInterface := typ.Underlying().(*types.Interface)
-	if isInterface {
-		hasMethodSet = false
-	}

 	// As defined in https://pkg.go.dev/reflect#Type:
 	// NumMethod returns the number of methods accessible using Method.
 	// For a non-interface type, it returns the number of exported methods.
 	// For an interface type, it returns the number of exported and unexported methods.
 	var numMethods int
-	for i := 0; i < ms.Len(); i++ {
-		if isInterface || ms.At(i).Obj().Exported() {
+	var hasMethodSet bool
+	for method := range ms.Methods() {
+		if isGenericMethod(method.Obj().(*types.Func)) {
+			continue
+		}
+		hasMethodSet = true
+		if isInterface || method.Obj().Exported() {
 			numMethods++
 		}
 	}
+	if isInterface {
+		hasMethodSet = false
+	}

 	// Short-circuit all the global pointer logic here for pointers to pointers.
 	if typ, ok := typ.(*types.Pointer); ok {
@@ -165,7 +191,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		}
 	}

-	typeCodeName, isLocal := getTypeCodeName(typ)
+	typeCodeName, isLocal := c.getTypeCodeName(typ)
 	globalName := "reflect/types.type:" + typeCodeName
 	var global llvm.Value
 	if isLocal {
@@ -192,8 +218,12 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		}
 		// Compute the method set value for types that support methods.
 		var methods []*types.Func
-		for i := 0; i < ms.Len(); i++ {
-			methods = append(methods, ms.At(i).Obj().(*types.Func))
+		for method := range ms.Methods() {
+			fn := method.Obj().(*types.Func)
+			if isGenericMethod(fn) {
+				continue
+			}
+			methods = append(methods, fn)
 		}
 		methodSetType := types.NewStruct([]*types.Var{
 			types.NewVar(token.NoPos, nil, "length", types.Typ[types.Uintptr]),
@@ -489,10 +519,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 				c.getTypeMethodSet(typ),
 			}, typeFields...)
 		}
-		alignment := c.targetData.TypeAllocSize(c.dataPtrType)
-		if alignment < 4 {
-			alignment = 4
-		}
+		alignment := max(c.targetData.TypeAllocSize(c.dataPtrType), 4)
 		globalValue := c.ctx.ConstStruct(typeFields, false)
 		global.SetInitializer(globalValue)
 		if isLocal {
@@ -580,20 +607,63 @@ var basicTypeNames = [...]string{
 // getTypeCodeName returns a name for this type that can be used in the
 // interface lowering pass to assign type codes as expected by the reflect
 // package. See getTypeCodeNum.
-func getTypeCodeName(t types.Type) (string, bool) {
+//
+// isLocal is true when the type is declared inside a function body.
+// Such types need a per-declaration (or per instantiation) suffix
+// because their printed names are not unique.
+//
+// Ordinary function-local types (TypeName.Parent() != nil) are
+// disambiguated lazily from their declaration position: every
+// declaration in the package has a distinct (file, line, column)
+// triple, and the position is taken un-//line-adjusted so it is
+// stable across builds. Such types are only nameable inside their
+// declaring package, so the name does not need to agree with anything
+// computed in another package.
+//
+// Synthetic locals (TypeName.Parent() == nil), produced by generic
+// instantiation, are pre-registered by scanLocalTypes because their
+// names must agree across packages that materialize the same instance.
+func (c *compilerContext) getTypeCodeName(t types.Type) (name string, isLocal bool) {
 	switch t := types.Unalias(t).(type) {
 	case *types.Named:
-		if t.Obj().Parent() != t.Obj().Pkg().Scope() {
-			return "named:" + t.String() + "$local", true
+		tn := t.Obj()
+		if tn.Pkg() == nil || tn.Parent() == tn.Pkg().Scope() {
+			name := tn.Name()
+			if tn.Pkg() != nil {
+				name = tn.Pkg().Path() + "." + name
+			}
+			isLocal := false
+			if targs := t.TypeArgs(); targs.Len() != 0 {
+				parts := make([]string, targs.Len())
+				for i := range parts {
+					var local bool
+					parts[i], local = c.getTypeCodeName(targs.At(i))
+					isLocal = isLocal || local
+				}
+				name += "[" + strings.Join(parts, ",") + "]"
+			}
+			return "named:" + name, isLocal
 		}
-		return "named:" + t.String(), false
+		if tn.Parent() != nil {
+			// Ordinary function-local type. Use the un-//line-adjusted
+			// declaration position as the disambiguator.
+			pos := c.program.Fset.PositionFor(tn.Pos(), false)
+			return fmt.Sprintf("named:%s$%s:%d:%d", t.String(), filepath.Base(pos.Filename), pos.Line, pos.Column), true
+		}
+		// Synthetic local from generic instantiation: must have been
+		// pre-registered by scanLocalTypes.
+		v := c.localTypeNames.At(t)
+		if v == nil {
+			panic("compiler: synthetic local type " + tn.Name() + " was not registered by scanLocalTypes")
+		}
+		return "named:" + v.(string), true
 	case *types.Array:
-		s, isLocal := getTypeCodeName(t.Elem())
+		s, isLocal := c.getTypeCodeName(t.Elem())
 		return "array:" + strconv.FormatInt(t.Len(), 10) + ":" + s, isLocal
 	case *types.Basic:
 		return "basic:" + basicTypeNames[t.Kind()], false
 	case *types.Chan:
-		s, isLocal := getTypeCodeName(t.Elem())
+		s, isLocal := c.getTypeCodeName(t.Elem())
 		var dir string
 		switch t.Dir() {
 		case types.SendOnly:
@@ -613,7 +683,7 @@ func getTypeCodeName(t types.Type) (string, bool) {
 			if !token.IsExported(name) {
 				name = t.Method(i).Pkg().Path() + "." + name
 			}
-			s, local := getTypeCodeName(t.Method(i).Type())
+			s, local := c.getTypeCodeName(t.Method(i).Type())
 			if local {
 				isLocal = true
 			}
@@ -621,17 +691,17 @@ func getTypeCodeName(t types.Type) (string, bool) {
 		}
 		return "interface:" + "{" + strings.Join(methods, ",") + "}", isLocal
 	case *types.Map:
-		keyType, keyLocal := getTypeCodeName(t.Key())
-		elemType, elemLocal := getTypeCodeName(t.Elem())
+		keyType, keyLocal := c.getTypeCodeName(t.Key())
+		elemType, elemLocal := c.getTypeCodeName(t.Elem())
 		return "map:" + "{" + keyType + "," + elemType + "}", keyLocal || elemLocal
 	case *types.Pointer:
-		s, isLocal := getTypeCodeName(t.Elem())
+		s, isLocal := c.getTypeCodeName(t.Elem())
 		return "pointer:" + s, isLocal
 	case *types.Signature:
 		isLocal := false
 		params := make([]string, t.Params().Len())
 		for i := 0; i < t.Params().Len(); i++ {
-			s, local := getTypeCodeName(t.Params().At(i).Type())
+			s, local := c.getTypeCodeName(t.Params().At(i).Type())
 			if local {
 				isLocal = true
 			}
@@ -639,7 +709,7 @@ func getTypeCodeName(t types.Type) (string, bool) {
 		}
 		results := make([]string, t.Results().Len())
 		for i := 0; i < t.Results().Len(); i++ {
-			s, local := getTypeCodeName(t.Results().At(i).Type())
+			s, local := c.getTypeCodeName(t.Results().At(i).Type())
 			if local {
 				isLocal = true
 			}
@@ -647,7 +717,7 @@ func getTypeCodeName(t types.Type) (string, bool) {
 		}
 		return "func:" + "{" + strings.Join(params, ",") + "}{" + strings.Join(results, ",") + "}", isLocal
 	case *types.Slice:
-		s, isLocal := getTypeCodeName(t.Elem())
+		s, isLocal := c.getTypeCodeName(t.Elem())
 		return "slice:" + s, isLocal
 	case *types.Struct:
 		elems := make([]string, t.NumFields())
@@ -657,7 +727,7 @@ func getTypeCodeName(t types.Type) (string, bool) {
 			if t.Field(i).Embedded() {
 				embedded = "#"
 			}
-			s, local := getTypeCodeName(t.Field(i).Type())
+			s, local := c.getTypeCodeName(t.Field(i).Type())
 			if local {
 				isLocal = true
 			}
@@ -672,18 +742,246 @@ func getTypeCodeName(t types.Type) (string, bool) {
 	}
 }

+// scanLocalTypes assigns names to every synthetic *types.TypeName
+// (TypeName.Parent() == nil) reachable from this package and stores
+// them in c.localTypeNames.
+//
+// Synthetic TypeNames are produced by generic instantiation: two
+// instantiations of the same generic function (e.g. F[int] and
+// F[string]) produce TypeNames with the same printed name and the
+// same source position, so each is named with the enclosing instance's
+// canonical function name as prefix. The function name encodes the type
+// arguments, matching Go's runtime behavior, where F[int].Inner and
+// F[string].Inner are distinct types even when Inner does not mention
+// the type parameter.
+//
+// A given instance may be materialized by several packages (the body
+// of F[int] is compiled in every package that calls F[int]); its
+// reflect/types.type:* global has LinkOnceODRLinkage and is merged by
+// name at link time. The chosen name therefore depends only on
+// intrinsic SSA properties (the canonical function name and raw token.Pos),
+// so any package compiling the same instance produces the same
+// identifier.
+//
+// Ordinary function-local TypeNames (TypeName.Parent() != nil) are
+// not handled here: they are nameable only inside their declaring
+// package, and getTypeCodeName derives a stable per-declaration name
+// for them directly from their source position.
+func (c *compilerContext) scanLocalTypes(ssaPkg *ssa.Package) {
+	// Locate every generic instance reachable from this package
+	// (including instances declared in imported packages and any
+	// function reached through an instance subtree).
+	var instances []*ssa.Function
+	seen := map[*ssa.Function]struct{}{}
+	var walk func(fn *ssa.Function, inInstance bool)
+	walk = func(fn *ssa.Function, inInstance bool) {
+		if fn == nil {
+			return
+		}
+		if _, ok := seen[fn]; ok {
+			return
+		}
+		// fn belongs to an instance subtree if it is itself an
+		// instantiation or if we reached it from one.
+		//
+		// len(TypeArgs()) is used instead of fn.Origin() because
+		// Origin() may call Build() on fn's declaring package, which
+		// would defeat per-package compilation.
+		isInstanceRoot := len(fn.TypeArgs()) > 0
+		if !isInstanceRoot && !inInstance && fn.Pkg != ssaPkg {
+			return
+		}
+		if fn.Blocks == nil && fn.AnonFuncs == nil {
+			return
+		}
+		seen[fn] = struct{}{}
+		isInInstance := inInstance || isInstanceRoot
+		if isInInstance {
+			instances = append(instances, fn)
+		}
+		for _, anon := range fn.AnonFuncs {
+			walk(anon, isInInstance)
+		}
+		var ops [10]*ssa.Value
+		for _, b := range fn.Blocks {
+			for _, instr := range b.Instrs {
+				for _, op := range instr.Operands(ops[:0]) {
+					if op == nil || *op == nil {
+						continue
+					}
+					if callee, ok := (*op).(*ssa.Function); ok {
+						walk(callee, isInInstance)
+					}
+				}
+			}
+		}
+	}
+	for _, member := range ssaPkg.Members {
+		switch m := member.(type) {
+		case *ssa.Function:
+			walk(m, false)
+		case *ssa.Type:
+			mset := c.program.MethodSets.MethodSet(m.Type())
+			for method := range mset.Methods() {
+				walk(c.program.MethodValue(method), false)
+			}
+			pmset := c.program.MethodSets.MethodSet(types.NewPointer(m.Type()))
+			for method := range pmset.Methods() {
+				walk(c.program.MethodValue(method), false)
+			}
+		}
+	}
+
+	// Registration is first-writer-wins (a synthetic TypeName may be
+	// reachable from several instances), so visit instances in a
+	// deterministic order. Pos() is a defensive tiebreaker.
+	sort.Slice(instances, func(i, j int) bool {
+		ri, rj := instances[i].RelString(nil), instances[j].RelString(nil)
+		if ri != rj {
+			return ri < rj
+		}
+		return instances[i].Pos() < instances[j].Pos()
+	})
+	for _, fn := range instances {
+		c.registerSyntheticLocalTypes(fn)
+	}
+}
+
+// registerSyntheticLocalTypes walks every type reachable from fn's
+// body and records each synthetic *types.Named (TypeName.Parent() ==
+// nil) in c.localTypeNames. Each is named with the canonical function
+// name plus a per-function counter assigned in source order.
+//
+// First-writer-wins: a *types.Named already present in
+// c.localTypeNames is left alone, so a synthetic type reachable from
+// several instances keeps the name assigned by the first (in
+// scanLocalTypes' deterministic order).
+func (c *compilerContext) registerSyntheticLocalTypes(fn *ssa.Function) {
+	var found []*types.Named
+	seen := map[types.Type]struct{}{}
+	var visit func(t types.Type)
+	visit = func(t types.Type) {
+		if t == nil {
+			return
+		}
+		if _, ok := seen[t]; ok {
+			return
+		}
+		seen[t] = struct{}{}
+		switch t := t.(type) {
+		case *types.Alias:
+			visit(types.Unalias(t))
+		case *types.Named:
+			tn := t.Obj()
+			if tn.Pkg() != nil && tn.Parent() == nil {
+				if c.localTypeNames.At(t) == nil {
+					// Reserve the slot so later calls within this
+					// scanLocalTypes invocation skip it; the final
+					// name is filled in after sorting, before any
+					// getTypeCodeName lookups happen.
+					c.localTypeNames.Set(t, "")
+					found = append(found, t)
+				}
+			}
+			targs := t.TypeArgs()
+			for t := range targs.Types() {
+				visit(t)
+			}
+			visit(t.Underlying())
+		case *types.Pointer:
+			visit(t.Elem())
+		case *types.Slice:
+			visit(t.Elem())
+		case *types.Array:
+			visit(t.Elem())
+		case *types.Chan:
+			visit(t.Elem())
+		case *types.Map:
+			visit(t.Key())
+			visit(t.Elem())
+		case *types.Struct:
+			for field := range t.Fields() {
+				visit(field.Type())
+			}
+		case *types.Signature:
+			if p := t.Params(); p != nil {
+				for v := range p.Variables() {
+					visit(v.Type())
+				}
+			}
+			if r := t.Results(); r != nil {
+				for v := range r.Variables() {
+					visit(v.Type())
+				}
+			}
+		case *types.Tuple:
+			for v := range t.Variables() {
+				visit(v.Type())
+			}
+		case *types.Interface:
+			// A synthetic local type can be reachable only through a
+			// local interface's method signature, so descend into
+			// them. getTypeCodeName encodes those signatures into
+			// the interface's identifier, and the seen map breaks
+			// cycles formed by methods that mention the interface
+			// itself.
+			for method := range t.Methods() {
+				visit(method.Type())
+			}
+		}
+	}
+	for _, p := range fn.Params {
+		visit(p.Type())
+	}
+	for _, fv := range fn.FreeVars {
+		visit(fv.Type())
+	}
+	for _, l := range fn.Locals {
+		visit(l.Type())
+	}
+	var ops [10]*ssa.Value
+	for _, b := range fn.Blocks {
+		for _, instr := range b.Instrs {
+			if v, ok := instr.(ssa.Value); ok {
+				visit(v.Type())
+			}
+			for _, op := range instr.Operands(ops[:0]) {
+				if op != nil && *op != nil {
+					visit((*op).Type())
+				}
+			}
+		}
+	}
+	if len(found) == 0 {
+		return
+	}
+	// Sort by raw token.Pos: this gives a total order on declarations
+	// that is stable across builds and unaffected by //line directives
+	// (which only adjust the human-facing position from Fset.Position).
+	sort.Slice(found, func(i, j int) bool {
+		return found[i].Obj().Pos() < found[j].Obj().Pos()
+	})
+	enclosing := c.canonicalFunctionName(fn)
+	for i, named := range found {
+		c.localTypeNames.Set(named, fmt.Sprintf("%s.%s$%d", enclosing, named.Obj().Name(), i))
+	}
+}
+
 // getTypeMethodSet returns a reference (GEP) to a global method set. This
 // method set should be unreferenced after the interface lowering pass.
 func (c *compilerContext) getTypeMethodSet(typ types.Type) llvm.Value {
-	globalName := typ.String() + "$methodset"
+	typeName, _ := c.getTypeCodeName(typ)
+	globalName := typeName + "$methodset"
 	global := c.mod.NamedGlobal(globalName)
 	if global.IsNil() {
 		ms := c.program.MethodSets.MethodSet(typ)

 		// Create method set.
 		var signatures, wrappers []llvm.Value
-		for i := 0; i < ms.Len(); i++ {
-			method := ms.At(i)
+		for method := range ms.Methods() {
+			if isGenericMethod(method.Obj().(*types.Func)) {
+				continue
+			}
 			signatureGlobal := c.getMethodSignature(method.Obj().(*types.Func))
 			signatures = append(signatures, signatureGlobal)
 			fn := c.program.MethodValue(method)
@@ -698,7 +996,7 @@ func (c *compilerContext) getTypeMethodSet(typ types.Type) llvm.Value {

 		// Construct global value.
 		globalValue := c.ctx.ConstStruct([]llvm.Value{
-			llvm.ConstInt(c.uintptrType, uint64(ms.Len()), false),
+			llvm.ConstInt(c.uintptrType, uint64(len(signatures)), false),
 			llvm.ConstArray(c.dataPtrType, signatures),
 			c.ctx.ConstStruct(wrappers, false),
 		}, false)
@@ -714,14 +1012,15 @@ func (c *compilerContext) getTypeMethodSet(typ types.Type) llvm.Value {
 // getMethodSignatureName returns a unique name (that can be used as the name of
 // a global) for the given method.
 func (c *compilerContext) getMethodSignatureName(method *types.Func) string {
-	signature := methodSignature(method)
-	var globalName string
+	name := method.Name()
+	var prefix string
 	if token.IsExported(method.Name()) {
-		globalName = "reflect/methods." + signature
+		prefix = "reflect/methods."
 	} else {
-		globalName = method.Type().(*types.Signature).Recv().Pkg().Path() + ".$methods." + signature
+		prefix = method.Type().(*types.Signature).Recv().Pkg().Path() + ".$methods."
 	}
-	return globalName
+	signature, _ := c.getTypeCodeName(method.Type())
+	return prefix + name + ":" + signature
 }

 // getMethodSignature returns a global variable which is a reference to an
@@ -769,7 +1068,7 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
 			commaOk = b.createInterfaceTypeAssert(intf, actualTypeNum)
 		}
 	} else {
-		name, _ := getTypeCodeName(expr.AssertedType)
+		name, _ := b.getTypeCodeName(expr.AssertedType)
 		globalName := "reflect/types.typeid:" + name
 		assertedTypeCodeGlobal := b.mod.NamedGlobal(globalName)
 		if assertedTypeCodeGlobal.IsNil() {
@@ -796,43 +1095,83 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {

 	prevBlock := b.GetInsertBlock()
 	okBlock := b.insertBasicBlock("typeassert.ok")
-	nextBlock := b.insertBasicBlock("typeassert.next")
-	b.currentBlockInfo.exit = nextBlock // adjust outgoing block for phi nodes
-	b.CreateCondBr(commaOk, okBlock, nextBlock)
-
-	// Retrieve the value from the interface if the type assert was
-	// successful.
-	b.SetInsertPointAtEnd(okBlock)
-	var valueOk llvm.Value
-	if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
-		// Type assert on interface type. Easy: just return the same
-		// interface value.
-		valueOk = itf
-	} else {
-		// Type assert on concrete type. Extract the underlying type from
-		// the interface (but only after checking it matches).
-		valueOk = b.extractValueFromInterface(itf, assertedType)
-	}
-	b.CreateBr(nextBlock)
-
-	// Continue after the if statement.
-	b.SetInsertPointAtEnd(nextBlock)
-	phi := b.CreatePHI(assertedType, "typeassert.value")
-	phi.AddIncoming([]llvm.Value{llvm.ConstNull(assertedType), valueOk}, []llvm.BasicBlock{prevBlock, okBlock})

 	if expr.CommaOk {
+		nextBlock := b.insertBasicBlock("typeassert.next")
+		b.currentBlockInfo.exit = nextBlock
+		if b.isIndirectAggregate(assertedType) {
+			resultType := b.getLLVMType(expr.Type())
+			result := b.createIndirectStorage(resultType, "typeassert.result")
+			b.zeroIndirectStorage(result, resultType)
+			b.CreateCondBr(commaOk, okBlock, nextBlock)
+
+			b.SetInsertPointAtEnd(okBlock)
+			valuePtr := b.extractValuePointerFromInterface(itf)
+			b.copyIndirectAggregate(b.CreateStructGEP(resultType, result, 0, ""), valuePtr, assertedType)
+			b.CreateBr(nextBlock)
+
+			b.SetInsertPointAtEnd(nextBlock)
+			b.CreateStore(commaOk, b.CreateStructGEP(resultType, result, 1, ""))
+			return result
+		}
+		b.CreateCondBr(commaOk, okBlock, nextBlock)
+
+		// Retrieve the value from the interface if the type assert was
+		// successful.
+		b.SetInsertPointAtEnd(okBlock)
+		var valueOk llvm.Value
+		if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
+			// Type assert on interface type. Easy: just return the same
+			// interface value.
+			valueOk = itf
+		} else {
+			// Type assert on concrete type. Extract the underlying type from
+			// the interface (but only after checking it matches).
+			valueOk = b.extractValueFromInterface(itf, assertedType)
+		}
+		b.CreateBr(nextBlock)
+
+		// Continue after the if statement.
+		b.SetInsertPointAtEnd(nextBlock)
+		phi := b.CreatePHI(assertedType, "typeassert.value")
+		phi.AddIncoming([]llvm.Value{llvm.ConstNull(assertedType), valueOk}, []llvm.BasicBlock{prevBlock, okBlock})
+
 		tuple := b.ctx.ConstStruct([]llvm.Value{llvm.Undef(assertedType), llvm.Undef(b.ctx.Int1Type())}, false) // create empty tuple
 		tuple = b.CreateInsertValue(tuple, phi, 0, "")                                                          // insert value
 		tuple = b.CreateInsertValue(tuple, commaOk, 1, "")                                                      // insert 'comma ok' boolean
 		return tuple
 	} else {
-		// This is kind of dirty as the branch above becomes mostly useless,
-		// but hopefully this gets optimized away.
-		b.createRuntimeCall("interfaceTypeAssert", []llvm.Value{commaOk}, "")
-		return phi
+		// Type assert without comma-ok. If it fails, panic.
+		faultBlock := b.getInterfaceAssertBlock()
+		b.currentBlockInfo.exit = okBlock
+		b.CreateCondBr(commaOk, okBlock, faultBlock)
+
+		// OK: extract the value from the interface.
+		b.SetInsertPointAtEnd(okBlock)
+		if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
+			return itf
+		}
+		return b.extractValueFromInterface(itf, assertedType)
 	}
 }

+func (b *builder) getInterfaceAssertBlock() llvm.BasicBlock {
+	if !b.interfaceAssertBlock.IsNil() {
+		return b.interfaceAssertBlock
+	}
+	savedBlock := b.GetInsertBlock()
+	block := b.ctx.AddBasicBlock(b.llvmFn, "typeassert.throw")
+	b.interfaceAssertBlock = block
+	b.SetInsertPointAtEnd(block)
+	if b.hasDeferFrame() {
+		b.createFaultCheckpoint()
+	}
+	b.createRuntimeCall("interfaceTypeAssert", []llvm.Value{llvm.ConstInt(b.ctx.Int1Type(), 0, false)}, "")
+	b.CreateUnreachable()
+	b.SetInsertPointAtEnd(savedBlock)
+	return block
+}
+
 // getMethodsString returns a string to be used in the "tinygo-methods" string
 // attribute for interface functions.
 func (c *compilerContext) getMethodsString(itf *types.Interface) string {
@@ -857,7 +1196,7 @@ func (c *compilerContext) getMethodSetValue(methods []*types.Func) llvm.Value {
 		if !token.IsExported(name) {
 			name = method.Pkg().Path() + "." + name
 		}
-		s, _ := getTypeCodeName(method.Type())
+		s, _ := c.getTypeCodeName(method.Type())
 		globalName := "reflect/types.signature:" + name + ":" + s
 		value := c.mod.NamedGlobal(globalName)
 		if value.IsNil() {
@@ -901,32 +1240,39 @@ func (c *compilerContext) getMethodSetValue(methods []*types.Func) llvm.Value {
 // thunk is declared, not defined: it will be defined by the interface lowering
 // pass.
 func (c *compilerContext) getInvokeFunction(instr *ssa.CallCommon) llvm.Value {
-	s, _ := getTypeCodeName(instr.Value.Type().Underlying())
-	fnName := s + "." + instr.Method.Name() + "$invoke"
+	fnName := c.getInvokeFunctionName(instr)
 	llvmFn := c.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
 		sig := instr.Method.Type().(*types.Signature)
 		var paramTuple []*types.Var
-		for i := 0; i < sig.Params().Len(); i++ {
-			paramTuple = append(paramTuple, sig.Params().At(i))
+		for v := range sig.Params().Variables() {
+			paramTuple = append(paramTuple, v)
 		}
 		paramTuple = append(paramTuple, types.NewVar(token.NoPos, nil, "$typecode", types.Typ[types.UnsafePointer]))
 		llvmFnType := c.getLLVMFunctionType(types.NewSignature(sig.Recv(), types.NewTuple(paramTuple...), sig.Results(), false))
 		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
 		c.addStandardDeclaredAttributes(llvmFn)
 		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-invoke", c.getMethodSignatureName(instr.Method)))
+		if _, indirect := c.hasIndirectResult(sig); indirect {
+			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-indirect-result", "true"))
+		}
 		methods := c.getMethodsString(instr.Value.Type().Underlying().(*types.Interface))
 		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-methods", methods))
 	}
 	return llvmFn
 }

+func (c *compilerContext) getInvokeFunctionName(instr *ssa.CallCommon) string {
+	s, _ := c.getTypeCodeName(instr.Value.Type().Underlying())
+	return s + "." + instr.Method.Name() + "$invoke"
+}
+
 // createInterfaceTypeAssert creates a call to a declared-but-not-defined
 // $typeassert function for the given interface. This function will be defined
 // by the interface lowering pass as a type-ID comparison chain, avoiding the
 // need for runtime.typeImplementsMethodSet at compile time.
 func (b *builder) createInterfaceTypeAssert(intf *types.Interface, actualType llvm.Value) llvm.Value {
-	s, _ := getTypeCodeName(intf)
+	s, _ := b.getTypeCodeName(intf)
 	fnName := s + ".$typeassert"
 	llvmFn := b.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
@@ -955,6 +1301,7 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFnType
 	// Get the expanded receiver type.
 	receiverType := c.getLLVMType(fn.Signature.Recv().Type())
 	var expandedReceiverType []llvm.Type
+	receiverIndirect := c.isIndirectAggregate(receiverType)
 	for _, info := range c.expandFormalParamType(receiverType, "", nil) {
 		expandedReceiverType = append(expandedReceiverType, info.llvmType)
 	}
@@ -969,7 +1316,13 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFnType
 	}

 	// create wrapper function
-	paramTypes := append([]llvm.Type{c.dataPtrType}, llvmFnType.ParamTypes()[len(expandedReceiverType):]...)
+	resultOffset := 0
+	if _, indirect := c.hasIndirectResult(fn.Signature); indirect {
+		resultOffset = 1
+	}
+	paramTypes := append([]llvm.Type{}, llvmFnType.ParamTypes()[:resultOffset]...)
+	paramTypes = append(paramTypes, c.dataPtrType)
+	paramTypes = append(paramTypes, llvmFnType.ParamTypes()[resultOffset+len(expandedReceiverType):]...)
 	wrapFnType := llvm.FunctionType(llvmFnType.ReturnType(), paramTypes, false)
 	wrapper = llvm.AddFunction(c.mod, wrapperName, wrapFnType)
 	c.addStandardAttributes(wrapper)
@@ -995,8 +1348,15 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFnType
 	block := b.ctx.AddBasicBlock(wrapper, "entry")
 	b.SetInsertPointAtEnd(block)

-	receiverValue := b.emitPointerUnpack(wrapper.Param(0), []llvm.Type{receiverType})[0]
-	params := append(b.expandFormalParam(receiverValue), wrapper.Params()[1:]...)
+	params := append([]llvm.Value{}, wrapper.Params()[:resultOffset]...)
+	receiverParam := wrapper.Param(resultOffset)
+	if receiverIndirect {
+		params = append(params, receiverParam)
+	} else {
+		receiverValue := b.emitPointerUnpack(receiverParam, []llvm.Type{receiverType})[0]
+		params = append(params, b.expandFormalParam(receiverValue)...)
+	}
+	params = append(params, wrapper.Params()[resultOffset+1:]...)
 	if llvmFnType.ReturnType().TypeKind() == llvm.VoidTypeKind {
 		b.CreateCall(llvmFnType, llvmFn, params, "")
 		b.CreateRetVoid()
@@ -1007,104 +1367,3 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFnType

 	return wrapper
 }
-
-// methodSignature creates a readable version of a method signature (including
-// the function name, excluding the receiver name). This string is used
-// internally to match interfaces and to call the correct method on an
-// interface. Examples:
-//
-//	String() string
-//	Read([]byte) (int, error)
-func methodSignature(method *types.Func) string {
-	return method.Name() + signature(method.Type().(*types.Signature))
-}
-
-// Make a readable version of a function (pointer) signature.
-// Examples:
-//
-//	() string
-//	(string, int) (int, error)
-func signature(sig *types.Signature) string {
-	s := ""
-	if sig.Params().Len() == 0 {
-		s += "()"
-	} else {
-		s += "("
-		for i := 0; i < sig.Params().Len(); i++ {
-			if i > 0 {
-				s += ", "
-			}
-			s += typestring(sig.Params().At(i).Type())
-		}
-		s += ")"
-	}
-	if sig.Results().Len() == 0 {
-		// keep as-is
-	} else if sig.Results().Len() == 1 {
-		s += " " + typestring(sig.Results().At(0).Type())
-	} else {
-		s += " ("
-		for i := 0; i < sig.Results().Len(); i++ {
-			if i > 0 {
-				s += ", "
-			}
-			s += typestring(sig.Results().At(i).Type())
-		}
-		s += ")"
-	}
-	return s
-}
-
-// typestring returns a stable (human-readable) type string for the given type
-// that can be used for interface equality checks. It is almost (but not
-// exactly) the same as calling t.String(). The main difference is some
-// normalization around `byte` vs `uint8` for example.
-func typestring(t types.Type) string {
-	// See: https://github.com/golang/go/blob/master/src/go/types/typestring.go
-	switch t := types.Unalias(t).(type) {
-	case *types.Array:
-		return "[" + strconv.FormatInt(t.Len(), 10) + "]" + typestring(t.Elem())
-	case *types.Basic:
-		return basicTypeNames[t.Kind()]
-	case *types.Chan:
-		switch t.Dir() {
-		case types.SendRecv:
-			return "chan (" + typestring(t.Elem()) + ")"
-		case types.SendOnly:
-			return "chan<- (" + typestring(t.Elem()) + ")"
-		case types.RecvOnly:
-			return "<-chan (" + typestring(t.Elem()) + ")"
-		default:
-			panic("unknown channel direction")
-		}
-	case *types.Interface:
-		methods := make([]string, t.NumMethods())
-		for i := range methods {
-			method := t.Method(i)
-			methods[i] = method.Name() + signature(method.Type().(*types.Signature))
-		}
-		return "interface{" + strings.Join(methods, ";") + "}"
-	case *types.Map:
-		return "map[" + typestring(t.Key()) + "]" + typestring(t.Elem())
-	case *types.Named:
-		return t.String()
-	case *types.Pointer:
-		return "*" + typestring(t.Elem())
-	case *types.Signature:
-		return "func" + signature(t)
-	case *types.Slice:
-		return "[]" + typestring(t.Elem())
-	case *types.Struct:
-		fields := make([]string, t.NumFields())
-		for i := range fields {
-			field := t.Field(i)
-			fields[i] = field.Name() + " " + typestring(field.Type())
-			if tag := t.Tag(i); tag != "" {
-				fields[i] += " " + strconv.Quote(tag)
-			}
-		}
-		return "struct{" + strings.Join(fields, ";") + "}"
-	default:
-		panic("unknown type: " + t.String())
-	}
-}
@@ -7,6 +7,7 @@ import (
 	"strconv"
 	"strings"

+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )

@@ -166,12 +167,22 @@ func (b *builder) createMachineKeepAliveImpl() {
 }

 var mathToLLVMMapping = map[string]string{
+	"math.Acos":  "llvm.acos.f64",
+	"math.Asin":  "llvm.asin.f64",
+	"math.Atan":  "llvm.atan.f64",
+	"math.Atan2": "llvm.atan2.f64",
 	"math.Ceil":  "llvm.ceil.f64",
+	"math.Cos":   "llvm.cos.f64",
+	"math.Cosh":  "llvm.cosh.f64",
 	"math.Exp":   "llvm.exp.f64",
 	"math.Exp2":  "llvm.exp2.f64",
 	"math.Floor": "llvm.floor.f64",
 	"math.Log":   "llvm.log.f64",
+	"math.Sin":   "llvm.sin.f64",
+	"math.Sinh":  "llvm.sinh.f64",
 	"math.Sqrt":  "llvm.sqrt.f64",
+	"math.Tan":   "llvm.tan.f64",
+	"math.Tanh":  "llvm.tanh.f64",
 	"math.Trunc": "llvm.trunc.f64",
 }

@@ -185,18 +196,40 @@ var mathToLLVMMapping = map[string]string{
 // float32(math.Sqrt(float64(v))) to a 32-bit floating point operation, which is
 // beneficial on architectures where 64-bit floating point operations are (much)
 // more expensive than 32-bit ones.
-func (b *builder) defineMathOp() {
-	b.createFunctionStart(true)
-	llvmName := mathToLLVMMapping[b.fn.RelString(nil)]
-	if llvmName == "" {
-		panic("unreachable: unknown math operation") // sanity check
+func (b *builder) defineMathOp() bool {
+	llvmName, ok := mathToLLVMMapping[b.fn.RelString(nil)]
+	if !ok {
+		return false
 	}
+	if strings.HasSuffix(b.Triple, "-wasi") || llvmutil.Version() < 19 {
+		// We don't have a real libc for wasip2. Until that is fixed, we need to
+		// limit math intrinsics on WASI to a subset supported natively in
+		// WebAssembly.
+		// Also, since we don't know the specific libc we will target, disallow
+		// these for all WASI targets.
+		//
+		// We also need to limit ourselves to LLVM 19 and above for the extended
+		// set of math intrinsics, see:
+		// https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294
+		switch b.fn.Name() {
+		case "Ceil", "Exp", "Exp2", "Floor", "Log", "Sqrt", "Trunc":
+		default:
+			return false
+		}
+	}
+	b.createFunctionStart(true)
 	llvmFn := b.mod.NamedFunction(llvmName)
 	if llvmFn.IsNil() {
 		// The intrinsic doesn't exist yet, so declare it.
-		// At the moment, all supported intrinsics have the form "double
-		// foo(double %x)" so we can hardcode the signature here.
-		llvmType := llvm.FunctionType(b.ctx.DoubleType(), []llvm.Type{b.ctx.DoubleType()}, false)
+		var llvmType llvm.Type
+		switch b.fn.Name() {
+		case "Atan2":
+			// double atan2(double %y, double %x)
+			llvmType = llvm.FunctionType(b.ctx.DoubleType(), []llvm.Type{b.ctx.DoubleType(), b.ctx.DoubleType()}, false)
+		default:
+			// double foo(double %x)
+			llvmType = llvm.FunctionType(b.ctx.DoubleType(), []llvm.Type{b.ctx.DoubleType()}, false)
+		}
 		llvmFn = llvm.AddFunction(b.mod, llvmName, llvmType)
 	}
 	// Create a call to the intrinsic.
@@ -206,6 +239,7 @@ func (b *builder) defineMathOp() {
 	}
 	result := b.CreateCall(llvmFn.GlobalValueType(), llvmFn, args, "")
 	b.CreateRet(result)
+	return true
 }

 func (b *builder) defineCryptoIntrinsic() bool {
@@ -54,7 +54,7 @@ func (b *builder) emitLifetimeEnd(ptr, size llvm.Value) {
 // pointer value directly. It returns the pointer with the packed data.
 // If the values are all constants, they are be stored in a constant global and
 // deduplicated.
-func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
+func (b *builder) emitPointerPack(values []llvm.Value, pos token.Pos) llvm.Value {
 	valueTypes := make([]llvm.Type, len(values))
 	for i, value := range values {
 		valueTypes[i] = value.Type()
@@ -128,14 +128,9 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {

 		// Packed data is bigger than a pointer, so allocate it on the heap.
 		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
+		layoutValue := b.createObjectLayout(packedType, pos)
 		align := b.targetData.ABITypeAlignment(packedType)
-		alloc := b.mod.NamedFunction("runtime.alloc")
-		packedAlloc := b.CreateCall(alloc.GlobalValueType(), alloc, []llvm.Value{
-			sizeValue,
-			llvm.ConstNull(b.dataPtrType),
-			llvm.Undef(b.dataPtrType), // unused context parameter
-		}, "")
-		packedAlloc.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align)))
+		packedAlloc := b.createAlloc(sizeValue, layoutValue, align, "")
 		if b.NeedsStackObjects {
 			b.trackPointer(packedAlloc)
 		}
@@ -212,7 +207,7 @@ func (c *compilerContext) makeGlobalArray(buf []byte, name string, elementType l
 	globalType := llvm.ArrayType(elementType, len(buf))
 	global := llvm.AddGlobal(c.mod, globalType, name)
 	value := llvm.Undef(globalType)
-	for i := 0; i < len(buf); i++ {
+	for i := range buf {
 		ch := uint64(buf[i])
 		value = c.builder.CreateInsertValue(value, llvm.ConstInt(elementType, ch, false), i, "")
 	}
@@ -396,7 +391,7 @@ func (c *compilerContext) buildPointerBitmap(
 			return
 		}
 		elementSize /= ptrAlign
-		for i := 0; i < len; i++ {
+		for i := range len {
 			c.buildPointerBitmap(
 				dst,
 				ptrAlign,
@@ -423,7 +418,7 @@ func (c *compilerContext) archFamily() string {
 // features string is not one for an ARM architecture.
 func (c *compilerContext) isThumb() bool {
 	var isThumb, isNotThumb bool
-	for _, feature := range strings.Split(c.Features, ",") {
+	for feature := range strings.SplitSeq(c.Features, ",") {
 		if feature == "+thumb-mode" {
 			isThumb = true
 		}
@@ -44,7 +44,7 @@ func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, n
 	alloca = CreateEntryBlockAlloca(builder, t, name)
 	size = llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
 	fnType, fn := getLifetimeStartFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
+	builder.CreateCall(fnType, fn, lifetimeCallArgs(size, alloca), "")
 	return
 }

@@ -58,14 +58,14 @@ func CreateInstructionAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type,
 	builder.SetInsertPointBefore(inst)
 	size := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
 	fnType, fn := getLifetimeStartFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
+	builder.CreateCall(fnType, fn, lifetimeCallArgs(size, alloca), "")
 	if next := llvm.NextInstruction(inst); !next.IsNil() {
 		builder.SetInsertPointBefore(next)
 	} else {
 		builder.SetInsertPointAtEnd(inst.InstructionParent())
 	}
 	fnType, fn = getLifetimeEndFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
+	builder.CreateCall(fnType, fn, lifetimeCallArgs(size, alloca), "")
 	return alloca
 }

@@ -74,7 +74,27 @@ func CreateInstructionAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type,
 // createTemporaryAlloca.
 func EmitLifetimeEnd(builder llvm.Builder, mod llvm.Module, ptr, size llvm.Value) {
 	fnType, fn := getLifetimeEndFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, ptr}, "")
+	builder.CreateCall(fnType, fn, lifetimeCallArgs(size, ptr), "")
+}
+
+// lifetimeCallArgs returns the arguments to pass to a call of the
+// llvm.lifetime.start/end intrinsics. LLVM 22 dropped the (redundant,
+// already required to match the alloca size) i64 size argument, so the
+// intrinsic now only takes the pointer.
+func lifetimeCallArgs(size, ptr llvm.Value) []llvm.Value {
+	if Version() >= 22 {
+		return []llvm.Value{ptr}
+	}
+	return []llvm.Value{size, ptr}
+}
+
+// lifetimeFuncType returns the function type of the llvm.lifetime.start/end
+// intrinsics, which lost their i64 size parameter in LLVM 22.
+func lifetimeFuncType(ctx llvm.Context, ptrType llvm.Type) llvm.Type {
+	if Version() >= 22 {
+		return llvm.FunctionType(ctx.VoidType(), []llvm.Type{ptrType}, false)
+	}
+	return llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
 }

 // getLifetimeStartFunc returns the llvm.lifetime.start intrinsic and creates it
@@ -84,7 +104,7 @@ func getLifetimeStartFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 	fn := mod.NamedFunction(fnName)
 	ctx := mod.Context()
 	ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
+	fnType := lifetimeFuncType(ctx, ptrType)
 	if fn.IsNil() {
 		fn = llvm.AddFunction(mod, fnName, fnType)
 	}
@@ -98,7 +118,7 @@ func getLifetimeEndFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 	fn := mod.NamedFunction(fnName)
 	ctx := mod.Context()
 	ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
+	fnType := lifetimeFuncType(ctx, ptrType)
 	if fn.IsNil() {
 		fn = llvm.AddFunction(mod, fnName, fnType)
 	}
@@ -218,6 +238,36 @@ func Version() int {
 	return major
 }

+// NoCaptureAttrName returns the name of the LLVM attribute kind that marks a
+// pointer parameter as guaranteed not to escape by capture.
+//
+// LLVM 21 removed the old boolean 'nocapture' enum attribute in favor of the
+// more expressive 'captures' int attribute, where 'captures(none)' (encoded
+// as the value 0) is the equivalent of the old 'nocapture'. LLVM 20 supports
+// both, but its own optimizer still emits 'nocapture', so the cutoff for
+// reading/writing the new name is LLVM 21.
+func NoCaptureAttrName() string {
+	if Version() >= 21 {
+		return "captures"
+	}
+	return "nocapture"
+}
+
+// IsNoCapture reports whether attr (looked up using the kind returned by
+// NoCaptureAttrName) indicates that the pointer it is attached to does not
+// escape by capture. It returns false for a nil attribute.
+func IsNoCapture(attr llvm.Attribute) bool {
+	if attr.IsNil() {
+		return false
+	}
+	if Version() >= 21 {
+		// captures(none) is encoded as the value 0; any other value permits
+		// some form of capture.
+		return attr.GetEnumValue() == 0
+	}
+	return true
+}
+
 // ByteOrder returns the byte order for the given target triple. Most targets are little
 // endian, but for example MIPS can be big-endian.
 func ByteOrder(target string) binary.ByteOrder {
@@ -3,42 +3,29 @@ package compiler
 // This file emits the correct map intrinsics for map operations.

 import (
+	"fmt"
 	"go/token"
 	"go/types"
-
-	"github.com/tinygo-org/tinygo/src/tinygo"
 	"golang.org/x/tools/go/ssa"
+	"strings"
 	"tinygo.org/x/go-llvm"
 )

+const hashArrayUnrollLimit = 4
+
 // createMakeMap creates a new map object (runtime.hashmap) by allocating and
 // initializing an appropriately sized object.
 func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
 	mapType := expr.Type().Underlying().(*types.Map)
 	keyType := mapType.Key().Underlying()
 	llvmValueType := b.getLLVMType(mapType.Elem().Underlying())
-	var llvmKeyType llvm.Type
-	var alg uint64
-	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
-		// String keys.
-		llvmKeyType = b.getLLVMType(keyType)
-		alg = uint64(tinygo.HashmapAlgorithmString)
-	} else if hashmapIsBinaryKey(keyType) {
-		// Trivially comparable keys.
-		llvmKeyType = b.getLLVMType(keyType)
-		alg = uint64(tinygo.HashmapAlgorithmBinary)
-	} else {
-		// All other keys. Implemented as map[interface{}]valueType for ease of
-		// implementation.
-		llvmKeyType = b.getLLVMRuntimeType("_interface")
-		alg = uint64(tinygo.HashmapAlgorithmInterface)
-	}
+	llvmKeyType := b.getLLVMType(keyType)
+
 	keySize := b.targetData.TypeAllocSize(llvmKeyType)
 	valueSize := b.targetData.TypeAllocSize(llvmValueType)
 	llvmKeySize := llvm.ConstInt(b.uintptrType, keySize, false)
 	llvmValueSize := llvm.ConstInt(b.uintptrType, valueSize, false)
 	sizeHint := llvm.ConstInt(b.uintptrType, 8, false)
-	algEnum := llvm.ConstInt(b.ctx.Int8Type(), alg, false)
 	if expr.Reserve != nil {
 		sizeHint = b.getValue(expr.Reserve, getPos(expr))
 		var err error
@@ -47,134 +34,129 @@ func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
 			return llvm.Value{}, err
 		}
 	}
-	hashmap := b.createRuntimeCall("hashmapMake", []llvm.Value{llvmKeySize, llvmValueSize, sizeHint, algEnum}, "")
+
+	// Resolve hash and equal functions for this key type. For string and
+	// binary key types, reference the corresponding runtime functions
+	// directly. For composite types, generate type-specific functions.
+	var hashFn, equalFn llvm.Value
+	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
+		hashFn = b.getRuntimeFunctionValue("hashmapStringPtrHash", hashmapKeyHashSignature())
+		equalFn = b.getRuntimeFunctionValue("hashmapStringEqual", hashmapKeyEqualSignature())
+	} else if hashmapIsBinaryKey(keyType) {
+		hashFn = b.getRuntimeFunctionValue("hash32", hashmapKeyHashSignature())
+		equalFn = b.getRuntimeFunctionValue("memequal", hashmapKeyEqualSignature())
+	} else {
+		fn := b.getOrGenerateKeyHashFunc(keyType)
+		hashFn = b.createFuncValue(fn, llvm.ConstNull(b.dataPtrType), hashmapKeyHashSignature())
+		fn = b.getOrGenerateKeyEqualFunc(keyType)
+		equalFn = b.createFuncValue(fn, llvm.ConstNull(b.dataPtrType), hashmapKeyEqualSignature())
+	}
+
+	hashmap := b.createRuntimeCall("hashmapMakeGeneric", []llvm.Value{
+		llvmKeySize, llvmValueSize, sizeHint,
+		hashFn, equalFn,
+	}, "")
 	return hashmap, nil
 }

+// getRuntimeFunctionValue returns a TinyGo function value (with nil context)
+// for the named runtime function.
+func (b *builder) getRuntimeFunctionValue(name string, sig *types.Signature) llvm.Value {
+	member := b.program.ImportedPackage("runtime").Members[name]
+	if member == nil {
+		panic("unknown runtime function: " + name)
+	}
+	_, llvmFn := b.getFunction(member.(*ssa.Function))
+	return b.createFuncValue(llvmFn, llvm.ConstNull(b.dataPtrType), sig)
+}
+
 // createMapLookup returns the value in a map. It calls a runtime function
 // depending on the map key type to load the map value and its comma-ok value.
-func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
+func (b *builder) createMapLookup(keyType, valueType types.Type, m llvm.Value, key ssa.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
 	llvmValueType := b.getLLVMType(valueType)

 	// Allocate the memory for the resulting type. Do not zero this memory: it
 	// will be zeroed by the hashmap get implementation if the key is not
 	// present in the map.
-	mapValueAlloca, mapValueAllocaSize := b.createTemporaryAlloca(llvmValueType, "hashmap.value")
+	result := b.createRuntimeValueResult(llvmValueType, commaOk, false, "hashmap")
+	mapValueAlloca := result.valuePtr

 	// We need the map size (with type uintptr) to pass to the hashmap*Get
 	// functions. This is necessary because those *Get functions are valid on
 	// nil maps, and they'll need to zero the value pointer by that number of
 	// bytes.
-	mapValueSize := mapValueAllocaSize
+	mapValueSize := result.valueSize
 	if mapValueSize.Type().IntTypeWidth() > b.uintptrType.IntTypeWidth() {
 		mapValueSize = llvm.ConstTrunc(mapValueSize, b.uintptrType)
 	}

 	// Do the lookup. How it is done depends on the key type.
 	var commaOkValue llvm.Value
-	origKeyType := keyType
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, mapValueAlloca, mapValueSize}
+		params := []llvm.Value{m, b.getValue(key, getPos(key)), mapValueAlloca, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapStringGet", params, "")
-	} else if hashmapIsBinaryKey(keyType) {
-		// key can be compared with runtime.memequal
-		// Store the key in an alloca, in the entry block to avoid dynamic stack
-		// growth.
-		mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
-		b.CreateStore(key, mapKeyAlloca)
-		b.zeroUndefBytes(b.getLLVMType(keyType), mapKeyAlloca)
-		// Fetch the value from the hashmap.
-		params := []llvm.Value{m, mapKeyAlloca, mapValueAlloca, mapValueSize}
-		commaOkValue = b.createRuntimeCall("hashmapBinaryGet", params, "")
-		b.emitLifetimeEnd(mapKeyAlloca, mapKeySize)
 	} else {
-		// Not trivially comparable using memcmp. Make it an interface instead.
-		itfKey := key
-		if _, ok := keyType.(*types.Interface); !ok {
-			// Not already an interface, so convert it to an interface now.
-			itfKey = b.createMakeInterface(key, origKeyType, pos)
+		// Key stored at actual type: either binary-comparable or with
+		// compiler-generated hash/equal.
+		mapKey := b.getValueStorage(key, "hashmap.key")
+		params := []llvm.Value{m, mapKey.ptr, mapValueAlloca, mapValueSize}
+		fnName := "hashmapBinaryGet"
+		if !hashmapIsBinaryKey(keyType) {
+			fnName = "hashmapGenericGet"
 		}
-		params := []llvm.Value{m, itfKey, mapValueAlloca, mapValueSize}
-		commaOkValue = b.createRuntimeCall("hashmapInterfaceGet", params, "")
+		commaOkValue = b.createRuntimeCall(fnName, params, "")
+		b.endValueStorage(mapKey)
 	}

-	// Load the resulting value from the hashmap. The value is set to the zero
-	// value if the key doesn't exist in the hashmap.
-	mapValue := b.CreateLoad(llvmValueType, mapValueAlloca, "")
-	b.emitLifetimeEnd(mapValueAlloca, mapValueAllocaSize)
-
-	if commaOk {
-		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{llvmValueType, b.ctx.Int1Type()}, false))
-		tuple = b.CreateInsertValue(tuple, mapValue, 0, "")
-		tuple = b.CreateInsertValue(tuple, commaOkValue, 1, "")
-		return tuple, nil
-	} else {
-		return mapValue, nil
-	}
+	// The value is set to the zero value if the key doesn't exist.
+	return result.finish(b, commaOkValue, ""), nil
 }

 // createMapUpdate updates a map key to a given value, by creating an
 // appropriate runtime call.
-func (b *builder) createMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
-	valueAlloca, valueSize := b.createTemporaryAlloca(value.Type(), "hashmap.value")
-	b.CreateStore(value, valueAlloca)
-	origKeyType := keyType
+func (b *builder) createMapUpdate(keyType types.Type, m llvm.Value, key, value ssa.Value, pos token.Pos) {
+	storedValue := b.getValueStorage(value, "hashmap.value")
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, valueAlloca}
-		b.createRuntimeCall("hashmapStringSet", params, "")
-	} else if hashmapIsBinaryKey(keyType) {
-		// key can be compared with runtime.memequal
-		keyAlloca, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
-		b.CreateStore(key, keyAlloca)
-		b.zeroUndefBytes(b.getLLVMType(keyType), keyAlloca)
-		params := []llvm.Value{m, keyAlloca, valueAlloca}
-		b.createRuntimeCall("hashmapBinarySet", params, "")
-		b.emitLifetimeEnd(keyAlloca, keySize)
+		params := []llvm.Value{m, b.getValue(key, getPos(key)), storedValue.ptr}
+		b.createRuntimeInvoke("hashmapStringSet", params, "")
 	} else {
-		// Key is not trivially comparable, so compare it as an interface instead.
-		itfKey := key
-		if _, ok := keyType.(*types.Interface); !ok {
-			// Not already an interface, so convert it to an interface first.
-			itfKey = b.createMakeInterface(key, origKeyType, pos)
+		// Key stored at actual type.
+		keyStorage := b.getValueStorage(key, "hashmap.key")
+		fnName := "hashmapBinarySet"
+		if !hashmapIsBinaryKey(keyType) {
+			fnName = "hashmapGenericSet"
 		}
-		params := []llvm.Value{m, itfKey, valueAlloca}
-		b.createRuntimeCall("hashmapInterfaceSet", params, "")
+		params := []llvm.Value{m, keyStorage.ptr, storedValue.ptr}
+		b.createRuntimeInvoke(fnName, params, "")
+		b.endValueStorage(keyStorage)
 	}
-	b.emitLifetimeEnd(valueAlloca, valueSize)
+	b.endValueStorage(storedValue)
 }

 // createMapDelete deletes a key from a map by calling the appropriate runtime
 // function. It is the implementation of the Go delete() builtin.
 func (b *builder) createMapDelete(keyType types.Type, m, key llvm.Value, pos token.Pos) error {
-	origKeyType := keyType
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
 		params := []llvm.Value{m, key}
 		b.createRuntimeCall("hashmapStringDelete", params, "")
 		return nil
-	} else if hashmapIsBinaryKey(keyType) {
+	} else {
+		// Key stored at actual type.
 		keyAlloca, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, keyAlloca)
-		b.zeroUndefBytes(b.getLLVMType(keyType), keyAlloca)
-		params := []llvm.Value{m, keyAlloca}
-		b.createRuntimeCall("hashmapBinaryDelete", params, "")
-		b.emitLifetimeEnd(keyAlloca, keySize)
-		return nil
-	} else {
-		// Key is not trivially comparable, so compare it as an interface
-		// instead.
-		itfKey := key
-		if _, ok := keyType.(*types.Interface); !ok {
-			// Not already an interface, so convert it to an interface first.
-			itfKey = b.createMakeInterface(key, origKeyType, pos)
+		fnName := "hashmapBinaryDelete"
+		if !hashmapIsBinaryKey(keyType) {
+			fnName = "hashmapGenericDelete"
 		}
-		params := []llvm.Value{m, itfKey}
-		b.createRuntimeCall("hashmapInterfaceDelete", params, "")
+		params := []llvm.Value{m, keyAlloca}
+		b.createRuntimeCall(fnName, params, "")
+		b.emitLifetimeEnd(keyAlloca, keySize)
 		return nil
 	}
 }
@@ -195,42 +177,15 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 	llvmKeyType := b.getLLVMType(keyType)
 	llvmValueType := b.getLLVMType(valueType)

-	// There is a special case in which keys are stored as an interface value
-	// instead of the value they normally are. This happens for non-trivially
-	// comparable types such as float32 or some structs.
-	isKeyStoredAsInterface := false
-	if t, ok := keyType.Underlying().(*types.Basic); ok && t.Info()&types.IsString != 0 {
-		// key is a string
-	} else if hashmapIsBinaryKey(keyType) {
-		// key can be compared with runtime.memequal
-	} else {
-		// The key is stored as an interface value, and may or may not be an
-		// interface type (for example, float32 keys are stored as an interface
-		// value).
-		if _, ok := keyType.Underlying().(*types.Interface); !ok {
-			isKeyStoredAsInterface = true
-		}
-	}
-
-	// Determine the type of the key as stored in the map.
-	llvmStoredKeyType := llvmKeyType
-	if isKeyStoredAsInterface {
-		llvmStoredKeyType = b.getLLVMRuntimeType("_interface")
-	}
+	// All key types are now stored at their declared type (no interface wrapping).

 	// Extract the key and value from the map.
-	mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(llvmStoredKeyType, "range.key")
+	mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(llvmKeyType, "range.key")
 	mapValueAlloca, mapValueSize := b.createTemporaryAlloca(llvmValueType, "range.value")
 	ok := b.createRuntimeCall("hashmapNext", []llvm.Value{llvmRangeVal, it, mapKeyAlloca, mapValueAlloca}, "range.next")
-	mapKey := b.CreateLoad(llvmStoredKeyType, mapKeyAlloca, "")
+	mapKey := b.CreateLoad(llvmKeyType, mapKeyAlloca, "")
 	mapValue := b.CreateLoad(llvmValueType, mapValueAlloca, "")

-	if isKeyStoredAsInterface {
-		// The key is stored as an interface but it isn't of interface type.
-		// Extract the underlying value.
-		mapKey = b.extractValueFromInterface(mapKey, llvmKeyType)
-	}
-
 	// End the lifetimes of the allocas, because we're done with them.
 	b.emitLifetimeEnd(mapKeyAlloca, mapKeySize)
 	b.emitLifetimeEnd(mapValueAlloca, mapValueSize)
@@ -250,20 +205,9 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 func hashmapIsBinaryKey(keyType types.Type) bool {
 	switch keyType := keyType.Underlying().(type) {
 	case *types.Basic:
-		// TODO: unsafe.Pointer is also a binary key, but to support that we
-		// need to fix an issue with interp first (see
-		// https://github.com/tinygo-org/tinygo/pull/4898).
-		return keyType.Info()&(types.IsBoolean|types.IsInteger) != 0
+		return keyType.Info()&(types.IsBoolean|types.IsInteger) != 0 || keyType.Kind() == types.UnsafePointer
 	case *types.Pointer:
 		return true
-	case *types.Struct:
-		for i := 0; i < keyType.NumFields(); i++ {
-			fieldType := keyType.Field(i).Type().Underlying()
-			if !hashmapIsBinaryKey(fieldType) {
-				return false
-			}
-		}
-		return true
 	case *types.Array:
 		return hashmapIsBinaryKey(keyType.Elem())
 	default:
@@ -271,68 +215,400 @@ func hashmapIsBinaryKey(keyType types.Type) bool {
 	}
 }

-func (b *builder) zeroUndefBytes(llvmType llvm.Type, ptr llvm.Value) error {
-	// We know that hashmapIsBinaryKey is true, so we only have to handle those types that can show up there.
-	// To zero all undefined bytes, we iterate over all the fields in the type.  For each element, compute the
-	// offset of that element.  If it's Basic type, there are no internal padding bytes.  For compound types, we recurse to ensure
-	// we handle nested types.  Next, we determine if there are any padding bytes before the next
-	// element and zero those as well.
+// hashmapKeyHashSignature returns the Go type signature for hashmap key hash
+// functions: func(key unsafe.Pointer, size, seed uintptr) uint32
+func hashmapKeyHashSignature() *types.Signature {
+	return types.NewSignatureType(nil, nil, nil,
+		types.NewTuple(
+			types.NewVar(token.NoPos, nil, "key", types.Typ[types.UnsafePointer]),
+			types.NewVar(token.NoPos, nil, "size", types.Typ[types.Uintptr]),
+			types.NewVar(token.NoPos, nil, "seed", types.Typ[types.Uintptr]),
+		),
+		types.NewTuple(
+			types.NewVar(token.NoPos, nil, "", types.Typ[types.Uint32]),
+		),
+		false,
+	)
+}

-	zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+// hashmapKeyEqualSignature returns the Go type signature for hashmap key equal
+// functions: func(x, y unsafe.Pointer, n uintptr) bool
+func hashmapKeyEqualSignature() *types.Signature {
+	return types.NewSignatureType(nil, nil, nil,
+		types.NewTuple(
+			types.NewVar(token.NoPos, nil, "x", types.Typ[types.UnsafePointer]),
+			types.NewVar(token.NoPos, nil, "y", types.Typ[types.UnsafePointer]),
+			types.NewVar(token.NoPos, nil, "n", types.Typ[types.Uintptr]),
+		),
+		types.NewTuple(
+			types.NewVar(token.NoPos, nil, "", types.Typ[types.Bool]),
+		),
+		false,
+	)
+}

-	switch llvmType.TypeKind() {
-	case llvm.IntegerTypeKind:
-		// no padding bytes
-		return nil
-	case llvm.PointerTypeKind:
-		// mo padding bytes
-		return nil
-	case llvm.ArrayTypeKind:
-		llvmArrayType := llvmType
-		llvmElemType := llvmType.ElementType()
+// hashmapKeyFuncName returns a canonical name for a generated hash or equal
+// function based on the key type's underlying structure. Named types are
+// replaced with their underlying types so that structurally identical key
+// types (e.g., struct{i1; str1} and struct{i2; str2} where both i1, i2 are
+// int and str1, str2 are string) share the same generated function.
+func hashmapKeyFuncName(prefix string, keyType types.Type) string {
+	return prefix + "." + hashmapCanonicalTypeName(keyType)
+}

-		for i := 0; i < llvmArrayType.ArrayLength(); i++ {
-			idx := llvm.ConstInt(b.uintptrType, uint64(i), false)
-			elemPtr := b.CreateInBoundsGEP(llvmArrayType, ptr, []llvm.Value{zero, idx}, "")
-
-			// zero any padding bytes in this element
-			b.zeroUndefBytes(llvmElemType, elemPtr)
+// hashmapCanonicalTypeName returns a string representation of the hash/equal
+// operations needed for a type, stripping named types where the operation does
+// not depend on the name. Pointer and channel names do not include the element
+// type because their hash/equal operations only use the pointer word.
+func hashmapCanonicalTypeName(t types.Type) string {
+	switch t := t.Underlying().(type) {
+	case *types.Basic:
+		return t.Name()
+	case *types.Pointer:
+		return "*"
+	case *types.Chan:
+		switch t.Dir() {
+		case types.SendRecv:
+			return "chan"
+		case types.SendOnly:
+			return "chan<-"
+		case types.RecvOnly:
+			return "<-chan"
 		}
-
-	case llvm.StructTypeKind:
-		llvmStructType := llvmType
-		numFields := llvmStructType.StructElementTypesCount()
-		llvmElementTypes := llvmStructType.StructElementTypes()
-
-		for i := 0; i < numFields; i++ {
-			idx := llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)
-			elemPtr := b.CreateInBoundsGEP(llvmStructType, ptr, []llvm.Value{zero, idx}, "")
-
-			// zero any padding bytes in this field
-			llvmElemType := llvmElementTypes[i]
-			b.zeroUndefBytes(llvmElemType, elemPtr)
-
-			// zero any padding bytes before the next field, if any
-			offset := b.targetData.ElementOffset(llvmStructType, i)
-			storeSize := b.targetData.TypeStoreSize(llvmElemType)
-			fieldEndOffset := offset + storeSize
-
-			var nextOffset uint64
-			if i < numFields-1 {
-				nextOffset = b.targetData.ElementOffset(llvmStructType, i+1)
-			} else {
-				// Last field?  Next offset is the total size of the allocate struct.
-				nextOffset = b.targetData.TypeAllocSize(llvmStructType)
-			}
-
-			if fieldEndOffset != nextOffset {
-				n := llvm.ConstInt(b.uintptrType, nextOffset-fieldEndOffset, false)
-				llvmStoreSize := llvm.ConstInt(b.uintptrType, storeSize, false)
-				paddingStart := b.CreateInBoundsGEP(b.ctx.Int8Type(), elemPtr, []llvm.Value{llvmStoreSize}, "")
-				b.createRuntimeCall("memzero", []llvm.Value{paddingStart, n}, "")
-			}
+	case *types.Interface:
+		if t.NumMethods() == 0 {
+			return "interface{}"
 		}
+		return t.String()
+	case *types.Struct:
+		var s strings.Builder
+		s.WriteString("struct{")
+		for i := 0; i < t.NumFields(); i++ {
+			if i > 0 {
+				s.WriteString("; ")
+			}
+			s.WriteString(hashmapCanonicalTypeName(t.Field(i).Type()))
+		}
+		return s.String() + "}"
+	case *types.Array:
+		return fmt.Sprintf("[%d]%s", t.Len(), hashmapCanonicalTypeName(t.Elem()))
+	}
+	return t.String()
+}
+
+// getOrGenerateKeyHashFunc returns an LLVM function that computes the hash
+// of a key of the given type. The function is generated on first call and
+// cached in the module.
+func (b *builder) getOrGenerateKeyHashFunc(keyType types.Type) llvm.Value {
+	name := hashmapKeyFuncName("hashmapKeyHash", keyType)
+	if fn := b.mod.NamedFunction(name); !fn.IsNil() {
+		return fn
 	}

-	return nil
+	// Create the LLVM function type:
+	// (key ptr, size uintptr, seed uintptr, context ptr) -> i32
+	fnType := llvm.FunctionType(b.ctx.Int32Type(), []llvm.Type{
+		b.dataPtrType, b.uintptrType, b.uintptrType, b.dataPtrType,
+	}, false)
+	fn := llvm.AddFunction(b.mod, name, fnType)
+	fn.SetLinkage(llvm.LinkOnceODRLinkage)
+	fn.SetUnnamedAddr(true)
+	b.addStandardAttributes(fn)
+
+	// Generate the function body.
+	savedBlock := b.GetInsertBlock()
+	defer b.SetInsertPointAtEnd(savedBlock)
+
+	entry := b.ctx.AddBasicBlock(fn, "entry")
+	b.SetInsertPointAtEnd(entry)
+
+	keyPtr := fn.Param(0)
+	seed := fn.Param(2)
+	llvmKeyType := b.getLLVMType(keyType)
+	hash := b.generateKeyHash(keyType, llvmKeyType, keyPtr, seed)
+	b.CreateRet(hash)
+
+	return fn
+}
+
+// getOrGenerateKeyEqualFunc returns an LLVM function that compares two keys
+// of the given type for equality. The function is generated on first call
+// and cached in the module.
+func (b *builder) getOrGenerateKeyEqualFunc(keyType types.Type) llvm.Value {
+	name := hashmapKeyFuncName("hashmapKeyEqual", keyType)
+	if fn := b.mod.NamedFunction(name); !fn.IsNil() {
+		return fn
+	}
+
+	// Create the LLVM function type:
+	// (x ptr, y ptr, n uintptr, context ptr) -> i1
+	fnType := llvm.FunctionType(b.ctx.Int1Type(), []llvm.Type{
+		b.dataPtrType, b.dataPtrType, b.uintptrType, b.dataPtrType,
+	}, false)
+	fn := llvm.AddFunction(b.mod, name, fnType)
+	fn.SetLinkage(llvm.LinkOnceODRLinkage)
+	fn.SetUnnamedAddr(true)
+	b.addStandardAttributes(fn)
+
+	// Generate the function body.
+	savedBlock := b.GetInsertBlock()
+	defer b.SetInsertPointAtEnd(savedBlock)
+
+	entry := b.ctx.AddBasicBlock(fn, "entry")
+	b.SetInsertPointAtEnd(entry)
+
+	xPtr := fn.Param(0)
+	yPtr := fn.Param(1)
+	llvmKeyType := b.getLLVMType(keyType)
+	result := b.generateKeyEqual(keyType, llvmKeyType, xPtr, yPtr, fn)
+	b.CreateRet(result)
+
+	return fn
+}
+
+// generateKeyHash generates IR that hashes a key value. Returns the i32 hash.
+func (b *builder) generateKeyHash(keyType types.Type, llvmKeyType llvm.Type, keyPtr llvm.Value, seed llvm.Value) llvm.Value {
+	switch keyType := keyType.Underlying().(type) {
+	case *types.Basic:
+		if keyType.Info()&types.IsString != 0 {
+			// Hash the string contents. The size parameter is unused by
+			// hashmapStringPtrHash (it dereferences the string header to
+			// get the actual length), but we pass it for signature
+			// consistency with other hash functions.
+			size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+			return b.createRuntimeCall("hashmapStringPtrHash", []llvm.Value{keyPtr, size, seed}, "hash")
+		}
+		if keyType.Info()&types.IsFloat != 0 {
+			// Float hash: normalizes -0 to +0 before hashing.
+			if keyType.Kind() == types.Float32 {
+				return b.createRuntimeCall("hashmapFloat32Hash", []llvm.Value{keyPtr, seed}, "hash")
+			}
+			return b.createRuntimeCall("hashmapFloat64Hash", []llvm.Value{keyPtr, seed}, "hash")
+		}
+		if keyType.Info()&types.IsComplex != 0 {
+			// Complex hash: hash real and imaginary parts as floats.
+			if keyType.Kind() == types.Complex64 {
+				realPtr := keyPtr
+				imagPtr := b.CreateInBoundsGEP(b.ctx.Int8Type(), keyPtr, []llvm.Value{
+					llvm.ConstInt(b.uintptrType, 4, false),
+				}, "")
+				realHash := b.createRuntimeCall("hashmapFloat32Hash", []llvm.Value{realPtr, seed}, "hash.real")
+				imagHash := b.createRuntimeCall("hashmapFloat32Hash", []llvm.Value{imagPtr, seed}, "hash.imag")
+				return b.CreateXor(realHash, imagHash, "")
+			}
+			realPtr := keyPtr
+			imagPtr := b.CreateInBoundsGEP(b.ctx.Int8Type(), keyPtr, []llvm.Value{
+				llvm.ConstInt(b.uintptrType, 8, false),
+			}, "")
+			realHash := b.createRuntimeCall("hashmapFloat64Hash", []llvm.Value{realPtr, seed}, "hash.real")
+			imagHash := b.createRuntimeCall("hashmapFloat64Hash", []llvm.Value{imagPtr, seed}, "hash.imag")
+			return b.CreateXor(realHash, imagHash, "")
+		}
+		// Integer/boolean: hash the raw bytes.
+		size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+		return b.createRuntimeCall("hash32", []llvm.Value{keyPtr, size, seed}, "hash")
+	case *types.Pointer, *types.Chan:
+		// Pointers and channels: hash as raw pointer-sized bytes.
+		size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+		return b.createRuntimeCall("hash32", []llvm.Value{keyPtr, size, seed}, "hash")
+	case *types.Interface:
+		// Interface: use runtime reflection-based hash.
+		size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+		return b.createRuntimeCall("hashmapInterfacePtrHash", []llvm.Value{keyPtr, size, seed}, "hash")
+	case *types.Struct:
+		hash := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+		zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+		for i := 0; i < keyType.NumFields(); i++ {
+			if keyType.Field(i).Name() == "_" {
+				continue // blank fields are ignored in Go equality
+			}
+			fieldType := keyType.Field(i).Type()
+			llvmFieldType := b.getLLVMType(fieldType)
+			if b.targetData.TypeAllocSize(llvmFieldType) == 0 {
+				continue // skip zero-sized fields
+			}
+			idx := llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)
+			fieldPtr := b.CreateInBoundsGEP(llvmKeyType, keyPtr, []llvm.Value{zero, idx}, "")
+			fieldHash := b.generateKeyHash(fieldType, llvmFieldType, fieldPtr, seed)
+			hash = b.CreateXor(hash, fieldHash, "")
+		}
+		return hash
+	case *types.Array:
+		elemType := keyType.Elem()
+		llvmElemType := b.getLLVMType(elemType)
+		arrayLen := keyType.Len()
+		if hashmapIsBinaryKey(elemType) {
+			// All elements are binary-comparable; hash the entire array as raw bytes.
+			size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+			return b.createRuntimeCall("hash32", []llvm.Value{keyPtr, size, seed}, "hash")
+		}
+		if arrayLen == 0 {
+			return llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+		}
+		if arrayLen <= hashArrayUnrollLimit {
+			hash := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+			for i := 0; i < int(arrayLen); i++ {
+				idx := llvm.ConstInt(b.uintptrType, uint64(i), false)
+				elemPtr := b.CreateInBoundsGEP(llvmKeyType, keyPtr, []llvm.Value{zero, idx}, "")
+				elemHash := b.generateKeyHash(elemType, llvmElemType, elemPtr, seed)
+				hash = b.CreateXor(hash, elemHash, "")
+			}
+			return hash
+		}
+		initHash := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+		zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+
+		loopEntry := b.GetInsertBlock()
+		loopBody := b.ctx.AddBasicBlock(loopEntry.Parent(), "hash.array.body")
+		loopDone := b.ctx.AddBasicBlock(loopEntry.Parent(), "hash.array.done")
+
+		b.CreateBr(loopBody)
+		b.SetInsertPointAtEnd(loopBody)
+
+		phiI := b.CreatePHI(b.uintptrType, "i")
+		phiHash := b.CreatePHI(b.ctx.Int32Type(), "hash.acc")
+
+		elemPtr := b.CreateInBoundsGEP(llvmKeyType, keyPtr, []llvm.Value{zero, phiI}, "")
+		elemHash := b.generateKeyHash(elemType, llvmElemType, elemPtr, seed)
+		newHash := b.CreateXor(phiHash, elemHash, "")
+		nextI := b.CreateAdd(phiI, llvm.ConstInt(b.uintptrType, 1, false), "")
+		cond := b.CreateICmp(llvm.IntULT, nextI, llvm.ConstInt(b.uintptrType, uint64(arrayLen), false), "")
+		b.CreateCondBr(cond, loopBody, loopDone)
+
+		bodyEnd := b.GetInsertBlock()
+		phiI.AddIncoming([]llvm.Value{llvm.ConstInt(b.uintptrType, 0, false), nextI},
+			[]llvm.BasicBlock{loopEntry, bodyEnd})
+		phiHash.AddIncoming([]llvm.Value{initHash, newHash},
+			[]llvm.BasicBlock{loopEntry, bodyEnd})
+
+		b.SetInsertPointAtEnd(loopDone)
+		return newHash
+	default:
+		panic(fmt.Sprintf("unhandled key type for hash generation: %T", keyType))
+	}
+}
+
+// generateKeyEqual generates IR that compares two key values for equality.
+// Returns an i1 result.
+func (b *builder) generateKeyEqual(keyType types.Type, llvmKeyType llvm.Type, xPtr, yPtr llvm.Value, fn llvm.Value) llvm.Value {
+	switch keyType := keyType.Underlying().(type) {
+	case *types.Basic:
+		if keyType.Info()&types.IsString != 0 {
+			// Compare strings: load both string headers and compare.
+			xStr := b.CreateLoad(llvmKeyType, xPtr, "x.str")
+			yStr := b.CreateLoad(llvmKeyType, yPtr, "y.str")
+			return b.createRuntimeCall("stringEqual", []llvm.Value{xStr, yStr}, "eq")
+		}
+		if keyType.Info()&types.IsFloat != 0 {
+			// Float equality: fcmp oeq handles -0==+0 (true) and NaN==NaN (false).
+			xVal := b.CreateLoad(llvmKeyType, xPtr, "x.float")
+			yVal := b.CreateLoad(llvmKeyType, yPtr, "y.float")
+			return b.CreateFCmp(llvm.FloatOEQ, xVal, yVal, "eq")
+		}
+		if keyType.Info()&types.IsComplex != 0 {
+			// Complex equality: both real and imaginary parts must be equal.
+			var floatType llvm.Type
+			if keyType.Kind() == types.Complex64 {
+				floatType = b.ctx.FloatType()
+			} else {
+				floatType = b.ctx.DoubleType()
+			}
+			floatSize := b.targetData.TypeAllocSize(floatType)
+			imagOffset := llvm.ConstInt(b.uintptrType, floatSize, false)
+			// Real parts
+			xReal := b.CreateLoad(floatType, xPtr, "x.real")
+			yReal := b.CreateLoad(floatType, yPtr, "y.real")
+			realEq := b.CreateFCmp(llvm.FloatOEQ, xReal, yReal, "eq.real")
+			// Imaginary parts
+			xImagPtr := b.CreateInBoundsGEP(b.ctx.Int8Type(), xPtr, []llvm.Value{imagOffset}, "")
+			yImagPtr := b.CreateInBoundsGEP(b.ctx.Int8Type(), yPtr, []llvm.Value{imagOffset}, "")
+			xImag := b.CreateLoad(floatType, xImagPtr, "x.imag")
+			yImag := b.CreateLoad(floatType, yImagPtr, "y.imag")
+			imagEq := b.CreateFCmp(llvm.FloatOEQ, xImag, yImag, "eq.imag")
+			return b.CreateAnd(realEq, imagEq, "")
+		}
+		// Integer/boolean: compare raw bytes.
+		size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+		return b.createRuntimeCall("memequal", []llvm.Value{xPtr, yPtr, size}, "eq")
+	case *types.Pointer, *types.Chan:
+		// Pointers and channels: compare as raw pointer-sized bytes.
+		size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+		return b.createRuntimeCall("memequal", []llvm.Value{xPtr, yPtr, size}, "eq")
+	case *types.Interface:
+		// Interface: use runtime interface equality.
+		size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+		return b.createRuntimeCall("hashmapInterfaceEqual", []llvm.Value{xPtr, yPtr, size}, "eq")
+	case *types.Struct:
+		result := llvm.ConstInt(b.ctx.Int1Type(), 1, false) // start with true
+		zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+		for i := 0; i < keyType.NumFields(); i++ {
+			if keyType.Field(i).Name() == "_" {
+				continue // blank fields are ignored in Go equality
+			}
+			fieldType := keyType.Field(i).Type()
+			llvmFieldType := b.getLLVMType(fieldType)
+			if b.targetData.TypeAllocSize(llvmFieldType) == 0 {
+				continue // skip zero-sized fields
+			}
+			idx := llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)
+			xFieldPtr := b.CreateInBoundsGEP(llvmKeyType, xPtr, []llvm.Value{zero, idx}, "")
+			yFieldPtr := b.CreateInBoundsGEP(llvmKeyType, yPtr, []llvm.Value{zero, idx}, "")
+			fieldEq := b.generateKeyEqual(fieldType, llvmFieldType, xFieldPtr, yFieldPtr, fn)
+			result = b.CreateAnd(result, fieldEq, "")
+		}
+		return result
+	case *types.Array:
+		elemType := keyType.Elem()
+		llvmElemType := b.getLLVMType(elemType)
+		arrayLen := keyType.Len()
+		if hashmapIsBinaryKey(elemType) {
+			// All elements are binary-comparable; compare the entire array.
+			size := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(llvmKeyType), false)
+			return b.createRuntimeCall("memequal", []llvm.Value{xPtr, yPtr, size}, "eq")
+		}
+		if arrayLen == 0 {
+			return llvm.ConstInt(b.ctx.Int1Type(), 1, false)
+		}
+		if arrayLen <= hashArrayUnrollLimit {
+			result := llvm.ConstInt(b.ctx.Int1Type(), 1, false)
+			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+			for i := 0; i < int(arrayLen); i++ {
+				idx := llvm.ConstInt(b.uintptrType, uint64(i), false)
+				xElemPtr := b.CreateInBoundsGEP(llvmKeyType, xPtr, []llvm.Value{zero, idx}, "")
+				yElemPtr := b.CreateInBoundsGEP(llvmKeyType, yPtr, []llvm.Value{zero, idx}, "")
+				elemEq := b.generateKeyEqual(elemType, llvmElemType, xElemPtr, yElemPtr, fn)
+				result = b.CreateAnd(result, elemEq, "")
+			}
+			return result
+		}
+		zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+
+		loopEntry := b.GetInsertBlock()
+		loopBody := b.ctx.AddBasicBlock(loopEntry.Parent(), "eq.array.body")
+		loopDone := b.ctx.AddBasicBlock(loopEntry.Parent(), "eq.array.done")
+
+		b.CreateBr(loopBody)
+		b.SetInsertPointAtEnd(loopBody)
+
+		phiI := b.CreatePHI(b.uintptrType, "i")
+
+		xElemPtr := b.CreateInBoundsGEP(llvmKeyType, xPtr, []llvm.Value{zero, phiI}, "")
+		yElemPtr := b.CreateInBoundsGEP(llvmKeyType, yPtr, []llvm.Value{zero, phiI}, "")
+		elemEq := b.generateKeyEqual(elemType, llvmElemType, xElemPtr, yElemPtr, fn)
+
+		nextI := b.CreateAdd(phiI, llvm.ConstInt(b.uintptrType, 1, false), "")
+		atEnd := b.CreateICmp(llvm.IntUGE, nextI, llvm.ConstInt(b.uintptrType, uint64(arrayLen), false), "")
+		exitLoop := b.CreateOr(atEnd, b.CreateNot(elemEq, ""), "")
+		b.CreateCondBr(exitLoop, loopDone, loopBody)
+
+		bodyEnd := b.GetInsertBlock()
+		phiI.AddIncoming([]llvm.Value{llvm.ConstInt(b.uintptrType, 0, false), nextI},
+			[]llvm.BasicBlock{loopEntry, bodyEnd})
+
+		b.SetInsertPointAtEnd(loopDone)
+		return elemEq
+	default:
+		panic(fmt.Sprintf("unhandled key type for equal generation: %T", keyType))
+	}
 }
@@ -29,8 +29,7 @@ func (s *stdSizes) Alignof(T types.Type) int64 {
 		// is the largest of the values unsafe.Alignof(x.f) for each
 		// field f of x, but at least 1."
 		max := int64(1)
-		for i := 0; i < t.NumFields(); i++ {
-			f := t.Field(i)
+		for f := range t.Fields() {
 			if a := s.Alignof(f.Type()); a > max {
 				max = a
 			}
@@ -8,6 +8,7 @@ import (
 	"go/ast"
 	"go/token"
 	"go/types"
+	"slices"
 	"strconv"
 	"strings"

@@ -34,6 +35,7 @@ type functionInfo struct {
 	interrupt     bool       // go:interrupt
 	nobounds      bool       // go:nobounds
 	noescape      bool       // go:noescape
+	noheap        bool       // go:noheap
 	variadic      bool       // go:variadic (CGo only)
 	inline        inlineType // go:inline
 }
@@ -77,24 +79,27 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		return llvmFn.GlobalValueType(), llvmFn
 	}

-	var retType llvm.Type
-	if fn.Signature.Results() == nil {
-		retType = c.ctx.VoidType()
-	} else if fn.Signature.Results().Len() == 1 {
-		retType = c.getLLVMType(fn.Signature.Results().At(0).Type())
-	} else {
-		results := make([]llvm.Type, 0, fn.Signature.Results().Len())
-		for i := 0; i < fn.Signature.Results().Len(); i++ {
-			results = append(results, c.getLLVMType(fn.Signature.Results().At(i).Type()))
-		}
-		retType = c.ctx.StructType(results, false)
+	retType, indirectResult := c.hasIndirectResult(fn.Signature)
+	if info.exported {
+		indirectResult = false
 	}

 	var paramInfos []paramInfo
+	if indirectResult {
+		paramInfos = append(paramInfos, paramInfo{
+			llvmType: c.dataPtrType,
+			name:     "return",
+			elemSize: c.targetData.TypeAllocSize(retType),
+		})
+		retType = c.ctx.VoidType()
+	}
 	for _, param := range getParams(fn.Signature) {
 		paramType := c.getLLVMType(param.Type())
-		paramFragmentInfos := c.expandFormalParamType(paramType, param.Name(), param.Type())
-		paramInfos = append(paramInfos, paramFragmentInfos...)
+		if info.exported {
+			paramInfos = append(paramInfos, c.expandDirectFormalParamType(paramType, param.Name(), param.Type())...)
+		} else {
+			paramInfos = append(paramInfos, c.expandFormalParamType(paramType, param.Name(), param.Type())...)
+		}
 	}

 	// Add an extra parameter as the function context. This context is used in
@@ -104,12 +109,20 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	}

 	var paramTypes []llvm.Type
+	hasIndirectABI := indirectResult
 	for _, info := range paramInfos {
 		paramTypes = append(paramTypes, info.llvmType)
+		hasIndirectABI = hasIndirectABI || info.flags&paramIsIndirect != 0
 	}

 	fnType := llvm.FunctionType(retType, paramTypes, info.variadic)
 	llvmFn = llvm.AddFunction(c.mod, info.linkName, fnType)
+	if hasIndirectABI {
+		// Argument promotion only rewrites functions whose uses are all direct
+		// calls. Keep an address use so LLVM cannot reconstruct the large
+		// aggregate signature that this ABI exists to avoid.
+		llvmutil.AppendToGlobal(c.mod, "llvm.used", llvmFn)
+	}
 	if strings.HasPrefix(c.Triple, "wasm") {
 		// C functions without prototypes like this:
 		//   void foo();
@@ -139,7 +152,7 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 			// not.
 			// (It may be safe to add the nocapture parameter to the context
 			// parameter, but I'd like to stay on the safe side here).
-			nocapture := c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0)
+			nocapture := c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0)
 			llvmFn.AddAttributeAtIndex(i+1, nocapture)
 		}
 		if paramInfo.flags&paramIsReadonly != 0 && paramInfo.llvmType.TypeKind() == llvm.PointerTypeKind {
@@ -158,10 +171,10 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		// Mark it as noreturn so LLVM can optimize away code.
 		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("noreturn"), 0))
 	case "internal/abi.NoEscape":
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
 	case "machine.keepAliveNoEscape", "machine.unsafeNoEscape":
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
-	case "runtime.alloc":
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+	case "runtime.alloc", "runtime.alloc_noheap", "runtime.alloc_zero":
 		// Tell the optimizer that runtime.alloc is an allocator, meaning that it
 		// returns values that are never null and never alias to an existing value.
 		for _, attrName := range []string{"noalias", "nonnull"} {
@@ -182,19 +195,44 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	case "runtime.sliceAppend":
 		// Appending a slice will only read the to-be-appended slice, it won't
 		// be modified.
-		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
 	case "runtime.stringFromBytes":
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
 	case "runtime.stringFromRunes":
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
+	case "runtime.hashmapSet":
+		// The key (param 2) and value (param 3) pointers are only read via
+		// memcpy/hash/equal and are never captured. The indirect calls
+		// through m.keyHash and m.keyEqual function pointers prevent LLVM's
+		// functionattrs pass from inferring this automatically.
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+		llvmFn.AddAttributeAtIndex(3, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+	case "runtime.hashmapGet":
+		// The key (param 2) is read-only and never captured.
+		// The value (param 3) is written to (receives the result) but never captured.
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+		llvmFn.AddAttributeAtIndex(3, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+	case "runtime.hashmapDelete":
+		// The key (param 2) is read-only and never captured.
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+	case "runtime.hashmapGenericSet":
+		// Same as hashmapBinarySet: key (param 2) and value (param 3) are
+		// not captured.
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+		llvmFn.AddAttributeAtIndex(3, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+	case "runtime.hashmapGenericGet":
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+		llvmFn.AddAttributeAtIndex(3, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
+	case "runtime.hashmapGenericDelete":
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
 	case "runtime.trackPointer":
 		// This function is necessary for tracking pointers on the stack in a
 		// portable way (see gc_stack_portable.go). Indicate to the optimizer
 		// that the only thing we'll do is read the pointer.
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(llvmutil.NoCaptureAttrName()), 0))
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
 	case "__mulsi3", "__divmodsi4", "__udivmodsi4":
 		if strings.Split(c.Triple, "-")[0] == "avr" {
@@ -229,7 +267,7 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)

 			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("wasm-import-name", info.wasmName))
 		}
-		nocaptureKind := llvm.AttributeKindID("nocapture")
+		nocaptureKind := llvm.AttributeKindID(llvmutil.NoCaptureAttrName())
 		nocapture := c.ctx.CreateEnumAttribute(nocaptureKind, 0)
 		for i, typ := range paramTypes {
 			if typ.TypeKind() == llvm.PointerTypeKind {
@@ -293,7 +331,7 @@ func (c *compilerContext) getFunctionInfo(f *ssa.Function) functionInfo {
 	}
 	info := functionInfo{
 		// Pick the default linkName.
-		linkName: f.RelString(nil),
+		linkName: c.canonicalFunctionName(f),
 	}

 	// Check for a few runtime functions that are treated specially.
@@ -320,6 +358,18 @@ func (c *compilerContext) getFunctionInfo(f *ssa.Function) functionInfo {
 	return info
 }

+func (c *compilerContext) canonicalFunctionName(f *ssa.Function) string {
+	typeArgs := f.TypeArgs()
+	if len(typeArgs) == 0 {
+		return f.RelString(nil)
+	}
+	parts := make([]string, len(typeArgs))
+	for i, ta := range typeArgs {
+		parts[i], _ = c.getTypeCodeName(ta)
+	}
+	return f.Origin().RelString(nil) + "[" + strings.Join(parts, ",") + "]"
+}
+
 // parsePragmas is used by getFunctionInfo to parse function pragmas such as
 // //export or //go:noinline.
 func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
@@ -346,6 +396,51 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 		}
 	}

+	// Also scan file-level //go:linkname directives. These appear as
+	// free-standing comments in *ast.File.Comments (not attached to any
+	// declaration), and are used by modern golang.org/x/sys/unix and others.
+	// Function-attached directives (above) take precedence — we only add
+	// file-level ones if no doc-comment linkname was found for this function.
+	//
+	// TODO: the hasUnsafeImport gate enforced downstream (see the
+	// //go:linkname case below) is package-level. gc enforces it per
+	// file, on the file containing the directive. For file-level
+	// linknames this is more important than for function-attached ones,
+	// because the directive can live in a file separate from the
+	// function. A stricter implementation would check whether the file
+	// returned by fileForFunc imports "unsafe", not whether any file in
+	// the package does.
+	hasFunctionLinkname := false
+	for _, comment := range pragmas {
+		if strings.HasPrefix(comment.Text, "//go:linkname ") {
+			parts := strings.Fields(comment.Text)
+			if len(parts) == 3 && parts[1] == f.Name() {
+				hasFunctionLinkname = true
+				break
+			}
+		}
+	}
+	if !hasFunctionLinkname {
+		if file := c.fileForFunc(f); file != nil {
+			for _, group := range file.Comments {
+				// Skip the function's own doc comment — already handled above.
+				if decl, ok := syntax.(*ast.FuncDecl); ok && group == decl.Doc {
+					continue
+				}
+				for _, comment := range group.List {
+					if !strings.HasPrefix(comment.Text, "//go:linkname ") {
+						continue
+					}
+					parts := strings.Fields(comment.Text)
+					if len(parts) != 3 || parts[1] != f.Name() {
+						continue
+					}
+					pragmas = append(pragmas, comment)
+				}
+			}
+		}
+	}
+
 	// Parse each pragma.
 	for _, comment := range pragmas {
 		parts := strings.Fields(comment.Text)
@@ -363,7 +458,7 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			info.wasmName = info.linkName
 			info.exported = true
 		case "//go:interrupt":
-			if hasUnsafeImport(f.Pkg.Pkg) {
+			if slices.Contains(f.Pkg.Pkg.Imports(), types.Unsafe) {
 				info.interrupt = true
 			}
 		case "//go:wasm-module":
@@ -417,7 +512,7 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			info.inline = inlineHint
 		case "//go:noinline":
 			info.inline = inlineNone
-		case "//go:linkname":
+		case "//go:linkname", "//go:linknamestd":
 			if len(parts) != 3 || parts[1] != f.Name() {
 				continue
 			}
@@ -425,14 +520,14 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			// This is a slightly looser requirement than what gc uses: gc
 			// requires the file to import "unsafe", not the package as a
 			// whole.
-			if hasUnsafeImport(f.Pkg.Pkg) {
+			if slices.Contains(f.Pkg.Pkg.Imports(), types.Unsafe) {
 				info.linkName = parts[2]
 			}
 		case "//go:section":
 			// Only enable go:section when the package imports "unsafe".
 			// go:section also implies go:noinline since inlining could
 			// move the code to a different section than that requested.
-			if len(parts) == 2 && hasUnsafeImport(f.Pkg.Pkg) {
+			if len(parts) == 2 && slices.Contains(f.Pkg.Pkg.Imports(), types.Unsafe) {
 				info.section = parts[1]
 				info.inline = inlineNone
 			}
@@ -441,7 +536,7 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			// runtime functions.
 			// This is somewhat dangerous and thus only imported in packages
 			// that import unsafe.
-			if hasUnsafeImport(f.Pkg.Pkg) {
+			if slices.Contains(f.Pkg.Pkg.Imports(), types.Unsafe) {
 				info.nobounds = true
 			}
 		case "//go:noescape":
@@ -451,6 +546,9 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			if len(f.Blocks) == 0 {
 				info.noescape = true
 			}
+		case "//go:noheap":
+			// Ensure this function does not allocate on the heap.
+			info.noheap = true
 		case "//go:variadic":
 			// The //go:variadic pragma is emitted by the CGo preprocessing
 			// pass for C variadic functions. This includes both explicit
@@ -538,8 +636,8 @@ func (c *compilerContext) isValidWasmType(typ types.Type, site wasmSite) bool {
 				hasHostLayout = false // package structs added in go1.23
 			}
 		}
-		for i := 0; i < typ.NumFields(); i++ {
-			ftyp := typ.Field(i).Type()
+		for field := range typ.Fields() {
+			ftyp := field.Type()
 			if types.Unalias(ftyp).String() == "structs.HostLayout" {
 				hasHostLayout = true
 				continue
@@ -571,8 +669,8 @@ func getParams(sig *types.Signature) []*types.Var {
 	if sig.Recv() != nil {
 		params = append(params, sig.Recv())
 	}
-	for i := 0; i < sig.Params().Len(); i++ {
-		params = append(params, sig.Params().At(i))
+	for v := range sig.Params().Variables() {
+		params = append(params, v)
 	}
 	return params
 }
@@ -637,6 +735,34 @@ type globalInfo struct {
 	section  string // go:section
 }

+// fileForFunc returns the *ast.File that contains the declaration of f, or
+// nil if it cannot be determined. File-level pragmas are only consulted for
+// functions in the package currently being compiled — functions imported from
+// other packages have their file-level pragmas processed when those packages
+// are compiled.
+func (c *compilerContext) fileForFunc(f *ssa.Function) *ast.File {
+	if c.loaderPkg == nil || f.Pkg == nil || f.Pkg.Pkg != c.loaderPkg.Pkg {
+		return nil
+	}
+	syntax := f.Syntax()
+	if f.Origin() != nil {
+		syntax = f.Origin().Syntax()
+	}
+	if syntax == nil {
+		return nil
+	}
+	pos := syntax.Pos()
+	if !pos.IsValid() {
+		return nil
+	}
+	for _, file := range c.loaderPkg.Files {
+		if file.FileStart <= pos && pos < file.FileEnd {
+			return file
+		}
+	}
+	return nil
+}
+
 // loadASTComments loads comments on globals from the AST, for use later in the
 // program. In particular, they are required for //go:extern pragmas on globals.
 func (c *compilerContext) loadASTComments(pkg *loader.Package) {
@@ -675,10 +801,7 @@ func (c *compilerContext) getGlobal(g *ssa.Global) llvm.Value {
 		llvmGlobal = llvm.AddGlobal(c.mod, llvmType, info.linkName)

 		// Set alignment from the //go:align comment.
-		alignment := c.targetData.ABITypeAlignment(llvmType)
-		if info.align > alignment {
-			alignment = info.align
-		}
+		alignment := max(info.align, c.targetData.ABITypeAlignment(llvmType))
 		if alignment <= 0 || alignment&(alignment-1) != 0 {
 			// Check for power-of-two (or 0).
 			// See: https://stackoverflow.com/a/108360
@@ -752,7 +875,7 @@ func (info *globalInfo) parsePragmas(doc *ast.CommentGroup, c *compilerContext,
 			// This is a slightly looser requirement than what gc uses: gc
 			// requires the file to import "unsafe", not the package as a
 			// whole.
-			if hasUnsafeImport(g.Pkg.Pkg) {
+			if slices.Contains(g.Pkg.Pkg.Imports(), types.Unsafe) {
 				info.linkName = parts[2]
 			}
 		}
@@ -768,13 +891,3 @@ func getAllMethods(prog *ssa.Program, typ types.Type) []*types.Selection {
 	}
 	return methods
 }
-
-// Return true if this package imports "unsafe", false otherwise.
-func hasUnsafeImport(pkg *types.Package) bool {
-	for _, imp := range pkg.Imports() {
-		if imp == types.Unsafe {
-			return true
-		}
-	}
-	return false
-}
@@ -26,24 +26,25 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		argTypes := []llvm.Type{b.uintptrType}
 		// Constraints will look something like:
 		//   "={rax},0,{rdi},{rsi},{rdx},{r10},{r8},{r9},~{rcx},~{r11}"
-		constraints := "={rax},0"
+		var constraints strings.Builder
+		constraints.WriteString("={rax},0")
 		for i, arg := range call.Args[1:] {
-			constraints += "," + [...]string{
+			constraints.WriteString("," + [...]string{
 				"{rdi}",
 				"{rsi}",
 				"{rdx}",
 				"{r10}",
 				"{r8}",
 				"{r9}",
-			}[i]
+			}[i])
 			llvmValue := b.getValue(arg, getPos(call))
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		// rcx and r11 are clobbered by the syscall, so make sure they are not used
-		constraints += ",~{rcx},~{r11}"
+		constraints.WriteString(",~{rcx},~{r11}")
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
-		target := llvm.InlineAsm(fnType, "syscall", constraints, true, false, llvm.InlineAsmDialectIntel, false)
+		target := llvm.InlineAsm(fnType, "syscall", constraints.String(), true, false, llvm.InlineAsmDialectIntel, false)
 		return b.CreateCall(fnType, target, args, ""), nil

 	case b.GOARCH == "386" && b.GOOS == "linux":
@@ -55,22 +56,23 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		argTypes := []llvm.Type{b.uintptrType}
 		// Constraints will look something like:
 		//   "={eax},0,{ebx},{ecx},{edx},{esi},{edi},{ebp}"
-		constraints := "={eax},0"
+		var constraints strings.Builder
+		constraints.WriteString("={eax},0")
 		for i, arg := range call.Args[1:] {
-			constraints += "," + [...]string{
+			constraints.WriteString("," + [...]string{
 				"{ebx}",
 				"{ecx}",
 				"{edx}",
 				"{esi}",
 				"{edi}",
 				"{ebp}",
-			}[i]
+			}[i])
 			llvmValue := b.getValue(arg, getPos(call))
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
-		target := llvm.InlineAsm(fnType, "int 0x80", constraints, true, false, llvm.InlineAsmDialectIntel, false)
+		target := llvm.InlineAsm(fnType, "int 0x80", constraints.String(), true, false, llvm.InlineAsmDialectIntel, false)
 		return b.CreateCall(fnType, target, args, ""), nil

 	case b.GOARCH == "arm" && b.GOOS == "linux":
@@ -88,9 +90,10 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		argTypes := []llvm.Type{}
 		// Constraints will look something like:
 		//   ={r0},0,{r1},{r2},{r7},~{r3}
-		constraints := "={r0}"
+		var constraints strings.Builder
+		constraints.WriteString("={r0}")
 		for i, arg := range call.Args[1:] {
-			constraints += "," + [...]string{
+			constraints.WriteString("," + [...]string{
 				"0", // tie to output
 				"{r1}",
 				"{r2}",
@@ -98,20 +101,20 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 				"{r4}",
 				"{r5}",
 				"{r6}",
-			}[i]
+			}[i])
 			llvmValue := b.getValue(arg, getPos(call))
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		args = append(args, num)
 		argTypes = append(argTypes, b.uintptrType)
-		constraints += ",{r7}" // syscall number
+		constraints.WriteString(",{r7}") // syscall number
 		for i := len(call.Args) - 1; i < 4; i++ {
 			// r0-r3 get clobbered after the syscall returns
-			constraints += ",~{r" + strconv.Itoa(i) + "}"
+			constraints.WriteString(",~{r" + strconv.Itoa(i) + "}")
 		}
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
-		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0, false)
+		target := llvm.InlineAsm(fnType, "svc #0", constraints.String(), true, false, 0, false)
 		return b.CreateCall(fnType, target, args, ""), nil

 	case b.GOARCH == "arm64" && b.GOOS == "linux":
@@ -120,31 +123,32 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		argTypes := []llvm.Type{}
 		// Constraints will look something like:
 		//   ={x0},0,{x1},{x2},{x8},~{x3},~{x4},~{x5},~{x6},~{x7},~{x16},~{x17}
-		constraints := "={x0}"
+		var constraints strings.Builder
+		constraints.WriteString("={x0}")
 		for i, arg := range call.Args[1:] {
-			constraints += "," + [...]string{
+			constraints.WriteString("," + [...]string{
 				"0", // tie to output
 				"{x1}",
 				"{x2}",
 				"{x3}",
 				"{x4}",
 				"{x5}",
-			}[i]
+			}[i])
 			llvmValue := b.getValue(arg, getPos(call))
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		args = append(args, num)
 		argTypes = append(argTypes, b.uintptrType)
-		constraints += ",{x8}" // syscall number
+		constraints.WriteString(",{x8}") // syscall number
 		for i := len(call.Args) - 1; i < 8; i++ {
 			// x0-x7 may get clobbered during the syscall following the aarch64
 			// calling convention.
-			constraints += ",~{x" + strconv.Itoa(i) + "}"
+			constraints.WriteString(",~{x" + strconv.Itoa(i) + "}")
 		}
-		constraints += ",~{x16},~{x17}" // scratch registers
+		constraints.WriteString(",~{x16},~{x17}") // scratch registers
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
-		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0, false)
+		target := llvm.InlineAsm(fnType, "svc #0", constraints.String(), true, false, 0, false)
 		return b.CreateCall(fnType, target, args, ""), nil

 	case (b.GOARCH == "mips" || b.GOARCH == "mipsle") && b.GOOS == "linux":
@@ -163,7 +167,8 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		// faster and smaller code.
 		args := []llvm.Value{num}
 		argTypes := []llvm.Type{b.uintptrType}
-		constraints := "={$2},={$7},0"
+		var constraints strings.Builder
+		constraints.WriteString("={$2},={$7},0")
 		syscallParams := call.Args[1:]
 		if len(syscallParams) > 7 {
 			// There is one syscall that uses 7 parameters: sync_file_range.
@@ -172,7 +177,7 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 			syscallParams = syscallParams[:7]
 		}
 		for i, arg := range syscallParams {
-			constraints += "," + [...]string{
+			constraints.WriteString("," + [...]string{
 				"{$4}", // arg1
 				"{$5}", // arg2
 				"{$6}", // arg3
@@ -180,7 +185,7 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 				"r",    // arg5 on the stack
 				"r",    // arg6 on the stack
 				"r",    // arg7 on the stack
-			}[i]
+			}[i])
 			llvmValue := b.getValue(arg, getPos(call))
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
@@ -221,10 +226,10 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 				"addu $$sp, $$sp, 32\n" +
 				".set at\n"
 		}
-		constraints += ",~{$3},~{$4},~{$5},~{$6},~{$8},~{$9},~{$10},~{$11},~{$12},~{$13},~{$14},~{$15},~{$24},~{$25},~{hi},~{lo},~{memory}"
+		constraints.WriteString(",~{$3},~{$4},~{$5},~{$6},~{$8},~{$9},~{$10},~{$11},~{$12},~{$13},~{$14},~{$15},~{$24},~{$25},~{hi},~{lo},~{memory}")
 		returnType := b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType}, false)
 		fnType := llvm.FunctionType(returnType, argTypes, false)
-		target := llvm.InlineAsm(fnType, asm, constraints, true, true, 0, false)
+		target := llvm.InlineAsm(fnType, asm, constraints.String(), true, true, 0, false)
 		call := b.CreateCall(fnType, target, args, "")
 		resultCode := b.CreateExtractValue(call, 0, "") // r2
 		errorFlag := b.CreateExtractValue(call, 1, "")  // r7
@@ -66,6 +66,11 @@ func complexSub(x, y complex64) complex64 {
 	return x - y
 }

+func shiftNested(x uint64) uint64 {
+	k := 3
+	return x >> (1 << k) // https://github.com/tinygo-org/tinygo/issues/5496
+}
+
 func complexMul(x, y complex64) complex64 {
 	return x * y
 }
@@ -10,33 +10,30 @@ target triple = "wasm32-unknown-wasi"
@main.a = hidden global { ptr, i32, i32 } zeroinitializer, align 4
@main.b = hidden global [2 x ptr] zeroinitializer, align 4

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.addInt(i32 %x, i32 %y, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.addInt(i32 %x, i32 %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = add i32 %x, %y
  ret i32 %0
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.equalInt(i32 %x, i32 %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.equalInt(i32 %x, i32 %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp eq i32 %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.divInt(i32 %x, i32 %y, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.divInt(i32 %x, i32 %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp eq i32 %y, 0
  br i1 %0, label %divbyzero.throw, label %divbyzero.next
@@ -50,14 +47,14 @@ divbyzero.next:                                   ; preds = %entry
  ret i32 %5

 divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(ptr undef) #3
+  call void @runtime.divideByZeroPanic(ptr undef) #2
  unreachable
 }

-declare void @runtime.divideByZeroPanic(ptr) #1
+declare void @runtime.divideByZeroPanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden i32 @main.divUint(i32 %x, i32 %y, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.divUint(i32 %x, i32 %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp eq i32 %y, 0
  br i1 %0, label %divbyzero.throw, label %divbyzero.next
@@ -67,12 +64,12 @@ divbyzero.next:                                   ; preds = %entry
  ret i32 %1

 divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(ptr undef) #3
+  call void @runtime.divideByZeroPanic(ptr undef) #2
  unreachable
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.remInt(i32 %x, i32 %y, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.remInt(i32 %x, i32 %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp eq i32 %y, 0
  br i1 %0, label %divbyzero.throw, label %divbyzero.next
@@ -86,12 +83,12 @@ divbyzero.next:                                   ; preds = %entry
  ret i32 %5

 divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(ptr undef) #3
+  call void @runtime.divideByZeroPanic(ptr undef) #2
  unreachable
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.remUint(i32 %x, i32 %y, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.remUint(i32 %x, i32 %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp eq i32 %y, 0
  br i1 %0, label %divbyzero.throw, label %divbyzero.next
@@ -101,66 +98,66 @@ divbyzero.next:                                   ; preds = %entry
  ret i32 %1

 divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(ptr undef) #3
+  call void @runtime.divideByZeroPanic(ptr undef) #2
  unreachable
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.floatEQ(float %x, float %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.floatEQ(float %x, float %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fcmp oeq float %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.floatNE(float %x, float %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.floatNE(float %x, float %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fcmp une float %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.floatLower(float %x, float %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.floatLower(float %x, float %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fcmp olt float %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.floatLowerEqual(float %x, float %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.floatLowerEqual(float %x, float %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fcmp ole float %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.floatGreater(float %x, float %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.floatGreater(float %x, float %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fcmp ogt float %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.floatGreaterEqual(float %x, float %y, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.floatGreaterEqual(float %x, float %y, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fcmp oge float %x, %y
  ret i1 %0
 }

 ; Function Attrs: nounwind
-define hidden float @main.complexReal(float %x.r, float %x.i, ptr %context) unnamed_addr #2 {
+define hidden float @main.complexReal(float %x.r, float %x.i, ptr %context) unnamed_addr #1 {
 entry:
  ret float %x.r
 }

 ; Function Attrs: nounwind
-define hidden float @main.complexImag(float %x.r, float %x.i, ptr %context) unnamed_addr #2 {
+define hidden float @main.complexImag(float %x.r, float %x.i, ptr %context) unnamed_addr #1 {
 entry:
  ret float %x.i
 }

 ; Function Attrs: nounwind
-define hidden { float, float } @main.complexAdd(float %x.r, float %x.i, float %y.r, float %y.i, ptr %context) unnamed_addr #2 {
+define hidden { float, float } @main.complexAdd(float %x.r, float %x.i, float %y.r, float %y.i, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fadd float %x.r, %y.r
  %1 = fadd float %x.i, %y.i
@@ -170,7 +167,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden { float, float } @main.complexSub(float %x.r, float %x.i, float %y.r, float %y.i, ptr %context) unnamed_addr #2 {
+define hidden { float, float } @main.complexSub(float %x.r, float %x.i, float %y.r, float %y.i, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fsub float %x.r, %y.r
  %1 = fsub float %x.i, %y.i
@@ -180,7 +177,14 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden { float, float } @main.complexMul(float %x.r, float %x.i, float %y.r, float %y.i, ptr %context) unnamed_addr #2 {
+define hidden i64 @main.shiftNested(i64 %x, ptr %context) unnamed_addr #1 {
+entry:
+  %0 = lshr i64 %x, 8
+  ret i64 %0
+}
+
+; Function Attrs: nounwind
+define hidden { float, float } @main.complexMul(float %x.r, float %x.i, float %y.r, float %y.i, ptr %context) unnamed_addr #1 {
 entry:
  %0 = fmul float %x.r, %y.r
  %1 = fmul float %x.i, %y.i
@@ -194,19 +198,18 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.foo(ptr %context) unnamed_addr #2 {
+define hidden void @main.foo(ptr %context) unnamed_addr #1 {
 entry:
  call void @"main.foo$1"(%main.kv.0 zeroinitializer, ptr undef)
  ret void
 }

 ; Function Attrs: nounwind
-define internal void @"main.foo$1"(%main.kv.0 %b, ptr %context) unnamed_addr #2 {
+define internal void @"main.foo$1"(%main.kv.0 %b, ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind }
@@ -6,76 +6,73 @@ target triple = "wasm32-unknown-wasi"
 %runtime.channelOp = type { ptr, ptr, i32, ptr }
 %runtime.chanSelectState = type { ptr, ptr }

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.chanIntSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanIntSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
  %chan.op = alloca %runtime.channelOp, align 8
  %chan.value = alloca i32, align 4
  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %chan.value)
  store i32 3, ptr %chan.value, align 4
  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
-  call void @runtime.chanSend(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #4
+  call void @runtime.chanSend(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #3
  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %chan.value)
  ret void
 }

 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
-declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #2

-declare void @runtime.chanSend(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #1
+declare void @runtime.chanSend(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #0

 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
-declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #3
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #2

 ; Function Attrs: nounwind
-define hidden void @main.chanIntRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanIntRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
  %chan.op = alloca %runtime.channelOp, align 8
  %chan.value = alloca i32, align 4
  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %chan.value)
  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
-  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #4
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #3
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %chan.value)
  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  ret void
 }

-declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #1
+declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.chanZeroSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanZeroSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
  %chan.op = alloca %runtime.channelOp, align 8
  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
-  call void @runtime.chanSend(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #4
+  call void @runtime.chanSend(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #3
  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.chanZeroRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanZeroRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
  %chan.op = alloca %runtime.channelOp, align 8
  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
-  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #4
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #3
  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.selectZeroRecv(ptr dereferenceable_or_null(36) %ch1, ptr dereferenceable_or_null(36) %ch2, ptr %context) unnamed_addr #2 {
+define hidden void @main.selectZeroRecv(ptr dereferenceable_or_null(36) %ch1, ptr dereferenceable_or_null(36) %ch2, ptr %context) unnamed_addr #1 {
 entry:
  %select.states.alloca = alloca [2 x %runtime.chanSelectState], align 8
  %select.send.value = alloca i32, align 4
@@ -88,7 +85,7 @@ entry:
  store ptr %ch2, ptr %0, align 4
  %.repack3 = getelementptr inbounds nuw i8, ptr %select.states.alloca, i32 12
  store ptr null, ptr %.repack3, align 4
-  %select.result = call { i32, i1 } @runtime.chanSelect(ptr undef, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr null, i32 0, i32 0, ptr undef) #4
+  %select.result = call { i32, i1 } @runtime.chanSelect(ptr undef, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr null, i32 0, i32 0, ptr undef) #3
  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %select.states.alloca)
  %1 = extractvalue { i32, i1 } %select.result, 0
  %2 = icmp eq i32 %1, 0
@@ -105,10 +102,9 @@ select.body:                                      ; preds = %select.next
  br label %select.done
 }

-declare { i32, i1 } @runtime.chanSelect(ptr, ptr, i32, i32, ptr, i32, i32, ptr) #1
+declare { i32, i1 } @runtime.chanSelect(ptr, ptr, i32, i32, ptr, i32, i32, ptr) #0

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
-attributes #4 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #3 = { nounwind }
@@ -3,32 +3,29 @@ source_filename = "defer.go"
 target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
 target triple = "thumbv7m-unknown-unknown-eabi"

-%runtime.deferFrame = type { ptr, ptr, [0 x ptr], ptr, i8, %runtime._interface }
+%runtime.deferFrame = type { ptr, ptr, [0 x ptr], ptr, i8, %runtime._interface, ptr }
 %runtime._interface = type { ptr, ptr }

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #1 {
+define hidden void @main.init(ptr %context) unnamed_addr #0 {
 entry:
  ret void
 }

-declare void @main.external(ptr) #2
+declare void @main.external(ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.deferSimple(ptr %context) unnamed_addr #1 {
+define hidden void @main.deferSimple(ptr %context) unnamed_addr #0 {
 entry:
  %defer.alloca = alloca { i32, ptr }, align 4
-  %deferPtr = alloca ptr, align 4
-  store ptr null, ptr %deferPtr, align 4
  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %deferPtr = getelementptr inbounds nuw i8, ptr %deferframe.buf, i32 24
  %0 = call ptr @llvm.stacksave.p0()
  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
+  %defer.next = load ptr, ptr %deferPtr, align 4
  store i32 0, ptr %defer.alloca, align 4
  %defer.alloca.repack15 = getelementptr inbounds nuw i8, ptr %defer.alloca, i32 4
-  store ptr null, ptr %defer.alloca.repack15, align 4
+  store ptr %defer.next, ptr %defer.alloca.repack15, align 4
  store ptr %defer.alloca, ptr %deferPtr, align 4
  %setjmp = call i32 asm "\0Amovs r0, #0\0Amov r2, pc\0Astr r2, [r1, #4]", "={r0},{r1},~{r1},~{r2},~{r3},~{r4},~{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{lr},~{q0},~{q1},~{q2},~{q3},~{q4},~{q5},~{q6},~{q7},~{q8},~{q9},~{q10},~{q11},~{q12},~{q13},~{q14},~{q15},~{cpsr},~{memory}"(ptr nonnull %deferframe.buf) #5
  %setjmp.result = icmp eq i32 %setjmp, 0
@@ -112,14 +109,14 @@ rundefers.end3:                                   ; preds = %rundefers.loophead6
 }

 ; Function Attrs: nocallback nofree nosync nounwind willreturn
-declare ptr @llvm.stacksave.p0() #3
+declare ptr @llvm.stacksave.p0() #2

-declare void @runtime.setupDeferFrame(ptr dereferenceable_or_null(24), ptr, ptr) #2
+declare void @runtime.setupDeferFrame(ptr dereferenceable_or_null(28), ptr, ptr) #1

-declare void @runtime.destroyDeferFrame(ptr dereferenceable_or_null(24), ptr) #2
+declare void @runtime.destroyDeferFrame(ptr dereferenceable_or_null(28), ptr) #1

 ; Function Attrs: nounwind
-define internal void @"main.deferSimple$1"(ptr %context) unnamed_addr #1 {
+define internal void @"main.deferSimple$1"(ptr %context) unnamed_addr #0 {
 entry:
  call void @runtime.printlock(ptr undef) #4
  call void @runtime.printint32(i32 3, ptr undef) #4
@@ -127,29 +124,29 @@ entry:
  ret void
 }

-declare void @runtime.printlock(ptr) #2
+declare void @runtime.printlock(ptr) #1

-declare void @runtime.printint32(i32, ptr) #2
+declare void @runtime.printint32(i32, ptr) #1

-declare void @runtime.printunlock(ptr) #2
+declare void @runtime.printunlock(ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.deferMultiple(ptr %context) unnamed_addr #1 {
+define hidden void @main.deferMultiple(ptr %context) unnamed_addr #0 {
 entry:
  %defer.alloca2 = alloca { i32, ptr }, align 4
  %defer.alloca = alloca { i32, ptr }, align 4
-  %deferPtr = alloca ptr, align 4
-  store ptr null, ptr %deferPtr, align 4
  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %deferPtr = getelementptr inbounds nuw i8, ptr %deferframe.buf, i32 24
  %0 = call ptr @llvm.stacksave.p0()
  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
+  %defer.next = load ptr, ptr %deferPtr, align 4
  store i32 0, ptr %defer.alloca, align 4
  %defer.alloca.repack22 = getelementptr inbounds nuw i8, ptr %defer.alloca, i32 4
-  store ptr null, ptr %defer.alloca.repack22, align 4
+  store ptr %defer.next, ptr %defer.alloca.repack22, align 4
  store ptr %defer.alloca, ptr %deferPtr, align 4
  store i32 1, ptr %defer.alloca2, align 4
-  %defer.alloca2.repack23 = getelementptr inbounds nuw i8, ptr %defer.alloca2, i32 4
-  store ptr %defer.alloca, ptr %defer.alloca2.repack23, align 4
+  %defer.alloca2.repack24 = getelementptr inbounds nuw i8, ptr %defer.alloca2, i32 4
+  store ptr %defer.alloca, ptr %defer.alloca2.repack24, align 4
  store ptr %defer.alloca2, ptr %deferPtr, align 4
  %setjmp = call i32 asm "\0Amovs r0, #0\0Amov r2, pc\0Astr r2, [r1, #4]", "={r0},{r1},~{r1},~{r2},~{r3},~{r4},~{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{lr},~{q0},~{q1},~{q2},~{q3},~{q4},~{q5},~{q6},~{q7},~{q8},~{q9},~{q10},~{q11},~{q12},~{q13},~{q14},~{q15},~{cpsr},~{memory}"(ptr nonnull %deferframe.buf) #5
  %setjmp.result = icmp eq i32 %setjmp, 0
@@ -253,7 +250,7 @@ rundefers.end7:                                   ; preds = %rundefers.loophead1
 }

 ; Function Attrs: nounwind
-define internal void @"main.deferMultiple$1"(ptr %context) unnamed_addr #1 {
+define internal void @"main.deferMultiple$1"(ptr %context) unnamed_addr #0 {
 entry:
  call void @runtime.printlock(ptr undef) #4
  call void @runtime.printint32(i32 3, ptr undef) #4
@@ -262,7 +259,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define internal void @"main.deferMultiple$2"(ptr %context) unnamed_addr #1 {
+define internal void @"main.deferMultiple$2"(ptr %context) unnamed_addr #0 {
 entry:
  call void @runtime.printlock(ptr undef) #4
  call void @runtime.printint32(i32 5, ptr undef) #4
@@ -271,18 +268,17 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.deferInfiniteLoop(ptr %context) unnamed_addr #1 {
+define hidden void @main.deferInfiniteLoop(ptr %context) unnamed_addr #0 {
 entry:
-  %deferPtr = alloca ptr, align 4
-  store ptr null, ptr %deferPtr, align 4
  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %deferPtr = getelementptr inbounds nuw i8, ptr %deferframe.buf, i32 24
  %0 = call ptr @llvm.stacksave.p0()
  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
  br label %for.body

 for.body:                                         ; preds = %for.body, %entry
  %defer.next = load ptr, ptr %deferPtr, align 4
-  %defer.alloc.call = call dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #4
+  %defer.alloc.call = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 135 to ptr), ptr undef) #4
  store i32 0, ptr %defer.alloc.call, align 4
  %defer.alloc.call.repack1 = getelementptr inbounds nuw i8, ptr %defer.alloc.call, i32 4
  store ptr %defer.next, ptr %defer.alloc.call.repack1, align 4
@@ -315,12 +311,14 @@ rundefers.end:                                    ; preds = %rundefers.loophead
  br label %recover
 }

+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #3
+
 ; Function Attrs: nounwind
-define hidden void @main.deferLoop(ptr %context) unnamed_addr #1 {
+define hidden void @main.deferLoop(ptr %context) unnamed_addr #0 {
 entry:
-  %deferPtr = alloca ptr, align 4
-  store ptr null, ptr %deferPtr, align 4
  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %deferPtr = getelementptr inbounds nuw i8, ptr %deferframe.buf, i32 24
  %0 = call ptr @llvm.stacksave.p0()
  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
  br label %for.loop
@@ -332,7 +330,7 @@ for.loop:                                         ; preds = %for.body, %entry

 for.body:                                         ; preds = %for.loop
  %defer.next = load ptr, ptr %deferPtr, align 4
-  %defer.alloc.call = call dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #4
+  %defer.alloc.call = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 135 to ptr), ptr undef) #4
  store i32 0, ptr %defer.alloc.call, align 4
  %defer.alloc.call.repack13 = getelementptr inbounds nuw i8, ptr %defer.alloc.call, i32 4
  store ptr %defer.next, ptr %defer.alloc.call.repack13, align 4
@@ -405,12 +403,11 @@ rundefers.end1:                                   ; preds = %rundefers.loophead4
 }

 ; Function Attrs: nounwind
-define hidden void @main.deferBetweenLoops(ptr %context) unnamed_addr #1 {
+define hidden void @main.deferBetweenLoops(ptr %context) unnamed_addr #0 {
 entry:
  %defer.alloca = alloca { i32, ptr, i32 }, align 4
-  %deferPtr = alloca ptr, align 4
-  store ptr null, ptr %deferPtr, align 4
  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %deferPtr = getelementptr inbounds nuw i8, ptr %deferframe.buf, i32 24
  %0 = call ptr @llvm.stacksave.p0()
  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
  br label %for.loop
@@ -505,9 +502,9 @@ rundefers.end4:                                   ; preds = %rundefers.loophead7
  br label %recover
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #1 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #2 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #3 = { nocallback nofree nosync nounwind willreturn }
+attributes #0 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #1 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #2 = { nocallback nofree nosync nounwind willreturn }
+attributes #3 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
 attributes #4 = { nounwind }
 attributes #5 = { nounwind returns_twice }
@@ -3,19 +3,16 @@ source_filename = "float.go"
 target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.f32tou32(float %v, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.f32tou32(float %v, ptr %context) unnamed_addr #1 {
 entry:
  %positive = fcmp oge float %v, 0.000000e+00
  %withinmax = fcmp ole float %v, 0x41EFFFFFC0000000
@@ -27,25 +24,25 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden float @main.maxu32f(ptr %context) unnamed_addr #2 {
+define hidden float @main.maxu32f(ptr %context) unnamed_addr #1 {
 entry:
  ret float 0x41F0000000000000
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.maxu32tof32(ptr %context) unnamed_addr #2 {
+define hidden i32 @main.maxu32tof32(ptr %context) unnamed_addr #1 {
 entry:
  ret i32 -1
 }

 ; Function Attrs: nounwind
-define hidden { i32, i32, i32, i32 } @main.inftoi32(ptr %context) unnamed_addr #2 {
+define hidden { i32, i32, i32, i32 } @main.inftoi32(ptr %context) unnamed_addr #1 {
 entry:
  ret { i32, i32, i32, i32 } { i32 -1, i32 0, i32 2147483647, i32 -2147483648 }
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.u32tof32tou32(i32 %v, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.u32tof32tou32(i32 %v, ptr %context) unnamed_addr #1 {
 entry:
  %0 = uitofp i32 %v to float
  %withinmax = fcmp ole float %0, 0x41EFFFFFC0000000
@@ -55,7 +52,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden float @main.f32tou32tof32(float %v, ptr %context) unnamed_addr #2 {
+define hidden float @main.f32tou32tof32(float %v, ptr %context) unnamed_addr #1 {
 entry:
  %positive = fcmp oge float %v, 0.000000e+00
  %withinmax = fcmp ole float %v, 0x41EFFFFFC0000000
@@ -68,7 +65,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden i8 @main.f32tou8(float %v, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.f32tou8(float %v, ptr %context) unnamed_addr #1 {
 entry:
  %positive = fcmp oge float %v, 0.000000e+00
  %withinmax = fcmp ole float %v, 2.550000e+02
@@ -80,7 +77,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden i8 @main.f32toi8(float %v, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.f32toi8(float %v, ptr %context) unnamed_addr #1 {
 entry:
  %abovemin = fcmp oge float %v, -1.280000e+02
  %belowmax = fcmp ole float %v, 1.270000e+02
@@ -93,6 +90,5 @@ entry:
  ret i8 %0
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
@@ -3,48 +3,44 @@ source_filename = "func.go"
 target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.foo(ptr %callback.context, ptr %callback.funcptr, ptr %context) unnamed_addr #2 {
+define hidden void @main.foo(ptr %callback.context, ptr %callback.funcptr, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp eq ptr %callback.funcptr, null
  br i1 %0, label %fpcall.throw, label %fpcall.next

 fpcall.next:                                      ; preds = %entry
-  call void %callback.funcptr(i32 3, ptr %callback.context) #3
+  call void %callback.funcptr(i32 3, ptr %callback.context) #2
  ret void

 fpcall.throw:                                     ; preds = %entry
-  call void @runtime.nilPanic(ptr undef) #3
+  call void @runtime.nilPanic(ptr undef) #2
  unreachable
 }

-declare void @runtime.nilPanic(ptr) #1
+declare void @runtime.nilPanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.bar(ptr %context) unnamed_addr #2 {
+define hidden void @main.bar(ptr %context) unnamed_addr #1 {
 entry:
  call void @main.foo(ptr undef, ptr nonnull @main.someFunc, ptr undef)
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.someFunc(i32 %arg0, ptr %context) unnamed_addr #2 {
+define hidden void @main.someFunc(i32 %arg0, ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind }
@@ -25,19 +25,16 @@ target triple = "wasm32-unknown-wasi"
@"reflect/types.type:basic:complex128" = linkonce_odr constant { i8, ptr } { i8 80, ptr @"reflect/types.type:pointer:basic:complex128" }, align 4
@"reflect/types.type:pointer:basic:complex128" = linkonce_odr constant { i8, i16, ptr } { i8 -43, i16 0, ptr @"reflect/types.type:basic:complex128" }, align 4

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.newScalar(ptr %context) unnamed_addr #2 {
+define hidden void @main.newScalar(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %new = call align 1 dereferenceable(1) ptr @runtime.alloc(i32 1, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
@@ -55,8 +52,11 @@ entry:
  ret void
 }

+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #2
+
 ; Function Attrs: nounwind
-define hidden void @main.newArray(ptr %context) unnamed_addr #2 {
+define hidden void @main.newArray(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %new = call align 1 dereferenceable(3) ptr @runtime.alloc(i32 3, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
@@ -72,10 +72,10 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.newStruct(ptr %context) unnamed_addr #2 {
+define hidden void @main.newStruct(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  %new = call align 1 ptr @runtime.alloc(i32 0, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new = call align 1 ptr @runtime.alloc_zero(i32 0, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
  call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #3
  store ptr %new, ptr @main.struct1, align 4
  %new1 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
@@ -93,8 +93,11 @@ entry:
  ret void
 }

+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc_zero(i32, ptr, ptr) #2
+
 ; Function Attrs: nounwind
-define hidden ptr @main.newFuncValue(ptr %context) unnamed_addr #2 {
+define hidden ptr @main.newFuncValue(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %new = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 197 to ptr), ptr undef) #3
@@ -103,7 +106,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.makeSlice(ptr %context) unnamed_addr #2 {
+define hidden void @main.makeSlice(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %makeslice = call align 1 dereferenceable(5) ptr @runtime.alloc(i32 5, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
@@ -125,10 +128,10 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden %runtime._interface @main.makeInterface(double %v.r, double %v.i, ptr %context) unnamed_addr #2 {
+define hidden %runtime._interface @main.makeInterface(double %v.r, double %v.i, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  %0 = call align 8 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #3
+  %0 = call align 8 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #3
  store double %v.r, ptr %0, align 8
  %.repack1 = getelementptr inbounds nuw i8, ptr %0, i32 8
@@ -139,7 +142,7 @@ entry:
  ret %runtime._interface %1
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #3 = { nounwind }
@@ -19,12 +19,49 @@ func Add[T Coord](a, b Point[T]) Point[T] {
 	}
 }

+func aliasSize[F float32 | float64]() uintptr {
+	return unsafe.Sizeof(F(0))
+}
+
+func aliasSize32() uintptr {
+	type F = float32
+	return aliasSize[F]()
+}
+
+func aliasSize64() uintptr {
+	type F = float64
+	return aliasSize[F]()
+}
+
+type aliasMethodResult[F float32 | float64] struct {
+	Value F
+}
+
+type aliasMethodValue[F float32 | float64] struct{}
+
+func (aliasMethodValue[F]) Get() aliasMethodResult[F] {
+	return aliasMethodResult[F]{}
+}
+
 func main() {
 	var af, bf Point[float32]
 	Add(af, bf)

 	var ai, bi Point[int]
 	Add(ai, bi)
+
+	checkSize(aliasSize32())
+	checkSize(aliasSize64())
 }

 func checkSize(uintptr)
+
+func checkBool(bool)
+
+func aliasMethod32(x any) {
+	type F = float32
+	_, ok := x.(interface {
+		Get() aliasMethodResult[F]
+	})
+	checkBool(ok)
+}
@@ -1,259 +1,196 @@
 ; ModuleID = 'generics.go'
 source_filename = "generics.go"
-target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-n32:64-S128-ni:1:10:20"
+target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"

-%"main.Point[int]" = type { i32, i32 }
 %"main.Point[float32]" = type { float, float }
+%"main.Point[int]" = type { i32, i32 }

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*) #0
-
-declare void @runtime.trackPointer(i8* nocapture readonly, i8*) #0
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(i8* %context) unnamed_addr #1 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.main(i8* %context) unnamed_addr #1 {
+define hidden i32 @main.aliasSize32(ptr %context) unnamed_addr #1 {
 entry:
-  %bi = alloca %"main.Point[int]", align 8
-  %ai = alloca %"main.Point[int]", align 8
-  %bf = alloca %"main.Point[float32]", align 8
-  %af = alloca %"main.Point[float32]", align 8
-  %af.repack = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %af, i32 0, i32 0
-  store float 0.000000e+00, float* %af.repack, align 8
-  %af.repack1 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %af, i32 0, i32 1
-  store float 0.000000e+00, float* %af.repack1, align 4
-  %0 = bitcast %"main.Point[float32]"* %af to i8*
-  call void @runtime.trackPointer(i8* nonnull %0, i8* undef) #2
-  %bf.repack = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %bf, i32 0, i32 0
-  store float 0.000000e+00, float* %bf.repack, align 8
-  %bf.repack2 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %bf, i32 0, i32 1
-  store float 0.000000e+00, float* %bf.repack2, align 4
-  %1 = bitcast %"main.Point[float32]"* %bf to i8*
-  call void @runtime.trackPointer(i8* nonnull %1, i8* undef) #2
-  %.elt = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %af, i32 0, i32 0
-  %.unpack = load float, float* %.elt, align 8
-  %.elt3 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %af, i32 0, i32 1
-  %.unpack4 = load float, float* %.elt3, align 4
-  %.elt5 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %bf, i32 0, i32 0
-  %.unpack6 = load float, float* %.elt5, align 8
-  %.elt7 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %bf, i32 0, i32 1
-  %.unpack8 = load float, float* %.elt7, align 4
-  %2 = call %"main.Point[float32]" @"main.Add[float32]"(float %.unpack, float %.unpack4, float %.unpack6, float %.unpack8, i8* undef)
-  %ai.repack = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %ai, i32 0, i32 0
-  store i32 0, i32* %ai.repack, align 8
-  %ai.repack9 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %ai, i32 0, i32 1
-  store i32 0, i32* %ai.repack9, align 4
-  %3 = bitcast %"main.Point[int]"* %ai to i8*
-  call void @runtime.trackPointer(i8* nonnull %3, i8* undef) #2
-  %bi.repack = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %bi, i32 0, i32 0
-  store i32 0, i32* %bi.repack, align 8
-  %bi.repack10 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %bi, i32 0, i32 1
-  store i32 0, i32* %bi.repack10, align 4
-  %4 = bitcast %"main.Point[int]"* %bi to i8*
-  call void @runtime.trackPointer(i8* nonnull %4, i8* undef) #2
-  %.elt11 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %ai, i32 0, i32 0
-  %.unpack12 = load i32, i32* %.elt11, align 8
-  %.elt13 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %ai, i32 0, i32 1
-  %.unpack14 = load i32, i32* %.elt13, align 4
-  %.elt15 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %bi, i32 0, i32 0
-  %.unpack16 = load i32, i32* %.elt15, align 8
-  %.elt17 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %bi, i32 0, i32 1
-  %.unpack18 = load i32, i32* %.elt17, align 4
-  %5 = call %"main.Point[int]" @"main.Add[int]"(i32 %.unpack12, i32 %.unpack14, i32 %.unpack16, i32 %.unpack18, i8* undef)
+  %0 = call i32 @"main.aliasSize[basic:float32]"(ptr undef)
+  ret i32 %0
+}
+
+; Function Attrs: nounwind
+define linkonce_odr hidden i32 @"main.aliasSize[basic:float32]"(ptr %context) unnamed_addr #1 {
+entry:
+  ret i32 4
+}
+
+; Function Attrs: nounwind
+define hidden i32 @main.aliasSize64(ptr %context) unnamed_addr #1 {
+entry:
+  %0 = call i32 @"main.aliasSize[basic:float64]"(ptr undef)
+  ret i32 %0
+}
+
+; Function Attrs: nounwind
+define linkonce_odr hidden i32 @"main.aliasSize[basic:float64]"(ptr %context) unnamed_addr #1 {
+entry:
+  ret i32 8
+}
+
+; Function Attrs: nounwind
+define hidden void @main.main(ptr %context) unnamed_addr #1 {
+entry:
+  %0 = call %"main.Point[float32]" @"main.Add[basic:float32]"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, ptr undef)
+  %1 = call %"main.Point[int]" @"main.Add[basic:int]"(i32 0, i32 0, i32 0, i32 0, ptr undef)
+  %2 = call i32 @main.aliasSize32(ptr undef)
+  call void @main.checkSize(i32 %2, ptr undef) #4
+  %3 = call i32 @main.aliasSize64(ptr undef)
+  call void @main.checkSize(i32 %3, ptr undef) #4
  ret void
 }

 ; Function Attrs: nounwind
-define linkonce_odr hidden %"main.Point[float32]" @"main.Add[float32]"(float %a.X, float %a.Y, float %b.X, float %b.Y, i8* %context) unnamed_addr #1 {
+define linkonce_odr hidden %"main.Point[float32]" @"main.Add[basic:float32]"(float %a.X, float %a.Y, float %b.X, float %b.Y, ptr %context) unnamed_addr #1 {
 entry:
-  %complit = alloca %"main.Point[float32]", align 8
-  %b = alloca %"main.Point[float32]", align 8
-  %a = alloca %"main.Point[float32]", align 8
-  %a.repack = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %a, i32 0, i32 0
-  store float 0.000000e+00, float* %a.repack, align 8
-  %a.repack9 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %a, i32 0, i32 1
-  store float 0.000000e+00, float* %a.repack9, align 4
-  %0 = bitcast %"main.Point[float32]"* %a to i8*
-  call void @runtime.trackPointer(i8* nonnull %0, i8* undef) #2
-  %a.repack10 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %a, i32 0, i32 0
-  store float %a.X, float* %a.repack10, align 8
-  %a.repack11 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %a, i32 0, i32 1
-  store float %a.Y, float* %a.repack11, align 4
-  %b.repack = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %b, i32 0, i32 0
-  store float 0.000000e+00, float* %b.repack, align 8
-  %b.repack13 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %b, i32 0, i32 1
-  store float 0.000000e+00, float* %b.repack13, align 4
-  %1 = bitcast %"main.Point[float32]"* %b to i8*
-  call void @runtime.trackPointer(i8* nonnull %1, i8* undef) #2
-  %b.repack14 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %b, i32 0, i32 0
-  store float %b.X, float* %b.repack14, align 8
-  %b.repack15 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %b, i32 0, i32 1
-  store float %b.Y, float* %b.repack15, align 4
-  call void @main.checkSize(i32 4, i8* undef) #2
-  call void @main.checkSize(i32 8, i8* undef) #2
-  %complit.repack = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %complit, i32 0, i32 0
-  store float 0.000000e+00, float* %complit.repack, align 8
-  %complit.repack17 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %complit, i32 0, i32 1
-  store float 0.000000e+00, float* %complit.repack17, align 4
-  %2 = bitcast %"main.Point[float32]"* %complit to i8*
-  call void @runtime.trackPointer(i8* nonnull %2, i8* undef) #2
-  %3 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %complit, i32 0, i32 0
+  %stackalloc = alloca i8, align 1
+  %a = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #4
+  call void @runtime.trackPointer(ptr nonnull %a, ptr nonnull %stackalloc, ptr undef) #4
+  store float %a.X, ptr %a, align 4
+  %a.repack5 = getelementptr inbounds nuw i8, ptr %a, i32 4
+  store float %a.Y, ptr %a.repack5, align 4
+  %b = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #4
+  call void @runtime.trackPointer(ptr nonnull %b, ptr nonnull %stackalloc, ptr undef) #4
+  store float %b.X, ptr %b, align 4
+  %b.repack7 = getelementptr inbounds nuw i8, ptr %b, i32 4
+  store float %b.Y, ptr %b.repack7, align 4
+  call void @main.checkSize(i32 4, ptr undef) #4
+  call void @main.checkSize(i32 8, ptr undef) #4
+  %complit = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #4
+  call void @runtime.trackPointer(ptr nonnull %complit, ptr nonnull %stackalloc, ptr undef) #4
  br i1 false, label %deref.throw, label %deref.next

 deref.next:                                       ; preds = %entry
-  br i1 false, label %deref.throw1, label %deref.next2
+  br i1 false, label %deref.throw, label %deref.next1

-deref.next2:                                      ; preds = %deref.next
-  %4 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %b, i32 0, i32 0
-  %5 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %a, i32 0, i32 0
-  %6 = load float, float* %5, align 8
-  %7 = load float, float* %4, align 8
-  %8 = fadd float %6, %7
-  br i1 false, label %deref.throw3, label %deref.next4
+deref.next1:                                      ; preds = %deref.next
+  %0 = load float, ptr %a, align 4
+  %1 = load float, ptr %b, align 4
+  %2 = fadd float %0, %1
+  br i1 false, label %deref.throw, label %deref.next2

-deref.next4:                                      ; preds = %deref.next2
-  br i1 false, label %deref.throw5, label %deref.next6
+deref.next2:                                      ; preds = %deref.next1
+  br i1 false, label %deref.throw, label %deref.next3

-deref.next6:                                      ; preds = %deref.next4
-  %9 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %b, i32 0, i32 1
-  %10 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %a, i32 0, i32 1
-  %11 = load float, float* %10, align 4
-  %12 = load float, float* %9, align 4
-  br i1 false, label %store.throw, label %store.next
+deref.next3:                                      ; preds = %deref.next2
+  %3 = getelementptr inbounds nuw i8, ptr %b, i32 4
+  %4 = getelementptr inbounds nuw i8, ptr %a, i32 4
+  %5 = load float, ptr %4, align 4
+  %6 = load float, ptr %3, align 4
+  br i1 false, label %deref.throw, label %store.next

-store.next:                                       ; preds = %deref.next6
-  store float %8, float* %3, align 8
-  br i1 false, label %store.throw7, label %store.next8
+store.next:                                       ; preds = %deref.next3
+  store float %2, ptr %complit, align 4
+  br i1 false, label %deref.throw, label %store.next4

-store.next8:                                      ; preds = %store.next
-  %13 = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %complit, i32 0, i32 1
-  %14 = fadd float %11, %12
-  store float %14, float* %13, align 4
-  %.elt = getelementptr inbounds %"main.Point[float32]", %"main.Point[float32]"* %complit, i32 0, i32 0
-  %.unpack = load float, float* %.elt, align 8
-  %15 = insertvalue %"main.Point[float32]" undef, float %.unpack, 0
-  %16 = insertvalue %"main.Point[float32]" %15, float %14, 1
-  ret %"main.Point[float32]" %16
+store.next4:                                      ; preds = %store.next
+  %7 = getelementptr inbounds nuw i8, ptr %complit, i32 4
+  %8 = fadd float %5, %6
+  store float %8, ptr %7, align 4
+  %.unpack = load float, ptr %complit, align 4
+  %9 = insertvalue %"main.Point[float32]" poison, float %.unpack, 0
+  %10 = insertvalue %"main.Point[float32]" %9, float %8, 1
+  ret %"main.Point[float32]" %10

-deref.throw:                                      ; preds = %entry
-  unreachable
-
-deref.throw1:                                     ; preds = %deref.next
-  unreachable
-
-deref.throw3:                                     ; preds = %deref.next2
-  unreachable
-
-deref.throw5:                                     ; preds = %deref.next4
-  unreachable
-
-store.throw:                                      ; preds = %deref.next6
-  unreachable
-
-store.throw7:                                     ; preds = %store.next
+deref.throw:                                      ; preds = %store.next, %deref.next3, %deref.next2, %deref.next1, %deref.next, %entry
  unreachable
 }

-declare void @main.checkSize(i32, i8*) #0
+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #2

-declare void @runtime.nilPanic(i8*) #0
+declare void @main.checkSize(i32, ptr) #0
+
+declare void @runtime.nilPanic(ptr) #0

 ; Function Attrs: nounwind
-define linkonce_odr hidden %"main.Point[int]" @"main.Add[int]"(i32 %a.X, i32 %a.Y, i32 %b.X, i32 %b.Y, i8* %context) unnamed_addr #1 {
+define linkonce_odr hidden %"main.Point[int]" @"main.Add[basic:int]"(i32 %a.X, i32 %a.Y, i32 %b.X, i32 %b.Y, ptr %context) unnamed_addr #1 {
 entry:
-  %complit = alloca %"main.Point[int]", align 8
-  %b = alloca %"main.Point[int]", align 8
-  %a = alloca %"main.Point[int]", align 8
-  %a.repack = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %a, i32 0, i32 0
-  store i32 0, i32* %a.repack, align 8
-  %a.repack9 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %a, i32 0, i32 1
-  store i32 0, i32* %a.repack9, align 4
-  %0 = bitcast %"main.Point[int]"* %a to i8*
-  call void @runtime.trackPointer(i8* nonnull %0, i8* undef) #2
-  %a.repack10 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %a, i32 0, i32 0
-  store i32 %a.X, i32* %a.repack10, align 8
-  %a.repack11 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %a, i32 0, i32 1
-  store i32 %a.Y, i32* %a.repack11, align 4
-  %b.repack = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %b, i32 0, i32 0
-  store i32 0, i32* %b.repack, align 8
-  %b.repack13 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %b, i32 0, i32 1
-  store i32 0, i32* %b.repack13, align 4
-  %1 = bitcast %"main.Point[int]"* %b to i8*
-  call void @runtime.trackPointer(i8* nonnull %1, i8* undef) #2
-  %b.repack14 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %b, i32 0, i32 0
-  store i32 %b.X, i32* %b.repack14, align 8
-  %b.repack15 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %b, i32 0, i32 1
-  store i32 %b.Y, i32* %b.repack15, align 4
-  call void @main.checkSize(i32 4, i8* undef) #2
-  call void @main.checkSize(i32 8, i8* undef) #2
-  %complit.repack = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %complit, i32 0, i32 0
-  store i32 0, i32* %complit.repack, align 8
-  %complit.repack17 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %complit, i32 0, i32 1
-  store i32 0, i32* %complit.repack17, align 4
-  %2 = bitcast %"main.Point[int]"* %complit to i8*
-  call void @runtime.trackPointer(i8* nonnull %2, i8* undef) #2
-  %3 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %complit, i32 0, i32 0
+  %stackalloc = alloca i8, align 1
+  %a = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #4
+  call void @runtime.trackPointer(ptr nonnull %a, ptr nonnull %stackalloc, ptr undef) #4
+  store i32 %a.X, ptr %a, align 4
+  %a.repack5 = getelementptr inbounds nuw i8, ptr %a, i32 4
+  store i32 %a.Y, ptr %a.repack5, align 4
+  %b = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #4
+  call void @runtime.trackPointer(ptr nonnull %b, ptr nonnull %stackalloc, ptr undef) #4
+  store i32 %b.X, ptr %b, align 4
+  %b.repack7 = getelementptr inbounds nuw i8, ptr %b, i32 4
+  store i32 %b.Y, ptr %b.repack7, align 4
+  call void @main.checkSize(i32 4, ptr undef) #4
+  call void @main.checkSize(i32 8, ptr undef) #4
+  %complit = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #4
+  call void @runtime.trackPointer(ptr nonnull %complit, ptr nonnull %stackalloc, ptr undef) #4
  br i1 false, label %deref.throw, label %deref.next

 deref.next:                                       ; preds = %entry
-  br i1 false, label %deref.throw1, label %deref.next2
+  br i1 false, label %deref.throw, label %deref.next1

-deref.next2:                                      ; preds = %deref.next
-  %4 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %b, i32 0, i32 0
-  %5 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %a, i32 0, i32 0
-  %6 = load i32, i32* %5, align 8
-  %7 = load i32, i32* %4, align 8
-  %8 = add i32 %6, %7
-  br i1 false, label %deref.throw3, label %deref.next4
+deref.next1:                                      ; preds = %deref.next
+  %0 = load i32, ptr %a, align 4
+  %1 = load i32, ptr %b, align 4
+  %2 = add i32 %0, %1
+  br i1 false, label %deref.throw, label %deref.next2

-deref.next4:                                      ; preds = %deref.next2
-  br i1 false, label %deref.throw5, label %deref.next6
+deref.next2:                                      ; preds = %deref.next1
+  br i1 false, label %deref.throw, label %deref.next3

-deref.next6:                                      ; preds = %deref.next4
-  %9 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %b, i32 0, i32 1
-  %10 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %a, i32 0, i32 1
-  %11 = load i32, i32* %10, align 4
-  %12 = load i32, i32* %9, align 4
-  br i1 false, label %store.throw, label %store.next
+deref.next3:                                      ; preds = %deref.next2
+  %3 = getelementptr inbounds nuw i8, ptr %b, i32 4
+  %4 = getelementptr inbounds nuw i8, ptr %a, i32 4
+  %5 = load i32, ptr %4, align 4
+  %6 = load i32, ptr %3, align 4
+  br i1 false, label %deref.throw, label %store.next

-store.next:                                       ; preds = %deref.next6
-  store i32 %8, i32* %3, align 8
-  br i1 false, label %store.throw7, label %store.next8
+store.next:                                       ; preds = %deref.next3
+  store i32 %2, ptr %complit, align 4
+  br i1 false, label %deref.throw, label %store.next4

-store.next8:                                      ; preds = %store.next
-  %13 = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %complit, i32 0, i32 1
-  %14 = add i32 %11, %12
-  store i32 %14, i32* %13, align 4
-  %.elt = getelementptr inbounds %"main.Point[int]", %"main.Point[int]"* %complit, i32 0, i32 0
-  %.unpack = load i32, i32* %.elt, align 8
-  %15 = insertvalue %"main.Point[int]" undef, i32 %.unpack, 0
-  %16 = insertvalue %"main.Point[int]" %15, i32 %14, 1
-  ret %"main.Point[int]" %16
+store.next4:                                      ; preds = %store.next
+  %7 = getelementptr inbounds nuw i8, ptr %complit, i32 4
+  %8 = add i32 %5, %6
+  store i32 %8, ptr %7, align 4
+  %.unpack = load i32, ptr %complit, align 4
+  %9 = insertvalue %"main.Point[int]" poison, i32 %.unpack, 0
+  %10 = insertvalue %"main.Point[int]" %9, i32 %8, 1
+  ret %"main.Point[int]" %10

-deref.throw:                                      ; preds = %entry
-  unreachable
-
-deref.throw1:                                     ; preds = %deref.next
-  unreachable
-
-deref.throw3:                                     ; preds = %deref.next2
-  unreachable
-
-deref.throw5:                                     ; preds = %deref.next4
-  unreachable
-
-store.throw:                                      ; preds = %deref.next6
-  unreachable
-
-store.throw7:                                     ; preds = %store.next
+deref.throw:                                      ; preds = %store.next, %deref.next3, %deref.next2, %deref.next1, %deref.next, %entry
  unreachable
 }

-attributes #0 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind }
+declare void @main.checkBool(i1, ptr) #0
+
+; Function Attrs: nounwind
+define hidden void @main.aliasMethod32(ptr %x.typecode, ptr %x.value, ptr %context) unnamed_addr #1 {
+entry:
+  %0 = call i1 @"interface:{Get:func:{}{named:main.aliasMethodResult[basic:float32]}}.$typeassert"(ptr %x.typecode) #4
+  br i1 %0, label %typeassert.ok, label %typeassert.next
+
+typeassert.next:                                  ; preds = %typeassert.ok, %entry
+  call void @main.checkBool(i1 %0, ptr undef) #4
+  ret void
+
+typeassert.ok:                                    ; preds = %entry
+  br label %typeassert.next
+}
+
+declare i1 @"interface:{Get:func:{}{named:main.aliasMethodResult[basic:float32]}}.$typeassert"(ptr) #3
+
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #3 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-methods"="reflect/methods.Get:func:{}{named:main.aliasMethodResult[basic:float32]}" }
+attributes #4 = { nounwind }
@@ -5,27 +5,24 @@ target triple = "wasm32-unknown-wasi"

 %runtime._string = type { ptr, i32 }

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.unsafeSliceData(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.unsafeSliceData(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr %s.data, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %s.data, ptr nonnull %stackalloc, ptr undef) #2
  ret ptr %s.data
 }

 ; Function Attrs: nounwind
-define hidden %runtime._string @main.unsafeString(ptr dereferenceable_or_null(1) %ptr, i16 %len, ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.unsafeString(ptr dereferenceable_or_null(1) %ptr, i16 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = icmp slt i16 %len, 0
@@ -39,25 +36,24 @@ unsafe.String.next:                               ; preds = %entry
  %5 = zext nneg i16 %len to i32
  %6 = insertvalue %runtime._string undef, ptr %ptr, 0
  %7 = insertvalue %runtime._string %6, i32 %5, 1
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #2
  ret %runtime._string %7

 unsafe.String.throw:                              ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #2
  unreachable
 }

-declare void @runtime.unsafeSlicePanic(ptr) #1
+declare void @runtime.unsafeSlicePanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden ptr @main.unsafeStringData(ptr readonly %s.data, i32 %s.len, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.unsafeStringData(ptr readonly %s.data, i32 %s.len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr %s.data, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %s.data, ptr nonnull %stackalloc, ptr undef) #2
  ret ptr %s.data
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind }
@@ -5,35 +5,32 @@ target triple = "wasm32-unknown-wasi"

 %runtime._string = type { ptr, i32 }

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.min1(i32 %a, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.min1(i32 %a, ptr %context) unnamed_addr #1 {
 entry:
  ret i32 %a
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.smin.i32(i32, i32) #3
+declare i32 @llvm.smin.i32(i32, i32) #2

 ; Function Attrs: nounwind
-define hidden i32 @main.min2(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.min2(i32 %a, i32 %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i32 @llvm.smin.i32(i32 %a, i32 %b)
  ret i32 %0
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.min3(i32 %a, i32 %b, i32 %c, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.min3(i32 %a, i32 %b, i32 %c, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i32 @llvm.smin.i32(i32 %a, i32 %b)
  %1 = call i32 @llvm.smin.i32(i32 %0, i32 %c)
@@ -41,7 +38,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.min4(i32 %a, i32 %b, i32 %c, i32 %d, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.min4(i32 %a, i32 %b, i32 %c, i32 %d, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i32 @llvm.smin.i32(i32 %a, i32 %b)
  %1 = call i32 @llvm.smin.i32(i32 %0, i32 %c)
@@ -50,109 +47,109 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden i8 @main.minUint8(i8 %a, i8 %b, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.minUint8(i8 %a, i8 %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i8 @llvm.umin.i8(i8 %a, i8 %b)
  ret i8 %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i8 @llvm.umin.i8(i8, i8) #3
+declare i8 @llvm.umin.i8(i8, i8) #2

 ; Function Attrs: nounwind
-define hidden i32 @main.minUnsigned(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.minUnsigned(i32 %a, i32 %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i32 @llvm.umin.i32(i32 %a, i32 %b)
  ret i32 %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.umin.i32(i32, i32) #3
+declare i32 @llvm.umin.i32(i32, i32) #2

 ; Function Attrs: nounwind
-define hidden float @main.minFloat32(float %a, float %b, ptr %context) unnamed_addr #2 {
+define hidden float @main.minFloat32(float %a, float %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call float @llvm.minimum.f32(float %a, float %b)
  ret float %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare float @llvm.minimum.f32(float, float) #3
+declare float @llvm.minimum.f32(float, float) #2

 ; Function Attrs: nounwind
-define hidden double @main.minFloat64(double %a, double %b, ptr %context) unnamed_addr #2 {
+define hidden double @main.minFloat64(double %a, double %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call double @llvm.minimum.f64(double %a, double %b)
  ret double %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare double @llvm.minimum.f64(double, double) #3
+declare double @llvm.minimum.f64(double, double) #2

 ; Function Attrs: nounwind
-define hidden %runtime._string @main.minString(ptr readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.minString(ptr readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #1 {
 entry:
  %0 = insertvalue %runtime._string zeroinitializer, ptr %a.data, 0
  %1 = insertvalue %runtime._string %0, i32 %a.len, 1
  %2 = insertvalue %runtime._string zeroinitializer, ptr %b.data, 0
  %3 = insertvalue %runtime._string %2, i32 %b.len, 1
  %stackalloc = alloca i8, align 1
-  %4 = call i1 @runtime.stringLess(ptr %a.data, i32 %a.len, ptr %b.data, i32 %b.len, ptr undef) #5
+  %4 = call i1 @runtime.stringLess(ptr %a.data, i32 %a.len, ptr %b.data, i32 %b.len, ptr undef) #4
  %5 = select i1 %4, %runtime._string %1, %runtime._string %3
  %6 = select i1 %4, ptr %a.data, ptr %b.data
-  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #5
+  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #4
  ret %runtime._string %5
 }

-declare i1 @runtime.stringLess(ptr readonly, i32, ptr readonly, i32, ptr) #1
+declare i1 @runtime.stringLess(ptr readonly, i32, ptr readonly, i32, ptr) #0

 ; Function Attrs: nounwind
-define hidden i32 @main.maxInt(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.maxInt(i32 %a, i32 %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i32 @llvm.smax.i32(i32 %a, i32 %b)
  ret i32 %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.smax.i32(i32, i32) #3
+declare i32 @llvm.smax.i32(i32, i32) #2

 ; Function Attrs: nounwind
-define hidden i32 @main.maxUint(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.maxUint(i32 %a, i32 %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call i32 @llvm.umax.i32(i32 %a, i32 %b)
  ret i32 %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.umax.i32(i32, i32) #3
+declare i32 @llvm.umax.i32(i32, i32) #2

 ; Function Attrs: nounwind
-define hidden float @main.maxFloat32(float %a, float %b, ptr %context) unnamed_addr #2 {
+define hidden float @main.maxFloat32(float %a, float %b, ptr %context) unnamed_addr #1 {
 entry:
  %0 = call float @llvm.maximum.f32(float %a, float %b)
  ret float %0
 }

 ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare float @llvm.maximum.f32(float, float) #3
+declare float @llvm.maximum.f32(float, float) #2

 ; Function Attrs: nounwind
-define hidden %runtime._string @main.maxString(ptr readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.maxString(ptr readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #1 {
 entry:
  %0 = insertvalue %runtime._string zeroinitializer, ptr %a.data, 0
  %1 = insertvalue %runtime._string %0, i32 %a.len, 1
  %2 = insertvalue %runtime._string zeroinitializer, ptr %b.data, 0
  %3 = insertvalue %runtime._string %2, i32 %b.len, 1
  %stackalloc = alloca i8, align 1
-  %4 = call i1 @runtime.stringLess(ptr %b.data, i32 %b.len, ptr %a.data, i32 %a.len, ptr undef) #5
+  %4 = call i1 @runtime.stringLess(ptr %b.data, i32 %b.len, ptr %a.data, i32 %a.len, ptr undef) #4
  %5 = select i1 %4, %runtime._string %1, %runtime._string %3
  %6 = select i1 %4, ptr %a.data, ptr %b.data
-  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #5
+  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #4
  ret %runtime._string %5
 }

 ; Function Attrs: nounwind
-define hidden void @main.clearSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
+define hidden void @main.clearSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #1 {
 entry:
  %0 = shl i32 %s.len, 2
  call void @llvm.memset.p0.i32(ptr align 4 %s.data, i8 0, i32 %0, i1 false)
@@ -160,26 +157,25 @@ entry:
 }

 ; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
-declare void @llvm.memset.p0.i32(ptr nocapture writeonly, i8, i32, i1 immarg) #4
+declare void @llvm.memset.p0.i32(ptr nocapture writeonly, i8, i32, i1 immarg) #3

 ; Function Attrs: nounwind
-define hidden void @main.clearZeroSizedSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
+define hidden void @main.clearZeroSizedSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.clearMap(ptr dereferenceable_or_null(40) %m, ptr %context) unnamed_addr #2 {
+define hidden void @main.clearMap(ptr dereferenceable_or_null(48) %m, ptr %context) unnamed_addr #1 {
 entry:
-  call void @runtime.hashmapClear(ptr %m, ptr undef) #5
+  call void @runtime.hashmapClear(ptr %m, ptr undef) #4
  ret void
 }

-declare void @runtime.hashmapClear(ptr dereferenceable_or_null(40), ptr) #1
+declare void @runtime.hashmapClear(ptr dereferenceable_or_null(48), ptr) #0

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
-attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: write) }
-attributes #5 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #3 = { nocallback nofree nounwind willreturn memory(argmem: write) }
+attributes #4 = { nounwind }
@@ -0,0 +1,24 @@
+package main
+
+// genericMethod has both a regular method and a "generic method": a method
+// with its own type parameter, independent of any type parameter on the
+// receiver. This is a Go 1.27 feature (see math/rand/v2.Rand.N for a
+// real-world example). Boxing such a value into an interface must not try to
+// encode the generic method's type-parameterized signature into a type code.
+type genericMethod struct{}
+
+func (t genericMethod) Regular(n int) int { return n }
+
+func (t genericMethod) GenericParam[X int | int64](n X) X { return n }
+
+// onlyGenericMethod's only method is generic, so its runtime type must have
+// an empty method set (hasMethodSet must become false, not just numMethods).
+type onlyGenericMethod struct{}
+
+func (t onlyGenericMethod) GenericParam[X int | int64](n X) X { return n }
+
+func useGenericMethods() (any, any) {
+	var g genericMethod
+	var o onlyGenericMethod
+	return any(g), any(o)
+}
@@ -0,0 +1,67 @@
+; ModuleID = 'go1.27.go'
+source_filename = "go1.27.go"
+target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
+target triple = "wasm32-unknown-wasi"
+
+%runtime.structField = type { ptr, ptr }
+%runtime._interface = type { ptr, ptr }
+
+@"reflect/types.signature:Regular:func:{basic:int}{basic:int}" = linkonce_odr constant i8 0, align 1
+@"reflect/types.type:named:main.genericMethod" = linkonce_odr constant { ptr, i8, i16, ptr, ptr, ptr, { i32, [1 x ptr] }, [19 x i8] } { ptr @"named:main.genericMethod$methodset", i8 122, i16 -32767, ptr getelementptr ({ ptr, i8, i16, ptr, { i32, [1 x ptr] } }, ptr @"reflect/types.type:pointer:named:main.genericMethod", i32 0, i32 1), ptr @"reflect/types.type:struct:{}", ptr @"reflect/types.type.pkgpath:main", { i32, [1 x ptr] } { i32 1, [1 x ptr] [ptr @"reflect/types.signature:Regular:func:{basic:int}{basic:int}"] }, [19 x i8] c"main.genericMethod\00" }, align 4
+@"reflect/types.type.pkgpath:main" = linkonce_odr unnamed_addr constant [5 x i8] c"main\00", align 1
+@"reflect/types.type:pointer:named:main.genericMethod" = linkonce_odr constant { ptr, i8, i16, ptr, { i32, [1 x ptr] } } { ptr @"pointer:named:main.genericMethod$methodset", i8 -43, i16 -32767, ptr getelementptr ({ ptr, i8, i16, ptr, ptr, ptr, { i32, [1 x ptr] }, [19 x i8] }, ptr @"reflect/types.type:named:main.genericMethod", i32 0, i32 1), { i32, [1 x ptr] } { i32 1, [1 x ptr] [ptr @"reflect/types.signature:Regular:func:{basic:int}{basic:int}"] } }, align 4
+@"reflect/methods.Regular:func:{basic:int}{basic:int}" = linkonce_odr constant i8 0, align 1
+@"main$string" = internal unnamed_addr constant [18 x i8] c"main.genericMethod", align 1
+@"main$string.1" = internal unnamed_addr constant [7 x i8] c"Regular", align 1
+@"pointer:named:main.genericMethod$methodset" = linkonce_odr unnamed_addr constant { i32, [1 x ptr], { ptr } } { i32 1, [1 x ptr] [ptr @"reflect/methods.Regular:func:{basic:int}{basic:int}"], { ptr } { ptr @"(*main.genericMethod).Regular" } }
+@"reflect/types.type:struct:{}" = linkonce_odr constant { i8, i16, ptr, ptr, i32, i16, [0 x %runtime.structField] } { i8 90, i16 0, ptr @"reflect/types.type:pointer:struct:{}", ptr @"reflect/types.type.pkgpath.empty", i32 0, i16 0, [0 x %runtime.structField] zeroinitializer }, align 4
+@"reflect/types.type.pkgpath.empty" = linkonce_odr unnamed_addr constant [1 x i8] zeroinitializer, align 1
+@"reflect/types.type:pointer:struct:{}" = linkonce_odr constant { i8, i16, ptr } { i8 -43, i16 0, ptr @"reflect/types.type:struct:{}" }, align 4
+@"named:main.genericMethod$methodset" = linkonce_odr unnamed_addr constant { i32, [1 x ptr], { ptr } } { i32 1, [1 x ptr] [ptr @"reflect/methods.Regular:func:{basic:int}{basic:int}"], { ptr } { ptr @"(main.genericMethod).Regular$invoke" } }
+@"reflect/types.type:named:main.onlyGenericMethod" = linkonce_odr constant { i8, i16, ptr, ptr, ptr, [23 x i8] } { i8 122, i16 0, ptr @"reflect/types.type:pointer:named:main.onlyGenericMethod", ptr @"reflect/types.type:struct:{}", ptr @"reflect/types.type.pkgpath:main", [23 x i8] c"main.onlyGenericMethod\00" }, align 4
+@"reflect/types.type:pointer:named:main.onlyGenericMethod" = linkonce_odr constant { i8, i16, ptr } { i8 -43, i16 0, ptr @"reflect/types.type:named:main.onlyGenericMethod" }, align 4
+
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0
+
+; Function Attrs: nounwind
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden i32 @"(main.genericMethod).Regular"(i32 %n, ptr %context) unnamed_addr #1 {
+entry:
+  ret i32 %n
+}
+
+; Function Attrs: nounwind
+define hidden { %runtime._interface, %runtime._interface } @main.useGenericMethods(ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  call void @runtime.trackPointer(ptr nonnull getelementptr inbounds nuw (i8, ptr @"reflect/types.type:named:main.genericMethod", i32 4), ptr nonnull %stackalloc, ptr undef) #2
+  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #2
+  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:named:main.onlyGenericMethod", ptr nonnull %stackalloc, ptr undef) #2
+  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #2
+  ret { %runtime._interface, %runtime._interface } { %runtime._interface { ptr getelementptr ({ ptr, i8, i16, ptr, ptr, ptr, { i32, [1 x ptr] }, [19 x i8] }, ptr @"reflect/types.type:named:main.genericMethod", i32 0, i32 1), ptr null }, %runtime._interface { ptr @"reflect/types.type:named:main.onlyGenericMethod", ptr null } }
+}
+
+; Function Attrs: nounwind
+define linkonce_odr hidden i32 @"(*main.genericMethod).Regular"(ptr %t, i32 %n, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  call void @runtime.trackPointer(ptr %t, ptr nonnull %stackalloc, ptr undef) #2
+  %0 = call i32 @"(main.genericMethod).Regular"(i32 %n, ptr undef)
+  ret i32 %0
+}
+
+; Function Attrs: nounwind
+define linkonce_odr i32 @"(main.genericMethod).Regular$invoke"(ptr %0, i32 %1, ptr %2) unnamed_addr #1 {
+entry:
+  %ret = call i32 @"(main.genericMethod).Regular"(i32 %1, ptr %2)
+  ret i32 %ret
+}
+
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind }
@@ -5,39 +5,36 @@ target triple = "thumbv7m-unknown-unknown-eabi"

@"main$string" = internal unnamed_addr constant [4 x i8] c"test", align 1

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #1 {
+define hidden void @main.init(ptr %context) unnamed_addr #0 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #1 {
+define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #0 {
 entry:
  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr undef) #11
  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #11
  ret void
 }

-declare void @main.regularFunction(i32, ptr) #2
+declare void @main.regularFunction(i32, ptr) #1

 ; Function Attrs: nounwind
-define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #3 {
+define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #2 {
 entry:
  %unpack.int = ptrtoint ptr %0 to i32
  call void @main.regularFunction(i32 %unpack.int, ptr undef) #11
  ret void
 }

-declare i32 @"internal/task.getGoroutineStackSize"(i32, ptr) #2
+declare i32 @"internal/task.getGoroutineStackSize"(i32, ptr) #1

-declare void @"internal/task.start"(i32, ptr, i32, ptr) #2
+declare void @"internal/task.start"(i32, ptr, i32, ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #1 {
+define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #0 {
 entry:
  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr undef) #11
  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #11
@@ -45,13 +42,13 @@ entry:
 }

 ; Function Attrs: nounwind
-define internal void @"main.inlineFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #1 {
+define internal void @"main.inlineFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #0 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(ptr %0) unnamed_addr #4 {
+define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(ptr %0) unnamed_addr #3 {
 entry:
  %unpack.int = ptrtoint ptr %0 to i32
  call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, ptr undef)
@@ -59,11 +56,11 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #1 {
+define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #0 {
 entry:
  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #11
  store i32 3, ptr %n, align 4
-  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #11
+  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 133 to ptr), ptr undef) #11
  store i32 5, ptr %0, align 4
  %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
  store ptr %n, ptr %1, align 4
@@ -76,8 +73,11 @@ entry:
  ret void
 }

+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #4
+
 ; Function Attrs: nounwind
-define internal void @"main.closureFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #1 {
+define internal void @"main.closureFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #0 {
 entry:
  store i32 7, ptr %context, align 4
  ret void
@@ -93,16 +93,16 @@ entry:
  ret void
 }

-declare void @runtime.printlock(ptr) #2
+declare void @runtime.printlock(ptr) #1

-declare void @runtime.printint32(i32, ptr) #2
+declare void @runtime.printint32(i32, ptr) #1

-declare void @runtime.printunlock(ptr) #2
+declare void @runtime.printunlock(ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
+define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #0 {
 entry:
-  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #11
+  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 391 to ptr), ptr undef) #11
  store i32 5, ptr %0, align 4
  %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
  store ptr %fn.context, ptr %1, align 4
@@ -126,13 +126,13 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.recoverBuiltinGoroutine(ptr %context) unnamed_addr #1 {
+define hidden void @main.recoverBuiltinGoroutine(ptr %context) unnamed_addr #0 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.copyBuiltinGoroutine(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #1 {
+define hidden void @main.copyBuiltinGoroutine(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #0 {
 entry:
  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
  call void @llvm.memmove.p0.p0.i32(ptr align 1 %dst.data, ptr align 1 %src.data, i32 %copy.n, i1 false)
@@ -146,18 +146,18 @@ declare i32 @llvm.umin.i32(i32, i32) #7
 declare void @llvm.memmove.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #8

 ; Function Attrs: nounwind
-define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
+define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #0 {
 entry:
  call void @runtime.chanClose(ptr %ch, ptr undef) #11
  ret void
 }

-declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #2
+declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
+define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #0 {
 entry:
-  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #11
+  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr nonnull inttoptr (i32 713 to ptr), ptr undef) #11
  store ptr %itf.value, ptr %0, align 4
  %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
  store ptr @"main$string", ptr %1, align 4
@@ -186,15 +186,15 @@ entry:
  ret void
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #1 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #2 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #3 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.regularFunction" }
-attributes #4 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #0 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #1 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #2 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.regularFunction" }
+attributes #3 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #4 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
 attributes #5 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
 attributes #6 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper" }
 attributes #7 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
 attributes #8 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
-attributes #9 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #9 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-invoke"="reflect/methods.Print:func:{basic:string}{}" "tinygo-methods"="reflect/methods.Print:func:{basic:string}{}" }
 attributes #10 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
 attributes #11 = { nounwind }
@@ -5,30 +5,27 @@ target triple = "wasm32-unknown-wasi"

@"main$string" = internal unnamed_addr constant [4 x i8] c"test", align 1

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #2 {
+define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #11
  ret void
 }

-declare void @main.regularFunction(i32, ptr) #1
+declare void @main.regularFunction(i32, ptr) #0

-declare void @runtime.deadlock(ptr) #1
+declare void @runtime.deadlock(ptr) #0

 ; Function Attrs: nounwind
-define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #3 {
+define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #2 {
 entry:
  %unpack.int = ptrtoint ptr %0 to i32
  call void @main.regularFunction(i32 %unpack.int, ptr undef) #11
@@ -36,23 +33,23 @@ entry:
  unreachable
 }

-declare void @"internal/task.start"(i32, ptr, i32, ptr) #1
+declare void @"internal/task.start"(i32, ptr, i32, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #2 {
+define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #11
  ret void
 }

 ; Function Attrs: nounwind
-define internal void @"main.inlineFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #2 {
+define internal void @"main.inlineFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(ptr %0) unnamed_addr #4 {
+define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(ptr %0) unnamed_addr #3 {
 entry:
  %unpack.int = ptrtoint ptr %0 to i32
  call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, ptr undef)
@@ -61,7 +58,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #2 {
+define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #11
@@ -69,7 +66,7 @@ entry:
  store i32 3, ptr %n, align 4
  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #11
  call void @runtime.trackPointer(ptr nonnull @"main.closureFunctionGoroutine$1", ptr nonnull %stackalloc, ptr undef) #11
-  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #11
+  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 133 to ptr), ptr undef) #11
  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #11
  store i32 5, ptr %0, align 4
  %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -82,8 +79,11 @@ entry:
  ret void
 }

+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #4
+
 ; Function Attrs: nounwind
-define internal void @"main.closureFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #2 {
+define internal void @"main.closureFunctionGoroutine$1"(i32 %x, ptr %context) unnamed_addr #1 {
 entry:
  store i32 7, ptr %context, align 4
  ret void
@@ -100,17 +100,17 @@ entry:
  unreachable
 }

-declare void @runtime.printlock(ptr) #1
+declare void @runtime.printlock(ptr) #0

-declare void @runtime.printint32(i32, ptr) #1
+declare void @runtime.printint32(i32, ptr) #0

-declare void @runtime.printunlock(ptr) #1
+declare void @runtime.printunlock(ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #2 {
+define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #11
+  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 391 to ptr), ptr undef) #11
  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #11
  store i32 5, ptr %0, align 4
  %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -135,13 +135,13 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.recoverBuiltinGoroutine(ptr %context) unnamed_addr #2 {
+define hidden void @main.recoverBuiltinGoroutine(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.copyBuiltinGoroutine(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #2 {
+define hidden void @main.copyBuiltinGoroutine(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #1 {
 entry:
  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
  call void @llvm.memmove.p0.p0.i32(ptr align 1 %dst.data, ptr align 1 %src.data, i32 %copy.n, i1 false)
@@ -155,19 +155,19 @@ declare i32 @llvm.umin.i32(i32, i32) #7
 declare void @llvm.memmove.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #8

 ; Function Attrs: nounwind
-define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
  call void @runtime.chanClose(ptr %ch, ptr undef) #11
  ret void
 }

-declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #1
+declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
+define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #11
+  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr nonnull inttoptr (i32 713 to ptr), ptr undef) #11
  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #11
  store ptr %itf.value, ptr %0, align 4
  %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -197,15 +197,15 @@ entry:
  unreachable
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.regularFunction" }
-attributes #4 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.regularFunction" }
+attributes #3 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #4 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #5 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
 attributes #6 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper" }
 attributes #7 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
 attributes #8 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
-attributes #9 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #9 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Print:func:{basic:string}{}" "tinygo-methods"="reflect/methods.Print:func:{basic:string}{}" }
 attributes #10 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
 attributes #11 = { nounwind }
@@ -19,57 +19,54 @@ target triple = "wasm32-unknown-wasi"
@"reflect/types.type:interface:{String:func:{}{basic:string}}" = linkonce_odr constant { i8, ptr, { i32, [1 x ptr] } } { i8 84, ptr @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}", { i32, [1 x ptr] } { i32 1, [1 x ptr] [ptr @"reflect/types.signature:String:func:{}{basic:string}"] } }, align 4
@"reflect/types.typeid:basic:int" = external constant i8

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden %runtime._interface @main.simpleType(ptr %context) unnamed_addr #2 {
+define hidden %runtime._interface @main.simpleType(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:basic:int", ptr nonnull %stackalloc, ptr undef) #7
-  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #7
+  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:basic:int", ptr nonnull %stackalloc, ptr undef) #6
+  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #6
  ret %runtime._interface { ptr @"reflect/types.type:basic:int", ptr null }
 }

 ; Function Attrs: nounwind
-define hidden %runtime._interface @main.pointerType(ptr %context) unnamed_addr #2 {
+define hidden %runtime._interface @main.pointerType(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:pointer:basic:int", ptr nonnull %stackalloc, ptr undef) #7
-  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #7
+  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:pointer:basic:int", ptr nonnull %stackalloc, ptr undef) #6
+  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #6
  ret %runtime._interface { ptr @"reflect/types.type:pointer:basic:int", ptr null }
 }

 ; Function Attrs: nounwind
-define hidden %runtime._interface @main.interfaceType(ptr %context) unnamed_addr #2 {
+define hidden %runtime._interface @main.interfaceType(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:pointer:named:error", ptr nonnull %stackalloc, ptr undef) #7
-  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #7
+  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:pointer:named:error", ptr nonnull %stackalloc, ptr undef) #6
+  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #6
  ret %runtime._interface { ptr @"reflect/types.type:pointer:named:error", ptr null }
 }

 ; Function Attrs: nounwind
-define hidden %runtime._interface @main.anonymousInterfaceType(ptr %context) unnamed_addr #2 {
+define hidden %runtime._interface @main.anonymousInterfaceType(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}", ptr nonnull %stackalloc, ptr undef) #7
-  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #7
+  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}", ptr nonnull %stackalloc, ptr undef) #6
+  call void @runtime.trackPointer(ptr null, ptr nonnull %stackalloc, ptr undef) #6
  ret %runtime._interface { ptr @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}", ptr null }
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.isInt(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.isInt(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
-  %typecode = call i1 @runtime.typeAssert(ptr %itf.typecode, ptr nonnull @"reflect/types.typeid:basic:int", ptr undef) #7
+  %typecode = call i1 @runtime.typeAssert(ptr %itf.typecode, ptr nonnull @"reflect/types.typeid:basic:int", ptr undef) #6
  br i1 %typecode, label %typeassert.ok, label %typeassert.next

 typeassert.next:                                  ; preds = %typeassert.ok, %entry
@@ -79,12 +76,12 @@ typeassert.ok:                                    ; preds = %entry
  br label %typeassert.next
 }

-declare i1 @runtime.typeAssert(ptr, ptr dereferenceable_or_null(1), ptr) #1
+declare i1 @runtime.typeAssert(ptr, ptr dereferenceable_or_null(1), ptr) #0

 ; Function Attrs: nounwind
-define hidden i1 @main.isError(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.isError(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(ptr %itf.typecode) #7
+  %0 = call i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(ptr %itf.typecode) #6
  br i1 %0, label %typeassert.ok, label %typeassert.next

 typeassert.next:                                  ; preds = %typeassert.ok, %entry
@@ -94,12 +91,12 @@ typeassert.ok:                                    ; preds = %entry
  br label %typeassert.next
 }

-declare i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(ptr) #3
+declare i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(ptr) #2

 ; Function Attrs: nounwind
-define hidden i1 @main.isStringer(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.isStringer(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(ptr %itf.typecode) #7
+  %0 = call i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(ptr %itf.typecode) #6
  br i1 %0, label %typeassert.ok, label %typeassert.next

 typeassert.next:                                  ; preds = %typeassert.ok, %entry
@@ -109,34 +106,33 @@ typeassert.ok:                                    ; preds = %entry
  br label %typeassert.next
 }

-declare i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(ptr) #4
+declare i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(ptr) #3

 ; Function Attrs: nounwind
-define hidden i8 @main.callFooMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.callFooMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call i8 @"interface:{String:func:{}{basic:string},main.foo:func:{basic:int}{basic:uint8}}.foo$invoke"(ptr %itf.value, i32 3, ptr %itf.typecode, ptr undef) #7
+  %0 = call i8 @"interface:{String:func:{}{basic:string},main.foo:func:{basic:int}{basic:uint8}}.foo$invoke"(ptr %itf.value, i32 3, ptr %itf.typecode, ptr undef) #6
  ret i8 %0
 }

-declare i8 @"interface:{String:func:{}{basic:string},main.foo:func:{basic:int}{basic:uint8}}.foo$invoke"(ptr, i32, ptr, ptr) #5
+declare i8 @"interface:{String:func:{}{basic:string},main.foo:func:{basic:int}{basic:uint8}}.foo$invoke"(ptr, i32, ptr, ptr) #4

 ; Function Attrs: nounwind
-define hidden %runtime._string @main.callErrorMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.callErrorMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  %0 = call %runtime._string @"interface:{Error:func:{}{basic:string}}.Error$invoke"(ptr %itf.value, ptr %itf.typecode, ptr undef) #7
+  %0 = call %runtime._string @"interface:{Error:func:{}{basic:string}}.Error$invoke"(ptr %itf.value, ptr %itf.typecode, ptr undef) #6
  %1 = extractvalue %runtime._string %0, 0
-  call void @runtime.trackPointer(ptr %1, ptr nonnull %stackalloc, ptr undef) #7
+  call void @runtime.trackPointer(ptr %1, ptr nonnull %stackalloc, ptr undef) #6
  ret %runtime._string %0
 }

-declare %runtime._string @"interface:{Error:func:{}{basic:string}}.Error$invoke"(ptr, ptr, ptr) #6
+declare %runtime._string @"interface:{Error:func:{}{basic:string}}.Error$invoke"(ptr, ptr, ptr) #5

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-methods"="reflect/methods.Error() string" }
-attributes #4 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-methods"="reflect/methods.String() string" }
-attributes #5 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="main.$methods.foo(int) uint8" "tinygo-methods"="reflect/methods.String() string; main.$methods.foo(int) uint8" }
-attributes #6 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Error() string" "tinygo-methods"="reflect/methods.Error() string" }
-attributes #7 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-methods"="reflect/methods.Error:func:{}{basic:string}" }
+attributes #3 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-methods"="reflect/methods.String:func:{}{basic:string}" }
+attributes #4 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="main.$methods.foo:func:{basic:int}{basic:uint8}" "tinygo-methods"="reflect/methods.String:func:{}{basic:string}; main.$methods.foo:func:{basic:int}{basic:uint8}" }
+attributes #5 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Error:func:{}{basic:string}" "tinygo-methods"="reflect/methods.Error:func:{}{basic:string}" }
+attributes #6 = { nounwind }
@@ -0,0 +1,20 @@
+package main
+
+type largeOptimizedValue [1025]byte
+
+type mixedLargeOptimizedValue struct {
+	value [1025]byte
+	any   any
+}
+
+func makeLargeOptimizedValue() largeOptimizedValue {
+	return largeOptimizedValue{}
+}
+
+func readLargeOptimizedValue(value largeOptimizedValue) byte {
+	return value[len(value)-1]
+}
+
+func readMixedLargeOptimizedValue(value mixedLargeOptimizedValue) byte {
+	return value.value[len(value.value)-1]
+}
@@ -0,0 +1,120 @@
+package main
+
+type largeValue [1025]byte
+
+type largeStruct struct {
+	value [1025]byte
+	ptr   *byte
+}
+
+type largeInterface interface {
+	makeLargeValue() largeValue
+	readLargeValue(largeValue) byte
+}
+
+type largeReceiver largeValue
+
+func makeLargeValue(value byte) largeValue {
+	var result largeValue
+	result[len(result)-1] = value
+	return result
+}
+
+func makeZeroLargeValue() largeValue {
+	return largeValue{}
+}
+
+func passZeroLargeValue() byte {
+	return readLargeValue(largeValue{})
+}
+
+func readLargeValue(value largeValue) byte {
+	return value[len(value)-1]
+}
+
+func useLargeValue() byte {
+	return readLargeValue(makeLargeValue(42))
+}
+
+func useLargeFunctionValue(fn func(largeValue) byte) byte {
+	return fn(makeLargeValue(42))
+}
+
+func (receiver largeReceiver) makeLargeValue() largeValue {
+	return largeValue(receiver)
+}
+
+func (receiver largeReceiver) readLargeValue(value largeValue) byte {
+	return value[len(value)-1]
+}
+
+func useLargeInterface(value largeInterface) byte {
+	return value.readLargeValue(value.makeLargeValue())
+}
+
+func deferLargeValue(value largeValue) {
+	defer readLargeValue(value)
+}
+
+func goLargeValue(value largeValue) {
+	go readLargeValue(value)
+}
+
+func makeLargeResults(value byte) (largeValue, byte) {
+	return makeLargeValue(value), value
+}
+
+func makeTwoLargeResults(value byte) (largeValue, largeValue) {
+	return makeLargeValue(value), makeLargeValue(value + 1)
+}
+
+func makeMixedLargeResults(value byte) (largeValue, byte, largeValue) {
+	return makeLargeValue(value), value + 1, makeLargeValue(value + 2)
+}
+
+func chooseLargeValue(flag bool) largeValue {
+	value := makeLargeValue(1)
+	if flag {
+		value = makeLargeValue(42)
+	}
+	return value
+}
+
+func makePointerLargeValue(value *byte) largeStruct {
+	return largeStruct{ptr: value}
+}
+
+func assertLargeValue(value any) byte {
+	large, ok := value.(largeValue)
+	if !ok {
+		return 0
+	}
+	return large[len(large)-1]
+}
+
+func useLargeMap(key, value largeValue) byte {
+	values := map[largeValue]largeValue{key: value}
+	result, ok := values[key]
+	if !ok {
+		return 0
+	}
+	return result[len(result)-1]
+}
+
+func useLargeChannel(ch chan largeValue, value largeValue) byte {
+	ch <- value
+	result, ok := <-ch
+	if !ok {
+		return 0
+	}
+	return result[len(result)-1]
+}
+
+func selectLargeChannel(ch chan largeValue, value largeValue) byte {
+	select {
+	case ch <- value:
+		return 0
+	case result := <-ch:
+		return result[len(result)-1]
+	}
+}
@@ -0,0 +1,498 @@
+; ModuleID = 'large.go'
+source_filename = "large.go"
+target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
+target triple = "wasm32-unknown-wasi"
+
+%runtime._string = type { ptr, i32 }
+%runtime.channelOp = type { ptr, ptr, i32, ptr }
+%runtime.chanSelectState = type { ptr, ptr }
+
+@"runtime/gc.layout:258-000000000000000000000000000000000000000000000000000000000000000002" = linkonce_odr unnamed_addr constant { i32, [33 x i8] } { i32 258, [33 x i8] c"\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\02" }
+@"reflect/types.typeid:named:main.largeValue" = external constant i8
+@llvm.used = appending global [15 x ptr] [ptr @"(main.largeReceiver).makeLargeValue", ptr @"(main.largeReceiver).readLargeValue", ptr @main.makeLargeValue, ptr @main.makeZeroLargeValue, ptr @main.readLargeValue, ptr @main.deferLargeValue, ptr @main.goLargeValue, ptr @main.makeLargeResults, ptr @main.makeTwoLargeResults, ptr @main.makeMixedLargeResults, ptr @main.chooseLargeValue, ptr @main.makePointerLargeValue, ptr @main.useLargeMap, ptr @main.useLargeChannel, ptr @main.selectLargeChannel]
+@"main$string" = internal unnamed_addr constant [31 x i8] c"blocking select matched no case", align 1
+@"main$pack" = internal unnamed_addr constant { %runtime._string } { %runtime._string { ptr @"main$string", i32 31 } }
+@"reflect/types.type:basic:string" = linkonce_odr constant { i8, ptr } { i8 81, ptr @"reflect/types.type:pointer:basic:string" }, align 4
+@"reflect/types.type:pointer:basic:string" = linkonce_odr constant { i8, i16, ptr } { i8 -43, i16 0, ptr @"reflect/types.type:basic:string" }, align 4
+
+declare void @runtime.trackPointer(ptr readonly captures(none), ptr, ptr) #0
+
+; Function Attrs: nounwind
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"(main.largeReceiver).makeLargeValue"(ptr dereferenceable_or_null(1025) %return, ptr readonly dereferenceable_or_null(1025) %receiver, ptr %context) unnamed_addr #1 {
+entry:
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %return, ptr noundef nonnull align 1 dereferenceable(1025) %receiver, i32 1025, i1 false)
+  ret void
+}
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.memcpy.p0.p0.i32(ptr noalias writeonly captures(none), ptr noalias readonly captures(none), i32, i1 immarg) #2
+
+; Function Attrs: nounwind
+define hidden i8 @"(main.largeReceiver).readLargeValue"(ptr readonly dereferenceable_or_null(1025) %receiver, ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %value1 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %value1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %value1, ptr noundef nonnull align 1 dereferenceable(1025) %value, i32 1025, i1 false)
+  %0 = getelementptr inbounds nuw i8, ptr %value1, i32 1024
+  %1 = load i8, ptr %0, align 1
+  ret i8 %1
+}
+
+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #3
+
+; Function Attrs: nounwind
+define hidden void @main.makeLargeValue(ptr dereferenceable_or_null(1025) %return, i8 %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %result, ptr nonnull %stackalloc, ptr undef) #9
+  %0 = getelementptr inbounds nuw i8, ptr %result, i32 1024
+  store i8 %value, ptr %0, align 1
+  %1 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %1, ptr noundef nonnull align 1 dereferenceable(1025) %result, i32 1025, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %return, ptr noundef nonnull align 1 dereferenceable(1025) %1, i32 1025, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.makeZeroLargeValue(ptr dereferenceable_or_null(1025) %return, ptr %context) unnamed_addr #1 {
+entry:
+  store [1025 x i8] zeroinitializer, ptr %return, align 1
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden i8 @main.passZeroLargeValue(ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %"main.largeValue{}:main.largeValue" = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %"main.largeValue{}:main.largeValue", ptr nonnull %stackalloc, ptr undef) #9
+  store [1025 x i8] zeroinitializer, ptr %"main.largeValue{}:main.largeValue", align 1
+  %0 = call i8 @main.readLargeValue(ptr nonnull %"main.largeValue{}:main.largeValue", ptr undef)
+  ret i8 %0
+}
+
+; Function Attrs: nounwind
+define hidden i8 @main.readLargeValue(ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %value1 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %value1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %value1, ptr noundef nonnull align 1 dereferenceable(1025) %value, i32 1025, i1 false)
+  %0 = getelementptr inbounds nuw i8, ptr %value1, i32 1024
+  %1 = load i8, ptr %0, align 1
+  ret i8 %1
+}
+
+; Function Attrs: nounwind
+define hidden i8 @main.useLargeValue(ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result, i8 42, ptr undef)
+  %0 = call i8 @main.readLargeValue(ptr nonnull %call.result, ptr undef)
+  ret i8 %0
+}
+
+; Function Attrs: nounwind
+define hidden i8 @main.useLargeFunctionValue(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result, i8 42, ptr undef)
+  %0 = icmp eq ptr %fn.funcptr, null
+  br i1 %0, label %fpcall.throw, label %fpcall.next
+
+fpcall.next:                                      ; preds = %entry
+  %1 = call i8 %fn.funcptr(ptr nonnull %call.result, ptr %fn.context) #9
+  ret i8 %1
+
+fpcall.throw:                                     ; preds = %entry
+  call void @runtime.nilPanic(ptr undef) #9
+  unreachable
+}
+
+declare void @runtime.nilPanic(ptr) #0
+
+; Function Attrs: nounwind
+define hidden i8 @main.useLargeInterface(ptr %value.typecode, ptr %value.value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @"interface:{main.makeLargeValue:func:{}{named:main.largeValue},main.readLargeValue:func:{named:main.largeValue}{basic:uint8}}.makeLargeValue$invoke"(ptr nonnull %call.result, ptr %value.value, ptr %value.typecode, ptr undef) #9
+  %0 = call i8 @"interface:{main.makeLargeValue:func:{}{named:main.largeValue},main.readLargeValue:func:{named:main.largeValue}{basic:uint8}}.readLargeValue$invoke"(ptr %value.value, ptr nonnull %call.result, ptr %value.typecode, ptr undef) #9
+  ret i8 %0
+}
+
+declare void @"interface:{main.makeLargeValue:func:{}{named:main.largeValue},main.readLargeValue:func:{named:main.largeValue}{basic:uint8}}.makeLargeValue$invoke"(ptr, ptr, ptr, ptr) #4
+
+declare i8 @"interface:{main.makeLargeValue:func:{}{named:main.largeValue},main.readLargeValue:func:{named:main.largeValue}{basic:uint8}}.readLargeValue$invoke"(ptr, ptr, ptr, ptr) #5
+
+; Function Attrs: nounwind
+define hidden void @main.deferLargeValue(ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %defer.alloca = alloca { i32, ptr, ptr }, align 8
+  %deferPtr = alloca ptr, align 4
+  store ptr null, ptr %deferPtr, align 4
+  %stackalloc = alloca i8, align 1
+  call void @runtime.trackPointer(ptr nonnull %defer.alloca, ptr nonnull %stackalloc, ptr undef) #9
+  store i32 0, ptr %defer.alloca, align 4
+  %defer.alloca.repack1 = getelementptr inbounds nuw i8, ptr %defer.alloca, i32 4
+  store ptr null, ptr %defer.alloca.repack1, align 4
+  %defer.alloca.repack3 = getelementptr inbounds nuw i8, ptr %defer.alloca, i32 8
+  store ptr %value, ptr %defer.alloca.repack3, align 4
+  store ptr %defer.alloca, ptr %deferPtr, align 4
+  br label %rundefers.block
+
+rundefers.after:                                  ; preds = %rundefers.end
+  ret void
+
+rundefers.block:                                  ; preds = %entry
+  br label %rundefers.loophead
+
+rundefers.loophead:                               ; preds = %rundefers.callback0, %rundefers.block
+  %0 = load ptr, ptr %deferPtr, align 4
+  %stackIsNil = icmp eq ptr %0, null
+  br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
+
+rundefers.loop:                                   ; preds = %rundefers.loophead
+  %stack.next.gep = getelementptr inbounds nuw i8, ptr %0, i32 4
+  %stack.next = load ptr, ptr %stack.next.gep, align 4
+  store ptr %stack.next, ptr %deferPtr, align 4
+  %callback = load i32, ptr %0, align 4
+  switch i32 %callback, label %rundefers.default [
+    i32 0, label %rundefers.callback0
+  ]
+
+rundefers.callback0:                              ; preds = %rundefers.loop
+  %gep = getelementptr inbounds nuw i8, ptr %0, i32 8
+  %param = load ptr, ptr %gep, align 4
+  %1 = call i8 @main.readLargeValue(ptr %param, ptr undef)
+  br label %rundefers.loophead
+
+rundefers.default:                                ; preds = %rundefers.loop
+  unreachable
+
+rundefers.end:                                    ; preds = %rundefers.loophead
+  br label %rundefers.after
+
+recover:                                          ; No predecessors!
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.goLargeValue(ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %go.param = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %go.param, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %go.param, ptr noundef nonnull align 1 dereferenceable(1025) %value, i32 1025, i1 false)
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.readLargeValue$gowrapper" to i32), ptr nonnull %go.param, i32 65536, ptr undef) #9
+  ret void
+}
+
+declare void @runtime.deadlock(ptr) #0
+
+; Function Attrs: nounwind
+define linkonce_odr void @"main.readLargeValue$gowrapper"(ptr %0) unnamed_addr #6 {
+entry:
+  %1 = call i8 @main.readLargeValue(ptr %0, ptr undef)
+  call void @runtime.deadlock(ptr undef) #9
+  unreachable
+}
+
+declare void @"internal/task.start"(i32, ptr, i32, ptr) #0
+
+; Function Attrs: nounwind
+define hidden void @main.makeLargeResults(ptr dereferenceable_or_null(1026) %return, i8 %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result, i8 %value, ptr undef)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %return, ptr noundef nonnull align 1 dereferenceable(1025) %call.result, i32 1025, i1 false)
+  %0 = getelementptr inbounds nuw i8, ptr %return, i32 1025
+  store i8 %value, ptr %0, align 1
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.makeTwoLargeResults(ptr dereferenceable_or_null(2050) %return, i8 %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result, i8 %value, ptr undef)
+  %0 = add i8 %value, 1
+  %call.result1 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result1, i8 %0, ptr undef)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %return, ptr noundef nonnull align 1 dereferenceable(1025) %call.result, i32 1025, i1 false)
+  %1 = getelementptr inbounds nuw i8, ptr %return, i32 1025
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %1, ptr noundef nonnull align 1 dereferenceable(1025) %call.result1, i32 1025, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.makeMixedLargeResults(ptr dereferenceable_or_null(2051) %return, i8 %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result, i8 %value, ptr undef)
+  %0 = add i8 %value, 1
+  %1 = add i8 %value, 2
+  %call.result1 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result1, i8 %1, ptr undef)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %return, ptr noundef nonnull align 1 dereferenceable(1025) %call.result, i32 1025, i1 false)
+  %2 = getelementptr inbounds nuw i8, ptr %return, i32 1025
+  store i8 %0, ptr %2, align 1
+  %3 = getelementptr inbounds nuw i8, ptr %return, i32 1026
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %3, ptr noundef nonnull align 1 dereferenceable(1025) %call.result1, i32 1025, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.chooseLargeValue(ptr dereferenceable_or_null(1025) %return, i1 %flag, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %call.result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result, i8 1, ptr undef)
+  br i1 %flag, label %if.then, label %if.done
+
+if.then:                                          ; preds = %entry
+  %call.result1 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %call.result1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @main.makeLargeValue(ptr nonnull %call.result1, i8 42, ptr undef)
+  br label %if.done
+
+if.done:                                          ; preds = %if.then, %entry
+  %0 = phi ptr [ %call.result, %entry ], [ %call.result1, %if.then ]
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %return, ptr noundef nonnull align 1 dereferenceable(1025) %0, i32 1025, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.makePointerLargeValue(ptr dereferenceable_or_null(1032) %return, ptr dereferenceable_or_null(1) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %complit = call align 4 dereferenceable(1032) ptr @runtime.alloc(i32 1032, ptr nonnull @"runtime/gc.layout:258-000000000000000000000000000000000000000000000000000000000000000002", ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %complit, ptr nonnull %stackalloc, ptr undef) #9
+  br i1 false, label %store.throw, label %store.next
+
+store.next:                                       ; preds = %entry
+  %0 = getelementptr inbounds nuw i8, ptr %complit, i32 1028
+  store ptr %value, ptr %0, align 4
+  %1 = call align 4 dereferenceable(1032) ptr @runtime.alloc(i32 1032, ptr nonnull @"runtime/gc.layout:258-000000000000000000000000000000000000000000000000000000000000000002", ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %1, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 4 dereferenceable(1032) %1, ptr noundef nonnull align 4 dereferenceable(1032) %complit, i32 1032, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1032) %return, ptr noundef nonnull align 4 dereferenceable(1032) %1, i32 1032, i1 false)
+  ret void
+
+store.throw:                                      ; preds = %entry
+  unreachable
+}
+
+; Function Attrs: nounwind
+define hidden i8 @main.assertLargeValue(ptr %value.typecode, ptr %value.value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %large = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %large, ptr nonnull %stackalloc, ptr undef) #9
+  %typecode = call i1 @runtime.typeAssert(ptr %value.typecode, ptr nonnull @"reflect/types.typeid:named:main.largeValue", ptr undef) #9
+  %typeassert.result = call align 1 dereferenceable(1026) ptr @runtime.alloc(i32 1026, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %typeassert.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memset.p0.i32(ptr noundef nonnull align 1 dereferenceable(1026) %typeassert.result, i8 0, i32 1026, i1 false)
+  br i1 %typecode, label %typeassert.ok, label %typeassert.next
+
+typeassert.next:                                  ; preds = %typeassert.ok, %entry
+  %0 = getelementptr inbounds nuw i8, ptr %typeassert.result, i32 1025
+  store i1 %typecode, ptr %0, align 1
+  %t2 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %t2, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %t2, ptr noundef nonnull align 1 dereferenceable(1025) %typeassert.result, i32 1025, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %large, ptr noundef nonnull align 1 dereferenceable(1025) %t2, i32 1025, i1 false)
+  br i1 %typecode, label %if.done, label %if.then
+
+typeassert.ok:                                    ; preds = %entry
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %typeassert.result, ptr noundef nonnull align 1 dereferenceable(1025) %value.value, i32 1025, i1 false)
+  br label %typeassert.next
+
+if.done:                                          ; preds = %typeassert.next
+  %1 = getelementptr inbounds nuw i8, ptr %large, i32 1024
+  %2 = load i8, ptr %1, align 1
+  ret i8 %2
+
+if.then:                                          ; preds = %typeassert.next
+  ret i8 0
+}
+
+declare i1 @runtime.typeAssert(ptr, ptr dereferenceable_or_null(1), ptr) #0
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
+declare void @llvm.memset.p0.i32(ptr writeonly captures(none), i8, i32, i1 immarg) #7
+
+; Function Attrs: nounwind
+define hidden i8 @main.useLargeMap(ptr readonly dereferenceable_or_null(1025) %key, ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %0 = call ptr @runtime.hashmapMakeGeneric(i32 1025, i32 1025, i32 1, ptr null, ptr nonnull @runtime.hash32, ptr null, ptr nonnull @runtime.memequal, ptr undef) #9
+  call void @runtime.trackPointer(ptr %0, ptr nonnull %stackalloc, ptr undef) #9
+  call void @runtime.hashmapBinarySet(ptr %0, ptr %key, ptr %value, ptr undef) #9
+  %result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %result, ptr nonnull %stackalloc, ptr undef) #9
+  %hashmap.result = call align 1 dereferenceable(1026) ptr @runtime.alloc(i32 1026, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %hashmap.result, ptr nonnull %stackalloc, ptr undef) #9
+  %1 = call i1 @runtime.hashmapBinaryGet(ptr %0, ptr %key, ptr nonnull %hashmap.result, i32 1025, ptr undef) #9
+  %2 = getelementptr inbounds nuw i8, ptr %hashmap.result, i32 1025
+  store i1 %1, ptr %2, align 1
+  %t3 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %t3, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %t3, ptr noundef nonnull align 1 dereferenceable(1025) %hashmap.result, i32 1025, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %result, ptr noundef nonnull align 1 dereferenceable(1025) %t3, i32 1025, i1 false)
+  %3 = getelementptr inbounds nuw i8, ptr %hashmap.result, i32 1025
+  %t4 = load i1, ptr %3, align 1
+  br i1 %t4, label %if.done, label %if.then
+
+if.done:                                          ; preds = %entry
+  %4 = getelementptr inbounds nuw i8, ptr %result, i32 1024
+  %5 = load i8, ptr %4, align 1
+  ret i8 %5
+
+if.then:                                          ; preds = %entry
+  ret i8 0
+}
+
+declare i32 @runtime.hash32(ptr, i32, i32, ptr) #0
+
+declare i1 @runtime.memequal(ptr, ptr, i32, ptr) #0
+
+declare ptr @runtime.hashmapMakeGeneric(i32, i32, i32, ptr, ptr, ptr, ptr, ptr) #0
+
+declare void @runtime.hashmapBinarySet(ptr dereferenceable_or_null(48), ptr, ptr, ptr) #0
+
+declare i1 @runtime.hashmapBinaryGet(ptr dereferenceable_or_null(48), ptr, ptr, i32, ptr) #0
+
+; Function Attrs: nounwind
+define hidden i8 @main.useLargeChannel(ptr dereferenceable_or_null(36) %ch, ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %chan.op1 = alloca %runtime.channelOp, align 8
+  %chan.op = alloca %runtime.channelOp, align 8
+  %stackalloc = alloca i8, align 1
+  call void @llvm.lifetime.start.p0(ptr nonnull %chan.op)
+  call void @runtime.chanSend(ptr %ch, ptr %value, ptr nonnull %chan.op, ptr undef) #9
+  call void @llvm.lifetime.end.p0(ptr nonnull %chan.op)
+  %result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %result, ptr nonnull %stackalloc, ptr undef) #9
+  %chan.result = call align 1 dereferenceable(1026) ptr @runtime.alloc(i32 1026, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %chan.result, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.lifetime.start.p0(ptr nonnull %chan.op1)
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.result, ptr nonnull %chan.op1, ptr undef) #9
+  %1 = getelementptr inbounds nuw i8, ptr %chan.result, i32 1025
+  store i1 %0, ptr %1, align 1
+  call void @llvm.lifetime.end.p0(ptr nonnull %chan.op1)
+  %t2 = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %t2, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %t2, ptr noundef nonnull align 1 dereferenceable(1025) %chan.result, i32 1025, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %result, ptr noundef nonnull align 1 dereferenceable(1025) %t2, i32 1025, i1 false)
+  %2 = getelementptr inbounds nuw i8, ptr %chan.result, i32 1025
+  %t3 = load i1, ptr %2, align 1
+  br i1 %t3, label %if.done, label %if.then
+
+if.done:                                          ; preds = %entry
+  %3 = getelementptr inbounds nuw i8, ptr %result, i32 1024
+  %4 = load i8, ptr %3, align 1
+  ret i8 %4
+
+if.then:                                          ; preds = %entry
+  ret i8 0
+}
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.start.p0(ptr captures(none)) #8
+
+declare void @runtime.chanSend(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #0
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.end.p0(ptr captures(none)) #8
+
+declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #0
+
+; Function Attrs: nounwind
+define hidden i8 @main.selectLargeChannel(ptr dereferenceable_or_null(36) %ch, ptr readonly dereferenceable_or_null(1025) %value, ptr %context) unnamed_addr #1 {
+entry:
+  %select.block.alloca = alloca [2 x %runtime.channelOp], align 8
+  %select.states.alloca = alloca [2 x %runtime.chanSelectState], align 8
+  %select.recvbuf.alloca = alloca [1025 x i8], align 1
+  %stackalloc = alloca i8, align 1
+  call void @llvm.lifetime.start.p0(ptr nonnull %select.recvbuf.alloca)
+  call void @llvm.lifetime.start.p0(ptr nonnull %select.states.alloca)
+  store ptr %ch, ptr %select.states.alloca, align 4
+  %select.states.alloca.repack3 = getelementptr inbounds nuw i8, ptr %select.states.alloca, i32 4
+  store ptr %value, ptr %select.states.alloca.repack3, align 4
+  %0 = getelementptr inbounds nuw i8, ptr %select.states.alloca, i32 8
+  store ptr %ch, ptr %0, align 4
+  %.repack5 = getelementptr inbounds nuw i8, ptr %select.states.alloca, i32 12
+  store ptr null, ptr %.repack5, align 4
+  call void @llvm.lifetime.start.p0(ptr nonnull %select.block.alloca)
+  %select.result = call { i32, i1 } @runtime.chanSelect(ptr nonnull %select.recvbuf.alloca, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr nonnull %select.block.alloca, i32 2, i32 2, ptr undef) #9
+  call void @llvm.lifetime.end.p0(ptr nonnull %select.block.alloca)
+  call void @llvm.lifetime.end.p0(ptr nonnull %select.states.alloca)
+  call void @runtime.trackPointer(ptr nonnull %select.recvbuf.alloca, ptr nonnull %stackalloc, ptr undef) #9
+  %1 = extractvalue { i32, i1 } %select.result, 0
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %select.body, label %select.next
+
+select.body:                                      ; preds = %entry
+  ret i8 0
+
+select.next:                                      ; preds = %entry
+  %3 = icmp eq i32 %1, 1
+  br i1 %3, label %select.body1, label %select.next2
+
+select.body1:                                     ; preds = %select.next
+  %result = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %result, ptr nonnull %stackalloc, ptr undef) #9
+  %select.received = call align 1 dereferenceable(1025) ptr @runtime.alloc(i32 1025, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %select.received, ptr nonnull %stackalloc, ptr undef) #9
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %select.received, ptr noundef nonnull align 1 dereferenceable(1025) %select.recvbuf.alloca, i32 1025, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(1025) %result, ptr noundef nonnull align 1 dereferenceable(1025) %select.received, i32 1025, i1 false)
+  %4 = getelementptr inbounds nuw i8, ptr %result, i32 1024
+  %5 = load i8, ptr %4, align 1
+  ret i8 %5
+
+select.next2:                                     ; preds = %select.next
+  call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:basic:string", ptr nonnull %stackalloc, ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull @"main$pack", ptr nonnull %stackalloc, ptr undef) #9
+  call void @runtime._panic(ptr nonnull @"reflect/types.type:basic:string", ptr nonnull @"main$pack", ptr undef) #9
+  unreachable
+}
+
+declare { i32, i1 } @runtime.chanSelect(ptr, ptr, i32, i32, ptr, i32, i32, ptr) #0
+
+declare void @runtime._panic(ptr, ptr, ptr) #0
+
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+attributes #3 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #4 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-indirect-result"="true" "tinygo-invoke"="main.$methods.makeLargeValue:func:{}{named:main.largeValue}" "tinygo-methods"="main.$methods.makeLargeValue:func:{}{named:main.largeValue}; main.$methods.readLargeValue:func:{named:main.largeValue}{basic:uint8}" }
+attributes #5 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="main.$methods.readLargeValue:func:{named:main.largeValue}{basic:uint8}" "tinygo-methods"="main.$methods.makeLargeValue:func:{}{named:main.largeValue}; main.$methods.readLargeValue:func:{named:main.largeValue}{basic:uint8}" }
+attributes #6 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.readLargeValue" }
+attributes #7 = { nocallback nofree nounwind willreturn memory(argmem: write) }
+attributes #8 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #9 = { nounwind }
@@ -3,48 +3,44 @@ source_filename = "pointer.go"
 target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden [0 x i32] @main.pointerDerefZero(ptr %x, ptr %context) unnamed_addr #2 {
+define hidden [0 x i32] @main.pointerDerefZero(ptr %x, ptr %context) unnamed_addr #1 {
 entry:
  ret [0 x i32] zeroinitializer
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.pointerCastFromUnsafe(ptr %x, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.pointerCastFromUnsafe(ptr %x, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr %x, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %x, ptr nonnull %stackalloc, ptr undef) #2
  ret ptr %x
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.pointerCastToUnsafe(ptr dereferenceable_or_null(4) %x, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.pointerCastToUnsafe(ptr dereferenceable_or_null(4) %x, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr %x, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %x, ptr nonnull %stackalloc, ptr undef) #2
  ret ptr %x
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.pointerCastToUnsafeNoop(ptr dereferenceable_or_null(1) %x, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.pointerCastToUnsafeNoop(ptr dereferenceable_or_null(1) %x, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr %x, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %x, ptr nonnull %stackalloc, ptr undef) #2
  ret ptr %x
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind }
@@ -35,6 +35,12 @@ func withLinkageName1() {
 //go:linkname withLinkageName2 somepkg.someFunction2
 func withLinkageName2()

+// Import a function from a different package using go:linknamestd (the standard
+// library variant of go:linkname introduced in Go 1.27).
+//
+//go:linknamestd withLinkageNameStd somepkg.someFunctionStd
+func withLinkageNameStd()
+
 // Function has an 'inline hint', similar to the inline keyword in C.
 //
 //go:inline
@@ -115,3 +121,36 @@ func doesNotEscapeParam(a *int, b []int, c chan int, d *[0]byte)
 //go:noescape
 func stillEscapes(a *int, b []int, c chan int, d *[0]byte) {
 }
+
+//go:noheap
+func doesHeapAlloc() *int {
+	return new(int)
+}
+
+// Define a function in a different package using a file-level go:linkname.
+// (Same as withLinkageName1, but with the //go:linkname directive detached
+// from the function declaration — see https://github.com/tinygo-org/tinygo/issues/4395)
+func withFileLevelLinkageName1() {
+}
+
+// Import a function from a different package using a file-level go:linkname.
+// (Same as withLinkageName2, but with the //go:linkname directive detached
+// from the function declaration.)
+func withFileLevelLinkageName2()
+
+//go:linkname withFileLevelLinkageName1 somepkg.someFileLevelFunction1
+//go:linkname withFileLevelLinkageName2 somepkg.someFileLevelFunction2
+
+// File-level linkname directives can also appear between two function
+// declarations, in which case Go's AST attaches them as the doc comment
+// of the following function — even when the directive's localname refers
+// to a different function. Exercise that case: the directive below names
+// withAdjacentLinkageName, but Go will attach it to
+// sentinelAfterAdjacentLinkname's Doc. The file-level scan must find it
+// by walking comment groups regardless of which decl they're attached to.
+func withAdjacentLinkageName() {
+}
+
+//go:linkname withAdjacentLinkageName somepkg.someAdjacentFunction
+func sentinelAfterAdjacentLinkname() {
+}
@@ -11,95 +11,127 @@ target triple = "wasm32-unknown-wasi"
@undefinedGlobalNotInSection = external global i32, align 4
@main.multipleGlobalPragmas = hidden global i32 0, section ".global_section", align 1024

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define void @extern_func() #3 {
+define void @extern_func() #2 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @somepkg.someFunction1(ptr %context) unnamed_addr #2 {
+define hidden void @somepkg.someFunction1(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

-declare void @somepkg.someFunction2(ptr) #1
+declare void @somepkg.someFunction2(ptr) #0
+
+declare void @somepkg.someFunctionStd(ptr) #0

 ; Function Attrs: inlinehint nounwind
-define hidden void @main.inlineFunc(ptr %context) unnamed_addr #4 {
+define hidden void @main.inlineFunc(ptr %context) unnamed_addr #3 {
 entry:
  ret void
 }

 ; Function Attrs: noinline nounwind
-define hidden void @main.noinlineFunc(ptr %context) unnamed_addr #5 {
+define hidden void @main.noinlineFunc(ptr %context) unnamed_addr #4 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.useGeneric(ptr %context) unnamed_addr #2 {
+define hidden void @main.useGeneric(ptr %context) unnamed_addr #1 {
 entry:
-  call void @"main.noinlineGenericFunc[int8]"(ptr undef)
+  call void @"main.noinlineGenericFunc[basic:int8]"(ptr undef)
  ret void
 }

 ; Function Attrs: noinline nounwind
-define linkonce_odr hidden void @"main.noinlineGenericFunc[int8]"(ptr %context) unnamed_addr #5 {
+define linkonce_odr hidden void @"main.noinlineGenericFunc[basic:int8]"(ptr %context) unnamed_addr #4 {
 entry:
  ret void
 }

 ; Function Attrs: noinline nounwind
-define hidden void @main.functionInSection(ptr %context) unnamed_addr #5 section ".special_function_section" {
+define hidden void @main.functionInSection(ptr %context) unnamed_addr #4 section ".special_function_section" {
 entry:
  ret void
 }

 ; Function Attrs: noinline nounwind
-define void @exportedFunctionInSection() #6 section ".special_function_section" {
+define void @exportedFunctionInSection() #5 section ".special_function_section" {
 entry:
  ret void
 }

-declare void @main.declaredImport() #7
+declare void @main.declaredImport() #6

-declare void @imported() #8
+declare void @imported() #7

 ; Function Attrs: nounwind
-define void @exported() #9 {
+define void @exported() #8 {
 entry:
  ret void
 }

-declare void @main.undefinedFunctionNotInSection(ptr) #1
+declare void @main.undefinedFunctionNotInSection(ptr) #0

-declare void @main.doesNotEscapeParam(ptr nocapture dereferenceable_or_null(4), ptr nocapture, i32, i32, ptr nocapture dereferenceable_or_null(36), ptr nocapture, ptr) #1
+declare void @main.doesNotEscapeParam(ptr nocapture dereferenceable_or_null(4), ptr nocapture, i32, i32, ptr nocapture dereferenceable_or_null(36), ptr nocapture, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.stillEscapes(ptr dereferenceable_or_null(4) %a, ptr %b.data, i32 %b.len, i32 %b.cap, ptr dereferenceable_or_null(36) %c, ptr %d, ptr %context) unnamed_addr #2 {
+define hidden void @main.stillEscapes(ptr dereferenceable_or_null(4) %a, ptr %b.data, i32 %b.len, i32 %b.cap, ptr dereferenceable_or_null(36) %c, ptr %d, ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-export-name"="extern_func" }
-attributes #4 = { inlinehint nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #5 = { noinline nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #6 = { noinline nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-export-name"="exportedFunctionInSection" }
-attributes #7 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-import-module"="modulename" "wasm-import-name"="import1" }
-attributes #8 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-import-module"="foobar" "wasm-import-name"="imported" }
-attributes #9 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-export-name"="exported" }
+; Function Attrs: nounwind
+define hidden ptr @main.doesHeapAlloc(ptr %context) unnamed_addr #1 {
+entry:
+  %stackalloc = alloca i8, align 1
+  %new = call align 4 dereferenceable(4) ptr @runtime.alloc_noheap(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #10
+  call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #10
+  ret ptr %new
+}
+
+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc_noheap(i32, ptr, ptr) #9
+
+; Function Attrs: nounwind
+define hidden void @somepkg.someFileLevelFunction1(ptr %context) unnamed_addr #1 {
+entry:
+  ret void
+}
+
+declare void @somepkg.someFileLevelFunction2(ptr) #0
+
+; Function Attrs: nounwind
+define hidden void @somepkg.someAdjacentFunction(ptr %context) unnamed_addr #1 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.sentinelAfterAdjacentLinkname(ptr %context) unnamed_addr #1 {
+entry:
+  ret void
+}
+
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-export-name"="extern_func" }
+attributes #3 = { inlinehint nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #4 = { noinline nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #5 = { noinline nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-export-name"="exportedFunctionInSection" }
+attributes #6 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-import-module"="modulename" "wasm-import-name"="import1" }
+attributes #7 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-import-module"="foobar" "wasm-import-name"="imported" }
+attributes #8 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "wasm-export-name"="exported" }
+attributes #9 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #10 = { nounwind }
@@ -3,31 +3,28 @@ source_filename = "slice.go"
 target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-i128:128-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.sliceLen(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.sliceLen(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #1 {
 entry:
  ret i32 %ints.len
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.sliceCap(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.sliceCap(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #1 {
 entry:
  ret i32 %ints.cap
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.sliceElement(ptr %ints.data, i32 %ints.len, i32 %ints.cap, i32 %index, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.sliceElement(ptr %ints.data, i32 %ints.len, i32 %ints.cap, i32 %index, ptr %context) unnamed_addr #1 {
 entry:
  %.not = icmp ult i32 %index, %ints.len
  br i1 %.not, label %lookup.next, label %lookup.throw
@@ -42,10 +39,10 @@ lookup.throw:                                     ; preds = %entry
  unreachable
 }

-declare void @runtime.lookupPanic(ptr) #1
+declare void @runtime.lookupPanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.sliceAppendValues(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.sliceAppendValues(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %varargs = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
@@ -66,10 +63,13 @@ entry:
  ret { ptr, i32, i32 } %4
 }

-declare { ptr, i32, i32 } @runtime.sliceAppend(ptr, ptr nocapture readonly, i32, i32, i32, i32, ptr, ptr) #1
+; Function Attrs: allockind("alloc,zeroed") allocsize(0)
+declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #2
+
+declare { ptr, i32, i32 } @runtime.sliceAppend(ptr, ptr nocapture readonly, i32, i32, i32, i32, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.sliceAppendSlice(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %added.data, i32 %added.len, i32 %added.cap, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.sliceAppendSlice(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %added.data, i32 %added.len, i32 %added.cap, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr %added.data, i32 %ints.len, i32 %ints.cap, i32 %added.len, i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
@@ -84,7 +84,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.sliceCopy(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.sliceCopy(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #1 {
 entry:
  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
  %copy.size = shl nuw i32 %copy.n, 2
@@ -99,7 +99,7 @@ declare i32 @llvm.umin.i32(i32, i32) #3
 declare void @llvm.memmove.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #4

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.makeByteSlice(i32 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.makeByteSlice(i32 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %slice.maxcap = icmp slt i32 %len, 0
@@ -118,10 +118,10 @@ slice.throw:                                      ; preds = %entry
  unreachable
 }

-declare void @runtime.slicePanic(ptr) #1
+declare void @runtime.slicePanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.makeInt16Slice(i32 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.makeInt16Slice(i32 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %slice.maxcap = icmp slt i32 %len, 0
@@ -142,7 +142,7 @@ slice.throw:                                      ; preds = %entry
 }

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.makeArraySlice(i32 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.makeArraySlice(i32 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %slice.maxcap = icmp ugt i32 %len, 1431655765
@@ -163,7 +163,7 @@ slice.throw:                                      ; preds = %entry
 }

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.makeInt32Slice(i32 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.makeInt32Slice(i32 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %slice.maxcap = icmp ugt i32 %len, 1073741823
@@ -184,7 +184,7 @@ slice.throw:                                      ; preds = %entry
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.Add32(ptr %p, i32 %len, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.Add32(ptr %p, i32 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = getelementptr i8, ptr %p, i32 %len
@@ -193,7 +193,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.Add64(ptr %p, i64 %len, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.Add64(ptr %p, i64 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = trunc i64 %len to i32
@@ -203,7 +203,7 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden ptr @main.SliceToArray(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.SliceToArray(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #1 {
 entry:
  %0 = icmp ult i32 %s.len, 4
  br i1 %0, label %slicetoarray.throw, label %slicetoarray.next
@@ -216,10 +216,10 @@ slicetoarray.throw:                               ; preds = %entry
  unreachable
 }

-declare void @runtime.sliceToArrayPointerPanic(ptr) #1
+declare void @runtime.sliceToArrayPointerPanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden ptr @main.SliceToArrayConst(ptr %context) unnamed_addr #2 {
+define hidden ptr @main.SliceToArrayConst(ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %makeslice = call align 4 dereferenceable(24) ptr @runtime.alloc(i32 24, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
@@ -234,7 +234,7 @@ slicetoarray.throw:                               ; preds = %entry
 }

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.SliceInt(ptr dereferenceable_or_null(4) %ptr, i32 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.SliceInt(ptr dereferenceable_or_null(4) %ptr, i32 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = icmp ugt i32 %len, 1073741823
@@ -256,10 +256,10 @@ unsafe.Slice.throw:                               ; preds = %entry
  unreachable
 }

-declare void @runtime.unsafeSlicePanic(ptr) #1
+declare void @runtime.unsafeSlicePanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.SliceUint16(ptr dereferenceable_or_null(1) %ptr, i16 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.SliceUint16(ptr dereferenceable_or_null(1) %ptr, i16 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = icmp eq ptr %ptr, null
@@ -281,7 +281,7 @@ unsafe.Slice.throw:                               ; preds = %entry
 }

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.SliceUint64(ptr dereferenceable_or_null(4) %ptr, i64 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.SliceUint64(ptr dereferenceable_or_null(4) %ptr, i64 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = icmp ugt i64 %len, 1073741823
@@ -305,7 +305,7 @@ unsafe.Slice.throw:                               ; preds = %entry
 }

 ; Function Attrs: nounwind
-define hidden { ptr, i32, i32 } @main.SliceInt64(ptr dereferenceable_or_null(4) %ptr, i64 %len, ptr %context) unnamed_addr #2 {
+define hidden { ptr, i32, i32 } @main.SliceInt64(ptr dereferenceable_or_null(4) %ptr, i64 %len, ptr %context) unnamed_addr #1 {
 entry:
  %stackalloc = alloca i8, align 1
  %0 = icmp ugt i64 %len, 1073741823
@@ -328,9 +328,9 @@ unsafe.Slice.throw:                               ; preds = %entry
  unreachable
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #3 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
 attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
 attributes #5 = { nounwind }
@@ -7,37 +7,34 @@ target triple = "wasm32-unknown-wasi"

@"main$string" = internal unnamed_addr constant [3 x i8] c"foo", align 1

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: nounwind
-define hidden %runtime._string @main.someString(ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.someString(ptr %context) unnamed_addr #1 {
 entry:
  ret %runtime._string { ptr @"main$string", i32 3 }
 }

 ; Function Attrs: nounwind
-define hidden %runtime._string @main.zeroLengthString(ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.zeroLengthString(ptr %context) unnamed_addr #1 {
 entry:
  ret %runtime._string zeroinitializer
 }

 ; Function Attrs: nounwind
-define hidden i32 @main.stringLen(ptr readonly %s.data, i32 %s.len, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.stringLen(ptr readonly %s.data, i32 %s.len, ptr %context) unnamed_addr #1 {
 entry:
  ret i32 %s.len
 }

 ; Function Attrs: nounwind
-define hidden i8 @main.stringIndex(ptr readonly %s.data, i32 %s.len, i32 %index, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.stringIndex(ptr readonly %s.data, i32 %s.len, i32 %index, ptr %context) unnamed_addr #1 {
 entry:
  %.not = icmp ult i32 %index, %s.len
  br i1 %.not, label %lookup.next, label %lookup.throw
@@ -48,40 +45,40 @@ lookup.next:                                      ; preds = %entry
  ret i8 %1

 lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(ptr undef) #3
+  call void @runtime.lookupPanic(ptr undef) #2
  unreachable
 }

-declare void @runtime.lookupPanic(ptr) #1
+declare void @runtime.lookupPanic(ptr) #0

 ; Function Attrs: nounwind
-define hidden i1 @main.stringCompareEqual(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.stringCompareEqual(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call i1 @runtime.stringEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr undef) #3
+  %0 = call i1 @runtime.stringEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr undef) #2
  ret i1 %0
 }

-declare i1 @runtime.stringEqual(ptr readonly, i32, ptr readonly, i32, ptr) #1
+declare i1 @runtime.stringEqual(ptr readonly, i32, ptr readonly, i32, ptr) #0

 ; Function Attrs: nounwind
-define hidden i1 @main.stringCompareUnequal(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.stringCompareUnequal(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call i1 @runtime.stringEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr undef) #3
+  %0 = call i1 @runtime.stringEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr undef) #2
  %1 = xor i1 %0, true
  ret i1 %1
 }

 ; Function Attrs: nounwind
-define hidden i1 @main.stringCompareLarger(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.stringCompareLarger(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call i1 @runtime.stringLess(ptr %s2.data, i32 %s2.len, ptr %s1.data, i32 %s1.len, ptr undef) #3
+  %0 = call i1 @runtime.stringLess(ptr %s2.data, i32 %s2.len, ptr %s1.data, i32 %s1.len, ptr undef) #2
  ret i1 %0
 }

-declare i1 @runtime.stringLess(ptr readonly, i32, ptr readonly, i32, ptr) #1
+declare i1 @runtime.stringLess(ptr readonly, i32, ptr readonly, i32, ptr) #0

 ; Function Attrs: nounwind
-define hidden i8 @main.stringLookup(ptr readonly %s.data, i32 %s.len, i8 %x, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.stringLookup(ptr readonly %s.data, i32 %s.len, i8 %x, ptr %context) unnamed_addr #1 {
 entry:
  %0 = zext i8 %x to i32
  %.not = icmp ugt i32 %s.len, %0
@@ -93,11 +90,10 @@ lookup.next:                                      ; preds = %entry
  ret i8 %2

 lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(ptr undef) #3
+  call void @runtime.lookupPanic(ptr undef) #2
  unreachable
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind }
@@ -5,19 +5,16 @@ target triple = "wasm32-unknown-wasi"

 %main.hasPadding = type { i1, i32, i1 }

-; Function Attrs: allockind("alloc,zeroed") allocsize(0)
-declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
-
-declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #1
+declare void @runtime.trackPointer(ptr nocapture readonly, ptr, ptr) #0

 ; Function Attrs: nounwind
-define hidden void @main.init(ptr %context) unnamed_addr #2 {
+define hidden void @main.init(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

 ; Function Attrs: noinline nounwind
-define hidden i32 @main.testZeroGet(ptr dereferenceable_or_null(40) %m, i1 %s.b1, i32 %s.i, i1 %s.b2, ptr %context) unnamed_addr #3 {
+define hidden i32 @main.testZeroGet(ptr dereferenceable_or_null(48) %m, i1 %s.b1, i32 %s.i, i1 %s.b2, ptr %context) unnamed_addr #2 {
 entry:
  %hashmap.key = alloca %main.hasPadding, align 8
  %hashmap.value = alloca i32, align 4
@@ -27,29 +24,23 @@ entry:
  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %hashmap.value)
  call void @llvm.lifetime.start.p0(i64 12, ptr nonnull %hashmap.key)
  store %main.hasPadding %2, ptr %hashmap.key, align 4
-  %3 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 1
-  call void @runtime.memzero(ptr nonnull %3, i32 3, ptr undef) #5
-  %4 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 9
-  call void @runtime.memzero(ptr nonnull %4, i32 3, ptr undef) #5
-  %5 = call i1 @runtime.hashmapBinaryGet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, i32 4, ptr undef) #5
+  %3 = call i1 @runtime.hashmapGenericGet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, i32 4, ptr undef) #4
  call void @llvm.lifetime.end.p0(i64 12, ptr nonnull %hashmap.key)
-  %6 = load i32, ptr %hashmap.value, align 4
+  %4 = load i32, ptr %hashmap.value, align 4
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %hashmap.value)
-  ret i32 %6
+  ret i32 %4
 }

 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
-declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #4
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3

-declare void @runtime.memzero(ptr, i32, ptr) #1
-
-declare i1 @runtime.hashmapBinaryGet(ptr dereferenceable_or_null(40), ptr, ptr, i32, ptr) #1
+declare i1 @runtime.hashmapGenericGet(ptr dereferenceable_or_null(48), ptr nocapture, ptr nocapture, i32, ptr) #0

 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
-declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #4
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #3

 ; Function Attrs: noinline nounwind
-define hidden void @main.testZeroSet(ptr dereferenceable_or_null(40) %m, i1 %s.b1, i32 %s.i, i1 %s.b2, ptr %context) unnamed_addr #3 {
+define hidden void @main.testZeroSet(ptr dereferenceable_or_null(48) %m, i1 %s.b1, i32 %s.i, i1 %s.b2, ptr %context) unnamed_addr #2 {
 entry:
  %hashmap.key = alloca %main.hasPadding, align 8
  %hashmap.value = alloca i32, align 4
@@ -60,20 +51,16 @@ entry:
  store i32 5, ptr %hashmap.value, align 4
  call void @llvm.lifetime.start.p0(i64 12, ptr nonnull %hashmap.key)
  store %main.hasPadding %2, ptr %hashmap.key, align 4
-  %3 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 1
-  call void @runtime.memzero(ptr nonnull %3, i32 3, ptr undef) #5
-  %4 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 9
-  call void @runtime.memzero(ptr nonnull %4, i32 3, ptr undef) #5
-  call void @runtime.hashmapBinarySet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, ptr undef) #5
+  call void @runtime.hashmapGenericSet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, ptr undef) #4
  call void @llvm.lifetime.end.p0(i64 12, ptr nonnull %hashmap.key)
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %hashmap.value)
  ret void
 }

-declare void @runtime.hashmapBinarySet(ptr dereferenceable_or_null(40), ptr, ptr, ptr) #1
+declare void @runtime.hashmapGenericSet(ptr dereferenceable_or_null(48), ptr nocapture, ptr nocapture, ptr) #0

 ; Function Attrs: noinline nounwind
-define hidden i32 @main.testZeroArrayGet(ptr dereferenceable_or_null(40) %m, [2 x %main.hasPadding] %s, ptr %context) unnamed_addr #3 {
+define hidden i32 @main.testZeroArrayGet(ptr dereferenceable_or_null(48) %m, [2 x %main.hasPadding] %s, ptr %context) unnamed_addr #2 {
 entry:
  %hashmap.key = alloca [2 x %main.hasPadding], align 8
  %hashmap.value = alloca i32, align 4
@@ -84,23 +71,15 @@ entry:
  %hashmap.key.repack1 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 12
  %s.elt2 = extractvalue [2 x %main.hasPadding] %s, 1
  store %main.hasPadding %s.elt2, ptr %hashmap.key.repack1, align 4
-  %0 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 1
-  call void @runtime.memzero(ptr nonnull %0, i32 3, ptr undef) #5
-  %1 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 9
-  call void @runtime.memzero(ptr nonnull %1, i32 3, ptr undef) #5
-  %2 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 13
-  call void @runtime.memzero(ptr nonnull %2, i32 3, ptr undef) #5
-  %3 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 21
-  call void @runtime.memzero(ptr nonnull %3, i32 3, ptr undef) #5
-  %4 = call i1 @runtime.hashmapBinaryGet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, i32 4, ptr undef) #5
+  %0 = call i1 @runtime.hashmapGenericGet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, i32 4, ptr undef) #4
  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %hashmap.key)
-  %5 = load i32, ptr %hashmap.value, align 4
+  %1 = load i32, ptr %hashmap.value, align 4
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %hashmap.value)
-  ret i32 %5
+  ret i32 %1
 }

 ; Function Attrs: noinline nounwind
-define hidden void @main.testZeroArraySet(ptr dereferenceable_or_null(40) %m, [2 x %main.hasPadding] %s, ptr %context) unnamed_addr #3 {
+define hidden void @main.testZeroArraySet(ptr dereferenceable_or_null(48) %m, [2 x %main.hasPadding] %s, ptr %context) unnamed_addr #2 {
 entry:
  %hashmap.key = alloca [2 x %main.hasPadding], align 8
  %hashmap.value = alloca i32, align 4
@@ -112,29 +91,20 @@ entry:
  %hashmap.key.repack1 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 12
  %s.elt2 = extractvalue [2 x %main.hasPadding] %s, 1
  store %main.hasPadding %s.elt2, ptr %hashmap.key.repack1, align 4
-  %0 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 1
-  call void @runtime.memzero(ptr nonnull %0, i32 3, ptr undef) #5
-  %1 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 9
-  call void @runtime.memzero(ptr nonnull %1, i32 3, ptr undef) #5
-  %2 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 13
-  call void @runtime.memzero(ptr nonnull %2, i32 3, ptr undef) #5
-  %3 = getelementptr inbounds nuw i8, ptr %hashmap.key, i32 21
-  call void @runtime.memzero(ptr nonnull %3, i32 3, ptr undef) #5
-  call void @runtime.hashmapBinarySet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, ptr undef) #5
+  call void @runtime.hashmapGenericSet(ptr %m, ptr nonnull %hashmap.key, ptr nonnull %hashmap.value, ptr undef) #4
  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %hashmap.key)
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %hashmap.value)
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.main(ptr %context) unnamed_addr #2 {
+define hidden void @main.main(ptr %context) unnamed_addr #1 {
 entry:
  ret void
 }

-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { noinline nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #4 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
-attributes #5 = { nounwind }
+attributes #0 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { noinline nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #3 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #4 = { nounwind }
@@ -59,7 +59,6 @@ func TestCorpus(t *testing.T) {
 	}

 	for _, repo := range repos {
-		repo := repo
 		name := repo.Repo
 		if repo.Tags != "" {
 			name += "(" + strings.ReplaceAll(repo.Tags, " ", "-") + ")"
@@ -132,7 +131,6 @@ func TestCorpus(t *testing.T) {
 			}

 			for _, dir := range repo.Subdirs {
-				dir := dir
 				t.Run(dir.Pkg, func(t *testing.T) {
 					t.Parallel()

@@ -116,10 +116,7 @@ func Diff(oldName string, old []byte, newName string, new []byte) []byte {

 		// End chunk with common lines for context.
 		if len(ctext) > 0 {
-			n := end.x - start.x
-			if n > C {
-				n = C
-			}
+			n := min(end.x-start.x, C)
 			for _, s := range x[start.x : start.x+n] {
 				ctext = append(ctext, " "+s)
 				count.x++
@@ -234,7 +231,7 @@ func tgs(x, y []string) []pair {
 	for i := range T {
 		T[i] = n + 1
 	}
-	for i := 0; i < n; i++ {
+	for i := range n {
 		k := sort.Search(n, func(k int) bool {
 			return T[k] >= J[i]
 		})
@@ -38,6 +38,7 @@ func TestErrors(t *testing.T) {
 		{name: "loader-invaliddep"},
 		{name: "loader-invalidpackage"},
 		{name: "loader-nopackage"},
+		{name: "noheap", target: "cortex-m-qemu"},
 		{name: "optimizer"},
 		{name: "syntax"},
 		{name: "types"},
@@ -135,7 +136,7 @@ func readErrorMessages(t *testing.T, file string) string {
 	}

 	var errors []string
-	for _, line := range strings.Split(string(data), "\n") {
+	for line := range strings.SplitSeq(string(data), "\n") {
 		if strings.HasPrefix(line, "// ERROR: ") {
 			errors = append(errors, strings.TrimRight(line[len("// ERROR: "):], "\r\n"))
 		}
@@ -1,6 +1,6 @@
 module github.com/tinygo-org/tinygo

-go 1.23.0
+go 1.25.0

 require (
 	github.com/aykevl/go-wasm v0.0.2-0.20250317121156-42b86c494139
@@ -15,22 +15,21 @@ require (
 	github.com/mgechev/revive v1.3.9
 	github.com/sigurn/crc16 v0.0.0-20211026045750-20ab5afb07e3
 	github.com/tetratelabs/wazero v1.9.0
-	go.bug.st/serial v1.6.4
+	go.bug.st/serial v1.8.0
 	go.bytecodealliance.org v0.6.2
 	go.bytecodealliance.org/cm v0.2.2
-	golang.org/x/net v0.35.0
-	golang.org/x/sys v0.30.0
-	golang.org/x/tools v0.30.0
+	golang.org/x/net v0.56.0
+	golang.org/x/sys v0.47.0
+	golang.org/x/tools v0.47.0
 	gopkg.in/yaml.v2 v2.4.0
-	tinygo.org/x/espflasher v0.6.0
-	tinygo.org/x/go-llvm v0.0.0-20250422114502-b8f170971e74
+	tinygo.org/x/espflasher v0.7.1
+	tinygo.org/x/go-llvm v0.0.0-20260721072906-185673ef46a5
 )

 require (
 	github.com/BurntSushi/toml v1.4.0 // indirect
 	github.com/chavacava/garif v0.1.0 // indirect
 	github.com/coreos/go-semver v0.3.1 // indirect
-	github.com/creack/goselect v0.1.2 // indirect
 	github.com/docker/libtrust v0.0.0-20160708172513-aabc10ec26b7 // indirect
 	github.com/fatih/color v1.17.0 // indirect
 	github.com/fatih/structtag v1.2.0 // indirect
@@ -48,6 +47,6 @@ require (
 	github.com/spf13/afero v1.11.0 // indirect
 	github.com/ulikunitz/xz v0.5.12 // indirect
 	github.com/urfave/cli/v3 v3.0.0-beta1 // indirect
-	golang.org/x/mod v0.23.0 // indirect
-	golang.org/x/text v0.22.0 // indirect
+	golang.org/x/mod v0.37.0 // indirect
+	golang.org/x/text v0.38.0 // indirect
 )
@@ -8,8 +8,6 @@ github.com/chavacava/garif v0.1.0 h1:2JHa3hbYf5D9dsgseMKAmc/MZ109otzgNFk5s87H9Pc
 github.com/chavacava/garif v0.1.0/go.mod h1:XMyYCkEL58DF0oyW4qDjjnPWONs2HBqYKI+UIPD+Gww=
 github.com/coreos/go-semver v0.3.1 h1:yi21YpKnrx1gt5R+la8n5WgS0kCrsPp33dmEyHReZr4=
 github.com/coreos/go-semver v0.3.1/go.mod h1:irMmmIw/7yzSRPWryHsK7EYSg09caPQL03VsM8rvUec=
-github.com/creack/goselect v0.1.2 h1:2DNy14+JPjRBgPzAd1thbQp4BSIihxcBf0IXhQXDRa0=
-github.com/creack/goselect v0.1.2/go.mod h1:a/NhLweNvqIYMuxcMOuWY516Cimucms3DglDzQP3hKY=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -23,6 +21,8 @@ github.com/gofrs/flock v0.8.1 h1:+gYjHKf32LDeiEEFhQaotPbLuUXjY5ZqxKgXy7n59aw=
 github.com/gofrs/flock v0.8.1/go.mod h1:F1TvTiK9OcQqauNUHlbJvyl9Qa1QvF/gOUDKA14jxHU=
 github.com/golangci/misspell v0.6.0 h1:JCle2HUTNWirNlDIAUO44hUsKhOFqGPoC4LZxlaSXDs=
 github.com/golangci/misspell v0.6.0/go.mod h1:keMNyY6R9isGaSAu+4Q8NMBwMPkh15Gtc8UCVoDtAWo=
+github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
+github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
 github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf h1:7+FW5aGwISbqUtkfmIpZJGRgNFg2ioYPvFaUxdqpDsg=
 github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf/go.mod h1:RpwtwJQFrIEPstU94h88MWPXP2ektJZ8cZ0YntAmXiE=
 github.com/hashicorp/go-version v1.7.0 h1:5tqGy27NaOTB8yJKUZELlFAS/LTKJkrmONwQKeRZfjY=
@@ -87,30 +87,30 @@ github.com/ulikunitz/xz v0.5.12 h1:37Nm15o69RwBkXM0J6A5OlE67RZTfzUxTj8fB3dfcsc=
 github.com/ulikunitz/xz v0.5.12/go.mod h1:nbz6k7qbPmH4IRqmfOplQw/tblSgqTqBwxkY0oWt/14=
 github.com/urfave/cli/v3 v3.0.0-beta1 h1:6DTaaUarcM0wX7qj5Hcvs+5Dm3dyUTBbEwIWAjcw9Zg=
 github.com/urfave/cli/v3 v3.0.0-beta1/go.mod h1:FnIeEMYu+ko8zP1F9Ypr3xkZMIDqW3DR92yUtY39q1Y=
-go.bug.st/serial v1.6.4 h1:7FmqNPgVp3pu2Jz5PoPtbZ9jJO5gnEnZIvnI1lzve8A=
-go.bug.st/serial v1.6.4/go.mod h1:nofMJxTeNVny/m6+KaafC6vJGj3miwQZ6vW4BZUGJPI=
+go.bug.st/serial v1.8.0 h1:ZtnmN8aYXtPlTghwSvDWPHKBHL9TM6oFDa+KpSn4SQE=
+go.bug.st/serial v1.8.0/go.mod h1:d0MmS16Qt9b1m06yoYRNUXhRRTJV5Qg2S5EKqQtnayQ=
 go.bytecodealliance.org v0.6.2 h1:Jy4u5DVmSkXgsnwojBhJ+AD/YsJsR3VzVnxF0xRCqTQ=
 go.bytecodealliance.org v0.6.2/go.mod h1:gqjTJm0y9NSksG4py/lSjIQ/SNuIlOQ+hCIEPQwtJgA=
 go.bytecodealliance.org/cm v0.2.2 h1:M9iHS6qs884mbQbIjtLX1OifgyPG9DuMs2iwz8G4WQA=
 go.bytecodealliance.org/cm v0.2.2/go.mod h1:JD5vtVNZv7sBoQQkvBvAAVKJPhR/bqBH7yYXTItMfZI=
-golang.org/x/mod v0.23.0 h1:Zb7khfcRGKk+kqfxFaP5tZqCnDZMjC5VtUBs87Hr6QM=
-golang.org/x/mod v0.23.0/go.mod h1:6SkKJ3Xj0I0BrPOZoBy3bdMptDDU9oJrpohJ3eWZ1fY=
-golang.org/x/net v0.35.0 h1:T5GQRQb2y08kTAByq9L4/bz8cipCdA8FbRTXewonqY8=
-golang.org/x/net v0.35.0/go.mod h1:EglIi67kWsHKlRzzVMUD93VMSWGFOMSZgxFjparz1Qk=
-golang.org/x/sync v0.11.0 h1:GGz8+XQP4FvTTrjZPzNKTMFtSXH80RAzG+5ghFPgK9w=
-golang.org/x/sync v0.11.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/mod v0.37.0 h1:vF1DjpVEshcIqoEaauuHebaLk1O1forxjxBaVn884JQ=
+golang.org/x/mod v0.37.0/go.mod h1:m8S8VeM9r4dzDwjrKO0a1sZP3YjeMamRRlD+fmR2Q/0=
+golang.org/x/net v0.56.0 h1:Rw8j/hFzGvJUZwNBXnAtf5sVDVt+65SK2C7IxCxZt5o=
+golang.org/x/net v0.56.0/go.mod h1:D3Ku6r+V6JROoZK144D2XfMHFcMq/0zSfLelVTCFKec=
+golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
+golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20190222072716-a9d3bda3a223/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20191008105621-543471e840be/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20191120155948-bd437916bb0e/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220811171246-fbc7d0a398ab/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
-golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/text v0.22.0 h1:bofq7m3/HAFvbF51jz3Q9wLg3jkvSPuiZu/pD1XwgtM=
-golang.org/x/text v0.22.0/go.mod h1:YRoo4H8PVmsu+E3Ou7cqLVH8oXWIHVoX0jqUWALQhfY=
-golang.org/x/tools v0.30.0 h1:BgcpHewrV5AUp2G9MebG4XPFI1E2W41zU1SaqVA9vJY=
-golang.org/x/tools v0.30.0/go.mod h1:c347cR/OJfw5TI+GfX7RUPNMdDRRbjvYTS0jPyvsVtY=
+golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/text v0.38.0 h1:sXmwo9DwP3OK9EZ7PqAdaooSGozfl/3a6/xJcbzPRhE=
+golang.org/x/text v0.38.0/go.mod h1:YXZt3QhHUKYT53r2lLKFIVi6Ao1jdzrTR/KQ09qyxF4=
+golang.org/x/tools v0.47.0 h1:7Kn5x/d1svx/PzryTsqeoZN4TZwqeH5pGWjefhLi/1Q=
+golang.org/x/tools v0.47.0/go.mod h1:dFHnyTvFWY212G+h7ZY4Vsp/K3U4/7W9TyVaAul8uCA=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v2 v2.4.0 h1:D8xgwECY7CYvx+Y2n4sBz93Jn9JRvxdiyyo8CTfuKaY=
@@ -118,7 +118,7 @@ gopkg.in/yaml.v2 v2.4.0/go.mod h1:RDklbk79AGWmwhnvt/jBztapEOGDOx6ZbXqjP6csGnQ=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
-tinygo.org/x/espflasher v0.6.0 h1:CHbGMHAIWq1tB8FKd/QwBpNFw1jvHHxpmvZiF+QOYUo=
-tinygo.org/x/espflasher v0.6.0/go.mod h1:tr5u08HoE67WD5zxJesCiiVF/R1b6Akz3yXwh5zah8U=
-tinygo.org/x/go-llvm v0.0.0-20250422114502-b8f170971e74 h1:ovavgTdIBWCH8YWlcfq9gkpoyT1+IxMKSn+Df27QwE8=
-tinygo.org/x/go-llvm v0.0.0-20250422114502-b8f170971e74/go.mod h1:GFbusT2VTA4I+l4j80b17KFK+6whv69Wtny5U+T8RR0=
+tinygo.org/x/espflasher v0.7.1 h1:oJ+tt58QltP+c5+iHGRdCNNnVaXZ9yYXyWK/F747BVY=
+tinygo.org/x/espflasher v0.7.1/go.mod h1:a3l/4jOSMf9vhaPtEmGbCwtrC5oiZiNKx1i8jUkyyRI=
+tinygo.org/x/go-llvm v0.0.0-20260721072906-185673ef46a5 h1:0PKRhM1INWAi7PdIng1BHMPTDaRcOmv1UymdTgZ76Jo=
+tinygo.org/x/go-llvm v0.0.0-20260721072906-185673ef46a5/go.mod h1:GFbusT2VTA4I+l4j80b17KFK+6whv69Wtny5U+T8RR0=
@@ -151,6 +151,8 @@ func Get(name string) string {
 			panic("could not find cache dir: " + err.Error())
 		}
 		return filepath.Join(dir, "tinygo")
+	case "CGO_CFLAGS":
+		return os.Getenv("CGO_CFLAGS")
 	case "CGO_ENABLED":
 		// Always enable CGo. It is required by a number of targets, including
 		// macOS and the rp2040.
--- a/Show More
+++ b/Show More