compiler: implement copy directly

The compiler now implements the copy builtin directly instead of calling sliceCopy.
The length is calculated with the llvm.umax.* intrinsics, and the move is performed by llvm.memmove.*.
Both of these operations are easily understood by LLVM's optimization passes.
The type's alignment is also provided to llvm.memmove.*, which is useful when rewriting the move.

Interp no longer needs to reimplement sliceCopy.
Some edge case handling was implemented by sliceCopy but not llvm.memmove.*/llvm.memcpy.*.
I copied this over, so copies of external slices should work now.
Volatile moves/copies are now run at runtime by interp.

There is a 4-byte size increase due to some confusing length logic in sendUSBPacket.
I will look at sendUSBPacket in a future PR.

builder/sizes_test.go

@@ -44,7 +44,7 @@ func TestBinarySize(t *testing.T) {
 		// microcontrollers
 		{"hifive1b", "examples/echo", 3668, 280, 0, 2244},
 		{"microbit", "examples/serial", 2694, 342, 8, 2248},
-		{"wioterminal", "examples/pininterrupt", 6833, 1491, 120, 6888},
+		{"wioterminal", "examples/pininterrupt", 6837, 1491, 120, 6888},
 
 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the

compiler/compiler.go

@@ -1682,13 +1682,41 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 	case "copy":
 		dst := argValues[0]
 		src := argValues[1]
+		// Fetch the lengths.
 		dstLen := b.CreateExtractValue(dst, 1, "copy.dstLen")
 		srcLen := b.CreateExtractValue(src, 1, "copy.srcLen")
-		dstBuf := b.CreateExtractValue(dst, 0, "copy.dstArray")
+		// Find the minimum of the lengths.
-		srcBuf := b.CreateExtractValue(src, 0, "copy.srcArray")
+		minFuncName := "llvm.umin.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+		minFunc := b.mod.NamedFunction(minFuncName)
+		if minFunc.IsNil() {
+			fnType := llvm.FunctionType(b.uintptrType, []llvm.Type{b.uintptrType, b.uintptrType}, false)
+			minFunc = llvm.AddFunction(b.mod, minFuncName, fnType)
+		}
+		minLen := b.CreateCall(minFunc.GlobalValueType(), minFunc, []llvm.Value{dstLen, srcLen}, "copy.n")
+		// Multiply the length by the element size.
 		elemType := b.getLLVMType(argTypes[0].Underlying().(*types.Slice).Elem())
 		elemSize := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(elemType), false)
-		return b.createRuntimeCall("sliceCopy", []llvm.Value{dstBuf, srcBuf, dstLen, srcLen, elemSize}, "copy.n"), nil
+		// NOTE: This is also NSW when uintptr is int, but we can only choose one through the C API?
+		size := b.CreateNUWMul(minLen, elemSize, "copy.size")
+		// Fetch the pointers.
+		dstBuf := b.CreateExtractValue(dst, 0, "copy.dstPtr")
+		srcBuf := b.CreateExtractValue(src, 0, "copy.srcPtr")
+		// Create a memcpy.
+		call := b.createMemCopy("memmove", dstBuf, srcBuf, size)
+		align := b.targetData.ABITypeAlignment(elemType)
+		if align > 1 {
+			// Apply the type's alignment to the arguments.
+			// LLVM sometimes turns constant-length moves into loads and stores.
+			// It may use this alignment for the created loads and stores.
+			alignAttr := b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align))
+			call.AddCallSiteAttribute(1, alignAttr)
+			call.AddCallSiteAttribute(2, alignAttr)
+		}
+		// Extend and return the copied length.
+		if b.targetData.TypeAllocSize(minLen.Type()) < b.targetData.TypeAllocSize(b.intType) {
+			minLen = b.CreateZExt(minLen, b.intType, "copy.n.zext")
+		}
+		return minLen, nil
 	case "delete":
 		m := argValues[0]
 		key := argValues[1]

compiler/intrinsics.go

@@ -50,19 +50,24 @@ func (b *builder) defineIntrinsicFunction() {
 // and will otherwise be lowered to regular libc memcpy/memmove calls.
 func (b *builder) createMemoryCopyImpl() {
 	b.createFunctionStart(true)
-	fnName := "llvm." + b.fn.Name() + ".p0.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+	params := b.fn.Params[0:3]
+	b.createMemCopy(
+		b.fn.Name(),
+		b.getValue(params[0], getPos(b.fn)),
+		b.getValue(params[1], getPos(b.fn)),
+		b.getValue(params[2], getPos(b.fn)),
+	)
+	b.CreateRetVoid()
+}
+
+func (b *builder) createMemCopy(kind string, dst, src, len llvm.Value) llvm.Value {
+	fnName := "llvm." + kind + ".p0.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
 	llvmFn := b.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
 		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType, b.dataPtrType, b.uintptrType, b.ctx.Int1Type()}, false)
 		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
 	}
-	var params []llvm.Value
+	return b.CreateCall(llvmFn.GlobalValueType(), llvmFn, []llvm.Value{dst, src, len, llvm.ConstInt(b.ctx.Int1Type(), 0, false)}, "")
-	for _, param := range b.fn.Params {
-		params = append(params, b.getValue(param, getPos(b.fn)))
-	}
-	params = append(params, llvm.ConstInt(b.ctx.Int1Type(), 0, false))
-	b.CreateCall(llvmFn.GlobalValueType(), llvmFn, params, "")
-	b.CreateRetVoid()
 }
 
 // createMemoryZeroImpl creates calls to llvm.memset.* to zero a block of

compiler/symbol.go

@@ -184,12 +184,6 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		// be modified.
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
-	case "runtime.sliceCopy":
-		// Copying a slice won't capture any of the parameters.
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("writeonly"), 0))
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
-		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
-		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 	case "runtime.stringFromBytes":
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))

compiler/testdata/goroutine-cortex-m-qemu-tasks.ll

@@ -17,8 +17,8 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr undef) #11
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #11
   ret void
 }
 
@@ -28,7 +28,7 @@ declare void @main.regularFunction(i32, ptr) #2
 define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #3 {
 entry:
   %unpack.int = ptrtoint ptr %0 to i32
-  call void @main.regularFunction(i32 %unpack.int, ptr undef) #9
+  call void @main.regularFunction(i32 %unpack.int, ptr undef) #11
   ret void
 }
 
@@ -39,8 +39,8 @@ declare void @"internal/task.start"(i32, ptr, i32, ptr) #2
 ; Function Attrs: nounwind
 define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr undef) #9
+  %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) #9
+  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
   ret void
 }
 
@@ -61,18 +61,18 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
-  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  %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) #9
+  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, 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
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr undef) #11
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #11
   %2 = load i32, ptr %n, align 4
-  call void @runtime.printlock(ptr undef) #9
+  call void @runtime.printlock(ptr undef) #11
-  call void @runtime.printint32(i32 %2, ptr undef) #9
+  call void @runtime.printint32(i32 %2, ptr undef) #11
-  call void @runtime.printunlock(ptr undef) #9
+  call void @runtime.printunlock(ptr undef) #11
   ret void
 }
 
@@ -102,14 +102,14 @@ declare void @runtime.printunlock(ptr) #2
 ; Function Attrs: nounwind
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, 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
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 8
   store ptr %fn.funcptr, ptr %2, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr undef) #11
-  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #11
   ret void
 }
 
@@ -121,7 +121,7 @@ entry:
   %3 = load ptr, ptr %2, align 4
   %4 = getelementptr inbounds nuw i8, ptr %0, i32 8
   %5 = load ptr, ptr %4, align 4
-  call void %5(i32 %1, ptr %3) #9
+  call void %5(i32 %1, ptr %3) #11
   ret void
 }
 
@@ -134,16 +134,21 @@ entry:
 ; 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 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(ptr %dst.data, ptr %src.data, i32 %dst.len, i32 %src.len, i32 1, ptr undef) #9
+  %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)
   ret void
 }
 
-declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #2
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #7
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+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 {
 entry:
-  call void @runtime.chanClose(ptr %ch, ptr undef) #9
+  call void @runtime.chanClose(ptr %ch, ptr undef) #11
   ret void
 }
 
@@ -152,7 +157,7 @@ declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #2
 ; Function Attrs: nounwind
 define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #11
   store ptr %itf.value, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr @"main$string", ptr %1, align 4
@@ -160,15 +165,15 @@ entry:
   store i32 4, ptr %2, align 4
   %3 = getelementptr inbounds nuw i8, ptr %0, i32 12
   store ptr %itf.typecode, ptr %3, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr undef) #11
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #11
   ret void
 }
 
-declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #7
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #9
 
 ; Function Attrs: nounwind
-define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #8 {
+define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #10 {
 entry:
   %1 = load ptr, ptr %0, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -177,7 +182,7 @@ entry:
   %5 = load i32, ptr %4, align 4
   %6 = getelementptr inbounds nuw i8, ptr %0, i32 12
   %7 = load ptr, ptr %6, align 4
-  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #9
+  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #11
   ret void
 }
 
@@ -188,6 +193,8 @@ attributes #3 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb
 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 #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$1" }
 attributes #6 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper" }
-attributes #7 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #7 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
-attributes #8 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
+attributes #8 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
-attributes #9 = { nounwind }
+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 #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 }

compiler/testdata/goroutine-wasm-asyncify.ll

@@ -19,7 +19,7 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #2 {
 entry:
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #11
   ret void
 }
 
@@ -31,8 +31,8 @@ declare void @runtime.deadlock(ptr) #1
 define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #3 {
 entry:
   %unpack.int = ptrtoint ptr %0 to i32
-  call void @main.regularFunction(i32 %unpack.int, ptr undef) #9
+  call void @main.regularFunction(i32 %unpack.int, ptr undef) #11
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -41,7 +41,7 @@ declare void @"internal/task.start"(i32, ptr, i32, ptr) #1
 ; Function Attrs: nounwind
 define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #2 {
 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) #9
+  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
 }
 
@@ -56,7 +56,7 @@ define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(ptr %0) unn
 entry:
   %unpack.int = ptrtoint ptr %0 to i32
   call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, ptr undef)
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -64,21 +64,21 @@ entry:
 define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #2 {
 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) #9
+  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #11
-  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #11
   store i32 3, ptr %n, align 4
-  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #9
+  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) #9
+  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) #9
+  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #11
-  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #9
+  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
   store ptr %n, ptr %1, align 4
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #11
   %2 = load i32, ptr %n, align 4
-  call void @runtime.printlock(ptr undef) #9
+  call void @runtime.printlock(ptr undef) #11
-  call void @runtime.printint32(i32 %2, ptr undef) #9
+  call void @runtime.printint32(i32 %2, ptr undef) #11
-  call void @runtime.printunlock(ptr undef) #9
+  call void @runtime.printunlock(ptr undef) #11
   ret void
 }
 
@@ -96,7 +96,7 @@ entry:
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 4
   %3 = load ptr, ptr %2, align 4
   call void @"main.closureFunctionGoroutine$1"(i32 %1, ptr %3)
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -110,14 +110,14 @@ declare void @runtime.printunlock(ptr) #1
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #11
-  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #9
+  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
   store ptr %fn.context, ptr %1, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 8
   store ptr %fn.funcptr, ptr %2, align 4
-  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 65536, ptr undef) #11
   ret void
 }
 
@@ -129,8 +129,8 @@ entry:
   %3 = load ptr, ptr %2, align 4
   %4 = getelementptr inbounds nuw i8, ptr %0, i32 8
   %5 = load ptr, ptr %4, align 4
-  call void %5(i32 %1, ptr %3) #9
+  call void %5(i32 %1, ptr %3) #11
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -143,16 +143,21 @@ entry:
 ; 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 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(ptr %dst.data, ptr %src.data, i32 %dst.len, i32 %src.len, i32 1, ptr undef) #9
+  %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)
   ret void
 }
 
-declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #1
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #7
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+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 {
 entry:
-  call void @runtime.chanClose(ptr %ch, ptr undef) #9
+  call void @runtime.chanClose(ptr %ch, ptr undef) #11
   ret void
 }
 
@@ -162,8 +167,8 @@ declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #1
 define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #11
-  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #9
+  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
   store ptr @"main$string", ptr %1, align 4
@@ -171,14 +176,14 @@ entry:
   store i32 4, ptr %2, align 4
   %3 = getelementptr inbounds nuw i8, ptr %0, i32 12
   store ptr %itf.typecode, ptr %3, align 4
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #11
   ret void
 }
 
-declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #7
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #9
 
 ; Function Attrs: nounwind
-define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #8 {
+define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #10 {
 entry:
   %1 = load ptr, ptr %0, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -187,8 +192,8 @@ entry:
   %5 = load i32, ptr %4, align 4
   %6 = getelementptr inbounds nuw i8, ptr %0, i32 12
   %7 = load ptr, ptr %6, align 4
-  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #9
+  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #11
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -199,6 +204,8 @@ attributes #3 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+cal
 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 #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$1" }
 attributes #6 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper" }
-attributes #7 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #7 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
-attributes #8 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
+attributes #8 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
-attributes #9 = { nounwind }
+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 #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 }

compiler/testdata/slice.ll

@@ -38,7 +38,7 @@ lookup.next:                                      ; preds = %entry
   ret i32 %1
 
 lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(ptr undef) #3
+  call void @runtime.lookupPanic(ptr undef) #5
   unreachable
 }
 
@@ -48,21 +48,21 @@ declare void @runtime.lookupPanic(ptr) #1
 define hidden { ptr, i32, i32 } @main.sliceAppendValues(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #2 {
 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) #3
+  %varargs = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
-  call void @runtime.trackPointer(ptr nonnull %varargs, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %varargs, ptr nonnull %stackalloc, ptr undef) #5
   store i32 1, ptr %varargs, align 4
   %0 = getelementptr inbounds nuw i8, ptr %varargs, i32 4
   store i32 2, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %varargs, i32 8
   store i32 3, ptr %1, align 4
-  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr nonnull %varargs, i32 %ints.len, i32 %ints.cap, i32 3, i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr nonnull %varargs, i32 %ints.len, i32 %ints.cap, i32 3, i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %append.newPtr = extractvalue { ptr, i32, i32 } %append.new, 0
   %append.newLen = extractvalue { ptr, i32, i32 } %append.new, 1
   %append.newCap = extractvalue { ptr, i32, i32 } %append.new, 2
   %2 = insertvalue { ptr, i32, i32 } undef, ptr %append.newPtr, 0
   %3 = insertvalue { ptr, i32, i32 } %2, i32 %append.newLen, 1
   %4 = insertvalue { ptr, i32, i32 } %3, i32 %append.newCap, 2
-  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %4
 }
 
@@ -72,25 +72,31 @@ declare { ptr, i32, i32 } @runtime.sliceAppend(ptr, ptr nocapture readonly, i32,
 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 {
 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) #3
+  %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
   %append.newPtr = extractvalue { ptr, i32, i32 } %append.new, 0
   %append.newLen = extractvalue { ptr, i32, i32 } %append.new, 1
   %append.newCap = extractvalue { ptr, i32, i32 } %append.new, 2
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %append.newPtr, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %append.newLen, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %append.newCap, 2
-  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 }
 
 ; 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 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(ptr %dst.data, ptr %src.data, i32 %dst.len, i32 %src.len, i32 4, ptr undef) #3
+  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
+  %copy.size = shl nuw i32 %copy.n, 2
+  call void @llvm.memmove.p0.p0.i32(ptr align 4 %dst.data, ptr align 4 %src.data, i32 %copy.size, i1 false)
   ret i32 %copy.n
 }
 
-declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #1
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #3
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+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 {
@@ -100,15 +106,15 @@ entry:
   br i1 %slice.maxcap, label %slice.throw, label %slice.next
 
 slice.next:                                       ; preds = %entry
-  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %len, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %len, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -123,15 +129,15 @@ entry:
 
 slice.next:                                       ; preds = %entry
   %makeslice.cap = shl nuw i32 %len, 1
-  %makeslice.buf = call align 2 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 2 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -144,15 +150,15 @@ entry:
 
 slice.next:                                       ; preds = %entry
   %makeslice.cap = mul i32 %len, 3
-  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -165,15 +171,15 @@ entry:
 
 slice.next:                                       ; preds = %entry
   %makeslice.cap = shl nuw i32 %len, 2
-  %makeslice.buf = call align 4 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 4 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -182,7 +188,7 @@ define hidden ptr @main.Add32(ptr %p, i32 %len, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
   %0 = getelementptr i8, ptr %p, i32 %len
-  call void @runtime.trackPointer(ptr %0, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %0, ptr nonnull %stackalloc, ptr undef) #5
   ret ptr %0
 }
 
@@ -192,7 +198,7 @@ entry:
   %stackalloc = alloca i8, align 1
   %0 = trunc i64 %len to i32
   %1 = getelementptr i8, ptr %p, i32 %0
-  call void @runtime.trackPointer(ptr %1, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %1, ptr nonnull %stackalloc, ptr undef) #5
   ret ptr %1
 }
 
@@ -206,7 +212,7 @@ slicetoarray.next:                                ; preds = %entry
   ret ptr %s.data
 
 slicetoarray.throw:                               ; preds = %entry
-  call void @runtime.sliceToArrayPointerPanic(ptr undef) #3
+  call void @runtime.sliceToArrayPointerPanic(ptr undef) #5
   unreachable
 }
 
@@ -216,8 +222,8 @@ declare void @runtime.sliceToArrayPointerPanic(ptr) #1
 define hidden ptr @main.SliceToArrayConst(ptr %context) unnamed_addr #2 {
 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) #3
+  %makeslice = call align 4 dereferenceable(24) ptr @runtime.alloc(i32 24, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
-  call void @runtime.trackPointer(ptr nonnull %makeslice, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice, ptr nonnull %stackalloc, ptr undef) #5
   br i1 false, label %slicetoarray.throw, label %slicetoarray.next
 
 slicetoarray.next:                                ; preds = %entry
@@ -242,11 +248,11 @@ unsafe.Slice.next:                                ; preds = %entry
   %5 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %6 = insertvalue { ptr, i32, i32 } %5, i32 %len, 1
   %7 = insertvalue { ptr, i32, i32 } %6, i32 %len, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %7
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
@@ -266,11 +272,11 @@ unsafe.Slice.next:                                ; preds = %entry
   %4 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %5 = insertvalue { ptr, i32, i32 } %4, i32 %3, 1
   %6 = insertvalue { ptr, i32, i32 } %5, i32 %3, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %6
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
@@ -290,11 +296,11 @@ unsafe.Slice.next:                                ; preds = %entry
   %6 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %7 = insertvalue { ptr, i32, i32 } %6, i32 %5, 1
   %8 = insertvalue { ptr, i32, i32 } %7, i32 %5, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %8
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
@@ -314,15 +320,17 @@ unsafe.Slice.next:                                ; preds = %entry
   %6 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %7 = insertvalue { ptr, i32, i32 } %6, i32 %5, 1
   %8 = insertvalue { ptr, i32, i32 } %7, i32 %5, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %8
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   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 #3 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+attributes #5 = { nounwind }

interp/interp_test.go

@@ -2,7 +2,6 @@ package interp
 
 import (
 	"os"
-	"strconv"
 	"strings"
 	"testing"
 	"time"
@@ -11,25 +10,17 @@ import (
 )
 
 func TestInterp(t *testing.T) {
-	llvmVersion, err := strconv.Atoi(strings.Split(llvm.Version, ".")[0])
-	if err != nil {
-		// Note: this should never happen and if it does, it will always happen
-		// for a particular build because llvm.Version is a constant.
-		panic(err)
-	}
 	for _, name := range []string{
 		"basic",
 		"phi",
-		"slice-copy",
 		"consteval",
+		"intrinsics",
+		"copy",
 		"interface",
 		"revert",
 		"alloc",
 	} {
 		name := name // make local to this closure
-		if name == "slice-copy" && llvmVersion < 14 {
-			continue
-		}
 		t.Run(name, func(t *testing.T) {
 			t.Parallel()
 			runTest(t, "testdata/"+name)

interp/interpreter.go

@@ -312,33 +312,28 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					fmt.Fprintln(os.Stderr, indent+"runtime.alloc:", size, "->", ptr)
 				}
 				locals[inst.localIndex] = ptr
-			case callFn.name == "runtime.sliceCopy":
+			case strings.HasPrefix(callFn.name, "llvm.umin."):
-				// sliceCopy implements the built-in copy function for slices.
+				locals[inst.localIndex] = makeLiteralInt(min(operands[1].Uint(r), operands[2].Uint(r)), inst.llvmInst.Type().IntTypeWidth())
-				// It is implemented here so that it can be used even if the
+			case strings.HasPrefix(callFn.name, "llvm.smin."):
-				// runtime implementation is not available. Doing it this way
+				locals[inst.localIndex] = makeLiteralInt(uint64(min(operands[1].Int(r), operands[2].Int(r))), inst.llvmInst.Type().IntTypeWidth())
-				// may also be faster.
+			case strings.HasPrefix(callFn.name, "llvm.umax."):
-				// Code:
+				locals[inst.localIndex] = makeLiteralInt(max(operands[1].Uint(r), operands[2].Uint(r)), inst.llvmInst.Type().IntTypeWidth())
-				// func sliceCopy(dst, src unsafe.Pointer, dstLen, srcLen uintptr, elemSize uintptr) int {
+			case strings.HasPrefix(callFn.name, "llvm.smax."):
-				//     n := srcLen
+				locals[inst.localIndex] = makeLiteralInt(uint64(max(operands[1].Int(r), operands[2].Int(r))), inst.llvmInst.Type().IntTypeWidth())
-				//     if n > dstLen {
+			case strings.HasPrefix(callFn.name, "llvm.memcpy.p0") || strings.HasPrefix(callFn.name, "llvm.memmove.p0"):
-				//         n = dstLen
+				// Copy a block of memory from one pointer to another.
-				//     }
+				if operands[4].Uint(r) != 0 {
-				//     memmove(dst, src, n*elemSize)
+					// This is a volatile copy/move.
-				//     return int(n)
+					err := r.runAtRuntime(fn, inst, locals, &mem, indent)
-				// }
+					if err != nil {
-				dstLen := operands[3].Uint(r)
+						return nil, mem, err
-				srcLen := operands[4].Uint(r)
+					}
-				elemSize := operands[5].Uint(r)
+					continue
-				n := srcLen
-				if n > dstLen {
-					n = dstLen
 				}
-				if r.debug {
+				nBytes := operands[3].Uint(r)
-					fmt.Fprintln(os.Stderr, indent+"copy:", operands[1], operands[2], n)
+				if nBytes != 0 {
-				}
-				if n != 0 {
 					// Only try to copy bytes when there are any bytes to copy.
-					// This is not just an optimization. If one of the slices
+					// This is not just an optimization. If one of the pointers
 					// (or both) are nil, the asPointer method call will fail
 					// even though copying a nil slice is allowed.
 					dst, err := operands[1].asPointer(r)
@@ -363,11 +358,10 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 						}
 						continue
 					}
-					nBytes := uint32(n * elemSize)
 					srcObj := mem.get(src.index())
 					dstObj := mem.getWritable(dst.index())
 					if srcObj.buffer == nil || dstObj.buffer == nil {
-						// If the buffer is nil, it means the slice is external.
+						// If the buffer is nil, it means the memory is external.
 						// This can happen for example when copying data out of
 						// a //go:embed slice, which is not available at interp
 						// time.
@@ -380,33 +374,10 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					}
 					dstBuf := dstObj.buffer.asRawValue(r)
 					srcBuf := srcObj.buffer.asRawValue(r)
-					copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
+					copy(dstBuf.buf[dst.offset():][:nBytes], srcBuf.buf[src.offset():][:nBytes])
 					dstObj.buffer = dstBuf
 					mem.put(dst.index(), dstObj)
 				}
-				locals[inst.localIndex] = makeLiteralInt(n, inst.llvmInst.Type().IntTypeWidth())
-			case strings.HasPrefix(callFn.name, "llvm.memcpy.p0") || strings.HasPrefix(callFn.name, "llvm.memmove.p0"):
-				// Copy a block of memory from one pointer to another.
-				dst, err := operands[1].asPointer(r)
-				if err != nil {
-					return nil, mem, r.errorAt(inst, err)
-				}
-				src, err := operands[2].asPointer(r)
-				if err != nil {
-					return nil, mem, r.errorAt(inst, err)
-				}
-				nBytes := uint32(operands[3].Uint(r))
-				dstObj := mem.getWritable(dst.index())
-				dstBuf := dstObj.buffer.asRawValue(r)
-				if mem.get(src.index()).buffer == nil {
-					// Looks like the source buffer is not defined.
-					// This can happen with //extern or //go:embed.
-					return nil, mem, r.errorAt(inst, errUnsupportedRuntimeInst)
-				}
-				srcBuf := mem.get(src.index()).buffer.asRawValue(r)
-				copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
-				dstObj.buffer = dstBuf
-				mem.put(dst.index(), dstObj)
 			case callFn.name == "runtime.typeAssert":
 				// This function must be implemented manually as it is normally
 				// implemented by the interface lowering pass.

interp/testdata/copy.ll

@@ -0,0 +1,68 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@string = internal unnamed_addr constant [3 x i8] c"foo"
+@moveDst = global [3 x i8] zeroinitializer
+@copyDst = global [3 x i8] zeroinitializer
+
+@externalSrc = external global [2 x i8]
+@moveExternalDst = global [2 x i8] zeroinitializer
+
+@moveEscapedSrc = global [4 x i8] c"abcd"
+@moveEscapedDst = global [4 x i8] zeroinitializer
+
+@volatileSrc = global [2 x i8] c"xy"
+@volatileDst = global [2 x i8] zeroinitializer
+
+declare void @use(ptr)
+
+define void @runtime.initAll() {
+  call void @main.init()
+  ret void
+}
+
+define internal void @main.init() {
+  call void @testMove()
+  call void @testCopy()
+  call void @testMoveExternal()
+  call void @testMoveEscaped()
+  call void @testVolatileCopy()
+  ret void
+}
+
+; Test a simple memmove between globals.
+define internal void @testMove() {
+  call void @llvm.memmove.p0.p0.i64(ptr @moveDst, ptr @string, i64 3, i1 false)
+  ret void
+}
+
+; Test a simple memcpy between globals.
+define internal void @testCopy() {
+  call void @llvm.memcpy.p0.p0.i64(ptr @copyDst, ptr @string, i64 3, i1 false)
+  ret void
+}
+
+; Test a memmove from an external global.
+; This should be run at runtime.
+define internal void @testMoveExternal() {
+  call void @llvm.memmove.p0.p0.i64(ptr @moveExternalDst, ptr @externalSrc, i64 2, i1 false)
+  ret void
+}
+
+; Test a memmove from an escaped (and potentially modified) source buffer.
+define internal void @testMoveEscaped() {
+  call void @use(ptr @moveEscapedSrc)
+  call void @llvm.memmove.p0.p0.i64(ptr @moveEscapedDst, ptr @moveEscapedSrc, i64 4, i1 false)
+  ret void
+}
+
+; Test a volatile memcpy.
+; This should always be run at runtime.
+define internal void @testVolatileCopy() {
+  call void @llvm.memcpy.p0.p0.i64(ptr @volatileDst, ptr @volatileSrc, i64 2, i1 true)
+  ret void
+}
+
+declare void @llvm.memmove.p0.p0.i64(ptr, ptr, i64, i1)
+
+declare void @llvm.memcpy.p0.p0.i64(ptr, ptr, i64, i1)

interp/testdata/copy.out.ll

@@ -0,0 +1,27 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@moveDst = local_unnamed_addr global [3 x i8] c"foo"
+@copyDst = local_unnamed_addr global [3 x i8] c"foo"
+@externalSrc = external local_unnamed_addr global [2 x i8]
+@moveExternalDst = local_unnamed_addr global [2 x i8] zeroinitializer
+@moveEscapedSrc = global [4 x i8] c"abcd"
+@moveEscapedDst = local_unnamed_addr global [4 x i8] zeroinitializer
+@volatileSrc = global [2 x i8] c"xy"
+@volatileDst = global [2 x i8] zeroinitializer
+
+declare void @use(ptr) local_unnamed_addr
+
+define void @runtime.initAll() local_unnamed_addr {
+  call void @llvm.memmove.p0.p0.i64(ptr @moveExternalDst, ptr @externalSrc, i64 2, i1 false)
+  call void @use(ptr @moveEscapedSrc)
+  call void @llvm.memmove.p0.p0.i64(ptr @moveEscapedDst, ptr @moveEscapedSrc, i64 4, i1 false)
+  call void @llvm.memcpy.p0.p0.i64(ptr @volatileDst, ptr @volatileSrc, i64 2, i1 true)
+  ret void
+}
+
+declare void @llvm.memmove.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1 immarg) #0
+
+declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg) #0
+
+attributes #0 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }

interp/testdata/intrinsics.ll

@@ -0,0 +1,52 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@uminResult = global i32 0
+@sminResult = global i32 0
+@umaxResult = global i32 0
+@smaxResult = global i32 0
+
+define void @runtime.initAll() {
+  call void @main.init()
+  ret void
+}
+
+define internal void @main.init() {
+  call void @testUMin()
+  call void @testSMin()
+  call void @testUMax()
+  call void @testSMax()
+  ret void
+}
+
+define internal void @testUMin() {
+  %umin = call i32 @llvm.umin.i32(i32 12, i32 -1)
+  store i32 %umin, ptr @uminResult
+  ret void
+}
+
+declare i32 @llvm.umin.i32(i32, i32)
+
+define internal void @testSMin() {
+  %smin = call i32 @llvm.smin.i32(i32 12, i32 -1)
+  store i32 %smin, ptr @sminResult
+  ret void
+}
+
+declare i32 @llvm.smin.i32(i32, i32)
+
+define internal void @testUMax() {
+  %umax = call i32 @llvm.umax.i32(i32 12, i32 -1)
+  store i32 %umax, ptr @umaxResult
+  ret void
+}
+
+declare i32 @llvm.umax.i32(i32, i32)
+
+define internal void @testSMax() {
+  %smax = call i32 @llvm.smax.i32(i32 12, i32 -1)
+  store i32 %smax, ptr @smaxResult
+  ret void
+}
+
+declare i32 @llvm.smax.i32(i32, i32)

interp/testdata/intrinsics.out.ll

@@ -0,0 +1,11 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@uminResult = local_unnamed_addr global i32 12
+@sminResult = local_unnamed_addr global i32 -1
+@umaxResult = local_unnamed_addr global i32 -1
+@smaxResult = local_unnamed_addr global i32 12
+
+define void @runtime.initAll() local_unnamed_addr {
+  ret void
+}

interp/testdata/slice-copy.ll

@@ -1,124 +0,0 @@
-target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64--linux"
-
-@main.uint8SliceSrc.buf = internal global [2 x i8] c"\03d"
-@main.uint8SliceSrc = internal unnamed_addr global { ptr, i64, i64 } { ptr @main.uint8SliceSrc.buf, i64 2, i64 2 }
-@main.uint8SliceDst = internal unnamed_addr global { ptr, i64, i64 } zeroinitializer
-@main.int16SliceSrc.buf = internal global [3 x i16] [i16 5, i16 123, i16 1024]
-@main.int16SliceSrc = internal unnamed_addr global { ptr, i64, i64 } { ptr @main.int16SliceSrc.buf, i64 3, i64 3 }
-@main.int16SliceDst = internal unnamed_addr global { ptr, i64, i64 } zeroinitializer
-@main.sliceSrcUntaint.buf = internal global [2 x i8] c"ab"
-@main.sliceDstUntaint.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcTaint.buf = internal global [2 x i8] c"cd"
-@main.sliceDstTaint.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal1.buf = external global [2 x i8]
-@main.sliceDstExternal1.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal2.buf = internal global [2 x i8] zeroinitializer
-@main.sliceDstExternal2.buf = external global [2 x i8]
-
-declare i64 @runtime.sliceCopy(ptr %dst, ptr %src, i64 %dstLen, i64 %srcLen, i64 %elemSize) unnamed_addr
-
-declare ptr @runtime.alloc(i64, ptr) unnamed_addr
-
-declare void @runtime.printuint8(i8)
-
-declare void @runtime.printint16(i16)
-
-declare void @use(ptr)
-
-define void @runtime.initAll() unnamed_addr {
-entry:
-  call void @main.init()
-  ret void
-}
-
-define void @main() unnamed_addr {
-entry:
-  ; print(uintSliceSrc[0])
-  %uint8SliceSrc.buf = load ptr, ptr @main.uint8SliceSrc
-  %uint8SliceSrc.val = load i8, ptr %uint8SliceSrc.buf
-  call void @runtime.printuint8(i8 %uint8SliceSrc.val)
-
-  ; print(uintSliceDst[0])
-  %uint8SliceDst.buf = load ptr, ptr @main.uint8SliceDst
-  %uint8SliceDst.val = load i8, ptr %uint8SliceDst.buf
-  call void @runtime.printuint8(i8 %uint8SliceDst.val)
-
-  ; print(int16SliceSrc[0])
-  %int16SliceSrc.buf = load ptr, ptr @main.int16SliceSrc
-  %int16SliceSrc.val = load i16, ptr %int16SliceSrc.buf
-  call void @runtime.printint16(i16 %int16SliceSrc.val)
-
-  ; print(int16SliceDst[0])
-  %int16SliceDst.buf = load ptr, ptr @main.int16SliceDst
-  %int16SliceDst.val = load i16, ptr %int16SliceDst.buf
-  call void @runtime.printint16(i16 %int16SliceDst.val)
-
-  ; print(sliceDstUntaint[0])
-  %sliceDstUntaint.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstUntaint.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstUntaint.val)
-
-  ; print(sliceDstTaint[0])
-  %sliceDstTaint.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstTaint.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstTaint.val)
-
-  ; print(sliceDstExternal1[0])
-  %sliceDstExternal1.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstExternal1.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstExternal1.val)
-
-  ; print(sliceDstExternal2[0])
-  %sliceDstExternal2.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstExternal2.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstExternal2.val)
-
-  ret void
-}
-
-define internal void @main.init() unnamed_addr {
-entry:
-  ; equivalent of:
-  ;     uint8SliceDst = make([]uint8, len(uint8SliceSrc))
-  %uint8SliceSrc = load { ptr, i64, i64 }, ptr @main.uint8SliceSrc
-  %uint8SliceSrc.len = extractvalue { ptr, i64, i64 } %uint8SliceSrc, 1
-  %uint8SliceDst.buf = call ptr @runtime.alloc(i64 %uint8SliceSrc.len, ptr null)
-  %0 = insertvalue { ptr, i64, i64 } undef, ptr %uint8SliceDst.buf, 0
-  %1 = insertvalue { ptr, i64, i64 } %0, i64 %uint8SliceSrc.len, 1
-  %2 = insertvalue { ptr, i64, i64 } %1, i64 %uint8SliceSrc.len, 2
-  store { ptr, i64, i64 } %2, ptr @main.uint8SliceDst
-
-  ; equivalent of:
-  ;     copy(uint8SliceDst, uint8SliceSrc)
-  %uint8SliceSrc.buf = extractvalue { ptr, i64, i64 } %uint8SliceSrc, 0
-  %copy.n = call i64 @runtime.sliceCopy(ptr %uint8SliceDst.buf, ptr %uint8SliceSrc.buf, i64 %uint8SliceSrc.len, i64 %uint8SliceSrc.len, i64 1)
-
-  ; equivalent of:
-  ;     int16SliceDst = make([]int16, len(int16SliceSrc))
-  %int16SliceSrc = load { ptr, i64, i64 }, ptr @main.int16SliceSrc
-  %int16SliceSrc.len = extractvalue { ptr, i64, i64 } %int16SliceSrc, 1
-  %int16SliceSrc.len.bytes = mul i64 %int16SliceSrc.len, 2
-  %int16SliceDst.buf = call ptr @runtime.alloc(i64 %int16SliceSrc.len.bytes, ptr null)
-  %3 = insertvalue { ptr, i64, i64 } undef, ptr %int16SliceDst.buf, 0
-  %4 = insertvalue { ptr, i64, i64 } %3, i64 %int16SliceSrc.len, 1
-  %5 = insertvalue { ptr, i64, i64 } %4, i64 %int16SliceSrc.len, 2
-  store { ptr, i64, i64 } %5, ptr @main.int16SliceDst
-
-  ; equivalent of:
-  ;     copy(int16SliceDst, int16SliceSrc)
-  %int16SliceSrc.buf = extractvalue { ptr, i64, i64 } %int16SliceSrc, 0
-  %copy.n2 = call i64 @runtime.sliceCopy(ptr %int16SliceDst.buf, ptr %int16SliceSrc.buf, i64 %int16SliceSrc.len, i64 %int16SliceSrc.len, i64 2)
-
-  ; Copy slice that has a known value.
-  %copy.n3 = call i64 @runtime.sliceCopy(ptr @main.sliceDstUntaint.buf, ptr @main.sliceSrcUntaint.buf, i64 2, i64 2, i64 1)
-
-  ; Copy slice that might have been modified by the external @use call.
-  ; This is a fix for https://github.com/tinygo-org/tinygo/issues/3890.
-  call void @use(ptr @main.sliceSrcTaint.buf)
-  %copy.n4 = call i64 @runtime.sliceCopy(ptr @main.sliceDstTaint.buf, ptr @main.sliceSrcTaint.buf, i64 2, i64 2, i64 1)
-
-  ; Test that copying from or into external buffers works correctly.
-  ; These copy operations must be done at runtime.
-  ; https://github.com/tinygo-org/tinygo/issues/4895
-  %copy.n5 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal1.buf, ptr @main.sliceSrcExternal1.buf, i64 2, i64 2, i64 1)
-  %copy.n6 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal2.buf, ptr @main.sliceSrcExternal2.buf, i64 2, i64 2, i64 1)
-
-  ret void
-}

interp/testdata/slice-copy.out.ll

@@ -1,42 +0,0 @@
-target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64--linux"
-
-@main.sliceSrcTaint.buf = internal global [2 x i8] c"cd"
-@main.sliceDstTaint.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal1.buf = external global [2 x i8]
-@main.sliceDstExternal1.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal2.buf = internal global [2 x i8] zeroinitializer
-@main.sliceDstExternal2.buf = external global [2 x i8]
-
-declare i64 @runtime.sliceCopy(ptr, ptr, i64, i64, i64) unnamed_addr
-
-declare void @runtime.printuint8(i8) local_unnamed_addr
-
-declare void @runtime.printint16(i16) local_unnamed_addr
-
-declare void @use(ptr) local_unnamed_addr
-
-define void @runtime.initAll() unnamed_addr {
-entry:
-  call void @use(ptr @main.sliceSrcTaint.buf)
-  %copy.n4 = call i64 @runtime.sliceCopy(ptr @main.sliceDstTaint.buf, ptr @main.sliceSrcTaint.buf, i64 2, i64 2, i64 1)
-  %copy.n5 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal1.buf, ptr @main.sliceSrcExternal1.buf, i64 2, i64 2, i64 1)
-  %copy.n6 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal2.buf, ptr @main.sliceSrcExternal2.buf, i64 2, i64 2, i64 1)
-  ret void
-}
-
-define void @main() unnamed_addr {
-entry:
-  call void @runtime.printuint8(i8 3)
-  call void @runtime.printuint8(i8 3)
-  call void @runtime.printint16(i16 5)
-  call void @runtime.printint16(i16 5)
-  call void @runtime.printuint8(i8 97)
-  %sliceDstTaint.val = load i8, ptr @main.sliceDstTaint.buf, align 1
-  call void @runtime.printuint8(i8 %sliceDstTaint.val)
-  %sliceDstExternal1.val = load i8, ptr @main.sliceDstExternal1.buf, align 1
-  call void @runtime.printuint8(i8 %sliceDstExternal1.val)
-  %sliceDstExternal2.val = load i8, ptr @main.sliceDstExternal2.buf, align 1
-  call void @runtime.printuint8(i8 %sliceDstExternal2.val)
-  ret void
-}

src/internal/reflectlite/value.go

@@ -1738,6 +1738,9 @@ func (e *ValueError) Error() string {
 //go:linkname memcpy runtime.memcpy
 func memcpy(dst, src unsafe.Pointer, size uintptr)
 
+//go:linkname memmove runtime.memmove
+func memmove(dst, src unsafe.Pointer, size uintptr)
+
 //go:linkname memzero runtime.memzero
 func memzero(ptr unsafe.Pointer, size uintptr)
 
@@ -1747,9 +1750,6 @@ func alloc(size uintptr, layout unsafe.Pointer) unsafe.Pointer
 //go:linkname sliceAppend runtime.sliceAppend
 func sliceAppend(srcBuf, elemsBuf unsafe.Pointer, srcLen, srcCap, elemsLen uintptr, elemSize uintptr, layout unsafe.Pointer) (unsafe.Pointer, uintptr, uintptr)
 
-//go:linkname sliceCopy runtime.sliceCopy
-func sliceCopy(dst, src unsafe.Pointer, dstLen, srcLen uintptr, elemSize uintptr) int
-
 // Copy copies the contents of src into dst until either
 // dst has been filled or src has been exhausted.
 func Copy(dst, src Value) int {
@@ -1779,7 +1779,10 @@ func Copy(dst, src Value) int {
 		dst.checkRO()
 	}
 
-	return sliceCopy(dstbuf, srcbuf, dstlen, srclen, dst.typecode.elem().Size())
+	minLen := min(dstlen, srclen)
+	elemSize := dst.typecode.elem().Size()
+	memmove(dstbuf, srcbuf, minLen*elemSize)
+	return int(minLen)
 }
 
 func buflen(v Value) (unsafe.Pointer, uintptr) {

src/runtime/slice.go

@@ -23,17 +23,6 @@ func sliceAppend(srcBuf, elemsBuf unsafe.Pointer, srcLen, srcCap, elemsLen, elem
 	return srcBuf, newLen, srcCap
 }
 
-// Builtin copy(dst, src) function: copy bytes from dst to src.
-func sliceCopy(dst, src unsafe.Pointer, dstLen, srcLen uintptr, elemSize uintptr) int {
-	// n = min(srcLen, dstLen)
-	n := srcLen
-	if n > dstLen {
-		n = dstLen
-	}
-	memmove(dst, src, n*elemSize)
-	return int(n)
-}
-
 // sliceGrow returns a new slice with space for at least newCap elements
 func sliceGrow(oldBuf unsafe.Pointer, oldLen, oldCap, newCap, elemSize uintptr, layout unsafe.Pointer) (unsafe.Pointer, uintptr, uintptr) {
 	if oldCap >= newCap {