compiler: add 'align' attribute to runtime.alloc calls

This adds something like 'align 4' to the runtime.alloc calls, so that
the compiler knows the alignment of the allocation and can optimize
accordingly.

The optimization is very small, only 0.01% in my small test. However, my
plan is to read the value in the heap-to-stack transform pass to make it
more correct and let it produce slightly more optimal code.
This commit is contained in:
Ayke van Laethem
2024-06-27 20:25:33 +02:00
committed by Ron Evans
parent 9cb263479c
commit 571447c7c1
6 changed files with 36 additions and 30 deletions
+16 -16
View File
@@ -39,16 +39,16 @@ entry:
define hidden void @main.newScalar(ptr %context) unnamed_addr #2 {
entry:
%stackalloc = alloca i8, align 1
%new = call dereferenceable(1) ptr @runtime.alloc(i32 1, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new = call align 1 dereferenceable(1) ptr @runtime.alloc(i32 1, 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.scalar1, align 4
%new1 = call dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new1 = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new1, ptr @main.scalar2, align 4
%new2 = call dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new2 = call align 8 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new2, ptr @main.scalar3, align 4
%new3 = call dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new3 = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new3, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new3, ptr @main.scalar4, align 4
ret void
@@ -58,13 +58,13 @@ entry:
define hidden void @main.newArray(ptr %context) unnamed_addr #2 {
entry:
%stackalloc = alloca i8, align 1
%new = call dereferenceable(3) ptr @runtime.alloc(i32 3, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new = call align 1 dereferenceable(3) ptr @runtime.alloc(i32 3, 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.array1, align 4
%new1 = call dereferenceable(71) ptr @runtime.alloc(i32 71, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new1 = call align 1 dereferenceable(71) ptr @runtime.alloc(i32 71, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new1, ptr @main.array2, align 4
%new2 = call dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
%new2 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new2, ptr @main.array3, align 4
ret void
@@ -74,16 +74,16 @@ entry:
define hidden void @main.newStruct(ptr %context) unnamed_addr #2 {
entry:
%stackalloc = alloca i8, align 1
%new = call ptr @runtime.alloc(i32 0, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new = call align 1 ptr @runtime.alloc(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 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%new1 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new1, ptr @main.struct2, align 4
%new2 = call dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-2000000000000001", ptr undef) #3
%new2 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-2000000000000001", ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new2, ptr @main.struct3, align 4
%new3 = call dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0001", ptr undef) #3
%new3 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0001", ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new3, ptr nonnull %stackalloc, ptr undef) #3
store ptr %new3, ptr @main.struct4, align 4
ret void
@@ -93,7 +93,7 @@ entry:
define hidden ptr @main.newFuncValue(ptr %context) unnamed_addr #2 {
entry:
%stackalloc = alloca i8, align 1
%new = call dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 197 to ptr), ptr undef) #3
%new = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 197 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #3
ret ptr %new
}
@@ -102,17 +102,17 @@ entry:
define hidden void @main.makeSlice(ptr %context) unnamed_addr #2 {
entry:
%stackalloc = alloca i8, align 1
%makeslice = call dereferenceable(5) ptr @runtime.alloc(i32 5, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
%makeslice = call align 1 dereferenceable(5) ptr @runtime.alloc(i32 5, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %makeslice, ptr nonnull %stackalloc, ptr undef) #3
store ptr %makeslice, ptr @main.slice1, align 4
store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice1, i32 0, i32 1), align 4
store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice1, i32 0, i32 2), align 4
%makeslice1 = call dereferenceable(20) ptr @runtime.alloc(i32 20, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
%makeslice1 = call align 4 dereferenceable(20) ptr @runtime.alloc(i32 20, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %makeslice1, ptr nonnull %stackalloc, ptr undef) #3
store ptr %makeslice1, ptr @main.slice2, align 4
store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice2, i32 0, i32 1), align 4
store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice2, i32 0, i32 2), align 4
%makeslice3 = call dereferenceable(60) ptr @runtime.alloc(i32 60, ptr nonnull inttoptr (i32 71 to ptr), ptr undef) #3
%makeslice3 = call align 4 dereferenceable(60) ptr @runtime.alloc(i32 60, ptr nonnull inttoptr (i32 71 to ptr), ptr undef) #3
call void @runtime.trackPointer(ptr nonnull %makeslice3, ptr nonnull %stackalloc, ptr undef) #3
store ptr %makeslice3, ptr @main.slice3, align 4
store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice3, i32 0, i32 1), align 4
@@ -124,7 +124,7 @@ entry:
define hidden %runtime._interface @main.makeInterface(double %v.r, double %v.i, ptr %context) unnamed_addr #2 {
entry:
%stackalloc = alloca i8, align 1
%0 = call dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #3
%0 = call align 8 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, 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 { double, double }, ptr %0, i32 0, i32 1