• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck %s
2
3declare void @llvm.memcpy.p0i8.p1i8.i32(i8* nocapture, i8 addrspace(1)* nocapture, i32, i1) #0
4declare void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* nocapture, i8* nocapture, i32, i1) #0
5
6declare void @llvm.memmove.p0i8.p1i8.i32(i8* nocapture, i8 addrspace(1)* nocapture, i32, i1) #0
7declare void @llvm.memmove.p1i8.p0i8.i32(i8 addrspace(1)* nocapture, i8* nocapture, i32, i1) #0
8
9declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i1) #0
10
11declare i32 @llvm.objectsize.i32.p0i8(i8*, i1, i1, i1) #1
12
13; CHECK-LABEL: @promote_with_memcpy(
14; CHECK: getelementptr inbounds [64 x [17 x i32]], [64 x [17 x i32]] addrspace(3)* @promote_with_memcpy.alloca, i32 0, i32 %{{[0-9]+}}
15; CHECK: call void @llvm.memcpy.p3i8.p1i8.i32(i8 addrspace(3)* align 4 %alloca.bc, i8 addrspace(1)* align 4 %in.bc, i32 68, i1 false)
16; CHECK: call void @llvm.memcpy.p1i8.p3i8.i32(i8 addrspace(1)* align 4 %out.bc, i8 addrspace(3)* align 4 %alloca.bc, i32 68, i1 false)
17define amdgpu_kernel void @promote_with_memcpy(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
18  %alloca = alloca [17 x i32], align 4
19  %alloca.bc = bitcast [17 x i32]* %alloca to i8*
20  %in.bc = bitcast i32 addrspace(1)* %in to i8 addrspace(1)*
21  %out.bc = bitcast i32 addrspace(1)* %out to i8 addrspace(1)*
22  call void @llvm.memcpy.p0i8.p1i8.i32(i8* align 4 %alloca.bc, i8 addrspace(1)* align 4 %in.bc, i32 68, i1 false)
23  call void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* align 4 %out.bc, i8* align 4 %alloca.bc, i32 68, i1 false)
24  ret void
25}
26
27; CHECK-LABEL: @promote_with_memmove(
28; CHECK: getelementptr inbounds [64 x [17 x i32]], [64 x [17 x i32]] addrspace(3)* @promote_with_memmove.alloca, i32 0, i32 %{{[0-9]+}}
29; CHECK: call void @llvm.memmove.p3i8.p1i8.i32(i8 addrspace(3)* align 4 %alloca.bc, i8 addrspace(1)* align 4 %in.bc, i32 68, i1 false)
30; CHECK: call void @llvm.memmove.p1i8.p3i8.i32(i8 addrspace(1)* align 4 %out.bc, i8 addrspace(3)* align 4 %alloca.bc, i32 68, i1 false)
31define amdgpu_kernel void @promote_with_memmove(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
32  %alloca = alloca [17 x i32], align 4
33  %alloca.bc = bitcast [17 x i32]* %alloca to i8*
34  %in.bc = bitcast i32 addrspace(1)* %in to i8 addrspace(1)*
35  %out.bc = bitcast i32 addrspace(1)* %out to i8 addrspace(1)*
36  call void @llvm.memmove.p0i8.p1i8.i32(i8* align 4 %alloca.bc, i8 addrspace(1)* align 4 %in.bc, i32 68, i1 false)
37  call void @llvm.memmove.p1i8.p0i8.i32(i8 addrspace(1)* align 4 %out.bc, i8* align 4 %alloca.bc, i32 68, i1 false)
38  ret void
39}
40
41; CHECK-LABEL: @promote_with_memset(
42; CHECK: getelementptr inbounds [64 x [17 x i32]], [64 x [17 x i32]] addrspace(3)* @promote_with_memset.alloca, i32 0, i32 %{{[0-9]+}}
43; CHECK: call void @llvm.memset.p3i8.i32(i8 addrspace(3)* align 4 %alloca.bc, i8 7, i32 68, i1 false)
44define amdgpu_kernel void @promote_with_memset(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
45  %alloca = alloca [17 x i32], align 4
46  %alloca.bc = bitcast [17 x i32]* %alloca to i8*
47  %in.bc = bitcast i32 addrspace(1)* %in to i8 addrspace(1)*
48  %out.bc = bitcast i32 addrspace(1)* %out to i8 addrspace(1)*
49  call void @llvm.memset.p0i8.i32(i8* align 4 %alloca.bc, i8 7, i32 68, i1 false)
50  ret void
51}
52
53; CHECK-LABEL: @promote_with_objectsize(
54; CHECK: [[PTR:%[0-9]+]] = getelementptr inbounds [64 x [17 x i32]], [64 x [17 x i32]] addrspace(3)* @promote_with_objectsize.alloca, i32 0, i32 %{{[0-9]+}}
55; CHECK: call i32 @llvm.objectsize.i32.p3i8(i8 addrspace(3)* %alloca.bc, i1 false, i1 false, i1 false)
56define amdgpu_kernel void @promote_with_objectsize(i32 addrspace(1)* %out) #0 {
57  %alloca = alloca [17 x i32], align 4
58  %alloca.bc = bitcast [17 x i32]* %alloca to i8*
59  %size = call i32 @llvm.objectsize.i32.p0i8(i8* %alloca.bc, i1 false, i1 false, i1 false)
60  store i32 %size, i32 addrspace(1)* %out
61  ret void
62}
63
64attributes #0 = { nounwind "amdgpu-flat-work-group-size"="64,64" "amdgpu-waves-per-eu"="1,3" }
65attributes #1 = { nounwind readnone }
66