150
|
1 ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-promote-alloca < %s | FileCheck %s
|
|
2
|
|
3 ; Make sure this allocates the correct size if the alloca has a non-0
|
|
4 ; number of elements.
|
|
5
|
|
6 ; CHECK-LABEL: @array_alloca(
|
|
7 ; CHECK: %stack = alloca i32, i32 5, align 4
|
|
8 define amdgpu_kernel void @array_alloca(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #0 {
|
|
9 entry:
|
|
10 %stack = alloca i32, i32 5, align 4
|
|
11 %ld0 = load i32, i32 addrspace(1)* %in, align 4
|
|
12 %arrayidx1 = getelementptr inbounds i32, i32* %stack, i32 %ld0
|
|
13 store i32 4, i32* %arrayidx1, align 4
|
|
14 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1
|
|
15 %ld1 = load i32, i32 addrspace(1)* %arrayidx2, align 4
|
|
16 %arrayidx3 = getelementptr inbounds i32, i32* %stack, i32 %ld1
|
|
17 store i32 5, i32* %arrayidx3, align 4
|
|
18 %arrayidx10 = getelementptr inbounds i32, i32* %stack, i32 0
|
|
19 %ld2 = load i32, i32* %arrayidx10, align 4
|
|
20 store i32 %ld2, i32 addrspace(1)* %out, align 4
|
|
21 %arrayidx12 = getelementptr inbounds i32, i32* %stack, i32 1
|
|
22 %ld3 = load i32, i32* %arrayidx12
|
|
23 %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1
|
|
24 store i32 %ld3, i32 addrspace(1)* %arrayidx13
|
|
25 ret void
|
|
26 }
|
|
27
|
|
28 ; CHECK-LABEL: @array_alloca_dynamic(
|
|
29 ; CHECK: %stack = alloca i32, i32 %size, align 4
|
|
30 define amdgpu_kernel void @array_alloca_dynamic(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in, i32 %size) #0 {
|
|
31 entry:
|
|
32 %stack = alloca i32, i32 %size, align 4
|
|
33 %ld0 = load i32, i32 addrspace(1)* %in, align 4
|
|
34 %arrayidx1 = getelementptr inbounds i32, i32* %stack, i32 %ld0
|
|
35 store i32 4, i32* %arrayidx1, align 4
|
|
36 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1
|
|
37 %ld1 = load i32, i32 addrspace(1)* %arrayidx2, align 4
|
|
38 %arrayidx3 = getelementptr inbounds i32, i32* %stack, i32 %ld1
|
|
39 store i32 5, i32* %arrayidx3, align 4
|
|
40 %arrayidx10 = getelementptr inbounds i32, i32* %stack, i32 0
|
|
41 %ld2 = load i32, i32* %arrayidx10, align 4
|
|
42 store i32 %ld2, i32 addrspace(1)* %out, align 4
|
|
43 %arrayidx12 = getelementptr inbounds i32, i32* %stack, i32 1
|
|
44 %ld3 = load i32, i32* %arrayidx12
|
|
45 %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1
|
|
46 store i32 %ld3, i32 addrspace(1)* %arrayidx13
|
|
47 ret void
|
|
48 }
|
|
49
|
|
50 attributes #0 = { nounwind }
|