1; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,W64 %s
2; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,W32 %s
3; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,W64 %s
4; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,W32 %s
5; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,W64 %s
6
7; RUN: opt -O3 -S < %s | FileCheck -check-prefixes=OPT,OPT-WXX %s
8; RUN: opt -mtriple=amdgcn-- -O3 -S < %s | FileCheck -check-prefixes=OPT,OPT-WXX %s
9; RUN: opt -mtriple=amdgcn-- -O3 -mattr=+wavefrontsize32 -S < %s | FileCheck -check-prefixes=OPT,OPT-W32 %s
10; RUN: opt -mtriple=amdgcn-- -passes='default<O3>' -mattr=+wavefrontsize32 -S < %s | FileCheck -check-prefixes=OPT,OPT-W32 %s
11; RUN: opt -mtriple=amdgcn-- -O3 -mattr=+wavefrontsize64 -S < %s | FileCheck -check-prefixes=OPT,OPT-W64 %s
12; RUN: opt -mtriple=amdgcn-- -mcpu=tonga -O3 -S < %s | FileCheck -check-prefixes=OPT,OPT-W64 %s
13; RUN: opt -mtriple=amdgcn-- -mcpu=gfx1010 -O3 -mattr=+wavefrontsize32,-wavefrontsize64 -S < %s | FileCheck -check-prefixes=OPT,OPT-W32 %s
14; RUN: opt -mtriple=amdgcn-- -mcpu=gfx1010 -O3 -mattr=-wavefrontsize32,+wavefrontsize64 -S < %s | FileCheck -check-prefixes=OPT,OPT-W64 %s
15; RUN: opt -mtriple=amdgcn-- -mcpu=gfx1100 -O3 -mattr=+wavefrontsize32,-wavefrontsize64 -S < %s | FileCheck -check-prefixes=OPT,OPT-W32 %s
16; RUN: opt -mtriple=amdgcn-- -mcpu=gfx1100 -O3 -mattr=-wavefrontsize32,+wavefrontsize64 -S < %s | FileCheck -check-prefixes=OPT,OPT-W64 %s
17
18; GCN-LABEL: {{^}}fold_wavefrontsize:
19; OPT-LABEL: define amdgpu_kernel void @fold_wavefrontsize(
20
21; W32:       v_mov_b32_e32 [[V:v[0-9]+]], 32
22; W64:       v_mov_b32_e32 [[V:v[0-9]+]], 64
23; GCN:       store_{{dword|b32}} v{{.+}}, [[V]]
24
25; OPT-W32:   store i32 32, i32 addrspace(1)* %arg, align 4
26; OPT-W64:   store i32 64, i32 addrspace(1)* %arg, align 4
27; OPT-WXX:   %tmp = tail call i32 @llvm.amdgcn.wavefrontsize()
28; OPT-WXX:   store i32 %tmp, i32 addrspace(1)* %arg, align 4
29; OPT-NEXT:  ret void
30
31define amdgpu_kernel void @fold_wavefrontsize(i32 addrspace(1)* nocapture %arg) {
32bb:
33  %tmp = tail call i32 @llvm.amdgcn.wavefrontsize() #0
34  store i32 %tmp, i32 addrspace(1)* %arg, align 4
35  ret void
36}
37
38; GCN-LABEL: {{^}}fold_and_optimize_wavefrontsize:
39; OPT-LABEL: define amdgpu_kernel void @fold_and_optimize_wavefrontsize(
40
41; W32:       v_mov_b32_e32 [[V:v[0-9]+]], 1{{$}}
42; W64:       v_mov_b32_e32 [[V:v[0-9]+]], 2{{$}}
43; GCN-NOT:   cndmask
44; GCN:       store_{{dword|b32}} v{{.+}}, [[V]]
45
46; OPT-W32:   store i32 1, i32 addrspace(1)* %arg, align 4
47; OPT-W64:   store i32 2, i32 addrspace(1)* %arg, align 4
48; OPT-WXX:   %tmp = tail call i32 @llvm.amdgcn.wavefrontsize()
49; OPT-WXX:   %tmp1 = icmp ugt i32 %tmp, 32
50; OPT-WXX:   %tmp2 = select i1 %tmp1, i32 2, i32 1
51; OPT-WXX:   store i32 %tmp2, i32 addrspace(1)* %arg
52; OPT-NEXT:  ret void
53
54define amdgpu_kernel void @fold_and_optimize_wavefrontsize(i32 addrspace(1)* nocapture %arg) {
55bb:
56  %tmp = tail call i32 @llvm.amdgcn.wavefrontsize() #0
57  %tmp1 = icmp ugt i32 %tmp, 32
58  %tmp2 = select i1 %tmp1, i32 2, i32 1
59  store i32 %tmp2, i32 addrspace(1)* %arg
60  ret void
61}
62
63; GCN-LABEL: {{^}}fold_and_optimize_if_wavefrontsize:
64; OPT-LABEL: define amdgpu_kernel void @fold_and_optimize_if_wavefrontsize(
65
66; OPT:       bb:
67; OPT-WXX:   %tmp = tail call i32 @llvm.amdgcn.wavefrontsize()
68; OPT-WXX:   %tmp1 = icmp ugt i32 %tmp, 32
69; OPT-WXX:   bb3:
70; OPT-W64:   store i32 1, i32 addrspace(1)* %arg, align 4
71; OPT-NEXT:  ret void
72
73define amdgpu_kernel void @fold_and_optimize_if_wavefrontsize(i32 addrspace(1)* nocapture %arg) {
74bb:
75  %tmp = tail call i32 @llvm.amdgcn.wavefrontsize() #0
76  %tmp1 = icmp ugt i32 %tmp, 32
77  br i1 %tmp1, label %bb2, label %bb3
78
79bb2:                                              ; preds = %bb
80  store i32 1, i32 addrspace(1)* %arg, align 4
81  br label %bb3
82
83bb3:                                              ; preds = %bb2, %bb
84  ret void
85}
86
87declare i32 @llvm.amdgcn.wavefrontsize() #0
88
89attributes #0 = { nounwind readnone speculatable }
90