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