1; RUN: llc -mtriple=amdgcn-- -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX7LESS %s
2; RUN: llc  -mtriple=amdgcn-- -mcpu=tonga -mattr=-flat-for-global -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX8MORE %s
3; RUN: llc -mtriple=amdgcn-- -mcpu=gfx900 -mattr=-flat-for-global -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX8MORE %s
4
5declare i1 @llvm.amdgcn.wqm.vote(i1)
6declare i32 @llvm.amdgcn.buffer.atomic.add(i32, <4 x i32>, i32, i32, i1)
7declare void @llvm.amdgcn.buffer.store.f32(float, <4 x i32>, i32, i32, i1, i1)
8
9; Show that what the atomic optimization pass will do for raw buffers.
10
11; GCN-LABEL: add_i32_constant:
12; GCN-LABEL: BB0_1:
13; GCN: v_cmp_ne_u32_e64 s{{\[}}[[exec_lo:[0-9]+]]:[[exec_hi:[0-9]+]]{{\]}}, 1, 0
14; GCN: v_mbcnt_lo_u32_b32{{(_e[0-9]+)?}} v[[mbcnt_lo:[0-9]+]], s[[exec_lo]], 0
15; GCN: v_mbcnt_hi_u32_b32{{(_e[0-9]+)?}} v[[mbcnt_hi:[0-9]+]], s[[exec_hi]], v[[mbcnt_lo]]
16; GCN: v_cmp_eq_u32{{(_e[0-9]+)?}} vcc, 0, v[[mbcnt_hi]]
17; GCN: s_bcnt1_i32_b64 s[[popcount:[0-9]+]], s{{\[}}[[exec_lo]]:[[exec_hi]]{{\]}}
18; GCN: v_mul_u32_u24{{(_e[0-9]+)?}} v[[value:[0-9]+]], s[[popcount]], 5
19; GCN: buffer_atomic_add v[[value]]
20; GCN: v_readfirstlane_b32 s{{[0-9]+}}, v[[value]]
21define amdgpu_ps void @add_i32_constant(<4 x i32> inreg %out, <4 x i32> inreg %inout) {
22entry:
23  %cond1 = call i1 @llvm.amdgcn.wqm.vote(i1 true)
24  %old = call i32 @llvm.amdgcn.buffer.atomic.add(i32 5, <4 x i32> %inout, i32 0, i32 0, i1 0)
25  %cond2 = call i1 @llvm.amdgcn.wqm.vote(i1 true)
26  %cond = and i1 %cond1, %cond2
27  br i1 %cond, label %if, label %else
28if:
29  %bitcast = bitcast i32 %old to float
30  call void @llvm.amdgcn.buffer.store.f32(float %bitcast, <4 x i32> %out, i32 0, i32 0, i1 0, i1 0)
31  ret void
32else:
33  ret void
34}
35
36; GCN-LABEL: add_i32_varying:
37; GFX7LESS-NOT: v_mbcnt_lo_u32_b32
38; GFX7LESS-NOT: v_mbcnt_hi_u32_b32
39; GFX8MORE: v_cmp_ne_u32_e64 s{{\[}}[[exec_lo:[0-9]+]]:[[exec_hi:[0-9]+]]{{\]}}, 1, 0
40; GFX8MORE: v_mbcnt_lo_u32_b32{{(_e[0-9]+)?}} v[[mbcnt_lo:[0-9]+]], s[[exec_lo]], 0
41; GFX8MORE: v_mbcnt_hi_u32_b32{{(_e[0-9]+)?}} v[[mbcnt_hi:[0-9]+]], s[[exec_hi]], v[[mbcnt_lo]]
42; GFX8MORE: v_readlane_b32 s[[scalar_value:[0-9]+]], v{{[0-9]+}}, 63
43; GFX8MORE: v_cmp_eq_u32{{(_e[0-9]+)?}} vcc, 0, v[[mbcnt_hi]]
44; GFX8MORE: v_mov_b32{{(_e[0-9]+)?}} v[[value:[0-9]+]], s[[scalar_value]]
45; GFX8MORE: buffer_atomic_add v[[value]]
46; GFX8MORE: v_readfirstlane_b32 s{{[0-9]+}}, v[[value]]
47define amdgpu_ps void @add_i32_varying(<4 x i32> inreg %out, <4 x i32> inreg %inout, i32 %val) {
48entry:
49  %cond1 = call i1 @llvm.amdgcn.wqm.vote(i1 true)
50  %old = call i32 @llvm.amdgcn.buffer.atomic.add(i32 %val, <4 x i32> %inout, i32 0, i32 0, i1 0)
51  %cond2 = call i1 @llvm.amdgcn.wqm.vote(i1 true)
52  %cond = and i1 %cond1, %cond2
53  br i1 %cond, label %if, label %else
54if:
55  %bitcast = bitcast i32 %old to float
56  call void @llvm.amdgcn.buffer.store.f32(float %bitcast, <4 x i32> %out, i32 0, i32 0, i1 0, i1 0)
57  ret void
58else:
59  ret void
60}
61