• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX900 %s
2; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX908 %s
3
4; GCN-LABEL: {{^}}global_atomic_fadd_ret_f32:
5; GCN: [[LOOP:BB[0-9]+_[0-9]+]]
6; GCN: v_add_f32_e32
7; GCN: global_atomic_cmpswap
8; GCN: s_andn2_b64 exec, exec,
9; GCN-NEXT: s_cbranch_execnz [[LOOP]]
10define amdgpu_kernel void @global_atomic_fadd_ret_f32(float addrspace(1)* %ptr) #0 {
11  %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
12  store float %result, float addrspace(1)* undef
13  ret void
14}
15
16; GCN-LABEL: {{^}}global_atomic_fadd_ret_f32_ieee:
17; GCN: [[LOOP:BB[0-9]+_[0-9]+]]
18; GCN: v_add_f32_e32
19; GCN: global_atomic_cmpswap
20; GCN: s_andn2_b64 exec, exec,
21; GCN-NEXT: s_cbranch_execnz [[LOOP]]
22define amdgpu_kernel void @global_atomic_fadd_ret_f32_ieee(float addrspace(1)* %ptr) {
23  %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
24  store float %result, float addrspace(1)* undef
25  ret void
26}
27
28; GCN-LABEL: {{^}}global_atomic_fadd_noret_f32:
29; GFX900: [[LOOP:BB[0-9]+_[0-9]+]]
30; GFX900: v_add_f32_e32
31; GFX900: global_atomic_cmpswap
32; GFX900: s_andn2_b64 exec, exec,
33; GFX900-NEXT: s_cbranch_execnz [[LOOP]]
34
35; GFX908-NOT: v_add_f32
36; GFX908:     global_atomic_add_f32 v{{[0-9]+}}, v{{[0-9]+}}, s
37; GFX908-NOT: s_cbranch_execnz
38define amdgpu_kernel void @global_atomic_fadd_noret_f32(float addrspace(1)* %ptr) #0 {
39  %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
40  ret void
41}
42
43; GCN-LABEL: {{^}}global_atomic_fadd_noret_f32_ieee:
44; GCN: global_atomic_cmpswap
45define amdgpu_kernel void @global_atomic_fadd_noret_f32_ieee(float addrspace(1)* %ptr) {
46  %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
47  ret void
48}
49
50; Make sure this artificially selects with an incorrect subtarget, but the feature set.
51; GCN-LABEL: {{^}}global_atomic_fadd_ret_f32_wrong_subtarget:
52define amdgpu_kernel void @global_atomic_fadd_ret_f32_wrong_subtarget(float addrspace(1)* %ptr) #1 {
53  %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
54  store float %result, float addrspace(1)* undef
55  ret void
56}
57
58; GCN-LABEL: {{^}}global_atomic_fadd_noret_f32_wrong_subtarget:
59define amdgpu_kernel void @global_atomic_fadd_noret_f32_wrong_subtarget(float addrspace(1)* %ptr) #1 {
60  %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
61  ret void
62}
63
64attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign"}
65attributes #1 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "target-cpu"="gfx803" "target-features"="+atomic-fadd-insts" }
66