xref: /llvm-project/llvm/test/CodeGen/AMDGPU/fp-min-max-num-flat-atomics.ll (revision ba52f06f9d92c7ca04b440f618f8d352ea121fcc) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs | FileCheck %s -check-prefixes=GFX12,GFX12-SDAG
3; RUN: llc < %s -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs | FileCheck %s -check-prefixes=GFX12,GFX12-GISEL
4
5declare float @llvm.amdgcn.flat.atomic.fmin.num.f32.p1.f32(ptr %ptr, float %data)
6declare float @llvm.amdgcn.flat.atomic.fmax.num.f32.p1.f32(ptr %ptr, float %data)
7
8define amdgpu_cs void @flat_atomic_fmin_num_f32_noret(ptr %ptr, float %data) {
9; GFX12-LABEL: flat_atomic_fmin_num_f32_noret:
10; GFX12:       ; %bb.0:
11; GFX12-NEXT:    flat_atomic_min_num_f32 v[0:1], v2
12; GFX12-NEXT:    s_endpgm
13  %ret = call float @llvm.amdgcn.flat.atomic.fmin.num.f32.p1.f32(ptr %ptr, float %data)
14  ret void
15}
16
17define amdgpu_cs void @flat_atomic_fmax_num_f32_noret(ptr %ptr, float %data) {
18; GFX12-LABEL: flat_atomic_fmax_num_f32_noret:
19; GFX12:       ; %bb.0:
20; GFX12-NEXT:    flat_atomic_max_num_f32 v[0:1], v2
21; GFX12-NEXT:    s_endpgm
22  %ret = call float @llvm.amdgcn.flat.atomic.fmax.num.f32.p1.f32(ptr %ptr, float %data)
23  ret void
24}
25
26define amdgpu_cs float @flat_atomic_fmin_num_f32_rtn(ptr %ptr, float %data, ptr %out) {
27; GFX12-LABEL: flat_atomic_fmin_num_f32_rtn:
28; GFX12:       ; %bb.0:
29; GFX12-NEXT:    flat_atomic_min_num_f32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN
30; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
31; GFX12-NEXT:    flat_store_b32 v[3:4], v0
32; GFX12-NEXT:    s_wait_dscnt 0x0
33; GFX12-NEXT:    ; return to shader part epilog
34  %ret = call float @llvm.amdgcn.flat.atomic.fmin.num.f32.p1.f32(ptr %ptr, float %data)
35  store float %ret, ptr %out
36  ret float %ret
37}
38
39define amdgpu_cs float @flat_atomic_fmax_num_f32_rtn(ptr %ptr, float %data, ptr %out) {
40; GFX12-LABEL: flat_atomic_fmax_num_f32_rtn:
41; GFX12:       ; %bb.0:
42; GFX12-NEXT:    flat_atomic_max_num_f32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN
43; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
44; GFX12-NEXT:    flat_store_b32 v[3:4], v0
45; GFX12-NEXT:    s_wait_dscnt 0x0
46; GFX12-NEXT:    ; return to shader part epilog
47  %ret = call float @llvm.amdgcn.flat.atomic.fmax.num.f32.p1.f32(ptr %ptr, float %data)
48  store float %ret, ptr %out
49  ret float %ret
50}
51;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
52; GFX12-GISEL: {{.*}}
53; GFX12-SDAG: {{.*}}
54