1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx908 < %s | FileCheck -enable-var-scope --check-prefixes=GCN,GFX908 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx908 -mattr=-mfma-inline-literal-bug < %s | FileCheck -enable-var-scope --check-prefixes=GCN,GFX908 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a < %s | FileCheck -enable-var-scope --check-prefixes=GCN,GFX90A %s
declare <16 x i32> @llvm.amdgcn.mfma.i32.32x32x8i8(i32, i32, <16 x i32>, i32, i32, i32)
declare <4 x i32> @llvm.amdgcn.mfma.i32.16x16x16i8(i32, i32, <4 x i32>, i32, i32, i32)
; GCN-LABEL: {{^}}test_mfma_i32_32x32x8i8:
; GCN-DAG: v_mov_b32_e32 [[TWO:v[0-9]+]], 2
; GCN-DAG: v_mov_b32_e32 [[ONE:v[0-9]+]], 1
; GCN-DAG: s_load_dwordx16
; GFX908-DAG-COUNT-16: v_accvgpr_write_b32 a{{[0-9]+}}, v{{[0-9]+}}
; GFX90A-COUNT-16: v_accvgpr_write_b32 a{{[0-9]+}}, s{{[0-9]+}}
; GCN: v_mfma_i32_32x32x8i8 a[{{[0-9]+:[0-9]+}}], [[ONE]], [[TWO]], a[{{[0-9]+:[0-9]+}}] cbsz:1 abid:2 blgp:3
; GFX908-COUNT-16: v_accvgpr_read_b32
; GFX908: global_store_dwordx4
; GFX90A-NOT: v_accvgpr_read_b32
; GFX90A-COUNT-4: global_store_dwordx4 v{{[0-9]+}}, a[{{[0-9:]+}}]
define amdgpu_kernel void @test_mfma_i32_32x32x8i8(ptr addrspace(1) %arg) #0 {
bb:
%in.1 = load <16 x i32>, ptr addrspace(1) %arg
%mai.1 = tail call <16 x i32> @llvm.amdgcn.mfma.i32.32x32x8i8(i32 1, i32 2, <16 x i32> %in.1, i32 1, i32 2, i32 3)
store <16 x i32> %mai.1, ptr addrspace(1) %arg
ret void
}
; GCN-LABEL: {{^}}test_mfma_i32_16x16x16i8:
; GCN-DAG: v_mov_b32_e32 [[TWO:v[0-9]+]], 2
; GCN-DAG: v_mov_b32_e32 [[ONE:v[0-9]+]], 1
; GCN: s_load_dwordx4
; GFX908-COUNT-4: v_accvgpr_write_b32 a{{[0-9]+}}, v{{[0-9]+}}
; GFX90A-COUNT-4: v_accvgpr_write_b32 a{{[0-9]+}}, s{{[0-9]+}}
; GCN: v_mfma_i32_16x16x16i8 [[RES:a\[[0-9]+:[0-9]+\]]], [[ONE]], [[TWO]], a[{{[0-9]+:[0-9]+}}] cbsz:1 abid:2 blgp:3
; GFX908-COUNT-4: v_accvgpr_read_b32
; GFX908: global_store_dwordx4
; GFX90A-NOT: v_accvgpr_read_b32
; GFX90A: global_store_dwordx4 v{{[0-9]+}}, [[RES]]
define amdgpu_kernel void @test_mfma_i32_16x16x16i8(ptr addrspace(1) %arg) #0 {
bb:
%in.1 = load <4 x i32>, ptr addrspace(1) %arg
%mai.1 = tail call <4 x i32> @llvm.amdgcn.mfma.i32.16x16x16i8(i32 1, i32 2, <4 x i32> %in.1, i32 1, i32 2, i32 3)
store <4 x i32> %mai.1, ptr addrspace(1) %arg
ret void
}
attributes #0 = { "amdgpu-flat-work-group-size"="1,256" }
|
