llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.i8.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn -mcpu=gfx908 < %s | FileCheck --check-prefixes=GCN,GFX908 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx908 -mattr=-mfma-inline-literal-bug < %s | FileCheck --check-prefixes=GCN,GFX908 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a < %s | FileCheck --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)

define amdgpu_kernel void @test_mfma_i32_32x32x8i8(ptr addrspace(1) %arg) #0 {
; GFX908-LABEL: test_mfma_i32_32x32x8i8:
; GFX908:       ; %bb.0: ; %bb
; GFX908-NEXT:    s_load_dwordx2 s[16:17], s[4:5], 0x24
; GFX908-NEXT:    v_mov_b32_e32 v0, 1
; GFX908-NEXT:    v_mov_b32_e32 v16, 0
; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
; GFX908-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
; GFX908-NEXT:    v_mov_b32_e32 v17, s0
; GFX908-NEXT:    v_mov_b32_e32 v1, s1
; GFX908-NEXT:    v_mov_b32_e32 v2, s2
; GFX908-NEXT:    v_accvgpr_write_b32 a0, v17
; GFX908-NEXT:    v_mov_b32_e32 v17, s3
; GFX908-NEXT:    v_accvgpr_write_b32 a1, v1
; GFX908-NEXT:    v_accvgpr_write_b32 a2, v2
; GFX908-NEXT:    v_accvgpr_write_b32 a3, v17
; GFX908-NEXT:    v_mov_b32_e32 v1, s4
; GFX908-NEXT:    v_mov_b32_e32 v2, s5
; GFX908-NEXT:    v_mov_b32_e32 v17, s6
; GFX908-NEXT:    v_accvgpr_write_b32 a4, v1
; GFX908-NEXT:    v_accvgpr_write_b32 a5, v2
; GFX908-NEXT:    v_accvgpr_write_b32 a6, v17
; GFX908-NEXT:    v_mov_b32_e32 v1, s7
; GFX908-NEXT:    v_mov_b32_e32 v2, s8
; GFX908-NEXT:    v_mov_b32_e32 v17, s9
; GFX908-NEXT:    v_accvgpr_write_b32 a7, v1
; GFX908-NEXT:    v_accvgpr_write_b32 a8, v2
; GFX908-NEXT:    v_accvgpr_write_b32 a9, v17
; GFX908-NEXT:    v_mov_b32_e32 v1, s10
; GFX908-NEXT:    v_mov_b32_e32 v2, s11
; GFX908-NEXT:    v_mov_b32_e32 v17, s12
; GFX908-NEXT:    v_accvgpr_write_b32 a10, v1
; GFX908-NEXT:    v_accvgpr_write_b32 a11, v2
; GFX908-NEXT:    v_accvgpr_write_b32 a12, v17
; GFX908-NEXT:    v_mov_b32_e32 v1, s13
; GFX908-NEXT:    v_mov_b32_e32 v2, s14
; GFX908-NEXT:    v_mov_b32_e32 v17, s15
; GFX908-NEXT:    v_accvgpr_write_b32 a13, v1
; GFX908-NEXT:    v_accvgpr_write_b32 a14, v2
; GFX908-NEXT:    v_accvgpr_write_b32 a15, v17
; GFX908-NEXT:    v_mov_b32_e32 v1, 2
; GFX908-NEXT:    s_nop 1
; GFX908-NEXT:    v_mfma_i32_32x32x8i8 a[0:15], v0, v1, a[0:15] cbsz:1 abid:2 blgp:3
; GFX908-NEXT:    s_nop 15
; GFX908-NEXT:    s_nop 1
; GFX908-NEXT:    v_accvgpr_read_b32 v15, a15
; GFX908-NEXT:    v_accvgpr_read_b32 v14, a14
; GFX908-NEXT:    v_accvgpr_read_b32 v13, a13
; GFX908-NEXT:    v_accvgpr_read_b32 v12, a12
; GFX908-NEXT:    v_accvgpr_read_b32 v3, a3
; GFX908-NEXT:    v_accvgpr_read_b32 v2, a2
; GFX908-NEXT:    v_accvgpr_read_b32 v1, a1
; GFX908-NEXT:    v_accvgpr_read_b32 v0, a0
; GFX908-NEXT:    v_accvgpr_read_b32 v7, a7
; GFX908-NEXT:    v_accvgpr_read_b32 v6, a6
; GFX908-NEXT:    v_accvgpr_read_b32 v5, a5
; GFX908-NEXT:    v_accvgpr_read_b32 v4, a4
; GFX908-NEXT:    v_accvgpr_read_b32 v11, a11
; GFX908-NEXT:    v_accvgpr_read_b32 v10, a10
; GFX908-NEXT:    v_accvgpr_read_b32 v9, a9
; GFX908-NEXT:    v_accvgpr_read_b32 v8, a8
; GFX908-NEXT:    global_store_dwordx4 v16, v[12:15], s[16:17] offset:48
; GFX908-NEXT:    s_nop 0
; GFX908-NEXT:    global_store_dwordx4 v16, v[8:11], s[16:17] offset:32
; GFX908-NEXT:    global_store_dwordx4 v16, v[4:7], s[16:17] offset:16
; GFX908-NEXT:    global_store_dwordx4 v16, v[0:3], s[16:17]
; GFX908-NEXT:    s_endpgm
;
; GFX90A-LABEL: test_mfma_i32_32x32x8i8:
; GFX90A:       ; %bb.0: ; %bb
; GFX90A-NEXT:    s_load_dwordx2 s[16:17], s[4:5], 0x24
; GFX90A-NEXT:    v_mov_b32_e32 v0, 1
; GFX90A-NEXT:    v_mov_b32_e32 v1, 2
; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
; GFX90A-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
; GFX90A-NEXT:    v_accvgpr_write_b32 a0, s0
; GFX90A-NEXT:    v_accvgpr_write_b32 a1, s1
; GFX90A-NEXT:    v_accvgpr_write_b32 a2, s2
; GFX90A-NEXT:    v_accvgpr_write_b32 a3, s3
; GFX90A-NEXT:    v_accvgpr_write_b32 a4, s4
; GFX90A-NEXT:    v_accvgpr_write_b32 a5, s5
; GFX90A-NEXT:    v_accvgpr_write_b32 a6, s6
; GFX90A-NEXT:    v_accvgpr_write_b32 a7, s7
; GFX90A-NEXT:    v_accvgpr_write_b32 a8, s8
; GFX90A-NEXT:    v_accvgpr_write_b32 a9, s9
; GFX90A-NEXT:    v_accvgpr_write_b32 a10, s10
; GFX90A-NEXT:    v_accvgpr_write_b32 a11, s11
; GFX90A-NEXT:    v_accvgpr_write_b32 a12, s12
; GFX90A-NEXT:    v_accvgpr_write_b32 a13, s13
; GFX90A-NEXT:    v_accvgpr_write_b32 a14, s14
; GFX90A-NEXT:    v_accvgpr_write_b32 a15, s15
; GFX90A-NEXT:    s_nop 1
; GFX90A-NEXT:    v_mfma_i32_32x32x8i8 a[0:15], v0, v1, a[0:15] cbsz:1 abid:2 blgp:3
; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
; GFX90A-NEXT:    s_nop 15
; GFX90A-NEXT:    s_nop 1
; GFX90A-NEXT:    global_store_dwordx4 v0, a[12:15], s[16:17] offset:48
; GFX90A-NEXT:    global_store_dwordx4 v0, a[8:11], s[16:17] offset:32
; GFX90A-NEXT:    global_store_dwordx4 v0, a[4:7], s[16:17] offset:16
; GFX90A-NEXT:    global_store_dwordx4 v0, a[0:3], s[16:17]
; GFX90A-NEXT:    s_endpgm
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
}

define amdgpu_kernel void @test_mfma_i32_16x16x16i8(ptr addrspace(1) %arg) #0 {
; GFX908-LABEL: test_mfma_i32_16x16x16i8:
; GFX908:       ; %bb.0: ; %bb
; GFX908-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
; GFX908-NEXT:    v_mov_b32_e32 v0, 1
; GFX908-NEXT:    v_mov_b32_e32 v1, 2
; GFX908-NEXT:    v_mov_b32_e32 v4, 0
; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
; GFX908-NEXT:    s_load_dwordx4 s[0:3], s[6:7], 0x0
; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
; GFX908-NEXT:    v_mov_b32_e32 v5, s0
; GFX908-NEXT:    v_mov_b32_e32 v2, s1
; GFX908-NEXT:    v_mov_b32_e32 v3, s2
; GFX908-NEXT:    v_accvgpr_write_b32 a0, v5
; GFX908-NEXT:    v_mov_b32_e32 v5, s3
; GFX908-NEXT:    v_accvgpr_write_b32 a1, v2
; GFX908-NEXT:    v_accvgpr_write_b32 a2, v3
; GFX908-NEXT:    v_accvgpr_write_b32 a3, v5
; GFX908-NEXT:    s_nop 0
; GFX908-NEXT:    v_mfma_i32_16x16x16i8 a[0:3], v0, v1, a[0:3] cbsz:1 abid:2 blgp:3
; GFX908-NEXT:    s_nop 9
; GFX908-NEXT:    v_accvgpr_read_b32 v0, a0
; GFX908-NEXT:    v_accvgpr_read_b32 v1, a1
; GFX908-NEXT:    v_accvgpr_read_b32 v2, a2
; GFX908-NEXT:    v_accvgpr_read_b32 v3, a3
; GFX908-NEXT:    s_nop 1
; GFX908-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
; GFX908-NEXT:    s_endpgm
;
; GFX90A-LABEL: test_mfma_i32_16x16x16i8:
; GFX90A:       ; %bb.0: ; %bb
; GFX90A-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
; GFX90A-NEXT:    v_mov_b32_e32 v0, 1
; GFX90A-NEXT:    v_mov_b32_e32 v2, 2
; GFX90A-NEXT:    v_mov_b32_e32 v1, 0
; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
; GFX90A-NEXT:    s_load_dwordx4 s[0:3], s[6:7], 0x0
; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
; GFX90A-NEXT:    v_accvgpr_write_b32 a0, s0
; GFX90A-NEXT:    v_accvgpr_write_b32 a1, s1
; GFX90A-NEXT:    v_accvgpr_write_b32 a2, s2
; GFX90A-NEXT:    v_accvgpr_write_b32 a3, s3
; GFX90A-NEXT:    s_nop 1
; GFX90A-NEXT:    v_mfma_i32_16x16x16i8 a[0:3], v0, v2, a[0:3] cbsz:1 abid:2 blgp:3
; GFX90A-NEXT:    s_nop 10
; GFX90A-NEXT:    global_store_dwordx4 v1, a[0:3], s[6:7]
; GFX90A-NEXT:    s_endpgm
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" }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}