llvm/test/CodeGen/AMDGPU/llvm.amdgcn.wmma.imod.gfx1250.w32.ll - llvm-project - Git at Google

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck %s --check-prefix=GFX1250
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1250 -global-isel -global-isel-abort=2 < %s | FileCheck %s --check-prefix=GISEL

 define amdgpu_ps void @test_wmma_f32_16x16x4_f32_negA(<2 x float> %A, <2 x float> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x4_f32_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x4_f32_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x4.f32.v8f32.v2f32(i1 1, <2 x float> %A, i1 0, <2 x float> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x4_f32_negB(<2 x float> %A, <2 x float> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x4_f32_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x4_f32_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x4.f32.v8f32.v2f32(i1 0, <2 x float> %A, i1 1, <2 x float> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x4_f32_negC(<2 x float> %A, <2 x float> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x4_f32_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x4_f32_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x4.f32.v8f32.v2f32(i1 0, <2 x float> %A, i1 0, <2 x float> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x4_f32_neg_absC(<2 x float> %A, <2 x float> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x4_f32_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x4_f32_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x4.f32.v8f32.v2f32(i1 0, <2 x float> %A, i1 0, <2 x float> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x4_f32_ignoreC(<2 x float> %A, <2 x float> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x4_f32_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x4_f32_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[4:7], off
 ; GISEL-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x4.f32.v8f32.v2f32(i1 0, <2 x float> %A, i1 0, <2 x float> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_bf16_negA(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_bf16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_bf16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.bf16.v8f32.v16bf16(i1 1, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_bf16_negB(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_bf16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_bf16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.bf16.v8f32.v16bf16(i1 0, <16 x bfloat> %A, i1 1, <16 x bfloat> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_bf16_negC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_bf16_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_bf16_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.bf16.v8f32.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_bf16_neg_absC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_bf16_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_bf16_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.bf16.v8f32.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_bf16_ignoreC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_bf16_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_bf16_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.bf16.v8f32.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16_16x16x32_bf16_negA(<16 x bfloat> %A, <16 x bfloat> %B, <8 x bfloat> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16_16x16x32_bf16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16_16x16x32_bf16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16.16x16x32.bf16.v8bf16.v16bf16(i1 1, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 0, <8 x bfloat> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16_16x16x32_bf16_negB(<16 x bfloat> %A, <16 x bfloat> %B, <8 x bfloat> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16_16x16x32_bf16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16_16x16x32_bf16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 1, <16 x bfloat> %B, i16 0, <8 x bfloat> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16_16x16x32_bf16_negC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x bfloat> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16_16x16x32_bf16_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16_16x16x32_bf16_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 1, <8 x bfloat> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16_16x16x32_bf16_neg_absC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x bfloat> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16_16x16x32_bf16_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16_16x16x32_bf16_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 3, <8 x bfloat> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16_16x16x32_bf16_ignoreC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x bfloat> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16_16x16x32_bf16_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16_16x16x32_bf16_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16_16x16x32_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 4, <8 x bfloat> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16f32_16x16x32_bf16_negA(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16f32_16x16x32_bf16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16f32_16x16x32_bf16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16f32.16x16x32.bf16.v8bf16.v16bf16(i1 1, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16f32_16x16x32_bf16_negB(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16f32_16x16x32_bf16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16f32_16x16x32_bf16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16f32.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 1, <16 x bfloat> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16f32_16x16x32_bf16_negC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16f32_16x16x32_bf16_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16f32_16x16x32_bf16_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16f32.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16f32_16x16x32_bf16_neg_absC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16f32_16x16x32_bf16_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16f32_16x16x32_bf16_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16f32.16x16x32.bf16.v8bf16.v16bf16(i1 0, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_bf16f32_16x16x32_bf16_ignoreC(<16 x bfloat> %A, <16 x bfloat> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_bf16f32_16x16x32_bf16_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_bf16f32_16x16x32_bf16_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_bf16f32_16x16x32_bf16 v[26:29], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[26:29], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.wmma.bf16f32.16x16x32.bf16.v8bf16.v16bf16(i1 1, <16 x bfloat> %A, i1 0, <16 x bfloat> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_fp8_fp8_negC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_fp8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_fp8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.fp8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_fp8_fp8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_fp8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_fp8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.fp8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_fp8_fp8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_fp8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_fp8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.fp8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_fp8_bf8_negC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_fp8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_fp8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_fp8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_fp8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.bf8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_fp8_bf8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_fp8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_fp8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_fp8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_fp8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.bf8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_fp8_bf8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_fp8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_fp8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_fp8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_fp8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.bf8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_bf8_fp8_negC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_bf8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_bf8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_bf8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_bf8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.fp8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_bf8_fp8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_bf8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_bf8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_bf8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_bf8_fp8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.fp8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_bf8_fp8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_bf8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_bf8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_bf8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_bf8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.fp8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_bf8_bf8_negC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_bf8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_bf8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.bf8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_bf8_bf8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_bf8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_bf8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.bf8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x64_bf8_bf8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x64_bf8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x64_bf8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.bf8.v8f32.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_fp8_fp8_negC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_fp8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_fp8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_fp8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_fp8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.fp8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_fp8_fp8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_fp8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_fp8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_fp8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_fp8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.fp8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_fp8_fp8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_fp8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_fp8_fp8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_fp8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_fp8_fp8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.fp8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_fp8_bf8_negC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_fp8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_fp8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_fp8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_fp8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.bf8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_fp8_bf8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_fp8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_fp8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_fp8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_fp8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.bf8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_fp8_bf8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_fp8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_fp8_bf8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_fp8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_fp8_bf8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.bf8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_bf8_fp8_negC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_bf8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_bf8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_bf8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_bf8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.fp8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_bf8_fp8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_bf8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_bf8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_bf8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_bf8_fp8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.fp8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_bf8_fp8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_bf8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_bf8_fp8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_bf8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_bf8_fp8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.fp8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_bf8_bf8_negC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_bf8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_bf8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_bf8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_bf8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.bf8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_bf8_bf8_neg_absC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_bf8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_bf8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_bf8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_bf8_bf8 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.bf8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x64_bf8_bf8_ignoreC(<8 x i32> %A, <8 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x64_bf8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x64_bf8_bf8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x64_bf8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x64_bf8_bf8 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.bf8.v8f16.v8i32(<8 x i32> %A, <8 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_i32_16x16x64_iu8_signedA(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_i32_16x16x64_iu8_signedA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_i32_16x16x64_iu8_signedA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 1, <8 x i32> %A, i1 0, <8 x i32> %B, <8 x i32> %C, i1 false, i1 false)
   store <8 x i32> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_i32_16x16x64_iu8_signedB(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_i32_16x16x64_iu8_signedB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_i32_16x16x64_iu8_signedB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 0, <8 x i32> %A, i1 1, <8 x i32> %B, <8 x i32> %C, i1 false, i1 false)
   store <8 x i32> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_f16_negA(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_f16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_f16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.f16.v8f32.v16f16(i1 1, <16 x half> %A, i1 0, <16 x half> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_f16_negB(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_f16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_f16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.f16.v8f32.v16f16(i1 0, <16 x half> %A, i1 1, <16 x half> %B, i16 0, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_f16_negC(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_f16_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_f16_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.f16.v8f32.v16f16(i1 0, <16 x half> %A, i1 0, <16 x half> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_f16_neg_absC(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_f16_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_f16_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.f16.v8f32.v16f16(i1 0, <16 x half> %A, i1 0, <16 x half> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x32_f16_ignoreC(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x32_f16_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x32_f16_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[16:19], off
 ; GISEL-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.f16.v8f32.v16f16(i1 0, <16 x half> %A, i1 0, <16 x half> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x32_f16_negA(<16 x half> %A, <16 x half> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x32_f16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x32_f16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x32.f16.v8f16.v16f16(i1 1, <16 x half> %A, i1 0, <16 x half> %B, i16 0, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x32_f16_negB(<16 x half> %A, <16 x half> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x32_f16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x32_f16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x32.f16.v8f16.v16f16(i1 0, <16 x half> %A, i1 1, <16 x half> %B, i16 0, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x32_f16_negC(<16 x half> %A, <16 x half> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x32_f16_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x32_f16_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x32.f16.v8f16.v16f16(i1 0, <16 x half> %A, i1 0, <16 x half> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x32_f16_neg_absC(<16 x half> %A, <16 x half> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x32_f16_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x32_f16_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x32.f16.v8f16.v16f16(i1 0, <16 x half> %A, i1 0, <16 x half> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x32_f16_ignoreC(<16 x half> %A, <16 x half> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x32_f16_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GFX1250-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x32_f16_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x32_f16 v[16:19], v[0:7], v[8:15], v[16:19]
 ; GISEL-NEXT:    global_store_b128 v[20:21], v[16:19], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x32.f16.v8f16.v16f16(i1 0, <16 x half> %A, i1 0, <16 x half> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_f8f6f4_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_f8f6f4_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_f8f6f4_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 1, <8 x float> %C)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_f8f6f4_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_f8f6f4_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_f8f6f4_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 3, <8 x float> %C)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_f8f6f4_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_f8f6f4_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_f8f6f4_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 4, <8 x float> %C)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale_f32_16x16x128_f8f6f4_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale_f32_16x16x128_f8f6f4_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale_f32_16x16x128_f8f6f4_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.scale.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 1, <8 x float> %C, i32 1, i32 0, i32 2, i32 1, i32 0, i32 4, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale_f32_16x16x128_f8f6f4_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale_f32_16x16x128_f8f6f4_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale_f32_16x16x128_f8f6f4_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.scale.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 3, <8 x float> %C, i32 1, i32 0, i32 2, i32 1, i32 0, i32 4, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale_f32_16x16x128_f8f6f4_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale_f32_16x16x128_f8f6f4_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale_f32_16x16x128_f8f6f4_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.scale.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 4, <8 x float> %C, i32 1, i32 0, i32 2, i32 1, i32 0, i32 4, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale16_f32_16x16x128_f8f6f4_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale16_f32_16x16x128_f8f6f4_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale16_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale16_f32_16x16x128_f8f6f4_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale16_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.scale16.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 1, <8 x float> %C, i32 1, i32 0, i64 2, i32 1, i32 0, i64 4, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale16_f32_16x16x128_f8f6f4_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale16_f32_16x16x128_f8f6f4_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale16_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale16_f32_16x16x128_f8f6f4_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale16_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.scale16.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 3, <8 x float> %C, i32 1, i32 0, i64 2, i32 1, i32 0, i64 4, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale16_f32_16x16x128_f8f6f4_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale16_f32_16x16x128_f8f6f4_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale16_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale16_f32_16x16x128_f8f6f4_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale16_f32_16x16x128_f8f6f4 v[32:39], v[0:15], v[16:31], v[32:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.scale16.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32 0, <16 x i32> %A, i32 0, <16 x i32> %B, i16 4, <8 x float> %C, i32 1, i32 0, i64 2, i32 1, i32 0, i64 4, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_fp8_fp8_negC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_fp8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_fp8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_fp8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_fp8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.fp8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_fp8_fp8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_fp8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_fp8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_fp8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_fp8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.fp8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_fp8_fp8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_fp8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_fp8_fp8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_fp8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_fp8_fp8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.fp8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_fp8_bf8_negC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_fp8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_fp8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_fp8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_fp8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.bf8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_fp8_bf8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_fp8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_fp8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_fp8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_fp8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.bf8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_fp8_bf8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_fp8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_fp8_bf8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_fp8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_fp8_bf8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.bf8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_bf8_fp8_negC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_bf8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_bf8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_bf8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_bf8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.fp8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_bf8_fp8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_bf8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_bf8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_bf8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_bf8_fp8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.fp8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_bf8_fp8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_bf8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_bf8_fp8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_bf8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_bf8_fp8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.fp8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_bf8_bf8_negC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_bf8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_bf8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_bf8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_bf8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.bf8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_bf8_bf8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_bf8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_bf8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_bf8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_bf8_bf8 v[32:35], v[0:15], v[16:31], v[32:35] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.bf8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f16_16x16x128_bf8_bf8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x half> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f16_16x16x128_bf8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f16_16x16x128_bf8_bf8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GFX1250-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f16_16x16x128_bf8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f16_16x16x128_bf8_bf8 v[32:35], v[0:15], v[16:31], v[32:35]
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[32:35], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.bf8.v8f16.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x half> %C, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_fp8_fp8_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_fp8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_fp8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_fp8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_fp8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.fp8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_fp8_fp8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_fp8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_fp8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_fp8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_fp8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.fp8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_fp8_fp8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_fp8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_fp8_fp8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_fp8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_fp8_fp8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.fp8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_fp8_bf8_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_fp8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_fp8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_fp8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_fp8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.bf8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_fp8_bf8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_fp8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_fp8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_fp8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_fp8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.bf8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_fp8_bf8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_fp8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_fp8_bf8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_fp8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_fp8_bf8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.bf8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_bf8_fp8_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_bf8_fp8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_bf8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_bf8_fp8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_bf8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.fp8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_bf8_fp8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_bf8_fp8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_bf8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_bf8_fp8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_bf8_fp8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.fp8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_bf8_fp8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_bf8_fp8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_bf8_fp8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_bf8_fp8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_bf8_fp8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.fp8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_bf8_bf8_negC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_bf8_bf8_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_bf8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_bf8_bf8_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_bf8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.bf8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 1, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_bf8_bf8_neg_absC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_bf8_bf8_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_bf8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_bf8_bf8_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_bf8_bf8 v[32:39], v[0:15], v[16:31], v[32:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.bf8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 3, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_16x16x128_bf8_bf8_ignoreC(<16 x i32> %A, <16 x i32> %B, <8 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_16x16x128_bf8_bf8_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_16x16x128_bf8_bf8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_16x16x128_bf8_bf8_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_16x16x128_bf8_bf8 v[32:39], v[0:15], v[16:31], v[32:39]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.bf8.v8f32.v16i32(<16 x i32> %A, <16 x i32> %B, i16 4, <8 x float> %C, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_32x16x128_f4_negC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_32x16x128_f4_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39] neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_32x16x128_f4_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39] neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.f32.32x16x128.f4.v16i32.v8i32.v16f32(<16 x i32> %A, <8 x i32> %B, i16 1, <16 x float> %C)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_32x16x128_f4_neg_absC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_32x16x128_f4_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_32x16x128_f4_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39] neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.f32.32x16x128.f4.v16i32.v8i32.v16f32(<16 x i32> %A, <8 x i32> %B, i16 3, <16 x float> %C)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_f32_32x16x128_f4_ignoreC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_f32_32x16x128_f4_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_f32_32x16x128_f4_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.f32.32x16x128.f4.v16i32.v8i32.v16f32(<16 x i32> %A, <8 x i32> %B, i16 4, <16 x float> %C)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale_f32_32x16x128_f4_negC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale_f32_32x16x128_f4_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale_f32_32x16x128_f4_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.scale.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> %A, <8 x i32> %B, i16 1, <16 x float> %C, i32 1, i32 0, i32 2, i32 1, i32 0, i32 4, i1 false, i1 false)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale_f32_32x16x128_f4_neg_absC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale_f32_32x16x128_f4_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale_f32_32x16x128_f4_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.scale.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> %A, <8 x i32> %B, i16 3, <16 x float> %C, i32 1, i32 0, i32 2, i32 1, i32 0, i32 4, i1 false, i1 false)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale_f32_32x16x128_f4_ignoreC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale_f32_32x16x128_f4_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale_f32_32x16x128_f4_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.scale.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> %A, <8 x i32> %B, i16 4, <16 x float> %C, i32 1, i32 0, i32 2, i32 1, i32 0, i32 4, i1 false, i1 false)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale16_f32_32x16x128_f4_negC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale16_f32_32x16x128_f4_negC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale16_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale16_f32_32x16x128_f4_negC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale16_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.scale16.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> %A, <8 x i32> %B, i16 1, <16 x float> %C, i32 1, i32 0, i64 2, i32 1, i32 0, i64 4, i1 false, i1 false)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale16_f32_32x16x128_f4_neg_absC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale16_f32_32x16x128_f4_neg_absC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale16_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale16_f32_32x16x128_f4_neg_absC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale16_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1 neg_lo:[0,0,1] neg_hi:[0,0,1]
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.scale16.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> %A, <8 x i32> %B, i16 3, <16 x float> %C, i32 1, i32 0, i64 2, i32 1, i32 0, i64 4, i1 false, i1 false)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_wmma_scale16_f32_32x16x128_f4_ignoreC(<16 x i32> %A, <8 x i32> %B, <16 x float> %C, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_wmma_scale16_f32_32x16x128_f4_ignoreC:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_wmma_scale16_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GFX1250-NEXT:    s_clause 0x3
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_wmma_scale16_f32_32x16x128_f4_ignoreC:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_wmma_scale16_f32_32x16x128_f4 v[24:39], v[0:15], v[16:23], v[24:39], 2, 4 matrix_a_scale:MATRIX_SCALE_ROW1 matrix_b_scale:MATRIX_SCALE_ROW1
 ; GISEL-NEXT:    s_clause 0x3
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[32:35], off offset:32
 ; GISEL-NEXT:    global_store_b128 v[40:41], v[36:39], off offset:48
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <16 x float> @llvm.amdgcn.wmma.scale16.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> %A, <8 x i32> %B, i16 4, <16 x float> %C, i32 1, i32 0, i64 2, i32 1, i32 0, i64 4, i1 false, i1 false)
   store <16 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_f32_16x16x64_bf16_negA(<16 x bfloat> %A, <32 x bfloat> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_f32_16x16x64_bf16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_f32_16x16x64_bf16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i16(i1 1, <16 x bfloat> %A, i1 0, <32 x bfloat> %B, <8 x float> %C, i16 %Index, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_f32_16x16x64_bf16_negB(<16 x bfloat> %A, <32 x bfloat> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_f32_16x16x64_bf16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_f32_16x16x64_bf16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i16(i1 0, <16 x bfloat> %A, i1 1, <32 x bfloat> %B, <8 x float> %C, i16 %Index, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_bf16_16x16x64_bf16_negA(<16 x bfloat> %A, <32 x bfloat> %B, <8 x bfloat> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_bf16_16x16x64_bf16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_bf16_16x16x64_bf16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v31, v30 :: v_dual_mov_b32 v30, v29
 ; GFX1250-NEXT:    global_store_b128 v[30:31], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_bf16_16x16x64_bf16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_bf16_16x16x64_bf16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v31, v30 :: v_dual_mov_b32 v30, v29
 ; GISEL-NEXT:    global_store_b128 v[30:31], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.swmmac.bf16.16x16x64.bf16.v8bf16.v16bf16.v32bf16.i16(i1 1, <16 x bfloat> %A, i1 0, <32 x bfloat> %B, <8 x bfloat> %C, i16 %Index, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_bf16_16x16x64_bf16_negB(<16 x bfloat> %A, <32 x bfloat> %B, <8 x bfloat> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_bf16_16x16x64_bf16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_bf16_16x16x64_bf16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v31, v30 :: v_dual_mov_b32 v30, v29
 ; GFX1250-NEXT:    global_store_b128 v[30:31], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_bf16_16x16x64_bf16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_bf16_16x16x64_bf16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v31, v30 :: v_dual_mov_b32 v30, v29
 ; GISEL-NEXT:    global_store_b128 v[30:31], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x bfloat> @llvm.amdgcn.swmmac.bf16.16x16x64.bf16.v8bf16.v16bf16.v32bf16.i16(i1 0, <16 x bfloat> %A, i1 1, <32 x bfloat> %B, <8 x bfloat> %C, i16 %Index, i1 false, i1 false)
   store <8 x bfloat> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_bf16f32_16x16x64_bf16_negA(<16 x bfloat> %A, <32 x bfloat> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_bf16f32_16x16x64_bf16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_bf16f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_bf16f32_16x16x64_bf16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_bf16f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.swmmac.bf16f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i16(i1 1, <16 x bfloat> %A, i1 0, <32 x bfloat> %B, <8 x float> %C, i16 %Index, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_bf16f32_16x16x64_bf16_negB(<16 x bfloat> %A, <32 x bfloat> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_bf16f32_16x16x64_bf16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_bf16f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_bf16f32_16x16x64_bf16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_bf16f32_16x16x64_bf16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.swmmac.bf16f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i16(i1 0, <16 x bfloat> %A, i1 1, <32 x bfloat> %B, <8 x float> %C, i16 %Index, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_i32_16x16x128_iu8_signedA(<8 x i32> %A, <16 x i32> %B, <8 x i32> %C, i64 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_i32_16x16x128_iu8_signedA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_i32_16x16x128_iu8 v[24:31], v[0:7], v[8:23], v[32:33] neg_lo:[1,0,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_i32_16x16x128_iu8_signedA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_i32_16x16x128_iu8 v[24:31], v[0:7], v[8:23], v[32:33] neg_lo:[1,0,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x128.iu8.v8i32.v8i32.v16i32.i64(i1 1, <8 x i32> %A, i1 0, <16 x i32> %B, <8 x i32> %C, i64 %Index, i1 false, i1 false)
   store <8 x i32> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_i32_16x16x128_iu8_signedB(<8 x i32> %A, <16 x i32> %B, <8 x i32> %C, i64 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_i32_16x16x128_iu8_signedB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_i32_16x16x128_iu8 v[24:31], v[0:7], v[8:23], v[32:33] neg_lo:[0,1,0]
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_i32_16x16x128_iu8_signedB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_i32_16x16x128_iu8 v[24:31], v[0:7], v[8:23], v[32:33] neg_lo:[0,1,0]
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x128.iu8.v8i32.v8i32.v16i32.i64(i1 0, <8 x i32> %A, i1 1, <16 x i32> %B, <8 x i32> %C, i64 %Index, i1 false, i1 false)
   store <8 x i32> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_f32_16x16x64_f16_negA(<16 x half> %A, <32 x half> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_f32_16x16x64_f16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_f32_16x16x64_f16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_f32_16x16x64_f16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_f32_16x16x64_f16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v36, v33 :: v_dual_mov_b32 v37, v34
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[28:31], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.f16.v8f32.v16f16.v32f16.i16(i1 1, <16 x half> %A, i1 0, <32 x half> %B, <8 x float> %C, i16 %Index, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_f32_16x16x64_f16_negB(<16 x half> %A, <32 x half> %B, <8 x float> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_f32_16x16x64_f16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_f32_16x16x64_f16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v35, v34 :: v_dual_mov_b32 v34, v33
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[28:31], off offset:16
 ; GFX1250-NEXT:    global_store_b128 v[34:35], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_f32_16x16x64_f16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_f32_16x16x64_f16 v[24:31], v[0:7], v[8:23], v32 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v36, v33 :: v_dual_mov_b32 v37, v34
 ; GISEL-NEXT:    s_clause 0x1
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[24:27], off
 ; GISEL-NEXT:    global_store_b128 v[36:37], v[28:31], off offset:16
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.f16.v8f32.v16f16.v32f16.i16(i1 0, <16 x half> %A, i1 1, <32 x half> %B, <8 x float> %C, i16 %Index, i1 false, i1 false)
   store <8 x float> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_f16_16x16x64_f16_negA(<16 x half> %A, <32 x half> %B, <8 x half> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_f16_16x16x64_f16_negA:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_f16_16x16x64_f16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v31, v30 :: v_dual_mov_b32 v30, v29
 ; GFX1250-NEXT:    global_store_b128 v[30:31], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_f16_16x16x64_f16_negA:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_f16_16x16x64_f16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[1,0,0] neg_hi:[1,0,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v32, v29 :: v_dual_mov_b32 v33, v30
 ; GISEL-NEXT:    global_store_b128 v[32:33], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.swmmac.f16.16x16x64.f16.v8f16.v16f16.v32f16.i16(i1 1, <16 x half> %A, i1 0, <32 x half> %B, <8 x half> %C, i16 %Index, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 define amdgpu_ps void @test_swmmac_f16_16x16x64_f16_negB(<16 x half> %A, <32 x half> %B, <8 x half> %C, i16 %Index, ptr addrspace(1) %out) {
 ; GFX1250-LABEL: test_swmmac_f16_16x16x64_f16_negB:
 ; GFX1250:       ; %bb.0: ; %bb
 ; GFX1250-NEXT:    v_swmmac_f16_16x16x64_f16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GFX1250-NEXT:    v_dual_mov_b32 v31, v30 :: v_dual_mov_b32 v30, v29
 ; GFX1250-NEXT:    global_store_b128 v[30:31], v[24:27], off
 ; GFX1250-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_swmmac_f16_16x16x64_f16_negB:
 ; GISEL:       ; %bb.0: ; %bb
 ; GISEL-NEXT:    v_swmmac_f16_16x16x64_f16 v[24:27], v[0:7], v[8:23], v28 neg_lo:[0,1,0] neg_hi:[0,1,0]
 ; GISEL-NEXT:    v_dual_mov_b32 v32, v29 :: v_dual_mov_b32 v33, v30
 ; GISEL-NEXT:    global_store_b128 v[32:33], v[24:27], off
 ; GISEL-NEXT:    s_endpgm
 bb:
   %res = call <8 x half> @llvm.amdgcn.swmmac.f16.16x16x64.f16.v8f16.v16f16.v32f16.i16(i1 0, <16 x half> %A, i1 1, <32 x half> %B, <8 x half> %C, i16 %Index, i1 false, i1 false)
   store <8 x half> %res, ptr addrspace(1) %out
   ret void
 }

 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x4.f32.v8f32.v2f32(i1, <2 x float>, i1, <2 x float>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.bf16.v8f32.v16bf16(i1, <16 x bfloat>, i1, <16 x bfloat>, i16, <8 x float>, i1, i1)
 declare <8 x bfloat> @llvm.amdgcn.wmma.bf16.16x16x32.bf16.v8bf16.v16bf16(i1, <16 x bfloat>, i1, <16 x bfloat>, i16, <8 x bfloat>, i1, i1)
 declare <8 x bfloat> @llvm.amdgcn.wmma.bf16f32.16x16x32.bf16.v8bf16.v16bf16(i1, <16 x bfloat>, i1, <16 x bfloat>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.fp8.v8f32.v8i32(<8 x i32>, <8 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.fp8.bf8.v8f32.v8i32(<8 x i32>, <8 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.fp8.v8f32.v8i32(<8 x i32>, <8 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x64.bf8.bf8.v8f32.v8i32(<8 x i32>, <8 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.fp8.v8f16.v8i32(<8 x i32>, <8 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.fp8.bf8.v8f16.v8i32(<8 x i32>, <8 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.fp8.v8f16.v8i32(<8 x i32>, <8 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x64.bf8.bf8.v8f16.v8i32(<8 x i32>, <8 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x64.iu8.v8i32.v8i32(i1 immarg, <8 x i32>, i1 immarg, <8 x i32>, <8 x i32>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x32.f16.v8f32.v16f16(i1, <16 x half>, i1, <16 x half>, i16, <8 x float>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x32.f16.v8f16.v16f16(i1, <16 x half>, i1, <16 x half>, i16, <8 x half>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32, <16 x i32>, i32, <16 x i32>, i16, <8 x float>)
 declare <8 x float> @llvm.amdgcn.wmma.scale.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32, <16 x i32>, i32, <16 x i32>, i16, <8 x float>, i32, i32, i32, i32, i32, i32, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.scale16.f32.16x16x128.f8f6f4.v8f32.v16i32.v16i32(i32, <16 x i32>, i32, <16 x i32>, i16, <8 x float>, i32, i32, i64, i32, i32, i64, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.fp8.v8f16.v16i32(<16 x i32>, <16 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.fp8.bf8.v8f16.v16i32(<16 x i32>, <16 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.fp8.v8f16.v16i32(<16 x i32>, <16 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x half> @llvm.amdgcn.wmma.f16.16x16x128.bf8.bf8.v8f16.v16i32(<16 x i32>, <16 x i32>, i16, <8 x half>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.fp8.v8f32.v16i32(<16 x i32>, <16 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.fp8.bf8.v8f32.v16i32(<16 x i32>, <16 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.fp8.v8f32.v16i32(<16 x i32>, <16 x i32>, i16, <8 x float>, i1, i1)
 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x128.bf8.bf8.v8f32.v16i32(<16 x i32>, <16 x i32>, i16, <8 x float>, i1, i1)
 declare <16 x float> @llvm.amdgcn.wmma.f32.32x16x128.f4.v16i32.v8i32.v16f32(<16 x i32>, <8 x i32>, i16, <16 x float>)
 declare <16 x float> @llvm.amdgcn.wmma.scale.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32>, <8 x i32>, i16, <16 x float>, i32, i32, i32, i32, i32, i32, i1, i1)
 declare <16 x float> @llvm.amdgcn.wmma.scale16.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32>, <8 x i32>, i16, <16 x float>, i32, i32, i64, i32, i32, i64, i1, i1)

 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i16(i1, <16 x bfloat>, i1, <32 x bfloat>, <8 x float>, i16, i1, i1)
 declare <8 x bfloat> @llvm.amdgcn.swmmac.bf16.16x16x64.bf16.v8bf16.v16bf16.v32bf16.i16(i1, <16 x bfloat>, i1, <32 x bfloat>, <8 x bfloat>, i16, i1, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.bf16f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i16(i1, <16 x bfloat>, i1, <32 x bfloat>, <8 x float>, i16, i1, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x128.fp8.fp8.v8f32.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x float>, i64, i1, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x128.fp8.bf8.v8f32.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x float>, i64, i1, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x128.bf8.fp8.v8f32.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x float>, i64, i1, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x128.bf8.bf8.v8f32.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x float>, i64, i1, i1)
 declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x128.fp8.fp8.v8f16.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x half>, i64, i1, i1)
 declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x128.fp8.bf8.v8f16.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x half>, i64, i1, i1)
 declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x128.bf8.fp8.v8f16.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x half>, i64, i1, i1)
 declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x128.bf8.bf8.v8f16.v8i32.v16i32.i64(<8 x i32>, <16 x i32>, <8 x half>, i64, i1, i1)
 declare <8 x i32> @llvm.amdgcn.swmmac.i32.16x16x128.iu8.v8i32.v8i32.v16i32.i64(i1 immarg, <8 x i32>, i1 immarg, <16 x i32>, <8 x i32>, i64, i1, i1)
 declare <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.f16.v8f32.v16f16.v32f16.i16(i1, <16 x half>, i1, <32 x half>, <8 x float>, i16, i1, i1)
 declare <8 x half> @llvm.amdgcn.swmmac.f16.16x16x64.f16.v8f16.v16f16.v32f16.i16(i1, <16 x half>, i1, <32 x half>, <8 x half>, i16, i1, i1)