diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll
index 7d85d3439eed9..beda16c17a5c9 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll
@@ -1,13 +1,9 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GFX942,GFX942-VGPRCD,GFX942-SDAG,GFX942-VGPRCD-SDAG %s
 ; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GFX942,GFX942-VGPRCD,GFX942-GISEL,GFX942-VGPRCD-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 -stress-regalloc=10 < %s | FileCheck --check-prefixes=GFX942,GFX942-AGPRCD,GFX942-SDAG,GFX942-AGPRCD-SDAG %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx942 -stress-regalloc=10 < %s | FileCheck --check-prefixes=GFX942,GFX942-AGPRCD,GFX942-GISEL,GFX942-AGPRCD-GISEL %s
 
 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx950 < %s | FileCheck --check-prefixes=GFX950,GFX950-VGPRCD,GFX950-SDAG,GFX950-VGPRCD-SDAG %s
 ; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx950 < %s | FileCheck --check-prefixes=GFX950,GFX950-VGPRCD,GFX950-GISEL,GFX950-VGPRCD-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx950 -stress-regalloc=10 < %s | FileCheck --check-prefixes=GFX950,GFX950-AGPRCD,GFX950-SDAG,GFX950-AGPRCD-SDAG %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx950 -stress-regalloc=10 < %s | FileCheck --check-prefixes=GFX950,GFX950-AGPRCD,GFX950-GISEL,GFX950-AGPRCD-GISEL %s
 
 declare <4 x i32> @llvm.amdgcn.mfma.i32.16x16x32.i8(i64, i64, <4 x i32>, i32, i32, i32)
 declare <16 x i32> @llvm.amdgcn.mfma.i32.32x32x16.i8(i64, i64, <16 x i32>, i32, i32, i32)
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll
index d358837452eab..8081a15b53bb7 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll
@@ -252,62 +252,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd(<8 x bfloat> %arg
 ; GCN-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; GCN-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
+; GCN-NEXT:    v_mov_b32_e32 v44, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT:    v_accvgpr_write_b32 a31, s23
-; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT:    v_accvgpr_write_b32 a30, s22
-; GCN-NEXT:    v_accvgpr_write_b32 a29, s21
-; GCN-NEXT:    v_accvgpr_write_b32 a28, s20
-; GCN-NEXT:    v_accvgpr_write_b32 a27, s19
-; GCN-NEXT:    v_accvgpr_write_b32 a26, s18
-; GCN-NEXT:    v_accvgpr_write_b32 a25, s17
-; GCN-NEXT:    v_accvgpr_write_b32 a24, s16
-; GCN-NEXT:    v_accvgpr_write_b32 a23, s15
-; GCN-NEXT:    v_accvgpr_write_b32 a22, s14
-; GCN-NEXT:    v_accvgpr_write_b32 a21, s13
-; GCN-NEXT:    v_accvgpr_write_b32 a20, s12
-; GCN-NEXT:    v_accvgpr_write_b32 a19, s11
-; GCN-NEXT:    v_accvgpr_write_b32 a18, s10
-; GCN-NEXT:    v_accvgpr_write_b32 a17, s9
-; GCN-NEXT:    v_accvgpr_write_b32 a16, s8
-; GCN-NEXT:    v_mov_b32_e32 v10, s20
-; GCN-NEXT:    v_mov_b32_e32 v11, s21
-; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[16:31]
-; GCN-NEXT:    v_mov_b32_e32 v12, s22
-; GCN-NEXT:    v_mov_b32_e32 v13, s23
-; GCN-NEXT:    v_mov_b32_e32 v0, s16
-; GCN-NEXT:    v_mov_b32_e32 v1, s17
-; GCN-NEXT:    v_mov_b32_e32 v2, s18
-; GCN-NEXT:    v_mov_b32_e32 v3, s19
-; GCN-NEXT:    global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; GCN-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; GCN-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; GCN-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; GCN-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; GCN-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; GCN-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; GCN-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; GCN-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; GCN-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; GCN-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GCN-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GCN-NEXT:    v_mov_b32_e32 v40, s20
+; GCN-NEXT:    v_mov_b32_e32 v41, s21
+; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 v[0:15], v[32:35], v[36:39], v[16:31]
+; GCN-NEXT:    v_mov_b32_e32 v42, s22
+; GCN-NEXT:    v_mov_b32_e32 v43, s23
+; GCN-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT:    s_waitcnt vmcnt(0)
+; GCN-NEXT:    s_nop 2
+; GCN-NEXT:    v_mov_b32_e32 v16, s16
+; GCN-NEXT:    v_mov_b32_e32 v17, s17
+; GCN-NEXT:    v_mov_b32_e32 v18, s18
+; GCN-NEXT:    v_mov_b32_e32 v19, s19
+; GCN-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_nop 0
-; GCN-NEXT:    v_mov_b32_e32 v0, s12
-; GCN-NEXT:    v_mov_b32_e32 v1, s13
-; GCN-NEXT:    v_mov_b32_e32 v2, s14
-; GCN-NEXT:    v_mov_b32_e32 v3, s15
-; GCN-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT:    v_mov_b32_e32 v16, s12
+; GCN-NEXT:    v_mov_b32_e32 v17, s13
+; GCN-NEXT:    v_mov_b32_e32 v18, s14
+; GCN-NEXT:    v_mov_b32_e32 v19, s15
+; GCN-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_nop 0
-; GCN-NEXT:    v_mov_b32_e32 v0, s8
-; GCN-NEXT:    v_mov_b32_e32 v1, s9
-; GCN-NEXT:    v_mov_b32_e32 v2, s10
-; GCN-NEXT:    v_mov_b32_e32 v3, s11
-; GCN-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; GCN-NEXT:    v_mov_b32_e32 v16, s8
+; GCN-NEXT:    v_mov_b32_e32 v17, s9
+; GCN-NEXT:    v_mov_b32_e32 v18, s10
+; GCN-NEXT:    v_mov_b32_e32 v19, s11
+; GCN-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0)
@@ -322,62 +315,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd__flags(<8 x bfloa
 ; GCN-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; GCN-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
+; GCN-NEXT:    v_mov_b32_e32 v44, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT:    v_accvgpr_write_b32 a31, s23
-; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT:    v_accvgpr_write_b32 a30, s22
-; GCN-NEXT:    v_accvgpr_write_b32 a29, s21
-; GCN-NEXT:    v_accvgpr_write_b32 a28, s20
-; GCN-NEXT:    v_accvgpr_write_b32 a27, s19
-; GCN-NEXT:    v_accvgpr_write_b32 a26, s18
-; GCN-NEXT:    v_accvgpr_write_b32 a25, s17
-; GCN-NEXT:    v_accvgpr_write_b32 a24, s16
-; GCN-NEXT:    v_accvgpr_write_b32 a23, s15
-; GCN-NEXT:    v_accvgpr_write_b32 a22, s14
-; GCN-NEXT:    v_accvgpr_write_b32 a21, s13
-; GCN-NEXT:    v_accvgpr_write_b32 a20, s12
-; GCN-NEXT:    v_accvgpr_write_b32 a19, s11
-; GCN-NEXT:    v_accvgpr_write_b32 a18, s10
-; GCN-NEXT:    v_accvgpr_write_b32 a17, s9
-; GCN-NEXT:    v_accvgpr_write_b32 a16, s8
-; GCN-NEXT:    v_mov_b32_e32 v10, s20
-; GCN-NEXT:    v_mov_b32_e32 v11, s21
-; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[16:31] cbsz:1 abid:2 blgp:3
-; GCN-NEXT:    v_mov_b32_e32 v12, s22
-; GCN-NEXT:    v_mov_b32_e32 v13, s23
-; GCN-NEXT:    v_mov_b32_e32 v0, s16
-; GCN-NEXT:    v_mov_b32_e32 v1, s17
-; GCN-NEXT:    v_mov_b32_e32 v2, s18
-; GCN-NEXT:    v_mov_b32_e32 v3, s19
-; GCN-NEXT:    global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; GCN-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; GCN-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; GCN-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; GCN-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; GCN-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; GCN-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; GCN-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; GCN-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; GCN-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; GCN-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GCN-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GCN-NEXT:    v_mov_b32_e32 v40, s20
+; GCN-NEXT:    v_mov_b32_e32 v41, s21
+; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; GCN-NEXT:    v_mov_b32_e32 v42, s22
+; GCN-NEXT:    v_mov_b32_e32 v43, s23
+; GCN-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT:    s_waitcnt vmcnt(0)
+; GCN-NEXT:    s_nop 2
+; GCN-NEXT:    v_mov_b32_e32 v16, s16
+; GCN-NEXT:    v_mov_b32_e32 v17, s17
+; GCN-NEXT:    v_mov_b32_e32 v18, s18
+; GCN-NEXT:    v_mov_b32_e32 v19, s19
+; GCN-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_nop 0
-; GCN-NEXT:    v_mov_b32_e32 v0, s12
-; GCN-NEXT:    v_mov_b32_e32 v1, s13
-; GCN-NEXT:    v_mov_b32_e32 v2, s14
-; GCN-NEXT:    v_mov_b32_e32 v3, s15
-; GCN-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT:    v_mov_b32_e32 v16, s12
+; GCN-NEXT:    v_mov_b32_e32 v17, s13
+; GCN-NEXT:    v_mov_b32_e32 v18, s14
+; GCN-NEXT:    v_mov_b32_e32 v19, s15
+; GCN-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_nop 0
-; GCN-NEXT:    v_mov_b32_e32 v0, s8
-; GCN-NEXT:    v_mov_b32_e32 v1, s9
-; GCN-NEXT:    v_mov_b32_e32 v2, s10
-; GCN-NEXT:    v_mov_b32_e32 v3, s11
-; GCN-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; GCN-NEXT:    v_mov_b32_e32 v16, s8
+; GCN-NEXT:    v_mov_b32_e32 v17, s9
+; GCN-NEXT:    v_mov_b32_e32 v18, s10
+; GCN-NEXT:    v_mov_b32_e32 v19, s11
+; GCN-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 1, i32 2, i32 3)
@@ -393,35 +379,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd_mac(<8 x bfloat>
 ; GCN-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT:    v_accvgpr_write_b32 a0, s8
-; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT:    v_accvgpr_write_b32 a1, s9
-; GCN-NEXT:    v_accvgpr_write_b32 a2, s10
-; GCN-NEXT:    v_accvgpr_write_b32 a3, s11
-; GCN-NEXT:    v_accvgpr_write_b32 a4, s12
-; GCN-NEXT:    v_accvgpr_write_b32 a5, s13
-; GCN-NEXT:    v_accvgpr_write_b32 a6, s14
-; GCN-NEXT:    v_accvgpr_write_b32 a7, s15
-; GCN-NEXT:    v_accvgpr_write_b32 a8, s16
-; GCN-NEXT:    v_accvgpr_write_b32 a9, s17
-; GCN-NEXT:    v_accvgpr_write_b32 a10, s18
-; GCN-NEXT:    v_accvgpr_write_b32 a11, s19
-; GCN-NEXT:    v_accvgpr_write_b32 a12, s20
-; GCN-NEXT:    v_accvgpr_write_b32 a13, s21
-; GCN-NEXT:    v_accvgpr_write_b32 a14, s22
-; GCN-NEXT:    v_accvgpr_write_b32 a15, s23
+; GCN-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; GCN-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GCN-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GCN-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GCN-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GCN-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GCN-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GCN-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; GCN-NEXT:    s_nop 1
-; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[0:15]
-; GCN-NEXT:    v_mov_b32_e32 v0, 0
+; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 v[0:15], v[16:19], v[20:23], v[0:15]
+; GCN-NEXT:    v_mov_b32_e32 v16, 0
 ; GCN-NEXT:    s_nop 7
 ; GCN-NEXT:    s_nop 2
-; GCN-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; GCN-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GCN-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GCN-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; GCN-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GCN-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GCN-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; GCN-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0)
   store <16 x float> %result, ptr addrspace(1) %out
@@ -435,40 +413,32 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd_mac_flags(<8 x bf
 ; GCN-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT:    v_accvgpr_write_b32 a0, s8
-; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT:    v_accvgpr_write_b32 a1, s9
-; GCN-NEXT:    v_accvgpr_write_b32 a2, s10
-; GCN-NEXT:    v_accvgpr_write_b32 a3, s11
-; GCN-NEXT:    v_accvgpr_write_b32 a4, s12
-; GCN-NEXT:    v_accvgpr_write_b32 a5, s13
-; GCN-NEXT:    v_accvgpr_write_b32 a6, s14
-; GCN-NEXT:    v_accvgpr_write_b32 a7, s15
-; GCN-NEXT:    v_accvgpr_write_b32 a8, s16
-; GCN-NEXT:    v_accvgpr_write_b32 a9, s17
-; GCN-NEXT:    v_accvgpr_write_b32 a10, s18
-; GCN-NEXT:    v_accvgpr_write_b32 a11, s19
-; GCN-NEXT:    v_accvgpr_write_b32 a12, s20
-; GCN-NEXT:    v_accvgpr_write_b32 a13, s21
-; GCN-NEXT:    v_accvgpr_write_b32 a14, s22
-; GCN-NEXT:    v_accvgpr_write_b32 a15, s23
+; GCN-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; GCN-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GCN-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GCN-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GCN-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GCN-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GCN-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GCN-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; GCN-NEXT:    s_nop 1
-; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; GCN-NEXT:    v_mov_b32_e32 v0, 0
+; GCN-NEXT:    v_mfma_f32_32x32x16_bf16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; GCN-NEXT:    v_mov_b32_e32 v16, 0
 ; GCN-NEXT:    s_nop 7
 ; GCN-NEXT:    s_nop 2
-; GCN-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; GCN-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GCN-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GCN-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; GCN-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GCN-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GCN-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; GCN-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 3, i32 2, i32 1)
   store <16 x float> %result, ptr addrspace(1) %out
   ret void
 }
 
-attributes #0 = { "amdgpu-flat-work-group-size"="512,512" }
+attributes #0 = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-agpr-alloc"="0,0" }
 attributes #1 = { "amdgpu-flat-work-group-size"="1,64" }
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
index 21465beb21de7..d81ec1c61634b 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
@@ -141,20 +141,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; SDAG-NEXT:    v_mov_b32_e32 v8, 0
+; SDAG-NEXT:    v_mov_b32_e32 v12, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3]
+; SDAG-NEXT:    v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11]
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd:
@@ -166,16 +164,14 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11]
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 6
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd:
@@ -183,20 +179,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp
 ; HEURRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; HEURRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; HEURRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; HEURRC-NEXT:    v_mov_b32_e32 v8, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v12, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s0
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3]
+; HEURRC-NEXT:    v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11]
 ; HEURRC-NEXT:    s_nop 7
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; HEURRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd:
@@ -266,20 +260,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; SDAG-NEXT:    v_mov_b32_e32 v8, 0
+; SDAG-NEXT:    v_mov_b32_e32 v12, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:3 abid:2 blgp:1
+; SDAG-NEXT:    v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags:
@@ -291,16 +283,14 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:3 abid:2 blgp:1
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 6
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags:
@@ -308,20 +298,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr
 ; HEURRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; HEURRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; HEURRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; HEURRC-NEXT:    v_mov_b32_e32 v8, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v12, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s0
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:3 abid:2 blgp:1
+; HEURRC-NEXT:    v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; HEURRC-NEXT:    s_nop 7
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; HEURRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags:
@@ -1505,62 +1493,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0,
 ; SDAG-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; SDAG-NEXT:    v_mov_b32_e32 v8, 0
+; SDAG-NEXT:    v_mov_b32_e32 v44, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; SDAG-NEXT:    v_accvgpr_write_b32 a31, s23
-; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; SDAG-NEXT:    v_accvgpr_write_b32 a30, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a29, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a28, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a27, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a26, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a25, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a24, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a23, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a22, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a21, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a20, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a19, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a18, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a17, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a16, s8
-; SDAG-NEXT:    v_mov_b32_e32 v10, s20
-; SDAG-NEXT:    v_mov_b32_e32 v11, s21
-; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[16:31]
-; SDAG-NEXT:    v_mov_b32_e32 v12, s22
-; SDAG-NEXT:    v_mov_b32_e32 v13, s23
-; SDAG-NEXT:    v_mov_b32_e32 v0, s16
-; SDAG-NEXT:    v_mov_b32_e32 v1, s17
-; SDAG-NEXT:    v_mov_b32_e32 v2, s18
-; SDAG-NEXT:    v_mov_b32_e32 v3, s19
-; SDAG-NEXT:    global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; SDAG-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; SDAG-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; SDAG-NEXT:    v_mov_b32_e32 v40, s20
+; SDAG-NEXT:    v_mov_b32_e32 v41, s21
+; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31]
+; SDAG-NEXT:    v_mov_b32_e32 v42, s22
+; SDAG-NEXT:    v_mov_b32_e32 v43, s23
+; SDAG-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    s_nop 2
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v0, s12
-; SDAG-NEXT:    v_mov_b32_e32 v1, s13
-; SDAG-NEXT:    v_mov_b32_e32 v2, s14
-; SDAG-NEXT:    v_mov_b32_e32 v3, s15
-; SDAG-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v0, s8
-; SDAG-NEXT:    v_mov_b32_e32 v1, s9
-; SDAG-NEXT:    v_mov_b32_e32 v2, s10
-; SDAG-NEXT:    v_mov_b32_e32 v3, s11
-; SDAG-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_endpgm
 ;
@@ -1569,52 +1550,44 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0,
 ; GISEL-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GISEL-NEXT:    v_mov_b32_e32 v24, 0
+; GISEL-NEXT:    v_mov_b32_e32 v56, 0
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[10:11]
-; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 a[16:31], v[0:3], v[4:7], a[0:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[16:17]
-; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[22:23]
-; GISEL-NEXT:    global_store_dwordx4 v24, v[8:11], s[0:1] sc0 sc1
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[42:43], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[40:41], s[8:9]
+; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 v[16:31], v[32:35], v[36:39], v[0:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[46:47], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[50:51], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[54:55], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[44:45], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[48:49], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[52:53], s[20:21]
+; GISEL-NEXT:    global_store_dwordx4 v56, v[40:43], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[12:15], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[44:47], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[16:19], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[48:51], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[20:23], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[52:55], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[16:19], s[0:1] sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[16:19], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[20:23], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[20:23], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[24:27], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[24:27], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[28:31], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[28:31], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_endpgm
 ;
@@ -1623,62 +1596,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0,
 ; HEURRC-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; HEURRC-NEXT:    v_mov_b32_e32 v8, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v44, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a31, s23
-; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a30, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a29, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a28, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a27, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a26, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a25, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a24, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a23, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a22, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a21, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a20, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a19, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a18, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a17, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a16, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v10, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v11, s21
-; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[16:31]
-; HEURRC-NEXT:    v_mov_b32_e32 v12, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v13, s23
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s16
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s17
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s18
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s19
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; HEURRC-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; HEURRC-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; HEURRC-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; HEURRC-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; HEURRC-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; HEURRC-NEXT:    v_mov_b32_e32 v40, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v41, s21
+; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31]
+; HEURRC-NEXT:    v_mov_b32_e32 v42, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v43, s23
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    s_nop 2
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s16
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s17
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s18
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s19
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s12
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s13
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s14
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s15
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s12
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s13
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s14
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s15
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s9
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s10
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s11
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s8
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s9
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s10
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s11
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_endpgm
 ;
@@ -1687,7 +1653,7 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0,
 ; VGPRRC-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; VGPRRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; VGPRRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; VGPRRC-NEXT:    v_mov_b32_e32 v40, 0
+; VGPRRC-NEXT:    v_mov_b32_e32 v44, 0
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
@@ -1701,41 +1667,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0,
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
-; VGPRRC-NEXT:    v_mov_b32_e32 v42, s20
-; VGPRRC-NEXT:    v_mov_b32_e32 v43, s21
+; VGPRRC-NEXT:    v_mov_b32_e32 v40, s20
+; VGPRRC-NEXT:    v_mov_b32_e32 v41, s21
 ; VGPRRC-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31]
-; VGPRRC-NEXT:    v_mov_b32_e32 v44, s22
-; VGPRRC-NEXT:    v_mov_b32_e32 v45, s23
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[42:45], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    v_mov_b32_e32 v42, s22
+; VGPRRC-NEXT:    v_mov_b32_e32 v43, s23
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 2
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s16
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s17
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s18
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s19
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s12
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s13
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s14
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s15
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s8
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s9
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s10
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s11
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_endpgm
 ; AGPR-LABEL: test_mfma_f32_32x32x16_f16__vgprcd:
@@ -1869,62 +1835,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half>
 ; SDAG-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; SDAG-NEXT:    v_mov_b32_e32 v8, 0
+; SDAG-NEXT:    v_mov_b32_e32 v44, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; SDAG-NEXT:    v_accvgpr_write_b32 a31, s23
-; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; SDAG-NEXT:    v_accvgpr_write_b32 a30, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a29, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a28, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a27, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a26, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a25, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a24, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a23, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a22, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a21, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a20, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a19, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a18, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a17, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a16, s8
-; SDAG-NEXT:    v_mov_b32_e32 v10, s20
-; SDAG-NEXT:    v_mov_b32_e32 v11, s21
-; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[16:31] cbsz:1 abid:2 blgp:3
-; SDAG-NEXT:    v_mov_b32_e32 v12, s22
-; SDAG-NEXT:    v_mov_b32_e32 v13, s23
-; SDAG-NEXT:    v_mov_b32_e32 v0, s16
-; SDAG-NEXT:    v_mov_b32_e32 v1, s17
-; SDAG-NEXT:    v_mov_b32_e32 v2, s18
-; SDAG-NEXT:    v_mov_b32_e32 v3, s19
-; SDAG-NEXT:    global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; SDAG-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; SDAG-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; SDAG-NEXT:    v_mov_b32_e32 v40, s20
+; SDAG-NEXT:    v_mov_b32_e32 v41, s21
+; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; SDAG-NEXT:    v_mov_b32_e32 v42, s22
+; SDAG-NEXT:    v_mov_b32_e32 v43, s23
+; SDAG-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    s_nop 2
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v0, s12
-; SDAG-NEXT:    v_mov_b32_e32 v1, s13
-; SDAG-NEXT:    v_mov_b32_e32 v2, s14
-; SDAG-NEXT:    v_mov_b32_e32 v3, s15
-; SDAG-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v0, s8
-; SDAG-NEXT:    v_mov_b32_e32 v1, s9
-; SDAG-NEXT:    v_mov_b32_e32 v2, s10
-; SDAG-NEXT:    v_mov_b32_e32 v3, s11
-; SDAG-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_endpgm
 ;
@@ -1933,52 +1892,44 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half>
 ; GISEL-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GISEL-NEXT:    v_mov_b32_e32 v24, 0
+; GISEL-NEXT:    v_mov_b32_e32 v56, 0
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[10:11]
-; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 a[16:31], v[0:3], v[4:7], a[0:15] cbsz:1 abid:2 blgp:3
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[16:17]
-; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[22:23]
-; GISEL-NEXT:    global_store_dwordx4 v24, v[8:11], s[0:1] sc0 sc1
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[42:43], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[40:41], s[8:9]
+; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 v[16:31], v[32:35], v[36:39], v[0:15] cbsz:1 abid:2 blgp:3
+; GISEL-NEXT:    v_mov_b64_e32 v[46:47], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[50:51], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[54:55], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[44:45], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[48:49], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[52:53], s[20:21]
+; GISEL-NEXT:    global_store_dwordx4 v56, v[40:43], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[12:15], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[44:47], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[16:19], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[48:51], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[20:23], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[52:55], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[16:19], s[0:1] sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[16:19], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[20:23], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[20:23], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[24:27], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[24:27], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[28:31], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[28:31], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_endpgm
 ;
@@ -1987,62 +1938,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half>
 ; HEURRC-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; HEURRC-NEXT:    v_mov_b32_e32 v8, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v44, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a31, s23
-; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a30, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a29, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a28, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a27, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a26, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a25, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a24, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a23, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a22, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a21, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a20, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a19, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a18, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a17, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a16, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v10, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v11, s21
-; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[16:31] cbsz:1 abid:2 blgp:3
-; HEURRC-NEXT:    v_mov_b32_e32 v12, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v13, s23
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s16
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s17
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s18
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s19
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; HEURRC-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; HEURRC-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; HEURRC-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; HEURRC-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; HEURRC-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; HEURRC-NEXT:    v_mov_b32_e32 v40, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v41, s21
+; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; HEURRC-NEXT:    v_mov_b32_e32 v42, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v43, s23
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    s_nop 2
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s16
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s17
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s18
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s19
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s12
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s13
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s14
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s15
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s12
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s13
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s14
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s15
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s9
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s10
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s11
-; HEURRC-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s8
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s9
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s10
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s11
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_endpgm
 ;
@@ -2051,7 +1995,7 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half>
 ; VGPRRC-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; VGPRRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; VGPRRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; VGPRRC-NEXT:    v_mov_b32_e32 v40, 0
+; VGPRRC-NEXT:    v_mov_b32_e32 v44, 0
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
@@ -2065,41 +2009,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half>
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
-; VGPRRC-NEXT:    v_mov_b32_e32 v42, s20
-; VGPRRC-NEXT:    v_mov_b32_e32 v43, s21
+; VGPRRC-NEXT:    v_mov_b32_e32 v40, s20
+; VGPRRC-NEXT:    v_mov_b32_e32 v41, s21
 ; VGPRRC-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
-; VGPRRC-NEXT:    v_mov_b32_e32 v44, s22
-; VGPRRC-NEXT:    v_mov_b32_e32 v45, s23
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[42:45], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    v_mov_b32_e32 v42, s22
+; VGPRRC-NEXT:    v_mov_b32_e32 v43, s23
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 2
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s16
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s17
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s18
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s19
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s12
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s13
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s14
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s15
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s8
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s9
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s10
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s11
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_endpgm
 ; AGPR-LABEL: test_mfma_f32_32x32x16_f16__vgprcd__flags:
@@ -2234,35 +2178,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac(<8 x half> %ar
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s8
-; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[0:15]
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15]
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac:
@@ -2271,35 +2207,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac(<8 x half> %ar
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[0:15]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15]
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
-; GISEL-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac:
@@ -2308,35 +2236,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac(<8 x half> %ar
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s8
-; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a4, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a5, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a6, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a7, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a8, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a9, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a10, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a11, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a12, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a13, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a14, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a15, s23
+; HEURRC-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; HEURRC-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; HEURRC-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; HEURRC-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[0:15]
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15]
+; HEURRC-NEXT:    v_mov_b32_e32 v16, 0
 ; HEURRC-NEXT:    s_nop 7
 ; HEURRC-NEXT:    s_nop 2
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac:
@@ -2443,35 +2363,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac_flags(<8 x hal
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s8
-; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac_flags:
@@ -2480,35 +2392,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac_flags(<8 x hal
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
-; GISEL-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac_flags:
@@ -2517,35 +2421,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd_mac_flags(<8 x hal
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s8
-; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a4, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a5, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a6, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a7, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a8, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a9, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a10, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a11, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a12, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a13, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a14, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a15, s23
+; HEURRC-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; HEURRC-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; HEURRC-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; HEURRC-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mfma_f32_32x32x16_f16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, 0
 ; HEURRC-NEXT:    s_nop 7
 ; HEURRC-NEXT:    s_nop 2
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_f32_32x32x16_f16__vgprcd_mac_flags:
@@ -2781,24 +2677,24 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd(ptr addrspa
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v12, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
+; SDAG-NEXT:    v_mov_b32_e32 v8, s0
+; SDAG-NEXT:    v_mov_b32_e32 v9, s1
+; SDAG-NEXT:    v_mov_b32_e32 v10, s2
+; SDAG-NEXT:    v_mov_b32_e32 v11, s3
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[2:5], v[6:9], a[0:3]
+; SDAG-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11]
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd:
@@ -2810,16 +2706,14 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd(ptr addrspa
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[4:7], a[0:3]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11]
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 6
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd:
@@ -2827,24 +2721,24 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd(ptr addrspa
 ; HEURRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; HEURRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v12, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s9
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s10
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s11
-; HEURRC-NEXT:    v_mov_b32_e32 v6, s12
-; HEURRC-NEXT:    v_mov_b32_e32 v7, s13
-; HEURRC-NEXT:    v_mov_b32_e32 v8, s14
-; HEURRC-NEXT:    v_mov_b32_e32 v9, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s0
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT:    v_mov_b32_e32 v0, s8
+; HEURRC-NEXT:    v_mov_b32_e32 v1, s9
+; HEURRC-NEXT:    v_mov_b32_e32 v2, s10
+; HEURRC-NEXT:    v_mov_b32_e32 v3, s11
+; HEURRC-NEXT:    v_mov_b32_e32 v4, s12
+; HEURRC-NEXT:    v_mov_b32_e32 v5, s13
+; HEURRC-NEXT:    v_mov_b32_e32 v6, s14
+; HEURRC-NEXT:    v_mov_b32_e32 v7, s15
+; HEURRC-NEXT:    v_mov_b32_e32 v8, s0
+; HEURRC-NEXT:    v_mov_b32_e32 v9, s1
+; HEURRC-NEXT:    v_mov_b32_e32 v10, s2
+; HEURRC-NEXT:    v_mov_b32_e32 v11, s3
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[2:5], v[6:9], a[0:3]
+; HEURRC-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11]
 ; HEURRC-NEXT:    s_nop 7
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; HEURRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd:
@@ -2852,24 +2746,24 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd(ptr addrspa
 ; VGPRRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; VGPRRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
 ; VGPRRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; VGPRRC-NEXT:    v_mov_b32_e32 v4, 0
+; VGPRRC-NEXT:    v_mov_b32_e32 v12, 0
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
-; VGPRRC-NEXT:    v_mov_b32_e32 v6, s8
-; VGPRRC-NEXT:    v_mov_b32_e32 v7, s9
-; VGPRRC-NEXT:    v_mov_b32_e32 v8, s10
-; VGPRRC-NEXT:    v_mov_b32_e32 v9, s11
-; VGPRRC-NEXT:    v_mov_b32_e32 v10, s12
-; VGPRRC-NEXT:    v_mov_b32_e32 v11, s13
-; VGPRRC-NEXT:    v_mov_b32_e32 v12, s14
-; VGPRRC-NEXT:    v_mov_b32_e32 v13, s15
-; VGPRRC-NEXT:    v_mov_b32_e32 v0, s0
-; VGPRRC-NEXT:    v_mov_b32_e32 v1, s1
-; VGPRRC-NEXT:    v_mov_b32_e32 v2, s2
-; VGPRRC-NEXT:    v_mov_b32_e32 v3, s3
+; VGPRRC-NEXT:    v_mov_b32_e32 v0, s8
+; VGPRRC-NEXT:    v_mov_b32_e32 v1, s9
+; VGPRRC-NEXT:    v_mov_b32_e32 v2, s10
+; VGPRRC-NEXT:    v_mov_b32_e32 v3, s11
+; VGPRRC-NEXT:    v_mov_b32_e32 v4, s12
+; VGPRRC-NEXT:    v_mov_b32_e32 v5, s13
+; VGPRRC-NEXT:    v_mov_b32_e32 v6, s14
+; VGPRRC-NEXT:    v_mov_b32_e32 v7, s15
+; VGPRRC-NEXT:    v_mov_b32_e32 v8, s0
+; VGPRRC-NEXT:    v_mov_b32_e32 v9, s1
+; VGPRRC-NEXT:    v_mov_b32_e32 v10, s2
+; VGPRRC-NEXT:    v_mov_b32_e32 v11, s3
 ; VGPRRC-NEXT:    s_nop 1
-; VGPRRC-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[6:9], v[10:13], v[0:3]
+; VGPRRC-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11]
 ; VGPRRC-NEXT:    s_nop 7
-; VGPRRC-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
+; VGPRRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; VGPRRC-NEXT:    s_endpgm
 ; AGPR-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd:
 ; AGPR:       ; %bb.0:
@@ -2930,24 +2824,24 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags(ptr
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v12, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
+; SDAG-NEXT:    v_mov_b32_e32 v8, s0
+; SDAG-NEXT:    v_mov_b32_e32 v9, s1
+; SDAG-NEXT:    v_mov_b32_e32 v10, s2
+; SDAG-NEXT:    v_mov_b32_e32 v11, s3
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[2:5], v[6:9], a[0:3] cbsz:3 abid:2 blgp:1
+; SDAG-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags:
@@ -2959,16 +2853,14 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags(ptr
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:3 abid:2 blgp:1
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 6
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags:
@@ -2976,24 +2868,24 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags(ptr
 ; HEURRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; HEURRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v12, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s9
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s10
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s11
-; HEURRC-NEXT:    v_mov_b32_e32 v6, s12
-; HEURRC-NEXT:    v_mov_b32_e32 v7, s13
-; HEURRC-NEXT:    v_mov_b32_e32 v8, s14
-; HEURRC-NEXT:    v_mov_b32_e32 v9, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s0
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT:    v_mov_b32_e32 v0, s8
+; HEURRC-NEXT:    v_mov_b32_e32 v1, s9
+; HEURRC-NEXT:    v_mov_b32_e32 v2, s10
+; HEURRC-NEXT:    v_mov_b32_e32 v3, s11
+; HEURRC-NEXT:    v_mov_b32_e32 v4, s12
+; HEURRC-NEXT:    v_mov_b32_e32 v5, s13
+; HEURRC-NEXT:    v_mov_b32_e32 v6, s14
+; HEURRC-NEXT:    v_mov_b32_e32 v7, s15
+; HEURRC-NEXT:    v_mov_b32_e32 v8, s0
+; HEURRC-NEXT:    v_mov_b32_e32 v9, s1
+; HEURRC-NEXT:    v_mov_b32_e32 v10, s2
+; HEURRC-NEXT:    v_mov_b32_e32 v11, s3
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_i32_16x16x64_i8 a[0:3], v[2:5], v[6:9], a[0:3] cbsz:3 abid:2 blgp:1
+; HEURRC-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; HEURRC-NEXT:    s_nop 7
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
+; HEURRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags:
@@ -3001,24 +2893,24 @@ define amdgpu_kernel void @test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags(ptr
 ; VGPRRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; VGPRRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
 ; VGPRRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; VGPRRC-NEXT:    v_mov_b32_e32 v4, 0
+; VGPRRC-NEXT:    v_mov_b32_e32 v12, 0
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
-; VGPRRC-NEXT:    v_mov_b32_e32 v6, s8
-; VGPRRC-NEXT:    v_mov_b32_e32 v7, s9
-; VGPRRC-NEXT:    v_mov_b32_e32 v8, s10
-; VGPRRC-NEXT:    v_mov_b32_e32 v9, s11
-; VGPRRC-NEXT:    v_mov_b32_e32 v10, s12
-; VGPRRC-NEXT:    v_mov_b32_e32 v11, s13
-; VGPRRC-NEXT:    v_mov_b32_e32 v12, s14
-; VGPRRC-NEXT:    v_mov_b32_e32 v13, s15
-; VGPRRC-NEXT:    v_mov_b32_e32 v0, s0
-; VGPRRC-NEXT:    v_mov_b32_e32 v1, s1
-; VGPRRC-NEXT:    v_mov_b32_e32 v2, s2
-; VGPRRC-NEXT:    v_mov_b32_e32 v3, s3
+; VGPRRC-NEXT:    v_mov_b32_e32 v0, s8
+; VGPRRC-NEXT:    v_mov_b32_e32 v1, s9
+; VGPRRC-NEXT:    v_mov_b32_e32 v2, s10
+; VGPRRC-NEXT:    v_mov_b32_e32 v3, s11
+; VGPRRC-NEXT:    v_mov_b32_e32 v4, s12
+; VGPRRC-NEXT:    v_mov_b32_e32 v5, s13
+; VGPRRC-NEXT:    v_mov_b32_e32 v6, s14
+; VGPRRC-NEXT:    v_mov_b32_e32 v7, s15
+; VGPRRC-NEXT:    v_mov_b32_e32 v8, s0
+; VGPRRC-NEXT:    v_mov_b32_e32 v9, s1
+; VGPRRC-NEXT:    v_mov_b32_e32 v10, s2
+; VGPRRC-NEXT:    v_mov_b32_e32 v11, s3
 ; VGPRRC-NEXT:    s_nop 1
-; VGPRRC-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1
+; VGPRRC-NEXT:    v_mfma_i32_16x16x64_i8 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; VGPRRC-NEXT:    s_nop 7
-; VGPRRC-NEXT:    global_store_dwordx4 v4, v[0:3], s[6:7]
+; VGPRRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; VGPRRC-NEXT:    s_endpgm
 ; AGPR-LABEL: test_mfma_i32_16x16x64_i8_no_agpr__vgprcd__flags:
 ; AGPR:       ; %bb.0:
@@ -4246,70 +4138,63 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd(<4 x i32> %arg0, <4
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v40, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s20
-; SDAG-NEXT:    v_mov_b32_e32 v3, s21
-; SDAG-NEXT:    v_mov_b32_e32 v4, s22
-; SDAG-NEXT:    v_mov_b32_e32 v5, s23
+; SDAG-NEXT:    v_mov_b32_e32 v32, s20
+; SDAG-NEXT:    v_mov_b32_e32 v33, s21
+; SDAG-NEXT:    v_mov_b32_e32 v34, s22
+; SDAG-NEXT:    v_mov_b32_e32 v35, s23
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; SDAG-NEXT:    v_mov_b32_e32 v6, s24
-; SDAG-NEXT:    v_mov_b32_e32 v7, s25
-; SDAG-NEXT:    v_mov_b32_e32 v8, s26
-; SDAG-NEXT:    v_mov_b32_e32 v9, s27
+; SDAG-NEXT:    v_mov_b32_e32 v36, s24
+; SDAG-NEXT:    v_mov_b32_e32 v37, s25
+; SDAG-NEXT:    v_mov_b32_e32 v38, s26
+; SDAG-NEXT:    v_mov_b32_e32 v39, s27
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a31, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a30, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a29, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a28, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a27, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a26, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a25, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a24, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a23, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a22, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a21, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a20, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a19, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a18, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a17, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a16, s8
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[2:5], v[6:9], a[16:31]
-; SDAG-NEXT:    v_mov_b32_e32 v2, s20
-; SDAG-NEXT:    v_mov_b32_e32 v3, s21
-; SDAG-NEXT:    v_mov_b32_e32 v4, s22
-; SDAG-NEXT:    v_mov_b32_e32 v5, s23
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31]
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_mov_b32_e32 v16, s20
+; SDAG-NEXT:    v_mov_b32_e32 v17, s21
+; SDAG-NEXT:    v_mov_b32_e32 v18, s22
+; SDAG-NEXT:    v_mov_b32_e32 v19, s23
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v2, s16
-; SDAG-NEXT:    v_mov_b32_e32 v3, s17
-; SDAG-NEXT:    v_mov_b32_e32 v4, s18
-; SDAG-NEXT:    v_mov_b32_e32 v5, s19
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_endpgm
 ;
@@ -4318,52 +4203,44 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd(<4 x i32> %arg0, <4
 ; GISEL-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GISEL-NEXT:    v_mov_b32_e32 v24, 0
+; GISEL-NEXT:    v_mov_b32_e32 v56, 0
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[10:11]
-; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 a[16:31], v[0:3], v[4:7], a[0:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[16:17]
-; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[22:23]
-; GISEL-NEXT:    global_store_dwordx4 v24, v[8:11], s[0:1] sc0 sc1
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[42:43], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[40:41], s[8:9]
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[16:31], v[32:35], v[36:39], v[0:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[46:47], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[50:51], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[54:55], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[44:45], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[48:49], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[52:53], s[20:21]
+; GISEL-NEXT:    global_store_dwordx4 v56, v[40:43], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[12:15], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[44:47], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[16:19], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[48:51], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[20:23], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[52:55], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[16:19], s[0:1] sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[16:19], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[20:23], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[20:23], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[24:27], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[24:27], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[28:31], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[28:31], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_endpgm
 ;
@@ -4371,70 +4248,63 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd(<4 x i32> %arg0, <4
 ; HEURRC:       ; %bb.0:
 ; HEURRC-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v40, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s21
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s23
+; HEURRC-NEXT:    v_mov_b32_e32 v32, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v33, s21
+; HEURRC-NEXT:    v_mov_b32_e32 v34, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v35, s23
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; HEURRC-NEXT:    v_mov_b32_e32 v6, s24
-; HEURRC-NEXT:    v_mov_b32_e32 v7, s25
-; HEURRC-NEXT:    v_mov_b32_e32 v8, s26
-; HEURRC-NEXT:    v_mov_b32_e32 v9, s27
+; HEURRC-NEXT:    v_mov_b32_e32 v36, s24
+; HEURRC-NEXT:    v_mov_b32_e32 v37, s25
+; HEURRC-NEXT:    v_mov_b32_e32 v38, s26
+; HEURRC-NEXT:    v_mov_b32_e32 v39, s27
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_accvgpr_write_b32 a31, s23
-; HEURRC-NEXT:    v_accvgpr_write_b32 a30, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a29, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a28, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a27, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a26, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a25, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a24, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a23, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a22, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a21, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a20, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a19, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a18, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a17, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a16, s8
+; HEURRC-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; HEURRC-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[2:5], v[6:9], a[16:31]
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s21
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s23
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31]
+; HEURRC-NEXT:    s_nop 6
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s21
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s23
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s16
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s17
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s18
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s19
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s16
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s17
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s18
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s19
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s12
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s13
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s14
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s15
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s12
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s13
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s14
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s15
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s9
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s10
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s11
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s8
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s9
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s10
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s11
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1] sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_endpgm
 ;
@@ -4442,17 +4312,17 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd(<4 x i32> %arg0, <4
 ; VGPRRC:       ; %bb.0:
 ; VGPRRC-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; VGPRRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; VGPRRC-NEXT:    v_mov_b32_e32 v32, 0
+; VGPRRC-NEXT:    v_mov_b32_e32 v40, 0
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
-; VGPRRC-NEXT:    v_mov_b32_e32 v34, s20
-; VGPRRC-NEXT:    v_mov_b32_e32 v35, s21
-; VGPRRC-NEXT:    v_mov_b32_e32 v36, s22
-; VGPRRC-NEXT:    v_mov_b32_e32 v37, s23
+; VGPRRC-NEXT:    v_mov_b32_e32 v32, s20
+; VGPRRC-NEXT:    v_mov_b32_e32 v33, s21
+; VGPRRC-NEXT:    v_mov_b32_e32 v34, s22
+; VGPRRC-NEXT:    v_mov_b32_e32 v35, s23
 ; VGPRRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; VGPRRC-NEXT:    v_mov_b32_e32 v38, s24
-; VGPRRC-NEXT:    v_mov_b32_e32 v39, s25
-; VGPRRC-NEXT:    v_mov_b32_e32 v40, s26
-; VGPRRC-NEXT:    v_mov_b32_e32 v41, s27
+; VGPRRC-NEXT:    v_mov_b32_e32 v36, s24
+; VGPRRC-NEXT:    v_mov_b32_e32 v37, s25
+; VGPRRC-NEXT:    v_mov_b32_e32 v38, s26
+; VGPRRC-NEXT:    v_mov_b32_e32 v39, s27
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
@@ -4463,42 +4333,42 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd(<4 x i32> %arg0, <4
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; VGPRRC-NEXT:    s_nop 1
-; VGPRRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[34:37], v[38:41], v[16:31]
+; VGPRRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31]
 ; VGPRRC-NEXT:    s_nop 6
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s20
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s21
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s22
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s23
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s16
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s17
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s18
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s19
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s12
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s13
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s14
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s15
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s8
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s9
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s10
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s11
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[8:11], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[12:15], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[0:3], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[4:7], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_endpgm
 ; AGPR-LABEL: test_mfma_i32_32x32x32_i8__vgprcd:
@@ -4645,70 +4515,63 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd__flags(<4 x i32> %a
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v40, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s20
-; SDAG-NEXT:    v_mov_b32_e32 v3, s21
-; SDAG-NEXT:    v_mov_b32_e32 v4, s22
-; SDAG-NEXT:    v_mov_b32_e32 v5, s23
+; SDAG-NEXT:    v_mov_b32_e32 v32, s20
+; SDAG-NEXT:    v_mov_b32_e32 v33, s21
+; SDAG-NEXT:    v_mov_b32_e32 v34, s22
+; SDAG-NEXT:    v_mov_b32_e32 v35, s23
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; SDAG-NEXT:    v_mov_b32_e32 v6, s24
-; SDAG-NEXT:    v_mov_b32_e32 v7, s25
-; SDAG-NEXT:    v_mov_b32_e32 v8, s26
-; SDAG-NEXT:    v_mov_b32_e32 v9, s27
+; SDAG-NEXT:    v_mov_b32_e32 v36, s24
+; SDAG-NEXT:    v_mov_b32_e32 v37, s25
+; SDAG-NEXT:    v_mov_b32_e32 v38, s26
+; SDAG-NEXT:    v_mov_b32_e32 v39, s27
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a31, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a30, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a29, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a28, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a27, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a26, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a25, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a24, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a23, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a22, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a21, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a20, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a19, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a18, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a17, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a16, s8
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[2:5], v[6:9], a[16:31] cbsz:1 abid:2 blgp:3
-; SDAG-NEXT:    v_mov_b32_e32 v2, s20
-; SDAG-NEXT:    v_mov_b32_e32 v3, s21
-; SDAG-NEXT:    v_mov_b32_e32 v4, s22
-; SDAG-NEXT:    v_mov_b32_e32 v5, s23
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_mov_b32_e32 v16, s20
+; SDAG-NEXT:    v_mov_b32_e32 v17, s21
+; SDAG-NEXT:    v_mov_b32_e32 v18, s22
+; SDAG-NEXT:    v_mov_b32_e32 v19, s23
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v2, s16
-; SDAG-NEXT:    v_mov_b32_e32 v3, s17
-; SDAG-NEXT:    v_mov_b32_e32 v4, s18
-; SDAG-NEXT:    v_mov_b32_e32 v5, s19
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] sc0 sc1
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1] sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16 sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_endpgm
 ;
@@ -4717,52 +4580,44 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd__flags(<4 x i32> %a
 ; GISEL-NEXT:    s_load_dwordx8 s[24:31], s[4:5], 0x24
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GISEL-NEXT:    v_mov_b32_e32 v24, 0
+; GISEL-NEXT:    v_mov_b32_e32 v56, 0
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[10:11]
-; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 a[16:31], v[0:3], v[4:7], a[0:15] cbsz:1 abid:2 blgp:3
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[16:17]
-; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[22:23]
-; GISEL-NEXT:    global_store_dwordx4 v24, v[8:11], s[0:1] sc0 sc1
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[42:43], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[40:41], s[8:9]
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[16:31], v[32:35], v[36:39], v[0:15] cbsz:1 abid:2 blgp:3
+; GISEL-NEXT:    v_mov_b64_e32 v[46:47], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[50:51], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[54:55], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[44:45], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[48:49], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[52:53], s[20:21]
+; GISEL-NEXT:    global_store_dwordx4 v56, v[40:43], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[12:15], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[44:47], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[16:19], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[48:51], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, v[20:23], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[52:55], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[16:19], s[0:1] sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[16:19], s[0:1] sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[20:23], s[0:1] offset:16 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[20:23], s[0:1] offset:16 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[24:27], s[0:1] offset:32 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[24:27], s[0:1] offset:32 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v24, a[28:31], s[0:1] offset:48 sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v56, v[28:31], s[0:1] offset:48 sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_endpgm
 ;
@@ -4770,70 +4625,63 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd__flags(<4 x i32> %a
 ; HEURRC:       ; %bb.0:
 ; HEURRC-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v40, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s21
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s23
+; HEURRC-NEXT:    v_mov_b32_e32 v32, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v33, s21
+; HEURRC-NEXT:    v_mov_b32_e32 v34, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v35, s23
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; HEURRC-NEXT:    v_mov_b32_e32 v6, s24
-; HEURRC-NEXT:    v_mov_b32_e32 v7, s25
-; HEURRC-NEXT:    v_mov_b32_e32 v8, s26
-; HEURRC-NEXT:    v_mov_b32_e32 v9, s27
+; HEURRC-NEXT:    v_mov_b32_e32 v36, s24
+; HEURRC-NEXT:    v_mov_b32_e32 v37, s25
+; HEURRC-NEXT:    v_mov_b32_e32 v38, s26
+; HEURRC-NEXT:    v_mov_b32_e32 v39, s27
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_accvgpr_write_b32 a31, s23
-; HEURRC-NEXT:    v_accvgpr_write_b32 a30, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a29, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a28, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a27, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a26, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a25, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a24, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a23, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a22, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a21, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a20, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a19, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a18, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a17, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a16, s8
+; HEURRC-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; HEURRC-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[2:5], v[6:9], a[16:31] cbsz:1 abid:2 blgp:3
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s21
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s23
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; HEURRC-NEXT:    s_nop 6
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s21
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s23
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s16
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s17
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s18
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s19
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s16
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s17
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s18
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s19
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s12
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s13
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s14
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s15
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s12
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s13
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s14
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s15
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_nop 0
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s8
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s9
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s10
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s11
-; HEURRC-NEXT:    global_store_dwordx4 v0, v[2:5], s[0:1] sc0 sc1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s8
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s9
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s10
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s11
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1] sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16 sc0 sc1
+; HEURRC-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
 ; HEURRC-NEXT:    s_waitcnt vmcnt(0)
 ; HEURRC-NEXT:    s_endpgm
 ;
@@ -4841,17 +4689,17 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd__flags(<4 x i32> %a
 ; VGPRRC:       ; %bb.0:
 ; VGPRRC-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; VGPRRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; VGPRRC-NEXT:    v_mov_b32_e32 v32, 0
+; VGPRRC-NEXT:    v_mov_b32_e32 v40, 0
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
-; VGPRRC-NEXT:    v_mov_b32_e32 v34, s20
-; VGPRRC-NEXT:    v_mov_b32_e32 v35, s21
-; VGPRRC-NEXT:    v_mov_b32_e32 v36, s22
-; VGPRRC-NEXT:    v_mov_b32_e32 v37, s23
+; VGPRRC-NEXT:    v_mov_b32_e32 v32, s20
+; VGPRRC-NEXT:    v_mov_b32_e32 v33, s21
+; VGPRRC-NEXT:    v_mov_b32_e32 v34, s22
+; VGPRRC-NEXT:    v_mov_b32_e32 v35, s23
 ; VGPRRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; VGPRRC-NEXT:    v_mov_b32_e32 v38, s24
-; VGPRRC-NEXT:    v_mov_b32_e32 v39, s25
-; VGPRRC-NEXT:    v_mov_b32_e32 v40, s26
-; VGPRRC-NEXT:    v_mov_b32_e32 v41, s27
+; VGPRRC-NEXT:    v_mov_b32_e32 v36, s24
+; VGPRRC-NEXT:    v_mov_b32_e32 v37, s25
+; VGPRRC-NEXT:    v_mov_b32_e32 v38, s26
+; VGPRRC-NEXT:    v_mov_b32_e32 v39, s27
 ; VGPRRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
@@ -4862,42 +4710,42 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd__flags(<4 x i32> %a
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
 ; VGPRRC-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; VGPRRC-NEXT:    s_nop 1
-; VGPRRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[34:37], v[38:41], v[16:31] cbsz:1 abid:2 blgp:3
+; VGPRRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
 ; VGPRRC-NEXT:    s_nop 6
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s20
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s21
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s22
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s23
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s16
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s17
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s18
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s19
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s12
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s13
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s14
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s15
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_nop 0
 ; VGPRRC-NEXT:    v_mov_b32_e32 v16, s8
 ; VGPRRC-NEXT:    v_mov_b32_e32 v17, s9
 ; VGPRRC-NEXT:    v_mov_b32_e32 v18, s10
 ; VGPRRC-NEXT:    v_mov_b32_e32 v19, s11
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[16:19], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[16:19], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[8:11], s[0:1] offset:32 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[8:11], s[0:1] offset:32 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[12:15], s[0:1] offset:48 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[12:15], s[0:1] offset:48 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[0:3], s[0:1] sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[0:3], s[0:1] sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
-; VGPRRC-NEXT:    global_store_dwordx4 v32, v[4:7], s[0:1] offset:16 sc0 sc1
+; VGPRRC-NEXT:    global_store_dwordx4 v40, v[4:7], s[0:1] offset:16 sc0 sc1
 ; VGPRRC-NEXT:    s_waitcnt vmcnt(0)
 ; VGPRRC-NEXT:    s_endpgm
 ; AGPR-LABEL: test_mfma_i32_32x32x32_i8__vgprcd__flags:
@@ -5045,41 +4893,33 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac(<4 x i32> %arg0
 ; SDAG-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v0, s20
-; SDAG-NEXT:    v_mov_b32_e32 v1, s21
-; SDAG-NEXT:    v_mov_b32_e32 v2, s22
-; SDAG-NEXT:    v_mov_b32_e32 v3, s23
+; SDAG-NEXT:    v_mov_b32_e32 v16, s20
+; SDAG-NEXT:    v_mov_b32_e32 v17, s21
+; SDAG-NEXT:    v_mov_b32_e32 v18, s22
+; SDAG-NEXT:    v_mov_b32_e32 v19, s23
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; SDAG-NEXT:    v_mov_b32_e32 v4, s24
-; SDAG-NEXT:    v_mov_b32_e32 v5, s25
-; SDAG-NEXT:    v_mov_b32_e32 v6, s26
-; SDAG-NEXT:    v_mov_b32_e32 v7, s27
+; SDAG-NEXT:    v_mov_b32_e32 v20, s24
+; SDAG-NEXT:    v_mov_b32_e32 v21, s25
+; SDAG-NEXT:    v_mov_b32_e32 v22, s26
+; SDAG-NEXT:    v_mov_b32_e32 v23, s27
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s8
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15]
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15]
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac:
@@ -5088,35 +4928,27 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac(<4 x i32> %arg0
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15]
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
-; GISEL-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac:
@@ -5124,41 +4956,33 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac(<4 x i32> %arg0
 ; HEURRC-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s21
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s23
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s21
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s23
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s24
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s25
-; HEURRC-NEXT:    v_mov_b32_e32 v6, s26
-; HEURRC-NEXT:    v_mov_b32_e32 v7, s27
+; HEURRC-NEXT:    v_mov_b32_e32 v20, s24
+; HEURRC-NEXT:    v_mov_b32_e32 v21, s25
+; HEURRC-NEXT:    v_mov_b32_e32 v22, s26
+; HEURRC-NEXT:    v_mov_b32_e32 v23, s27
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s8
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a4, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a5, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a6, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a7, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a8, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a9, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a10, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a11, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a12, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a13, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a14, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a15, s23
+; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15]
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15]
+; HEURRC-NEXT:    v_mov_b32_e32 v16, 0
 ; HEURRC-NEXT:    s_nop 7
 ; HEURRC-NEXT:    s_nop 2
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac:
@@ -5279,41 +5103,33 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac_flags(<4 x i32>
 ; SDAG-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v0, s20
-; SDAG-NEXT:    v_mov_b32_e32 v1, s21
-; SDAG-NEXT:    v_mov_b32_e32 v2, s22
-; SDAG-NEXT:    v_mov_b32_e32 v3, s23
+; SDAG-NEXT:    v_mov_b32_e32 v16, s20
+; SDAG-NEXT:    v_mov_b32_e32 v17, s21
+; SDAG-NEXT:    v_mov_b32_e32 v18, s22
+; SDAG-NEXT:    v_mov_b32_e32 v19, s23
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; SDAG-NEXT:    v_mov_b32_e32 v4, s24
-; SDAG-NEXT:    v_mov_b32_e32 v5, s25
-; SDAG-NEXT:    v_mov_b32_e32 v6, s26
-; SDAG-NEXT:    v_mov_b32_e32 v7, s27
+; SDAG-NEXT:    v_mov_b32_e32 v20, s24
+; SDAG-NEXT:    v_mov_b32_e32 v21, s25
+; SDAG-NEXT:    v_mov_b32_e32 v22, s26
+; SDAG-NEXT:    v_mov_b32_e32 v23, s27
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s8
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac_flags:
@@ -5322,35 +5138,27 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac_flags(<4 x i32>
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[24:25]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[26:27]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[28:29]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[30:31]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[28:29]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[30:31]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
-; GISEL-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
 ; GISEL-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac_flags:
@@ -5358,41 +5166,33 @@ define amdgpu_kernel void @test_mfma_i32_32x32x32_i8__vgprcd_mac_flags(<4 x i32>
 ; HEURRC-NEXT:    s_load_dwordx8 s[20:27], s[4:5], 0x24
 ; HEURRC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa4
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_mov_b32_e32 v0, s20
-; HEURRC-NEXT:    v_mov_b32_e32 v1, s21
-; HEURRC-NEXT:    v_mov_b32_e32 v2, s22
-; HEURRC-NEXT:    v_mov_b32_e32 v3, s23
+; HEURRC-NEXT:    v_mov_b32_e32 v16, s20
+; HEURRC-NEXT:    v_mov_b32_e32 v17, s21
+; HEURRC-NEXT:    v_mov_b32_e32 v18, s22
+; HEURRC-NEXT:    v_mov_b32_e32 v19, s23
 ; HEURRC-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x64
-; HEURRC-NEXT:    v_mov_b32_e32 v4, s24
-; HEURRC-NEXT:    v_mov_b32_e32 v5, s25
-; HEURRC-NEXT:    v_mov_b32_e32 v6, s26
-; HEURRC-NEXT:    v_mov_b32_e32 v7, s27
+; HEURRC-NEXT:    v_mov_b32_e32 v20, s24
+; HEURRC-NEXT:    v_mov_b32_e32 v21, s25
+; HEURRC-NEXT:    v_mov_b32_e32 v22, s26
+; HEURRC-NEXT:    v_mov_b32_e32 v23, s27
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s8
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s9
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s10
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s11
-; HEURRC-NEXT:    v_accvgpr_write_b32 a4, s12
-; HEURRC-NEXT:    v_accvgpr_write_b32 a5, s13
-; HEURRC-NEXT:    v_accvgpr_write_b32 a6, s14
-; HEURRC-NEXT:    v_accvgpr_write_b32 a7, s15
-; HEURRC-NEXT:    v_accvgpr_write_b32 a8, s16
-; HEURRC-NEXT:    v_accvgpr_write_b32 a9, s17
-; HEURRC-NEXT:    v_accvgpr_write_b32 a10, s18
-; HEURRC-NEXT:    v_accvgpr_write_b32 a11, s19
-; HEURRC-NEXT:    v_accvgpr_write_b32 a12, s20
-; HEURRC-NEXT:    v_accvgpr_write_b32 a13, s21
-; HEURRC-NEXT:    v_accvgpr_write_b32 a14, s22
-; HEURRC-NEXT:    v_accvgpr_write_b32 a15, s23
+; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; HEURRC-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; HEURRC-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; HEURRC-NEXT:    v_mov_b32_e32 v0, 0
+; HEURRC-NEXT:    v_mfma_i32_32x32x32_i8 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; HEURRC-NEXT:    v_mov_b32_e32 v16, 0
 ; HEURRC-NEXT:    s_nop 7
 ; HEURRC-NEXT:    s_nop 2
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; HEURRC-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; HEURRC-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_i32_32x32x32_i8__vgprcd_mac_flags:
@@ -5643,20 +5443,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd(ptr addrs
 ; GCN-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; GCN-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
+; GCN-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GCN-NEXT:    v_accvgpr_write_b32 a0, s0
+; GCN-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GCN-NEXT:    v_accvgpr_write_b32 a1, s1
-; GCN-NEXT:    v_accvgpr_write_b32 a2, s2
-; GCN-NEXT:    v_accvgpr_write_b32 a3, s3
+; GCN-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; GCN-NEXT:    s_nop 1
-; GCN-NEXT:    v_mfma_f32_16x16x32_bf16 a[0:3], v[0:3], v[4:7], a[0:3]
+; GCN-NEXT:    v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11]
 ; GCN-NEXT:    s_nop 7
-; GCN-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; GCN-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; GCN-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd:
@@ -5664,20 +5462,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd(ptr addrs
 ; HEURRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; HEURRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; HEURRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; HEURRC-NEXT:    v_mov_b32_e32 v8, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v12, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s0
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_f32_16x16x32_bf16 a[0:3], v[0:3], v[4:7], a[0:3]
+; HEURRC-NEXT:    v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11]
 ; HEURRC-NEXT:    s_nop 7
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; HEURRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd:
@@ -5747,20 +5543,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags(pt
 ; GCN-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; GCN-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
+; GCN-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GCN-NEXT:    v_accvgpr_write_b32 a0, s0
+; GCN-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GCN-NEXT:    v_accvgpr_write_b32 a1, s1
-; GCN-NEXT:    v_accvgpr_write_b32 a2, s2
-; GCN-NEXT:    v_accvgpr_write_b32 a3, s3
+; GCN-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; GCN-NEXT:    s_nop 1
-; GCN-NEXT:    v_mfma_f32_16x16x32_bf16 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:3 abid:2 blgp:1
+; GCN-NEXT:    v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; GCN-NEXT:    s_nop 7
-; GCN-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; GCN-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; GCN-NEXT:    s_endpgm
 ;
 ; HEURRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags:
@@ -5768,20 +5562,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags(pt
 ; HEURRC-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
 ; HEURRC-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; HEURRC-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; HEURRC-NEXT:    v_mov_b32_e32 v8, 0
+; HEURRC-NEXT:    v_mov_b32_e32 v12, 0
 ; HEURRC-NEXT:    s_waitcnt lgkmcnt(0)
 ; HEURRC-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a0, s0
+; HEURRC-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
 ; HEURRC-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; HEURRC-NEXT:    v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT:    v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT:    v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
 ; HEURRC-NEXT:    s_nop 1
-; HEURRC-NEXT:    v_mfma_f32_16x16x32_bf16 a[0:3], v[0:3], v[4:7], a[0:3] cbsz:3 abid:2 blgp:1
+; HEURRC-NEXT:    v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1
 ; HEURRC-NEXT:    s_nop 7
-; HEURRC-NEXT:    global_store_dwordx4 v8, a[0:3], s[6:7]
+; HEURRC-NEXT:    global_store_dwordx4 v12, v[0:3], s[6:7]
 ; HEURRC-NEXT:    s_endpgm
 ;
 ; VGPRRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags:
@@ -5845,5 +5637,5 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags(pt
   ret void
 }
 
-attributes #0 = { "amdgpu-flat-work-group-size"="512,512" }
+attributes #0 = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-agpr-alloc"="0,0" }
 attributes #1 = { "amdgpu-flat-work-group-size"="1,64" }
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.ll
index 37809da10241b..f78ea92b4840b 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.ll
@@ -1895,36 +1895,36 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd(<8 x i32
 ; SDAG-LABEL: test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd:
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x0
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v20, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x40
-; SDAG-NEXT:    v_mov_b32_e32 v10, s16
-; SDAG-NEXT:    v_mov_b32_e32 v11, s17
-; SDAG-NEXT:    v_mov_b32_e32 v12, s18
-; SDAG-NEXT:    v_mov_b32_e32 v13, s19
-; SDAG-NEXT:    v_mov_b32_e32 v14, s20
-; SDAG-NEXT:    v_mov_b32_e32 v15, s21
-; SDAG-NEXT:    v_mov_b32_e32 v16, s22
-; SDAG-NEXT:    v_mov_b32_e32 v17, s23
+; SDAG-NEXT:    v_mov_b32_e32 v8, s16
+; SDAG-NEXT:    v_mov_b32_e32 v9, s17
+; SDAG-NEXT:    v_mov_b32_e32 v10, s18
+; SDAG-NEXT:    v_mov_b32_e32 v11, s19
+; SDAG-NEXT:    v_mov_b32_e32 v12, s20
+; SDAG-NEXT:    v_mov_b32_e32 v13, s21
+; SDAG-NEXT:    v_mov_b32_e32 v14, s22
+; SDAG-NEXT:    v_mov_b32_e32 v15, s23
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s8
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s11
-; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    v_mov_b32_e32 v21, s13
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[2:9], v[10:17], a[0:3], s12, v1 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
+; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], s12, v21 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 3
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[14:15]
+; SDAG-NEXT:    global_store_dwordx4 v20, v[0:3], s[14:15]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd:
@@ -1937,20 +1937,18 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd(<8 x i32
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
 ; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s24
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[24:25]
 ; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
 ; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s25
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s26
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s27
-; GISEL-NEXT:    v_mov_b32_e32 v16, s29
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[26:27]
+; GISEL-NEXT:    v_mov_b32_e32 v20, s29
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[0:7], v[8:15], a[0:3], s28, v16 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], s28, v20 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[30:31]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[30:31]
 ; GISEL-NEXT:    s_endpgm
   %result = call <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 0, i32 2, i32 3, i32 %scale0, i32 1, i32 %scale1)
   store <4 x float> %result, ptr addrspace(1) %ptr, align 16
@@ -1964,40 +1962,38 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x40
 ; SDAG-NEXT:    s_movk_i32 s6, 0x41
 ; SDAG-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v20, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
-; SDAG-NEXT:    v_mov_b32_e32 v10, s16
-; SDAG-NEXT:    v_mov_b32_e32 v11, s17
-; SDAG-NEXT:    v_mov_b32_e32 v12, s18
-; SDAG-NEXT:    v_mov_b32_e32 v13, s19
-; SDAG-NEXT:    v_mov_b32_e32 v14, s20
-; SDAG-NEXT:    v_mov_b32_e32 v15, s21
-; SDAG-NEXT:    v_mov_b32_e32 v16, s22
-; SDAG-NEXT:    v_mov_b32_e32 v17, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
+; SDAG-NEXT:    v_mov_b32_e32 v8, s16
+; SDAG-NEXT:    v_mov_b32_e32 v9, s17
+; SDAG-NEXT:    v_mov_b32_e32 v10, s18
+; SDAG-NEXT:    v_mov_b32_e32 v11, s19
+; SDAG-NEXT:    v_mov_b32_e32 v12, s20
+; SDAG-NEXT:    v_mov_b32_e32 v13, s21
+; SDAG-NEXT:    v_mov_b32_e32 v14, s22
+; SDAG-NEXT:    v_mov_b32_e32 v15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[2:9], v[10:17], a[0:3], s6, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], s6, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 3
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; SDAG-NEXT:    global_store_dwordx4 v20, v[0:3], s[4:5]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA_kimm__scaleB__inlineimm:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x0
 ; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x40
-; GISEL-NEXT:    v_mov_b32_e32 v16, 0x41
+; GISEL-NEXT:    v_mov_b32_e32 v20, 0x41
 ; GISEL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
@@ -2005,19 +2001,17 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
 ; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
 ; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[0:7], v[8:15], a[0:3], v16, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], v20, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[4:5]
 ; GISEL-NEXT:    s_endpgm
   %result = call <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 0, i32 0, i32 3, i32 65, i32 1, i32 -2)
   store <4 x float> %result, ptr addrspace(1) %ptr, align 16
@@ -2031,40 +2025,38 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x40
 ; SDAG-NEXT:    s_movk_i32 s6, 0x41
 ; SDAG-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v20, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
-; SDAG-NEXT:    v_mov_b32_e32 v10, s16
-; SDAG-NEXT:    v_mov_b32_e32 v11, s17
-; SDAG-NEXT:    v_mov_b32_e32 v12, s18
-; SDAG-NEXT:    v_mov_b32_e32 v13, s19
-; SDAG-NEXT:    v_mov_b32_e32 v14, s20
-; SDAG-NEXT:    v_mov_b32_e32 v15, s21
-; SDAG-NEXT:    v_mov_b32_e32 v16, s22
-; SDAG-NEXT:    v_mov_b32_e32 v17, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
+; SDAG-NEXT:    v_mov_b32_e32 v8, s16
+; SDAG-NEXT:    v_mov_b32_e32 v9, s17
+; SDAG-NEXT:    v_mov_b32_e32 v10, s18
+; SDAG-NEXT:    v_mov_b32_e32 v11, s19
+; SDAG-NEXT:    v_mov_b32_e32 v12, s20
+; SDAG-NEXT:    v_mov_b32_e32 v13, s21
+; SDAG-NEXT:    v_mov_b32_e32 v14, s22
+; SDAG-NEXT:    v_mov_b32_e32 v15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[2:9], v[10:17], a[0:3], s6, 1.0 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], s6, 1.0 op_sel:[1,1,0] op_sel_hi:[1,0,0]
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 3
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; SDAG-NEXT:    global_store_dwordx4 v20, v[0:3], s[4:5]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA_kimm__scaleB__FP_literal:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x0
 ; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x40
-; GISEL-NEXT:    v_mov_b32_e32 v16, 0x41
+; GISEL-NEXT:    v_mov_b32_e32 v20, 0x41
 ; GISEL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
@@ -2072,19 +2064,17 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
 ; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
 ; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[0:7], v[8:15], a[0:3], v16, 1.0 op_sel:[1,1,0] op_sel_hi:[1,0,0]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], v20, 1.0 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[4:5]
 ; GISEL-NEXT:    s_endpgm
   %result = call <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 0, i32 0, i32 3, i32 65, i32 1, i32 1065353216)
   store <4 x float> %result, ptr addrspace(1) %ptr, align 16
@@ -2096,34 +2086,32 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x0
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x40
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v20, 0
 ; SDAG-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
-; SDAG-NEXT:    v_mov_b32_e32 v10, s16
-; SDAG-NEXT:    v_mov_b32_e32 v11, s17
-; SDAG-NEXT:    v_mov_b32_e32 v12, s18
-; SDAG-NEXT:    v_mov_b32_e32 v13, s19
-; SDAG-NEXT:    v_mov_b32_e32 v14, s20
-; SDAG-NEXT:    v_mov_b32_e32 v15, s21
-; SDAG-NEXT:    v_mov_b32_e32 v16, s22
-; SDAG-NEXT:    v_mov_b32_e32 v17, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
+; SDAG-NEXT:    v_mov_b32_e32 v8, s16
+; SDAG-NEXT:    v_mov_b32_e32 v9, s17
+; SDAG-NEXT:    v_mov_b32_e32 v10, s18
+; SDAG-NEXT:    v_mov_b32_e32 v11, s19
+; SDAG-NEXT:    v_mov_b32_e32 v12, s20
+; SDAG-NEXT:    v_mov_b32_e32 v13, s21
+; SDAG-NEXT:    v_mov_b32_e32 v14, s22
+; SDAG-NEXT:    v_mov_b32_e32 v15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[2:9], v[10:17], a[0:3], 1.0, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], 1.0, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 3
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; SDAG-NEXT:    global_store_dwordx4 v20, v[0:3], s[4:5]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA_FP_literal__scaleB__inline_imm:
@@ -2136,21 +2124,19 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
 ; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
 ; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; GISEL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
 ; GISEL-NEXT:    s_nop 0
-; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[0:7], v[8:15], a[0:3], 1.0, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], 1.0, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[4:5]
 ; GISEL-NEXT:    s_endpgm
   %result = call <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 0, i32 0, i32 3, i32 1065353216, i32 1, i32 -2)
   store <4 x float> %result, ptr addrspace(1) %ptr, align 16
@@ -2162,34 +2148,32 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x0
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x40
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v20, 0
 ; SDAG-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s0
-; SDAG-NEXT:    v_mov_b32_e32 v10, s16
-; SDAG-NEXT:    v_mov_b32_e32 v11, s17
-; SDAG-NEXT:    v_mov_b32_e32 v12, s18
-; SDAG-NEXT:    v_mov_b32_e32 v13, s19
-; SDAG-NEXT:    v_mov_b32_e32 v14, s20
-; SDAG-NEXT:    v_mov_b32_e32 v15, s21
-; SDAG-NEXT:    v_mov_b32_e32 v16, s22
-; SDAG-NEXT:    v_mov_b32_e32 v17, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT:    v_mov_b32_e32 v0, s8
+; SDAG-NEXT:    v_mov_b32_e32 v1, s9
+; SDAG-NEXT:    v_mov_b32_e32 v2, s10
+; SDAG-NEXT:    v_mov_b32_e32 v3, s11
+; SDAG-NEXT:    v_mov_b32_e32 v4, s12
+; SDAG-NEXT:    v_mov_b32_e32 v5, s13
+; SDAG-NEXT:    v_mov_b32_e32 v6, s14
+; SDAG-NEXT:    v_mov_b32_e32 v7, s15
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
+; SDAG-NEXT:    v_mov_b32_e32 v8, s16
+; SDAG-NEXT:    v_mov_b32_e32 v9, s17
+; SDAG-NEXT:    v_mov_b32_e32 v10, s18
+; SDAG-NEXT:    v_mov_b32_e32 v11, s19
+; SDAG-NEXT:    v_mov_b32_e32 v12, s20
+; SDAG-NEXT:    v_mov_b32_e32 v13, s21
+; SDAG-NEXT:    v_mov_b32_e32 v14, s22
+; SDAG-NEXT:    v_mov_b32_e32 v15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[2:9], v[10:17], a[0:3], 1.0, 0.15915494 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; SDAG-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], 1.0, 0.15915494 op_sel:[1,1,0] op_sel_hi:[1,0,0]
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 3
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; SDAG-NEXT:    global_store_dwordx4 v20, v[0:3], s[4:5]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA_FP_literal__scaleB__FP_literal:
@@ -2202,21 +2186,19 @@ define amdgpu_kernel void @test_mfma_scale_f32_16x16x128_f8f6f4__vgprcd___scaleA
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
 ; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[2:3]
 ; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
 ; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[0:1]
 ; GISEL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x50
 ; GISEL-NEXT:    s_nop 0
-; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 a[0:3], v[0:7], v[8:15], a[0:3], 1.0, 0.15915494 op_sel:[1,1,0] op_sel_hi:[1,0,0]
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_16x16x128_f8f6f4 v[0:3], v[0:7], v[8:15], v[16:19], 1.0, 0.15915494 op_sel:[1,1,0] op_sel_hi:[1,0,0]
+; GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[4:5]
+; GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[4:5]
 ; GISEL-NEXT:    s_endpgm
   %result = call <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 0, i32 0, i32 3, i32 1065353216, i32 1, i32 1042479491)
   store <4 x float> %result, ptr addrspace(1) %ptr, align 16
@@ -2559,5 +2541,5 @@ declare <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v6i32.v8i32(<6
 declare <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v4i32(<8 x i32>, <4 x i32>, <4 x float>, i32 immarg, i32 immarg, i32 immarg, i32, i32 immarg, i32) #1
 declare <4 x float> @llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.v8i32.v6i32(<8 x i32>, <6 x i32>, <4 x float>, i32 immarg, i32 immarg, i32 immarg, i32, i32 immarg, i32) #1
 
-attributes #0 = { "amdgpu-flat-work-group-size"="512,512" }
+attributes #0 = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-agpr-alloc"="0,0" }
 attributes #1 = { convergent nocallback nofree nosync nounwind willreturn memory(none) }
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.ll
index bc50058778dbf..0b2818f38149d 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.ll
@@ -4539,49 +4539,41 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4__vgprcd(<8 x i32>
 ; SDAG-NEXT:    s_load_dwordx16 s[36:51], s[4:5], 0x40
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x80
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s36
-; SDAG-NEXT:    v_mov_b32_e32 v2, s8
-; SDAG-NEXT:    v_mov_b32_e32 v3, s9
-; SDAG-NEXT:    v_mov_b32_e32 v4, s10
-; SDAG-NEXT:    v_mov_b32_e32 v5, s11
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_mov_b32_e32 v10, s16
-; SDAG-NEXT:    v_mov_b32_e32 v11, s17
-; SDAG-NEXT:    v_mov_b32_e32 v12, s18
-; SDAG-NEXT:    v_mov_b32_e32 v13, s19
-; SDAG-NEXT:    v_mov_b32_e32 v14, s20
-; SDAG-NEXT:    v_mov_b32_e32 v15, s21
-; SDAG-NEXT:    v_mov_b32_e32 v16, s22
-; SDAG-NEXT:    v_mov_b32_e32 v17, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s37
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s38
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s39
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s40
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s41
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s42
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s43
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s44
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s45
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s46
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s47
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s48
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s49
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s50
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s51
-; SDAG-NEXT:    v_mov_b32_e32 v0, s1
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[36:37]
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    v_mov_b32_e32 v20, s12
+; SDAG-NEXT:    v_mov_b32_e32 v21, s13
+; SDAG-NEXT:    v_mov_b32_e32 v22, s14
+; SDAG-NEXT:    v_mov_b32_e32 v23, s15
+; SDAG-NEXT:    v_mov_b32_e32 v24, s16
+; SDAG-NEXT:    v_mov_b32_e32 v25, s17
+; SDAG-NEXT:    v_mov_b32_e32 v26, s18
+; SDAG-NEXT:    v_mov_b32_e32 v27, s19
+; SDAG-NEXT:    v_mov_b32_e32 v28, s20
+; SDAG-NEXT:    v_mov_b32_e32 v29, s21
+; SDAG-NEXT:    v_mov_b32_e32 v30, s22
+; SDAG-NEXT:    v_mov_b32_e32 v31, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[38:39]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[40:41]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[42:43]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[44:45]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[46:47]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[48:49]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[50:51]
+; SDAG-NEXT:    v_mov_b32_e32 v32, s1
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[2:9], v[10:17], a[0:15], s0, v0 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[0:15], s0, v32 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[2:3] offset:48
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[2:3] offset:32
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[2:3] offset:16
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[2:3]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[2:3] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[2:3] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[2:3] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[2:3]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_32x32x64_f8f6f4__vgprcd:
@@ -4590,41 +4582,33 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4__vgprcd(<8 x i32>
 ; GISEL-NEXT:    s_load_dwordx16 s[36:51], s[4:5], 0x40
 ; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x80
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s36
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s37
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s38
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s39
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s40
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s41
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s42
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s43
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s44
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s45
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s46
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s47
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s48
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s49
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s50
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s51
-; GISEL-NEXT:    v_mov_b32_e32 v16, s1
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[36:37]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[38:39]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[40:41]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[42:43]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[44:45]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[46:47]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[48:49]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[50:51]
+; GISEL-NEXT:    v_mov_b32_e32 v32, s1
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[0:15], s0, v16 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[0:15], s0, v32 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[2:3]
-; GISEL-NEXT:    global_store_dwordx4 v0, a[4:7], s[2:3] offset:16
-; GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[2:3] offset:32
-; GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[2:3] offset:48
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[2:3]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[2:3] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[2:3] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[2:3] offset:48
 ; GISEL-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 0, i32 2, i32 3, i32 %scale0, i32 1, i32 %scale1)
   store <16 x float> %result, ptr addrspace(1) %ptr, align 64
@@ -4639,91 +4623,75 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4__vgprcd___scaleA_
 ; SDAG-NEXT:    s_movk_i32 s2, 0x41
 ; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x80
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v0, s8
-; SDAG-NEXT:    v_mov_b32_e32 v1, s9
-; SDAG-NEXT:    v_mov_b32_e32 v2, s10
-; SDAG-NEXT:    v_mov_b32_e32 v3, s11
-; SDAG-NEXT:    v_mov_b32_e32 v4, s12
-; SDAG-NEXT:    v_mov_b32_e32 v5, s13
-; SDAG-NEXT:    v_mov_b32_e32 v6, s14
-; SDAG-NEXT:    v_mov_b32_e32 v7, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s36
-; SDAG-NEXT:    v_mov_b32_e32 v8, s16
-; SDAG-NEXT:    v_mov_b32_e32 v9, s17
-; SDAG-NEXT:    v_mov_b32_e32 v10, s18
-; SDAG-NEXT:    v_mov_b32_e32 v11, s19
-; SDAG-NEXT:    v_mov_b32_e32 v12, s20
-; SDAG-NEXT:    v_mov_b32_e32 v13, s21
-; SDAG-NEXT:    v_mov_b32_e32 v14, s22
-; SDAG-NEXT:    v_mov_b32_e32 v15, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s37
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s38
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s39
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s40
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s41
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s42
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s43
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s44
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s45
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s46
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s47
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s48
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s49
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s50
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s51
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    v_mov_b32_e32 v20, s12
+; SDAG-NEXT:    v_mov_b32_e32 v21, s13
+; SDAG-NEXT:    v_mov_b32_e32 v22, s14
+; SDAG-NEXT:    v_mov_b32_e32 v23, s15
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[36:37]
+; SDAG-NEXT:    v_mov_b32_e32 v24, s16
+; SDAG-NEXT:    v_mov_b32_e32 v25, s17
+; SDAG-NEXT:    v_mov_b32_e32 v26, s18
+; SDAG-NEXT:    v_mov_b32_e32 v27, s19
+; SDAG-NEXT:    v_mov_b32_e32 v28, s20
+; SDAG-NEXT:    v_mov_b32_e32 v29, s21
+; SDAG-NEXT:    v_mov_b32_e32 v30, s22
+; SDAG-NEXT:    v_mov_b32_e32 v31, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[38:39]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[40:41]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[42:43]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[44:45]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[46:47]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[48:49]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[50:51]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[0:15], s2, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[0:15], s2, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
-; SDAG-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; SDAG-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; SDAG-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; SDAG-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; SDAG-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; SDAG-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_mfma_scale_f32_32x32x64_f8f6f4__vgprcd___scaleA_kimm__scaleB__inlineimm:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x0
 ; GISEL-NEXT:    s_load_dwordx16 s[36:51], s[4:5], 0x40
-; GISEL-NEXT:    v_mov_b32_e32 v16, 0x41
+; GISEL-NEXT:    v_mov_b32_e32 v32, 0x41
 ; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x80
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s36
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s37
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s38
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s39
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s40
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s41
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s42
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s43
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s44
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s45
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s46
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s47
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s48
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s49
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s50
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s51
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[36:37]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[38:39]
+; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[40:41]
+; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[42:43]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[44:45]
+; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[46:47]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[48:49]
+; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[50:51]
 ; GISEL-NEXT:    s_nop 1
-; GISEL-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[0:15], v16, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
-; GISEL-NEXT:    v_mov_b32_e32 v0, 0
+; GISEL-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[0:15], v32, -2 op_sel:[1,1,0] op_sel_hi:[1,0,0] blgp:2
+; GISEL-NEXT:    v_mov_b32_e32 v16, 0
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 7
 ; GISEL-NEXT:    s_nop 2
-; GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[0:1]
-; GISEL-NEXT:    global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
+; GISEL-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GISEL-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GISEL-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GISEL-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
 ; GISEL-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 0, i32 2, i32 3, i32 65, i32 1, i32 -2)
   store <16 x float> %result, ptr addrspace(1) %ptr, align 64
@@ -5031,77 +4999,72 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4_0_0__vgprcd_nonma
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx16 s[12:27], s[4:5], 0x0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v0, s12
-; SDAG-NEXT:    v_mov_b32_e32 v1, s13
-; SDAG-NEXT:    v_mov_b32_e32 v2, s14
-; SDAG-NEXT:    v_mov_b32_e32 v3, s15
-; SDAG-NEXT:    v_mov_b32_e32 v4, s16
-; SDAG-NEXT:    v_mov_b32_e32 v5, s17
-; SDAG-NEXT:    v_mov_b32_e32 v6, s18
-; SDAG-NEXT:    v_mov_b32_e32 v7, s19
-; SDAG-NEXT:    v_mov_b32_e32 v8, s20
-; SDAG-NEXT:    v_mov_b32_e32 v9, s21
-; SDAG-NEXT:    v_mov_b32_e32 v10, s22
-; SDAG-NEXT:    v_mov_b32_e32 v11, s23
+; SDAG-NEXT:    v_mov_b32_e32 v32, s12
+; SDAG-NEXT:    v_mov_b32_e32 v33, s13
+; SDAG-NEXT:    v_mov_b32_e32 v34, s14
+; SDAG-NEXT:    v_mov_b32_e32 v35, s15
+; SDAG-NEXT:    v_mov_b32_e32 v36, s16
+; SDAG-NEXT:    v_mov_b32_e32 v37, s17
+; SDAG-NEXT:    v_mov_b32_e32 v38, s18
+; SDAG-NEXT:    v_mov_b32_e32 v39, s19
+; SDAG-NEXT:    v_mov_b32_e32 v40, s20
+; SDAG-NEXT:    v_mov_b32_e32 v41, s21
+; SDAG-NEXT:    v_mov_b32_e32 v42, s22
+; SDAG-NEXT:    v_mov_b32_e32 v43, s23
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x40
-; SDAG-NEXT:    v_mov_b32_e32 v12, s24
-; SDAG-NEXT:    v_mov_b32_e32 v13, s25
-; SDAG-NEXT:    v_mov_b32_e32 v14, s26
-; SDAG-NEXT:    v_mov_b32_e32 v15, s27
+; SDAG-NEXT:    v_mov_b32_e32 v44, s24
+; SDAG-NEXT:    v_mov_b32_e32 v45, s25
+; SDAG-NEXT:    v_mov_b32_e32 v46, s26
+; SDAG-NEXT:    v_mov_b32_e32 v47, s27
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a31, s23
-; SDAG-NEXT:    v_accvgpr_write_b32 a30, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a29, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a28, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a27, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a26, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a25, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a24, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a23, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a22, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a21, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a20, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a19, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a18, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a17, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a16, s8
+; SDAG-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; SDAG-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[16:31] blgp:2
-; SDAG-NEXT:    v_mov_b32_e32 v2, s20
-; SDAG-NEXT:    v_mov_b32_e32 v3, s21
-; SDAG-NEXT:    v_mov_b32_e32 v4, s22
-; SDAG-NEXT:    v_mov_b32_e32 v5, s23
-; SDAG-NEXT:    v_mov_b64_e32 v[0:1], 48
-; SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off sc0 sc1
+; SDAG-NEXT:    v_mfma_f32_32x32x64_f8f6f4 v[0:15], v[32:39], v[40:47], v[16:31] blgp:2
+; SDAG-NEXT:    s_nop 7
+; SDAG-NEXT:    s_nop 6
+; SDAG-NEXT:    v_mov_b32_e32 v16, s20
+; SDAG-NEXT:    v_mov_b32_e32 v17, s21
+; SDAG-NEXT:    v_mov_b32_e32 v18, s22
+; SDAG-NEXT:    v_mov_b32_e32 v19, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], 48
+; SDAG-NEXT:    global_store_dwordx4 v[20:21], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s18
-; SDAG-NEXT:    v_mov_b32_e32 v7, s19
-; SDAG-NEXT:    v_mov_b32_e32 v4, s16
-; SDAG-NEXT:    v_mov_b32_e32 v5, s17
-; SDAG-NEXT:    v_mov_b64_e32 v[2:3], 32
-; SDAG-NEXT:    global_store_dwordx4 v[2:3], v[4:7], off sc0 sc1
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], 32
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], 16
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v[22:23], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b64_e32 v[4:5], 16
-; SDAG-NEXT:    global_store_dwordx4 v[4:5], v[6:9], off sc0 sc1
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v[24:25], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v10, s10
-; SDAG-NEXT:    v_mov_b32_e32 v11, s11
-; SDAG-NEXT:    v_mov_b32_e32 v8, s8
-; SDAG-NEXT:    v_mov_b32_e32 v9, s9
-; SDAG-NEXT:    v_mov_b64_e32 v[6:7], 0
-; SDAG-NEXT:    global_store_dwordx4 v[6:7], v[8:11], off sc0 sc1
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v[26:27], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[2:3], a[8:11], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[22:23], v[8:11], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[0:1], a[12:15], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[20:21], v[12:15], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[6:7], a[0:3], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[26:27], v[0:3], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[4:5], a[4:7], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[24:25], v[4:7], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_endpgm
 ;
@@ -5109,61 +5072,45 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4_0_0__vgprcd_nonma
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_load_dwordx16 s[36:51], s[4:5], 0x0
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x40
-; GISEL-NEXT:    v_mov_b64_e32 v[16:17], 0
-; GISEL-NEXT:    v_mov_b64_e32 v[18:19], 16
-; GISEL-NEXT:    v_mov_b64_e32 v[20:21], 32
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[36:37]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[38:39]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[40:41]
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[42:43]
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[44:45]
-; GISEL-NEXT:    v_accvgpr_write_b32 a31, s23
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[46:47]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[48:49]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[50:51]
-; GISEL-NEXT:    v_accvgpr_write_b32 a30, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a29, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a28, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a27, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a26, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a25, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a24, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a23, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a22, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a21, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a20, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a19, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a18, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a17, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a16, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[22:23], 48
-; GISEL-NEXT:    s_nop 0
-; GISEL-NEXT:    v_mfma_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[16:31] blgp:2
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    global_store_dwordx4 v[16:17], v[0:3], off sc0 sc1
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], s[36:37]
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], s[38:39]
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], s[40:41]
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], s[42:43]
+; GISEL-NEXT:    v_mov_b64_e32 v[40:41], s[44:45]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[42:43], s[46:47]
+; GISEL-NEXT:    v_mov_b64_e32 v[44:45], s[48:49]
+; GISEL-NEXT:    v_mov_b64_e32 v[46:47], s[50:51]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GISEL-NEXT:    s_nop 1
+; GISEL-NEXT:    v_mfma_f32_32x32x64_f8f6f4 v[0:15], v[32:39], v[40:47], v[16:31] blgp:2
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], 0
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], 16
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], 32
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], 48
+; GISEL-NEXT:    global_store_dwordx4 v[32:33], v[16:19], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[18:19], v[4:7], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[34:35], v[20:23], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[20:21], v[8:11], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[36:37], v[24:27], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[22:23], v[12:15], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[38:39], v[28:31], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    s_nop 3
-; GISEL-NEXT:    global_store_dwordx4 v[16:17], a[0:3], off sc0 sc1
+; GISEL-NEXT:    s_nop 7
+; GISEL-NEXT:    global_store_dwordx4 v[32:33], v[0:3], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[18:19], a[4:7], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[34:35], v[4:7], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[20:21], a[8:11], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[36:37], v[8:11], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[22:23], a[12:15], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[38:39], v[12:15], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 0, i32 2, i32 0, i32 0, i32 0, i32 0)
@@ -5177,77 +5124,70 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4_25_42__vgprcd_non
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_load_dwordx16 s[12:27], s[4:5], 0x0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v0, s12
-; SDAG-NEXT:    v_mov_b32_e32 v1, s13
-; SDAG-NEXT:    v_mov_b32_e32 v2, s14
-; SDAG-NEXT:    v_mov_b32_e32 v3, s15
-; SDAG-NEXT:    v_mov_b32_e32 v4, s16
-; SDAG-NEXT:    v_mov_b32_e32 v5, s17
-; SDAG-NEXT:    v_mov_b32_e32 v6, s18
-; SDAG-NEXT:    v_mov_b32_e32 v7, s19
-; SDAG-NEXT:    v_mov_b32_e32 v8, s20
-; SDAG-NEXT:    v_mov_b32_e32 v9, s21
-; SDAG-NEXT:    v_mov_b32_e32 v10, s22
-; SDAG-NEXT:    v_mov_b32_e32 v11, s23
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    v_mov_b32_e32 v20, s16
+; SDAG-NEXT:    v_mov_b32_e32 v21, s17
+; SDAG-NEXT:    v_mov_b32_e32 v22, s18
+; SDAG-NEXT:    v_mov_b32_e32 v23, s19
+; SDAG-NEXT:    v_mov_b32_e32 v24, s20
+; SDAG-NEXT:    v_mov_b32_e32 v25, s21
+; SDAG-NEXT:    v_mov_b32_e32 v26, s22
+; SDAG-NEXT:    v_mov_b32_e32 v27, s23
 ; SDAG-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x40
-; SDAG-NEXT:    v_mov_b32_e32 v12, s24
-; SDAG-NEXT:    v_mov_b32_e32 v13, s25
-; SDAG-NEXT:    v_mov_b32_e32 v14, s26
-; SDAG-NEXT:    v_mov_b32_e32 v15, s27
+; SDAG-NEXT:    v_mov_b32_e32 v28, s24
+; SDAG-NEXT:    v_mov_b32_e32 v29, s25
+; SDAG-NEXT:    v_mov_b32_e32 v30, s26
+; SDAG-NEXT:    v_mov_b32_e32 v31, s27
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_accvgpr_write_b32 a0, s8
-; SDAG-NEXT:    v_accvgpr_write_b32 a1, s9
-; SDAG-NEXT:    v_accvgpr_write_b32 a2, s10
-; SDAG-NEXT:    v_accvgpr_write_b32 a3, s11
-; SDAG-NEXT:    v_accvgpr_write_b32 a4, s12
-; SDAG-NEXT:    v_accvgpr_write_b32 a5, s13
-; SDAG-NEXT:    v_accvgpr_write_b32 a6, s14
-; SDAG-NEXT:    v_accvgpr_write_b32 a7, s15
-; SDAG-NEXT:    v_accvgpr_write_b32 a8, s16
-; SDAG-NEXT:    v_accvgpr_write_b32 a9, s17
-; SDAG-NEXT:    v_accvgpr_write_b32 a10, s18
-; SDAG-NEXT:    v_accvgpr_write_b32 a11, s19
-; SDAG-NEXT:    v_accvgpr_write_b32 a12, s20
-; SDAG-NEXT:    v_accvgpr_write_b32 a13, s21
-; SDAG-NEXT:    v_accvgpr_write_b32 a14, s22
-; SDAG-NEXT:    v_accvgpr_write_b32 a15, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
+; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
+; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
+; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
 ; SDAG-NEXT:    s_nop 1
-; SDAG-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[0:15], 25, 42 op_sel_hi:[0,0,0] blgp:2
-; SDAG-NEXT:    v_mov_b32_e32 v2, s20
-; SDAG-NEXT:    v_mov_b32_e32 v3, s21
-; SDAG-NEXT:    v_mov_b32_e32 v4, s22
-; SDAG-NEXT:    v_mov_b32_e32 v5, s23
-; SDAG-NEXT:    v_mov_b64_e32 v[0:1], 48
-; SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off sc0 sc1
+; SDAG-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[0:15], 25, 42 op_sel_hi:[0,0,0] blgp:2
+; SDAG-NEXT:    v_mov_b32_e32 v16, s20
+; SDAG-NEXT:    v_mov_b32_e32 v17, s21
+; SDAG-NEXT:    v_mov_b32_e32 v18, s22
+; SDAG-NEXT:    v_mov_b32_e32 v19, s23
+; SDAG-NEXT:    v_mov_b64_e32 v[20:21], 48
+; SDAG-NEXT:    global_store_dwordx4 v[20:21], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s18
-; SDAG-NEXT:    v_mov_b32_e32 v7, s19
-; SDAG-NEXT:    v_mov_b32_e32 v4, s16
-; SDAG-NEXT:    v_mov_b32_e32 v5, s17
-; SDAG-NEXT:    v_mov_b64_e32 v[2:3], 32
-; SDAG-NEXT:    global_store_dwordx4 v[2:3], v[4:7], off sc0 sc1
+; SDAG-NEXT:    v_mov_b64_e32 v[22:23], 32
+; SDAG-NEXT:    v_mov_b64_e32 v[24:25], 16
+; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s17
+; SDAG-NEXT:    v_mov_b32_e32 v18, s18
+; SDAG-NEXT:    v_mov_b32_e32 v19, s19
+; SDAG-NEXT:    global_store_dwordx4 v[22:23], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v8, s14
-; SDAG-NEXT:    v_mov_b32_e32 v9, s15
-; SDAG-NEXT:    v_mov_b32_e32 v6, s12
-; SDAG-NEXT:    v_mov_b32_e32 v7, s13
-; SDAG-NEXT:    v_mov_b64_e32 v[4:5], 16
-; SDAG-NEXT:    global_store_dwordx4 v[4:5], v[6:9], off sc0 sc1
+; SDAG-NEXT:    v_mov_b64_e32 v[26:27], 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    global_store_dwordx4 v[24:25], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v10, s10
-; SDAG-NEXT:    v_mov_b32_e32 v11, s11
-; SDAG-NEXT:    v_mov_b32_e32 v8, s8
-; SDAG-NEXT:    v_mov_b32_e32 v9, s9
-; SDAG-NEXT:    v_mov_b64_e32 v[6:7], 0
-; SDAG-NEXT:    global_store_dwordx4 v[6:7], v[8:11], off sc0 sc1
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, s8
+; SDAG-NEXT:    v_mov_b32_e32 v17, s9
+; SDAG-NEXT:    v_mov_b32_e32 v18, s10
+; SDAG-NEXT:    v_mov_b32_e32 v19, s11
+; SDAG-NEXT:    global_store_dwordx4 v[26:27], v[16:19], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[2:3], a[8:11], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[22:23], v[8:11], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[0:1], a[12:15], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[20:21], v[12:15], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[6:7], a[0:3], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[26:27], v[0:3], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
-; SDAG-NEXT:    global_store_dwordx4 v[4:5], a[4:7], off sc0 sc1
+; SDAG-NEXT:    global_store_dwordx4 v[24:25], v[4:7], off sc0 sc1
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_endpgm
 ;
@@ -5255,61 +5195,53 @@ define amdgpu_kernel void @test_mfma_scale_f32_32x32x64_f8f6f4_25_42__vgprcd_non
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_load_dwordx16 s[36:51], s[4:5], 0x0
 ; GISEL-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x40
-; GISEL-NEXT:    v_mov_b64_e32 v[16:17], 0
-; GISEL-NEXT:    v_mov_b64_e32 v[18:19], 16
-; GISEL-NEXT:    v_mov_b64_e32 v[20:21], 32
+; GISEL-NEXT:    v_mov_b64_e32 v[32:33], 0
+; GISEL-NEXT:    v_mov_b64_e32 v[34:35], 16
+; GISEL-NEXT:    v_mov_b64_e32 v[36:37], 32
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[36:37]
-; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[38:39]
-; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[40:41]
-; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[42:43]
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[44:45]
-; GISEL-NEXT:    v_accvgpr_write_b32 a0, s8
-; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[46:47]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[48:49]
-; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[50:51]
-; GISEL-NEXT:    v_accvgpr_write_b32 a1, s9
-; GISEL-NEXT:    v_accvgpr_write_b32 a2, s10
-; GISEL-NEXT:    v_accvgpr_write_b32 a3, s11
-; GISEL-NEXT:    v_accvgpr_write_b32 a4, s12
-; GISEL-NEXT:    v_accvgpr_write_b32 a5, s13
-; GISEL-NEXT:    v_accvgpr_write_b32 a6, s14
-; GISEL-NEXT:    v_accvgpr_write_b32 a7, s15
-; GISEL-NEXT:    v_accvgpr_write_b32 a8, s16
-; GISEL-NEXT:    v_accvgpr_write_b32 a9, s17
-; GISEL-NEXT:    v_accvgpr_write_b32 a10, s18
-; GISEL-NEXT:    v_accvgpr_write_b32 a11, s19
-; GISEL-NEXT:    v_accvgpr_write_b32 a12, s20
-; GISEL-NEXT:    v_accvgpr_write_b32 a13, s21
-; GISEL-NEXT:    v_accvgpr_write_b32 a14, s22
-; GISEL-NEXT:    v_accvgpr_write_b32 a15, s23
-; GISEL-NEXT:    v_mov_b64_e32 v[22:23], 48
-; GISEL-NEXT:    s_nop 0
-; GISEL-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[0:15], 25, 42 op_sel_hi:[0,0,0] blgp:2
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[36:37]
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[38:39]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[40:41]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[42:43]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[44:45]
 ; GISEL-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[46:47]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[48:49]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[50:51]
 ; GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
-; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[16:17]
 ; GISEL-NEXT:    v_mov_b64_e32 v[10:11], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[12:13], s[20:21]
 ; GISEL-NEXT:    v_mov_b64_e32 v[14:15], s[22:23]
-; GISEL-NEXT:    global_store_dwordx4 v[16:17], v[0:3], off sc0 sc1
+; GISEL-NEXT:    v_mov_b64_e32 v[38:39], 48
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[0:15], 25, 42 op_sel_hi:[0,0,0] blgp:2
+; GISEL-NEXT:    v_mov_b64_e32 v[18:19], s[10:11]
+; GISEL-NEXT:    v_mov_b64_e32 v[16:17], s[8:9]
+; GISEL-NEXT:    v_mov_b64_e32 v[22:23], s[14:15]
+; GISEL-NEXT:    v_mov_b64_e32 v[26:27], s[18:19]
+; GISEL-NEXT:    v_mov_b64_e32 v[30:31], s[22:23]
+; GISEL-NEXT:    v_mov_b64_e32 v[20:21], s[12:13]
+; GISEL-NEXT:    v_mov_b64_e32 v[24:25], s[16:17]
+; GISEL-NEXT:    v_mov_b64_e32 v[28:29], s[20:21]
+; GISEL-NEXT:    global_store_dwordx4 v[32:33], v[16:19], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[18:19], v[4:7], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[34:35], v[20:23], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[20:21], v[8:11], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[36:37], v[24:27], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[22:23], v[12:15], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[38:39], v[28:31], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_nop 3
-; GISEL-NEXT:    global_store_dwordx4 v[16:17], a[0:3], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[32:33], v[0:3], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[18:19], a[4:7], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[34:35], v[4:7], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[20:21], a[8:11], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[36:37], v[8:11], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GISEL-NEXT:    global_store_dwordx4 v[22:23], a[12:15], off sc0 sc1
+; GISEL-NEXT:    global_store_dwordx4 v[38:39], v[12:15], off sc0 sc1
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_endpgm
   %result = call <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 0, i32 2, i32 0, i32 25, i32 0, i32 42)
@@ -6298,6 +6230,6 @@ declare <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v6i32.v8i32(<6
 declare <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v4i32(<8 x i32>, <4 x i32>, <16 x float>, i32 immarg, i32 immarg, i32 immarg, i32, i32 immarg, i32) #2
 declare <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v6i32(<8 x i32>, <6 x i32>, <16 x float>, i32 immarg, i32 immarg, i32 immarg, i32, i32 immarg, i32) #2
 
-attributes #0 = { "amdgpu-flat-work-group-size"="512,512" }
+attributes #0 = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-agpr-alloc"="0,0" }
 attributes #1 = { "amdgpu-flat-work-group-size"="128,128" }
 attributes #2 = { convergent nocallback nofree nosync nounwind willreturn memory(none) }
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.xf32.gfx942.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.xf32.gfx942.ll
index ea9334a6c74d3..31a48de4133ac 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.xf32.gfx942.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.xf32.gfx942.ll
@@ -1,8 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GFX942,GFX942-SDAG %s
 ; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GFX942,GFX942-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 -stress-regalloc=10 < %s | FileCheck --check-prefixes=GFX942-STRESS,GFX942-SDAG-STRESS %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx942 -stress-regalloc=10 < %s | FileCheck --check-prefixes=GFX942-STRESS,GFX942-GISEL-STRESS %s
 
 declare <4 x float> @llvm.amdgcn.mfma.f32.16x16x8.xf32(<2 x float>, <2 x float>, <4 x float>, i32, i32, i32)
 declare <16 x float> @llvm.amdgcn.mfma.f32.32x32x4.xf32(<2 x float>, <2 x float>, <16 x float>, i32, i32, i32)
@@ -51,50 +49,6 @@ define amdgpu_kernel void @test_mfma_f32_16x16x8xf32(ptr addrspace(1) %arg) #0 {
 ; GFX942-GISEL-NEXT:    s_nop 5
 ; GFX942-GISEL-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
 ; GFX942-GISEL-NEXT:    s_endpgm
-;
-; GFX942-SDAG-STRESS-LABEL: test_mfma_f32_16x16x8xf32:
-; GFX942-SDAG-STRESS:       ; %bb.0: ; %bb
-; GFX942-SDAG-STRESS-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v0, 1.0
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v1, 2.0
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v2, 0x40400000
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v3, 4.0
-; GFX942-SDAG-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-SDAG-STRESS-NEXT:    s_load_dwordx4 s[0:3], s[6:7], 0x0
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v4, 0
-; GFX942-SDAG-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a0, s0
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a1, s1
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a2, s2
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a3, s3
-; GFX942-SDAG-STRESS-NEXT:    s_nop 1
-; GFX942-SDAG-STRESS-NEXT:    v_mfma_f32_16x16x8_xf32 a[0:3], v[0:1], v[2:3], a[0:3] cbsz:1 abid:2 blgp:3
-; GFX942-SDAG-STRESS-NEXT:    s_nop 6
-; GFX942-SDAG-STRESS-NEXT:    global_store_dwordx4 v4, a[0:3], s[6:7]
-; GFX942-SDAG-STRESS-NEXT:    s_endpgm
-;
-; GFX942-GISEL-STRESS-LABEL: test_mfma_f32_16x16x8xf32:
-; GFX942-GISEL-STRESS:       ; %bb.0: ; %bb
-; GFX942-GISEL-STRESS-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s0, 1.0
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s2, 0x40400000
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s1, 2.0
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s3, 4.0
-; GFX942-GISEL-STRESS-NEXT:    v_mov_b64_e32 v[0:1], s[0:1]
-; GFX942-GISEL-STRESS-NEXT:    v_mov_b64_e32 v[2:3], s[2:3]
-; GFX942-GISEL-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-STRESS-NEXT:    s_load_dwordx4 s[0:3], s[6:7], 0x0
-; GFX942-GISEL-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a0, s0
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a1, s1
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a2, s2
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a3, s3
-; GFX942-GISEL-STRESS-NEXT:    s_nop 1
-; GFX942-GISEL-STRESS-NEXT:    v_mfma_f32_16x16x8_xf32 a[0:3], v[0:1], v[2:3], a[0:3] cbsz:1 abid:2 blgp:3
-; GFX942-GISEL-STRESS-NEXT:    v_mov_b32_e32 v0, 0
-; GFX942-GISEL-STRESS-NEXT:    s_nop 5
-; GFX942-GISEL-STRESS-NEXT:    global_store_dwordx4 v0, a[0:3], s[6:7]
-; GFX942-GISEL-STRESS-NEXT:    s_endpgm
 bb:
   %in.1 = load <4 x float>, ptr addrspace(1) %arg
   %mai.1 = tail call <4 x float> @llvm.amdgcn.mfma.f32.16x16x8.xf32(<2 x float> <float 1.0, float 2.0>, <2 x float> <float 3.0, float 4.0>, <4 x float> %in.1, i32 1, i32 2, i32 3)
@@ -178,82 +132,6 @@ define amdgpu_kernel void @test_mfma_f32_32x32x4xf32(ptr addrspace(1) %arg) #0 {
 ; GFX942-GISEL-NEXT:    global_store_dwordx4 v0, a[8:11], s[16:17] offset:32
 ; GFX942-GISEL-NEXT:    global_store_dwordx4 v0, a[12:15], s[16:17] offset:48
 ; GFX942-GISEL-NEXT:    s_endpgm
-;
-; GFX942-SDAG-STRESS-LABEL: test_mfma_f32_32x32x4xf32:
-; GFX942-SDAG-STRESS:       ; %bb.0: ; %bb
-; GFX942-SDAG-STRESS-NEXT:    s_load_dwordx2 s[16:17], s[4:5], 0x24
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v0, 1.0
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v1, 2.0
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v2, 0x40400000
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v3, 4.0
-; GFX942-SDAG-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-SDAG-STRESS-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
-; GFX942-SDAG-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a0, s0
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a1, s1
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a2, s2
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a3, s3
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a4, s4
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a5, s5
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a6, s6
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a7, s7
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a8, s8
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a9, s9
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a10, s10
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a11, s11
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a12, s12
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a13, s13
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a14, s14
-; GFX942-SDAG-STRESS-NEXT:    v_accvgpr_write_b32 a15, s15
-; GFX942-SDAG-STRESS-NEXT:    s_nop 1
-; GFX942-SDAG-STRESS-NEXT:    v_mfma_f32_32x32x4_xf32 a[0:15], v[0:1], v[2:3], a[0:15] cbsz:1 abid:2 blgp:3
-; GFX942-SDAG-STRESS-NEXT:    v_mov_b32_e32 v0, 0
-; GFX942-SDAG-STRESS-NEXT:    s_nop 7
-; GFX942-SDAG-STRESS-NEXT:    s_nop 1
-; GFX942-SDAG-STRESS-NEXT:    global_store_dwordx4 v0, a[12:15], s[16:17] offset:48
-; GFX942-SDAG-STRESS-NEXT:    global_store_dwordx4 v0, a[8:11], s[16:17] offset:32
-; GFX942-SDAG-STRESS-NEXT:    global_store_dwordx4 v0, a[4:7], s[16:17] offset:16
-; GFX942-SDAG-STRESS-NEXT:    global_store_dwordx4 v0, a[0:3], s[16:17]
-; GFX942-SDAG-STRESS-NEXT:    s_endpgm
-;
-; GFX942-GISEL-STRESS-LABEL: test_mfma_f32_32x32x4xf32:
-; GFX942-GISEL-STRESS:       ; %bb.0: ; %bb
-; GFX942-GISEL-STRESS-NEXT:    s_load_dwordx2 s[16:17], s[4:5], 0x24
-; GFX942-GISEL-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-STRESS-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
-; GFX942-GISEL-STRESS-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a0, s0
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a1, s1
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a2, s2
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a3, s3
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a4, s4
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a5, s5
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a6, s6
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a7, s7
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a8, s8
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a9, s9
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a10, s10
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a11, s11
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a12, s12
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a13, s13
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a14, s14
-; GFX942-GISEL-STRESS-NEXT:    v_accvgpr_write_b32 a15, s15
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s0, 1.0
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s1, 2.0
-; GFX942-GISEL-STRESS-NEXT:    v_mov_b64_e32 v[0:1], s[0:1]
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s0, 0x40400000
-; GFX942-GISEL-STRESS-NEXT:    s_mov_b32 s1, 4.0
-; GFX942-GISEL-STRESS-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX942-GISEL-STRESS-NEXT:    s_nop 1
-; GFX942-GISEL-STRESS-NEXT:    v_mfma_f32_32x32x4_xf32 a[0:15], v[0:1], v[2:3], a[0:15] cbsz:1 abid:2 blgp:3
-; GFX942-GISEL-STRESS-NEXT:    v_mov_b32_e32 v0, 0
-; GFX942-GISEL-STRESS-NEXT:    s_nop 7
-; GFX942-GISEL-STRESS-NEXT:    s_nop 1
-; GFX942-GISEL-STRESS-NEXT:    global_store_dwordx4 v0, a[0:3], s[16:17]
-; GFX942-GISEL-STRESS-NEXT:    global_store_dwordx4 v0, a[4:7], s[16:17] offset:16
-; GFX942-GISEL-STRESS-NEXT:    global_store_dwordx4 v0, a[8:11], s[16:17] offset:32
-; GFX942-GISEL-STRESS-NEXT:    global_store_dwordx4 v0, a[12:15], s[16:17] offset:48
-; GFX942-GISEL-STRESS-NEXT:    s_endpgm
 bb:
   %in.1 = load <16 x float>, ptr addrspace(1) %arg
   %mai.1 = tail call <16 x float> @llvm.amdgcn.mfma.f32.32x32x4.xf32(<2 x float> <float 1.0, float 2.0>, <2 x float> <float 3.0, float 4.0>, <16 x float> %in.1, i32 1, i32 2, i32 3)
@@ -264,4 +142,3 @@ bb:
 attributes #0 = { "amdgpu-flat-work-group-size"="1,256" }
 ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
 ; GFX942: {{.*}}
-; GFX942-STRESS: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
index 80568810e42b5..b25fe8392a60e 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
@@ -17,24 +17,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x64_f16__vgpr(ptr addrspace(1) %
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
 ; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; SDAG-NEXT:    v_mov_b32_e32 v12, 0
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    global_load_dwordx4 v[14:17], v0, s[6:7]
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x44
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
-; SDAG-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
-; SDAG-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
+; SDAG-NEXT:    v_mov_b64_e32 v[14:15], s[2:3]
+; SDAG-NEXT:    v_mov_b64_e32 v[12:13], s[0:1]
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; SDAG-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; SDAG-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; SDAG-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; SDAG-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; SDAG-NEXT:    v_mov_b32_e32 v13, s16
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_16x16x64_f16 v[14:17], v[8:11], v[0:7], v13 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_16x16x64_f16 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v12, v[14:17], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__vgpr:
@@ -547,24 +547,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x64_bf16__vgpr(ptr addrspace(1)
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
 ; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GCN-NEXT:    v_mov_b32_e32 v12, 0
+; GCN-NEXT:    v_mov_b32_e32 v16, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    global_load_dwordx4 v[14:17], v0, s[6:7]
+; GCN-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; GCN-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x44
 ; GCN-NEXT:    s_load_dword s16, s[4:5], 0x64
-; GCN-NEXT:    v_mov_b64_e32 v[10:11], s[2:3]
-; GCN-NEXT:    v_mov_b64_e32 v[8:9], s[0:1]
+; GCN-NEXT:    v_mov_b64_e32 v[14:15], s[2:3]
+; GCN-NEXT:    v_mov_b64_e32 v[12:13], s[0:1]
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_mov_b64_e32 v[0:1], s[8:9]
 ; GCN-NEXT:    v_mov_b64_e32 v[2:3], s[10:11]
 ; GCN-NEXT:    v_mov_b64_e32 v[4:5], s[12:13]
 ; GCN-NEXT:    v_mov_b64_e32 v[6:7], s[14:15]
-; GCN-NEXT:    v_mov_b32_e32 v13, s16
+; GCN-NEXT:    v_mov_b32_e32 v17, s16
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_nop 0
-; GCN-NEXT:    v_smfmac_f32_16x16x64_bf16 v[14:17], v[8:11], v[0:7], v13 cbsz:1 abid:2
+; GCN-NEXT:    v_smfmac_f32_16x16x64_bf16 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; GCN-NEXT:    s_nop 7
-; GCN-NEXT:    global_store_dwordx4 v12, v[14:17], s[6:7]
+; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; GCN-NEXT:    s_endpgm
 bb:
   %id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -855,30 +855,30 @@ define amdgpu_kernel void @test_smfmac_i32_16x16x128_i8__vgpr(ptr addrspace(1) %
 ; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    global_load_dwordx4 v[10:13], v0, s[6:7]
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
-; SDAG-NEXT:    v_mov_b32_e32 v14, s8
-; SDAG-NEXT:    v_mov_b32_e32 v15, s9
-; SDAG-NEXT:    v_mov_b32_e32 v16, s10
-; SDAG-NEXT:    v_mov_b32_e32 v17, s11
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
+; SDAG-NEXT:    v_mov_b32_e32 v12, s8
+; SDAG-NEXT:    v_mov_b32_e32 v13, s9
+; SDAG-NEXT:    v_mov_b32_e32 v14, s10
+; SDAG-NEXT:    v_mov_b32_e32 v15, s11
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s0
-; SDAG-NEXT:    v_mov_b32_e32 v7, s1
-; SDAG-NEXT:    v_mov_b32_e32 v8, s2
-; SDAG-NEXT:    v_mov_b32_e32 v9, s3
-; SDAG-NEXT:    v_mov_b32_e32 v1, s16
+; SDAG-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-NEXT:    v_mov_b32_e32 v5, s1
+; SDAG-NEXT:    v_mov_b32_e32 v6, s2
+; SDAG-NEXT:    v_mov_b32_e32 v7, s3
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_i32_16x16x128_i8 v[10:13], v[14:17], v[2:9], v1 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_i32_16x16x128_i8 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, v[10:13], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_smfmac_i32_16x16x128_i8__vgpr:
@@ -1032,22 +1032,22 @@ define amdgpu_kernel void @test_smfmac_i32_32x32x64_i8__vgpr(ptr addrspace(1) %a
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v26, s8
-; SDAG-NEXT:    v_mov_b32_e32 v27, s9
-; SDAG-NEXT:    v_mov_b32_e32 v28, s10
-; SDAG-NEXT:    v_mov_b32_e32 v29, s11
-; SDAG-NEXT:    v_mov_b32_e32 v18, s12
-; SDAG-NEXT:    v_mov_b32_e32 v19, s13
-; SDAG-NEXT:    v_mov_b32_e32 v20, s14
-; SDAG-NEXT:    v_mov_b32_e32 v21, s15
-; SDAG-NEXT:    v_mov_b32_e32 v22, s0
-; SDAG-NEXT:    v_mov_b32_e32 v23, s1
-; SDAG-NEXT:    v_mov_b32_e32 v24, s2
-; SDAG-NEXT:    v_mov_b32_e32 v25, s3
-; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v24, s8
+; SDAG-NEXT:    v_mov_b32_e32 v25, s9
+; SDAG-NEXT:    v_mov_b32_e32 v26, s10
+; SDAG-NEXT:    v_mov_b32_e32 v27, s11
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    v_mov_b32_e32 v20, s0
+; SDAG-NEXT:    v_mov_b32_e32 v21, s1
+; SDAG-NEXT:    v_mov_b32_e32 v22, s2
+; SDAG-NEXT:    v_mov_b32_e32 v23, s3
+; SDAG-NEXT:    v_mov_b32_e32 v28, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_i32_32x32x64_i8 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_i32_32x32x64_i8 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2
 ; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
@@ -1397,30 +1397,30 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_bf8_bf8__vgpr(ptr addrspace
 ; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    global_load_dwordx4 v[10:13], v0, s[6:7]
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
-; SDAG-NEXT:    v_mov_b32_e32 v14, s8
-; SDAG-NEXT:    v_mov_b32_e32 v15, s9
-; SDAG-NEXT:    v_mov_b32_e32 v16, s10
-; SDAG-NEXT:    v_mov_b32_e32 v17, s11
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
+; SDAG-NEXT:    v_mov_b32_e32 v12, s8
+; SDAG-NEXT:    v_mov_b32_e32 v13, s9
+; SDAG-NEXT:    v_mov_b32_e32 v14, s10
+; SDAG-NEXT:    v_mov_b32_e32 v15, s11
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s0
-; SDAG-NEXT:    v_mov_b32_e32 v7, s1
-; SDAG-NEXT:    v_mov_b32_e32 v8, s2
-; SDAG-NEXT:    v_mov_b32_e32 v9, s3
-; SDAG-NEXT:    v_mov_b32_e32 v1, s16
+; SDAG-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-NEXT:    v_mov_b32_e32 v5, s1
+; SDAG-NEXT:    v_mov_b32_e32 v6, s2
+; SDAG-NEXT:    v_mov_b32_e32 v7, s3
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_bf8 v[10:13], v[14:17], v[2:9], v1 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_bf8 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, v[10:13], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__vgpr:
@@ -1566,30 +1566,30 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_bf8_fp8__vgpr(ptr addrspace
 ; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    global_load_dwordx4 v[10:13], v0, s[6:7]
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
-; SDAG-NEXT:    v_mov_b32_e32 v14, s8
-; SDAG-NEXT:    v_mov_b32_e32 v15, s9
-; SDAG-NEXT:    v_mov_b32_e32 v16, s10
-; SDAG-NEXT:    v_mov_b32_e32 v17, s11
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
+; SDAG-NEXT:    v_mov_b32_e32 v12, s8
+; SDAG-NEXT:    v_mov_b32_e32 v13, s9
+; SDAG-NEXT:    v_mov_b32_e32 v14, s10
+; SDAG-NEXT:    v_mov_b32_e32 v15, s11
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s0
-; SDAG-NEXT:    v_mov_b32_e32 v7, s1
-; SDAG-NEXT:    v_mov_b32_e32 v8, s2
-; SDAG-NEXT:    v_mov_b32_e32 v9, s3
-; SDAG-NEXT:    v_mov_b32_e32 v1, s16
+; SDAG-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-NEXT:    v_mov_b32_e32 v5, s1
+; SDAG-NEXT:    v_mov_b32_e32 v6, s2
+; SDAG-NEXT:    v_mov_b32_e32 v7, s3
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[10:13], v[14:17], v[2:9], v1 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_bf8_fp8 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, v[10:13], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__vgpr:
@@ -1735,30 +1735,30 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_fp8_bf8__vgpr(ptr addrspace
 ; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    global_load_dwordx4 v[10:13], v0, s[6:7]
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
-; SDAG-NEXT:    v_mov_b32_e32 v14, s8
-; SDAG-NEXT:    v_mov_b32_e32 v15, s9
-; SDAG-NEXT:    v_mov_b32_e32 v16, s10
-; SDAG-NEXT:    v_mov_b32_e32 v17, s11
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
+; SDAG-NEXT:    v_mov_b32_e32 v12, s8
+; SDAG-NEXT:    v_mov_b32_e32 v13, s9
+; SDAG-NEXT:    v_mov_b32_e32 v14, s10
+; SDAG-NEXT:    v_mov_b32_e32 v15, s11
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s0
-; SDAG-NEXT:    v_mov_b32_e32 v7, s1
-; SDAG-NEXT:    v_mov_b32_e32 v8, s2
-; SDAG-NEXT:    v_mov_b32_e32 v9, s3
-; SDAG-NEXT:    v_mov_b32_e32 v1, s16
+; SDAG-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-NEXT:    v_mov_b32_e32 v5, s1
+; SDAG-NEXT:    v_mov_b32_e32 v6, s2
+; SDAG-NEXT:    v_mov_b32_e32 v7, s3
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_16x16x128_fp8_bf8 v[10:13], v[14:17], v[2:9], v1 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_fp8_bf8 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, v[10:13], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__vgpr:
@@ -1904,30 +1904,30 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_fp8_fp8__vgpr(ptr addrspace
 ; SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; SDAG-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SDAG-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x34
+; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    global_load_dwordx4 v[10:13], v0, s[6:7]
+; SDAG-NEXT:    global_load_dwordx4 v[8:11], v0, s[6:7]
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
-; SDAG-NEXT:    v_mov_b32_e32 v0, 0
-; SDAG-NEXT:    v_mov_b32_e32 v14, s8
-; SDAG-NEXT:    v_mov_b32_e32 v15, s9
-; SDAG-NEXT:    v_mov_b32_e32 v16, s10
-; SDAG-NEXT:    v_mov_b32_e32 v17, s11
-; SDAG-NEXT:    v_mov_b32_e32 v2, s12
-; SDAG-NEXT:    v_mov_b32_e32 v3, s13
-; SDAG-NEXT:    v_mov_b32_e32 v4, s14
-; SDAG-NEXT:    v_mov_b32_e32 v5, s15
+; SDAG-NEXT:    v_mov_b32_e32 v12, s8
+; SDAG-NEXT:    v_mov_b32_e32 v13, s9
+; SDAG-NEXT:    v_mov_b32_e32 v14, s10
+; SDAG-NEXT:    v_mov_b32_e32 v15, s11
+; SDAG-NEXT:    v_mov_b32_e32 v0, s12
+; SDAG-NEXT:    v_mov_b32_e32 v1, s13
+; SDAG-NEXT:    v_mov_b32_e32 v2, s14
+; SDAG-NEXT:    v_mov_b32_e32 v3, s15
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v6, s0
-; SDAG-NEXT:    v_mov_b32_e32 v7, s1
-; SDAG-NEXT:    v_mov_b32_e32 v8, s2
-; SDAG-NEXT:    v_mov_b32_e32 v9, s3
-; SDAG-NEXT:    v_mov_b32_e32 v1, s16
+; SDAG-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-NEXT:    v_mov_b32_e32 v5, s1
+; SDAG-NEXT:    v_mov_b32_e32 v6, s2
+; SDAG-NEXT:    v_mov_b32_e32 v7, s3
+; SDAG-NEXT:    v_mov_b32_e32 v17, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_16x16x128_fp8_fp8 v[10:13], v[14:17], v[2:9], v1 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_16x16x128_fp8_fp8 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
 ; SDAG-NEXT:    s_nop 7
-; SDAG-NEXT:    global_store_dwordx4 v0, v[10:13], s[6:7]
+; SDAG-NEXT:    global_store_dwordx4 v16, v[8:11], s[6:7]
 ; SDAG-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__vgpr:
@@ -2081,22 +2081,22 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x64_bf8_bf8__vgpr(ptr addrspace(
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v26, s8
-; SDAG-NEXT:    v_mov_b32_e32 v27, s9
-; SDAG-NEXT:    v_mov_b32_e32 v28, s10
-; SDAG-NEXT:    v_mov_b32_e32 v29, s11
-; SDAG-NEXT:    v_mov_b32_e32 v18, s12
-; SDAG-NEXT:    v_mov_b32_e32 v19, s13
-; SDAG-NEXT:    v_mov_b32_e32 v20, s14
-; SDAG-NEXT:    v_mov_b32_e32 v21, s15
-; SDAG-NEXT:    v_mov_b32_e32 v22, s0
-; SDAG-NEXT:    v_mov_b32_e32 v23, s1
-; SDAG-NEXT:    v_mov_b32_e32 v24, s2
-; SDAG-NEXT:    v_mov_b32_e32 v25, s3
-; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v24, s8
+; SDAG-NEXT:    v_mov_b32_e32 v25, s9
+; SDAG-NEXT:    v_mov_b32_e32 v26, s10
+; SDAG-NEXT:    v_mov_b32_e32 v27, s11
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    v_mov_b32_e32 v20, s0
+; SDAG-NEXT:    v_mov_b32_e32 v21, s1
+; SDAG-NEXT:    v_mov_b32_e32 v22, s2
+; SDAG-NEXT:    v_mov_b32_e32 v23, s3
+; SDAG-NEXT:    v_mov_b32_e32 v28, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_32x32x64_bf8_bf8 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_32x32x64_bf8_bf8 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2
 ; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
@@ -2454,22 +2454,22 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x64_bf8_fp8__vgpr(ptr addrspace(
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v26, s8
-; SDAG-NEXT:    v_mov_b32_e32 v27, s9
-; SDAG-NEXT:    v_mov_b32_e32 v28, s10
-; SDAG-NEXT:    v_mov_b32_e32 v29, s11
-; SDAG-NEXT:    v_mov_b32_e32 v18, s12
-; SDAG-NEXT:    v_mov_b32_e32 v19, s13
-; SDAG-NEXT:    v_mov_b32_e32 v20, s14
-; SDAG-NEXT:    v_mov_b32_e32 v21, s15
-; SDAG-NEXT:    v_mov_b32_e32 v22, s0
-; SDAG-NEXT:    v_mov_b32_e32 v23, s1
-; SDAG-NEXT:    v_mov_b32_e32 v24, s2
-; SDAG-NEXT:    v_mov_b32_e32 v25, s3
-; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v24, s8
+; SDAG-NEXT:    v_mov_b32_e32 v25, s9
+; SDAG-NEXT:    v_mov_b32_e32 v26, s10
+; SDAG-NEXT:    v_mov_b32_e32 v27, s11
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    v_mov_b32_e32 v20, s0
+; SDAG-NEXT:    v_mov_b32_e32 v21, s1
+; SDAG-NEXT:    v_mov_b32_e32 v22, s2
+; SDAG-NEXT:    v_mov_b32_e32 v23, s3
+; SDAG-NEXT:    v_mov_b32_e32 v28, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_32x32x64_bf8_fp8 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_32x32x64_bf8_fp8 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2
 ; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
@@ -2827,22 +2827,22 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x64_fp8_bf8__vgpr(ptr addrspace(
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v26, s8
-; SDAG-NEXT:    v_mov_b32_e32 v27, s9
-; SDAG-NEXT:    v_mov_b32_e32 v28, s10
-; SDAG-NEXT:    v_mov_b32_e32 v29, s11
-; SDAG-NEXT:    v_mov_b32_e32 v18, s12
-; SDAG-NEXT:    v_mov_b32_e32 v19, s13
-; SDAG-NEXT:    v_mov_b32_e32 v20, s14
-; SDAG-NEXT:    v_mov_b32_e32 v21, s15
-; SDAG-NEXT:    v_mov_b32_e32 v22, s0
-; SDAG-NEXT:    v_mov_b32_e32 v23, s1
-; SDAG-NEXT:    v_mov_b32_e32 v24, s2
-; SDAG-NEXT:    v_mov_b32_e32 v25, s3
-; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v24, s8
+; SDAG-NEXT:    v_mov_b32_e32 v25, s9
+; SDAG-NEXT:    v_mov_b32_e32 v26, s10
+; SDAG-NEXT:    v_mov_b32_e32 v27, s11
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    v_mov_b32_e32 v20, s0
+; SDAG-NEXT:    v_mov_b32_e32 v21, s1
+; SDAG-NEXT:    v_mov_b32_e32 v22, s2
+; SDAG-NEXT:    v_mov_b32_e32 v23, s3
+; SDAG-NEXT:    v_mov_b32_e32 v28, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_32x32x64_fp8_bf8 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_32x32x64_fp8_bf8 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2
 ; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
@@ -3200,22 +3200,22 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x64_fp8_fp8__vgpr(ptr addrspace(
 ; SDAG-NEXT:    s_load_dword s16, s[4:5], 0x64
 ; SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x54
 ; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
-; SDAG-NEXT:    v_mov_b32_e32 v26, s8
-; SDAG-NEXT:    v_mov_b32_e32 v27, s9
-; SDAG-NEXT:    v_mov_b32_e32 v28, s10
-; SDAG-NEXT:    v_mov_b32_e32 v29, s11
-; SDAG-NEXT:    v_mov_b32_e32 v18, s12
-; SDAG-NEXT:    v_mov_b32_e32 v19, s13
-; SDAG-NEXT:    v_mov_b32_e32 v20, s14
-; SDAG-NEXT:    v_mov_b32_e32 v21, s15
-; SDAG-NEXT:    v_mov_b32_e32 v22, s0
-; SDAG-NEXT:    v_mov_b32_e32 v23, s1
-; SDAG-NEXT:    v_mov_b32_e32 v24, s2
-; SDAG-NEXT:    v_mov_b32_e32 v25, s3
-; SDAG-NEXT:    v_mov_b32_e32 v16, s16
+; SDAG-NEXT:    v_mov_b32_e32 v24, s8
+; SDAG-NEXT:    v_mov_b32_e32 v25, s9
+; SDAG-NEXT:    v_mov_b32_e32 v26, s10
+; SDAG-NEXT:    v_mov_b32_e32 v27, s11
+; SDAG-NEXT:    v_mov_b32_e32 v16, s12
+; SDAG-NEXT:    v_mov_b32_e32 v17, s13
+; SDAG-NEXT:    v_mov_b32_e32 v18, s14
+; SDAG-NEXT:    v_mov_b32_e32 v19, s15
+; SDAG-NEXT:    v_mov_b32_e32 v20, s0
+; SDAG-NEXT:    v_mov_b32_e32 v21, s1
+; SDAG-NEXT:    v_mov_b32_e32 v22, s2
+; SDAG-NEXT:    v_mov_b32_e32 v23, s3
+; SDAG-NEXT:    v_mov_b32_e32 v28, s16
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_nop 0
-; SDAG-NEXT:    v_smfmac_f32_32x32x64_fp8_fp8 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2
+; SDAG-NEXT:    v_smfmac_f32_32x32x64_fp8_fp8 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2
 ; SDAG-NEXT:    v_mov_b32_e32 v16, 0
 ; SDAG-NEXT:    s_nop 7
 ; SDAG-NEXT:    s_nop 2
@@ -3552,4 +3552,4 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_fp8__sgpr(<4 x i32> inreg %arg
   ret <16 x float> %result
 }
 
-attributes #0 = { "amdgpu-flat-work-group-size"="1,256" }
+attributes #0 = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-agpr-alloc"="0,0" }