[llvm] AMDGPU/GlobalISel: RegBankLegalize rules for gfx950 smfmac intrinsics (PR #203287)
Petar Avramovic via llvm-commits
llvm-commits at lists.llvm.org
Thu Jun 11 07:15:07 PDT 2026
https://github.com/petar-avramovic created https://github.com/llvm/llvm-project/pull/203287
None
>From 41fa982c49abfbd31341cf0d1b8f686a60e5af56 Mon Sep 17 00:00:00 2001
From: Petar Avramovic <Petar.Avramovic at amd.com>
Date: Thu, 11 Jun 2026 15:09:36 +0200
Subject: [PATCH] AMDGPU/GlobalISel: RegBankLegalize rules for gfx950 smfmac
intrinsics
---
.../AMDGPU/AMDGPURegBankLegalizeRules.cpp | 12 +-
.../AMDGPU/llvm.amdgcn.smfmac.gfx950.ll | 1408 +++++++----------
2 files changed, 624 insertions(+), 796 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
index 838951bcd8d26..45322296c943b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
@@ -2521,13 +2521,21 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
{IntrId, VgprAnyTy, VgprAnyTy, VgprOrAgprAnyTy}}});
addRulesForIOpcs(
- {amdgcn_smfmac_f32_16x16x32_f16, amdgcn_smfmac_f32_32x32x16_f16,
+ {// gfx942+
+ amdgcn_smfmac_f32_16x16x32_f16, amdgcn_smfmac_f32_32x32x16_f16,
amdgcn_smfmac_f32_16x16x32_bf16, amdgcn_smfmac_f32_32x32x16_bf16,
amdgcn_smfmac_i32_16x16x64_i8, amdgcn_smfmac_i32_32x32x32_i8,
amdgcn_smfmac_f32_16x16x64_bf8_bf8, amdgcn_smfmac_f32_16x16x64_bf8_fp8,
amdgcn_smfmac_f32_16x16x64_fp8_bf8, amdgcn_smfmac_f32_16x16x64_fp8_fp8,
amdgcn_smfmac_f32_32x32x32_bf8_bf8, amdgcn_smfmac_f32_32x32x32_bf8_fp8,
- amdgcn_smfmac_f32_32x32x32_fp8_bf8, amdgcn_smfmac_f32_32x32x32_fp8_fp8})
+ amdgcn_smfmac_f32_32x32x32_fp8_bf8, amdgcn_smfmac_f32_32x32x32_fp8_fp8,
+ // gfx950+
+ amdgcn_smfmac_f32_16x16x64_f16, amdgcn_smfmac_f32_32x32x32_f16,
+ amdgcn_smfmac_i32_16x16x128_i8, amdgcn_smfmac_i32_32x32x64_i8,
+ amdgcn_smfmac_f32_16x16x128_bf8_bf8, amdgcn_smfmac_f32_16x16x128_bf8_fp8,
+ amdgcn_smfmac_f32_16x16x128_fp8_bf8, amdgcn_smfmac_f32_16x16x128_fp8_fp8,
+ amdgcn_smfmac_f32_32x32x64_bf8_bf8, amdgcn_smfmac_f32_32x32x64_bf8_fp8,
+ amdgcn_smfmac_f32_32x32x64_fp8_bf8, amdgcn_smfmac_f32_32x32x64_fp8_fp8})
.Any({{DivAnyTy},
{{VgprOrAgprAnyTy},
{IntrId, VgprAnyTy, VgprAnyTy, VgprOrAgprAnyTy, VgprAnyTy}}});
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
index 6b287385414e0..a2ff88ec6874d 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
@@ -1,9 +1,9 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -mtriple=amdgcn -mcpu=gfx950 -global-isel=0 -amdgpu-mfma-vgpr-form=0 < %s | FileCheck -enable-var-scope --check-prefixes=GCN,SDAG %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx950 -global-isel=1 -global-isel-abort=2 -amdgpu-mfma-vgpr-form=0 < %s | FileCheck -enable-var-scope --check-prefixes=GCN,GISEL %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx950 -global-isel=1 -new-reg-bank-select -global-isel-abort=2 -amdgpu-mfma-vgpr-form=0 < %s | FileCheck -enable-var-scope --check-prefixes=GCN,GISEL %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx950 -global-isel=0 < %s | FileCheck -enable-var-scope --check-prefixes=GCN-VGPR,SDAG-VGPR %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx950 -global-isel=1 -global-isel-abort=2 < %s | FileCheck -enable-var-scope --check-prefixes=GCN-VGPR,GISEL-VGPR %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx950 -global-isel=1 -new-reg-bank-select -global-isel-abort=2 < %s | FileCheck -enable-var-scope --check-prefixes=GCN-VGPR,GISEL-VGPR %s
declare i32 @llvm.amdgcn.workitem.id.x()
@@ -121,16 +121,25 @@ bb:
}
define <4 x float> @test_smfmac_f32_16x16x64_f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x64_f16:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x64_f16:
; GCN-VGPR: ; %bb.0:
@@ -147,16 +156,25 @@ define <4 x float> @test_smfmac_f32_16x16x64_f16(<8 x half> %arg0, <16 x half> %
}
define <4 x float> @test_smfmac_f32_16x16x64_f16__flags0(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x64_f16__flags0:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x64_f16__flags0:
; GCN-VGPR: ; %bb.0:
@@ -173,16 +191,25 @@ define <4 x float> @test_smfmac_f32_16x16x64_f16__flags0(<8 x half> %arg0, <16 x
}
define <4 x float> @test_smfmac_f32_16x16x64_f16__flags1(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x64_f16__flags1:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x64_f16__flags1:
; GCN-VGPR: ; %bb.0:
@@ -226,19 +253,22 @@ define <4 x float> @test_smfmac_f32_16x16x64_f16__sgpr(<8 x half> inreg %arg0, <
; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__sgpr:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
-; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
-; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
-; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
-; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
-; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
-; GISEL-NEXT: v_mov_b32_e32 v0, s24
-; GISEL-NEXT: v_mov_b32_e32 v1, s25
-; GISEL-NEXT: v_mov_b32_e32 v2, s26
-; GISEL-NEXT: v_mov_b32_e32 v3, s27
-; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[22:23]
+; GISEL-NEXT: v_mov_b32_e32 v14, s24
+; GISEL-NEXT: v_mov_b32_e32 v15, s25
+; GISEL-NEXT: v_mov_b32_e32 v16, s26
+; GISEL-NEXT: v_mov_b32_e32 v17, s27
+; GISEL-NEXT: v_mov_b32_e32 v12, s28
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[14:17], v[8:11], v[0:7], v12
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[16:17]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; SDAG-VGPR-LABEL: test_smfmac_f32_16x16x64_f16__sgpr:
@@ -450,36 +480,16 @@ define <16 x float> @test_smfmac_f32_32x32x32_f16(<8 x half> %arg0, <16 x half>
; GISEL-LABEL: test_smfmac_f32_32x32x32_f16:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x32_f16 v[0:15], v[48:51], v[30:37], v28
+; GISEL-NEXT: v_smfmac_f32_32x32x32_f16 v[12:27], v[0:3], v[4:11], v28
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x32_f16:
@@ -535,36 +545,16 @@ define <16 x float> @test_smfmac_f32_32x32x32_f16__flags0(<8 x half> %arg0, <16
; GISEL-LABEL: test_smfmac_f32_32x32x32_f16__flags0:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x32_f16 v[0:15], v[48:51], v[30:37], v28 cbsz:1 abid:3
+; GISEL-NEXT: v_smfmac_f32_32x32x32_f16 v[12:27], v[0:3], v[4:11], v28 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x32_f16__flags0:
@@ -620,36 +610,16 @@ define <16 x float> @test_smfmac_f32_32x32x32_f16__flags1(<8 x half> %arg0, <16
; GISEL-LABEL: test_smfmac_f32_32x32x32_f16__flags1:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x32_f16 v[0:15], v[48:51], v[30:37], v28 cbsz:3 abid:1
+; GISEL-NEXT: v_smfmac_f32_32x32x32_f16 v[12:27], v[0:3], v[4:11], v28 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x32_f16__flags1:
@@ -1528,16 +1498,25 @@ bb:
}
define <4 x i32> @test_smfmac_i32_16x16x128_i8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x i32> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_i32_16x16x128_i8:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_i32_16x16x128_i8:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_i32_16x16x128_i8:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_i32_16x16x128_i8:
; GCN-VGPR: ; %bb.0:
@@ -1554,16 +1533,25 @@ define <4 x i32> @test_smfmac_i32_16x16x128_i8(<4 x i32> %arg0, <8 x i32> %arg1,
}
define <4 x i32> @test_smfmac_i32_16x16x128_i8__flags0(<4 x i32> %arg0, <8 x i32> %arg1, <4 x i32> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_i32_16x16x128_i8__flags0:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_i32_16x16x128_i8__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_i32_16x16x128_i8__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_i32_16x16x128_i8__flags0:
; GCN-VGPR: ; %bb.0:
@@ -1580,16 +1568,25 @@ define <4 x i32> @test_smfmac_i32_16x16x128_i8__flags0(<4 x i32> %arg0, <8 x i32
}
define <4 x i32> @test_smfmac_i32_16x16x128_i8__flags1(<4 x i32> %arg0, <8 x i32> %arg1, <4 x i32> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_i32_16x16x128_i8__flags1:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_i32_16x16x128_i8__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_i32_16x16x128_i8__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_i32_16x16x128_i8__flags1:
; GCN-VGPR: ; %bb.0:
@@ -1633,19 +1630,22 @@ define <4 x i32> @test_smfmac_i32_16x16x128_i8__sgpr(<4 x i32> inreg %arg0, <8 x
; GISEL-LABEL: test_smfmac_i32_16x16x128_i8__sgpr:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
-; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
-; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
-; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
-; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
-; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
-; GISEL-NEXT: v_mov_b32_e32 v0, s24
-; GISEL-NEXT: v_mov_b32_e32 v1, s25
-; GISEL-NEXT: v_mov_b32_e32 v2, s26
-; GISEL-NEXT: v_mov_b32_e32 v3, s27
-; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[22:23]
+; GISEL-NEXT: v_mov_b32_e32 v14, s24
+; GISEL-NEXT: v_mov_b32_e32 v15, s25
+; GISEL-NEXT: v_mov_b32_e32 v16, s26
+; GISEL-NEXT: v_mov_b32_e32 v17, s27
+; GISEL-NEXT: v_mov_b32_e32 v12, s28
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[14:17], v[8:11], v[0:7], v12
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[16:17]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; SDAG-VGPR-LABEL: test_smfmac_i32_16x16x128_i8__sgpr:
@@ -1869,36 +1869,16 @@ define <16 x i32> @test_smfmac_i32_32x32x64_i8(<4 x i32> %arg0, <8 x i32> %arg1,
; GISEL-LABEL: test_smfmac_i32_32x32x64_i8:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_i32_32x32x64_i8 v[0:15], v[48:51], v[30:37], v28
+; GISEL-NEXT: v_smfmac_i32_32x32x64_i8 v[12:27], v[0:3], v[4:11], v28
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_i32_32x32x64_i8:
@@ -1954,36 +1934,16 @@ define <16 x i32> @test_smfmac_i32_32x32x64_i8__flags0(<4 x i32> %arg0, <8 x i32
; GISEL-LABEL: test_smfmac_i32_32x32x64_i8__flags0:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_i32_32x32x64_i8 v[0:15], v[48:51], v[30:37], v28 cbsz:1 abid:3
+; GISEL-NEXT: v_smfmac_i32_32x32x64_i8 v[12:27], v[0:3], v[4:11], v28 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_i32_32x32x64_i8__flags0:
@@ -2039,36 +1999,16 @@ define <16 x i32> @test_smfmac_i32_32x32x64_i8__flags1(<4 x i32> %arg0, <8 x i32
; GISEL-LABEL: test_smfmac_i32_32x32x64_i8__flags1:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_i32_32x32x64_i8 v[0:15], v[48:51], v[30:37], v28 cbsz:3 abid:1
+; GISEL-NEXT: v_smfmac_i32_32x32x64_i8 v[12:27], v[0:3], v[4:11], v28 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_i32_32x32x64_i8__flags1:
@@ -2427,16 +2367,25 @@ bb:
}
define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_bf8_bf8:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_bf8:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_bf8:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_bf8:
; GCN-VGPR: ; %bb.0:
@@ -2453,16 +2402,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8(<4 x i32> %arg0, <8 x i32>
}
define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8__flags0(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags0:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags0:
; GCN-VGPR: ; %bb.0:
@@ -2479,16 +2437,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8__flags0(<4 x i32> %arg0, <
}
define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8__flags1(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags1:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__flags1:
; GCN-VGPR: ; %bb.0:
@@ -2532,19 +2499,22 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8__sgpr(<4 x i32> inreg %arg
; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__sgpr:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
-; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
-; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
-; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
-; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
-; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
-; GISEL-NEXT: v_mov_b32_e32 v0, s24
-; GISEL-NEXT: v_mov_b32_e32 v1, s25
-; GISEL-NEXT: v_mov_b32_e32 v2, s26
-; GISEL-NEXT: v_mov_b32_e32 v3, s27
-; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[22:23]
+; GISEL-NEXT: v_mov_b32_e32 v14, s24
+; GISEL-NEXT: v_mov_b32_e32 v15, s25
+; GISEL-NEXT: v_mov_b32_e32 v16, s26
+; GISEL-NEXT: v_mov_b32_e32 v17, s27
+; GISEL-NEXT: v_mov_b32_e32 v12, s28
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[14:17], v[8:11], v[0:7], v12
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[16:17]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; SDAG-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__sgpr:
@@ -2718,16 +2688,25 @@ bb:
}
define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_bf8_fp8:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_fp8:
; GCN-VGPR: ; %bb.0:
@@ -2744,16 +2723,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8(<4 x i32> %arg0, <8 x i32>
}
define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__flags0(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags0:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags0:
; GCN-VGPR: ; %bb.0:
@@ -2770,16 +2758,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__flags0(<4 x i32> %arg0, <
}
define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__flags1(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags1:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__flags1:
; GCN-VGPR: ; %bb.0:
@@ -2823,19 +2820,22 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__sgpr(<4 x i32> inreg %arg
; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__sgpr:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
-; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
-; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
-; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
-; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
-; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
-; GISEL-NEXT: v_mov_b32_e32 v0, s24
-; GISEL-NEXT: v_mov_b32_e32 v1, s25
-; GISEL-NEXT: v_mov_b32_e32 v2, s26
-; GISEL-NEXT: v_mov_b32_e32 v3, s27
-; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[22:23]
+; GISEL-NEXT: v_mov_b32_e32 v14, s24
+; GISEL-NEXT: v_mov_b32_e32 v15, s25
+; GISEL-NEXT: v_mov_b32_e32 v16, s26
+; GISEL-NEXT: v_mov_b32_e32 v17, s27
+; GISEL-NEXT: v_mov_b32_e32 v12, s28
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[14:17], v[8:11], v[0:7], v12
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[16:17]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; SDAG-VGPR-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__sgpr:
@@ -3009,16 +3009,25 @@ bb:
}
define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_fp8_bf8:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_bf8:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_bf8:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_bf8:
; GCN-VGPR: ; %bb.0:
@@ -3035,16 +3044,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8(<4 x i32> %arg0, <8 x i32>
}
define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8__flags0(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags0:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags0:
; GCN-VGPR: ; %bb.0:
@@ -3061,16 +3079,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8__flags0(<4 x i32> %arg0, <
}
define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8__flags1(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags1:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__flags1:
; GCN-VGPR: ; %bb.0:
@@ -3114,19 +3141,22 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8__sgpr(<4 x i32> inreg %arg
; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__sgpr:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
-; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
-; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
-; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
-; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
-; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
-; GISEL-NEXT: v_mov_b32_e32 v0, s24
-; GISEL-NEXT: v_mov_b32_e32 v1, s25
-; GISEL-NEXT: v_mov_b32_e32 v2, s26
-; GISEL-NEXT: v_mov_b32_e32 v3, s27
-; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[22:23]
+; GISEL-NEXT: v_mov_b32_e32 v14, s24
+; GISEL-NEXT: v_mov_b32_e32 v15, s25
+; GISEL-NEXT: v_mov_b32_e32 v16, s26
+; GISEL-NEXT: v_mov_b32_e32 v17, s27
+; GISEL-NEXT: v_mov_b32_e32 v12, s28
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[14:17], v[8:11], v[0:7], v12
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[16:17]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; SDAG-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__sgpr:
@@ -3300,16 +3330,25 @@ bb:
}
define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_fp8_fp8:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_fp8:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_fp8:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_fp8:
; GCN-VGPR: ; %bb.0:
@@ -3326,16 +3365,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8(<4 x i32> %arg0, <8 x i32>
}
define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8__flags0(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags0:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags0:
; GCN-VGPR: ; %bb.0:
@@ -3352,16 +3400,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8__flags0(<4 x i32> %arg0, <
}
define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8__flags1(<4 x i32> %arg0, <8 x i32> %arg1, <4 x float> %arg2, i32 %arg3) #1 {
-; GCN-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags1:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
-; GCN-NEXT: s_nop 7
-; GCN-NEXT: v_mov_b32_e32 v0, v12
-; GCN-NEXT: v_mov_b32_e32 v1, v13
-; GCN-NEXT: v_mov_b32_e32 v2, v14
-; GCN-NEXT: v_mov_b32_e32 v3, v15
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 7
+; SDAG-NEXT: v_mov_b32_e32 v0, v12
+; SDAG-NEXT: v_mov_b32_e32 v1, v13
+; SDAG-NEXT: v_mov_b32_e32 v2, v14
+; SDAG-NEXT: v_mov_b32_e32 v3, v15
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__flags1:
; GCN-VGPR: ; %bb.0:
@@ -3405,19 +3462,22 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8__sgpr(<4 x i32> inreg %arg
; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__sgpr:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
-; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
-; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
-; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
-; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
-; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
-; GISEL-NEXT: v_mov_b32_e32 v0, s24
-; GISEL-NEXT: v_mov_b32_e32 v1, s25
-; GISEL-NEXT: v_mov_b32_e32 v2, s26
-; GISEL-NEXT: v_mov_b32_e32 v3, s27
-; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[22:23]
+; GISEL-NEXT: v_mov_b32_e32 v14, s24
+; GISEL-NEXT: v_mov_b32_e32 v15, s25
+; GISEL-NEXT: v_mov_b32_e32 v16, s26
+; GISEL-NEXT: v_mov_b32_e32 v17, s27
+; GISEL-NEXT: v_mov_b32_e32 v12, s28
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[14:17], v[8:11], v[0:7], v12
+; GISEL-NEXT: s_nop 7
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[16:17]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; SDAG-VGPR-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__sgpr:
@@ -3641,36 +3701,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_bf8(<4 x i32> %arg0, <8 x i32>
; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_bf8:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[0:15], v[48:51], v[30:37], v28
+; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[12:27], v[0:3], v[4:11], v28
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_bf8_bf8:
@@ -3726,36 +3766,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_bf8__flags0(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_bf8__flags0:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[0:15], v[48:51], v[30:37], v28 cbsz:1 abid:3
+; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[12:27], v[0:3], v[4:11], v28 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_bf8_bf8__flags0:
@@ -3811,36 +3831,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_bf8__flags1(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_bf8__flags1:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[0:15], v[48:51], v[30:37], v28 cbsz:3 abid:1
+; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[12:27], v[0:3], v[4:11], v28 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_bf8_bf8__flags1:
@@ -4249,36 +4249,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_fp8(<4 x i32> %arg0, <8 x i32>
; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_fp8:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[0:15], v[48:51], v[30:37], v28
+; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[12:27], v[0:3], v[4:11], v28
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_bf8_fp8:
@@ -4334,36 +4314,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_fp8__flags0(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_fp8__flags0:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[0:15], v[48:51], v[30:37], v28 cbsz:1 abid:3
+; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[12:27], v[0:3], v[4:11], v28 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_bf8_fp8__flags0:
@@ -4419,36 +4379,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_fp8__flags1(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_fp8__flags1:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[0:15], v[48:51], v[30:37], v28 cbsz:3 abid:1
+; GISEL-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[12:27], v[0:3], v[4:11], v28 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_bf8_fp8__flags1:
@@ -4857,36 +4797,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_bf8(<4 x i32> %arg0, <8 x i32>
; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_bf8:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[0:15], v[48:51], v[30:37], v28
+; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[12:27], v[0:3], v[4:11], v28
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_fp8_bf8:
@@ -4942,36 +4862,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_bf8__flags0(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_bf8__flags0:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[0:15], v[48:51], v[30:37], v28 cbsz:1 abid:3
+; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[12:27], v[0:3], v[4:11], v28 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_fp8_bf8__flags0:
@@ -5027,36 +4927,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_bf8__flags1(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_bf8__flags1:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[0:15], v[48:51], v[30:37], v28 cbsz:3 abid:1
+; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[12:27], v[0:3], v[4:11], v28 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_fp8_bf8__flags1:
@@ -5465,36 +5345,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_fp8(<4 x i32> %arg0, <8 x i32>
; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_fp8:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[0:15], v[48:51], v[30:37], v28
+; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[12:27], v[0:3], v[4:11], v28
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_fp8_fp8:
@@ -5550,36 +5410,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_fp8__flags0(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_fp8__flags0:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[0:15], v[48:51], v[30:37], v28 cbsz:1 abid:3
+; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[12:27], v[0:3], v[4:11], v28 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_fp8_fp8__flags0:
@@ -5635,36 +5475,16 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_fp8__flags1(<4 x i32> %arg0, <
; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_fp8__flags1:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_mov_b32_e32 v48, v0
-; GISEL-NEXT: v_mov_b32_e32 v49, v1
-; GISEL-NEXT: v_mov_b32_e32 v50, v2
-; GISEL-NEXT: v_mov_b32_e32 v51, v3
-; GISEL-NEXT: v_mov_b32_e32 v30, v4
-; GISEL-NEXT: v_mov_b32_e32 v31, v5
-; GISEL-NEXT: v_mov_b32_e32 v32, v6
-; GISEL-NEXT: v_mov_b32_e32 v33, v7
-; GISEL-NEXT: v_mov_b32_e32 v34, v8
-; GISEL-NEXT: v_mov_b32_e32 v35, v9
-; GISEL-NEXT: v_mov_b32_e32 v36, v10
-; GISEL-NEXT: v_mov_b32_e32 v37, v11
-; GISEL-NEXT: v_mov_b32_e32 v0, v12
-; GISEL-NEXT: v_mov_b32_e32 v1, v13
-; GISEL-NEXT: v_mov_b32_e32 v2, v14
-; GISEL-NEXT: v_mov_b32_e32 v3, v15
-; GISEL-NEXT: v_mov_b32_e32 v4, v16
-; GISEL-NEXT: v_mov_b32_e32 v5, v17
-; GISEL-NEXT: v_mov_b32_e32 v6, v18
-; GISEL-NEXT: v_mov_b32_e32 v7, v19
-; GISEL-NEXT: v_mov_b32_e32 v8, v20
-; GISEL-NEXT: v_mov_b32_e32 v9, v21
-; GISEL-NEXT: v_mov_b32_e32 v10, v22
-; GISEL-NEXT: v_mov_b32_e32 v11, v23
-; GISEL-NEXT: v_mov_b32_e32 v12, v24
-; GISEL-NEXT: v_mov_b32_e32 v13, v25
-; GISEL-NEXT: v_mov_b32_e32 v14, v26
-; GISEL-NEXT: v_mov_b32_e32 v15, v27
-; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[0:15], v[48:51], v[30:37], v28 cbsz:3 abid:1
+; GISEL-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[12:27], v[0:3], v[4:11], v28 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 11
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], v[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], v[14:15]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], v[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], v[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], v[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], v[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], v[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], v[26:27]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GCN-VGPR-LABEL: test_smfmac_f32_32x32x64_fp8_fp8__flags1:
More information about the llvm-commits
mailing list