[llvm] [AMDGPU] Extend MFMA padding option to gfx90a+ (PR #86768)

Austin Kerbow via llvm-commits llvm-commits at lists.llvm.org
Tue Mar 26 22:06:08 PDT 2024


https://github.com/kerbowa created https://github.com/llvm/llvm-project/pull/86768

It was shown experimentally that this may have some benefit on newer HW.

>From 663981a9fd5f4b4ce59913d4ead8ad6d9dbb8a71 Mon Sep 17 00:00:00 2001
From: Austin Kerbow <Austin.Kerbow at amd.com>
Date: Tue, 26 Mar 2024 21:37:17 -0700
Subject: [PATCH] [AMDGPU] Extend MFMA padding option to gfx90a+

It was shown experimentally that this may have some benefit on newer HW.
---
 .../lib/Target/AMDGPU/GCNHazardRecognizer.cpp |   3 +
 .../AMDGPU/neighboring-mfma-padding.mir       | 504 ++++++++++++++++++
 2 files changed, 507 insertions(+)

diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
index e515b729e7d7e8..75766b11ca8229 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -2357,6 +2357,9 @@ int GCNHazardRecognizer::checkMAIHazards90A(MachineInstr *MI) {
       break;
   }
 
+  // Pad neighboring MFMA with noops for better inter-wave performance.
+  WaitStatesNeeded = std::max(WaitStatesNeeded, checkMFMAPadding(MI));
+
   return WaitStatesNeeded;
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/neighboring-mfma-padding.mir b/llvm/test/CodeGen/AMDGPU/neighboring-mfma-padding.mir
index 3de258bb52a556..bf2cf6aeb990d7 100644
--- a/llvm/test/CodeGen/AMDGPU/neighboring-mfma-padding.mir
+++ b/llvm/test/CodeGen/AMDGPU/neighboring-mfma-padding.mir
@@ -5,6 +5,14 @@
 # RUN: llc -mtriple=amdgcn -mcpu=gfx908 -amdgpu-mfma-padding-ratio=75 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx908-PAD75 %s
 # RUN: llc -mtriple=amdgcn -mcpu=gfx908 -amdgpu-mfma-padding-ratio=100 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx908-PAD100 %s
 
+# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx90a-DEFAULT %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -amdgpu-mfma-padding-ratio=50 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx90a-PAD50 %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -amdgpu-mfma-padding-ratio=100 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx90a-PAD100 %s
+
+# RUN: llc -mtriple=amdgcn -mcpu=gfx940 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx940-DEFAULT %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx940 -amdgpu-mfma-padding-ratio=50 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx940-PAD50 %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx940 -amdgpu-mfma-padding-ratio=100 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=gfx940-PAD100 %s
+
 ---
 name: mfma_padding_2_pass
 body: |
@@ -31,6 +39,35 @@ body: |
     ; gfx908-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     ; gfx908-PAD100-NEXT: S_NOP 1
     ; gfx908-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_2_pass
+    ; gfx90a-DEFAULT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_2_pass
+    ; gfx90a-PAD50: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: S_NOP 0
+    ; gfx90a-PAD50-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_2_pass
+    ; gfx90a-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: S_NOP 1
+    ; gfx90a-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_2_pass
+    ; gfx940-DEFAULT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: S_NOP 1
+    ; gfx940-DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_2_pass
+    ; gfx940-PAD50: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: S_NOP 1
+    ; gfx940-PAD50-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_2_pass
+    ; gfx940-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: S_NOP 1
+    ; gfx940-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
 ...
@@ -64,6 +101,40 @@ body: |
     ; gfx908-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
     ; gfx908-PAD100-NEXT: S_NOP 0
     ; gfx908-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_2_pass_1_intervening_valu
+    ; gfx90a-DEFAULT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_2_pass_1_intervening_valu
+    ; gfx90a-PAD50: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_2_pass_1_intervening_valu
+    ; gfx90a-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: S_NOP 0
+    ; gfx90a-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_2_pass_1_intervening_valu
+    ; gfx940-DEFAULT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: S_NOP 0
+    ; gfx940-DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_2_pass_1_intervening_valu
+    ; gfx940-PAD50: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: S_NOP 0
+    ; gfx940-PAD50-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_2_pass_1_intervening_valu
+    ; gfx940-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: S_NOP 0
+    ; gfx940-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     $vgpr2 = V_MOV_B32_e32 1, implicit $exec
     $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
@@ -100,6 +171,41 @@ body: |
     ; gfx908-PAD100-NEXT: DBG_VALUE
     ; gfx908-PAD100-NEXT: S_NOP 1
     ; gfx908-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_2_pass_dbg
+    ; gfx90a-DEFAULT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: DBG_VALUE
+    ; gfx90a-DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_2_pass_dbg
+    ; gfx90a-PAD50: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: DBG_VALUE
+    ; gfx90a-PAD50-NEXT: S_NOP 0
+    ; gfx90a-PAD50-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_2_pass_dbg
+    ; gfx90a-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: DBG_VALUE
+    ; gfx90a-PAD100-NEXT: S_NOP 1
+    ; gfx90a-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_2_pass_dbg
+    ; gfx940-DEFAULT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: DBG_VALUE
+    ; gfx940-DEFAULT-NEXT: S_NOP 1
+    ; gfx940-DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_2_pass_dbg
+    ; gfx940-PAD50: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: DBG_VALUE
+    ; gfx940-PAD50-NEXT: S_NOP 1
+    ; gfx940-PAD50-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_2_pass_dbg
+    ; gfx940-PAD100: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: DBG_VALUE
+    ; gfx940-PAD100-NEXT: S_NOP 1
+    ; gfx940-PAD100-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
     DBG_VALUE
     $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
@@ -132,6 +238,34 @@ body: |
     ; gfx908-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     ; gfx908-PAD100-NEXT: S_NOP 7
     ; gfx908-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_8_pass
+    ; gfx90a-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_8_pass
+    ; gfx90a-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: S_NOP 3
+    ; gfx90a-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_8_pass
+    ; gfx90a-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: S_NOP 7
+    ; gfx90a-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_8_pass
+    ; gfx940-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_8_pass
+    ; gfx940-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: S_NOP 3
+    ; gfx940-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_8_pass
+    ; gfx940-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: S_NOP 7
+    ; gfx940-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
 ...
@@ -172,6 +306,46 @@ body: |
     ; gfx908-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
     ; gfx908-PAD100-NEXT: S_NOP 5
     ; gfx908-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_8_pass_2_intervening_valu
+    ; gfx90a-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_8_pass_2_intervening_valu
+    ; gfx90a-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: S_NOP 1
+    ; gfx90a-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_8_pass_2_intervening_valu
+    ; gfx90a-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: S_NOP 5
+    ; gfx90a-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_8_pass_2_intervening_valu
+    ; gfx940-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_8_pass_2_intervening_valu
+    ; gfx940-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: S_NOP 1
+    ; gfx940-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_8_pass_2_intervening_valu
+    ; gfx940-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: S_NOP 5
+    ; gfx940-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $vgpr2 = V_MOV_B32_e32 1, implicit $exec
     $vgpr3 = V_MOV_B32_e32 1, implicit $exec
@@ -207,6 +381,36 @@ body: |
     ; gfx908-PAD100-NEXT: S_NOP 7
     ; gfx908-PAD100-NEXT: S_NOP 7
     ; gfx908-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_16_pass
+    ; gfx90a-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_16_pass
+    ; gfx90a-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: S_NOP 7
+    ; gfx90a-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_16_pass
+    ; gfx90a-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: S_NOP 7
+    ; gfx90a-PAD100-NEXT: S_NOP 7
+    ; gfx90a-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_16_pass
+    ; gfx940-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_16_pass
+    ; gfx940-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: S_NOP 7
+    ; gfx940-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_16_pass
+    ; gfx940-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: S_NOP 7
+    ; gfx940-PAD100-NEXT: S_NOP 7
+    ; gfx940-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
 ...
@@ -258,6 +462,60 @@ body: |
     ; gfx908-PAD100-NEXT: S_NOP 7
     ; gfx908-PAD100-NEXT: S_NOP 3
     ; gfx908-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_16_pass_4_intervening_valu
+    ; gfx90a-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_16_pass_4_intervening_valu
+    ; gfx90a-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: S_NOP 3
+    ; gfx90a-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_16_pass_4_intervening_valu
+    ; gfx90a-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: S_NOP 7
+    ; gfx90a-PAD100-NEXT: S_NOP 3
+    ; gfx90a-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_16_pass_4_intervening_valu
+    ; gfx940-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_16_pass_4_intervening_valu
+    ; gfx940-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: S_NOP 3
+    ; gfx940-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_16_pass_4_intervening_valu
+    ; gfx940-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: S_NOP 7
+    ; gfx940-PAD100-NEXT: S_NOP 3
+    ; gfx940-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $vgpr2 = V_MOV_B32_e32 1, implicit $exec
     $vgpr3 = V_MOV_B32_e32 1, implicit $exec
@@ -369,6 +627,126 @@ body: |
     ; gfx908-PAD100-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
     ; gfx908-PAD100-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
     ; gfx908-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_16_pass_16_intervening_valu
+    ; gfx90a-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr6 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr7 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr8 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr10 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr11 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr12 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr13 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr14 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr15 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_16_pass_16_intervening_valu
+    ; gfx90a-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr6 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr7 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr8 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr10 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr11 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr12 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr13 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr14 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr15 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_16_pass_16_intervening_valu
+    ; gfx90a-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr6 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr7 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr8 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr10 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr11 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr12 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr13 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr14 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr15 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx90a-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_16_pass_16_intervening_valu
+    ; gfx940-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr6 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr7 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr8 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr10 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr11 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr12 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr13 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr14 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr15 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_16_pass_16_intervening_valu
+    ; gfx940-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr6 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr7 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr8 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr10 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr11 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr12 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr13 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr14 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr15 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_16_pass_16_intervening_valu
+    ; gfx940-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr4 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr5 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr6 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr7 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr8 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr10 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr11 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr12 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr13 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr14 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr15 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr16 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
+    ; gfx940-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $vgpr2 = V_MOV_B32_e32 1, implicit $exec
     $vgpr3 = V_MOV_B32_e32 1, implicit $exec
@@ -414,6 +792,30 @@ body: |
     ; gfx908-PAD100-LABEL: name: mfma_padding_16_pass_occ_1
     ; gfx908-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     ; gfx908-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-DEFAULT-LABEL: name: mfma_padding_16_pass_occ_1
+    ; gfx90a-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD50-LABEL: name: mfma_padding_16_pass_occ_1
+    ; gfx90a-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx90a-PAD100-LABEL: name: mfma_padding_16_pass_occ_1
+    ; gfx90a-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx90a-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-DEFAULT-LABEL: name: mfma_padding_16_pass_occ_1
+    ; gfx940-DEFAULT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-DEFAULT-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD50-LABEL: name: mfma_padding_16_pass_occ_1
+    ; gfx940-PAD50: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD50-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ;
+    ; gfx940-PAD100-LABEL: name: mfma_padding_16_pass_occ_1
+    ; gfx940-PAD100: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+    ; gfx940-PAD100-NEXT: early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
 ...
@@ -506,6 +908,108 @@ body: |
   ; gfx908-PAD100-NEXT:   S_NOP 7
   ; gfx908-PAD100-NEXT:   S_NOP 5
   ; gfx908-PAD100-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ;
+  ; gfx90a-DEFAULT-LABEL: name: mfma_padding_16_pass_2_preds
+  ; gfx90a-DEFAULT: bb.0:
+  ; gfx90a-DEFAULT-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; gfx90a-DEFAULT-NEXT: {{  $}}
+  ; gfx90a-DEFAULT-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ; gfx90a-DEFAULT-NEXT:   S_CBRANCH_VCCZ %bb.2, implicit undef $vcc
+  ; gfx90a-DEFAULT-NEXT: {{  $}}
+  ; gfx90a-DEFAULT-NEXT: bb.1:
+  ; gfx90a-DEFAULT-NEXT:   successors: %bb.2(0x80000000)
+  ; gfx90a-DEFAULT-NEXT: {{  $}}
+  ; gfx90a-DEFAULT-NEXT:   $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx90a-DEFAULT-NEXT: {{  $}}
+  ; gfx90a-DEFAULT-NEXT: bb.2:
+  ; gfx90a-DEFAULT-NEXT:   $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx90a-DEFAULT-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ;
+  ; gfx90a-PAD50-LABEL: name: mfma_padding_16_pass_2_preds
+  ; gfx90a-PAD50: bb.0:
+  ; gfx90a-PAD50-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; gfx90a-PAD50-NEXT: {{  $}}
+  ; gfx90a-PAD50-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ; gfx90a-PAD50-NEXT:   S_CBRANCH_VCCZ %bb.2, implicit undef $vcc
+  ; gfx90a-PAD50-NEXT: {{  $}}
+  ; gfx90a-PAD50-NEXT: bb.1:
+  ; gfx90a-PAD50-NEXT:   successors: %bb.2(0x80000000)
+  ; gfx90a-PAD50-NEXT: {{  $}}
+  ; gfx90a-PAD50-NEXT:   $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx90a-PAD50-NEXT: {{  $}}
+  ; gfx90a-PAD50-NEXT: bb.2:
+  ; gfx90a-PAD50-NEXT:   $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx90a-PAD50-NEXT:   S_NOP 5
+  ; gfx90a-PAD50-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ;
+  ; gfx90a-PAD100-LABEL: name: mfma_padding_16_pass_2_preds
+  ; gfx90a-PAD100: bb.0:
+  ; gfx90a-PAD100-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; gfx90a-PAD100-NEXT: {{  $}}
+  ; gfx90a-PAD100-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ; gfx90a-PAD100-NEXT:   S_CBRANCH_VCCZ %bb.2, implicit undef $vcc
+  ; gfx90a-PAD100-NEXT: {{  $}}
+  ; gfx90a-PAD100-NEXT: bb.1:
+  ; gfx90a-PAD100-NEXT:   successors: %bb.2(0x80000000)
+  ; gfx90a-PAD100-NEXT: {{  $}}
+  ; gfx90a-PAD100-NEXT:   $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx90a-PAD100-NEXT: {{  $}}
+  ; gfx90a-PAD100-NEXT: bb.2:
+  ; gfx90a-PAD100-NEXT:   $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx90a-PAD100-NEXT:   S_NOP 7
+  ; gfx90a-PAD100-NEXT:   S_NOP 5
+  ; gfx90a-PAD100-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ;
+  ; gfx940-DEFAULT-LABEL: name: mfma_padding_16_pass_2_preds
+  ; gfx940-DEFAULT: bb.0:
+  ; gfx940-DEFAULT-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; gfx940-DEFAULT-NEXT: {{  $}}
+  ; gfx940-DEFAULT-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ; gfx940-DEFAULT-NEXT:   S_CBRANCH_VCCZ %bb.2, implicit undef $vcc
+  ; gfx940-DEFAULT-NEXT: {{  $}}
+  ; gfx940-DEFAULT-NEXT: bb.1:
+  ; gfx940-DEFAULT-NEXT:   successors: %bb.2(0x80000000)
+  ; gfx940-DEFAULT-NEXT: {{  $}}
+  ; gfx940-DEFAULT-NEXT:   $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx940-DEFAULT-NEXT: {{  $}}
+  ; gfx940-DEFAULT-NEXT: bb.2:
+  ; gfx940-DEFAULT-NEXT:   $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx940-DEFAULT-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ;
+  ; gfx940-PAD50-LABEL: name: mfma_padding_16_pass_2_preds
+  ; gfx940-PAD50: bb.0:
+  ; gfx940-PAD50-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; gfx940-PAD50-NEXT: {{  $}}
+  ; gfx940-PAD50-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ; gfx940-PAD50-NEXT:   S_CBRANCH_VCCZ %bb.2, implicit undef $vcc
+  ; gfx940-PAD50-NEXT: {{  $}}
+  ; gfx940-PAD50-NEXT: bb.1:
+  ; gfx940-PAD50-NEXT:   successors: %bb.2(0x80000000)
+  ; gfx940-PAD50-NEXT: {{  $}}
+  ; gfx940-PAD50-NEXT:   $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx940-PAD50-NEXT: {{  $}}
+  ; gfx940-PAD50-NEXT: bb.2:
+  ; gfx940-PAD50-NEXT:   $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx940-PAD50-NEXT:   S_NOP 5
+  ; gfx940-PAD50-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ;
+  ; gfx940-PAD100-LABEL: name: mfma_padding_16_pass_2_preds
+  ; gfx940-PAD100: bb.0:
+  ; gfx940-PAD100-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; gfx940-PAD100-NEXT: {{  $}}
+  ; gfx940-PAD100-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
+  ; gfx940-PAD100-NEXT:   S_CBRANCH_VCCZ %bb.2, implicit undef $vcc
+  ; gfx940-PAD100-NEXT: {{  $}}
+  ; gfx940-PAD100-NEXT: bb.1:
+  ; gfx940-PAD100-NEXT:   successors: %bb.2(0x80000000)
+  ; gfx940-PAD100-NEXT: {{  $}}
+  ; gfx940-PAD100-NEXT:   $vgpr2 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx940-PAD100-NEXT: {{  $}}
+  ; gfx940-PAD100-NEXT: bb.2:
+  ; gfx940-PAD100-NEXT:   $vgpr3 = V_MOV_B32_e32 1, implicit $exec
+  ; gfx940-PAD100-NEXT:   S_NOP 7
+  ; gfx940-PAD100-NEXT:   S_NOP 5
+  ; gfx940-PAD100-NEXT:   early-clobber $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
   bb.0:
     $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
     S_CBRANCH_VCCZ %bb.2, implicit undef $vcc



More information about the llvm-commits mailing list