[llvm] f2c99ea - AMDGPU: Use reserved VGPR for AGPR spills to memory

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Tue Feb 8 08:27:05 PST 2022


Author: Matt Arsenault
Date: 2022-02-08T11:26:59-05:00
New Revision: f2c99ea47da3ded530fa94fe96f486338fffc99c

URL: https://github.com/llvm/llvm-project/commit/f2c99ea47da3ded530fa94fe96f486338fffc99c
DIFF: https://github.com/llvm/llvm-project/commit/f2c99ea47da3ded530fa94fe96f486338fffc99c.diff

LOG: AMDGPU: Use reserved VGPR for AGPR spills to memory

Previously would reuse the VGPR used for large frame offsets with the
one needed for copying from the AGPR. Fix this by reusing the register
we already reserved for handling AGPR to AGPR copies.

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
    llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir
    llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
    llvm/test/CodeGen/AMDGPU/pei-build-av-spill.mir
    llvm/test/CodeGen/AMDGPU/pei-build-spill.mir
    llvm/test/CodeGen/AMDGPU/spill-agpr-partially-undef.mir
    llvm/test/CodeGen/AMDGPU/spill-agpr.ll
    llvm/test/CodeGen/AMDGPU/spill-agpr.mir

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
index ab1580d7b296..81742a845bfe 100644
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -1557,16 +1557,8 @@ void SIRegisterInfo::buildSpillLoadStore(
       assert(EltSize == 4);
 
       if (!TmpIntermediateVGPR) {
-        bool AllowSpill = !UseVGPROffset;
-
-        assert(RS && "Needs to have RegScavenger to spill an AGPR!");
-        // FIXME: change to scavengeRegisterBackwards()
-        TmpIntermediateVGPR = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass,
-                                                   MI, 0, AllowSpill);
-        if (!TmpIntermediateVGPR)
-          TmpIntermediateVGPR = TmpOffsetVGPR;
-        else
-          RS->setRegUsed(TmpIntermediateVGPR);
+        assert(MF->getRegInfo().isReserved(AMDGPU::VGPR32));
+        TmpIntermediateVGPR = AMDGPU::VGPR32;
       }
       if (IsStore) {
         auto AccRead = BuildMI(MBB, MI, DL,

diff  --git a/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir b/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir
index d37794137b21..724b09025b1a 100644
--- a/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir
+++ b/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir
@@ -25,8 +25,8 @@ body:             |
   ; GFX908-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
-  ; GFX908-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+  ; GFX908-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
   ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-NEXT: {{  $}}
@@ -516,8 +516,8 @@ body:             |
   ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-FLATSCR-NEXT: {{  $}}
@@ -1035,11 +1035,10 @@ body:             |
   ; GFX908-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
-  ; GFX908-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
-  ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
-  ; GFX908-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
+  ; GFX908-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
+  ; GFX908-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
+  ; GFX908-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
   ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-NEXT: {{  $}}
@@ -1530,12 +1529,10 @@ body:             |
   ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 4, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = SCRATCH_LOAD_DWORD $vgpr0, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 4, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-FLATSCR-NEXT: {{  $}}
@@ -2053,14 +2050,12 @@ body:             |
   ; GFX908-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
-  ; GFX908-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-  ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
-  ; GFX908-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-  ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 8, addrspace 5)
-  ; GFX908-NEXT:   $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
+  ; GFX908-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 8, addrspace 5)
+  ; GFX908-NEXT:   $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
   ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-NEXT: {{  $}}
@@ -2552,16 +2547,12 @@ body:             |
   ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 4, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD $vgpr0, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 8, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = SCRATCH_LOAD_DWORD $vgpr0, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 4, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = SCRATCH_LOAD_DWORD $vgpr0, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 8, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-FLATSCR-NEXT: {{  $}}
@@ -3079,8 +3070,8 @@ body:             |
   ; GFX908-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec
-  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
+  ; GFX908-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr32, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-NEXT: {{  $}}
@@ -3570,8 +3561,8 @@ body:             |
   ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr0, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec
+  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr32, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-FLATSCR-NEXT: {{  $}}
@@ -4088,11 +4079,10 @@ body:             |
   ; GFX908-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.1, addrspace 5)
-  ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
-  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.1 + 4, addrspace 5)
+  ; GFX908-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr32, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.1, addrspace 5)
+  ; GFX908-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr32, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.1 + 4, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-NEXT: {{  $}}
@@ -4583,12 +4573,10 @@ body:             |
   ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr0, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.1, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr0, $vgpr0, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.1 + 4, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr32, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.1, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
+  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr32, $vgpr0, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.1 + 4, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-FLATSCR-NEXT: {{  $}}
@@ -5104,14 +5092,12 @@ body:             |
   ; GFX908-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1, addrspace 5)
-  ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
-  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 4, addrspace 5)
-  ; GFX908-NEXT:   $vgpr0 = V_MOV_B32_e32 8200, implicit $exec
-  ; GFX908-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec
-  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 8, addrspace 5)
+  ; GFX908-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr32, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1, addrspace 5)
+  ; GFX908-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr32, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 4, addrspace 5)
+  ; GFX908-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN $vgpr32, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 8, addrspace 5)
   ; GFX908-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-NEXT: {{  $}}
@@ -5603,16 +5589,12 @@ body:             |
   ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr0, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr0, $vgpr0, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 4, addrspace 5)
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ADD_U32_e32 8200, $vgpr0, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec
-  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr0, $vgpr0, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 8, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr32, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
+  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr32, $vgpr0, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 4, addrspace 5)
+  ; GFX908-FLATSCR-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec
+  ; GFX908-FLATSCR-NEXT:   SCRATCH_STORE_DWORD $vgpr32, $vgpr0, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 8, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
   ; GFX908-FLATSCR-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; GFX908-FLATSCR-NEXT: {{  $}}

diff  --git a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
index 27b844b4a182..8ec375cf93b6 100644
--- a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
+++ b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
@@ -65,80 +65,81 @@ define void @no_free_vgprs_at_agpr_copy(float %v0, float %v1) #0 {
 ; GFX908-NEXT:    v_mfma_f32_16x16x1f32 a[0:15], v33, v34, a[16:31]
 ; GFX908-NEXT:    s_nop 7
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a0 ; Reload Reuse
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a0 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_read_b32 v39, a11 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_read_b32 v38, a12 ; Reload Reuse
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a1 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a1 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_read_b32 v37, a13 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_read_b32 v36, a14 ; Reload Reuse
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a2 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a2 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_read_b32 v35, a15 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 0
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a3 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a3 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a4 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a4 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a5 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a5 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a6 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a6 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a7 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a7 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a8 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a8 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a9 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a9 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
-; GFX908-NEXT:    v_accvgpr_read_b32 v34, a10 ; Reload Reuse
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GFX908-NEXT:    v_accvgpr_read_b32 v32, a10 ; Reload Reuse
 ; GFX908-NEXT:    s_nop 1
-; GFX908-NEXT:    buffer_store_dword v34, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GFX908-NEXT:    buffer_store_dword v32, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
 ; GFX908-NEXT:    ;;#ASMSTART
 ; GFX908-NEXT:    ; copy
 ; GFX908-NEXT:    ;;#ASMEND
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GFX908-NEXT:    v_accvgpr_read_b32 v32, a1
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a0, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX908-NEXT:    s_nop 1
 ; GFX908-NEXT:    v_accvgpr_write_b32 a16, v32
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GFX908-NEXT:    v_accvgpr_write_b32 a0, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a1, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a1, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a2, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a2, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a3, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a3, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a4, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a4, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a5, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a5, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a6, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a6, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a7, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a7, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a8, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a8, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a9, v34 ; Reload Reuse
-; GFX908-NEXT:    buffer_load_dword v34, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GFX908-NEXT:    v_accvgpr_write_b32 a9, v32 ; Reload Reuse
+; GFX908-NEXT:    buffer_load_dword v32, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_accvgpr_write_b32 a10, v34 ; Reload Reuse
+; GFX908-NEXT:    v_accvgpr_write_b32 a10, v32 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_write_b32 a11, v39 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_write_b32 a12, v38 ; Reload Reuse
 ; GFX908-NEXT:    v_accvgpr_write_b32 a13, v37 ; Reload Reuse

diff  --git a/llvm/test/CodeGen/AMDGPU/pei-build-av-spill.mir b/llvm/test/CodeGen/AMDGPU/pei-build-av-spill.mir
index 7514a01ba77e..348c665332f8 100644
--- a/llvm/test/CodeGen/AMDGPU/pei-build-av-spill.mir
+++ b/llvm/test/CodeGen/AMDGPU/pei-build-av-spill.mir
@@ -1625,10 +1625,10 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a1
     ; MUBUF: $agpr0 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a1
     ; MUBUF-V2A: liveins: $vgpr0
@@ -1639,10 +1639,10 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a1
     ; FLATSCR: $agpr0 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a1
     ; FLATSCR-V2A: liveins: $vgpr0
@@ -1695,14 +1695,14 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a2
     ; MUBUF: $agpr0_agpr1 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a2
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1
@@ -1715,14 +1715,14 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a2
     ; FLATSCR: $agpr0_agpr1 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a2
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1
@@ -1783,18 +1783,18 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a3
     ; MUBUF: $agpr0_agpr1_agpr2 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a3
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2
@@ -1809,18 +1809,18 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a3
     ; FLATSCR: $agpr0_agpr1_agpr2 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a3
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2
@@ -1889,22 +1889,22 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a4
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a4
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
@@ -1921,22 +1921,22 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a4
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a4
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
@@ -2013,26 +2013,26 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a5
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a5
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
@@ -2051,26 +2051,26 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a5
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a5
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
@@ -2157,30 +2157,30 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a6
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a6
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
@@ -2201,30 +2201,30 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a6
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a6
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
@@ -2319,34 +2319,34 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a7
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a7
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
@@ -2369,34 +2369,34 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a7
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a7
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
@@ -2499,38 +2499,38 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a8
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a8
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
@@ -2555,38 +2555,38 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a8
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a8
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
@@ -2697,70 +2697,70 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a16
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a16
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15
@@ -2801,70 +2801,70 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a16
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a16
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15
@@ -3043,134 +3043,134 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_av_a32
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 64, addrspace 5)
-    ; MUBUF-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 68, addrspace 5)
-    ; MUBUF-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 72, addrspace 5)
-    ; MUBUF-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 76, addrspace 5)
-    ; MUBUF-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 80, addrspace 5)
-    ; MUBUF-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 84, addrspace 5)
-    ; MUBUF-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 88, addrspace 5)
-    ; MUBUF-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 92, addrspace 5)
-    ; MUBUF-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 96, addrspace 5)
-    ; MUBUF-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 100, addrspace 5)
-    ; MUBUF-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 104, addrspace 5)
-    ; MUBUF-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 108, addrspace 5)
-    ; MUBUF-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 112, addrspace 5)
-    ; MUBUF-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 116, addrspace 5)
-    ; MUBUF-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 120, addrspace 5)
-    ; MUBUF-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 124, addrspace 5)
-    ; MUBUF-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 64, addrspace 5)
+    ; MUBUF-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 68, addrspace 5)
+    ; MUBUF-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 72, addrspace 5)
+    ; MUBUF-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 76, addrspace 5)
+    ; MUBUF-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 80, addrspace 5)
+    ; MUBUF-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 84, addrspace 5)
+    ; MUBUF-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 88, addrspace 5)
+    ; MUBUF-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 92, addrspace 5)
+    ; MUBUF-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 96, addrspace 5)
+    ; MUBUF-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 100, addrspace 5)
+    ; MUBUF-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 104, addrspace 5)
+    ; MUBUF-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 108, addrspace 5)
+    ; MUBUF-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 112, addrspace 5)
+    ; MUBUF-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 116, addrspace 5)
+    ; MUBUF-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 120, addrspace 5)
+    ; MUBUF-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 124, addrspace 5)
+    ; MUBUF-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_av_a32
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31
@@ -3243,134 +3243,134 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_av_a32
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 64, addrspace 5)
-    ; FLATSCR-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 68, addrspace 5)
-    ; FLATSCR-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 72, addrspace 5)
-    ; FLATSCR-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 76, addrspace 5)
-    ; FLATSCR-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 80, addrspace 5)
-    ; FLATSCR-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 84, addrspace 5)
-    ; FLATSCR-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 88, addrspace 5)
-    ; FLATSCR-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 92, addrspace 5)
-    ; FLATSCR-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 96, addrspace 5)
-    ; FLATSCR-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 100, addrspace 5)
-    ; FLATSCR-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 104, addrspace 5)
-    ; FLATSCR-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 108, addrspace 5)
-    ; FLATSCR-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 112, addrspace 5)
-    ; FLATSCR-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 116, addrspace 5)
-    ; FLATSCR-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 120, addrspace 5)
-    ; FLATSCR-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 124, addrspace 5)
-    ; FLATSCR-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 64, addrspace 5)
+    ; FLATSCR-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 68, addrspace 5)
+    ; FLATSCR-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 72, addrspace 5)
+    ; FLATSCR-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 76, addrspace 5)
+    ; FLATSCR-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 80, addrspace 5)
+    ; FLATSCR-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 84, addrspace 5)
+    ; FLATSCR-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 88, addrspace 5)
+    ; FLATSCR-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 92, addrspace 5)
+    ; FLATSCR-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 96, addrspace 5)
+    ; FLATSCR-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 100, addrspace 5)
+    ; FLATSCR-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 104, addrspace 5)
+    ; FLATSCR-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 108, addrspace 5)
+    ; FLATSCR-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 112, addrspace 5)
+    ; FLATSCR-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 116, addrspace 5)
+    ; FLATSCR-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 120, addrspace 5)
+    ; FLATSCR-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 124, addrspace 5)
+    ; FLATSCR-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_av_a32
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31

diff  --git a/llvm/test/CodeGen/AMDGPU/pei-build-spill.mir b/llvm/test/CodeGen/AMDGPU/pei-build-spill.mir
index 981133ac1408..afab15b53658 100644
--- a/llvm/test/CodeGen/AMDGPU/pei-build-spill.mir
+++ b/llvm/test/CodeGen/AMDGPU/pei-build-spill.mir
@@ -1483,10 +1483,10 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a1
     ; MUBUF: $agpr0 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a1
     ; MUBUF-V2A: liveins: $vgpr0
@@ -1497,10 +1497,10 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a1
     ; FLATSCR: $agpr0 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a1
     ; FLATSCR-V2A: liveins: $vgpr0
@@ -1553,14 +1553,14 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a2
     ; MUBUF: $agpr0_agpr1 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a2
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1
@@ -1573,14 +1573,14 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a2
     ; FLATSCR: $agpr0_agpr1 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a2
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1
@@ -1641,18 +1641,18 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a3
     ; MUBUF: $agpr0_agpr1_agpr2 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a3
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2
@@ -1667,18 +1667,18 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a3
     ; FLATSCR: $agpr0_agpr1_agpr2 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a3
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2
@@ -1747,22 +1747,22 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a4
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a4
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
@@ -1779,22 +1779,22 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a4
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a4
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
@@ -1871,26 +1871,26 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a5
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a5
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
@@ -1909,26 +1909,26 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a5
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a5
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
@@ -2015,30 +2015,30 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a6
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a6
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
@@ -2059,30 +2059,30 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a6
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a6
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
@@ -2177,38 +2177,38 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a8
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a8
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
@@ -2233,38 +2233,38 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a8
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a8
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
@@ -2375,70 +2375,70 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a16
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a16
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15
@@ -2479,70 +2479,70 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a16
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a16
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15
@@ -2721,134 +2721,134 @@ body:             |
   bb.0.entry:
     ; MUBUF-LABEL: name: test_spill_a32
     ; MUBUF: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 = IMPLICIT_DEF
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
-    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 64, addrspace 5)
-    ; MUBUF-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 68, addrspace 5)
-    ; MUBUF-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 72, addrspace 5)
-    ; MUBUF-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 76, addrspace 5)
-    ; MUBUF-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 80, addrspace 5)
-    ; MUBUF-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 84, addrspace 5)
-    ; MUBUF-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 88, addrspace 5)
-    ; MUBUF-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 92, addrspace 5)
-    ; MUBUF-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 96, addrspace 5)
-    ; MUBUF-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 100, addrspace 5)
-    ; MUBUF-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 104, addrspace 5)
-    ; MUBUF-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 108, addrspace 5)
-    ; MUBUF-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 112, addrspace 5)
-    ; MUBUF-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 116, addrspace 5)
-    ; MUBUF-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 120, addrspace 5)
-    ; MUBUF-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 124, addrspace 5)
-    ; MUBUF-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
+    ; MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; MUBUF-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; MUBUF-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; MUBUF-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; MUBUF-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; MUBUF-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; MUBUF-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; MUBUF-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; MUBUF-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; MUBUF-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; MUBUF-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; MUBUF-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; MUBUF-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; MUBUF-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; MUBUF-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; MUBUF-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; MUBUF-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 64, addrspace 5)
+    ; MUBUF-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 68, addrspace 5)
+    ; MUBUF-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 72, addrspace 5)
+    ; MUBUF-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 76, addrspace 5)
+    ; MUBUF-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 80, addrspace 5)
+    ; MUBUF-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 84, addrspace 5)
+    ; MUBUF-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 88, addrspace 5)
+    ; MUBUF-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 92, addrspace 5)
+    ; MUBUF-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 96, addrspace 5)
+    ; MUBUF-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 100, addrspace 5)
+    ; MUBUF-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 104, addrspace 5)
+    ; MUBUF-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 108, addrspace 5)
+    ; MUBUF-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 112, addrspace 5)
+    ; MUBUF-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 116, addrspace 5)
+    ; MUBUF-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 120, addrspace 5)
+    ; MUBUF-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; MUBUF-NEXT: $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0 + 124, addrspace 5)
+    ; MUBUF-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
     ; MUBUF-NEXT: S_ENDPGM 0
     ; MUBUF-V2A-LABEL: name: test_spill_a32
     ; MUBUF-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31
@@ -2921,134 +2921,134 @@ body:             |
     ; MUBUF-V2A-NEXT: S_ENDPGM 0
     ; FLATSCR-LABEL: name: test_spill_a32
     ; FLATSCR: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 = IMPLICIT_DEF
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
-    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
-    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
-    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
-    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
-    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
-    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
-    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
-    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
-    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
-    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
-    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
-    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
-    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
-    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
-    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
-    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 64, addrspace 5)
-    ; FLATSCR-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 68, addrspace 5)
-    ; FLATSCR-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 72, addrspace 5)
-    ; FLATSCR-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 76, addrspace 5)
-    ; FLATSCR-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 80, addrspace 5)
-    ; FLATSCR-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 84, addrspace 5)
-    ; FLATSCR-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 88, addrspace 5)
-    ; FLATSCR-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 92, addrspace 5)
-    ; FLATSCR-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 96, addrspace 5)
-    ; FLATSCR-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 100, addrspace 5)
-    ; FLATSCR-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 104, addrspace 5)
-    ; FLATSCR-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 108, addrspace 5)
-    ; FLATSCR-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 112, addrspace 5)
-    ; FLATSCR-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 116, addrspace 5)
-    ; FLATSCR-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 120, addrspace 5)
-    ; FLATSCR-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
-    ; FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 124, addrspace 5)
-    ; FLATSCR-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr2, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr3, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr5, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr6, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr7, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr8, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr9, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr10, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr11, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr12, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr13, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr14, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr15, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr16, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 64, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr17, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 68, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr18, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 72, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr19, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 76, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr20, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 80, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr21, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 84, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr22, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 88, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr23, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 92, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr24, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 96, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr25, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 100, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr26, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 104, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr27, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 108, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr28, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 112, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr29, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 116, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr30, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 120, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr31, implicit $exec
+    ; FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr32, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr, implicit killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (store (s32) into %stack.0 + 124, addrspace 5)
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 4, addrspace 5)
+    ; FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 8, addrspace 5)
+    ; FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 12, addrspace 5)
+    ; FLATSCR-NEXT: $agpr3 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 16, addrspace 5)
+    ; FLATSCR-NEXT: $agpr4 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 20, addrspace 5)
+    ; FLATSCR-NEXT: $agpr5 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 24, addrspace 5)
+    ; FLATSCR-NEXT: $agpr6 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 28, addrspace 5)
+    ; FLATSCR-NEXT: $agpr7 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 32, addrspace 5)
+    ; FLATSCR-NEXT: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 36, addrspace 5)
+    ; FLATSCR-NEXT: $agpr9 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 40, addrspace 5)
+    ; FLATSCR-NEXT: $agpr10 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 44, addrspace 5)
+    ; FLATSCR-NEXT: $agpr11 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 48, addrspace 5)
+    ; FLATSCR-NEXT: $agpr12 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 52, addrspace 5)
+    ; FLATSCR-NEXT: $agpr13 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 56, addrspace 5)
+    ; FLATSCR-NEXT: $agpr14 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 60, addrspace 5)
+    ; FLATSCR-NEXT: $agpr15 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 64, addrspace 5)
+    ; FLATSCR-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 68, addrspace 5)
+    ; FLATSCR-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 72, addrspace 5)
+    ; FLATSCR-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 76, addrspace 5)
+    ; FLATSCR-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 80, addrspace 5)
+    ; FLATSCR-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 84, addrspace 5)
+    ; FLATSCR-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 88, addrspace 5)
+    ; FLATSCR-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 92, addrspace 5)
+    ; FLATSCR-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 96, addrspace 5)
+    ; FLATSCR-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 100, addrspace 5)
+    ; FLATSCR-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 104, addrspace 5)
+    ; FLATSCR-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 108, addrspace 5)
+    ; FLATSCR-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 112, addrspace 5)
+    ; FLATSCR-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 116, addrspace 5)
+    ; FLATSCR-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 120, addrspace 5)
+    ; FLATSCR-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; FLATSCR-NEXT: $vgpr32 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0 + 124, addrspace 5)
+    ; FLATSCR-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
     ; FLATSCR-NEXT: S_ENDPGM 0
     ; FLATSCR-V2A-LABEL: name: test_spill_a32
     ; FLATSCR-V2A: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31

diff  --git a/llvm/test/CodeGen/AMDGPU/spill-agpr-partially-undef.mir b/llvm/test/CodeGen/AMDGPU/spill-agpr-partially-undef.mir
index 494b7587d5f0..6fdbfeebb9a9 100644
--- a/llvm/test/CodeGen/AMDGPU/spill-agpr-partially-undef.mir
+++ b/llvm/test/CodeGen/AMDGPU/spill-agpr-partially-undef.mir
@@ -17,10 +17,10 @@ body:             |
     ; CHECK-LABEL: name: spill_a64_kill
     ; CHECK: liveins: $agpr0_agpr1
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; CHECK-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; CHECK-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; CHECK-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
     SI_SPILL_A64_SAVE killed $agpr0_agpr1, %stack.0, $sgpr32, 0, implicit $exec :: (store (s64) into %stack.0, addrspace 5)
 ...
 
@@ -42,10 +42,10 @@ body:             |
     ; CHECK-LABEL: name: spill_a64_undef_sub1_killed
     ; CHECK: liveins: $agpr0
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; CHECK-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; CHECK-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; CHECK-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
     SI_SPILL_A64_SAVE killed $agpr0_agpr1, %stack.0, $sgpr32, 0, implicit $exec :: (store (s64) into %stack.0, addrspace 5)
 ...
 
@@ -65,9 +65,9 @@ body:             |
     ; CHECK-LABEL: name: spill_a64_undef_sub0_killed
     ; CHECK: liveins: $agpr1
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
-    ; CHECK-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
+    ; CHECK-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec, implicit-def $agpr0_agpr1
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.0, addrspace 5)
+    ; CHECK-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec, implicit killed $agpr0_agpr1 :: (store (s32) into %stack.0 + 4, addrspace 5)
     SI_SPILL_A64_SAVE killed $agpr0_agpr1, %stack.0, $sgpr32, 0, implicit $exec :: (store (s64) into %stack.0, addrspace 5)
 ...

diff  --git a/llvm/test/CodeGen/AMDGPU/spill-agpr.ll b/llvm/test/CodeGen/AMDGPU/spill-agpr.ll
index c51f3e369ba8..6ec439280835 100644
--- a/llvm/test/CodeGen/AMDGPU/spill-agpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/spill-agpr.ll
@@ -74,14 +74,14 @@ use:
 ; GCN: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
 ; GCN: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
 
-; GFX908-DAG:  v_accvgpr_read_b32 v1, a0 ; Reload Reuse
-; GFX908-DAG:  buffer_store_dword v1, off, s[{{[0-9:]+}}], 0 offset:4 ; 4-byte Folded Spill
-; GFX908-DAG:  v_accvgpr_read_b32 v1, a1 ; Reload Reuse
-; GFX908-DAG:  buffer_store_dword v1, off, s[{{[0-9:]+}}], 0 offset:8 ; 4-byte Folded Spill
-; GFX908-DAG:  v_accvgpr_read_b32 v1, a2 ; Reload Reuse
-; GFX908-DAG:  buffer_store_dword v1, off, s[{{[0-9:]+}}], 0 offset:12 ; 4-byte Folded Spill
-; GFX908-DAG:  v_accvgpr_read_b32 v1, a3 ; Reload Reuse
-; GFX908-DAG:  buffer_store_dword v1, off, s[{{[0-9:]+}}], 0 offset:16 ; 4-byte Folded Spill
+; GFX908-DAG:  v_accvgpr_read_b32 v32, a0 ; Reload Reuse
+; GFX908-DAG:  buffer_store_dword v32, off, s[{{[0-9:]+}}], 0 offset:4 ; 4-byte Folded Spill
+; GFX908-DAG:  v_accvgpr_read_b32 v32, a1 ; Reload Reuse
+; GFX908-DAG:  buffer_store_dword v32, off, s[{{[0-9:]+}}], 0 offset:8 ; 4-byte Folded Spill
+; GFX908-DAG:  v_accvgpr_read_b32 v32, a2 ; Reload Reuse
+; GFX908-DAG:  buffer_store_dword v32, off, s[{{[0-9:]+}}], 0 offset:12 ; 4-byte Folded Spill
+; GFX908-DAG:  v_accvgpr_read_b32 v32, a3 ; Reload Reuse
+; GFX908-DAG:  buffer_store_dword v32, off, s[{{[0-9:]+}}], 0 offset:16 ; 4-byte Folded Spill
 
 ; GFX90A-DAG:  buffer_store_dword a0, off, s[{{[0-9:]+}}], 0 offset:4 ; 4-byte Folded Spill
 ; GFX90A-DAG:  buffer_store_dword a1, off, s[{{[0-9:]+}}], 0 offset:8 ; 4-byte Folded Spill

diff  --git a/llvm/test/CodeGen/AMDGPU/spill-agpr.mir b/llvm/test/CodeGen/AMDGPU/spill-agpr.mir
index 3199b96fb636..af92d86200b8 100644
--- a/llvm/test/CodeGen/AMDGPU/spill-agpr.mir
+++ b/llvm/test/CodeGen/AMDGPU/spill-agpr.mir
@@ -308,10 +308,8 @@ body: |
   ; GFX908-EXPANDED-NEXT:   liveins: $vgpr240_vgpr241_vgpr242_vgpr243_vgpr244_vgpr245_vgpr246_vgpr247, $vgpr248_vgpr249_vgpr250_vgpr251_vgpr252_vgpr253_vgpr254_vgpr255, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79, $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95, $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111, $vgpr112_vgpr113_vgpr114_vgpr115_vgpr116_vgpr117_vgpr118_vgpr119_vgpr120_vgpr121_vgpr122_vgpr123_vgpr124_vgpr125_vgpr126_vgpr127, $vgpr128_vgpr129_vgpr130_vgpr131_vgpr132_vgpr133_vgpr134_vgpr135_vgpr136_vgpr137_vgpr138_vgpr139_vgpr140_vgpr141_vgpr142_vgpr143, $vgpr144_vgpr145_vgpr146_vgpr147_vgpr148_vgpr149_vgpr150_vgpr151_vgpr152_vgpr153_vgpr154_vgpr155_vgpr156_vgpr157_vgpr158_vgpr159, $vgpr160_vgpr161_vgpr162_vgpr163_vgpr164_vgpr165_vgpr166_vgpr167_vgpr168_vgpr169_vgpr170_vgpr171_vgpr172_vgpr173_vgpr174_vgpr175, $vgpr176_vgpr177_vgpr178_vgpr179_vgpr180_vgpr181_vgpr182_vgpr183_vgpr184_vgpr185_vgpr186_vgpr187_vgpr188_vgpr189_vgpr190_vgpr191, $vgpr192_vgpr193_vgpr194_vgpr195_vgpr196_vgpr197_vgpr198_vgpr199_vgpr200_vgpr201_vgpr202_vgpr203_vgpr204_vgpr205_vgpr206_vgpr207, $vgpr208_vgpr209_vgpr210_vgpr211_vgpr212_vgpr213_vgpr214_vgpr215_vgpr216_vgpr217_vgpr218_vgpr219_vgpr220_vgpr221_vgpr222_vgpr223, $vgpr224_vgpr225_vgpr226_vgpr227_vgpr228_vgpr229_vgpr230_vgpr231_vgpr232_vgpr233_vgpr234_vgpr235_vgpr236_vgpr237_vgpr238_vgpr239
   ; GFX908-EXPANDED-NEXT: {{  $}}
   ; GFX908-EXPANDED-NEXT:   S_NOP 0, implicit-def renamable $agpr0
-  ; GFX908-EXPANDED-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
-  ; GFX908-EXPANDED-NEXT:   $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
-  ; GFX908-EXPANDED-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-  ; GFX908-EXPANDED-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
+  ; GFX908-EXPANDED-NEXT:   $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec
+  ; GFX908-EXPANDED-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr32, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
   ; GFX908-EXPANDED-NEXT:   S_CBRANCH_SCC1 %bb.1, implicit undef $scc
   ; GFX908-EXPANDED-NEXT: {{  $}}
   ; GFX908-EXPANDED-NEXT: bb.1:
@@ -319,8 +317,8 @@ body: |
   ; GFX908-EXPANDED-NEXT: {{  $}}
   ; GFX908-EXPANDED-NEXT: {{  $}}
   ; GFX908-EXPANDED-NEXT: bb.2:
-  ; GFX908-EXPANDED-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-  ; GFX908-EXPANDED-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
+  ; GFX908-EXPANDED-NEXT:   $vgpr32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+  ; GFX908-EXPANDED-NEXT:   $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr32, implicit $exec
   ; GFX908-EXPANDED-NEXT:   S_NOP 0, implicit undef $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
   ; GFX908-EXPANDED-NEXT:   S_NOP 0, implicit undef $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47
   ; GFX908-EXPANDED-NEXT:   S_NOP 0, implicit undef $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63


        


More information about the llvm-commits mailing list