[llvm-branch-commits] [llvm] AMDGPU: Stop introducing v_accvgpr_write_b32 for reg-to-reg copy (PR #129059)

[Matt Arsenault via llvm-branch-commits]

https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/129059

This was trying to hack around the intermediate VGPR requirement
to copy to AGPRs on gfx908. We should still use a copy for all
reg-to-reg cases. This should matter less these days, as we
reserve a VGPR to handle it when required (and no end to end tests
need updating).

This was also an obstacle to handling this fold for input registers
which are larger than 32-bits.

>From 12e4b8a7d60c043ff5fd442bdf629288d720d271 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Thu, 27 Feb 2025 20:40:52 +0700
Subject: [PATCH] AMDGPU: Stop introducing v_accvgpr_write_b32 for reg-to-reg
 copy

This was trying to hack around the intermediate VGPR requirement
to copy to AGPRs on gfx908. We should still use a copy for all
reg-to-reg cases. This should matter less these days, as we
reserve a VGPR to handle it when required (and no end to end tests
need updating).

This was also an obstacle to handling this fold for input registers
which are larger than 32-bits.
---
 llvm/lib/Target/AMDGPU/SIFoldOperands.cpp      |  5 ++---
 ...si-fold-operands-agpr-copy-reg-sequence.mir | 18 +++++++++---------
 2 files changed, 11 insertions(+), 12 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
index 6cb6863068b5f..eb9aabf8b6317 100644
--- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -1573,9 +1573,8 @@ bool SIFoldOperandsImpl::foldCopyToAGPRRegSequence(MachineInstr *CopyMI) const {
         Vgpr = MRI->createVirtualRegister(&AMDGPU::VGPR_32RegClass);
         BuildMI(MBB, CopyMI, DL, TII->get(AMDGPU::COPY), Vgpr).add(*Def);
       }
-      auto Tmp = MRI->createVirtualRegister(&AMDGPU::AGPR_32RegClass);
-      BuildMI(MBB, CopyMI, DL, TII->get(AMDGPU::V_ACCVGPR_WRITE_B32_e64), Tmp)
-          .addReg(Vgpr);
+      Register Tmp = MRI->createVirtualRegister(&AMDGPU::AGPR_32RegClass);
+      BuildMI(MBB, CopyMI, DL, TII->get(AMDGPU::COPY), Tmp).addReg(Vgpr);
       B.addReg(Tmp);
     }
 
diff --git a/llvm/test/CodeGen/AMDGPU/si-fold-operands-agpr-copy-reg-sequence.mir b/llvm/test/CodeGen/AMDGPU/si-fold-operands-agpr-copy-reg-sequence.mir
index 95112826b7112..493138c933686 100644
--- a/llvm/test/CodeGen/AMDGPU/si-fold-operands-agpr-copy-reg-sequence.mir
+++ b/llvm/test/CodeGen/AMDGPU/si-fold-operands-agpr-copy-reg-sequence.mir
@@ -206,11 +206,11 @@ body:             |
     ; CHECK-LABEL: name: s_mov_b32_999_splat_sgpr_128_copy_vgpr_copy_agpr
     ; CHECK: [[S_MOV_B32_:%[0-9]+]]:sgpr_32 = S_MOV_B32 999
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 [[COPY]], implicit $exec
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_1:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 [[COPY]], implicit $exec
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_2:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 [[COPY]], implicit $exec
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_3:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 [[COPY]], implicit $exec
-    ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:areg_128 = REG_SEQUENCE [[V_ACCVGPR_WRITE_B32_e64_]], %subreg.sub0, [[V_ACCVGPR_WRITE_B32_e64_1]], %subreg.sub1, [[V_ACCVGPR_WRITE_B32_e64_2]], %subreg.sub2, [[V_ACCVGPR_WRITE_B32_e64_3]], %subreg.sub3
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:agpr_32 = COPY [[COPY]]
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:agpr_32 = COPY [[COPY]]
+    ; CHECK-NEXT: [[COPY3:%[0-9]+]]:agpr_32 = COPY [[COPY]]
+    ; CHECK-NEXT: [[COPY4:%[0-9]+]]:agpr_32 = COPY [[COPY]]
+    ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:areg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
     ; CHECK-NEXT: $agpr0_agpr1_agpr2_agpr3 = COPY [[REG_SEQUENCE]]
     ; CHECK-NEXT: S_ENDPGM 0, implicit $agpr0_agpr1_agpr2_agpr3
     %0:sgpr_32 = S_MOV_B32 999
@@ -232,10 +232,10 @@ body:             |
     ; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sgpr_32 = S_MOV_B32 1
     ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 1, implicit $exec
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_1:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 [[COPY]], implicit $exec
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_2:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 [[COPY]], implicit $exec
-    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_3:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 1, implicit $exec
-    ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:areg_128 = REG_SEQUENCE [[V_ACCVGPR_WRITE_B32_e64_]], %subreg.sub0, [[V_ACCVGPR_WRITE_B32_e64_1]], %subreg.sub1, [[V_ACCVGPR_WRITE_B32_e64_2]], %subreg.sub2, [[V_ACCVGPR_WRITE_B32_e64_3]], %subreg.sub3
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:agpr_32 = COPY [[COPY]]
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:agpr_32 = COPY [[COPY]]
+    ; CHECK-NEXT: [[V_ACCVGPR_WRITE_B32_e64_1:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 1, implicit $exec
+    ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:areg_128 = REG_SEQUENCE [[V_ACCVGPR_WRITE_B32_e64_]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[V_ACCVGPR_WRITE_B32_e64_1]], %subreg.sub3
     ; CHECK-NEXT: $agpr0_agpr1_agpr2_agpr3 = COPY [[REG_SEQUENCE]]
     ; CHECK-NEXT: S_ENDPGM 0, implicit $agpr0_agpr1_agpr2_agpr3
     %0:sgpr_32 = S_MOV_B32 999