[llvm] [AMDGPU] Fix sgpr to vreg_1 copy analysis (PR #149181)
Jorn Tuyls via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 17 00:39:26 PDT 2025
https://github.com/jtuyls updated https://github.com/llvm/llvm-project/pull/149181
>From 3f8c9b20ec4750048b4ad6e7361472ce0eff282a Mon Sep 17 00:00:00 2001
From: Jorn Tuyls <jorn.tuyls at gmail.com>
Date: Wed, 16 Jul 2025 15:48:22 +0000
Subject: [PATCH] [AMDGPU] Fix sgpr to vreg_1 copy
---
llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp | 19 ++--
.../test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.ll | 104 ++++++++++++++++++
.../CodeGen/AMDGPU/sgpr-to-vreg1-copy.mir | 31 ++++++
3 files changed, 147 insertions(+), 7 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.ll
create mode 100644 llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.mir
diff --git a/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp b/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
index 9b5a46395695d..343d203cbe0de 100644
--- a/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
@@ -946,13 +946,18 @@ void SIFixSGPRCopies::analyzeVGPRToSGPRCopy(MachineInstr* MI) {
// Copies and REG_SEQUENCE do not contribute to the final assembly
// So, skip them but take care of the SGPR to VGPR copies bookkeeping.
- if (Inst->isCopy() || Inst->isRegSequence()) {
- if (TRI->isVGPR(*MRI, Inst->getOperand(0).getReg())) {
- if (!Inst->isCopy() ||
- !tryChangeVGPRtoSGPRinCopy(*Inst, TRI, TII)) {
- Info.NumSVCopies++;
- continue;
- }
+ if (Inst->isRegSequence() &&
+ TRI->isVGPR(*MRI, Inst->getOperand(0).getReg())) {
+ Info.NumSVCopies++;
+ continue;
+ }
+ if (Inst->isCopy()) {
+ const TargetRegisterClass *SrcRC, *DstRC;
+ std::tie(SrcRC, DstRC) = getCopyRegClasses(*Inst, *TRI, *MRI);
+ if (isSGPRToVGPRCopy(SrcRC, DstRC, *TRI) &&
+ !tryChangeVGPRtoSGPRinCopy(*Inst, TRI, TII)) {
+ Info.NumSVCopies++;
+ continue;
}
}
diff --git a/llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.ll b/llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.ll
new file mode 100644
index 0000000000000..f3c5e5ec1daff
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.ll
@@ -0,0 +1,104 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GCN %s
+
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128:128:48-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
+target triple = "amdgcn-amd-amdhsa"
+
+define amdgpu_kernel void @copy_to_vreg_1(i32 %0) {
+; GCN-LABEL: copy_to_vreg_1:
+; GCN: ; %bb.0: ; %._crit_edge
+; GCN-NEXT: s_load_dword s4, s[4:5], 0x24
+; GCN-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; GCN-NEXT: v_mov_b64_e32 v[2:3], 0
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_sub_i32 s5, 1, s4
+; GCN-NEXT: s_cmp_lt_u32 s4, 2
+; GCN-NEXT: s_cselect_b64 s[0:1], -1, 0
+; GCN-NEXT: s_and_b64 s[2:3], s[0:1], exec
+; GCN-NEXT: s_cselect_b32 s3, s5, 1
+; GCN-NEXT: s_cmp_lg_u64 s[0:1], 0
+; GCN-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[0:1]
+; GCN-NEXT: s_addc_u32 s0, 1, 0
+; GCN-NEXT: v_readfirstlane_b32 s2, v1
+; GCN-NEXT: s_cmp_ge_u32 s3, s4
+; GCN-NEXT: s_cselect_b32 s4, s0, s2
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cmp_lg_u64 0, 0
+; GCN-NEXT: s_mov_b64 s[0:1], 0
+; GCN-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
+; GCN-NEXT: s_cselect_b64 s[2:3], -1, 0
+; GCN-NEXT: s_branch .LBB0_3
+; GCN-NEXT: .LBB0_1: ; %Flow
+; GCN-NEXT: ; in Loop: Header=BB0_3 Depth=1
+; GCN-NEXT: s_or_b64 exec, exec, s[6:7]
+; GCN-NEXT: s_xor_b64 s[8:9], exec, -1
+; GCN-NEXT: .LBB0_2: ; %Flow2
+; GCN-NEXT: ; in Loop: Header=BB0_3 Depth=1
+; GCN-NEXT: s_or_b64 exec, exec, s[4:5]
+; GCN-NEXT: s_and_b64 s[4:5], exec, s[8:9]
+; GCN-NEXT: s_or_b64 s[0:1], s[4:5], s[0:1]
+; GCN-NEXT: s_mov_b32 s4, 0
+; GCN-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GCN-NEXT: s_cbranch_execz .LBB0_8
+; GCN-NEXT: .LBB0_3: ; %.lr.ph27
+; GCN-NEXT: ; =>This Inner Loop Header: Depth=1
+; GCN-NEXT: s_cmp_lg_u32 s4, 0
+; GCN-NEXT: s_cselect_b64 s[4:5], -1, 0
+; GCN-NEXT: s_or_b64 s[8:9], vcc, s[4:5]
+; GCN-NEXT: s_xor_b64 s[6:7], s[8:9], -1
+; GCN-NEXT: s_and_saveexec_b64 s[4:5], s[8:9]
+; GCN-NEXT: s_cbranch_execz .LBB0_5
+; GCN-NEXT: ; %bb.4: ; %pred.store.if
+; GCN-NEXT: ; in Loop: Header=BB0_3 Depth=1
+; GCN-NEXT: s_or_b64 s[6:7], s[6:7], exec
+; GCN-NEXT: global_store_byte v[2:3], v1, off
+; GCN-NEXT: .LBB0_5: ; %Flow1
+; GCN-NEXT: ; in Loop: Header=BB0_3 Depth=1
+; GCN-NEXT: s_or_b64 exec, exec, s[4:5]
+; GCN-NEXT: s_mov_b64 s[8:9], -1
+; GCN-NEXT: s_and_saveexec_b64 s[4:5], s[6:7]
+; GCN-NEXT: s_cbranch_execz .LBB0_2
+; GCN-NEXT: ; %bb.6: ; %pred.store.continue
+; GCN-NEXT: ; in Loop: Header=BB0_3 Depth=1
+; GCN-NEXT: s_and_saveexec_b64 s[6:7], s[2:3]
+; GCN-NEXT: s_xor_b64 s[6:7], exec, s[6:7]
+; GCN-NEXT: s_cbranch_execz .LBB0_1
+; GCN-NEXT: ; %bb.7: ; %pred.store.if41
+; GCN-NEXT: ; in Loop: Header=BB0_3 Depth=1
+; GCN-NEXT: global_store_byte v[2:3], v1, off
+; GCN-NEXT: s_branch .LBB0_1
+; GCN-NEXT: .LBB0_8: ; %DummyReturnBlock
+; GCN-NEXT: s_endpgm
+._crit_edge:
+ %1 = tail call i32 @llvm.amdgcn.workitem.id.x()
+ %2 = udiv i32 1, %0
+ br label %.lr.ph27
+
+.lr.ph27: ; preds = %pred.store.if41, %pred.store.continue, %._crit_edge
+ %3 = phi i32 [ %2, %._crit_edge ], [ 0, %pred.store.if41 ], [ 0, %pred.store.continue ]
+ %4 = icmp ugt i32 %3, 0
+ %broadcast.splatinsert37 = insertelement <4 x i1> zeroinitializer, i1 %4, i64 0
+ %.zext = zext i32 %1 to i64
+ %broadcast.splatinsert39 = insertelement <4 x i64> zeroinitializer, i64 %.zext, i64 0
+ %5 = icmp uge <4 x i64> %broadcast.splatinsert39, splat (i64 1)
+ %6 = or <4 x i1> %5, %broadcast.splatinsert37
+ %7 = extractelement <4 x i1> %6, i64 0
+ br i1 %7, label %pred.store.if, label %pred.store.continue
+
+pred.store.if: ; preds = %.lr.ph27
+ store i8 0, ptr addrspace(1) null, align 64
+ br label %pred.store.continue
+
+pred.store.continue: ; preds = %pred.store.if, %.lr.ph27
+ %8 = extractelement <4 x i1> %6, i64 1
+ br i1 %8, label %pred.store.if41, label %.lr.ph27
+
+pred.store.if41: ; preds = %pred.store.continue
+ store i8 0, ptr addrspace(1) null, align 64
+ br label %.lr.ph27
+}
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x() #0
+
+attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
diff --git a/llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.mir b/llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.mir
new file mode 100644
index 0000000000000..2daea2b2eeb74
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/sgpr-to-vreg1-copy.mir
@@ -0,0 +1,31 @@
+# RUN: llc -mtriple=amdgcn -run-pass si-fix-sgpr-copies -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
+
+---
+name: copy_to_vreg_1
+tracksRegLiveness: true
+body: |
+ ; GCN-LABEL: name: copy_to_vreg_1
+ ; GCN: bb.0:
+ ; GCN-NEXT: successors: %bb.1(0x80000000)
+ ; GCN-NEXT: liveins: $vgpr0, $vgpr1
+ ; GCN-NEXT: {{ $}}
+ ; GCN-NEXT: [[V_CVT_U32_F32_e64:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, killed $vgpr0, 0, 0, implicit $mode, implicit $exec
+ ; GCN-NEXT: [[IMPLICIT_DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+ ; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GCN-NEXT: [[V_CMP_GT_U32_e64:%[0-9]+]]:sreg_64_xexec = samesign V_CMP_GT_U32_e64 [[V_CVT_U32_F32_e64]], killed [[COPY1]], implicit $exec
+ ; GCN-NEXT: [[VREG1:%[0-9]+]]:vreg_1 = COPY [[V_CMP_GT_U32_e64]]
+ ; GCN-NEXT: {{ $}}
+ ; GCN-NEXT: bb.1:
+ ; GCN-NEXT: S_ENDPGM 0
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ %0:vgpr_32 = nofpexcept V_CVT_U32_F32_e64 0, killed $vgpr0, 0, 0, implicit $mode, implicit $exec
+ %1:sreg_32 = COPY %0:vgpr_32
+ %2:sreg_32 = COPY $vgpr1
+ samesign S_CMP_GT_U32 %1:sreg_32, killed %2:sreg_32, implicit-def $scc
+ %3:sreg_64 = COPY $scc
+ %4:vreg_1 = COPY %3:sreg_64
+
+ bb.1:
+ S_ENDPGM 0
+...
More information about the llvm-commits
mailing list