PeepholeOpt: Try to constrain uses to support subregister #161338

arsenm · 2025-09-30T09:20:42Z

This allows removing a special case hack in ARM. ARM's implementation
of getExtractSubregLikeInputs has the strange property that it reports
a register with a class that does not support the reported subregister
index. We can however reconstrain the register to support this usage.

This is an alternative to #159600. I've included the test, but
the output is different. In this case version the VMOVSR is
replaced with an ordinary subregister extract copy.

This allows removing a special case hack in ARM. ARM's implementation of getExtractSubregLikeInputs has the strange property that it reports a register with a class that does not support the reported subregister index. We can however reconstrain the register to support this usage. This is an alternative to #159600. I've included the test, but the output is different. In this case version the VMOVSR is replaced with an ordinary subregister extract copy.

arsenm · 2025-09-30T09:21:01Z

PeepholeOpt: Try to constrain uses to support subregister #161338 👈 (View in Graphite)
main

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-09-30T09:22:18Z

@llvm/pr-subscribers-backend-arm

Author: Matt Arsenault (arsenm)

Changes

This allows removing a special case hack in ARM. ARM's implementation
of getExtractSubregLikeInputs has the strange property that it reports
a register with a class that does not support the reported subregister
index. We can however reconstrain the register to support this usage.

This is an alternative to #159600. I've included the test, but
the output is different. In this case version the VMOVSR is
replaced with an ordinary subregister extract copy.

Full diff: https://github.com/llvm/llvm-project/pull/161338.diff

6 Files Affected:

(modified) llvm/lib/CodeGen/PeepholeOptimizer.cpp (+24)
(modified) llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp (-14)
(modified) llvm/lib/Target/ARM/ARMBaseRegisterInfo.h (-5)
(added) llvm/test/CodeGen/ARM/issue159343.ll (+55)
(added) llvm/test/CodeGen/ARM/pr159343.mir (+31)
(modified) llvm/test/CodeGen/ARM/shouldRewriteCopySrc.ll (+2-2)

diff --git a/llvm/lib/CodeGen/PeepholeOptimizer.cpp b/llvm/lib/CodeGen/PeepholeOptimizer.cpp
index fb3e6482bb096..729a57ef23b1e 100644
--- a/llvm/lib/CodeGen/PeepholeOptimizer.cpp
+++ b/llvm/lib/CodeGen/PeepholeOptimizer.cpp
@@ -1203,6 +1203,18 @@ bool PeepholeOptimizer::optimizeCoalescableCopyImpl(Rewriter &&CpyRewriter) {
     if (!NewSrc.Reg)
       continue;
 
+    if (NewSrc.SubReg) {
+      // Verify the register class supports the subregister index. ARM's
+      // copy-like queries return register:subreg pairs where the register's
+      // current class does not directly support the subregister index.
+      const TargetRegisterClass *RC = MRI->getRegClass(NewSrc.Reg);
+      const TargetRegisterClass *WithSubRC =
+          TRI->getSubClassWithSubReg(RC, NewSrc.SubReg);
+      if (!MRI->constrainRegClass(NewSrc.Reg, WithSubRC))
+        continue;
+      Changed = true;
+    }
+
     // Rewrite source.
     if (CpyRewriter.RewriteCurrentSource(NewSrc.Reg, NewSrc.SubReg)) {
       // We may have extended the live-range of NewSrc, account for that.
@@ -1275,6 +1287,18 @@ MachineInstr &PeepholeOptimizer::rewriteSource(MachineInstr &CopyLike,
   const TargetRegisterClass *DefRC = MRI->getRegClass(Def.Reg);
   Register NewVReg = MRI->createVirtualRegister(DefRC);
 
+  if (NewSrc.SubReg) {
+    const TargetRegisterClass *NewSrcRC = MRI->getRegClass(NewSrc.Reg);
+    const TargetRegisterClass *WithSubRC =
+        TRI->getSubClassWithSubReg(NewSrcRC, NewSrc.SubReg);
+
+    // The new source may not directly support the subregister, but we should be
+    // able to assume it is constrainable to support the subregister (otherwise
+    // ValueTracker was lying and reported a useless value).
+    if (!MRI->constrainRegClass(NewSrc.Reg, WithSubRC))
+      llvm_unreachable("replacement register cannot support subregister");
+  }
+
   MachineInstr *NewCopy =
       BuildMI(*CopyLike.getParent(), &CopyLike, CopyLike.getDebugLoc(),
               TII->get(TargetOpcode::COPY), NewVReg)
diff --git a/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp
index e94220af05a0d..2e8a676269a74 100644
--- a/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp
@@ -960,17 +960,3 @@ bool ARMBaseRegisterInfo::shouldCoalesce(MachineInstr *MI,
   }
   return false;
 }
-
-bool ARMBaseRegisterInfo::shouldRewriteCopySrc(const TargetRegisterClass *DefRC,
-                                               unsigned DefSubReg,
-                                               const TargetRegisterClass *SrcRC,
-                                               unsigned SrcSubReg) const {
-  // We can't extract an SPR from an arbitary DPR (as opposed to a DPR_VFP2).
-  if (DefRC == &ARM::SPRRegClass && DefSubReg == 0 &&
-      SrcRC == &ARM::DPRRegClass &&
-      (SrcSubReg == ARM::ssub_0 || SrcSubReg == ARM::ssub_1))
-    return false;
-
-  return TargetRegisterInfo::shouldRewriteCopySrc(DefRC, DefSubReg,
-                                                  SrcRC, SrcSubReg);
-}
diff --git a/llvm/lib/Target/ARM/ARMBaseRegisterInfo.h b/llvm/lib/Target/ARM/ARMBaseRegisterInfo.h
index 5b67b34089d7e..03b0fa0d1ee08 100644
--- a/llvm/lib/Target/ARM/ARMBaseRegisterInfo.h
+++ b/llvm/lib/Target/ARM/ARMBaseRegisterInfo.h
@@ -158,11 +158,6 @@ class ARMBaseRegisterInfo : public ARMGenRegisterInfo {
                       const TargetRegisterClass *NewRC,
                       LiveIntervals &LIS) const override;
 
-  bool shouldRewriteCopySrc(const TargetRegisterClass *DefRC,
-                            unsigned DefSubReg,
-                            const TargetRegisterClass *SrcRC,
-                            unsigned SrcSubReg) const override;
-
   int getSEHRegNum(unsigned i) const { return getEncodingValue(i); }
 };
 
diff --git a/llvm/test/CodeGen/ARM/issue159343.ll b/llvm/test/CodeGen/ARM/issue159343.ll
new file mode 100644
index 0000000000000..03292582918a9
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/issue159343.ll
@@ -0,0 +1,55 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s | FileCheck %s
+
+; Make sure there's no assertion from peephole-opt introducing illegal
+; subregister index uses.
+
+target triple = "thumbv7-unknown-linux-android29"
+
+define void @_ZN11VersionEdit10DecodeFromEv(i1 %call4, ptr %__profc__ZN11VersionEdit10DecodeFromEv) nounwind {
+; CHECK-LABEL: _ZN11VersionEdit10DecodeFromEv:
+; CHECK:       @ %bb.0: @ %land.rhs.lr.ph
+; CHECK-NEXT:    lsls r0, r0, #31
+; CHECK-NEXT:    beq .LBB0_2
+; CHECK-NEXT:  @ %bb.1:
+; CHECK-NEXT:    adr r0, .LCPI0_0
+; CHECK-NEXT:    vld1.64 {d0, d1}, [r0:128]
+; CHECK-NEXT:    b .LBB0_3
+; CHECK-NEXT:  .LBB0_2: @ %select.false
+; CHECK-NEXT:    vmov.i32 q0, #0x0
+; CHECK-NEXT:  .LBB0_3: @ %select.end
+; CHECK-NEXT:    vldr s5, .LCPI0_1
+; CHECK-NEXT:    vldr s4, .LCPI0_2
+; CHECK-NEXT:    vmov.f32 s6, s0
+; CHECK-NEXT:    vmov.f32 s7, s1
+; CHECK-NEXT:    vst1.64 {d2, d3}, [r1]
+; CHECK-NEXT:    bx lr
+; CHECK-NEXT:    .p2align 4
+; CHECK-NEXT:  @ %bb.4:
+; CHECK-NEXT:  .LCPI0_0:
+; CHECK-NEXT:    .long 1 @ 0x1
+; CHECK-NEXT:    .long 0 @ 0x0
+; CHECK-NEXT:    .long 1 @ 0x1
+; CHECK-NEXT:    .long 0 @ 0x0
+; CHECK-NEXT:  .LCPI0_1:
+; CHECK-NEXT:    .long 0x00000000 @ float 0
+; CHECK-NEXT:  .LCPI0_2:
+; CHECK-NEXT:    .long 0x00000001 @ float 1.40129846E-45
+land.rhs.lr.ph:
+  br i1 %call4, label %sw.bb, label %while.cond.while.end_crit_edge.split.loop.exit43
+
+while.cond.while.end_crit_edge.split.loop.exit43: ; preds = %land.rhs.lr.ph
+  %ext0 = extractelement <4 x i64> zeroinitializer, i64 0
+  br label %while.cond.while.end_crit_edge
+
+while.cond.while.end_crit_edge:                   ; preds = %sw.bb, %while.cond.while.end_crit_edge.split.loop.exit43
+  %pgocount5374.ph = phi i64 [ %ext1, %sw.bb ], [ %ext0, %while.cond.while.end_crit_edge.split.loop.exit43 ]
+  %ins = insertelement <2 x i64> splat (i64 1), i64 %pgocount5374.ph, i64 1
+  store <2 x i64> %ins, ptr %__profc__ZN11VersionEdit10DecodeFromEv, align 8
+  ret void
+
+sw.bb:                                            ; preds = %land.rhs.lr.ph
+  %ext1 = extractelement <4 x i64> splat (i64 1), i64 0
+  br label %while.cond.while.end_crit_edge
+}
+
diff --git a/llvm/test/CodeGen/ARM/pr159343.mir b/llvm/test/CodeGen/ARM/pr159343.mir
new file mode 100644
index 0000000000000..9b71b1ad94b2f
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/pr159343.mir
@@ -0,0 +1,31 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -run-pass=peephole-opt -verify-machineinstrs -mtriple=thumbv7-unknown-linux-android29 %s -o - | FileCheck %s
+---
+name: Test_shouldRewriteCopySrc_Invalid_SubReg
+tracksRegLiveness: true
+body:             |
+  bb.1:
+    liveins: $r0, $r1
+
+    ; CHECK-LABEL: name: Test_shouldRewriteCopySrc_Invalid_SubReg
+    ; CHECK: liveins: $r0, $r1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:dpair = IMPLICIT_DEF
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:dpr_vfp2 = COPY [[DEF]].dsub_0
+    ; CHECK-NEXT: [[VMOVRRD:%[0-9]+]]:gpr, [[VMOVRRD1:%[0-9]+]]:gpr = VMOVRRD [[COPY]], 14 /* CC::al */, $noreg
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:spr = COPY [[COPY]].ssub_1
+    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:spr = IMPLICIT_DEF
+    ; CHECK-NEXT: [[DEF2:%[0-9]+]]:spr = IMPLICIT_DEF
+    ; CHECK-NEXT: [[DEF3:%[0-9]+]]:spr = IMPLICIT_DEF
+    ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:mqpr = REG_SEQUENCE killed [[DEF2]], %subreg.ssub_0, killed [[DEF1]], %subreg.ssub_1, killed [[DEF3]], %subreg.ssub_2, [[COPY]].ssub_1, %subreg.ssub_3
+    ; CHECK-NEXT: VST1q64 $r1, 0, killed [[REG_SEQUENCE]], 14 /* CC::al */, $noreg
+    %0:dpair = IMPLICIT_DEF
+    %1:dpr = COPY %0.dsub_0
+    %2:gpr, %3:gpr = VMOVRRD killed %1, 14 /* CC::al */, $noreg
+    %4:spr = VMOVSR killed %3, 14 /* CC::al */, $noreg
+    %5:spr = IMPLICIT_DEF
+    %6:spr = IMPLICIT_DEF
+    %7:spr = IMPLICIT_DEF
+    %8:mqpr = REG_SEQUENCE killed %6, %subreg.ssub_0, killed %5, %subreg.ssub_1, killed %7, %subreg.ssub_2, killed %4, %subreg.ssub_3
+    VST1q64 $r1, 0, killed %8, 14 /* CC::al */, $noreg
+...
diff --git a/llvm/test/CodeGen/ARM/shouldRewriteCopySrc.ll b/llvm/test/CodeGen/ARM/shouldRewriteCopySrc.ll
index e653aaa316fed..2bf8f29eccb40 100644
--- a/llvm/test/CodeGen/ARM/shouldRewriteCopySrc.ll
+++ b/llvm/test/CodeGen/ARM/shouldRewriteCopySrc.ll
@@ -12,8 +12,8 @@ define float @shouldRewriteCopySrc(double %arg) #0 {
 ; CHECK-NEXT:    @APP
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    @NO_APP
-; CHECK-NEXT:    vmov r0, r1, d16
-; CHECK-NEXT:    vmov s0, r0
+; CHECK-NEXT:    vmov.f64 d0, d16
+; CHECK-NEXT:    @ kill: def $s0 killed $s0 killed $d0
 ; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
 ; CHECK-NEXT:    bx lr
 bb:

This allows removing a special case hack in ARM. ARM's implementation of getExtractSubregLikeInputs has the strange property that it reports a register with a class that does not support the reported subregister index. We can however reconstrain the register to support this usage. This is an alternative to llvm#159600. I've included the test, but the output is different. In this case version the VMOVSR is replaced with an ordinary subregister extract copy.

arsenm added the llvm:codegen label Sep 30, 2025 — with Graphite App

arsenm requested review from MacDue, frederik-h, jayfoad, jrbyrnes, lukel97, perlfu, preames, qcolombet and sdesmalen-arm September 30, 2025 09:21

arsenm marked this pull request as ready for review September 30, 2025 09:21

llvmbot added the backend:ARM label Sep 30, 2025

sdesmalen-arm approved these changes Sep 30, 2025

View reviewed changes

sdesmalen-arm mentioned this pull request Sep 30, 2025

[CodeGen][ARM] Test validity of SubRC + SubReg in shouldRewriteCopySrc #159600

Closed

arsenm merged commit 9811226 into main Sep 30, 2025
14 checks passed

arsenm deleted the users/arsenm/peephole-opt/try-constraining-subreg-uses branch September 30, 2025 15:18

arsenm mentioned this pull request Sep 30, 2025

VirtRegRewriter::rewrite(): Assertion `PhysReg.isValid() && "Invalid SubReg for physical register"' failed. #159343

Closed

aemerson mentioned this pull request Oct 16, 2025

[clang-reorder-fields] Check for flexible array member (#160262) #163697

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

PeepholeOpt: Try to constrain uses to support subregister #161338

PeepholeOpt: Try to constrain uses to support subregister #161338

Uh oh!

arsenm commented Sep 30, 2025

Uh oh!

arsenm commented Sep 30, 2025

Uh oh!

llvmbot commented Sep 30, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

PeepholeOpt: Try to constrain uses to support subregister #161338

PeepholeOpt: Try to constrain uses to support subregister #161338

Uh oh!

Conversation

arsenm commented Sep 30, 2025

Uh oh!

arsenm commented Sep 30, 2025

Uh oh!

llvmbot commented Sep 30, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants