[llvm] [AArch64] Have isel just do neg directly (PR #145185)

[via llvm-commits]

https://github.com/AZero13 created https://github.com/llvm/llvm-project/pull/145185

This means we do not have to rely on register coalescing to fix this, which matches gcc outputting things like neg on -O0

>From f738b906e2852ea9c8fa563b8fe9fb678e40fd63 Mon Sep 17 00:00:00 2001
From: Rose <gfunni234 at gmail.com>
Date: Sat, 21 Jun 2025 15:05:23 -0400
Subject: [PATCH 1/2] Pre-commit test (NFC)

---
 llvm/test/CodeGen/AArch64/instruct-neg.ll | 12 ++++++++++++
 1 file changed, 12 insertions(+)
 create mode 100644 llvm/test/CodeGen/AArch64/instruct-neg.ll

diff --git a/llvm/test/CodeGen/AArch64/instruct-neg.ll b/llvm/test/CodeGen/AArch64/instruct-neg.ll
new file mode 100644
index 0000000000000..1edce89fb7d90
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/instruct-neg.ll
@@ -0,0 +1,12 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64-linux-gnu -O0 -fast-isel -fast-isel-abort=1 -verify-machineinstrs < %s | FileCheck %s --check-prefix=CHECK
+
+define i32 @negb(i32 %b) {
+; CHECK-LABEL: negb:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, wzr
+; CHECK-NEXT:    sub w0, w8, w0
+; CHECK-NEXT:    ret
+  %sub = sub nsw i32 0, %b
+  ret i32 %sub
+}

>From c4257d359ddc0ad67ae76981f219600f7e8f53f2 Mon Sep 17 00:00:00 2001
From: Rose <gfunni234 at gmail.com>
Date: Sat, 21 Jun 2025 15:36:00 -0400
Subject: [PATCH 2/2] [AArch64] Have isel just do neg directly

This means we do not have to rely on register coalescing to fix this, which matches gcc outputting things like neg on -O0
---
 llvm/lib/Target/AArch64/AArch64FastISel.cpp      | 13 +++++++++++++
 .../AArch64/GISel/AArch64InstructionSelector.cpp | 16 ++++++++++++++++
 llvm/test/CodeGen/AArch64/instruct-neg.ll        |  3 +--
 3 files changed, 30 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64FastISel.cpp b/llvm/lib/Target/AArch64/AArch64FastISel.cpp
index 9d74bb5a8661d..85f9140f34e68 100644
--- a/llvm/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/llvm/lib/Target/AArch64/AArch64FastISel.cpp
@@ -1201,6 +1201,19 @@ Register AArch64FastISel::emitAddSub(bool UseAdd, MVT RetVT, const Value *LHS,
             SI->getOpcode() == Instruction::AShr   )
           std::swap(LHS, RHS);
 
+  // Special case: sub 0, x -> neg x (use zero register directly)
+  if (!UseAdd && isa<Constant>(LHS) && cast<Constant>(LHS)->isNullValue()) {
+    Register RHSReg = getRegForValue(RHS);
+    if (!RHSReg)
+      return Register();
+    
+    if (NeedExtend)
+      RHSReg = emitIntExt(SrcVT, RHSReg, RetVT, IsZExt);
+    
+    Register ZeroReg = RetVT == MVT::i64 ? AArch64::XZR : AArch64::WZR;
+    return emitAddSub_rr(UseAdd, RetVT, ZeroReg, RHSReg, SetFlags, WantResult);
+  }
+
   Register LHSReg = getRegForValue(LHS);
   if (!LHSReg)
     return Register();
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index d55ff5acb3dca..cd6e7aa300043 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -4409,6 +4409,22 @@ MachineInstr *AArch64InstructionSelector::emitAddSub(
   assert((Size == 32 || Size == 64) && "Expected a 32-bit or 64-bit type only");
   bool Is32Bit = Size == 32;
 
+  // Special case: sub 0, x -> neg x (use zero register directly)
+  // Check if this is a SUB operation by examining the base register-register opcode
+  unsigned BaseOpc = AddrModeAndSizeToOpcode[2][Is32Bit];
+  bool IsSubtraction = (BaseOpc == AArch64::SUBWrr || BaseOpc == AArch64::SUBXrr ||
+                        BaseOpc == AArch64::SUBSWrr || BaseOpc == AArch64::SUBSXrr);
+  if (IsSubtraction) {
+    if (auto LHSImm = getIConstantVRegValWithLookThrough(LHS.getReg(), MRI)) {
+      if (LHSImm->Value.isZero()) {
+        // Replace LHS with the appropriate zero register
+        Register ZeroReg = Is32Bit ? AArch64::WZR : AArch64::XZR;
+        MachineOperand ZeroMO = MachineOperand::CreateReg(ZeroReg, false);
+        return emitInstr(BaseOpc, {Dst}, {ZeroMO, RHS}, MIRBuilder);
+      }
+    }
+  }
+
   // INSTRri form with positive arithmetic immediate.
   if (auto Fns = selectArithImmed(RHS))
     return emitInstr(AddrModeAndSizeToOpcode[0][Is32Bit], {Dst}, {LHS},
diff --git a/llvm/test/CodeGen/AArch64/instruct-neg.ll b/llvm/test/CodeGen/AArch64/instruct-neg.ll
index 1edce89fb7d90..4d60fb768901e 100644
--- a/llvm/test/CodeGen/AArch64/instruct-neg.ll
+++ b/llvm/test/CodeGen/AArch64/instruct-neg.ll
@@ -4,8 +4,7 @@
 define i32 @negb(i32 %b) {
 ; CHECK-LABEL: negb:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, wzr
-; CHECK-NEXT:    sub w0, w8, w0
+; CHECK-NEXT:    neg w0, w0
 ; CHECK-NEXT:    ret
   %sub = sub nsw i32 0, %b
   ret i32 %sub