diff options
| -rw-r--r-- | llvm/lib/Target/X86/X86FrameLowering.cpp | 9 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/merge-sp-update-lea.ll | 32 |
2 files changed, 39 insertions, 2 deletions
diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp index a6d395b92a0..c394383dce0 100644 --- a/llvm/lib/Target/X86/X86FrameLowering.cpp +++ b/llvm/lib/Target/X86/X86FrameLowering.cpp @@ -376,12 +376,17 @@ int X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB, int Offset = 0; if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 || - Opc == X86::ADD32ri || Opc == X86::ADD32ri8 || - Opc == X86::LEA32r || Opc == X86::LEA64_32r) && + Opc == X86::ADD32ri || Opc == X86::ADD32ri8) && PI->getOperand(0).getReg() == StackPtr){ Offset += PI->getOperand(2).getImm(); MBB.erase(PI); if (!doMergeWithPrevious) MBBI = NI; + } else if ((Opc == X86::LEA32r || Opc == X86::LEA64_32r) && + PI->getOperand(0).getReg() == StackPtr) { + // For LEAs we have: def = lea SP, FI, noreg, Offset, noreg. + Offset += PI->getOperand(4).getImm(); + MBB.erase(PI); + if (!doMergeWithPrevious) MBBI = NI; } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 || Opc == X86::SUB32ri || Opc == X86::SUB32ri8) && PI->getOperand(0).getReg() == StackPtr) { diff --git a/llvm/test/CodeGen/X86/merge-sp-update-lea.ll b/llvm/test/CodeGen/X86/merge-sp-update-lea.ll new file mode 100644 index 00000000000..cd2be44e1c0 --- /dev/null +++ b/llvm/test/CodeGen/X86/merge-sp-update-lea.ll @@ -0,0 +1,32 @@ +; RUN: llc %s -o - | FileCheck %s +target datalayout = "e-m:o-p:32:32-f64:32:64-f80:128-n8:16:32-S128" +target triple = "i386-apple-macosx" + +; Check that the merging of SP updates, when LEAs are involved, happen +; correctly. +; CHECK-LABEL: useLEA: +; CHECK: calll L_realloc +; Make sure that the offset we get here is 8 + 16. +; We used to have 8 + 1 because we were not reading the right immediate form +; the LEA instruction. +; CHECK-NEXT: leal 24(%esp), %esp +define noalias i8* @useLEA(i8* nocapture %p, i32 %nbytes) #0 { +entry: + %cmp = icmp slt i32 %nbytes, 0 + br i1 %cmp, label %cond.end.3, label %cond.false + +cond.false: ; preds = %entry + %tobool = icmp ne i32 %nbytes, 0 + %cond = select i1 %tobool, i32 %nbytes, i32 1 + %call = tail call i8* @realloc(i8* %p, i32 %cond) + br label %cond.end.3 + +cond.end.3: ; preds = %entry, %cond.false + %cond4 = phi i8* [ %call, %cond.false ], [ null, %entry ] + ret i8* %cond4 +} + +; Function Attrs: nounwind optsize +declare noalias i8* @realloc(i8* nocapture, i32) + +attributes #0 = { nounwind optsize ssp "disable-tail-calls"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "target-features"="+lea-sp" } |
