From 634b8b0117493f4dbd45d6e87f407f8373e20642 Mon Sep 17 00:00:00 2001 From: Saleem Abdulrasool Date: Fri, 9 Oct 2015 03:19:03 +0000 Subject: [PATCH] ARM: tweak WoA frame lowering Accept r11 when targeting Windows on ARM rather than just low registers. Because we are in a thumb-2 only mode, this may be slightly more expensive in code size, but results in better code for the environment since it spills the frame register, which is generally desired for fast stack walking as per the ABI. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249804 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/ARM/ARMFrameLowering.cpp | 16 ++++++++-------- test/CodeGen/ARM/Windows/no-frame-register.ll | 22 ++++++++++++++++++++++ 2 files changed, 30 insertions(+), 8 deletions(-) create mode 100644 test/CodeGen/ARM/Windows/no-frame-register.ll diff --git a/lib/Target/ARM/ARMFrameLowering.cpp b/lib/Target/ARM/ARMFrameLowering.cpp index 476a12fa287..802946aa8d9 100644 --- a/lib/Target/ARM/ARMFrameLowering.cpp +++ b/lib/Target/ARM/ARMFrameLowering.cpp @@ -1605,13 +1605,11 @@ void ARMFrameLowering::determineCalleeSaves(MachineFunction &MF, // FIXME: We could add logic to be more precise about negative offsets // and which instructions will need a scratch register for them. Is it // worth the effort and added fragility? - bool BigStack = - (RS && - (MFI->estimateStackSize(MF) + - ((hasFP(MF) && AFI->hasStackFrame()) ? 4:0) >= - estimateRSStackSizeLimit(MF, this))) - || MFI->hasVarSizedObjects() - || (MFI->adjustsStack() && !canSimplifyCallFramePseudos(MF)); + bool BigStack = (RS && (MFI->estimateStackSize(MF) + + ((hasFP(MF) && AFI->hasStackFrame()) ? 4 : 0) >= + estimateRSStackSizeLimit(MF, this))) || + MFI->hasVarSizedObjects() || + (MFI->adjustsStack() && !canSimplifyCallFramePseudos(MF)); bool ExtraCSSpill = false; if (BigStack || !CanEliminateFrame || RegInfo->cannotEliminateFrame(MF)) { @@ -1649,8 +1647,10 @@ void ARMFrameLowering::determineCalleeSaves(MachineFunction &MF, if (CS1Spilled && !UnspilledCS1GPRs.empty()) { for (unsigned i = 0, e = UnspilledCS1GPRs.size(); i != e; ++i) { unsigned Reg = UnspilledCS1GPRs[i]; - // Don't spill high register if the function is thumb + // Don't spill high register if the function is thumb. In the case of + // Windows on ARM, accept R11 (frame pointer) if (!AFI->isThumbFunction() || + (STI.isTargetWindows() && Reg == ARM::R11) || isARMLowRegister(Reg) || Reg == ARM::LR) { SavedRegs.set(Reg); if (!MRI.isReserved(Reg)) diff --git a/test/CodeGen/ARM/Windows/no-frame-register.ll b/test/CodeGen/ARM/Windows/no-frame-register.ll new file mode 100644 index 00000000000..80187af7ef2 --- /dev/null +++ b/test/CodeGen/ARM/Windows/no-frame-register.ll @@ -0,0 +1,22 @@ +; RUN: llc -mtriple thumbv7-windows -filetype asm -o - %s | FileCheck %s + +declare void @callee(i32) + +define i32 @calleer(i32 %i) { +entry: + %i.addr = alloca i32, align 4 + %j = alloca i32, align 4 + store i32 %i, i32* %i.addr, align 4 + %0 = load i32, i32* %i.addr, align 4 + %add = add nsw i32 %0, 1 + store i32 %add, i32* %j, align 4 + %1 = load i32, i32* %j, align 4 + call void @callee(i32 %1) + %2 = load i32, i32* %j, align 4 + %add1 = add nsw i32 %2, 1 + ret i32 %add1 +} + +; CHECK-NOT: push.w {r7, lr} +; CHECK: push.w {r11, lr} + -- 2.11.0