From 526522728ccac69dc8df2aa91735ec5c7495fa20 Mon Sep 17 00:00:00 2001 From: Reed Kotler Date: Tue, 10 Dec 2013 14:29:38 +0000 Subject: [PATCH] Next step in Mips16 prologue/epilogue cleanup. Save S2(reg 18) only when we are calling floating point stubs that have a return value of float or complex. Some more work to make this better but this is the first step. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196921 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/Mips/Mips16FrameLowering.cpp | 27 +++++-- lib/Target/Mips/Mips16HardFloat.cpp | 14 ++-- lib/Target/Mips/Mips16InstrInfo.cpp | 30 +++++--- lib/Target/Mips/MipsRegisterInfo.cpp | 3 + test/CodeGen/Mips/align16.ll | 4 +- test/CodeGen/Mips/alloca16.ll | 2 +- test/CodeGen/Mips/ex2.ll | 7 +- test/CodeGen/Mips/helloworld.ll | 4 +- test/CodeGen/Mips/s2rem.ll | 95 +++++++++++++++++++++++++ 9 files changed, 159 insertions(+), 27 deletions(-) create mode 100644 test/CodeGen/Mips/s2rem.ll diff --git a/lib/Target/Mips/Mips16FrameLowering.cpp b/lib/Target/Mips/Mips16FrameLowering.cpp index 6655ff98e03..ae6be05cdfb 100644 --- a/lib/Target/Mips/Mips16FrameLowering.cpp +++ b/lib/Target/Mips/Mips16FrameLowering.cpp @@ -15,6 +15,7 @@ #include "MCTargetDesc/MipsBaseInfo.h" #include "Mips16InstrInfo.h" #include "MipsInstrInfo.h" +#include "MipsRegisterInfo.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" @@ -56,17 +57,31 @@ void Mips16FrameLowering::emitPrologue(MachineFunction &MF) const { MCSymbol *CSLabel = MMI.getContext().CreateTempSymbol(); BuildMI(MBB, MBBI, dl, TII.get(TargetOpcode::PROLOG_LABEL)).addSym(CSLabel); - unsigned S2 = MRI->getDwarfRegNum(Mips::S2, true); - MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, S2, -8)); + + + const MipsRegisterInfo &RI = TII.getRegisterInfo(); + const BitVector Reserved = RI.getReservedRegs(MF); + bool SaveS2 = Reserved[Mips::S2]; + int Offset=-4; + unsigned RA = MRI->getDwarfRegNum(Mips::RA, true); + MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, RA, Offset)); + Offset -= 4; + + if (SaveS2) { + unsigned S2 = MRI->getDwarfRegNum(Mips::S2, true); + MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, S2, Offset)); + Offset -= 4; + } + unsigned S1 = MRI->getDwarfRegNum(Mips::S1, true); - MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, S1, -12)); + MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, S1, Offset)); + Offset -= 4; unsigned S0 = MRI->getDwarfRegNum(Mips::S0, true); - MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, S0, -16)); + MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, S0, Offset)); + - unsigned RA = MRI->getDwarfRegNum(Mips::RA, true); - MMI.addFrameInst(MCCFIInstruction::createOffset(CSLabel, RA, -4)); if (hasFP(MF)) BuildMI(MBB, MBBI, dl, TII.get(Mips::MoveR3216), Mips::S0) diff --git a/lib/Target/Mips/Mips16HardFloat.cpp b/lib/Target/Mips/Mips16HardFloat.cpp index 81bf18cd09d..65dbd74a1c8 100644 --- a/lib/Target/Mips/Mips16HardFloat.cpp +++ b/lib/Target/Mips/Mips16HardFloat.cpp @@ -400,13 +400,19 @@ static bool fixupFPReturnAndCall Value *F = (M->getOrInsertFunction(Name, A, MyVoid, T, NULL)); CallInst::Create(F, Params, "", &Inst ); } else if (const CallInst *CI = dyn_cast(I)) { + Function *F_ = CI->getCalledFunction(); + if (F_ && !isIntrinsicInline(F_)) { // pic mode calls are handled by already defined // helper functions - if (Subtarget.getRelocationModel() != Reloc::PIC_ ) { - Function *F_ = CI->getCalledFunction(); - if (F_ && !isIntrinsicInline(F_) && needsFPHelperFromSig(*F_)) { - assureFPCallStub(*F_, M, Subtarget); + if (needsFPReturnHelper(*F_)) { Modified=true; + F.addFnAttr("saveS2"); + } + if (Subtarget.getRelocationModel() != Reloc::PIC_ ) { + if (needsFPHelperFromSig(*F_)) { + assureFPCallStub(*F_, M, Subtarget); + Modified=true; + } } } } diff --git a/lib/Target/Mips/Mips16InstrInfo.cpp b/lib/Target/Mips/Mips16InstrInfo.cpp index 3736358bb6d..2df83e88506 100644 --- a/lib/Target/Mips/Mips16InstrInfo.cpp +++ b/lib/Target/Mips/Mips16InstrInfo.cpp @@ -173,22 +173,29 @@ void Mips16InstrInfo::makeFrame(unsigned SP, int64_t FrameSize, MachineBasicBlock &MBB, MachineBasicBlock::iterator I) const { DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); + const BitVector Reserved = RI.getReservedRegs(*MBB.getParent()); + bool SaveS2 = Reserved[Mips::S2]; + MachineInstrBuilder MIB; if (isUInt<11>(FrameSize)) - BuildMI(MBB, I, DL, get(Mips::SaveX16)).addReg(Mips::RA). + MIB = BuildMI( + MBB, I, DL, get(Mips::SaveX16)).addReg(Mips::RA). addReg(Mips::S0). - addReg(Mips::S1).addReg(Mips::S2).addImm(FrameSize); + addReg(Mips::S1).addImm(FrameSize); else { int Base = 2040; // should create template function like isUInt that // returns largest possible n bit unsigned integer int64_t Remainder = FrameSize - Base; - BuildMI(MBB, I, DL, get(Mips::SaveX16)).addReg(Mips::RA). + MIB = BuildMI( + MBB, I, DL, get(Mips::SaveX16)).addReg(Mips::RA). addReg(Mips::S0). - addReg(Mips::S1).addReg(Mips::S2).addImm(Base); + addReg(Mips::S1).addImm(Base); if (isInt<16>(-Remainder)) BuildAddiuSpImm(MBB, I, -Remainder); else adjustStackPtrBig(SP, -Remainder, MBB, I, Mips::V0, Mips::V1); } + if (SaveS2) + MIB.addReg(Mips::S2); } // Adjust SP by FrameSize bytes. Restore RA, S0, S1 @@ -196,12 +203,16 @@ void Mips16InstrInfo::restoreFrame(unsigned SP, int64_t FrameSize, MachineBasicBlock &MBB, MachineBasicBlock::iterator I) const { DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); + const BitVector Reserved = RI.getReservedRegs(*MBB.getParent()); + bool SaveS2 = Reserved[Mips::S2]; + MachineInstrBuilder MIB; if (isUInt<11>(FrameSize)) - BuildMI(MBB, I, DL, get(Mips::RestoreX16)). + MIB = BuildMI( + MBB, I, DL, get(Mips::RestoreX16)). addReg(Mips::RA, RegState::Define). addReg(Mips::S0, RegState::Define). addReg(Mips::S1, RegState::Define). - addReg(Mips::S2, RegState::Define).addImm(FrameSize); + addImm(FrameSize); else { int Base = 2040; // should create template function like isUInt that // returns largest possible n bit unsigned integer @@ -210,12 +221,15 @@ void Mips16InstrInfo::restoreFrame(unsigned SP, int64_t FrameSize, BuildAddiuSpImm(MBB, I, Remainder); else adjustStackPtrBig(SP, Remainder, MBB, I, Mips::A0, Mips::A1); - BuildMI(MBB, I, DL, get(Mips::RestoreX16)). + MIB = BuildMI( + MBB, I, DL, get(Mips::RestoreX16)). addReg(Mips::RA, RegState::Define). addReg(Mips::S0, RegState::Define). addReg(Mips::S1, RegState::Define). - addReg(Mips::S2, RegState::Define).addImm(Base); + addImm(Base); } + if (SaveS2) + MIB.addReg(Mips::S2, RegState::Define); } // Adjust SP by Amount bytes where bytes can be up to 32bit number. diff --git a/lib/Target/Mips/MipsRegisterInfo.cpp b/lib/Target/Mips/MipsRegisterInfo.cpp index 3105b020845..65b1f8cf2d1 100644 --- a/lib/Target/Mips/MipsRegisterInfo.cpp +++ b/lib/Target/Mips/MipsRegisterInfo.cpp @@ -27,6 +27,7 @@ #include "llvm/CodeGen/ValueTypes.h" #include "llvm/DebugInfo.h" #include "llvm/IR/Constants.h" +#include "llvm/IR/Function.h" #include "llvm/IR/Type.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" @@ -183,6 +184,8 @@ getReservedRegs(const MachineFunction &MF) const { Reserved.set(Mips::RA_64); Reserved.set(Mips::T0); Reserved.set(Mips::T1); + if (MF.getFunction()->hasFnAttribute("saveS2")) + Reserved.set(Mips::S2); } // Reserve GP if small section is used. diff --git a/test/CodeGen/Mips/align16.ll b/test/CodeGen/Mips/align16.ll index 1b724e9fc86..32ec94bd20c 100644 --- a/test/CodeGen/Mips/align16.ll +++ b/test/CodeGen/Mips/align16.ll @@ -25,7 +25,7 @@ entry: call void @p(i32* %arrayidx1) ret void } -; 16: save $ra, $16, $17, $18, 2040 +; 16: save $ra, $16, $17, 2040 ; 16: addiu $sp, -56 # 16 bit inst ; 16: addiu $sp, 56 # 16 bit inst -; 16: restore $ra, $16, $17, $18, 2040 +; 16: restore $ra, $16, $17, 2040 diff --git a/test/CodeGen/Mips/alloca16.ll b/test/CodeGen/Mips/alloca16.ll index a836d124889..5bdf4dd7835 100644 --- a/test/CodeGen/Mips/alloca16.ll +++ b/test/CodeGen/Mips/alloca16.ll @@ -20,7 +20,7 @@ entry: define void @test() nounwind { entry: ; 16: .frame $sp,24,$ra -; 16: save $ra, $16, $17, $18, 24 +; 16: save $ra, $16, $17, 24 ; 16: move $16, $sp ; 16: move ${{[0-9]+}}, $sp ; 16: subu $[[REGISTER:[0-9]+]], ${{[0-9]+}}, ${{[0-9]+}} diff --git a/test/CodeGen/Mips/ex2.ll b/test/CodeGen/Mips/ex2.ll index 5b3463e5094..75562156cf0 100644 --- a/test/CodeGen/Mips/ex2.ll +++ b/test/CodeGen/Mips/ex2.ll @@ -6,12 +6,11 @@ define i32 @main() { ; 16-LABEL: main: ; 16: .cfi_startproc -; 16: save $ra, $16, $17, $18, 40 +; 16: save $ra, $16, $17, 40 ; 16: .cfi_def_cfa_offset 40 -; 16: .cfi_offset 18, -8 -; 16: .cfi_offset 17, -12 -; 16: .cfi_offset 16, -16 ; 16: .cfi_offset 31, -4 +; 16: .cfi_offset 17, -8 +; 16: .cfi_offset 16, -12 ; 16: .cfi_endproc entry: %retval = alloca i32, align 4 diff --git a/test/CodeGen/Mips/helloworld.ll b/test/CodeGen/Mips/helloworld.ll index 32bc45fba38..d363006a550 100644 --- a/test/CodeGen/Mips/helloworld.ll +++ b/test/CodeGen/Mips/helloworld.ll @@ -25,7 +25,7 @@ entry: ; SR32: .set noreorder ; SR32: .set nomacro ; SR32: .set noat -; SR: save $ra, $16, $17, $18, [[FS:[0-9]+]] +; SR: save $ra, $16, $17, [[FS:[0-9]+]] ; PE: .ent main ; PE: .align 2 ; PE-NEXT: li $[[T1:[0-9]+]], %hi(_gp_disp) @@ -37,7 +37,7 @@ entry: ; C2: move $25, ${{[0-9]+}} ; C1: move $gp, ${{[0-9]+}} ; C1: jalrc ${{[0-9]+}} -; SR: restore $ra, $16, $17, $18, [[FS]] +; SR: restore $ra, $16, $17, [[FS]] ; PE: li $2, 0 ; PE: jrc $ra diff --git a/test/CodeGen/Mips/s2rem.ll b/test/CodeGen/Mips/s2rem.ll new file mode 100644 index 00000000000..0f0b3edade0 --- /dev/null +++ b/test/CodeGen/Mips/s2rem.ll @@ -0,0 +1,95 @@ +; RUN: llc -mtriple=mipsel-linux-gnu -march=mipsel -mcpu=mips16 -relocation-model=pic < %s | FileCheck %s -check-prefix=NEG + +; RUN: llc -mtriple=mipsel-linux-gnu -march=mipsel -mcpu=mips16 -relocation-model=static < %s | FileCheck %s -check-prefix=NEG + +; RUN: llc -mtriple=mipsel-linux-gnu -march=mipsel -mcpu=mips16 -relocation-model=pic < %s | FileCheck %s + +; RUN: llc -mtriple=mipsel-linux-gnu -march=mipsel -mcpu=mips16 -relocation-model=static < %s | FileCheck %s + +@xi = common global i32 0, align 4 +@x = common global float 0.000000e+00, align 4 +@xd = common global double 0.000000e+00, align 8 + +; Function Attrs: nounwind +define void @it() #0 { +entry: + %call = call i32 @i(i32 1) + store i32 %call, i32* @xi, align 4 + ret void +; CHECK: .ent it +; NEG: .ent it +; CHECK: save $ra, $16, $17, [[FS:[0-9]+]] +; NEG-NOT: save $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: restore $ra, $16, $17, [[FS]] +; NEG-NOT: restore $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: .end it +; NEG: .end it +} + +declare i32 @i(i32) #1 + +; Function Attrs: nounwind +define void @ft() #0 { +entry: + %call = call float @f() + store float %call, float* @x, align 4 + ret void +; CHECK: .ent ft +; CHECK: save $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: restore $ra, $16, $17, [[FS]], $18 +; CHECK: .end ft +} + +declare float @f() #1 + +; Function Attrs: nounwind +define void @dt() #0 { +entry: + %call = call double @d() + store double %call, double* @xd, align 8 + ret void +; CHECK: .ent dt +; CHECK: save $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: restore $ra, $16, $17, [[FS]], $18 +; CHECK: .end dt +} + +declare double @d() #1 + +; Function Attrs: nounwind +define void @fft() #0 { +entry: + %0 = load float* @x, align 4 + %call = call float @ff(float %0) + store float %call, float* @x, align 4 + ret void +; CHECK: .ent fft +; CHECK: save $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: restore $ra, $16, $17, [[FS]], $18 +; CHECK: .end fft +} + +declare float @ff(float) #1 + +; Function Attrs: nounwind +define void @vft() #0 { +entry: + %0 = load float* @x, align 4 + call void @vf(float %0) + ret void +; CHECK: .ent vft +; NEG: .ent vft +; CHECK: save $ra, $16, $17, [[FS:[0-9]+]] +; NEG-NOT: save $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: restore $ra, $16, $17, [[FS]] +; NEG-NOT: restore $ra, $16, $17, [[FS:[0-9]+]], $18 +; CHECK: .end vft +; NEG: .end vft +} + +declare void @vf(float) #1 + +attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #1 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } + + -- 2.34.1