1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-25 05:52:53 +02:00
llvm-mirror/lib/Target/PowerPC/PPCVSXCopy.cpp
Nemanja Ivanovic 285f278c18 Scalar to vector conversions using direct moves
This patch corresponds to review:
http://reviews.llvm.org/D11471

It improves the code generated for converting a scalar to a vector value. With
direct moves from GPRs to VSRs, we no longer require expensive stack operations
for this. Subsequent patches will handle the reverse case and more general
operations between vectors and their scalar elements.

llvm-svn: 244921
2015-08-13 17:40:44 +00:00

182 lines
6.1 KiB
C++

//===-------------- PPCVSXCopy.cpp - VSX Copy Legalization ----------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// A pass which deals with the complexity of generating legal VSX register
// copies to/from register classes which partially overlap with the VSX
// register file.
//
//===----------------------------------------------------------------------===//
#include "PPCInstrInfo.h"
#include "MCTargetDesc/PPCPredicates.h"
#include "PPC.h"
#include "PPCHazardRecognizers.h"
#include "PPCInstrBuilder.h"
#include "PPCMachineFunctionInfo.h"
#include "PPCTargetMachine.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/MC/MCAsmInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/TargetRegistry.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "ppc-vsx-copy"
namespace llvm {
void initializePPCVSXCopyPass(PassRegistry&);
}
namespace {
// PPCVSXCopy pass - For copies between VSX registers and non-VSX registers
// (Altivec and scalar floating-point registers), we need to transform the
// copies into subregister copies with other restrictions.
struct PPCVSXCopy : public MachineFunctionPass {
static char ID;
PPCVSXCopy() : MachineFunctionPass(ID) {
initializePPCVSXCopyPass(*PassRegistry::getPassRegistry());
}
const TargetInstrInfo *TII;
bool IsRegInClass(unsigned Reg, const TargetRegisterClass *RC,
MachineRegisterInfo &MRI) {
if (TargetRegisterInfo::isVirtualRegister(Reg)) {
return RC->hasSubClassEq(MRI.getRegClass(Reg));
} else if (RC->contains(Reg)) {
return true;
}
return false;
}
bool IsVSReg(unsigned Reg, MachineRegisterInfo &MRI) {
return IsRegInClass(Reg, &PPC::VSRCRegClass, MRI);
}
bool IsVRReg(unsigned Reg, MachineRegisterInfo &MRI) {
return IsRegInClass(Reg, &PPC::VRRCRegClass, MRI);
}
bool IsF8Reg(unsigned Reg, MachineRegisterInfo &MRI) {
return IsRegInClass(Reg, &PPC::F8RCRegClass, MRI);
}
bool IsVSFReg(unsigned Reg, MachineRegisterInfo &MRI) {
return IsRegInClass(Reg, &PPC::VSFRCRegClass, MRI);
}
protected:
bool processBlock(MachineBasicBlock &MBB) {
bool Changed = false;
MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
for (MachineBasicBlock::iterator I = MBB.begin(), IE = MBB.end();
I != IE; ++I) {
MachineInstr *MI = I;
if (!MI->isFullCopy())
continue;
MachineOperand &DstMO = MI->getOperand(0);
MachineOperand &SrcMO = MI->getOperand(1);
if ( IsVSReg(DstMO.getReg(), MRI) &&
!IsVSReg(SrcMO.getReg(), MRI)) {
// This is a copy *to* a VSX register from a non-VSX register.
Changed = true;
const TargetRegisterClass *SrcRC =
IsVRReg(SrcMO.getReg(), MRI) ? &PPC::VSHRCRegClass :
&PPC::VSLRCRegClass;
assert((IsF8Reg(SrcMO.getReg(), MRI) ||
IsVRReg(SrcMO.getReg(), MRI) ||
IsVSFReg(SrcMO.getReg(), MRI)) &&
"Unknown source for a VSX copy");
unsigned NewVReg = MRI.createVirtualRegister(SrcRC);
BuildMI(MBB, MI, MI->getDebugLoc(),
TII->get(TargetOpcode::SUBREG_TO_REG), NewVReg)
.addImm(1) // add 1, not 0, because there is no implicit clearing
// of the high bits.
.addOperand(SrcMO)
.addImm(IsVRReg(SrcMO.getReg(), MRI) ? PPC::sub_128 :
PPC::sub_64);
// The source of the original copy is now the new virtual register.
SrcMO.setReg(NewVReg);
} else if (!IsVSReg(DstMO.getReg(), MRI) &&
IsVSReg(SrcMO.getReg(), MRI)) {
// This is a copy *from* a VSX register to a non-VSX register.
Changed = true;
const TargetRegisterClass *DstRC =
IsVRReg(DstMO.getReg(), MRI) ? &PPC::VSHRCRegClass :
&PPC::VSLRCRegClass;
assert((IsF8Reg(DstMO.getReg(), MRI) ||
IsVRReg(DstMO.getReg(), MRI)) &&
"Unknown destination for a VSX copy");
// Copy the VSX value into a new VSX register of the correct subclass.
unsigned NewVReg = MRI.createVirtualRegister(DstRC);
BuildMI(MBB, MI, MI->getDebugLoc(),
TII->get(TargetOpcode::COPY), NewVReg)
.addOperand(SrcMO);
// Transform the original copy into a subregister extraction copy.
SrcMO.setReg(NewVReg);
SrcMO.setSubReg(IsVRReg(DstMO.getReg(), MRI) ? PPC::sub_128 :
PPC::sub_64);
}
}
return Changed;
}
public:
bool runOnMachineFunction(MachineFunction &MF) override {
// If we don't have VSX on the subtarget, don't do anything.
const PPCSubtarget &STI = MF.getSubtarget<PPCSubtarget>();
if (!STI.hasVSX())
return false;
TII = STI.getInstrInfo();
bool Changed = false;
for (MachineFunction::iterator I = MF.begin(); I != MF.end();) {
MachineBasicBlock &B = *I++;
if (processBlock(B))
Changed = true;
}
return Changed;
}
void getAnalysisUsage(AnalysisUsage &AU) const override {
MachineFunctionPass::getAnalysisUsage(AU);
}
};
}
INITIALIZE_PASS(PPCVSXCopy, DEBUG_TYPE,
"PowerPC VSX Copy Legalization", false, false)
char PPCVSXCopy::ID = 0;
FunctionPass*
llvm::createPPCVSXCopyPass() { return new PPCVSXCopy(); }