llvm-6502/lib/Target/X86/X86FloatingPointRegKill.cpp
Chris Lattner a26a8471bd now that fp reg kill insertion stuff happens as a separate
pass after isel instead of being interlaced with it, we can
trust that all the code for a function has been isel'd before
it is run.

The practical impact of this is that we can scan for machine
instr phis instead of doing a fuzzy match on the LLVM BB for
phi nodes.  Doing the fuzzy match required knowing when isel
would produce an fp reg stack phi which was gross.  It was
also wrong in cases where select got lowered to a branch
tree because cmovs aren't available (PR6828).

Just do the scan on machine phis which is simpler, faster
and more correct.  This fixes PR6828.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104333 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-21 18:17:54 +00:00

157 lines
5.2 KiB
C++

//===-- X86FloatingPoint.cpp - FP_REG_KILL inserter -----------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the pass which inserts FP_REG_KILL instructions.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "x86-codegen"
#include "X86.h"
#include "X86InstrInfo.h"
#include "llvm/Instructions.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/CFG.h"
#include "llvm/ADT/Statistic.h"
using namespace llvm;
STATISTIC(NumFPKill, "Number of FP_REG_KILL instructions added");
namespace {
struct FPRegKiller : public MachineFunctionPass {
static char ID;
FPRegKiller() : MachineFunctionPass(&ID) {}
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesCFG();
AU.addPreservedID(MachineLoopInfoID);
AU.addPreservedID(MachineDominatorsID);
MachineFunctionPass::getAnalysisUsage(AU);
}
virtual bool runOnMachineFunction(MachineFunction &MF);
virtual const char *getPassName() const {
return "X86 FP_REG_KILL inserter";
}
};
char FPRegKiller::ID = 0;
}
FunctionPass *llvm::createX87FPRegKillInserterPass() {
return new FPRegKiller();
}
/// isFPStackVReg - Return true if the specified vreg is from a fp stack
/// register class.
static bool isFPStackVReg(unsigned RegNo, const MachineRegisterInfo &MRI) {
if (!TargetRegisterInfo::isVirtualRegister(RegNo))
return false;
switch (MRI.getRegClass(RegNo)->getID()) {
default: return false;
case X86::RFP32RegClassID:
case X86::RFP64RegClassID:
case X86::RFP80RegClassID:
return true;
}
}
/// ContainsFPStackCode - Return true if the specific MBB has floating point
/// stack code, and thus needs an FP_REG_KILL.
static bool ContainsFPStackCode(MachineBasicBlock *MBB,
const MachineRegisterInfo &MRI) {
// Scan the block, looking for instructions that define fp stack vregs.
for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end();
I != E; ++I) {
if (I->getNumOperands() == 0 || !I->getOperand(0).isReg())
continue;
for (unsigned op = 0, e = I->getNumOperands(); op != e; ++op) {
if (!I->getOperand(op).isReg() || !I->getOperand(op).isDef())
continue;
if (isFPStackVReg(I->getOperand(op).getReg(), MRI))
return true;
}
}
// Check PHI nodes in successor blocks. These PHI's will be lowered to have
// a copy of the input value in this block, which is a definition of the
// value.
for (MachineBasicBlock::succ_iterator SI = MBB->succ_begin(),
E = MBB->succ_end(); SI != E; ++ SI) {
MachineBasicBlock *SuccBB = *SI;
for (MachineBasicBlock::iterator I = SuccBB->begin(), E = SuccBB->end();
I != E; ++I) {
// All PHI nodes are at the top of the block.
if (!I->isPHI()) break;
if (isFPStackVReg(I->getOperand(0).getReg(), MRI))
return true;
}
}
return false;
}
bool FPRegKiller::runOnMachineFunction(MachineFunction &MF) {
// If we are emitting FP stack code, scan the basic block to determine if this
// block defines any FP values. If so, put an FP_REG_KILL instruction before
// the terminator of the block.
// Note that FP stack instructions are used in all modes for long double,
// so we always need to do this check.
// Also note that it's possible for an FP stack register to be live across
// an instruction that produces multiple basic blocks (SSE CMOV) so we
// must check all the generated basic blocks.
// Scan all of the machine instructions in these MBBs, checking for FP
// stores. (RFP32 and RFP64 will not exist in SSE mode, but RFP80 might.)
// Fast-path: If nothing is using the x87 registers, we don't need to do
// any scanning.
const MachineRegisterInfo &MRI = MF.getRegInfo();
if (MRI.getRegClassVirtRegs(X86::RFP80RegisterClass).empty() &&
MRI.getRegClassVirtRegs(X86::RFP64RegisterClass).empty() &&
MRI.getRegClassVirtRegs(X86::RFP32RegisterClass).empty())
return false;
bool Changed = false;
MachineFunction::iterator MBBI = MF.begin();
MachineFunction::iterator EndMBB = MF.end();
for (; MBBI != EndMBB; ++MBBI) {
MachineBasicBlock *MBB = MBBI;
// If this block returns, ignore it. We don't want to insert an FP_REG_KILL
// before the return.
if (!MBB->empty()) {
MachineBasicBlock::iterator EndI = MBB->end();
--EndI;
if (EndI->getDesc().isReturn())
continue;
}
// If we find any FP stack code, emit the FP_REG_KILL instruction.
if (ContainsFPStackCode(MBB, MRI)) {
BuildMI(*MBB, MBBI->getFirstTerminator(), DebugLoc(),
MF.getTarget().getInstrInfo()->get(X86::FP_REG_KILL));
++NumFPKill;
Changed = true;
}
}
return Changed;
}