Files
clang-p2996/llvm/lib/Target/PowerPC/PPCTLSDynamicCall.cpp
Serge Pavlov d526b13e61 Add extra operand to CALLSEQ_START to keep frame part set up previously
Using arguments with attribute inalloca creates problems for verification
of machine representation. This attribute instructs the backend that the
argument is prepared in stack prior to  CALLSEQ_START..CALLSEQ_END
sequence (see http://llvm.org/docs/InAlloca.htm for details). Frame size
stored in CALLSEQ_START in this case does not count the size of this
argument. However CALLSEQ_END still keeps total frame size, as caller can
be responsible for cleanup of entire frame. So CALLSEQ_START and
CALLSEQ_END keep different frame size and the difference is treated by
MachineVerifier as stack error. Currently there is no way to distinguish
this case from actual errors.

This patch adds additional argument to CALLSEQ_START and its
target-specific counterparts to keep size of stack that is set up prior to
the call frame sequence. This argument allows MachineVerifier to calculate
actual frame size associated with frame setup instruction and correctly
process the case of inalloca arguments.

The changes made by the patch are:
- Frame setup instructions get the second mandatory argument. It
  affects all targets that use frame pseudo instructions and touched many
  files although the changes are uniform.
- Access to frame properties are implemented using special instructions
  rather than calls getOperand(N).getImm(). For X86 and ARM such
  replacement was made previously.
- Changes that reflect appearance of additional argument of frame setup
  instruction. These involve proper instruction initialization and
  methods that access instruction arguments.
- MachineVerifier retrieves frame size using method, which reports sum of
  frame parts initialized inside frame instruction pair and outside it.

The patch implements approach proposed by Quentin Colombet in
https://bugs.llvm.org/show_bug.cgi?id=27481#c1.
It fixes 9 tests failed with machine verifier enabled and listed
in PR27481.

Differential Revision: https://reviews.llvm.org/D32394

llvm-svn: 302527
2017-05-09 13:35:13 +00:00

176 lines
5.8 KiB
C++

//===---------- PPCTLSDynamicCall.cpp - TLS Dynamic Call Fixup ------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This pass expands ADDItls{ld,gd}LADDR[32] machine instructions into
// separate ADDItls[gd]L[32] and GETtlsADDR[32] instructions, both of
// which define GPR3. A copy is added from GPR3 to the target virtual
// register of the original instruction. The GETtlsADDR[32] is really
// a call instruction, so its target register is constrained to be GPR3.
// This is not true of ADDItls[gd]L[32], but there is a legacy linker
// optimization bug that requires the target register of the addi of
// a local- or general-dynamic TLS access sequence to be GPR3.
//
// This is done in a late pass so that TLS variable accesses can be
// fully commoned by MachineCSE.
//
//===----------------------------------------------------------------------===//
#include "PPCInstrInfo.h"
#include "PPC.h"
#include "PPCInstrBuilder.h"
#include "PPCTargetMachine.h"
#include "llvm/CodeGen/LiveIntervalAnalysis.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "ppc-tls-dynamic-call"
namespace llvm {
void initializePPCTLSDynamicCallPass(PassRegistry&);
}
namespace {
struct PPCTLSDynamicCall : public MachineFunctionPass {
static char ID;
PPCTLSDynamicCall() : MachineFunctionPass(ID) {
initializePPCTLSDynamicCallPass(*PassRegistry::getPassRegistry());
}
const PPCInstrInfo *TII;
LiveIntervals *LIS;
protected:
bool processBlock(MachineBasicBlock &MBB) {
bool Changed = false;
bool Is64Bit = MBB.getParent()->getSubtarget<PPCSubtarget>().isPPC64();
for (MachineBasicBlock::iterator I = MBB.begin(), IE = MBB.end();
I != IE;) {
MachineInstr &MI = *I;
if (MI.getOpcode() != PPC::ADDItlsgdLADDR &&
MI.getOpcode() != PPC::ADDItlsldLADDR &&
MI.getOpcode() != PPC::ADDItlsgdLADDR32 &&
MI.getOpcode() != PPC::ADDItlsldLADDR32) {
++I;
continue;
}
DEBUG(dbgs() << "TLS Dynamic Call Fixup:\n " << MI);
unsigned OutReg = MI.getOperand(0).getReg();
unsigned InReg = MI.getOperand(1).getReg();
DebugLoc DL = MI.getDebugLoc();
unsigned GPR3 = Is64Bit ? PPC::X3 : PPC::R3;
unsigned Opc1, Opc2;
const unsigned OrigRegs[] = {OutReg, InReg, GPR3};
switch (MI.getOpcode()) {
default:
llvm_unreachable("Opcode inconsistency error");
case PPC::ADDItlsgdLADDR:
Opc1 = PPC::ADDItlsgdL;
Opc2 = PPC::GETtlsADDR;
break;
case PPC::ADDItlsldLADDR:
Opc1 = PPC::ADDItlsldL;
Opc2 = PPC::GETtlsldADDR;
break;
case PPC::ADDItlsgdLADDR32:
Opc1 = PPC::ADDItlsgdL32;
Opc2 = PPC::GETtlsADDR32;
break;
case PPC::ADDItlsldLADDR32:
Opc1 = PPC::ADDItlsldL32;
Opc2 = PPC::GETtlsldADDR32;
break;
}
// Don't really need to save data to the stack - the clobbered
// registers are already saved when the SDNode (e.g. PPCaddiTlsgdLAddr)
// gets translated to the pseudo instruction (e.g. ADDItlsgdLADDR).
BuildMI(MBB, I, DL, TII->get(PPC::ADJCALLSTACKDOWN)).addImm(0)
.addImm(0);
// Expand into two ops built prior to the existing instruction.
MachineInstr *Addi = BuildMI(MBB, I, DL, TII->get(Opc1), GPR3)
.addReg(InReg);
Addi->addOperand(MI.getOperand(2));
// The ADDItls* instruction is the first instruction in the
// repair range.
MachineBasicBlock::iterator First = I;
--First;
MachineInstr *Call = (BuildMI(MBB, I, DL, TII->get(Opc2), GPR3)
.addReg(GPR3));
Call->addOperand(MI.getOperand(3));
BuildMI(MBB, I, DL, TII->get(PPC::ADJCALLSTACKUP)).addImm(0).addImm(0);
BuildMI(MBB, I, DL, TII->get(TargetOpcode::COPY), OutReg)
.addReg(GPR3);
// The COPY is the last instruction in the repair range.
MachineBasicBlock::iterator Last = I;
--Last;
// Move past the original instruction and remove it.
++I;
MI.removeFromParent();
// Repair the live intervals.
LIS->repairIntervalsInRange(&MBB, First, Last, OrigRegs);
Changed = true;
}
return Changed;
}
public:
bool runOnMachineFunction(MachineFunction &MF) override {
TII = MF.getSubtarget<PPCSubtarget>().getInstrInfo();
LIS = &getAnalysis<LiveIntervals>();
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 {
AU.addRequired<LiveIntervals>();
AU.addPreserved<LiveIntervals>();
AU.addRequired<SlotIndexes>();
AU.addPreserved<SlotIndexes>();
MachineFunctionPass::getAnalysisUsage(AU);
}
};
}
INITIALIZE_PASS_BEGIN(PPCTLSDynamicCall, DEBUG_TYPE,
"PowerPC TLS Dynamic Call Fixup", false, false)
INITIALIZE_PASS_DEPENDENCY(LiveIntervals)
INITIALIZE_PASS_DEPENDENCY(SlotIndexes)
INITIALIZE_PASS_END(PPCTLSDynamicCall, DEBUG_TYPE,
"PowerPC TLS Dynamic Call Fixup", false, false)
char PPCTLSDynamicCall::ID = 0;
FunctionPass*
llvm::createPPCTLSDynamicCallPass() { return new PPCTLSDynamicCall(); }