SIFixSGPRCopies.cpp File Reference

#include "AMDGPU.h"
#include "AMDGPUSubtarget.h"
#include "SIInstrInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"

Include dependency graph for SIFixSGPRCopies.cpp:

Go to the source code of this file.

Defines
#define	DEBUG_TYPE   "sgpr-copies"
Functions
static bool	hasVGPROperands (const MachineInstr &MI, const SIRegisterInfo *TRI)

---

Detailed Description

Copies from VGPR to SGPR registers are illegal and the register coalescer will sometimes generate these illegal copies in situations like this:

Register Class <vsrc> is the union of <vgpr> and <sgpr>

BB0: vreg0 <sgpr> = SCALAR_INST vreg1 <vsrc> = COPY vreg0 <sgpr> ... BRANCH cond BB1, BB2 BB1: vreg2 <vgpr> = VECTOR_INST vreg3 <vsrc> = COPY vreg2 <vgpr> BB2: vreg4 <vsrc> = PHI vreg1 <vsrc>, <BB#0>, vreg3 <vrsc>, <BB#1> vreg5 <vgpr> = VECTOR_INST vreg4 <vsrc>

The coalescer will begin at BB0 and eliminate its copy, then the resulting code will look like this:

BB0: vreg0 <sgpr> = SCALAR_INST ... BRANCH cond BB1, BB2 BB1: vreg2 <vgpr> = VECTOR_INST vreg3 <vsrc> = COPY vreg2 <vgpr> BB2: vreg4 <sgpr> = PHI vreg0 <sgpr>, <BB#0>, vreg3 <vsrc>, <BB#1> vreg5 <vgpr> = VECTOR_INST vreg4 <sgpr>

Now that the result of the PHI instruction is an SGPR, the register allocator is now forced to constrain the register class of vreg3 to <sgpr> so we end up with final code like this:

BB0: vreg0 <sgpr> = SCALAR_INST ... BRANCH cond BB1, BB2 BB1: vreg2 <vgpr> = VECTOR_INST vreg3 <sgpr> = COPY vreg2 <vgpr> BB2: vreg4 <sgpr> = PHI vreg0 <sgpr>, <BB#0>, vreg3 <sgpr>, <BB#1> vreg5 <vgpr> = VECTOR_INST vreg4 <sgpr>

Now this code contains an illegal copy from a VGPR to an SGPR.

In order to avoid this problem, this pass searches for PHI instructions which define a <vsrc> register and constrains its definition class to <vgpr> if the user of the PHI's definition register is a vector instruction. If the PHI's definition class is constrained to <vgpr> then the coalescer will be unable to perform the COPY removal from the above example which ultimately led to the creation of an illegal COPY.

Definition in file SIFixSGPRCopies.cpp.

---

Define Documentation

#define DEBUG_TYPE   "sgpr-copies"

Definition at line 80 of file SIFixSGPRCopies.cpp.

---

Function Documentation

static bool hasVGPROperands	(	const MachineInstr &	MI,
		const SIRegisterInfo *	TRI
	)		[static]

Definition at line 118 of file SIFixSGPRCopies.cpp.

References llvm::MachineInstr::getNumOperands(), llvm::MachineInstr::getOperand(), llvm::MachineInstr::getParent(), llvm::MachineBasicBlock::getParent(), llvm::MachineOperand::getReg(), llvm::MachineRegisterInfo::getRegClass(), llvm::MachineFunction::getRegInfo(), llvm::SIRegisterInfo::hasVGPRs(), llvm::MachineOperand::isReg(), and llvm::TargetRegisterInfo::isVirtualRegister().