llvm.org GIT mirror llvm / release_39 lib / Target / AMDGPU / SIFrameLowering.cpp

Tree @release_39 (Download .tar.gz)

SIFrameLowering.cpp @release_39raw · history · blame

//===----------------------- SIFrameLowering.cpp --------------------------===//
//                     The LLVM Compiler Infrastructure
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.

#include "SIFrameLowering.h"
#include "SIInstrInfo.h"
#include "SIMachineFunctionInfo.h"
#include "SIRegisterInfo.h"
#include "AMDGPUSubtarget.h"

#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/RegisterScavenging.h"

using namespace llvm;

static bool hasOnlySGPRSpills(const SIMachineFunctionInfo *FuncInfo,
                              const MachineFrameInfo *FrameInfo) {
  return FuncInfo->hasSpilledSGPRs() &&
    (!FuncInfo->hasSpilledVGPRs() && !FuncInfo->hasNonSpillStackObjects());

static ArrayRef<MCPhysReg> getAllSGPR128() {
  return makeArrayRef(AMDGPU::SGPR_128RegClass.begin(),

static ArrayRef<MCPhysReg> getAllSGPRs() {
  return makeArrayRef(AMDGPU::SGPR_32RegClass.begin(),

void SIFrameLowering::emitPrologue(MachineFunction &MF,
                                   MachineBasicBlock &MBB) const {
  // Emit debugger prologue if "amdgpu-debugger-emit-prologue" attribute was
  // specified.
  const SISubtarget &ST = MF.getSubtarget<SISubtarget>();
  if (ST.debuggerEmitPrologue())
    emitDebuggerPrologue(MF, MBB);

  if (!MF.getFrameInfo()->hasStackObjects())

  assert(&MF.front() == &MBB && "Shrink-wrapping not yet supported");

  SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();

  // If we only have SGPR spills, we won't actually be using scratch memory
  // since these spill to VGPRs.
  // FIXME: We should be cleaning up these unused SGPR spill frame indices
  // somewhere.
  if (hasOnlySGPRSpills(MFI, MF.getFrameInfo()))

  const SIInstrInfo *TII = ST.getInstrInfo();
  const SIRegisterInfo *TRI = &TII->getRegisterInfo();
  MachineRegisterInfo &MRI = MF.getRegInfo();
  MachineBasicBlock::iterator I = MBB.begin();

  // We need to insert initialization of the scratch resource descriptor.
  unsigned ScratchRsrcReg = MFI->getScratchRSrcReg();
  assert(ScratchRsrcReg != AMDGPU::NoRegister);

  unsigned ScratchWaveOffsetReg = MFI->getScratchWaveOffsetReg();
  assert(ScratchWaveOffsetReg != AMDGPU::NoRegister);

  unsigned PreloadedScratchWaveOffsetReg = TRI->getPreloadedValue(

  unsigned PreloadedPrivateBufferReg = AMDGPU::NoRegister;
  if (ST.isAmdHsaOS()) {
    PreloadedPrivateBufferReg = TRI->getPreloadedValue(

  if (MFI->hasFlatScratchInit()) {
    // We don't need this if we only have spills since there is no user facing
    // scratch.

    // TODO: If we know we don't have flat instructions earlier, we can omit
    // this from the input registers.
    // TODO: We only need to know if we access scratch space through a flat
    // pointer. Because we only detect if flat instructions are used at all,
    // this will be used more often than necessary on VI.

    // Debug location must be unknown since the first debug location is used to
    // determine the end of the prologue.
    DebugLoc DL;

    unsigned FlatScratchInitReg
      = TRI->getPreloadedValue(MF, SIRegisterInfo::FLAT_SCRATCH_INIT);


    // Copy the size in bytes.
    unsigned FlatScrInitHi = TRI->getSubReg(FlatScratchInitReg, AMDGPU::sub1);
      .addReg(FlatScrInitHi, RegState::Kill);

    unsigned FlatScrInitLo = TRI->getSubReg(FlatScratchInitReg, AMDGPU::sub0);

    // Add wave offset in bytes to private base offset.
    // See comment in AMDKernelCodeT.h for enable_sgpr_flat_scratch_init.
    BuildMI(MBB, I, DL, TII->get(AMDGPU::S_ADD_U32), FlatScrInitLo)

    // Convert offset to 256-byte units.
      .addReg(FlatScrInitLo, RegState::Kill)

  // If we reserved the original input registers, we don't need to copy to the
  // reserved registers.
  if (ScratchRsrcReg == PreloadedPrivateBufferReg) {
    // We should always reserve these 5 registers at the same time.
    assert(ScratchWaveOffsetReg == PreloadedScratchWaveOffsetReg &&
           "scratch wave offset and private segment buffer inconsistent");

  // We added live-ins during argument lowering, but since they were not used
  // they were deleted. We're adding the uses now, so add them back.

  if (ST.isAmdHsaOS()) {

  if (!ST.hasSGPRInitBug()) {
    // We reserved the last registers for this. Shift it down to the end of those
    // which were actually used.
    // FIXME: It might be safer to use a pseudoregister before replacement.

    // FIXME: We should be able to eliminate unused input registers. We only
    // cannot do this for the resources required for scratch access. For now we
    // skip over user SGPRs and may leave unused holes.

    // We find the resource first because it has an alignment requirement.
    if (ScratchRsrcReg == TRI->reservedPrivateSegmentBufferReg(MF)) {
      MachineRegisterInfo &MRI = MF.getRegInfo();

      unsigned NumPreloaded = MFI->getNumPreloadedSGPRs() / 4;
      // Skip the last 2 elements because the last one is reserved for VCC, and
      // this is the 2nd to last element already.
      for (MCPhysReg Reg : getAllSGPR128().drop_back(2).slice(NumPreloaded)) {
        // Pick the first unallocated one. Make sure we don't clobber the other
        // reserved input we needed.
        if (!MRI.isPhysRegUsed(Reg)) {
          MRI.replaceRegWith(ScratchRsrcReg, Reg);
          ScratchRsrcReg = Reg;

    if (ScratchWaveOffsetReg == TRI->reservedPrivateSegmentWaveByteOffsetReg(MF)) {
      MachineRegisterInfo &MRI = MF.getRegInfo();
      unsigned NumPreloaded = MFI->getNumPreloadedSGPRs();

      // We need to drop register from the end of the list that we cannot use
      // for the scratch wave offset.
      // + 2 s102 and s103 do not exist on VI.
      // + 2 for vcc
      // + 2 for xnack_mask
      // + 2 for flat_scratch
      // + 4 for registers reserved for scratch resource register
      // + 1 for register reserved for scratch wave offset.  (By exluding this
      //     register from the list to consider, it means that when this
      //     register is being used for the scratch wave offset and there
      //     are no other free SGPRs, then the value will stay in this register.
      // ----
      //  13
      for (MCPhysReg Reg : getAllSGPRs().drop_back(13).slice(NumPreloaded)) {
        // Pick the first unallocated SGPR. Be careful not to pick an alias of the
        // scratch descriptor, since we haven’t added its uses yet.
        if (!MRI.isPhysRegUsed(Reg)) {
          if (!MRI.isAllocatable(Reg) ||
              TRI->isSubRegisterEq(ScratchRsrcReg, Reg))

          MRI.replaceRegWith(ScratchWaveOffsetReg, Reg);
          ScratchWaveOffsetReg = Reg;

  assert(!TRI->isSubRegister(ScratchRsrcReg, ScratchWaveOffsetReg));

  const MCInstrDesc &SMovB32 = TII->get(AMDGPU::S_MOV_B32);
  DebugLoc DL;

  if (PreloadedScratchWaveOffsetReg != ScratchWaveOffsetReg) {
    // Make sure we emit the copy for the offset first. We may have chosen to copy
    // the buffer resource into a register that aliases the input offset register.
    BuildMI(MBB, I, DL, SMovB32, ScratchWaveOffsetReg)
      .addReg(PreloadedScratchWaveOffsetReg, RegState::Kill);

  if (ST.isAmdHsaOS()) {
    // Insert copies from argument register.
      !TRI->isSubRegisterEq(PreloadedPrivateBufferReg, ScratchRsrcReg) &&
      !TRI->isSubRegisterEq(PreloadedPrivateBufferReg, ScratchWaveOffsetReg));

    unsigned Rsrc01 = TRI->getSubReg(ScratchRsrcReg, AMDGPU::sub0_sub1);
    unsigned Rsrc23 = TRI->getSubReg(ScratchRsrcReg, AMDGPU::sub2_sub3);

    unsigned Lo = TRI->getSubReg(PreloadedPrivateBufferReg, AMDGPU::sub0_sub1);
    unsigned Hi = TRI->getSubReg(PreloadedPrivateBufferReg, AMDGPU::sub2_sub3);

    const MCInstrDesc &SMovB64 = TII->get(AMDGPU::S_MOV_B64);

    BuildMI(MBB, I, DL, SMovB64, Rsrc01)
      .addReg(Lo, RegState::Kill);
    BuildMI(MBB, I, DL, SMovB64, Rsrc23)
      .addReg(Hi, RegState::Kill);
  } else {
    unsigned Rsrc0 = TRI->getSubReg(ScratchRsrcReg, AMDGPU::sub0);
    unsigned Rsrc1 = TRI->getSubReg(ScratchRsrcReg, AMDGPU::sub1);
    unsigned Rsrc2 = TRI->getSubReg(ScratchRsrcReg, AMDGPU::sub2);
    unsigned Rsrc3 = TRI->getSubReg(ScratchRsrcReg, AMDGPU::sub3);

    // Use relocations to get the pointer, and setup the other bits manually.
    uint64_t Rsrc23 = TII->getScratchRsrcWords23();
    BuildMI(MBB, I, DL, SMovB32, Rsrc0)
      .addReg(ScratchRsrcReg, RegState::ImplicitDefine);

    BuildMI(MBB, I, DL, SMovB32, Rsrc1)
      .addReg(ScratchRsrcReg, RegState::ImplicitDefine);

    BuildMI(MBB, I, DL, SMovB32, Rsrc2)
      .addImm(Rsrc23 & 0xffffffff)
      .addReg(ScratchRsrcReg, RegState::ImplicitDefine);

    BuildMI(MBB, I, DL, SMovB32, Rsrc3)
      .addImm(Rsrc23 >> 32)
      .addReg(ScratchRsrcReg, RegState::ImplicitDefine);

  // Make the register selected live throughout the function.
  for (MachineBasicBlock &OtherBB : MF) {
    if (&OtherBB == &MBB)


void SIFrameLowering::emitEpilogue(MachineFunction &MF,
                                   MachineBasicBlock &MBB) const {


void SIFrameLowering::processFunctionBeforeFrameFinalized(
  MachineFunction &MF,
  RegScavenger *RS) const {
  MachineFrameInfo *MFI = MF.getFrameInfo();

  if (!MFI->hasStackObjects())

  bool MayNeedScavengingEmergencySlot = MFI->hasStackObjects();

  assert((RS || !MayNeedScavengingEmergencySlot) &&
         "RegScavenger required if spilling");

  if (MayNeedScavengingEmergencySlot) {
    int ScavengeFI = MFI->CreateSpillStackObject(

void SIFrameLowering::emitDebuggerPrologue(MachineFunction &MF,
                                           MachineBasicBlock &MBB) const {
  const SISubtarget &ST = MF.getSubtarget<SISubtarget>();
  const SIInstrInfo *TII = ST.getInstrInfo();
  const SIRegisterInfo *TRI = &TII->getRegisterInfo();
  const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();

  MachineBasicBlock::iterator I = MBB.begin();
  DebugLoc DL;

  // For each dimension:
  for (unsigned i = 0; i < 3; ++i) {
    // Get work group ID SGPR, and make it live-in again.
    unsigned WorkGroupIDSGPR = MFI->getWorkGroupIDSGPR(i);

    // Since SGPRs are spilled into VGPRs, copy work group ID SGPR to VGPR in
    // order to spill it to scratch.
    unsigned WorkGroupIDVGPR =
    BuildMI(MBB, I, DL, TII->get(AMDGPU::V_MOV_B32_e32), WorkGroupIDVGPR)

    // Spill work group ID.
    int WorkGroupIDObjectIdx = MFI->getDebuggerWorkGroupIDStackObjectIndex(i);
    TII->storeRegToStackSlot(MBB, I, WorkGroupIDVGPR, false,
      WorkGroupIDObjectIdx, &AMDGPU::VGPR_32RegClass, TRI);

    // Get work item ID VGPR, and make it live-in again.
    unsigned WorkItemIDVGPR = MFI->getWorkItemIDVGPR(i);

    // Spill work item ID.
    int WorkItemIDObjectIdx = MFI->getDebuggerWorkItemIDStackObjectIndex(i);
    TII->storeRegToStackSlot(MBB, I, WorkItemIDVGPR, false,
      WorkItemIDObjectIdx, &AMDGPU::VGPR_32RegClass, TRI);