aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp')
-rw-r--r--lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp134
1 files changed, 68 insertions, 66 deletions
diff --git a/lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp b/lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp
index 0e788df1c9c0..a19020276f35 100644
--- a/lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp
+++ b/lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp
@@ -19,6 +19,7 @@
#include "R600InstrInfo.h"
#include "R600MachineFunctionInfo.h"
#include "R600RegisterInfo.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringRef.h"
@@ -93,7 +94,7 @@ bool CFStack::branchStackContains(CFStack::StackItem Item) {
}
bool CFStack::requiresWorkAroundForInst(unsigned Opcode) {
- if (Opcode == AMDGPU::CF_ALU_PUSH_BEFORE && ST->hasCaymanISA() &&
+ if (Opcode == R600::CF_ALU_PUSH_BEFORE && ST->hasCaymanISA() &&
getLoopDepth() > 1)
return true;
@@ -102,10 +103,10 @@ bool CFStack::requiresWorkAroundForInst(unsigned Opcode) {
switch(Opcode) {
default: return false;
- case AMDGPU::CF_ALU_PUSH_BEFORE:
- case AMDGPU::CF_ALU_ELSE_AFTER:
- case AMDGPU::CF_ALU_BREAK:
- case AMDGPU::CF_ALU_CONTINUE:
+ case R600::CF_ALU_PUSH_BEFORE:
+ case R600::CF_ALU_ELSE_AFTER:
+ case R600::CF_ALU_BREAK:
+ case R600::CF_ALU_CONTINUE:
if (CurrentSubEntries == 0)
return false;
if (ST->getWavefrontSize() == 64) {
@@ -136,7 +137,7 @@ unsigned CFStack::getSubEntrySize(CFStack::StackItem Item) {
return 0;
case CFStack::FIRST_NON_WQM_PUSH:
assert(!ST->hasCaymanISA());
- if (ST->getGeneration() <= R600Subtarget::R700) {
+ if (ST->getGeneration() <= AMDGPUSubtarget::R700) {
// +1 For the push operation.
// +2 Extra space required.
return 3;
@@ -149,7 +150,7 @@ unsigned CFStack::getSubEntrySize(CFStack::StackItem Item) {
return 2;
}
case CFStack::FIRST_NON_WQM_PUSH_W_FULL_ENTRY:
- assert(ST->getGeneration() >= R600Subtarget::EVERGREEN);
+ assert(ST->getGeneration() >= AMDGPUSubtarget::EVERGREEN);
// +1 For the push operation.
// +1 Extra space required.
return 2;
@@ -167,8 +168,8 @@ void CFStack::updateMaxStackSize() {
void CFStack::pushBranch(unsigned Opcode, bool isWQM) {
CFStack::StackItem Item = CFStack::ENTRY;
switch(Opcode) {
- case AMDGPU::CF_PUSH_EG:
- case AMDGPU::CF_ALU_PUSH_BEFORE:
+ case R600::CF_PUSH_EG:
+ case R600::CF_ALU_PUSH_BEFORE:
if (!isWQM) {
if (!ST->hasCaymanISA() &&
!branchStackContains(CFStack::FIRST_NON_WQM_PUSH))
@@ -176,7 +177,7 @@ void CFStack::pushBranch(unsigned Opcode, bool isWQM) {
// See comment in
// CFStack::getSubEntrySize()
else if (CurrentEntries > 0 &&
- ST->getGeneration() > R600Subtarget::EVERGREEN &&
+ ST->getGeneration() > AMDGPUSubtarget::EVERGREEN &&
!ST->hasCaymanISA() &&
!branchStackContains(CFStack::FIRST_NON_WQM_PUSH_W_FULL_ENTRY))
Item = CFStack::FIRST_NON_WQM_PUSH_W_FULL_ENTRY;
@@ -239,8 +240,8 @@ private:
bool IsTrivialInst(MachineInstr &MI) const {
switch (MI.getOpcode()) {
- case AMDGPU::KILL:
- case AMDGPU::RETURN:
+ case R600::KILL:
+ case R600::RETURN:
return true;
default:
return false;
@@ -249,44 +250,44 @@ private:
const MCInstrDesc &getHWInstrDesc(ControlFlowInstruction CFI) const {
unsigned Opcode = 0;
- bool isEg = (ST->getGeneration() >= R600Subtarget::EVERGREEN);
+ bool isEg = (ST->getGeneration() >= AMDGPUSubtarget::EVERGREEN);
switch (CFI) {
case CF_TC:
- Opcode = isEg ? AMDGPU::CF_TC_EG : AMDGPU::CF_TC_R600;
+ Opcode = isEg ? R600::CF_TC_EG : R600::CF_TC_R600;
break;
case CF_VC:
- Opcode = isEg ? AMDGPU::CF_VC_EG : AMDGPU::CF_VC_R600;
+ Opcode = isEg ? R600::CF_VC_EG : R600::CF_VC_R600;
break;
case CF_CALL_FS:
- Opcode = isEg ? AMDGPU::CF_CALL_FS_EG : AMDGPU::CF_CALL_FS_R600;
+ Opcode = isEg ? R600::CF_CALL_FS_EG : R600::CF_CALL_FS_R600;
break;
case CF_WHILE_LOOP:
- Opcode = isEg ? AMDGPU::WHILE_LOOP_EG : AMDGPU::WHILE_LOOP_R600;
+ Opcode = isEg ? R600::WHILE_LOOP_EG : R600::WHILE_LOOP_R600;
break;
case CF_END_LOOP:
- Opcode = isEg ? AMDGPU::END_LOOP_EG : AMDGPU::END_LOOP_R600;
+ Opcode = isEg ? R600::END_LOOP_EG : R600::END_LOOP_R600;
break;
case CF_LOOP_BREAK:
- Opcode = isEg ? AMDGPU::LOOP_BREAK_EG : AMDGPU::LOOP_BREAK_R600;
+ Opcode = isEg ? R600::LOOP_BREAK_EG : R600::LOOP_BREAK_R600;
break;
case CF_LOOP_CONTINUE:
- Opcode = isEg ? AMDGPU::CF_CONTINUE_EG : AMDGPU::CF_CONTINUE_R600;
+ Opcode = isEg ? R600::CF_CONTINUE_EG : R600::CF_CONTINUE_R600;
break;
case CF_JUMP:
- Opcode = isEg ? AMDGPU::CF_JUMP_EG : AMDGPU::CF_JUMP_R600;
+ Opcode = isEg ? R600::CF_JUMP_EG : R600::CF_JUMP_R600;
break;
case CF_ELSE:
- Opcode = isEg ? AMDGPU::CF_ELSE_EG : AMDGPU::CF_ELSE_R600;
+ Opcode = isEg ? R600::CF_ELSE_EG : R600::CF_ELSE_R600;
break;
case CF_POP:
- Opcode = isEg ? AMDGPU::POP_EG : AMDGPU::POP_R600;
+ Opcode = isEg ? R600::POP_EG : R600::POP_R600;
break;
case CF_END:
if (ST->hasCaymanISA()) {
- Opcode = AMDGPU::CF_END_CM;
+ Opcode = R600::CF_END_CM;
break;
}
- Opcode = isEg ? AMDGPU::CF_END_EG : AMDGPU::CF_END_R600;
+ Opcode = isEg ? R600::CF_END_EG : R600::CF_END_R600;
break;
}
assert (Opcode && "No opcode selected");
@@ -304,21 +305,21 @@ private:
continue;
if (MO.isDef()) {
unsigned Reg = MO.getReg();
- if (AMDGPU::R600_Reg128RegClass.contains(Reg))
+ if (R600::R600_Reg128RegClass.contains(Reg))
DstMI = Reg;
else
DstMI = TRI->getMatchingSuperReg(Reg,
- TRI->getSubRegFromChannel(TRI->getHWRegChan(Reg)),
- &AMDGPU::R600_Reg128RegClass);
+ AMDGPURegisterInfo::getSubRegFromChannel(TRI->getHWRegChan(Reg)),
+ &R600::R600_Reg128RegClass);
}
if (MO.isUse()) {
unsigned Reg = MO.getReg();
- if (AMDGPU::R600_Reg128RegClass.contains(Reg))
+ if (R600::R600_Reg128RegClass.contains(Reg))
SrcMI = Reg;
else
SrcMI = TRI->getMatchingSuperReg(Reg,
- TRI->getSubRegFromChannel(TRI->getHWRegChan(Reg)),
- &AMDGPU::R600_Reg128RegClass);
+ AMDGPURegisterInfo::getSubRegFromChannel(TRI->getHWRegChan(Reg)),
+ &R600::R600_Reg128RegClass);
}
}
if ((DstRegs.find(SrcMI) == DstRegs.end())) {
@@ -358,15 +359,15 @@ private:
void getLiteral(MachineInstr &MI, std::vector<MachineOperand *> &Lits) const {
static const unsigned LiteralRegs[] = {
- AMDGPU::ALU_LITERAL_X,
- AMDGPU::ALU_LITERAL_Y,
- AMDGPU::ALU_LITERAL_Z,
- AMDGPU::ALU_LITERAL_W
+ R600::ALU_LITERAL_X,
+ R600::ALU_LITERAL_Y,
+ R600::ALU_LITERAL_Z,
+ R600::ALU_LITERAL_W
};
const SmallVector<std::pair<MachineOperand *, int64_t>, 3> Srcs =
TII->getSrcs(MI);
for (const auto &Src:Srcs) {
- if (Src.first->getReg() != AMDGPU::ALU_LITERAL_X)
+ if (Src.first->getReg() != R600::ALU_LITERAL_X)
continue;
int64_t Imm = Src.second;
std::vector<MachineOperand *>::iterator It =
@@ -376,7 +377,7 @@ private:
// Get corresponding Operand
MachineOperand &Operand = MI.getOperand(
- TII->getOperandIdx(MI.getOpcode(), AMDGPU::OpName::literal));
+ TII->getOperandIdx(MI.getOpcode(), R600::OpName::literal));
if (It != Lits.end()) {
// Reuse existing literal reg
@@ -399,7 +400,7 @@ private:
unsigned LiteralPair0 = Literals[i];
unsigned LiteralPair1 = (i + 1 < e)?Literals[i + 1]:0;
InsertPos = BuildMI(MBB, InsertPos->getDebugLoc(),
- TII->get(AMDGPU::LITERALS))
+ TII->get(R600::LITERALS))
.addImm(LiteralPair0)
.addImm(LiteralPair1);
}
@@ -441,7 +442,7 @@ private:
}
for (unsigned i = 0, e = Literals.size(); i < e; i += 2) {
MachineInstrBuilder MILit = BuildMI(MBB, I, I->getDebugLoc(),
- TII->get(AMDGPU::LITERALS));
+ TII->get(R600::LITERALS));
if (Literals[i]->isImm()) {
MILit.addImm(Literals[i]->getImm());
} else {
@@ -470,7 +471,7 @@ private:
unsigned &CfCount) {
CounterPropagateAddr(*Clause.first, CfCount);
MachineBasicBlock *BB = Clause.first->getParent();
- BuildMI(BB, DL, TII->get(AMDGPU::FETCH_CLAUSE)).addImm(CfCount);
+ BuildMI(BB, DL, TII->get(R600::FETCH_CLAUSE)).addImm(CfCount);
for (unsigned i = 0, e = Clause.second.size(); i < e; ++i) {
BB->splice(InsertPos, BB, Clause.second[i]);
}
@@ -482,7 +483,7 @@ private:
Clause.first->getOperand(0).setImm(0);
CounterPropagateAddr(*Clause.first, CfCount);
MachineBasicBlock *BB = Clause.first->getParent();
- BuildMI(BB, DL, TII->get(AMDGPU::ALU_CLAUSE)).addImm(CfCount);
+ BuildMI(BB, DL, TII->get(R600::ALU_CLAUSE)).addImm(CfCount);
for (unsigned i = 0, e = Clause.second.size(); i < e; ++i) {
BB->splice(InsertPos, BB, Clause.second[i]);
}
@@ -531,7 +532,7 @@ public:
for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
I != E;) {
if (TII->usesTextureCache(*I) || TII->usesVertexCache(*I)) {
- DEBUG(dbgs() << CfCount << ":"; I->dump(););
+ LLVM_DEBUG(dbgs() << CfCount << ":"; I->dump(););
FetchClauses.push_back(MakeFetchClause(MBB, I));
CfCount++;
LastAlu.back() = nullptr;
@@ -539,33 +540,34 @@ public:
}
MachineBasicBlock::iterator MI = I;
- if (MI->getOpcode() != AMDGPU::ENDIF)
+ if (MI->getOpcode() != R600::ENDIF)
LastAlu.back() = nullptr;
- if (MI->getOpcode() == AMDGPU::CF_ALU)
+ if (MI->getOpcode() == R600::CF_ALU)
LastAlu.back() = &*MI;
I++;
bool RequiresWorkAround =
CFStack.requiresWorkAroundForInst(MI->getOpcode());
switch (MI->getOpcode()) {
- case AMDGPU::CF_ALU_PUSH_BEFORE:
+ case R600::CF_ALU_PUSH_BEFORE:
if (RequiresWorkAround) {
- DEBUG(dbgs() << "Applying bug work-around for ALU_PUSH_BEFORE\n");
- BuildMI(MBB, MI, MBB.findDebugLoc(MI), TII->get(AMDGPU::CF_PUSH_EG))
+ LLVM_DEBUG(dbgs()
+ << "Applying bug work-around for ALU_PUSH_BEFORE\n");
+ BuildMI(MBB, MI, MBB.findDebugLoc(MI), TII->get(R600::CF_PUSH_EG))
.addImm(CfCount + 1)
.addImm(1);
- MI->setDesc(TII->get(AMDGPU::CF_ALU));
+ MI->setDesc(TII->get(R600::CF_ALU));
CfCount++;
- CFStack.pushBranch(AMDGPU::CF_PUSH_EG);
+ CFStack.pushBranch(R600::CF_PUSH_EG);
} else
- CFStack.pushBranch(AMDGPU::CF_ALU_PUSH_BEFORE);
+ CFStack.pushBranch(R600::CF_ALU_PUSH_BEFORE);
LLVM_FALLTHROUGH;
- case AMDGPU::CF_ALU:
+ case R600::CF_ALU:
I = MI;
AluClauses.push_back(MakeALUClause(MBB, I));
- DEBUG(dbgs() << CfCount << ":"; MI->dump(););
+ LLVM_DEBUG(dbgs() << CfCount << ":"; MI->dump(););
CfCount++;
break;
- case AMDGPU::WHILELOOP: {
+ case R600::WHILELOOP: {
CFStack.pushLoop();
MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
getHWInstrDesc(CF_WHILE_LOOP))
@@ -578,7 +580,7 @@ public:
CfCount++;
break;
}
- case AMDGPU::ENDLOOP: {
+ case R600::ENDLOOP: {
CFStack.popLoop();
std::pair<unsigned, std::set<MachineInstr *>> Pair =
std::move(LoopStack.back());
@@ -590,19 +592,19 @@ public:
CfCount++;
break;
}
- case AMDGPU::IF_PREDICATE_SET: {
+ case R600::IF_PREDICATE_SET: {
LastAlu.push_back(nullptr);
MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
getHWInstrDesc(CF_JUMP))
.addImm(0)
.addImm(0);
IfThenElseStack.push_back(MIb);
- DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
+ LLVM_DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
MI->eraseFromParent();
CfCount++;
break;
}
- case AMDGPU::ELSE: {
+ case R600::ELSE: {
MachineInstr * JumpInst = IfThenElseStack.back();
IfThenElseStack.pop_back();
CounterPropagateAddr(*JumpInst, CfCount);
@@ -610,13 +612,13 @@ public:
getHWInstrDesc(CF_ELSE))
.addImm(0)
.addImm(0);
- DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
+ LLVM_DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
IfThenElseStack.push_back(MIb);
MI->eraseFromParent();
CfCount++;
break;
}
- case AMDGPU::ENDIF: {
+ case R600::ENDIF: {
CFStack.popBranch();
if (LastAlu.back()) {
ToPopAfter.push_back(LastAlu.back());
@@ -626,7 +628,7 @@ public:
.addImm(CfCount + 1)
.addImm(1);
(void)MIb;
- DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
+ LLVM_DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
CfCount++;
}
@@ -638,7 +640,7 @@ public:
MI->eraseFromParent();
break;
}
- case AMDGPU::BREAK: {
+ case R600::BREAK: {
CfCount ++;
MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
getHWInstrDesc(CF_LOOP_BREAK))
@@ -647,7 +649,7 @@ public:
MI->eraseFromParent();
break;
}
- case AMDGPU::CONTINUE: {
+ case R600::CONTINUE: {
MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
getHWInstrDesc(CF_LOOP_CONTINUE))
.addImm(0);
@@ -656,12 +658,12 @@ public:
CfCount++;
break;
}
- case AMDGPU::RETURN: {
+ case R600::RETURN: {
DebugLoc DL = MBB.findDebugLoc(MI);
BuildMI(MBB, MI, DL, getHWInstrDesc(CF_END));
CfCount++;
if (CfCount % 2) {
- BuildMI(MBB, I, DL, TII->get(AMDGPU::PAD));
+ BuildMI(MBB, I, DL, TII->get(R600::PAD));
CfCount++;
}
MI->eraseFromParent();
@@ -673,7 +675,7 @@ public:
}
default:
if (TII->isExport(MI->getOpcode())) {
- DEBUG(dbgs() << CfCount << ":"; MI->dump(););
+ LLVM_DEBUG(dbgs() << CfCount << ":"; MI->dump(););
CfCount++;
}
break;
@@ -682,7 +684,7 @@ public:
for (unsigned i = 0, e = ToPopAfter.size(); i < e; ++i) {
MachineInstr *Alu = ToPopAfter[i];
BuildMI(MBB, Alu, MBB.findDebugLoc((MachineBasicBlock::iterator)Alu),
- TII->get(AMDGPU::CF_ALU_POP_AFTER))
+ TII->get(R600::CF_ALU_POP_AFTER))
.addImm(Alu->getOperand(0).getImm())
.addImm(Alu->getOperand(1).getImm())
.addImm(Alu->getOperand(2).getImm())
generated by cgit v1.2.3 (git 2.25.1) at 2024-05-12 22:00:25 +0000