Skip to content

Commit f001bfc

Browse files
krystian-andrzejewskikrystian-andrzejewski
committed
[Autobackout][FuncReg]Revert of change: c7d4a55
Improve IGCConstProp pass to support constant folding in rounding-mode-specific GenISA intrinsics. Add a new GenISA intrinsic for floating point conversion with the round-to-nearest rounding mode. The use case is when the default shader rounding mode is not set to round to nearest. Change-Id: I812417d15be705d8bbe02c8ed2f43af425c44006
1 parent 2785473 commit f001bfc

File tree

5 files changed

+13
-144
lines changed

5 files changed

+13
-144
lines changed

‎IGC/Compiler/CISACodeGen/EmitVISAPass.cpp‎

Lines changed: 0 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -7825,7 +7825,6 @@ void EmitPass::EmitGenIntrinsicMessage(llvm::GenIntrinsicInst* inst)
78257825
case GenISAIntrinsic::GenISA_uitof_rtn:
78267826
case GenISAIntrinsic::GenISA_uitof_rtp:
78277827
case GenISAIntrinsic::GenISA_uitof_rtz:
7828-
case GenISAIntrinsic::GenISA_ftof_rte:
78297828
case GenISAIntrinsic::GenISA_ftof_rtn:
78307829
case GenISAIntrinsic::GenISA_ftof_rtp:
78317830
case GenISAIntrinsic::GenISA_ftof_rtz:
@@ -13914,9 +13913,6 @@ ERoundingMode EmitPass::GetRoundingMode_FP(Instruction* inst)
1391413913
case GenISAIntrinsic::GenISA_uitof_rtp:
1391513914
RM = ERoundingMode::ROUND_TO_POSITIVE;
1391613915
break;
13917-
case GenISAIntrinsic::GenISA_ftof_rte:
13918-
RM = ERoundingMode::ROUND_TO_NEAREST_EVEN;
13919-
break;
1392013916
default:
1392113917
break;
1392213918
}

‎IGC/Compiler/CISACodeGen/opCode.h‎

Lines changed: 0 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -62,9 +62,7 @@ DECLARE_OPCODE(GenISA_PullCentroidBarys, GenISAIntrinsic, llvm_pul
6262
DECLARE_OPCODE(sqrt, Intrinsic, llvm_sqrt, true, true, true, false, true, false, false)
6363
DECLARE_OPCODE(GenISA_rsq, GenISAIntrinsic, llvm_rsq, true, true, true, false, true, false, false)
6464
DECLARE_OPCODE(FMul, Instruction, llvm_fmul, true, true, true, true, false, false, false)
65-
DECLARE_OPCODE(GenISA_mul_rtz, GenISAIntrinsic, llvm_fmul_rtz, true, true, true, true, false, false, false)
6665
DECLARE_OPCODE(FAdd, Instruction, llvm_fadd, true, true, true, true, false, false, false)
67-
DECLARE_OPCODE(GenISA_add_rtz, GenISAIntrinsic, llvm_fadd_rtz, true, true, true, true, false, false, false)
6866
DECLARE_OPCODE(FSub, Instruction, llvm_fsub, true, true, true, true, false, false, false)
6967
DECLARE_OPCODE(FDiv, Instruction, llvm_fdiv, true, true, true, false, false, false, false)
7068
DECLARE_OPCODE(FRem, Instruction, llvm_frem, false, false, false, false, false, false, false)
@@ -119,10 +117,6 @@ DECLARE_OPCODE(SExt, Instruction, llvm_sext, true, true, false, false, false, fa
119117
DECLARE_OPCODE(ZExt, Instruction, llvm_zext, false, true, false, false, false, false, false)
120118
DECLARE_OPCODE(Trunc, Instruction, llvm_trunc, true, false, false, false, false, false, false)
121119
DECLARE_OPCODE(FPTrunc, Instruction, llvm_fptrunc, true, false, false, false, false, false, false)
122-
DECLARE_OPCODE(GenISA_ftof_rte, GenISAIntrinsic, llvm_fptrunc_rte, true, false, false, false, false, false, false)
123-
DECLARE_OPCODE(GenISA_ftof_rtz, GenISAIntrinsic, llvm_fptrunc_rtz, true, false, false, false, false, false, false)
124-
DECLARE_OPCODE(GenISA_ftof_rtp, GenISAIntrinsic, llvm_fptrunc_rtp, true, false, false, false, false, false, false)
125-
DECLARE_OPCODE(GenISA_ftof_rtn, GenISAIntrinsic, llvm_fptrunc_rtn, true, false, false, false, false, false, false)
126120
DECLARE_OPCODE(BitCast, Instruction, llvm_bitcast, false, false, false, false, false, false, false)
127121
DECLARE_OPCODE(PtrToInt, Instruction, llvm_ptrtoint, false, false, false, false, false, false, false)
128122
DECLARE_OPCODE(IntToPtr, Instruction, llvm_inttoptr, false, false, false, false, false, false, false)

‎IGC/Compiler/CustomSafeOptPass.cpp‎

Lines changed: 13 additions & 130 deletions
Original file line numberDiff line numberDiff line change
@@ -2308,84 +2308,6 @@ void GenSpecificPattern::visitFNeg(llvm::UnaryOperator& I)
23082308
}
23092309
#endif
23102310

2311-
llvm::Constant* IGC::IGCConstantFolder::CreateFAdd(llvm::Constant* C0, llvm::Constant* C1, llvm::APFloatBase::roundingMode roundingMode) const
2312-
{
2313-
if (llvm::isa<llvm::UndefValue>(C0) || llvm::isa<llvm::UndefValue>(C1))
2314-
{
2315-
return llvm::ConstantFolder::CreateFAdd(C0, C1);
2316-
}
2317-
llvm::ConstantFP* CFP0 = llvm::cast<ConstantFP>(C0);
2318-
llvm::ConstantFP* CFP1 = llvm::cast<ConstantFP>(C1);
2319-
APFloat firstOperand = CFP0->getValueAPF();
2320-
APFloat secondOperand = CFP1->getValueAPF();
2321-
APFloat::opStatus status = firstOperand.add(secondOperand, roundingMode);
2322-
if (status != APFloat::opInvalidOp)
2323-
{
2324-
return llvm::ConstantFP::get(C0->getContext(), firstOperand);
2325-
}
2326-
else
2327-
{
2328-
return nullptr;
2329-
}
2330-
}
2331-
2332-
llvm::Constant* IGC::IGCConstantFolder::CreateFMul(llvm::Constant* C0, llvm::Constant* C1, llvm::APFloatBase::roundingMode roundingMode) const
2333-
{
2334-
if (llvm::isa<llvm::UndefValue>(C0) || llvm::isa<llvm::UndefValue>(C1))
2335-
{
2336-
return llvm::ConstantFolder::CreateFMul(C0, C1);
2337-
}
2338-
llvm::ConstantFP* CFP0 = llvm::cast<ConstantFP>(C0);
2339-
llvm::ConstantFP* CFP1 = llvm::cast<ConstantFP>(C1);
2340-
APFloat firstOperand = CFP0->getValueAPF();
2341-
APFloat secondOperand = CFP1->getValueAPF();
2342-
APFloat::opStatus status = firstOperand.multiply(secondOperand, roundingMode);
2343-
if (status != APFloat::opInvalidOp)
2344-
{
2345-
return llvm::ConstantFP::get(C0->getContext(), firstOperand);
2346-
}
2347-
else
2348-
{
2349-
return nullptr;
2350-
}
2351-
}
2352-
2353-
llvm::Constant* IGC::IGCConstantFolder::CreateFPTrunc(llvm::Constant* C0, llvm::Type* dstType, llvm::APFloatBase::roundingMode roundingMode) const
2354-
{
2355-
if (llvm::isa<llvm::UndefValue>(C0))
2356-
{
2357-
return llvm::ConstantFolder::CreateFPCast(C0, dstType);
2358-
}
2359-
APFloat APF = llvm::cast<ConstantFP>(C0)->getValueAPF();
2360-
const fltSemantics& outputSemantics = dstType->isHalfTy() ? APFloatBase::IEEEhalf() :
2361-
dstType->isFloatTy() ? APFloatBase::IEEEsingle() :
2362-
APFloatBase::IEEEdouble();
2363-
bool losesInfo = false;
2364-
APFloat::opStatus status = APF.convert(outputSemantics, roundingMode, &losesInfo);
2365-
if (status != APFloat::opInvalidOp)
2366-
{
2367-
return llvm::ConstantFP::get(C0->getContext(), APF);
2368-
}
2369-
else
2370-
{
2371-
return nullptr;
2372-
}
2373-
}
2374-
2375-
llvm::Constant* IGC::IGCConstantFolder::CreateCanonicalize(llvm::Constant* C0, bool flushDenorms /*= true*/) const
2376-
{
2377-
if (llvm::isa<llvm::UndefValue>(C0))
2378-
{
2379-
return C0;
2380-
}
2381-
auto APF = llvm::cast<ConstantFP>(C0)->getValueAPF();
2382-
if (flushDenorms && APF.isDenormal())
2383-
{
2384-
APF = APFloat::getZero(APF.getSemantics(), APF.isNegative());
2385-
}
2386-
return ConstantFP::get(C0->getContext(), APF);
2387-
}
2388-
23892311
// Register pass to igc-opt
23902312
#define PASS_FLAG3 "igc-const-prop"
23912313
#define PASS_DESCRIPTION3 "Custom Const-prop Pass"
@@ -2592,6 +2514,17 @@ Constant* IGCConstProp::replaceShaderConstant(LoadInst* inst)
25922514
return nullptr;
25932515
}
25942516

2517+
llvm::Constant* IGC::IGCConstantFolder::CreateCanonicalize(llvm::Constant* C0, bool flushDenorms /*= true*/) const
2518+
{
2519+
assert(llvm::isa<ConstantFP>(C0));
2520+
auto APF = llvm::cast<ConstantFP>(C0)->getValueAPF();
2521+
if (flushDenorms && APF.isDenormal())
2522+
{
2523+
APF = APFloat::getZero(APF.getSemantics(), APF.isNegative());
2524+
}
2525+
return ConstantFP::get(C0->getContext(), APF);
2526+
}
2527+
25952528
Constant* IGCConstProp::ConstantFoldCallInstruction(CallInst* inst)
25962529
{
25972530
Constant* C = nullptr;
@@ -2601,7 +2534,6 @@ Constant* IGCConstProp::ConstantFoldCallInstruction(CallInst* inst)
26012534
// used for GenISA_sqrt, GenISA_rsq and GenISA_ROUNDNE
26022535
ConstantFP* C0 = dyn_cast<ConstantFP>(inst->getOperand(0));
26032536
EOPCODE igcop = GetOpCode(inst);
2604-
IGCConstantFolder folder;
26052537

26062538
// special case of gen-intrinsic
26072539
switch (igcop)
@@ -2696,67 +2628,18 @@ Constant* IGCConstProp::ConstantFoldCallInstruction(CallInst* inst)
26962628
}
26972629
}
26982630
break;
2699-
case llvm_fptrunc_rte:
2700-
{
2701-
if (C0)
2702-
{
2703-
C = folder.CreateFPTrunc(C0, inst->getType(), llvm::APFloatBase::rmNearestTiesToEven);
2704-
}
2705-
}
2706-
break;
2707-
case llvm_fptrunc_rtz:
2708-
{
2709-
if (C0)
2710-
{
2711-
C = folder.CreateFPTrunc(C0, inst->getType(), llvm::APFloatBase::rmTowardZero);
2712-
}
2713-
}
2714-
break;
2715-
case llvm_fptrunc_rtp:
2716-
{
2717-
if (C0)
2718-
{
2719-
C = folder.CreateFPTrunc(C0, inst->getType(), llvm::APFloatBase::rmTowardPositive);
2720-
}
2721-
}
2722-
break;
2723-
case llvm_fptrunc_rtn:
2724-
{
2725-
if (C0)
2726-
{
2727-
C = folder.CreateFPTrunc(C0, inst->getType(), llvm::APFloatBase::rmTowardNegative);
2728-
}
2729-
}
2730-
break;
2731-
case llvm_fadd_rtz:
2732-
{
2733-
Constant* C1 = dyn_cast<Constant>(inst->getOperand(1));
2734-
if (C0 && C1)
2735-
{
2736-
C = folder.CreateFAdd(C0, C1, llvm::APFloatBase::rmTowardZero);
2737-
}
2738-
}
2739-
break;
2740-
case llvm_fmul_rtz:
2741-
{
2742-
Constant* C1 = dyn_cast<Constant>(inst->getOperand(1));
2743-
if (C0 && C1)
2744-
{
2745-
C = folder.CreateFMul(C0, C1, llvm::APFloatBase::rmTowardZero);
2746-
}
2747-
}
2748-
break;
27492631
case llvm_canonicalize:
27502632
{
27512633
// If the instruction should be emitted anyway, then remove the condition.
27522634
// Please, be aware of the fact that clients can understand the term canonical FP value in other way.
27532635
if (C0)
27542636
{
2637+
IGCConstantFolder constantFolder;
27552638
CodeGenContext* pCodeGenContext = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
27562639
bool flushVal = pCodeGenContext->m_floatDenormMode16 == ::IGC::FLOAT_DENORM_FLUSH_TO_ZERO && inst->getType()->isHalfTy();
27572640
flushVal = flushVal || (pCodeGenContext->m_floatDenormMode32 == ::IGC::FLOAT_DENORM_FLUSH_TO_ZERO && inst->getType()->isFloatTy());
27582641
flushVal = flushVal || (pCodeGenContext->m_floatDenormMode64 == ::IGC::FLOAT_DENORM_FLUSH_TO_ZERO && inst->getType()->isDoubleTy());
2759-
C = folder.CreateCanonicalize(C0, flushVal);
2642+
C = constantFolder.CreateCanonicalize(C0, flushVal);
27602643
}
27612644
}
27622645
break;

‎IGC/Compiler/CustomSafeOptPass.hpp‎

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -212,9 +212,6 @@ namespace IGC
212212
{}
213213

214214
llvm::Constant* CreateCanonicalize(llvm::Constant* C0, bool flushDenorms = true) const;
215-
llvm::Constant* CreateFAdd(llvm::Constant* C0, llvm::Constant* C1, llvm::APFloatBase::roundingMode roundingMode) const;
216-
llvm::Constant* CreateFMul(llvm::Constant* C0, llvm::Constant* C1, llvm::APFloatBase::roundingMode roundingMode) const;
217-
llvm::Constant* CreateFPTrunc(llvm::Constant* C0, llvm::Type* dstType, llvm::APFloatBase::roundingMode roundingMode) const;
218215
};
219216

220217
class IGCConstProp : public llvm::FunctionPass

‎IGC/GenISAIntrinsics/Intrinsic_definitions.py‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -66,7 +66,6 @@
6666
"GenISA_ftoui_rte": ["anyint",["anyfloat"],"NoMem"],
6767
"GenISA_ftoui_rtp": ["anyint",["anyfloat"],"NoMem"],
6868
"GenISA_ftoui_rtn": ["anyint",["anyfloat"],"NoMem"],
69-
"GenISA_ftof_rte": ["anyfloat",["anyfloat"],"NoMem"],
7069
"GenISA_ftof_rtn": ["anyfloat",["anyfloat"],"NoMem"],
7170
"GenISA_ftof_rtp": ["anyfloat",["anyfloat"],"NoMem"],
7271
"GenISA_ftof_rtz": ["anyfloat",["anyfloat"],"NoMem"],

0 commit comments

Comments
 (0)