[llvm] r337022 - AMDGPU: Properly handle shader inputs with split arguments
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 13 09:40:37 PDT 2018
Author: arsenm
Date: Fri Jul 13 09:40:37 2018
New Revision: 337022
URL: http://llvm.org/viewvc/llvm-project?rev=337022&view=rev
Log:
AMDGPU: Properly handle shader inputs with split arguments
This needs to refer to arguments by their original argument
index, not the argument split index which depends on what
the type splitting decides to do.
Also avoid increment PSInputNum for each split piece.
Modified:
llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
Modified: llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp?rev=337022&r1=337021&r2=337022&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp Fri Jul 13 09:40:37 2018
@@ -1265,36 +1265,51 @@ static void processShaderInputArgs(Small
FunctionType *FType,
SIMachineFunctionInfo *Info) {
for (unsigned I = 0, E = Ins.size(), PSInputNum = 0; I != E; ++I) {
- const ISD::InputArg &Arg = Ins[I];
+ const ISD::InputArg *Arg = &Ins[I];
// First check if it's a PS input addr.
- if (CallConv == CallingConv::AMDGPU_PS && !Arg.Flags.isInReg() &&
- !Arg.Flags.isByVal() && PSInputNum <= 15) {
+ if (CallConv == CallingConv::AMDGPU_PS &&
+ !Arg->Flags.isInReg() && !Arg->Flags.isByVal() && PSInputNum <= 15) {
- if (!Arg.Used && !Info->isPSInputAllocated(PSInputNum)) {
+ bool SkipArg = !Arg->Used && !Info->isPSInputAllocated(PSInputNum);
+
+ // Inconveniently only the first part of the split is marked as isSplit,
+ // so skip to the end. We only want to increment PSInputNum once for the
+ // entire split argument.
+ if (Arg->Flags.isSplit()) {
+ while (!Arg->Flags.isSplitEnd()) {
+ assert(!Arg->VT.isVector() &&
+ "unexpected vector split in ps argument type");
+ if (!SkipArg)
+ Splits.push_back(*Arg);
+ Arg = &Ins[++I];
+ }
+ }
+
+ if (SkipArg) {
// We can safely skip PS inputs.
- Skipped.set(I);
+ Skipped.set(Arg->getOrigArgIndex());
++PSInputNum;
continue;
}
Info->markPSInputAllocated(PSInputNum);
- if (Arg.Used)
+ if (Arg->Used)
Info->markPSInputEnabled(PSInputNum);
++PSInputNum;
}
// Second split vertices into their elements.
- if (Arg.VT.isVector()) {
- ISD::InputArg NewArg = Arg;
+ if (Arg->VT.isVector()) {
+ ISD::InputArg NewArg = *Arg;
NewArg.Flags.setSplit();
- NewArg.VT = Arg.VT.getVectorElementType();
+ NewArg.VT = Arg->VT.getVectorElementType();
// We REALLY want the ORIGINAL number of vertex elements here, e.g. a
// three or five element vertex only needs three or five registers,
// NOT four or eight.
- Type *ParamType = FType->getParamType(Arg.getOrigArgIndex());
+ Type *ParamType = FType->getParamType(Arg->getOrigArgIndex());
unsigned NumElements = ParamType->getVectorNumElements();
for (unsigned J = 0; J != NumElements; ++J) {
@@ -1302,7 +1317,7 @@ static void processShaderInputArgs(Small
NewArg.PartOffset += NewArg.VT.getStoreSize();
}
} else {
- Splits.push_back(Arg);
+ Splits.push_back(*Arg);
}
}
}
@@ -1784,7 +1799,7 @@ SDValue SITargetLowering::LowerFormalArg
for (unsigned i = 0, e = Ins.size(), ArgIdx = 0; i != e; ++i) {
const ISD::InputArg &Arg = Ins[i];
- if (Skipped[i]) {
+ if (Arg.isOrigArg() && Skipped[Arg.getOrigArgIndex()]) {
InVals.push_back(DAG.getUNDEF(Arg.VT));
continue;
}
More information about the llvm-commits
mailing list