From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Fri, 13 Jul 2018 16:40:37 +0000 (+0000)
Subject: AMDGPU: Properly handle shader inputs with split arguments
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9c21e67b4cbd6e8ab29f61e95a393b2a4ae60da7;p=llvm

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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337022 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp b/lib/Target/AMDGPU/SIISelLowering.cpp
index 177cec982f3..ea8578fb19d 100644
--- a/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1265,36 +1265,51 @@ static void processShaderInputArgs(SmallVectorImpl<ISD::InputArg> &Splits,
                                    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) {
+
+      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 (!Arg.Used && !Info->isPSInputAllocated(PSInputNum)) {
+      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(SmallVectorImpl<ISD::InputArg> &Splits,
         NewArg.PartOffset += NewArg.VT.getStoreSize();
       }
     } else {
-      Splits.push_back(Arg);
+      Splits.push_back(*Arg);
     }
   }
 }
@@ -1784,7 +1799,7 @@ SDValue SITargetLowering::LowerFormalArguments(
 
    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;
     }