Merge pull request #126 from sx-aurora-dev/merge/vp-opaque-ptr

kaz7 · web-flow · commit ebe1bd022a54 · 2021-12-17T11:45:03.000+09:00
Merge/vp opaque ptr
diff --git a/llvm/include/llvm/IR/IRBuilder.h b/llvm/include/llvm/IR/IRBuilder.h
@@ -783,9 +783,10 @@ class IRBuilderBase {
 
 
   /// Call an arithmetic VP intrinsic.
-  Instruction *CreateVectorPredicatedInst(unsigned OC, ArrayRef<Value *>,
-                             Instruction *FMFSource = nullptr,
-                             const Twine &Name = "");
+  Instruction *CreateVectorPredicatedInst(unsigned OC, Type *ReturnTy,
+                                          ArrayRef<Value *>,
+                                          Instruction *FMFSource = nullptr,
+                                          const Twine &Name = "");
 
   /// Call an comparison VP intrinsic.
   Instruction *CreateVectorPredicatedCmp(CmpInst::Predicate Pred,
diff --git a/llvm/include/llvm/IR/IntrinsicInst.h b/llvm/include/llvm/IR/IntrinsicInst.h
@@ -399,10 +399,11 @@ class VPIntrinsic : public IntrinsicInst {
   static bool HasExceptionMode(Intrinsic::ID VPID);
 
   /// \brief Declares a llvm.vp.* intrinsic in \p M that matches the parameters
-  /// \p Params.
+  /// \p Params. Additionally, the load and gather intrinsics require
+  /// \p ReturnType to be specified.
   static Function *getDeclarationForParams(Module *M, Intrinsic::ID,
-                                           ArrayRef<Value *> Params,
-                                           Type *VecRetTy = nullptr);
+                                           Type *ReturnType,
+                                           ArrayRef<Value *> Params);
 
   static Optional<unsigned> getMaskParamPos(Intrinsic::ID IntrinsicID);
   static Optional<unsigned> getVectorLengthParamPos(Intrinsic::ID IntrinsicID);
diff --git a/llvm/include/llvm/IR/VPBuilder.h b/llvm/include/llvm/IR/VPBuilder.h
@@ -67,9 +67,9 @@ class VPBuilder {
 
   // Memory
   Value& CreateContiguousStore(Value & Val, Value & Pointer, MaybeAlign Alignment);
-  Value& CreateContiguousLoad(Value & Pointer, MaybeAlign Alignment);
+  Value& CreateContiguousLoad(Type *ReturnTy, Value & Pointer, MaybeAlign Alignment);
   Value& CreateScatter(Value & Val, Value & PointerVec, MaybeAlign Alignment);
-  Value& CreateGather(Value & PointerVec, MaybeAlign Alignment);
+  Value& CreateGather(Type *ReturnTy, Value & PointerVec, MaybeAlign Alignment);
 };
 
 
diff --git a/llvm/lib/IR/IRBuilder.cpp b/llvm/lib/IR/IRBuilder.cpp
@@ -563,14 +563,15 @@ CallInst *IRBuilderBase::CreateMaskedIntrinsic(Intrinsic::ID Id,
 /// \p FMFSource     - Copy source for Fast Math Flags
 /// \p Name          - name of the result variable
 Instruction *IRBuilderBase::CreateVectorPredicatedInst(unsigned OC,
+                                                       Type *ReturnTy,
                                                        ArrayRef<Value *> Params,
                                                        Instruction *FMFSource,
                                                        const Twine &Name) {
 
   Module *M = BB->getParent()->getParent();
 
   Intrinsic::ID VPID = VPIntrinsic::getForOpcode(OC);
-  auto VPFunc = VPIntrinsic::getDeclarationForParams(M, VPID, Params);
+  auto VPFunc = VPIntrinsic::getDeclarationForParams(M, VPID, ReturnTy, Params);
   auto *VPCall = createCallHelper(VPFunc, Params, this, Name);
 
   // transfer fast math flags
@@ -594,7 +595,9 @@ Instruction *IRBuilderBase::CreateVectorPredicatedCmp(
 
   Module *M = BB->getParent()->getParent();
 
-  // encode comparison predicate as MD
+  auto *ReturnTy = MaskParam->getType();
+
+  // Encode comparison predicate as MD
   uint8_t RawPred = static_cast<uint8_t>(Pred);
   auto Int8Ty = Type::getInt8Ty(getContext());
   auto PredParam = ConstantInt::get(Int8Ty, RawPred, false);
@@ -604,7 +607,8 @@ Instruction *IRBuilderBase::CreateVectorPredicatedCmp(
                            : Intrinsic::vp_fcmp;
 
   auto VPFunc = VPIntrinsic::getDeclarationForParams(
-      M, VPID, {FirstParam, SndParam, PredParam, MaskParam, VectorLengthParam});
+      M, VPID, ReturnTy,
+      {FirstParam, SndParam, PredParam, MaskParam, VectorLengthParam});
 
   return createCallHelper(
       VPFunc, {FirstParam, SndParam, PredParam, MaskParam, VectorLengthParam},
diff --git a/llvm/lib/IR/IntrinsicInst.cpp b/llvm/lib/IR/IntrinsicInst.cpp
@@ -541,6 +541,9 @@ using VPTypeTokenVec = SmallVector<VPTypeToken, 4>;
 static VPIntrinsic::ShortTypeVec getVPIntrinsicTypes(Intrinsic::ID ID,
                                                  Type *VecRetTy, Type *VecPtrTy,
                                                  Type *VectorTy) {
+  if (VPReductionIntrinsic::isVPReduction(ID))
+    return VPIntrinsic::ShortTypeVec{VectorTy};
+
   switch (ID) {
   default:
     llvm_unreachable("not implemented!");
@@ -616,9 +619,7 @@ static VPIntrinsic::ShortTypeVec getVPIntrinsicTypes(Intrinsic::ID ID,
 
   case Intrinsic::vp_gather:
   case Intrinsic::vp_load:
-    assert(VecRetTy);
     return VPIntrinsic::ShortTypeVec{VecRetTy, VecPtrTy};
-
   case Intrinsic::vp_scatter:
   case Intrinsic::vp_store:
     return VPIntrinsic::ShortTypeVec{VectorTy, VecPtrTy};
@@ -639,8 +640,8 @@ static VPIntrinsic::ShortTypeVec getVPIntrinsicTypes(Intrinsic::ID ID,
 }
 
 Function *VPIntrinsic::getDeclarationForParams(Module *M, Intrinsic::ID VPID,
-                                               ArrayRef<Value *> Params,
-                                               Type *VecRetTy) {
+                                               Type *VecRetTy,
+                                               ArrayRef<Value *> Params) {
   assert(VPID != Intrinsic::not_intrinsic && "todo dispatch to default insts");
 
   bool IsArithOp = VPIntrinsic::IsBinaryVPOp(VPID) ||
@@ -683,15 +684,6 @@ Function *VPIntrinsic::getDeclarationForParams(Module *M, Intrinsic::ID VPID,
     if (DataPosOpt.hasValue()) {
       // store-kind operation
       VecTy = Params[DataPosOpt.getValue()]->getType();
-    } else if (VPID == Intrinsic::vp_gather) {
-      // Construct data vector from vector of pointer to element.
-      auto *ElemTy =
-          cast<VectorType>(VecPtrTy)->getElementType()->getPointerElementType();
-      VecTy = VectorType::get(ElemTy,
-                                 cast<VectorType>(VecPtrTy)->getElementCount());
-    } else {
-      // load-kind operation
-      VecTy = VecPtrTy->getPointerElementType();
     }
 
   } else if (IsShuffleOp) {
diff --git a/llvm/lib/IR/PredicatedInst.cpp b/llvm/lib/IR/PredicatedInst.cpp
@@ -79,7 +79,7 @@ Instruction *PredicatedUnaryOperator::Create(
   // Fetch the VP intrinsic
   auto &VecTy = cast<VectorType>(*V->getType());
   auto *VPFunc =
-      VPIntrinsic::getDeclarationForParams(Mod, VPID, {V}, &VecTy);
+      VPIntrinsic::getDeclarationForParams(Mod, VPID, &VecTy, {V});
 
   // Encode default environment fp behavior
 
@@ -139,7 +139,7 @@ Instruction *PredicatedBinaryOperator::Create(
   // Fetch the VP intrinsic
   auto &VecTy = cast<VectorType>(*V1->getType());
   auto *VPFunc =
-      VPIntrinsic::getDeclarationForParams(Mod, VPID, {V1, V2}, &VecTy);
+      VPIntrinsic::getDeclarationForParams(Mod, VPID, &VecTy, {V1, V2});
 
   // Encode default environment fp behavior
 
diff --git a/llvm/lib/IR/VPBuilder.cpp b/llvm/lib/IR/VPBuilder.cpp
@@ -89,7 +89,7 @@ Value *VPBuilder::CreateVectorCopy(Instruction &Inst, ValArray VecOpArray) {
   Type *VecRetTy = ScaRetTy->isVoidTy() ? ScaRetTy : &getVectorType(*ScaRetTy);
   auto &M = *Builder.GetInsertBlock()->getParent()->getParent();
   auto VPDecl =
-      VPIntrinsic::getDeclarationForParams(&M, VPID, VecParams, VecRetTy);
+      VPIntrinsic::getDeclarationForParams(&M, VPID, VecRetTy, VecParams);
 
   // Prepare constraint fp params
   // FIXME: \p Inst could also be just another VP intrinsic.
@@ -150,16 +150,12 @@ Value &VPBuilder::CreateContiguousStore(Value &Val, Value &ElemPointer,
   return StoreCall;
 }
 
-Value &VPBuilder::CreateContiguousLoad(Value &ElemPointer,
+Value &VPBuilder::CreateContiguousLoad(Type *ReturnTy,
+                                       Value &ElemPointer,
                                        MaybeAlign AlignOpt) {
-  auto &PointerTy = cast<PointerType>(*ElemPointer.getType());
-  auto &VecTy = getVectorType(*PointerTy.getPointerElementType());
-  auto *VecPtrTy = VecTy.getPointerTo(PointerTy.getAddressSpace());
-  auto *VecPtr = Builder.CreatePointerCast(&ElemPointer, VecPtrTy);
-
-  auto *LoadFunc = Intrinsic::getDeclaration(&getModule(), Intrinsic::vp_load,
-                                             {&VecTy, VecPtrTy});
-  ShortValueVec Args{VecPtr, &RequestPred(), &RequestEVL()};
+  auto *LoadFunc = VPIntrinsic::getDeclarationForParams(
+      &getModule(), Intrinsic::vp_load, ReturnTy, {&ElemPointer});
+  ShortValueVec Args{&ElemPointer, &RequestPred(), &RequestEVL()};
   CallInst &LoadCall = *Builder.CreateCall(LoadFunc, Args);
   if (AlignOpt.hasValue()) {
     unsigned PtrPos =
@@ -173,7 +169,7 @@ Value &VPBuilder::CreateContiguousLoad(Value &ElemPointer,
 Value &VPBuilder::CreateScatter(Value &Val, Value &PointerVec,
                                 MaybeAlign AlignOpt) {
   auto *ScatterFunc =
-      Intrinsic::getDeclaration(&getModule(), Intrinsic::vp_scatter,
+      Intrinsic::getDeclaration(&getModule(), Intrinsic::vp_scatter, 
                                 {Val.getType(), PointerVec.getType()});
   ShortValueVec Args{&Val, &PointerVec, &RequestPred(), &RequestEVL()};
   CallInst &ScatterCall = *Builder.CreateCall(ScatterFunc, Args);
@@ -189,13 +185,9 @@ Value &VPBuilder::CreateScatter(Value &Val, Value &PointerVec,
   return ScatterCall;
 }
 
-Value &VPBuilder::CreateGather(Value &PointerVec, MaybeAlign AlignOpt) {
-  auto &PointerVecTy = cast<VectorType>(*PointerVec.getType());
-  auto &ElemTy = *cast<PointerType>(*PointerVecTy.getElementType())
-                      .getPointerElementType();
-  auto &VecTy = *VectorType::get(&ElemTy, PointerVecTy.getElementCount());
-  auto *GatherFunc = Intrinsic::getDeclaration(
-      &getModule(), Intrinsic::vp_gather, {&VecTy, &PointerVecTy});
+Value &VPBuilder::CreateGather(Type *RetTy, Value &PointerVec, MaybeAlign AlignOpt) {
+  auto *GatherFunc = VPIntrinsic::getDeclarationForParams(
+      &getModule(), Intrinsic::vp_gather, RetTy, {&PointerVec});
 
   ShortValueVec Args{&PointerVec, &RequestPred(), &RequestEVL()};
   CallInst &GatherCall = *Builder.CreateCall(GatherFunc, Args);
@@ -213,7 +205,7 @@ Value &VPBuilder::CreateGather(Value &PointerVec, MaybeAlign AlignOpt) {
 
 Value *VPBuilder::CreateVectorShift(Value *SrcVal, Value *Amount, Twine Name) {
   auto D = VPIntrinsic::getDeclarationForParams(
-      &getModule(), Intrinsic::vp_vshift, {SrcVal, Amount});
+      &getModule(), Intrinsic::vp_vshift, SrcVal->getType(), {SrcVal, Amount});
   return Builder.CreateCall(D, {SrcVal, Amount, &RequestPred(), &RequestEVL()},
                             Name);
 }
diff --git a/llvm/unittests/IR/VPIntrinsicTest.cpp b/llvm/unittests/IR/VPIntrinsicTest.cpp
@@ -336,7 +336,7 @@ TEST_F(VPIntrinsicTest, VPIntrinsicDeclarationForParams) {
 
     ASSERT_NE(F.getIntrinsicID(), Intrinsic::not_intrinsic);
     auto *NewDecl = VPIntrinsic::getDeclarationForParams(
-        OutM.get(), F.getIntrinsicID(), Values, VecRetTy);
+        OutM.get(), F.getIntrinsicID(), FuncTy->getReturnType(), Values);
     ASSERT_TRUE(NewDecl);
 
     // Check that 'old decl' == 'new decl'.
