[2/11][POC][Clang][RISCV] Define RVV tuple types

[eopXD]

Summary

For the cover letter of this patch-set, please checkout D146872.

Depends on D146872.

This is the 2nd patch of the patch-set. This patch originates from
D97264. This patch further allows local variable declaration and
function parameter passing by adjustment in clang lowering.

Test cases are provided to demonstrate the LLVM IR generated.

Note: This patch is currently only a proof-of-concept with only a
single RVV tuple type declared here, the rest will be added when
the concept of this patch-set is accepted.

Authored-by: eop Chen <[e-mail]>
Co-Authored-by: Hsiangkai Wang <[e-mail]>

[eopXD]

Update code.

[eopXD]

Rebase upon change of parent patch.

[craig.topper]

clang/lib/CodeGen/CodeGenTypes.cpp

632

Why did this code need to move into the macro body? It was written the way it was before to avoid code duplication.

641

Can you unify tuple and non-tuple by checking Info.NumVectors != 1?

645

llvm::SmallVector<llvm::Type *, 4> EltTys(Info.NumVectors, EltTy)

[eopXD]

Address comments from Craig.

[craig.topper]

clang/lib/CodeGen/CGCall.cpp

3141

The Dstty, SrcSize and DstSize variables are unused in released builds.

3148

What are we loading here?

Is there a test for this code?

[craig.topper]

clang/include/clang/Basic/RISCVVTypes.def

149

Is this macro still needed?

[eopXD]

Address comments from Craig.

• Remove unused variables that only associate with assertions.
• Remove unused macro

clang/lib/CodeGen/CGCall.cpp

3148

You are right, should be a poison here.

Test case clang/test/CodeGen/RISCV/rvv-intrinsics-handcrafted/rvv-tuple-type-0.c and clang/test/CodeGen/RISCV/rvv-intrinsics-handcrafted/rvv-tuple-type-1.c covers the code here.

[craig.topper]

clang/lib/CodeGen/CGCall.cpp

3148

Thanks. I had tried to scan the tests to see if I could find the extra load, but I guess I missed it.

[eopXD]

clang/lib/CodeGen/CGCall.cpp

3148

Marking this as done.

[eopXD]

Bump CI.

[craig.topper]

clang/lib/CodeGen/CGCall.cpp

3135–3152

Can we call CGM.getDataLayout().getTypeAllocSize(STy) and check if the TypeSize returned isScalable instead of calling containsScalableVectorType?

clang/test/CodeGen/RISCV/rvv-intrinsics-handcrafted/rvv-tuple-type-1.c

1 ↗

(On Diff #523323)

Do these two tests have the same content but different RUN lines? Can we merge them with 2 different RUN lines?

[eopXD]

Address comments from Craig.

clang/lib/CodeGen/CGCall.cpp

3135–3152

Switched to using CGM.getDataLayout().getTypeAllocSize(STy), though calling containsScalableVectorType won't cost too much because the function has memoization.

[craig.topper]

clang/lib/CodeGen/CGCall.cpp

3135–3152

Can cache GM.getDataLayout().getTypeAllocSize(STy) in a variable? We call it here and inside both the if and the else body.

[eopXD]

Address comment from Craig.

[craig.topper]

clang/lib/CodeGen/CGCall.cpp

3156

This can use StructSize.getFixedValue()

[eopXD]

Address comment from Craig.

[craig.topper]

LGTM

[eopXD]

Rebase to latest main before landing this.

[eopXD]

Committing this patch caused build failure in non-RISC-V buildbot CI-s because of the missing REQUIRES line in the test case clang/test/CodeGen/RISCV/rvv-intrinsics-handcrafted/rvv-tuple-type.c.

This is addressed in the 5a61920ed8b3e76d8f9bf39f2c1e18d552fcc976.

clang/include/clang/Basic/RISCVVTypes.def

 RVV_PREDICATE_TYPE("__rvv_bool1_t",  RvvBool1,  RvvBool1Ty,  64)
 RVV_PREDICATE_TYPE("__rvv_bool2_t",  RvvBool2,  RvvBool2Ty,  32)
 RVV_PREDICATE_TYPE("__rvv_bool4_t",  RvvBool4,  RvvBool4Ty,  16)
 RVV_PREDICATE_TYPE("__rvv_bool8_t",  RvvBool8,  RvvBool8Ty,  8)
 RVV_PREDICATE_TYPE("__rvv_bool16_t", RvvBool16, RvvBool16Ty, 4)
 RVV_PREDICATE_TYPE("__rvv_bool32_t", RvvBool32, RvvBool32Ty, 2)
 RVV_PREDICATE_TYPE("__rvv_bool64_t", RvvBool64, RvvBool64Ty, 1)
 
+//===- Tuple vector types -------------------------------------------------===//
+
+RVV_VECTOR_TYPE_INT("__rvv_int32m1x2_t", RvvInt32m1x2, RvvInt32m1x2Ty, 2,  32, 2, true)
+
 #undef RVV_VECTOR_TYPE_FLOAT
 #undef RVV_VECTOR_TYPE_INT
 #undef RVV_VECTOR_TYPE
 #undef RVV_PREDICATE_TYPE
 #undef RVV_TYPE

clang/lib/CodeGen/CGCall.cpp

case ABIArgInfo::Direct: {
       // Pointer to store into.
       Address Ptr = emitAddressAtOffset(*this, Alloca, ArgI);
 
       // Fast-isel and the optimizer generally like scalar values better than
       // FCAs, so we flatten them if this is safe to do for this argument.
       llvm::StructType *STy = dyn_cast<llvm::StructType>(ArgI.getCoerceToType());
       if (ArgI.isDirect() && ArgI.getCanBeFlattened() && STy &&
           STy->getNumElements() > 1) {
+        llvm::TypeSize StructSize = CGM.getDataLayout().getTypeAllocSize(STy);
+        if (StructSize.isScalable()) {
+          assert(STy->containsHomogeneousScalableVectorTypes() &&
+                 "ABI only supports structure with homogeneous scalable vector "
+                 "type");
+          assert(StructSize == CGM.getDataLayout().getTypeAllocSize(
+                                   Ptr.getElementType()) &&
+                 "Only allow non-fractional movement of structure with"
+                 "homogeneous scalable vector type");
+          assert(STy->getNumElements() == NumIRArgs);
+
+          llvm::Value *LoadedStructValue = llvm::PoisonValue::get(STy);
+          for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+            auto *AI = Fn->getArg(FirstIRArg + i);
+            AI->setName(Arg->getName() + ".coerce" + Twine(i));
+            LoadedStructValue =
+                Builder.CreateInsertValue(LoadedStructValue, AI, i);
+          }
+
+          Builder.CreateStore(LoadedStructValue, Ptr);
+        } else {
           uint64_t SrcSize = CGM.getDataLayout().getTypeAllocSize(STy);
           llvm::Type *DstTy = Ptr.getElementType();
           uint64_t DstSize = CGM.getDataLayout().getTypeAllocSize(DstTy);
 
           Address AddrToStoreInto = Address::invalid();
           if (SrcSize <= DstSize) {
             AddrToStoreInto = Builder.CreateElementBitCast(Ptr, STy);
           } else {
             AddrToStoreInto =
                 CreateTempAlloca(STy, Alloca.getAlignment(), "coerce");
           }
 
           assert(STy->getNumElements() == NumIRArgs);
           for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
             auto AI = Fn->getArg(FirstIRArg + i);
             AI->setName(Arg->getName() + ".coerce" + Twine(i));
             Address EltPtr = Builder.CreateStructGEP(AddrToStoreInto, i);
             Builder.CreateStore(AI, EltPtr);
           }
 
           if (SrcSize > DstSize) {
             Builder.CreateMemCpy(Ptr, AddrToStoreInto, DstSize);
           }
-
+        }
       } else {
         // Simple case, just do a coerced store of the argument into the alloca.
         assert(NumIRArgs == 1);
         auto AI = Fn->getArg(FirstIRArg);
         AI->setName(Arg->getName() + ".coerce");
         CreateCoercedStore(AI, Ptr, /*DstIsVolatile=*/false, *this);
       }
 

clang/lib/CodeGen/CodeGenTypes.cpp

case BuiltinType::SveCount:
       return llvm::TargetExtType::get(getLLVMContext(), "aarch64.svcount");
 #define PPC_VECTOR_TYPE(Name, Id, Size) \
     case BuiltinType::Id: \
       ResultType = \
         llvm::FixedVectorType::get(ConvertType(Context.BoolTy), Size); \
       break;
 #include "clang/Basic/PPCTypes.def"
 #define RVV_TYPE(Name, Id, SingletonId) case BuiltinType::Id:
 #include "clang/Basic/RISCVVTypes.def"
       {
         ASTContext::BuiltinVectorTypeInfo Info =
             Context.getBuiltinVectorTypeInfo(cast<BuiltinType>(Ty));
+        // Tuple types are expressed as aggregregate types of the same scalable
+        // vector type (e.g. vint32m1x2_t is two vint32m1_t, which is {<vscale x
+        // 2 x i32>, <vscale x 2 x i32>}).
+        if (Info.NumVectors != 1) {
+          llvm::Type *EltTy = llvm::ScalableVectorType::get(
+              ConvertType(Info.ElementType), Info.EC.getKnownMinValue());
+          llvm::SmallVector<llvm::Type *, 4> EltTys(Info.NumVectors, EltTy);
+          return llvm::StructType::get(getLLVMContext(), EltTys);
+        }
         return llvm::ScalableVectorType::get(ConvertType(Info.ElementType),
                                              Info.EC.getKnownMinValue() *
                                                  Info.NumVectors);
       }
 #define WASM_REF_TYPE(Name, MangledName, Id, SingletonId, AS)                  \
   case BuiltinType::Id: {                                                      \
     if (BuiltinType::Id == BuiltinType::WasmExternRef)                         \
       ResultType = CGM.getTargetCodeGenInfo().getWasmExternrefReferenceType(); \
     else                                                                       \
       llvm_unreachable("Unexpected wasm reference builtin type!");             \
   } break;
 #include "clang/Basic/WebAssemblyReferenceTypes.def"

clang/test/CodeGen/RISCV/rvv-intrinsics-handcrafted/rvv-tuple-type.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
+// RUN: %clang_cc1 -triple riscv64 -target-feature +zve32x -O0 \
+// RUN:   -emit-llvm %s -o - | FileCheck %s --check-prefix=O0
+// RUN: %clang_cc1 -triple riscv64 -target-feature +zve32x -disable-O0-optnone \
+// RUN:   -emit-llvm %s -o - | opt -S -passes=mem2reg | FileCheck %s --check-prefix=AFTER_MEM2REG
+
+
+#include <riscv_vector.h>
+
+// Declare local variable
+// O0-LABEL: define dso_local void @foo
+// O0-SAME: () #[[ATTR0:[0-9]+]] {
+// O0-NEXT:  entry:
+// O0-NEXT:    [[V_TUPLE:%.*]] = alloca { <vscale x 2 x i32>, <vscale x 2 x i32> }, align 4
+// O0-NEXT:    ret void
+//
+// AFTER_MEM2REG-LABEL: define dso_local void @foo
+// AFTER_MEM2REG-SAME: () #[[ATTR0:[0-9]+]] {
+// AFTER_MEM2REG-NEXT:  entry:
+// AFTER_MEM2REG-NEXT:    ret void
+//
+void foo() {
+  __rvv_int32m1x2_t v_tuple;
+}
+
+// Declare local variable and return
+// O0-LABEL: define dso_local { <vscale x 2 x i32>, <vscale x 2 x i32> } @bar
+// O0-SAME: () #[[ATTR0]] {
+// O0-NEXT:  entry:
+// O0-NEXT:    [[V_TUPLE:%.*]] = alloca { <vscale x 2 x i32>, <vscale x 2 x i32> }, align 4
+// O0-NEXT:    [[TMP0:%.*]] = load { <vscale x 2 x i32>, <vscale x 2 x i32> }, ptr [[V_TUPLE]], align 4
+// O0-NEXT:    ret { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP0]]
+//
+// AFTER_MEM2REG-LABEL: define dso_local { <vscale x 2 x i32>, <vscale x 2 x i32> } @bar
+// AFTER_MEM2REG-SAME: () #[[ATTR0]] {
+// AFTER_MEM2REG-NEXT:  entry:
+// AFTER_MEM2REG-NEXT:    ret { <vscale x 2 x i32>, <vscale x 2 x i32> } undef
+//
+__rvv_int32m1x2_t bar() {
+  __rvv_int32m1x2_t v_tuple;
+  return v_tuple;
+}
+
+// Pass as function parameter
+// O0-LABEL: define dso_local void @baz
+// O0-SAME: (<vscale x 2 x i32> [[V_TUPLE_COERCE0:%.*]], <vscale x 2 x i32> [[V_TUPLE_COERCE1:%.*]]) #[[ATTR0]] {
+// O0-NEXT:  entry:
+// O0-NEXT:    [[V_TUPLE:%.*]] = alloca { <vscale x 2 x i32>, <vscale x 2 x i32> }, align 4
+// O0-NEXT:    [[V_TUPLE_ADDR:%.*]] = alloca { <vscale x 2 x i32>, <vscale x 2 x i32> }, align 4
+// O0-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } poison, <vscale x 2 x i32> [[V_TUPLE_COERCE0]], 0
+// O0-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP0]], <vscale x 2 x i32> [[V_TUPLE_COERCE1]], 1
+// O0-NEXT:    store { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP1]], ptr [[V_TUPLE]], align 4
+// O0-NEXT:    [[V_TUPLE1:%.*]] = load { <vscale x 2 x i32>, <vscale x 2 x i32> }, ptr [[V_TUPLE]], align 4
+// O0-NEXT:    store { <vscale x 2 x i32>, <vscale x 2 x i32> } [[V_TUPLE1]], ptr [[V_TUPLE_ADDR]], align 4
+// O0-NEXT:    ret void
+//
+// AFTER_MEM2REG-LABEL: define dso_local void @baz
+// AFTER_MEM2REG-SAME: (<vscale x 2 x i32> [[V_TUPLE_COERCE0:%.*]], <vscale x 2 x i32> [[V_TUPLE_COERCE1:%.*]]) #[[ATTR0]] {
+// AFTER_MEM2REG-NEXT:  entry:
+// AFTER_MEM2REG-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } poison, <vscale x 2 x i32> [[V_TUPLE_COERCE0]], 0
+// AFTER_MEM2REG-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP0]], <vscale x 2 x i32> [[V_TUPLE_COERCE1]], 1
+// AFTER_MEM2REG-NEXT:    ret void
+//
+void baz(__rvv_int32m1x2_t v_tuple) {
+}
+
+// Pass as function parameter and return
+// O0-LABEL: define dso_local { <vscale x 2 x i32>, <vscale x 2 x i32> } @qux
+// O0-SAME: (<vscale x 2 x i32> [[V_TUPLE_COERCE0:%.*]], <vscale x 2 x i32> [[V_TUPLE_COERCE1:%.*]]) #[[ATTR0]] {
+// O0-NEXT:  entry:
+// O0-NEXT:    [[V_TUPLE:%.*]] = alloca { <vscale x 2 x i32>, <vscale x 2 x i32> }, align 4
+// O0-NEXT:    [[V_TUPLE_ADDR:%.*]] = alloca { <vscale x 2 x i32>, <vscale x 2 x i32> }, align 4
+// O0-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } poison, <vscale x 2 x i32> [[V_TUPLE_COERCE0]], 0
+// O0-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP0]], <vscale x 2 x i32> [[V_TUPLE_COERCE1]], 1
+// O0-NEXT:    store { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP1]], ptr [[V_TUPLE]], align 4
+// O0-NEXT:    [[V_TUPLE1:%.*]] = load { <vscale x 2 x i32>, <vscale x 2 x i32> }, ptr [[V_TUPLE]], align 4
+// O0-NEXT:    store { <vscale x 2 x i32>, <vscale x 2 x i32> } [[V_TUPLE1]], ptr [[V_TUPLE_ADDR]], align 4
+// O0-NEXT:    [[TMP2:%.*]] = load { <vscale x 2 x i32>, <vscale x 2 x i32> }, ptr [[V_TUPLE_ADDR]], align 4
+// O0-NEXT:    ret { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP2]]
+//
+// AFTER_MEM2REG-LABEL: define dso_local { <vscale x 2 x i32>, <vscale x 2 x i32> } @qux
+// AFTER_MEM2REG-SAME: (<vscale x 2 x i32> [[V_TUPLE_COERCE0:%.*]], <vscale x 2 x i32> [[V_TUPLE_COERCE1:%.*]]) #[[ATTR0]] {
+// AFTER_MEM2REG-NEXT:  entry:
+// AFTER_MEM2REG-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } poison, <vscale x 2 x i32> [[V_TUPLE_COERCE0]], 0
+// AFTER_MEM2REG-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP0]], <vscale x 2 x i32> [[V_TUPLE_COERCE1]], 1
+// AFTER_MEM2REG-NEXT:    ret { <vscale x 2 x i32>, <vscale x 2 x i32> } [[TMP1]]
+//
+__rvv_int32m1x2_t qux(__rvv_int32m1x2_t v_tuple) {
+  return v_tuple;
+}

clang/test/Sema/riscv-types.c

void bar(void) {
   // CHECK: __rvv_int8mf8_t x41;
   __rvv_int8mf8_t x41;
 
   // CHECK: __rvv_int8mf4_t x42;
   __rvv_int8mf4_t x42;
 
   // CHECK: __rvv_int8mf2_t x43;
   __rvv_int8mf2_t x43;
+
+  // CHECK: __rvv_int32m1x2_t x44;
+  __rvv_int32m1x2_t x44;
 }
 
 typedef __rvv_bool4_t vbool4_t;
 __rvv_bool4_t get_rvv_bool4(void);
 vbool4_t get_vbool4_t(void);
 
 void func1(int sel) {
   // CHECK: vbool4_t t0 = sel ? get_rvv_bool4() : get_vbool4_t();
   vbool4_t t0 = sel ? get_rvv_bool4() : get_vbool4_t();
 }