[PowerPC][MMA] Allow MMA builtin types in pre-P10 compilation units

[kamaub]

Summary

This patch allows the use of vector_quad and vector_pair, PPC MMA builtin
types, on all PowerPC 64-bit compilation units. When these types are
made available the builtins that use them automatically become available
so semantic checking for mma and pair vector memop __builtins is also
expanded to ensure these builtin function call are only allowed on
Power10 and new architectures. All related test cases are updated to
ensure test coverage.

[kamaub]

The condition for failing semantic chequing on mma builtins was incorrect,
updating this patch to correctly check the semantics and associated testing.

[NeHuang]

clang/test/Sema/ppc-mma-builtins.c

1 ↗

(On Diff #371954)

can you please add // REQUIRES: powerpc-registered-target for the ppc specific test?

[kamaub]

Rebasing this patch and addressing review comments

An upstream commit removed the need for this patch to modify a few
test cases so it has been rebased to remain up-to-date, also addressing
review comments.

[nemanjai]

LGTM as long as you add the back end test.

clang/test/CodeGen/ppc-mma-types.c

2–6

Please add this IR test to the back end codegen tests (for older CPUs).

[kamaub]

Updated an existing test for the backend with pwr8 and pwr9 be/le targets during commit

clang/include/clang/Sema/Sema.h

private:
   bool SemaBuiltinConstantArgShiftedByte(CallExpr *TheCall, int ArgNum,
                                          unsigned ArgBits);
   bool SemaBuiltinConstantArgShiftedByteOrXXFF(CallExpr *TheCall, int ArgNum,
                                                unsigned ArgBits);
   bool SemaBuiltinARMSpecialReg(unsigned BuiltinID, CallExpr *TheCall,
                                 int ArgNum, unsigned ExpectedFieldNum,
                                 bool AllowName);
   bool SemaBuiltinARMMemoryTaggingCall(unsigned BuiltinID, CallExpr *TheCall);
-  bool SemaBuiltinPPCMMACall(CallExpr *TheCall, const char *TypeDesc);
+  bool SemaBuiltinPPCMMACall(CallExpr *TheCall, unsigned BuiltinID,
+                             const char *TypeDesc);
 
   bool CheckPPCMMAType(QualType Type, SourceLocation TypeLoc);
 
   // Matrix builtin handling.
   ExprResult SemaBuiltinMatrixTranspose(CallExpr *TheCall,
                                         ExprResult CallResult);
   ExprResult SemaBuiltinMatrixColumnMajorLoad(CallExpr *TheCall,
                                               ExprResult CallResult);

clang/lib/AST/ASTContext.cpp

#include "clang/Basic/OpenCLExtensionTypes.def"
   }
 
   if (Target.hasAArch64SVETypes()) {
 #define SVE_TYPE(Name, Id, SingletonId) \
     InitBuiltinType(SingletonId, BuiltinType::Id);
 #include "clang/Basic/AArch64SVEACLETypes.def"
   }
 
-  if (Target.getTriple().isPPC64() &&
-      Target.hasFeature("paired-vector-memops")) {
-    if (Target.hasFeature("mma")) {
+  if (Target.getTriple().isPPC64()) {
 #define PPC_VECTOR_MMA_TYPE(Name, Id, Size) \
       InitBuiltinType(Id##Ty, BuiltinType::Id);
 #include "clang/Basic/PPCTypes.def"
-    }
 #define PPC_VECTOR_VSX_TYPE(Name, Id, Size) \
     InitBuiltinType(Id##Ty, BuiltinType::Id);
 #include "clang/Basic/PPCTypes.def"
   }
 
   if (Target.hasRISCVVTypes()) {
 #define RVV_TYPE(Name, Id, SingletonId)                                        \
   InitBuiltinType(SingletonId, BuiltinType::Id);

clang/lib/Sema/Sema.cpp

#include "clang/Basic/OpenCLExtensionTypes.def"
   }
 
   if (Context.getTargetInfo().hasAArch64SVETypes()) {
 #define SVE_TYPE(Name, Id, SingletonId) \
     addImplicitTypedef(Name, Context.SingletonId);
 #include "clang/Basic/AArch64SVEACLETypes.def"
   }
 
-  if (Context.getTargetInfo().getTriple().isPPC64() &&
-      Context.getTargetInfo().hasFeature("paired-vector-memops")) {
-    if (Context.getTargetInfo().hasFeature("mma")) {
+  if (Context.getTargetInfo().getTriple().isPPC64()) {
 #define PPC_VECTOR_MMA_TYPE(Name, Id, Size) \
       addImplicitTypedef(#Name, Context.Id##Ty);
 #include "clang/Basic/PPCTypes.def"
-    }
 #define PPC_VECTOR_VSX_TYPE(Name, Id, Size) \
     addImplicitTypedef(#Name, Context.Id##Ty);
 #include "clang/Basic/PPCTypes.def"
   }
 
   if (Context.getTargetInfo().hasRISCVVTypes()) {
 #define RVV_TYPE(Name, Id, SingletonId)                                        \
   addImplicitTypedef(Name, Context.SingletonId);

clang/lib/Sema/SemaChecking.cpp

 //===- SemaChecking.cpp - Extra Semantic Checking -------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 //  This file implements extra semantic analysis beyond what is enforced
bool Sema::CheckPPCBuiltinFunctionCall(const TargetInfo &TI, unsigned BuiltinID,
   case PPC::BI__builtin_ppc_lbarx:
     return SemaFeatureCheck(*this, TheCall, "isa-v207-instructions",
                             diag::err_ppc_builtin_only_on_arch, "8");
   case PPC::BI__builtin_vsx_ldrmb:
   case PPC::BI__builtin_vsx_strmb:
     return SemaFeatureCheck(*this, TheCall, "isa-v207-instructions",
                             diag::err_ppc_builtin_only_on_arch, "8") ||
            SemaBuiltinConstantArgRange(TheCall, 1, 1, 16);
 #define CUSTOM_BUILTIN(Name, Intr, Types, Acc) \
   case PPC::BI__builtin_##Name: \
-    return SemaBuiltinPPCMMACall(TheCall, Types);
+    return SemaBuiltinPPCMMACall(TheCall, BuiltinID, Types);
 #include "clang/Basic/BuiltinsPPC.def"
   }
   return SemaBuiltinConstantArgRange(TheCall, i, l, u);
 }
 
 // Check if the given type is a non-pointer PPC MMA type. This function is used
 // in Sema to prevent invalid uses of restricted PPC MMA types.
 bool Sema::CheckPPCMMAType(QualType Type, SourceLocation TypeLoc) {
bool Sema::SemaBuiltinARMSpecialReg(unsigned BuiltinID, CallExpr *TheCall,
 
   return false;
 }
 
 /// SemaBuiltinPPCMMACall - Check the call to a PPC MMA builtin for validity.
 /// Emit an error and return true on failure; return false on success.
 /// TypeStr is a string containing the type descriptor of the value returned by
 /// the builtin and the descriptors of the expected type of the arguments.
-bool Sema::SemaBuiltinPPCMMACall(CallExpr *TheCall, const char *TypeStr) {
+bool Sema::SemaBuiltinPPCMMACall(CallExpr *TheCall, unsigned BuiltinID,
+                                 const char *TypeStr) {
 
   assert((TypeStr[0] != '\0') &&
          "Invalid types in PPC MMA builtin declaration");
 
+  switch(BuiltinID) {
+  default:
+    // This function is called CheckPPCBuiltinFunctionCall where the BuiltinID
+    // is guaranteed to be an MMA or pair vector memop builtin, here we are
+    // isolating the pair vector memop builtins that can be used with mma off
+    // so the default case is every builtin that requires mma and paired vector
+    // memops.
+    if (SemaFeatureCheck(*this, TheCall, "paired-vector-memops",
+                         diag::err_ppc_builtin_only_on_arch, "10") &&
+        SemaFeatureCheck(*this, TheCall, "mma",
+                         diag::err_ppc_builtin_only_on_arch, "10"))
+      return true;
+    break;
+  case PPC::BI__builtin_vsx_lxvp:
+  case PPC::BI__builtin_vsx_stxvp:
+  case PPC::BI__builtin_vsx_assemble_pair:
+  case PPC::BI__builtin_vsx_disassemble_pair:
+    if (SemaFeatureCheck(*this, TheCall, "paired-vector-memops",
+                            diag::err_ppc_builtin_only_on_arch, "10"))
+      return true;
+    break;
+  }
+
   unsigned Mask = 0;
   unsigned ArgNum = 0;
 
   // The first type in TypeStr is the type of the value returned by the
   // builtin. So we first read that type and change the type of TheCall.
   QualType type = DecodePPCMMATypeFromStr(Context, TypeStr, Mask);
   TheCall->setType(type);
 

clang/test/AST/ast-dump-ppc-types.c

-// RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -target-cpu future \
+// RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -target-cpu pwr10 \
 // RUN:   -ast-dump -ast-dump-filter __vector %s | FileCheck %s
-// RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -target-cpu future \
-// RUN:   -target-feature -mma -ast-dump %s | FileCheck %s \
-// RUN:   --check-prefix=CHECK-NO-MMA
-// RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -target-cpu future \
-// RUN:   -target-feature -paired-vector-memops -ast-dump %s | FileCheck %s \
-// RUN:   --check-prefix=CHECK-NO-PAIRED
 // RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -target-cpu pwr9 \
-// RUN:   -ast-dump %s | FileCheck %s --check-prefix=CHECK-PWR9
+// RUN:   -ast-dump -ast-dump-filter __vector %s | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -target-cpu pwr8 \
+// RUN:   -ast-dump -ast-dump-filter __vector %s | FileCheck %s
 // RUN: %clang_cc1 -triple x86_64-unknown-unknown -ast-dump %s | FileCheck %s \
 // RUN:   --check-prefix=CHECK-X86_64
 // RUN: %clang_cc1 -triple arm-unknown-unknown -ast-dump %s | FileCheck %s \
 // RUN:   --check-prefix=CHECK-ARM
 // RUN: %clang_cc1 -triple riscv64-unknown-unknown -ast-dump %s | FileCheck %s \
 // RUN:   --check-prefix=CHECK-RISCV64
 
 // This test case checks that the PowerPC __vector_pair and __vector_quad types
 // are correctly defined. We also added checks on a couple of other targets to
 // ensure the types are target-dependent.
 
 // CHECK: TypedefDecl {{.*}} implicit __vector_quad '__vector_quad'
 // CHECK-NEXT: -BuiltinType {{.*}} '__vector_quad'
 // CHECK: TypedefDecl {{.*}} implicit __vector_pair '__vector_pair'
 // CHECK-NEXT: -BuiltinType {{.*}} '__vector_pair'
 
-// CHECK-NO-MMA-NOT: __vector_quad
-// CHECK-NO-MMA: __vector_pair
-
-// CHECK-NO-PAIRED-NOT: __vector_quad
-// CHECK-NO-PAIRED-NOT: __vector_pair
-
-// CHECK-PWR9-NOT: __vector_quad
-// CHECK-PWR9-NOT: __vector_pair
-
 // CHECK-X86_64-NOT: __vector_quad
 // CHECK-X86_64-NOT: __vector_pair
 
 // CHECK-ARM-NOT: __vector_quad
 // CHECK-ARM-NOT: __vector_pair
 
 // CHECK-RISCV64-NOT: __vector_quad
 // CHECK-RISCV64-NOT: __vector_pair

clang/test/CodeGen/builtins-ppc-pair-mma.c

 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
-// RUN: %clang_cc1 -O3 -triple powerpc64le-unknown-unknown -target-cpu future -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -O3 -triple powerpc64le-unknown-unknown -target-cpu pwr10 -emit-llvm %s -o - | FileCheck %s
 
 // CHECK-LABEL: @test1(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8> [[VC:%.*]], <16 x i8> [[VC]], <16 x i8> [[VC]], <16 x i8> [[VC]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i8* [[RESP:%.*]] to <512 x i1>*
 // CHECK-NEXT:    store <512 x i1> [[TMP0]], <512 x i1>* [[TMP1]], align 64, !tbaa [[TBAA2:![0-9]+]]
 // CHECK-NEXT:    ret void
 //

clang/test/CodeGen/ppc-mma-types.c

 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
-// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu future \
+// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu pwr10 \
+// RUN:   -emit-llvm -O3 -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu pwr9 \
+// RUN:   -emit-llvm -O3 -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu pwr8 \
 // RUN:   -emit-llvm -O3 -o - %s | FileCheck %s
 
 // CHECK-LABEL: @test1(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds <512 x i1>, <512 x i1>* [[PTR1:%.*]], i64 2
 // CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, <512 x i1>* [[ADD_PTR]], align 64, [[TBAA2:!tbaa !.*]]
 // CHECK-NEXT:    [[ADD_PTR1:%.*]] = getelementptr inbounds <512 x i1>, <512 x i1>* [[PTR2:%.*]], i64 1
 // CHECK-NEXT:    store <512 x i1> [[TMP0]], <512 x i1>* [[ADD_PTR1]], align 64, [[TBAA2]]

clang/test/CodeGenCXX/ppc-mangle-mma-types.cpp

-// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu future %s \
+// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu pwr10 %s \
+// RUN:   -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu pwr9 %s \
+// RUN:   -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64le-linux-unknown -target-cpu pwr8 %s \
 // RUN:   -emit-llvm -o - | FileCheck %s
 
 // CHECK: _Z2f1Pu13__vector_quad
 void f1(__vector_quad *vq) {}
 
 // CHECK: _Z2f2Pu13__vector_pair
 void f2(__vector_pair *vp) {}
 
 // CHECK: _Z2f3Pu13__vector_quad
 void f3(__vector_quad *vq) {}
 
 // CHECK: _Z2f3Pu13__vector_pair
 void f3(__vector_pair *vp) {}

clang/test/Sema/ppc-pair-mma-types.c

 // RUN: %clang_cc1 -triple powerpc64le-unknown-unknown -fsyntax-only \
-// RUN:   -target-cpu future %s -verify
+// RUN:   -target-cpu pwr10 %s -verify
 
 // The use of PPC MMA types is strongly restricted. Non-pointer MMA variables
 // can only be declared in functions and a limited number of operations are
 // supported on these types. This test case checks that invalid uses of MMA
 // types are correctly prevented.
 
 // vector quad