[clang] add sqrt{pd|ps}512

Signed-off-by: Shreeyash Pandey <[email protected]>

Author: bojle

clang/include/clang/Basic/BuiltinsX86.td

@@ -950,7 +950,7 @@ let Features = "pku", Attributes = [NoThrow] in {
   def wrpkru : X86Builtin<"void(unsigned int)">;
 }
 
-let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in {
   def sqrtpd512 : X86Builtin<"_Vector<8, double>(_Vector<8, double>, _Constant int)">;
   def sqrtps512 : X86Builtin<"_Vector<16, float>(_Vector<16, float>, _Constant int)">;
 }

clang/lib/AST/ByteCode/InterpBuiltin.cpp

@@ -3016,7 +3016,12 @@ static bool interp__builtin_x86_sqrt(InterpState &S, CodePtr OpPC,
                                                  unsigned ID) {
   llvm::errs() << "Entering x86 sqrtpd/ps interpretbuiltin\n";
 
-  assert(Call->getNumArgs() == 1);
+  llvm::errs() << "BI__builtin_ia32_sqrtpd512 " << X86::BI__builtin_ia32_sqrtpd512 << '\n';
+  llvm::errs() << "BI__builtin_ia32_sqrtps512 " << X86::BI__builtin_ia32_sqrtps512 << '\n';
+  llvm::errs() << "Current ID " << ID << '\n';
+  llvm::errs() << "GetNumArgs " << Call->getNumArgs() << '\n';
+  unsigned NumArgs = Call->getNumArgs();
+  assert(NumArgs == 1 || NumArgs == 2);
   const Expr *ArgExpr = Call->getArg(0);
   QualType ArgTy = ArgExpr->getType();
   QualType ResultTy = Call->getType();
@@ -3033,6 +3038,16 @@ static bool interp__builtin_x86_sqrt(InterpState &S, CodePtr OpPC,
     SemanticsPtr = &S.getContext().getFloatSemantics(ArgTy);
   const llvm::fltSemantics &Semantics = *SemanticsPtr;
 
+  if (NumArgs == 2) {
+    if (!Call->getArg(1)->getType()->isIntegerType()) {
+      return false;
+    }
+    APSInt RoundingMode = popToAPSInt(S, Call->getArg(1));
+    if (RoundingMode.getZExtValue() != 4) {
+      return false;
+    }
+  }
+
 
   // Scalar case
   if (!ArgTy->isVectorType()) {
@@ -3831,6 +3846,8 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
   case X86::BI__builtin_ia32_sqrtps:
   case X86::BI__builtin_ia32_sqrtpd256:
   case X86::BI__builtin_ia32_sqrtps256:
+  case X86::BI__builtin_ia32_sqrtps512:
+  case X86::BI__builtin_ia32_sqrtpd512:
     return interp__builtin_x86_sqrt(S, OpPC, Call, BuiltinID);
 
   default:

clang/lib/AST/ExprConstant.cpp

@@ -12239,7 +12239,9 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
   case X86::BI__builtin_ia32_sqrtpd: 
   case X86::BI__builtin_ia32_sqrtps: 
   case X86::BI__builtin_ia32_sqrtpd256:
-  case X86::BI__builtin_ia32_sqrtps256: {
+  case X86::BI__builtin_ia32_sqrtps256: 
+  case X86::BI__builtin_ia32_sqrtps512:
+  case X86::BI__builtin_ia32_sqrtpd512: {
     llvm::errs() << "We are inside sqrtpd/sqrtps\n";
     APValue Source;
     if (!EvaluateAsRValue(Info, E->getArg(0), Source))

clang/lib/Headers/avx512fintrin.h

@@ -1458,7 +1458,7 @@ _mm512_mask_mullox_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m512i __B) {
                                        (__v8df)_mm512_sqrt_round_pd((A), (R)), \
                                        (__v8df)_mm512_setzero_pd()))
 
-static  __inline__ __m512d __DEFAULT_FN_ATTRS512
+static  __inline__ __m512d __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_sqrt_pd(__m512d __A)
 {
   return (__m512d)__builtin_ia32_sqrtpd512((__v8df)__A,
@@ -1494,7 +1494,7 @@ _mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
                                       (__v16sf)_mm512_sqrt_round_ps((A), (R)), \
                                       (__v16sf)_mm512_setzero_ps()))
 
-static  __inline__ __m512 __DEFAULT_FN_ATTRS512
+static  __inline__ __m512 __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_sqrt_ps(__m512 __A)
 {
   return (__m512)__builtin_ia32_sqrtps512((__v16sf)__A,

clang/test/CodeGen/X86/avx512f-builtins.c

@@ -18,6 +18,8 @@ __m512d test_mm512_sqrt_pd(__m512d a)
   return _mm512_sqrt_pd(a);
 }
 
+TEST_CONSTEXPR(match_m512d(_mm512_sqrt_pd(_mm512_set_pd(16.0, 9.0, 4.0, 1.0, 16.0, 9.0, 4.0, 1.0)), 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0));
+
 __m512d test_mm512_mask_sqrt_pd (__m512d __W, __mmask8 __U, __m512d __A)
 {
   // CHECK-LABEL: test_mm512_mask_sqrt_pd 
@@ -68,6 +70,8 @@ __m512 test_mm512_sqrt_ps(__m512 a)
   return _mm512_sqrt_ps(a);
 }
 
+TEST_CONSTEXPR(match_m512(_mm512_sqrt_ps(_mm512_set_ps(64.0f, 49.0f, 36.0f, 25.0f, 16.0f, 9.0f, 4.0f, 1.0f, 64.0f, 49.0f, 36.0f, 25.0f, 16.0f, 9.0f, 4.0f, 1.0f)), 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f));
+
 __m512 test_mm512_mask_sqrt_ps(__m512 __W, __mmask16 __U, __m512 __A)
 {
   // CHECK-LABEL: test_mm512_mask_sqrt_ps