[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr #161056

chaitanyav · 2025-09-28T07:35:54Z

Resolves #160498

github-actions · 2025-09-28T07:39:24Z

✅ With the latest revision this PR passed the C/C++ code formatter.

llvmbot · 2025-09-29T03:02:19Z

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-x86

Author: NagaChaitanya Vellanki (chaitanyav)

Changes

Resolves #160498

Patch is 27.25 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/161056.diff

7 Files Affected:

(modified) clang/include/clang/Basic/BuiltinsX86.td (+21-21)
(modified) clang/lib/AST/ByteCode/InterpBuiltin.cpp (+22)
(modified) clang/lib/AST/ExprConstant.cpp (+51)
(modified) clang/lib/Headers/avx512ifmaintrin.h (+37-34)
(modified) clang/lib/Headers/avxifmaintrin.h (+14-4)
(added) clang/test/AST/ByteCode/x86-ifma-constexpr.cpp (+36)
(added) clang/test/Sema/x86-ifma-constexpr.cpp (+262)

diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index 77e599587edc3..a5247629e255f 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -2101,27 +2101,6 @@ let Features = "avx512vl", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
   def movdqa64store256_mask : X86Builtin<"void(_Vector<4, long long int *>, _Vector<4, long long int>, unsigned char)">;
 }
 
-let Features = "avx512ifma", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
-  def vpmadd52huq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">;
-  def vpmadd52luq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">;
-}
-
-let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
-  def vpmadd52huq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">;
-}
-
-let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def vpmadd52huq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">;
-}
-
-let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
-  def vpmadd52luq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">;
-}
-
-let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def vpmadd52luq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">;
-}
-
 let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
   def vcomisd : X86Builtin<"int(_Vector<2, double>, _Vector<2, double>, _Constant int, _Constant int)">;
   def vcomiss : X86Builtin<"int(_Vector<4, float>, _Vector<4, float>, _Constant int, _Constant int)">;
@@ -3128,6 +3107,27 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
   def kordi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">;
 }
 
+let Features = "avx512ifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in {
+  def vpmadd52huq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">;
+  def vpmadd52luq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">;
+}
+
+let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
+  def vpmadd52huq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">;
+}
+
+let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
+  def vpmadd52huq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">;
+}
+
+let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
+  def vpmadd52luq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">;
+}
+
+let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
+  def vpmadd52luq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">;
+}
+
 let Features = "avx512dq", Attributes = [NoThrow, Const] in {
   def kortestcqi : X86Builtin<"int(unsigned char, unsigned char)">;
   def kortestzqi : X86Builtin<"int(unsigned char, unsigned char)">;
diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
index 891344d4e6ed0..cf6a739cc5c60 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3564,6 +3564,28 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
           return F;
         });
 
+  case X86::BI__builtin_ia32_vpmadd52luq128:
+  case X86::BI__builtin_ia32_vpmadd52luq256:
+  case X86::BI__builtin_ia32_vpmadd52luq512:
+    return interp__builtin_elementwise_triop(
+        S, OpPC, Call, [](const APSInt &A, const APSInt &B, const APSInt &C) {
+          APSInt result = A * B + C;
+          APSInt mask(APSInt::getAllOnes(52).zext(64), false);
+          APSInt masked_result = result & mask;
+          return APSInt(masked_result, true); // unsigned result
+        });
+  case X86::BI__builtin_ia32_vpmadd52huq128:
+  case X86::BI__builtin_ia32_vpmadd52huq256:
+  case X86::BI__builtin_ia32_vpmadd52huq512:
+    return interp__builtin_elementwise_triop(
+        S, OpPC, Call, [](const APSInt &A, const APSInt &B, const APSInt &C) {
+          APSInt result = A * B + C;
+          APSInt mask(APSInt::getAllOnes(52).zext(64), false);
+          APSInt shifted_result = result >> 52;
+          APSInt masked_result = shifted_result & mask;
+          return APSInt(masked_result, true); // unsigned result
+        });
+
   case X86::BI__builtin_ia32_vpshldd128:
   case X86::BI__builtin_ia32_vpshldd256:
   case X86::BI__builtin_ia32_vpshldd512:
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index b706b14945b6d..c9d8a2b01dd74 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -60,10 +60,12 @@
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/LSP/Logging.h"
 #include "llvm/Support/SaveAndRestore.h"
 #include "llvm/Support/SipHash.h"
 #include "llvm/Support/TimeProfiler.h"
 #include "llvm/Support/raw_ostream.h"
+
 #include <cstring>
 #include <functional>
 #include <limits>
@@ -11869,6 +11871,55 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
     return Success(APValue(ResultElements.data(), ResultElements.size()), E);
   }
 
+  case X86::BI__builtin_ia32_vpmadd52luq128:
+  case X86::BI__builtin_ia32_vpmadd52luq256:
+  case X86::BI__builtin_ia32_vpmadd52luq512: {
+    APValue A, B, C;
+    if (!EvaluateAsRValue(Info, E->getArg(0), A) ||
+        !EvaluateAsRValue(Info, E->getArg(1), B) ||
+        !EvaluateAsRValue(Info, E->getArg(2), C))
+      return false;
+
+    unsigned ALen = A.getVectorLength();
+    SmallVector<APValue, 4> ResultElements;
+    ResultElements.reserve(ALen);
+
+    for (unsigned EltNum = 0; EltNum < ALen; EltNum += 1) {
+      APInt AElt = A.getVectorElt(EltNum).getInt();
+      APInt BElt = B.getVectorElt(EltNum).getInt();
+      APInt CElt = C.getVectorElt(EltNum).getInt();
+      APInt ResElt(AElt.zext(128) * BElt.zext(128) + CElt.zext(128));
+      APInt Mask(64, 0x000FFFFFFFFFFFFFULL);
+      ResultElements.push_back(APValue(APSInt(ResElt.trunc(64) & Mask, false)));
+    }
+
+    return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+  }
+  case X86::BI__builtin_ia32_vpmadd52huq128:
+  case X86::BI__builtin_ia32_vpmadd52huq256:
+  case X86::BI__builtin_ia32_vpmadd52huq512: {
+    APValue A, B, C;
+    if (!EvaluateAsRValue(Info, E->getArg(0), A) ||
+        !EvaluateAsRValue(Info, E->getArg(1), B) ||
+        !EvaluateAsRValue(Info, E->getArg(2), C))
+      return false;
+
+    unsigned ALen = A.getVectorLength();
+    SmallVector<APValue, 4> ResultElements;
+    ResultElements.reserve(ALen);
+
+    for (unsigned EltNum = 0; EltNum < ALen; EltNum += 1) {
+      APInt AElt = A.getVectorElt(EltNum).getInt();
+      APInt BElt = B.getVectorElt(EltNum).getInt();
+      APInt CElt = C.getVectorElt(EltNum).getInt();
+      APInt ResElt(AElt.zext(128) * BElt.zext(128) + CElt.zext(128));
+      APInt Mask(64, 0x000FFFFFFFFFFFFFULL);
+      ResultElements.push_back(
+          APValue(APSInt(ResElt.lshr(52).trunc(64) & Mask, false)));
+    }
+
+    return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+  }
   case clang::X86::BI__builtin_ia32_vprotbi:
   case clang::X86::BI__builtin_ia32_vprotdi:
   case clang::X86::BI__builtin_ia32_vprotqi:
diff --git a/clang/lib/Headers/avx512ifmaintrin.h b/clang/lib/Headers/avx512ifmaintrin.h
index f01b322ce7787..6d800f25e5798 100644
--- a/clang/lib/Headers/avx512ifmaintrin.h
+++ b/clang/lib/Headers/avx512ifmaintrin.h
@@ -19,52 +19,55 @@
   __attribute__((__always_inline__, __nodebug__, __target__("avx512ifma"),     \
                  __min_vector_width__(512)))
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_madd52hi_epu64 (__m512i __X, __m512i __Y, __m512i __Z)
-{
-  return (__m512i)__builtin_ia32_vpmadd52huq512((__v8di) __X, (__v8di) __Y,
-                                                (__v8di) __Z);
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS_CONSTEXPR __DEFAULT_FN_ATTRS constexpr
+#else
+#define __DEFAULT_FN_ATTRS_CONSTEXPR __DEFAULT_FN_ATTRS
+#endif
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_madd52hi_epu64(__m512i __X, __m512i __Y, __m512i __Z) {
+  return (__m512i)__builtin_ia32_vpmadd52huq512((__v8di)__X, (__v8di)__Y,
+                                                (__v8di)__Z);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_madd52hi_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
-{
-  return (__m512i)__builtin_ia32_selectq_512(__M,
-                                   (__v8di)_mm512_madd52hi_epu64(__W, __X, __Y),
-                                   (__v8di)__W);
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_mask_madd52hi_epu64(__m512i __W, __mmask8 __M, __m512i __X,
+                           __m512i __Y) {
+  return (__m512i)__builtin_ia32_selectq_512(
+      __M, (__v8di)_mm512_madd52hi_epu64(__W, __X, __Y), (__v8di)__W);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_madd52hi_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z)
-{
-  return (__m512i)__builtin_ia32_selectq_512(__M,
-                                   (__v8di)_mm512_madd52hi_epu64(__X, __Y, __Z),
-                                   (__v8di)_mm512_setzero_si512());
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_maskz_madd52hi_epu64(__mmask8 __M, __m512i __X, __m512i __Y,
+                            __m512i __Z) {
+  return (__m512i)__builtin_ia32_selectq_512(
+      __M, (__v8di)_mm512_madd52hi_epu64(__X, __Y, __Z),
+      (__v8di)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_madd52lo_epu64 (__m512i __X, __m512i __Y, __m512i __Z)
-{
-  return (__m512i)__builtin_ia32_vpmadd52luq512((__v8di) __X, (__v8di) __Y,
-                                                (__v8di) __Z);
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_madd52lo_epu64(__m512i __X, __m512i __Y, __m512i __Z) {
+  return (__m512i)__builtin_ia32_vpmadd52luq512((__v8di)__X, (__v8di)__Y,
+                                                (__v8di)__Z);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_madd52lo_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
-{
-  return (__m512i)__builtin_ia32_selectq_512(__M,
-                                   (__v8di)_mm512_madd52lo_epu64(__W, __X, __Y),
-                                   (__v8di)__W);
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_mask_madd52lo_epu64(__m512i __W, __mmask8 __M, __m512i __X,
+                           __m512i __Y) {
+  return (__m512i)__builtin_ia32_selectq_512(
+      __M, (__v8di)_mm512_madd52lo_epu64(__W, __X, __Y), (__v8di)__W);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_madd52lo_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z)
-{
-  return (__m512i)__builtin_ia32_selectq_512(__M,
-                                   (__v8di)_mm512_madd52lo_epu64(__X, __Y, __Z),
-                                   (__v8di)_mm512_setzero_si512());
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_maskz_madd52lo_epu64(__mmask8 __M, __m512i __X, __m512i __Y,
+                            __m512i __Z) {
+  return (__m512i)__builtin_ia32_selectq_512(
+      __M, (__v8di)_mm512_madd52lo_epu64(__X, __Y, __Z),
+      (__v8di)_mm512_setzero_si512());
 }
 
 #undef __DEFAULT_FN_ATTRS
+#undef __DEFAULT_FN_ATTRS_CONSTEXPR
 
 #endif
diff --git a/clang/lib/Headers/avxifmaintrin.h b/clang/lib/Headers/avxifmaintrin.h
index 5c782d2a5b865..1a9aaaf53affa 100644
--- a/clang/lib/Headers/avxifmaintrin.h
+++ b/clang/lib/Headers/avxifmaintrin.h
@@ -22,6 +22,14 @@
   __attribute__((__always_inline__, __nodebug__, __target__("avxifma"),        \
                  __min_vector_width__(256)))
 
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256 constexpr
+#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128 constexpr
+#else
+#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256
+#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128
+#endif
+
 // must vex-encoding
 
 /// Multiply packed unsigned 52-bit integers in each 64-bit element of \a __Y
@@ -55,7 +63,7 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
 _mm_madd52hi_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
   return (__m128i)__builtin_ia32_vpmadd52huq128((__v2di)__X, (__v2di)__Y,
                                                 (__v2di)__Z);
@@ -92,7 +100,7 @@ _mm_madd52hi_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
 _mm256_madd52hi_avx_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
   return (__m256i)__builtin_ia32_vpmadd52huq256((__v4di)__X, (__v4di)__Y,
                                                 (__v4di)__Z);
@@ -129,7 +137,7 @@ _mm256_madd52hi_avx_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
 _mm_madd52lo_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
   return (__m128i)__builtin_ia32_vpmadd52luq128((__v2di)__X, (__v2di)__Y,
                                                 (__v2di)__Z);
@@ -166,12 +174,14 @@ _mm_madd52lo_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
 _mm256_madd52lo_avx_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
   return (__m256i)__builtin_ia32_vpmadd52luq256((__v4di)__X, (__v4di)__Y,
                                                 (__v4di)__Z);
 }
 #undef __DEFAULT_FN_ATTRS128
 #undef __DEFAULT_FN_ATTRS256
+#undef __DEFAULT_FN_ATTRS256_CONSTEXPR
+#undef __DEFAULT_FN_ATTRS128_CONSTEXPR
 
 #endif // __AVXIFMAINTRIN_H
diff --git a/clang/test/AST/ByteCode/x86-ifma-constexpr.cpp b/clang/test/AST/ByteCode/x86-ifma-constexpr.cpp
new file mode 100644
index 0000000000000..72dde7d27b3ec
--- /dev/null
+++ b/clang/test/AST/ByteCode/x86-ifma-constexpr.cpp
@@ -0,0 +1,36 @@
+// RUN: %clang_cc1 -fexperimental-new-constant-interpreter -std=c++2a -fsyntax-only \
+// RUN:   -triple x86_64-unknown-unknown -target-feature +avxifma -ffreestanding \
+// RUN:   -verify %s
+
+// Test constexpr evaluation of X86 IFMA intrinsics with the ByteCode interpreter.
+
+typedef long long __m128i __attribute__((__vector_size__(16), __aligned__(16)));
+
+// Declare required IFMA builtin functions.
+extern "C" {
+__m128i __builtin_ia32_vpmadd52luq128(__m128i, __m128i, __m128i);
+__m128i __builtin_ia32_vpmadd52huq128(__m128i, __m128i, __m128i);
+}
+
+// Intrinsic wrapper functions.
+static constexpr __inline__ __m128i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm_madd52lo_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
+  return __builtin_ia32_vpmadd52luq128(__X, __Y, __Z);
+}
+
+// Simple test to check if IFMA intrinsics can be used in constexpr context
+constexpr bool test_basic_ifma() {
+  __m128i a = (__m128i){5ULL, 3ULL};
+  __m128i b = (__m128i){7ULL, 4ULL};
+  __m128i c = (__m128i){2ULL, 1ULL};
+
+  // Just test that we can call the intrinsic in constexpr context
+  __m128i result = _mm_madd52lo_epu64(a, b, c);
+  (void)result; // Suppress unused variable warning
+  return true;
+}
+
+// Basic test to verify constexpr evaluation works
+static_assert(test_basic_ifma(), "Basic IFMA constexpr test failed");
+
+// expected-no-diagnostics
\ No newline at end of file
diff --git a/clang/test/Sema/x86-ifma-constexpr.cpp b/clang/test/Sema/x86-ifma-constexpr.cpp
new file mode 100644
index 0000000000000..eb1fca13dfea5
--- /dev/null
+++ b/clang/test/Sema/x86-ifma-constexpr.cpp
@@ -0,0 +1,262 @@
+// RUN: %clang_cc1 -std=c++2a -fsyntax-only -triple x86_64-unknown-unknown \
+// RUN:   -target-feature +avxifma -ffreestanding -verify %s
+
+typedef long long __m128i __attribute__((__vector_size__(16), __aligned__(16)));
+typedef long long __m256i __attribute__((__vector_size__(32), __aligned__(32)));
+typedef long long __m512i __attribute__((__vector_size__(64), __aligned__(64)));
+
+typedef unsigned long long __v2du __attribute__((__vector_size__(16)));
+typedef unsigned long long __v4du __attribute__((__vector_size__(32)));
+typedef unsigned long long __v8du __attribute__((__vector_size__(64)));
+
+extern "C" {
+__m128i __builtin_ia32_vpmadd52luq128(__m128i, __m128i, __m128i);
+__m128i __builtin_ia32_vpmadd52huq128(__m128i, __m128i, __m128i);
+__m256i __builtin_ia32_vpmadd52luq256(__m256i, __m256i, __m256i);
+__m256i __builtin_ia32_vpmadd52huq256(__m256i, __m256i, __m256i);
+__m512i __builtin_ia32_vpmadd52luq512(__m512i, __m512i, __m512i);
+__m512i __builtin_ia32_vpmadd52huq512(__m512i, __m512i, __m512i);
+}
+static constexpr __inline__ __m128i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm_madd52lo_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
+  return __builtin_ia32_vpmadd52luq128(__X, __Y, __Z);
+}
+
+static constexpr __inline__ __m128i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm_madd52hi_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
+  return __builtin_ia32_vpmadd52huq128(__X, __Y, __Z);
+}
+
+static constexpr __inline__ __m256i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm256_madd52lo_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
+  return __builtin_ia32_vpmadd52luq256(__X, __Y, __Z);
+}
+
+static constexpr __inline__ __m256i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm256_madd52hi_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
+  return __builtin_ia32_vpmadd52huq256(__X, __Y, __Z);
+}
+
+static constexpr __inline__ __m512i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm512_madd52lo_epu64(__m512i __X, __m512i __Y, __m512i __Z) {
+  return __builtin_ia32_vpmadd52luq512(__X, __Y, __Z);
+}
+
+static constexpr __inline__ __m512i __attribute__((__always_inline__, __nodebug__, __target__("avxifma")))
+_mm512_madd52hi_epu64(__m512i __X, __m512i __Y, __m512i __Z) {
+  return __builtin_ia32_vpmadd52huq512(__X, __Y, __Z);
+}
+
+#define TEST_CONSTEXPR(expr) static_assert(expr, "constexpr test failed")
+
+constexpr bool match_v2du(__m128i result, unsigned long long e0, unsigned long long e1) {
+  __v2du v = (__v2du)result;
+  return v[0] == e0 && v[1] == e1;
+}
+
+constexpr bool match_v4du(__m256i result, unsigned long long e0, unsigned long long e1,
+                          unsigned long long e2, unsigned long long e3) {
+  __v4du v = (__v4du)result;
+  return v[0] == e0 && v[1] == e1 && v[2] == e2 && v[3] == e3;
+}
+
+constexpr bool match_v8du(__m512i result, unsigned long long e0, unsigned long long e1,
+                          unsigned long long e2, unsigned long long e3,
+                          unsigned long long e4, unsigned long long e5,
+                          unsigned long long e6, unsigned long long ...
[truncated]

clang/test/AST/ByteCode/x86-ifma-constexpr.cpp

clang/lib/AST/ByteCode/InterpBuiltin.cpp

RKSimon

Please take a look at the Intel Intrinsics Guide to get a better understanding of the madd52 instructions

clang/include/clang/Basic/BuiltinsX86.td

clang/lib/AST/ByteCode/InterpBuiltin.cpp

clang/test/AST/ByteCode/x86-ifma-constexpr.cpp

RKSimon

Missing avx512ifmavl-builtins.c coverage

clang/lib/AST/ByteCode/InterpBuiltin.cpp

clang/test/CodeGen/X86/avx512ifma-builtins.c

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

clang/lib/Headers/avx512ifmaintrin.h

RKSimon

avx512ifmavlintrin.h ?

clang/lib/AST/ByteCode/InterpBuiltin.cpp

… intrinsics test coverage (#161684) Inspired by test problems encountered on #161056

RKSimon

you should now be able to cleanup your tests and get rid of the macosx hacks

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

… intrinsics test coverage (llvm#161684) Inspired by test problems encountered on llvm#161056

clang/lib/AST/ByteCode/InterpBuiltin.cpp

clang/lib/AST/ExprConstant.cpp

clang/lib/Headers/avxifmaintrin.h

clang/test/CodeGen/X86/avx512ifma-builtins.c

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

clang/test/CodeGen/X86/avx512ifmavl-builtins.c

clang/test/CodeGen/X86/avx512ifma-builtins.c

XChy

APInt calculation looks good to me. Please wait for @RKSimon to approve the frontend-specific part.

RKSimon

a couple of minors

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

chaitanyav · 2025-10-08T16:45:29Z

@RKSimon i have addressed the comments. Also i had to use __attribute__((target("avx512ifma,avx512vl"))) in avxifma-builtins.c for those 4 tests since we specify the -target-feature +avxifma in that test file.

RKSimon · 2025-10-09T09:17:45Z

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

+#include "builtin_test_helpers.h"
+

+__attribute__((target("avx512ifma,avx512vl")))


I don't understand why this is necessary

+ /mnt/f/llvm-project/build/bin/FileCheck /mnt/f/llvm-project/clang/test/CodeGen/X86/avxifma-builtins.c /mnt/f/llvm-project/clang/test/CodeGen/X86/avxifma-builtins.c:18:10: error: always_inline function '_mm_madd52hi_epu64' requires target feature 'avx512ifma', but would be inlined into function 'test_mm_madd52hi_epu64' that is compiled without support for 'avx512ifma' 18 | return _mm_madd52hi_epu64(__X, __Y, __Z); | ^ 1 error generated. FileCheck error: '<stdin>' is empty. FileCheck command line: /mnt/f/llvm-project/build/bin/FileCheck /mnt/f/llvm-project/clang/test/CodeGen/X86/avxifma-builtins.c

The fact that it didn't require this beforehand is very suspicous but I can't see the mistake right now @FreddyLeaf @phoebewang any ideas?

i think i understand why, in clang/lib/Headers/avx512ifmavlintrin.hpreviously it was a macro, now it's a static function. That is why we need the __attribute__

#define _mm_madd52hi_epu64(X, Y, Z) \ ((__m128i)__builtin_ia32_vpmadd52huq128((__v2di)(X), (__v2di)(Y), \ (__v2di)(Z)))

static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_madd52hi_epu64(__m128i __X, __m128i __Y, __m128i __Z) { return (__m128i)__builtin_ia32_vpmadd52huq128((__v2di)__X, (__v2di)__Y, (__v2di)__Z); }

We cannot change it to static functions directly. I have a workaround that may help here: #162760

i will rebase and resolve once those changes are merged

clang/lib/Headers/avx512ifmavlintrin.h

…Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr Resolves llvm#160498

RKSimon

LGTM - cheers

…Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr (llvm#161056) Resolves llvm#160498

chaitanyav force-pushed the issue_160498 branch 2 times, most recently from 4b022cd to bc897b3 Compare September 28, 2025 10:08

chaitanyav marked this pull request as ready for review September 29, 2025 03:01

chaitanyav requested review from FreddyLeaf, RKSimon and tbaederr and removed request for RKSimon and tbaederr September 29, 2025 03:01

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:headers Headers provided by Clang, e.g. for intrinsics clang:bytecode Issues for the clang bytecode constexpr interpreter labels Sep 29, 2025

chaitanyav force-pushed the issue_160498 branch from bc897b3 to 1920f6d Compare September 29, 2025 04:50

tbaederr reviewed Sep 29, 2025

View reviewed changes

RKSimon requested changes Sep 29, 2025

View reviewed changes

chaitanyav force-pushed the issue_160498 branch from 1920f6d to 14c11d6 Compare September 29, 2025 23:15

chaitanyav requested review from RKSimon and tbaederr September 30, 2025 00:21

chaitanyav self-assigned this Sep 30, 2025

RKSimon requested changes Sep 30, 2025

View reviewed changes

clang/lib/AST/ByteCode/InterpBuiltin.cpp Outdated Show resolved Hide resolved

clang/test/CodeGen/X86/avx512ifma-builtins.c Outdated Show resolved Hide resolved

clang/test/CodeGen/X86/avxifma-builtins.c Outdated Show resolved Hide resolved

RKSimon requested a review from XChy September 30, 2025 07:49

chaitanyav force-pushed the issue_160498 branch 2 times, most recently from f11c3bd to 5bddd56 Compare September 30, 2025 16:50

RKSimon self-requested a review October 2, 2025 09:16

RKSimon requested changes Oct 2, 2025

View reviewed changes

clang/lib/Headers/avx512ifmaintrin.h Show resolved Hide resolved

RKSimon requested changes Oct 2, 2025

View reviewed changes

XChy reviewed Oct 2, 2025

View reviewed changes

clang/lib/AST/ByteCode/InterpBuiltin.cpp Outdated Show resolved Hide resolved

RKSimon mentioned this pull request Oct 2, 2025

[X86] Add bytecode / -fexperimental-new-constant-interpreter VPMADD52 intrinsics test coverage #161684

Merged

RKSimon added a commit that referenced this pull request Oct 2, 2025

[X86] Add bytecode / -fexperimental-new-constant-interpreter VPMADD52…

5bbf724

… intrinsics test coverage (#161684) Inspired by test problems encountered on #161056

RKSimon requested changes Oct 2, 2025

View reviewed changes

clang/test/CodeGen/X86/avxifma-builtins.c Outdated Show resolved Hide resolved

clang/test/CodeGen/X86/avxifma-builtins.c Outdated Show resolved Hide resolved

clang/test/CodeGen/X86/avxifma-builtins.c Outdated Show resolved Hide resolved

mahesh-attarde pushed a commit to mahesh-attarde/llvm-project that referenced this pull request Oct 3, 2025

[X86] Add bytecode / -fexperimental-new-constant-interpreter VPMADD52…

1e20d80

… intrinsics test coverage (llvm#161684) Inspired by test problems encountered on llvm#161056

chaitanyav force-pushed the issue_160498 branch from 5bddd56 to 0613d83 Compare October 3, 2025 05:15

chaitanyav requested review from RKSimon and XChy October 3, 2025 05:22

chaitanyav force-pushed the issue_160498 branch from 0613d83 to 83f379b Compare October 3, 2025 05:35

XChy reviewed Oct 3, 2025

View reviewed changes

clang/lib/AST/ByteCode/InterpBuiltin.cpp Outdated Show resolved Hide resolved

clang/lib/AST/ExprConstant.cpp Outdated Show resolved Hide resolved

RKSimon requested changes Oct 3, 2025

View reviewed changes

chaitanyav force-pushed the issue_160498 branch from 83f379b to 7692612 Compare October 3, 2025 18:22

XChy reviewed Oct 4, 2025

View reviewed changes

RKSimon requested changes Oct 6, 2025

View reviewed changes

clang/test/CodeGen/X86/avxifma-builtins.c Show resolved Hide resolved

clang/test/CodeGen/X86/avxifma-builtins.c Show resolved Hide resolved

chaitanyav force-pushed the issue_160498 branch 2 times, most recently from 3fb4fb0 to 314265f Compare October 8, 2025 03:45

RKSimon reviewed Oct 9, 2025

View reviewed changes

RKSimon reviewed Oct 12, 2025

View reviewed changes

clang/lib/Headers/avx512ifmavlintrin.h Outdated Show resolved Hide resolved

[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - …

1a0a6e4

…Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr Resolves llvm#160498

chaitanyav force-pushed the issue_160498 branch from 314265f to 1a0a6e4 Compare October 13, 2025 00:13

Merge branch 'main' into issue_160498

78abc73

RKSimon approved these changes Oct 14, 2025

View reviewed changes

RKSimon enabled auto-merge (squash) October 14, 2025 13:21

RKSimon merged commit b110b7d into llvm:main Oct 14, 2025
9 of 10 checks passed

akadutta pushed a commit to akadutta/llvm-project that referenced this pull request Oct 14, 2025

[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - …

f0ff17c

…Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr (llvm#161056) Resolves llvm#160498

		#include "builtin_test_helpers.h"


		__attribute__((target("avx512ifma,avx512vl")))

[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr #161056

[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr #161056

Uh oh!

Conversation

chaitanyav commented Sep 28, 2025

Uh oh!

github-actions bot commented Sep 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Sep 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

XChy left a comment

Choose a reason for hiding this comment

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

chaitanyav commented Oct 8, 2025

Uh oh!

RKSimon Oct 9, 2025

Choose a reason for hiding this comment

Uh oh!

chaitanyav Oct 9, 2025

Choose a reason for hiding this comment

Uh oh!

RKSimon Oct 9, 2025

Choose a reason for hiding this comment

Uh oh!

chaitanyav Oct 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

phoebewang Oct 10, 2025

Choose a reason for hiding this comment

Uh oh!

chaitanyav Oct 10, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Sep 28, 2025 •

edited

Loading

llvmbot commented Sep 29, 2025 •

edited

Loading

chaitanyav Oct 9, 2025 •

edited

Loading