[Clang] feat: Allow AVX/AVX2 permute operations in constexpr

AhmedYasser5 · AhmedYasser5 · commit 732aa344e16f · 2025-12-14T21:00:06.000+08:00
Enables the usage of the following X86 intrinsics in `constexpr`:
```c
_mm256_permute2f128_pd	    _mm256_permute2f128_ps
_mm256_permute2f128_si256   _mm256_permute2x128_si256
```
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
@@ -462,15 +462,19 @@ let Features = "avx", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in
   def cvttpd2dq256 : X86Builtin<"_Vector<4, int>(_Vector<4, double>)">;
   def cvtpd2dq256 : X86Builtin<"_Vector<4, int>(_Vector<4, double>)">;
   def cvttps2dq256 : X86Builtin<"_Vector<8, int>(_Vector<8, float>)">;
-  def vperm2f128_pd256 : X86Builtin<"_Vector<4, double>(_Vector<4, double>, _Vector<4, double>, _Constant int)">;
-  def vperm2f128_ps256 : X86Builtin<"_Vector<8, float>(_Vector<8, float>, _Vector<8, float>, _Constant int)">;
-  def vperm2f128_si256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Constant int)">;
   foreach Op = ["max", "min"] in {
     def Op#pd256 : X86Builtin<"_Vector<4, double>(_Vector<4, double>, _Vector<4, double>)">;
     def Op#ps256 : X86Builtin<"_Vector<8, float>(_Vector<8, float>, _Vector<8, float>)">;
   }
 }
 
+let Features = "avx",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
+  def vperm2f128_pd256 : X86Builtin<"_Vector<4, double>(_Vector<4, double>, _Vector<4, double>, _Constant int)">;
+  def vperm2f128_ps256 : X86Builtin<"_Vector<8, float>(_Vector<8, float>, _Vector<8, float>, _Constant int)">;
+  def vperm2f128_si256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Constant int)">;
+}
+
 let Features = "avx", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
   def vpermilpd : X86Builtin<"_Vector<2, double>(_Vector<2, double>, _Constant int)">;
   def vpermilps : X86Builtin<"_Vector<4, float>(_Vector<4, float>, _Constant int)">;
@@ -567,14 +571,14 @@ let Features = "avx2", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] i
   def psadbw256
       : X86Builtin<
             "_Vector<4, long long int>(_Vector<32, char>, _Vector<32, char>)">;
-  def permti256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Constant int)">;
 }
 
 let Features = "avx2", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
   def permdf256
       : X86Builtin<"_Vector<4, double>(_Vector<4, double>, _Constant int)">;
   def permdi256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long "
                              "int>, _Constant int)">;
+  def permti256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Constant int)">;
   def pmovmskb256 : X86Builtin<"int(_Vector<32, char>)">;
   def pavgb256 : X86Builtin<"_Vector<32, unsigned char>(_Vector<32, unsigned char>, _Vector<32, unsigned char>)">;
   def pavgw256 : X86Builtin<"_Vector<16, unsigned short>(_Vector<16, unsigned short>, _Vector<16, unsigned short>)">;
diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -5101,6 +5101,29 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
           unsigned SrcIdx = (ShuffleMask >> 6) & 0x1;
           return std::pair<unsigned, int>{SrcIdx, Offset};
         });
+  case X86::BI__builtin_ia32_vperm2f128_pd256:
+  case X86::BI__builtin_ia32_vperm2f128_ps256:
+  case X86::BI__builtin_ia32_vperm2f128_si256:
+  case X86::BI__builtin_ia32_permti256:
+    return interp__builtin_ia32_shuffle_generic(
+        S, OpPC, Call,
+        [BuiltinID, Call](unsigned DstIdx, unsigned ShuffleMask) {
+          unsigned NumElements =
+              Call->getArg(0)->getType()->getAs<VectorType>()->getNumElements();
+          unsigned PreservedBitsCnt = NumElements >> 2;
+          unsigned ControlBitsCnt = DstIdx >> PreservedBitsCnt << 2;
+          unsigned ControlBits = ShuffleMask >> ControlBitsCnt;
+
+          if (BuiltinID == X86::BI__builtin_ia32_permti256 &&
+              (ControlBits & 0b1000))
+            return std::make_pair(0u, -1);
+
+          unsigned SrcVecIdx = (ControlBits & 0b10) >> 1;
+          unsigned PreservedBitsMask = (1 << PreservedBitsCnt) - 1;
+          int SrcIdx = ((ControlBits & 0b1) << PreservedBitsCnt) |
+                       (DstIdx & PreservedBitsMask);
+          return std::make_pair(SrcVecIdx, SrcIdx);
+        });
   case X86::BI__builtin_ia32_pshufb128:
   case X86::BI__builtin_ia32_pshufb256:
   case X86::BI__builtin_ia32_pshufb512:
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
@@ -14330,6 +14330,34 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
 
     return Success(ResultElements, E);
   }
+  case X86::BI__builtin_ia32_vperm2f128_pd256:
+  case X86::BI__builtin_ia32_vperm2f128_ps256:
+  case X86::BI__builtin_ia32_vperm2f128_si256:
+  case X86::BI__builtin_ia32_permti256: {
+    APValue R;
+    if (!evalShuffleGeneric(
+            Info, E, R, [E](unsigned DstIdx, unsigned ShuffleMask) {
+              unsigned NumElements = E->getArg(0)
+                                         ->getType()
+                                         ->getAs<VectorType>()
+                                         ->getNumElements();
+              unsigned PreservedBitsCnt = NumElements >> 2;
+              unsigned ControlBitsCnt = DstIdx >> PreservedBitsCnt << 2;
+              unsigned ControlBits = ShuffleMask >> ControlBitsCnt;
+
+              if (E->getBuiltinCallee() == X86::BI__builtin_ia32_permti256 &&
+                  (ControlBits & 0b1000))
+                return std::make_pair(0u, -1);
+
+              unsigned SrcVecIdx = (ControlBits & 0b10) >> 1;
+              unsigned PreservedBitsMask = (1 << PreservedBitsCnt) - 1;
+              int SrcIdx = ((ControlBits & 0b1) << PreservedBitsCnt) |
+                           (DstIdx & PreservedBitsMask);
+              return std::make_pair(SrcVecIdx, SrcIdx);
+            }))
+      return false;
+    return Success(R, E);
+  }
   }
 }
 
diff --git a/clang/test/CodeGen/X86/avx-builtins.c b/clang/test/CodeGen/X86/avx-builtins.c
@@ -1439,18 +1439,33 @@ __m256d test_mm256_permute2f128_pd(__m256d A, __m256d B) {
   return _mm256_permute2f128_pd(A, B, 0x31);
 }
 
+TEST_CONSTEXPR(match_m256d(_mm256_permute2f128_pd(((__m256d){1.0, 2.0, 3.0, 4.0}), ((__m256d){5.0, 6.0, 7.0, 8.0}), 0xA7), 7.0, 8.0, 5.0, 6.0));
+TEST_CONSTEXPR(match_m256d(_mm256_permute2f128_pd(((__m256d){1.0, 2.0, 3.0, 4.0}), ((__m256d){5.0, 6.0, 7.0, 8.0}), 0x5F), 7.0, 8.0, 3.0, 4.0));
+TEST_CONSTEXPR(match_m256d(_mm256_permute2f128_pd(((__m256d){1.0, 2.0, 3.0, 4.0}), ((__m256d){5.0, 6.0, 7.0, 8.0}), 0x37), 7.0, 8.0, 7.0, 8.0));
+TEST_CONSTEXPR(match_m256d(_mm256_permute2f128_pd(((__m256d){1.0, 2.0, 3.0, 4.0}), ((__m256d){5.0, 6.0, 7.0, 8.0}), 0x12), 5.0, 6.0, 3.0, 4.0));
+
 __m256 test_mm256_permute2f128_ps(__m256 A, __m256 B) {
   // CHECK-LABEL: test_mm256_permute2f128_ps
   // CHECK: shufflevector <8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
   return _mm256_permute2f128_ps(A, B, 0x13);
 }
 
+TEST_CONSTEXPR(match_m256(_mm256_permute2f128_ps(((__m256){1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f}), ((__m256){9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}), 0xA7), 13.0f, 14.0f, 15.0f, 16.0f, 9.0f, 10.0f, 11.0f, 12.0f));
+TEST_CONSTEXPR(match_m256(_mm256_permute2f128_ps(((__m256){1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f}), ((__m256){9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}), 0x5F), 13.0f, 14.0f, 15.0f, 16.0f, 5.0f, 6.0f, 7.0f, 8.0f));
+TEST_CONSTEXPR(match_m256(_mm256_permute2f128_ps(((__m256){1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f}), ((__m256){9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}), 0x37), 13.0f, 14.0f, 15.0f, 16.0f, 13.0f, 14.0f, 15.0f, 16.0f));
+TEST_CONSTEXPR(match_m256(_mm256_permute2f128_ps(((__m256){1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f}), ((__m256){9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}), 0x12), 9.0f, 10.0f, 11.0f, 12.0f, 5.0f, 6.0f, 7.0f, 8.0f));
+
 __m256i test_mm256_permute2f128_si256(__m256i A, __m256i B) {
   // CHECK-LABEL: test_mm256_permute2f128_si256
   // CHECK: shufflevector <8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
   return _mm256_permute2f128_si256(A, B, 0x20);
 }
 
+TEST_CONSTEXPR(match_m256i(_mm256_permute2f128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0xA7), 7LL, 8LL, 5LL, 6LL));
+TEST_CONSTEXPR(match_m256i(_mm256_permute2f128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0x5F), 7LL, 8LL, 3LL, 4LL));
+TEST_CONSTEXPR(match_m256i(_mm256_permute2f128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0x37), 7LL, 8LL, 7LL, 8LL));
+TEST_CONSTEXPR(match_m256i(_mm256_permute2f128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0x12), 5LL, 6LL, 3LL, 4LL));
+
 __m128d test_mm_permutevar_pd(__m128d A, __m128i B) {
   // CHECK-LABEL: test_mm_permutevar_pd
   // CHECK: call {{.*}}<2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %{{.*}}, <2 x i64> %{{.*}})
diff --git a/clang/test/CodeGen/X86/avx2-builtins.c b/clang/test/CodeGen/X86/avx2-builtins.c
@@ -1106,6 +1106,11 @@ __m256i test_mm256_permute2x128_si256(__m256i a, __m256i b) {
   return _mm256_permute2x128_si256(a, b, 0x38);
 }
 
+TEST_CONSTEXPR(match_m256i(_mm256_permute2x128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0xA7), 7LL, 8LL, 0LL, 0LL));
+TEST_CONSTEXPR(match_m256i(_mm256_permute2x128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0x5F), 0LL, 0LL, 3LL, 4LL));
+TEST_CONSTEXPR(match_m256i(_mm256_permute2x128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0x37), 7LL, 8LL, 7LL, 8LL));
+TEST_CONSTEXPR(match_m256i(_mm256_permute2x128_si256(((__m256i){1LL, 2LL, 3LL, 4LL}), ((__m256i){5LL, 6LL, 7LL, 8LL}), 0x12), 5LL, 6LL, 3LL, 4LL));
+
 __m256i test_mm256_permute4x64_epi64(__m256i a) {
   // CHECK-LABEL: test_mm256_permute4x64_epi64
   // CHECK: shufflevector <4 x i64> %{{.*}}, <4 x i64> poison, <4 x i32> <i32 3, i32 0, i32 2, i32 0>
