[X86] Add helper enum for ternary intrinsics

[RKSimon]

Are similar macros used by any other compilers at all?

clang/lib/Headers/avx512fintrin.h

5932

Add a doxygen comment for these?

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

6010

check the immediate value?

[pengfei]

In D120307#3337352, @RKSimon wrote:

Are similar macros used by any other compilers at all?

Not so far. GCC folks agreed to add them. I'll add them to intrinsic guide as well.

[craig.topper]

clang/lib/Headers/avx512fintrin.h

5944

Why do we need to change the type on imm here? The builtin takes an int, and the enum you created defaults to int. Is this so we truncate ~ to 8-bits?

[pengfei]

clang/lib/Headers/avx512fintrin.h

5944

Yes. By the way, ICC and MSVC don't error out for it. https://godbolt.org/z/s1bG1sq9q

[pengfei]

Ping.

[LuoYuanke]

LGTM, but pls wait for 1 or 2 days to see if there are any comments from others.

[RKSimon]

LGTM

clang/lib/Headers/avx512fintrin.h

 static __inline__ __m512i __DEFAULT_FN_ATTRS512
 _mm512_maskz_srlv_epi64(__mmask8 __U, __m512i __X, __m512i __Y)
 {
   return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U,
                                             (__v8di)_mm512_srlv_epi64(__X, __Y),
                                             (__v8di)_mm512_setzero_si512());
 }
 
-#define _mm512_ternarylogic_epi32(A, B, C, imm) \
-  ((__m512i)__builtin_ia32_pternlogd512_mask((__v16si)(__m512i)(A), \
-                                             (__v16si)(__m512i)(B), \
-                                             (__v16si)(__m512i)(C), (int)(imm), \
-                                             (__mmask16)-1))
+#define _MM_TERNLOG_A 0xF0
+#define _MM_TERNLOG_B 0xCC
+#define _MM_TERNLOG_C 0xAA
+
+#define _mm512_ternarylogic_epi32(A, B, C, imm)                                \
+  ((__m512i)__builtin_ia32_pternlogd512_mask(                                  \
+      (__v16si)(__m512i)(A), (__v16si)(__m512i)(B), (__v16si)(__m512i)(C),     \
+      (unsigned char)(imm), (__mmask16)-1))
 
-#define _mm512_mask_ternarylogic_epi32(A, U, B, C, imm) \
-  ((__m512i)__builtin_ia32_pternlogd512_mask((__v16si)(__m512i)(A), \
-                                             (__v16si)(__m512i)(B), \
-                                             (__v16si)(__m512i)(C), (int)(imm), \
-                                             (__mmask16)(U)))
+#define _mm512_mask_ternarylogic_epi32(A, U, B, C, imm)                        \
+  ((__m512i)__builtin_ia32_pternlogd512_mask(                                  \
+      (__v16si)(__m512i)(A), (__v16si)(__m512i)(B), (__v16si)(__m512i)(C),     \
+      (unsigned char)(imm), (__mmask16)(U)))
 
-#define _mm512_maskz_ternarylogic_epi32(U, A, B, C, imm) \
-  ((__m512i)__builtin_ia32_pternlogd512_maskz((__v16si)(__m512i)(A), \
-                                              (__v16si)(__m512i)(B), \
-                                              (__v16si)(__m512i)(C), \
-                                              (int)(imm), (__mmask16)(U)))
+#define _mm512_maskz_ternarylogic_epi32(U, A, B, C, imm)                       \
+  ((__m512i)__builtin_ia32_pternlogd512_maskz(                                 \
+      (__v16si)(__m512i)(A), (__v16si)(__m512i)(B), (__v16si)(__m512i)(C),     \
+      (unsigned char)(imm), (__mmask16)(U)))
 
-#define _mm512_ternarylogic_epi64(A, B, C, imm) \
-  ((__m512i)__builtin_ia32_pternlogq512_mask((__v8di)(__m512i)(A), \
-                                             (__v8di)(__m512i)(B), \
-                                             (__v8di)(__m512i)(C), (int)(imm), \
-                                             (__mmask8)-1))
+#define _mm512_ternarylogic_epi64(A, B, C, imm)                                \
+  ((__m512i)__builtin_ia32_pternlogq512_mask(                                  \
+      (__v8di)(__m512i)(A), (__v8di)(__m512i)(B), (__v8di)(__m512i)(C),        \
+      (unsigned char)(imm), (__mmask8)-1))
 
-#define _mm512_mask_ternarylogic_epi64(A, U, B, C, imm) \
-  ((__m512i)__builtin_ia32_pternlogq512_mask((__v8di)(__m512i)(A), \
-                                             (__v8di)(__m512i)(B), \
-                                             (__v8di)(__m512i)(C), (int)(imm), \
-                                             (__mmask8)(U)))
+#define _mm512_mask_ternarylogic_epi64(A, U, B, C, imm)                        \
+  ((__m512i)__builtin_ia32_pternlogq512_mask(                                  \
+      (__v8di)(__m512i)(A), (__v8di)(__m512i)(B), (__v8di)(__m512i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm512_maskz_ternarylogic_epi64(U, A, B, C, imm) \
-  ((__m512i)__builtin_ia32_pternlogq512_maskz((__v8di)(__m512i)(A), \
-                                              (__v8di)(__m512i)(B), \
-                                              (__v8di)(__m512i)(C), (int)(imm), \
-                                              (__mmask8)(U)))
+#define _mm512_maskz_ternarylogic_epi64(U, A, B, C, imm)                       \
+  ((__m512i)__builtin_ia32_pternlogq512_maskz(                                 \
+      (__v8di)(__m512i)(A), (__v8di)(__m512i)(B), (__v8di)(__m512i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
 #ifdef __x86_64__
 #define _mm_cvt_roundsd_i64(A, R) \
   ((long long)__builtin_ia32_vcvtsd2si64((__v2df)(__m128d)(A), (int)(R)))
 #endif
 
 #define _mm_cvt_roundsd_si32(A, R) \
   ((int)__builtin_ia32_vcvtsd2si32((__v2df)(__m128d)(A), (int)(R)))

clang/lib/Headers/avx512vlintrin.h

 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_maskz_srai_epi64(__mmask8 __U, __m256i __A, unsigned int __imm)
 {
   return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
                                         (__v4di)_mm256_srai_epi64(__A, __imm), \
                                         (__v4di)_mm256_setzero_si256());
 }
 
-#define _mm_ternarylogic_epi32(A, B, C, imm) \
-  ((__m128i)__builtin_ia32_pternlogd128_mask((__v4si)(__m128i)(A), \
-                                             (__v4si)(__m128i)(B), \
-                                             (__v4si)(__m128i)(C), (int)(imm), \
-                                             (__mmask8)-1))
+#define _mm_ternarylogic_epi32(A, B, C, imm)                                   \
+  ((__m128i)__builtin_ia32_pternlogd128_mask(                                  \
+      (__v4si)(__m128i)(A), (__v4si)(__m128i)(B), (__v4si)(__m128i)(C),        \
+      (unsigned char)(imm), (__mmask8)-1))
 
-#define _mm_mask_ternarylogic_epi32(A, U, B, C, imm) \
-  ((__m128i)__builtin_ia32_pternlogd128_mask((__v4si)(__m128i)(A), \
-                                             (__v4si)(__m128i)(B), \
-                                             (__v4si)(__m128i)(C), (int)(imm), \
-                                             (__mmask8)(U)))
+#define _mm_mask_ternarylogic_epi32(A, U, B, C, imm)                           \
+  ((__m128i)__builtin_ia32_pternlogd128_mask(                                  \
+      (__v4si)(__m128i)(A), (__v4si)(__m128i)(B), (__v4si)(__m128i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm_maskz_ternarylogic_epi32(U, A, B, C, imm) \
-  ((__m128i)__builtin_ia32_pternlogd128_maskz((__v4si)(__m128i)(A), \
-                                              (__v4si)(__m128i)(B), \
-                                              (__v4si)(__m128i)(C), (int)(imm), \
-                                              (__mmask8)(U)))
+#define _mm_maskz_ternarylogic_epi32(U, A, B, C, imm)                          \
+  ((__m128i)__builtin_ia32_pternlogd128_maskz(                                 \
+      (__v4si)(__m128i)(A), (__v4si)(__m128i)(B), (__v4si)(__m128i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm256_ternarylogic_epi32(A, B, C, imm) \
-  ((__m256i)__builtin_ia32_pternlogd256_mask((__v8si)(__m256i)(A), \
-                                             (__v8si)(__m256i)(B), \
-                                             (__v8si)(__m256i)(C), (int)(imm), \
-                                             (__mmask8)-1))
+#define _mm256_ternarylogic_epi32(A, B, C, imm)                                \
+  ((__m256i)__builtin_ia32_pternlogd256_mask(                                  \
+      (__v8si)(__m256i)(A), (__v8si)(__m256i)(B), (__v8si)(__m256i)(C),        \
+      (unsigned char)(imm), (__mmask8)-1))
 
-#define _mm256_mask_ternarylogic_epi32(A, U, B, C, imm) \
-  ((__m256i)__builtin_ia32_pternlogd256_mask((__v8si)(__m256i)(A), \
-                                             (__v8si)(__m256i)(B), \
-                                             (__v8si)(__m256i)(C), (int)(imm), \
-                                             (__mmask8)(U)))
+#define _mm256_mask_ternarylogic_epi32(A, U, B, C, imm)                        \
+  ((__m256i)__builtin_ia32_pternlogd256_mask(                                  \
+      (__v8si)(__m256i)(A), (__v8si)(__m256i)(B), (__v8si)(__m256i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm256_maskz_ternarylogic_epi32(U, A, B, C, imm) \
-  ((__m256i)__builtin_ia32_pternlogd256_maskz((__v8si)(__m256i)(A), \
-                                              (__v8si)(__m256i)(B), \
-                                              (__v8si)(__m256i)(C), (int)(imm), \
-                                              (__mmask8)(U)))
+#define _mm256_maskz_ternarylogic_epi32(U, A, B, C, imm)                       \
+  ((__m256i)__builtin_ia32_pternlogd256_maskz(                                 \
+      (__v8si)(__m256i)(A), (__v8si)(__m256i)(B), (__v8si)(__m256i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm_ternarylogic_epi64(A, B, C, imm) \
-  ((__m128i)__builtin_ia32_pternlogq128_mask((__v2di)(__m128i)(A), \
-                                             (__v2di)(__m128i)(B), \
-                                             (__v2di)(__m128i)(C), (int)(imm), \
-                                             (__mmask8)-1))
+#define _mm_ternarylogic_epi64(A, B, C, imm)                                   \
+  ((__m128i)__builtin_ia32_pternlogq128_mask(                                  \
+      (__v2di)(__m128i)(A), (__v2di)(__m128i)(B), (__v2di)(__m128i)(C),        \
+      (unsigned char)(imm), (__mmask8)-1))
 
-#define _mm_mask_ternarylogic_epi64(A, U, B, C, imm) \
-  ((__m128i)__builtin_ia32_pternlogq128_mask((__v2di)(__m128i)(A), \
-                                             (__v2di)(__m128i)(B), \
-                                             (__v2di)(__m128i)(C), (int)(imm), \
-                                             (__mmask8)(U)))
+#define _mm_mask_ternarylogic_epi64(A, U, B, C, imm)                           \
+  ((__m128i)__builtin_ia32_pternlogq128_mask(                                  \
+      (__v2di)(__m128i)(A), (__v2di)(__m128i)(B), (__v2di)(__m128i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm_maskz_ternarylogic_epi64(U, A, B, C, imm) \
-  ((__m128i)__builtin_ia32_pternlogq128_maskz((__v2di)(__m128i)(A), \
-                                              (__v2di)(__m128i)(B), \
-                                              (__v2di)(__m128i)(C), (int)(imm), \
-                                              (__mmask8)(U)))
+#define _mm_maskz_ternarylogic_epi64(U, A, B, C, imm)                          \
+  ((__m128i)__builtin_ia32_pternlogq128_maskz(                                 \
+      (__v2di)(__m128i)(A), (__v2di)(__m128i)(B), (__v2di)(__m128i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm256_ternarylogic_epi64(A, B, C, imm) \
-  ((__m256i)__builtin_ia32_pternlogq256_mask((__v4di)(__m256i)(A), \
-                                             (__v4di)(__m256i)(B), \
-                                             (__v4di)(__m256i)(C), (int)(imm), \
-                                             (__mmask8)-1))
+#define _mm256_ternarylogic_epi64(A, B, C, imm)                                \
+  ((__m256i)__builtin_ia32_pternlogq256_mask(                                  \
+      (__v4di)(__m256i)(A), (__v4di)(__m256i)(B), (__v4di)(__m256i)(C),        \
+      (unsigned char)(imm), (__mmask8)-1))
 
-#define _mm256_mask_ternarylogic_epi64(A, U, B, C, imm) \
-  ((__m256i)__builtin_ia32_pternlogq256_mask((__v4di)(__m256i)(A), \
-                                             (__v4di)(__m256i)(B), \
-                                             (__v4di)(__m256i)(C), (int)(imm), \
-                                             (__mmask8)(U)))
+#define _mm256_mask_ternarylogic_epi64(A, U, B, C, imm)                        \
+  ((__m256i)__builtin_ia32_pternlogq256_mask(                                  \
+      (__v4di)(__m256i)(A), (__v4di)(__m256i)(B), (__v4di)(__m256i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
-#define _mm256_maskz_ternarylogic_epi64(U, A, B, C, imm) \
-  ((__m256i)__builtin_ia32_pternlogq256_maskz((__v4di)(__m256i)(A), \
-                                              (__v4di)(__m256i)(B), \
-                                              (__v4di)(__m256i)(C), (int)(imm), \
-                                              (__mmask8)(U)))
-
-
+#define _mm256_maskz_ternarylogic_epi64(U, A, B, C, imm)                       \
+  ((__m256i)__builtin_ia32_pternlogq256_maskz(                                 \
+      (__v4di)(__m256i)(A), (__v4di)(__m256i)(B), (__v4di)(__m256i)(C),        \
+      (unsigned char)(imm), (__mmask8)(U)))
 
 #define _mm256_shuffle_f32x4(A, B, imm) \
   ((__m256)__builtin_ia32_shuf_f32x4_256((__v8sf)(__m256)(A), \
                                          (__v8sf)(__m256)(B), (int)(imm)))
 
 #define _mm256_mask_shuffle_f32x4(W, U, A, B, imm) \
   ((__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
                                        (__v8sf)_mm256_shuffle_f32x4((A), (B), (imm)), \

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

__m512i test_mm512_maskz_srlv_epi64(__mmask8 __U, __m512i __X, __m512i __Y) {
   // CHECK-LABEL: @test_mm512_maskz_srlv_epi64
   // CHECK: @llvm.x86.avx512.psrlv.q.512
   // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
   return _mm512_maskz_srlv_epi64(__U, __X, __Y); 
 }
 
 __m512i test_mm512_ternarylogic_epi32(__m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_ternarylogic_epi32
   // CHECK: @llvm.x86.avx512.pternlog.d.512
-  return _mm512_ternarylogic_epi32(__A, __B, __C, 4); 
+  return _mm512_ternarylogic_epi32(__A, __B, __C, _MM_TERNLOG_A);
 }
 
 __m512i test_mm512_mask_ternarylogic_epi32(__m512i __A, __mmask16 __U, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_mask_ternarylogic_epi32
   // CHECK: @llvm.x86.avx512.pternlog.d.512
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
-  return _mm512_mask_ternarylogic_epi32(__A, __U, __B, __C, 4); 
+  return _mm512_mask_ternarylogic_epi32(__A, __U, __B, __C, _MM_TERNLOG_B);
 }
 
 __m512i test_mm512_maskz_ternarylogic_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_maskz_ternarylogic_epi32
   // CHECK: @llvm.x86.avx512.pternlog.d.512
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> zeroinitializer
-  return _mm512_maskz_ternarylogic_epi32(__U, __A, __B, __C, 4); 
+  return _mm512_maskz_ternarylogic_epi32(__U, __A, __B, __C, _MM_TERNLOG_C);
 }
 
 __m512i test_mm512_ternarylogic_epi64(__m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_ternarylogic_epi64
   // CHECK: @llvm.x86.avx512.pternlog.q.512
-  return _mm512_ternarylogic_epi64(__A, __B, __C, 4); 
+  return _mm512_ternarylogic_epi64(__A, __B, __C, _MM_TERNLOG_A & _MM_TERNLOG_B);
 }
 
 __m512i test_mm512_mask_ternarylogic_epi64(__m512i __A, __mmask8 __U, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_mask_ternarylogic_epi64
   // CHECK: @llvm.x86.avx512.pternlog.q.512
   // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
-  return _mm512_mask_ternarylogic_epi64(__A, __U, __B, __C, 4); 
+  return _mm512_mask_ternarylogic_epi64(__A, __U, __B, __C, _MM_TERNLOG_B | _MM_TERNLOG_C);
 }
 
 __m512i test_mm512_maskz_ternarylogic_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_maskz_ternarylogic_epi64
   // CHECK: @llvm.x86.avx512.pternlog.q.512
   // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> zeroinitializer
-  return _mm512_maskz_ternarylogic_epi64(__U, __A, __B, __C, 4); 
+  return _mm512_maskz_ternarylogic_epi64(__U, __A, __B, __C, ~_MM_TERNLOG_A | (_MM_TERNLOG_B ^ _MM_TERNLOG_C));
 }
 
 __m512 test_mm512_shuffle_f32x4(__m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_shuffle_f32x4
   // CHECK: shufflevector <16 x float> %{{.*}}, <16 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 16, i32 17, i32 18, i32 19>
   return _mm512_shuffle_f32x4(__A, __B, 4); 
 }