@zigc/lib 0.17.0-dev.27 → 0.17.0-dev.296

package/include/avx512ifmavlintrin.h

@@ -8,22 +8,35 @@
  *===-----------------------------------------------------------------------===
  */
 #ifndef __IMMINTRIN_H
-#error "Never use <avx512ifmavlintrin.h> directly; include <immintrin.h> instead."
+#error                                                                         \
+    "Never use <avx512ifmavlintrin.h> directly; include <immintrin.h> instead."
 #endif
 
 #ifndef __IFMAVLINTRIN_H
 #define __IFMAVLINTRIN_H
 
 /* Define the default attributes for the functions in this file. */
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
 #define __DEFAULT_FN_ATTRS128                                                  \
   __attribute__((__always_inline__, __nodebug__,                               \
-                 __target__("avx512ifma,avx512vl,no-evex512"),                 \
+                 __target__("avx512ifma,avx512vl"),                            \
+                 __min_vector_width__(128))) constexpr
+#define __DEFAULT_FN_ATTRS256                                                  \
+  __attribute__((__always_inline__, __nodebug__,                               \
+                 __target__("avx512ifma,avx512vl"),                            \
+                 __min_vector_width__(256))) constexpr
+#else
+#define __DEFAULT_FN_ATTRS128                                                  \
+  __attribute__((__always_inline__, __nodebug__,                               \
+                 __target__("avx512ifma,avx512vl"),                            \
                  __min_vector_width__(128)))
 #define __DEFAULT_FN_ATTRS256                                                  \
   __attribute__((__always_inline__, __nodebug__,                               \
-                 __target__("avx512ifma,avx512vl,no-evex512"),                 \
+                 __target__("avx512ifma,avx512vl"),                            \
                  __min_vector_width__(256)))
+#endif
 
+#if !(defined(__AVXIFMA__) || defined(__AVX512IFMA__))
 #define _mm_madd52hi_epu64(X, Y, Z)                                            \
   ((__m128i)__builtin_ia32_vpmadd52huq128((__v2di)(X), (__v2di)(Y),            \
                                           (__v2di)(Z)))
@@ -39,71 +52,85 @@
 #define _mm256_madd52lo_epu64(X, Y, Z)                                         \
   ((__m256i)__builtin_ia32_vpmadd52luq256((__v4di)(X), (__v4di)(Y),            \
                                           (__v4di)(Z)))
+#endif
 
+#if defined(__AVX512IFMA__)
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_madd52hi_epu64 (__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
-{
-  return (__m128i)__builtin_ia32_selectq_128(__M,
-                                      (__v2di)_mm_madd52hi_epu64(__W, __X, __Y),
-                                      (__v2di)__W);
+_mm_madd52hi_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
+  return (__m128i)__builtin_ia32_vpmadd52huq128((__v2di)__X, (__v2di)__Y,
+                                                (__v2di)__Z);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS256
+_mm256_madd52hi_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
+  return (__m256i)__builtin_ia32_vpmadd52huq256((__v4di)__X, (__v4di)__Y,
+                                                (__v4di)__Z);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_maskz_madd52hi_epu64 (__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z)
-{
-  return (__m128i)__builtin_ia32_selectq_128(__M,
-                                      (__v2di)_mm_madd52hi_epu64(__X, __Y, __Z),
-                                      (__v2di)_mm_setzero_si128());
+_mm_madd52lo_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
+  return (__m128i)__builtin_ia32_vpmadd52luq128((__v2di)__X, (__v2di)__Y,
+                                                (__v2di)__Z);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_madd52hi_epu64 (__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y)
-{
-  return (__m256i)__builtin_ia32_selectq_256(__M,
-                                   (__v4di)_mm256_madd52hi_epu64(__W, __X, __Y),
-                                   (__v4di)__W);
+_mm256_madd52lo_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
+  return (__m256i)__builtin_ia32_vpmadd52luq256((__v4di)__X, (__v4di)__Y,
+                                                (__v4di)__Z);
 }
+#endif
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_maskz_madd52hi_epu64 (__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z)
-{
-  return (__m256i)__builtin_ia32_selectq_256(__M,
-                                   (__v4di)_mm256_madd52hi_epu64(__X, __Y, __Z),
-                                   (__v4di)_mm256_setzero_si256());
+static __inline__ __m128i __DEFAULT_FN_ATTRS128
+_mm_mask_madd52hi_epu64(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y) {
+  return (__m128i)__builtin_ia32_selectq_128(
+      __M, (__v2di)__builtin_ia32_vpmadd52huq128(__W, __X, __Y), (__v2di)__W);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_madd52lo_epu64 (__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
-{
-  return (__m128i)__builtin_ia32_selectq_128(__M,
-                                      (__v2di)_mm_madd52lo_epu64(__W, __X, __Y),
-                                      (__v2di)__W);
+_mm_maskz_madd52hi_epu64(__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z) {
+  return (__m128i)__builtin_ia32_selectq_128(
+      __M, (__v2di)__builtin_ia32_vpmadd52huq128(__X, __Y, __Z),
+      (__v2di)_mm_setzero_si128());
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_madd52hi_epu64(
+    __m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) {
+  return (__m256i)__builtin_ia32_selectq_256(
+      __M, (__v4di)__builtin_ia32_vpmadd52huq256(__W, __X, __Y), (__v4di)__W);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_maskz_madd52hi_epu64(
+    __mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z) {
+  return (__m256i)__builtin_ia32_selectq_256(
+      __M, (__v4di)__builtin_ia32_vpmadd52huq256(__X, __Y, __Z),
+      (__v4di)_mm256_setzero_si256());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_maskz_madd52lo_epu64 (__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z)
-{
-  return (__m128i)__builtin_ia32_selectq_128(__M,
-                                      (__v2di)_mm_madd52lo_epu64(__X, __Y, __Z),
-                                      (__v2di)_mm_setzero_si128());
+_mm_mask_madd52lo_epu64(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y) {
+  return (__m128i)__builtin_ia32_selectq_128(
+      __M, (__v2di)__builtin_ia32_vpmadd52luq128(__W, __X, __Y), (__v2di)__W);
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_madd52lo_epu64 (__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y)
-{
-  return (__m256i)__builtin_ia32_selectq_256(__M,
-                                   (__v4di)_mm256_madd52lo_epu64(__W, __X, __Y),
-                                   (__v4di)__W);
+static __inline__ __m128i __DEFAULT_FN_ATTRS128
+_mm_maskz_madd52lo_epu64(__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z) {
+  return (__m128i)__builtin_ia32_selectq_128(
+      __M, (__v2di)__builtin_ia32_vpmadd52luq128(__X, __Y, __Z),
+      (__v2di)_mm_setzero_si128());
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_maskz_madd52lo_epu64 (__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z)
-{
-  return (__m256i)__builtin_ia32_selectq_256(__M,
-                                   (__v4di)_mm256_madd52lo_epu64(__X, __Y, __Z),
-                                   (__v4di)_mm256_setzero_si256());
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_madd52lo_epu64(
+    __m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) {
+  return (__m256i)__builtin_ia32_selectq_256(
+      __M, (__v4di)__builtin_ia32_vpmadd52luq256(__W, __X, __Y), (__v4di)__W);
 }
 
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_maskz_madd52lo_epu64(
+    __mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z) {
+  return (__m256i)__builtin_ia32_selectq_256(
+      __M, (__v4di)__builtin_ia32_vpmadd52luq256(__X, __Y, __Z),
+      (__v4di)_mm256_setzero_si256());
+}
 
 #undef __DEFAULT_FN_ATTRS128
 #undef __DEFAULT_FN_ATTRS256

package/include/avx512vbmi2intrin.h

@@ -15,8 +15,15 @@
 #define __AVX512VBMI2INTRIN_H
 
 /* Define the default attributes for the functions in this file. */
-#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi2,evex512"), __min_vector_width__(512)))
-
+#define __DEFAULT_FN_ATTRS                                                     \
+  __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi2"),    \
+                 __min_vector_width__(512)))
+
+#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
 _mm512_mask_compress_epi16(__m512i __S, __mmask32 __U, __m512i __D)
@@ -212,14 +219,14 @@ _mm512_maskz_expandloadu_epi8(__mmask64 __U, void const *__P)
                                     (__v32hi)_mm512_shrdi_epi16((A), (B), (I)), \
                                     (__v32hi)_mm512_setzero_si512()))
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_shldv_epi64(__m512i __A, __m512i __B, __m512i __C)
 {
-  return (__m512i)__builtin_ia32_vpshldvq512((__v8di)__A, (__v8di)__B,
-                                             (__v8di)__C);
+  return (__m512i)__builtin_elementwise_fshl((__v8du)__A, (__v8du)__B,
+                                             (__v8du)__C);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_mask_shldv_epi64(__m512i __A, __mmask8 __U, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectq_512(__U,
@@ -227,7 +234,7 @@ _mm512_mask_shldv_epi64(__m512i __A, __mmask8 __U, __m512i __B, __m512i __C)
                                       (__v8di)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_maskz_shldv_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectq_512(__U,
@@ -235,14 +242,14 @@ _mm512_maskz_shldv_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C)
                                       (__v8di)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_shldv_epi32(__m512i __A, __m512i __B, __m512i __C)
 {
-  return (__m512i)__builtin_ia32_vpshldvd512((__v16si)__A, (__v16si)__B,
-                                             (__v16si)__C);
+  return (__m512i)__builtin_elementwise_fshl((__v16su)__A, (__v16su)__B,
+                                             (__v16su)__C);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_mask_shldv_epi32(__m512i __A, __mmask16 __U, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectd_512(__U,
@@ -250,7 +257,7 @@ _mm512_mask_shldv_epi32(__m512i __A, __mmask16 __U, __m512i __B, __m512i __C)
                                      (__v16si)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_maskz_shldv_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectd_512(__U,
@@ -258,14 +265,14 @@ _mm512_maskz_shldv_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C)
                                      (__v16si)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_shldv_epi16(__m512i __A, __m512i __B, __m512i __C)
 {
-  return (__m512i)__builtin_ia32_vpshldvw512((__v32hi)__A, (__v32hi)__B,
-                                             (__v32hi)__C);
+  return (__m512i)__builtin_elementwise_fshl((__v32hu)__A, (__v32hu)__B,
+                                             (__v32hu)__C);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_mask_shldv_epi16(__m512i __A, __mmask32 __U, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectw_512(__U,
@@ -273,7 +280,7 @@ _mm512_mask_shldv_epi16(__m512i __A, __mmask32 __U, __m512i __B, __m512i __C)
                                      (__v32hi)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_maskz_shldv_epi16(__mmask32 __U, __m512i __A, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectw_512(__U,
@@ -281,14 +288,15 @@ _mm512_maskz_shldv_epi16(__mmask32 __U, __m512i __A, __m512i __B, __m512i __C)
                                      (__v32hi)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_shrdv_epi64(__m512i __A, __m512i __B, __m512i __C)
 {
-  return (__m512i)__builtin_ia32_vpshrdvq512((__v8di)__A, (__v8di)__B,
-                                             (__v8di)__C);
+  // Ops __A and __B are swapped.
+  return (__m512i)__builtin_elementwise_fshr((__v8du)__B, (__v8du)__A,
+                                             (__v8du)__C);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_mask_shrdv_epi64(__m512i __A, __mmask8 __U, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectq_512(__U,
@@ -296,7 +304,7 @@ _mm512_mask_shrdv_epi64(__m512i __A, __mmask8 __U, __m512i __B, __m512i __C)
                                       (__v8di)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_maskz_shrdv_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectq_512(__U,
@@ -304,14 +312,15 @@ _mm512_maskz_shrdv_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C)
                                       (__v8di)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_shrdv_epi32(__m512i __A, __m512i __B, __m512i __C)
 {
-  return (__m512i)__builtin_ia32_vpshrdvd512((__v16si)__A, (__v16si)__B,
-                                             (__v16si)__C);
+  // Ops __A and __B are swapped.
+  return (__m512i)__builtin_elementwise_fshr((__v16su)__B, (__v16su)__A,
+                                             (__v16su)__C);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_mask_shrdv_epi32(__m512i __A, __mmask16 __U, __m512i __B, __m512i __C)
 {
   return (__m512i) __builtin_ia32_selectd_512(__U,
@@ -319,7 +328,7 @@ _mm512_mask_shrdv_epi32(__m512i __A, __mmask16 __U, __m512i __B, __m512i __C)
                                      (__v16si)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_maskz_shrdv_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C)
 {
   return (__m512i) __builtin_ia32_selectd_512(__U,
@@ -327,14 +336,15 @@ _mm512_maskz_shrdv_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C)
                                      (__v16si)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_shrdv_epi16(__m512i __A, __m512i __B, __m512i __C)
 {
-  return (__m512i)__builtin_ia32_vpshrdvw512((__v32hi)__A, (__v32hi)__B,
-                                             (__v32hi)__C);
+  // Ops __A and __B are swapped.
+  return (__m512i)__builtin_elementwise_fshr((__v32hu)__B, (__v32hu)__A,
+                                             (__v32hu)__C);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_mask_shrdv_epi16(__m512i __A, __mmask32 __U, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectw_512(__U,
@@ -342,7 +352,7 @@ _mm512_mask_shrdv_epi16(__m512i __A, __mmask32 __U, __m512i __B, __m512i __C)
                                      (__v32hi)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
+static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_maskz_shrdv_epi16(__mmask32 __U, __m512i __A, __m512i __B, __m512i __C)
 {
   return (__m512i)__builtin_ia32_selectw_512(__U,
@@ -352,6 +362,7 @@ _mm512_maskz_shrdv_epi16(__mmask32 __U, __m512i __A, __m512i __B, __m512i __C)
 
 
 #undef __DEFAULT_FN_ATTRS
+#undef __DEFAULT_FN_ATTRS_CONSTEXPR
 
 #endif
 

package/include/avx512vbmiintrin.h

@@ -15,63 +15,57 @@
 #define __VBMIINTRIN_H
 
 /* Define the default attributes for the functions in this file. */
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
 #define __DEFAULT_FN_ATTRS                                                     \
-  __attribute__((__always_inline__, __nodebug__,                               \
-                 __target__("avx512vbmi,evex512"), __min_vector_width__(512)))
+  __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi"),     \
+                 __min_vector_width__(512))) constexpr
+#else
+#define __DEFAULT_FN_ATTRS                                                     \
+  __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi"),     \
+                 __min_vector_width__(512)))
+#endif
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_permutex2var_epi8(__m512i __A, __m512i __I, __m512i __B)
-{
+_mm512_permutex2var_epi8(__m512i __A, __m512i __I, __m512i __B) {
   return (__m512i)__builtin_ia32_vpermi2varqi512((__v64qi)__A, (__v64qi)__I,
                                                  (__v64qi) __B);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_permutex2var_epi8(__m512i __A, __mmask64 __U, __m512i __I,
-                              __m512i __B)
-{
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_permutex2var_epi8(
+    __m512i __A, __mmask64 __U, __m512i __I, __m512i __B) {
   return (__m512i)__builtin_ia32_selectb_512(__U,
                                (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),
                                (__v64qi)__A);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask2_permutex2var_epi8(__m512i __A, __m512i __I, __mmask64 __U,
-                               __m512i __B)
-{
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask2_permutex2var_epi8(
+    __m512i __A, __m512i __I, __mmask64 __U, __m512i __B) {
   return (__m512i)__builtin_ia32_selectb_512(__U,
                                (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),
                                (__v64qi)__I);
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_permutex2var_epi8(__mmask64 __U, __m512i __A, __m512i __I,
-                               __m512i __B)
-{
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_permutex2var_epi8(
+    __mmask64 __U, __m512i __A, __m512i __I, __m512i __B) {
   return (__m512i)__builtin_ia32_selectb_512(__U,
                                (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),
                                (__v64qi)_mm512_setzero_si512());
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_permutexvar_epi8 (__m512i __A, __m512i __B)
-{
+_mm512_permutexvar_epi8(__m512i __A, __m512i __B) {
   return (__m512i)__builtin_ia32_permvarqi512((__v64qi) __B, (__v64qi) __A);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_permutexvar_epi8 (__mmask64 __M, __m512i __A,
-        __m512i __B)
-{
+_mm512_maskz_permutexvar_epi8(__mmask64 __M, __m512i __A, __m512i __B) {
   return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,
                                      (__v64qi)_mm512_permutexvar_epi8(__A, __B),
                                      (__v64qi)_mm512_setzero_si512());
 }
 
-static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_permutexvar_epi8 (__m512i __W, __mmask64 __M, __m512i __A,
-             __m512i __B)
-{
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_permutexvar_epi8(
+    __m512i __W, __mmask64 __M, __m512i __A, __m512i __B) {
   return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,
                                      (__v64qi)_mm512_permutexvar_epi8(__A, __B),
                                      (__v64qi)__W);
@@ -100,7 +94,5 @@ _mm512_maskz_multishift_epi64_epi8(__mmask64 __M, __m512i __X, __m512i __Y)
                                 (__v64qi)_mm512_setzero_si512());
 }
 
-
 #undef __DEFAULT_FN_ATTRS
-
 #endif

package/include/avx512vbmivlintrin.h

@@ -15,126 +15,114 @@
 #define __VBMIVLINTRIN_H
 
 /* Define the default attributes for the functions in this file. */
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
 #define __DEFAULT_FN_ATTRS128                                                  \
   __attribute__((__always_inline__, __nodebug__,                               \
-                 __target__("avx512vbmi,avx512vl,no-evex512"),                 \
+                 __target__("avx512vbmi,avx512vl"),                            \
+                 __min_vector_width__(128))) constexpr
+#define __DEFAULT_FN_ATTRS256                                                  \
+  __attribute__((__always_inline__, __nodebug__,                               \
+                 __target__("avx512vbmi,avx512vl"),                            \
+                 __min_vector_width__(256))) constexpr
+#else
+#define __DEFAULT_FN_ATTRS128                                                  \
+  __attribute__((__always_inline__, __nodebug__,                               \
+                 __target__("avx512vbmi,avx512vl"),                            \
                  __min_vector_width__(128)))
 #define __DEFAULT_FN_ATTRS256                                                  \
   __attribute__((__always_inline__, __nodebug__,                               \
-                 __target__("avx512vbmi,avx512vl,no-evex512"),                 \
+                 __target__("avx512vbmi,avx512vl"),                            \
                  __min_vector_width__(256)))
+#endif
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_permutex2var_epi8(__m128i __A, __m128i __I, __m128i __B)
-{
+_mm_permutex2var_epi8(__m128i __A, __m128i __I, __m128i __B) {
   return (__m128i)__builtin_ia32_vpermi2varqi128((__v16qi)__A,
                                                  (__v16qi)__I,
                                                  (__v16qi)__B);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_permutex2var_epi8(__m128i __A, __mmask16 __U, __m128i __I,
-                           __m128i __B)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_mask_permutex2var_epi8(
+    __m128i __A, __mmask16 __U, __m128i __I, __m128i __B) {
   return (__m128i)__builtin_ia32_selectb_128(__U,
                                   (__v16qi)_mm_permutex2var_epi8(__A, __I, __B),
                                   (__v16qi)__A);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask2_permutex2var_epi8(__m128i __A, __m128i __I, __mmask16 __U,
-                            __m128i __B)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_mask2_permutex2var_epi8(
+    __m128i __A, __m128i __I, __mmask16 __U, __m128i __B) {
   return (__m128i)__builtin_ia32_selectb_128(__U,
                                   (__v16qi)_mm_permutex2var_epi8(__A, __I, __B),
                                   (__v16qi)__I);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_maskz_permutex2var_epi8(__mmask16 __U, __m128i __A, __m128i __I,
-                            __m128i __B)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_maskz_permutex2var_epi8(
+    __mmask16 __U, __m128i __A, __m128i __I, __m128i __B) {
   return (__m128i)__builtin_ia32_selectb_128(__U,
                                   (__v16qi)_mm_permutex2var_epi8(__A, __I, __B),
                                   (__v16qi)_mm_setzero_si128());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_permutex2var_epi8(__m256i __A, __m256i __I, __m256i __B)
-{
+_mm256_permutex2var_epi8(__m256i __A, __m256i __I, __m256i __B) {
   return (__m256i)__builtin_ia32_vpermi2varqi256((__v32qi)__A, (__v32qi)__I,
                                                  (__v32qi)__B);
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_permutex2var_epi8(__m256i __A, __mmask32 __U, __m256i __I,
-                              __m256i __B)
-{
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_permutex2var_epi8(
+    __m256i __A, __mmask32 __U, __m256i __I, __m256i __B) {
   return (__m256i)__builtin_ia32_selectb_256(__U,
                                (__v32qi)_mm256_permutex2var_epi8(__A, __I, __B),
                                (__v32qi)__A);
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask2_permutex2var_epi8(__m256i __A, __m256i __I, __mmask32 __U,
-                               __m256i __B)
-{
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask2_permutex2var_epi8(
+    __m256i __A, __m256i __I, __mmask32 __U, __m256i __B) {
   return (__m256i)__builtin_ia32_selectb_256(__U,
                                (__v32qi)_mm256_permutex2var_epi8(__A, __I, __B),
                                (__v32qi)__I);
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_maskz_permutex2var_epi8(__mmask32 __U, __m256i __A, __m256i __I,
-                               __m256i __B)
-{
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_maskz_permutex2var_epi8(
+    __mmask32 __U, __m256i __A, __m256i __I, __m256i __B) {
   return (__m256i)__builtin_ia32_selectb_256(__U,
                                (__v32qi)_mm256_permutex2var_epi8(__A, __I, __B),
                                (__v32qi)_mm256_setzero_si256());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_permutexvar_epi8 (__m128i __A, __m128i __B)
-{
+_mm_permutexvar_epi8(__m128i __A, __m128i __B) {
   return (__m128i)__builtin_ia32_permvarqi128((__v16qi)__B, (__v16qi)__A);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_maskz_permutexvar_epi8 (__mmask16 __M, __m128i __A, __m128i __B)
-{
+_mm_maskz_permutexvar_epi8(__mmask16 __M, __m128i __A, __m128i __B) {
   return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
                                         (__v16qi)_mm_permutexvar_epi8(__A, __B),
                                         (__v16qi)_mm_setzero_si128());
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_permutexvar_epi8 (__m128i __W, __mmask16 __M, __m128i __A,
-          __m128i __B)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_mask_permutexvar_epi8(
+    __m128i __W, __mmask16 __M, __m128i __A, __m128i __B) {
   return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
                                         (__v16qi)_mm_permutexvar_epi8(__A, __B),
                                         (__v16qi)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_permutexvar_epi8 (__m256i __A, __m256i __B)
-{
+_mm256_permutexvar_epi8(__m256i __A, __m256i __B) {
   return (__m256i)__builtin_ia32_permvarqi256((__v32qi) __B, (__v32qi) __A);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_maskz_permutexvar_epi8 (__mmask32 __M, __m256i __A,
-        __m256i __B)
-{
+_mm256_maskz_permutexvar_epi8(__mmask32 __M, __m256i __A, __m256i __B) {
   return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
                                      (__v32qi)_mm256_permutexvar_epi8(__A, __B),
                                      (__v32qi)_mm256_setzero_si256());
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_permutexvar_epi8 (__m256i __W, __mmask32 __M, __m256i __A,
-             __m256i __B)
-{
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_permutexvar_epi8(
+    __m256i __W, __mmask32 __M, __m256i __A, __m256i __B) {
   return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
                                      (__v32qi)_mm256_permutexvar_epi8(__A, __B),
                                      (__v32qi)__W);
@@ -186,8 +174,6 @@ _mm256_maskz_multishift_epi64_epi8(__mmask32 __M, __m256i __X, __m256i __Y)
                                 (__v32qi)_mm256_setzero_si256());
 }
 
-
 #undef __DEFAULT_FN_ATTRS128
 #undef __DEFAULT_FN_ATTRS256
-
 #endif