v26.1

simd-utils/simd-128.h

@@ -642,6 +642,15 @@ static inline void v128_memcpy( v128_t *dst, const v128_t *src, const int n )
 #define v128_shuflr32(v)    _mm_shuffle_epi32( v, 0x39 )
 #define v128_shufll32(v)    _mm_shuffle_epi32( v, 0x93 )
 
+/* Zen6 AMD only
+// Reverse bits in bytes
+#if defined(__AVX512VL__) && defined(__AVX512BMM__)
+
+#define v128_bitrev8           _mm_vbitrevb_epi8
+
+#endif
+*/
+
 // Endian byte swap.
 
 #if defined(__SSSE3__)

simd-utils/simd-256.h

@@ -409,6 +409,15 @@ static inline __m256i mm256_shuflr128_x8( const __m256i v, const int c )
 { return _mm256_alignr_epi8( v, v, c ); }
 */
 
+/* Zen6 AMD only
+// Reverse bits in bytes
+#if defined(__AVX512VL__) && defined(__AVX512BMM__)
+
+#define mm256_bitrev8           _mm256_vbitrevb_epi8
+
+#endif
+*/
+
 // Reverse byte order in elements, endian bswap.
 #define mm256_bswap_64( v )     _mm256_shuffle_epi8( v, V256_BSWAP64 )
 

simd-utils/simd-512.h

@@ -257,6 +257,15 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
 #endif
 */
 
+/* Zen6 AMD only
+// Reverse bits in bytes
+#if defined(__AVX512BMM__)
+
+#define mm512_bitrev8           _mm512_vbitrevb_epi8
+
+#endif
+*/
+
 //
 // Reverse byte order of packed elements, vectorized endian conversion.
 

simd-utils/simd-neon.h

@@ -431,8 +431,10 @@ static inline uint32x4_t v128_shuflr32( uint32x4_t v )
 static inline uint32x4_t v128_shufll32( uint32x4_t v )
 {   return vextq_u32( v, v, 3 ); }
 
-// reverse bits in bytes, nothing like it in x86_64
+/* not used
+// reverse bits in bytes, nothing like it in x86_64 until Zen6
 #define v128_bitrev8           vrbitq_u8
+*/
 
 // reverse byte order
 #define v128_bswap16(v)        (uint16x8_t)vrev16q_u8( (uint8x16_t)(v) )