This commit is contained in:
Jay D Dee
2021-10-10 22:50:19 -04:00
parent 2cd1507c2e
commit 47cc5dcff5
14 changed files with 2057 additions and 2827 deletions

View File

@@ -535,7 +535,6 @@ static inline __m512i mm512_shufll_x32( const __m512i v, const int n )
// Rotate 256 bit lanes by one 64 bit element
#define mm512_shuflr256_64( v ) _mm512_permutex_epi64( v, 0x39 )
#define mm512_shufll256_64( v ) _mm512_permutex_epi64( v, 0x93 )
// Rotate 256 bit lanes by one 32 bit element
@@ -611,9 +610,6 @@ static inline __m512i mm512_shuflr128_8( const __m512i v, const int c )
// shufl2r is 2 input ...
// Drop macros? They can easilly be rebuilt using shufl2 functions
// add shuflr shufll functions performing rotate, returning first arg
// They're faster than doing both, when both not needed.
// Shuffle concatenated { v1, v2 ) right or left by 256 bits and return
// rotated v1
// visually confusing for shif2r because of arg order. First arg is always