Line data    Source code

       1             : #ifndef HEADER_fd_src_util_simd_fd_avx512_h
       2             : #error "Do not include this directly; use fd_avx512.h"
       3             : #endif
       4             : 
       5             : /* Vector ushort API **************************************************/
       6             : 
       7             : /* A wwh_t is a vector where each 16-bit wide lane holds an unsigned
       8             :    16-bit integer (a "ushort").
       9             : 
      10             :    These mirror the other APIs as much as possible.  Macros are
      11             :    preferred over static inlines when it is possible to do it robustly
      12             :    to reduce the risk of the compiler mucking it up. */
      13             : 
      14             : #define wwh_t __m512i
      15             : 
      16             : /* Constructors */
      17             : 
      18             : /* Given the ushort values, return ... */
      19             : 
      20   144303745 : #define wwh_bcast(b0) _mm512_set1_epi16( (ushort)(b0) ) /* [ b0 b0 ... b0 ] */
      21             : 
      22             : 
      23             : /* Predefined constants */
      24             : 
      25             : #define wwh_zero()           _mm512_setzero_si512()  /* wwh(0, 0, ... 0) */
      26             : #define wwh_one()            _mm512_set1_epi32( 1 )  /* wwh(1, 1, ... 1) */
      27             : 
      28             : /* Memory operations */
      29             : /* Note: wwh_{ld,st} assume m is 64-byte aligned while wwh_{ldu,stu}
      30             :    allow m to have arbitrary alignment */
      31             : 
      32           0 : static inline wwh_t wwh_ld( ushort const * m ) { return _mm512_load_epi32( m ); }  /* wwh( m[0], m[1], ... m[15] ) */
      33    11816192 : static inline void  wwh_st( ushort * m, wwh_t x ) { _mm512_store_epi32( m, x ); }  /* does m[0] = x0, m[1] = x1, ... m[15] = xf */
      34             : 
      35           0 : static inline wwh_t wwh_ldu( void const * m ) { return _mm512_loadu_epi32( m ); } /* wwh( m[0], m[1], ... m[15]) */
      36           0 : static inline void  wwh_stu( void * m, wwh_t x ) { _mm512_storeu_epi32( m, x ); } /* does m[0] = x0, m[1] = x1, ... m[15] = xf */
      37             : 
      38             : /* Arithmetic operations */
      39             : 
      40   177242880 : #define wwh_add(x,y) _mm512_add_epi16( (x), (y) ) /* wwh( x0+y0, x1+y1, ... xf+y31 ) */
      41             : #define wwh_sub(x,y) _mm512_sub_epi16( (x), (y) ) /* wwh( x0-y0, x1-y1, ... xf-y31 ) */
      42             : 
      43             : 
      44             : /* Bit operations */
      45             : 
      46             : #define wwh_shl(a,imm) _mm512_slli_epi16( (a), (imm) ) /* [ a0<<imm a1<<imm ... a31<<imm ] */
      47             : #define wwh_shr(a,imm) _mm512_srli_epi16( (a), (imm) ) /* [ a0>>imm a1>>imm ... a31>>imm ] */