database/postgres
database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-04-25 21:42:33 +03:00
Code

Introduce helper SIMD functions for small byte arrays

Browse Source 
vector8_min - helper for emulating ">=" semantics

vector8_highbit_mask - used to turn the result of a vector
comparison into a bitmask

Masahiko Sawada

Reviewed by Nathan Bossart, with additional adjustments by me
Discussion: https://postgr.es/m/CAFBsxsHbBm_M22gLBO%2BAZT4mfMq3L_oX3wdKZxjeNnT7fHsYMQ%40mail.gmail.com
This commit is contained in:
John Naylor 2024-03-06 14:22:15 +07:00
parent 60c07820d6
commit 9f225e992b
1 changed files with 47 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
src/include/port/simd.h
View File

@ -79,6 +79,7 @@ static inline bool vector8_has_le(const Vector8 v, const uint8 c);
static inline bool vector8_is_highbit_set(const Vector8 v); static inline bool vector8_is_highbit_set(const Vector8 v);
#ifndef USE_NO_SIMD #ifndef USE_NO_SIMD
static inline bool vector32_is_highbit_set(const Vector32 v); static inline bool vector32_is_highbit_set(const Vector32 v);
static inline uint32 vector8_highbit_mask(const Vector8 v);
#endif #endif
/* arithmetic operations */ /* arithmetic operations */
@ -96,6 +97,7 @@ static inline Vector8 vector8_ssub(const Vector8 v1, const Vector8 v2);
*/ */
#ifndef USE_NO_SIMD #ifndef USE_NO_SIMD
static inline Vector8 vector8_eq(const Vector8 v1, const Vector8 v2); static inline Vector8 vector8_eq(const Vector8 v1, const Vector8 v2);
static inline Vector8 vector8_min(const Vector8 v1, const Vector8 v2);
static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2); static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
#endif #endif
@ -299,6 +301,36 @@ vector32_is_highbit_set(const Vector32 v)
} }
#endif /* ! USE_NO_SIMD */ #endif /* ! USE_NO_SIMD */
/*
* Return a bitmask formed from the high-bit of each element.
*/
#ifndef USE_NO_SIMD
static inline uint32
vector8_highbit_mask(const Vector8 v)
{
#ifdef USE_SSE2
return (uint32) _mm_movemask_epi8(v);
#elif defined(USE_NEON)
/*
* Note: It would be faster to use vget_lane_u64 and vshrn_n_u16, but that
* returns a uint64, making it inconvenient to combine mask values from
* multiple vectors.
*/
static const uint8 mask[16] = {
1 << 0, 1 << 1, 1 << 2, 1 << 3,
1 << 4, 1 << 5, 1 << 6, 1 << 7,
1 << 0, 1 << 1, 1 << 2, 1 << 3,
1 << 4, 1 << 5, 1 << 6, 1 << 7,
};
uint8x16_t masked = vandq_u8(vld1q_u8(mask), (uint8x16_t) vshrq_n_s8(v, 7));
uint8x16_t maskedhi = vextq_u8(masked, masked, 8);
return (uint32) vaddvq_u16((uint16x8_t) vzip1q_u8(masked, maskedhi));
#endif
}
#endif /* ! USE_NO_SIMD */
/* /*
* Return the bitwise OR of the inputs * Return the bitwise OR of the inputs
*/ */
@ -372,4 +404,19 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
} }
#endif /* ! USE_NO_SIMD */ #endif /* ! USE_NO_SIMD */
/*
* Given two vectors, return a vector with the minimum element of each.
*/
#ifndef USE_NO_SIMD
static inline Vector8
vector8_min(const Vector8 v1, const Vector8 v2)
{
#ifdef USE_SSE2
return _mm_min_epu8(v1, v2);
#elif defined(USE_NEON)
return vminq_u8(v1, v2);
#endif
}
#endif /* ! USE_NO_SIMD */
#endif /* SIMD_H */ #endif /* SIMD_H */