mirror of
https://github.com/Karaka-Management/cOMS.git
synced 2026-01-11 03:08:41 +00:00
838 lines
16 KiB
C
838 lines
16 KiB
C
/**
|
|
* Jingga
|
|
*
|
|
* @copyright Jingga
|
|
* @license OMS License 2.0
|
|
* @version 1.0.0
|
|
* @link https://jingga.app
|
|
*/
|
|
#ifndef COMS_TOS_STDLIB_SIMD_I16_H
|
|
#define COMS_TOS_STDLIB_SIMD_I16_H
|
|
|
|
#include <immintrin.h>
|
|
#include <xmmintrin.h>
|
|
|
|
#include "../../../stdlib/Types.h"
|
|
|
|
struct int16_8 {
|
|
union {
|
|
#if ARM
|
|
svint16_t s;
|
|
#else
|
|
__m128i s;
|
|
#endif
|
|
|
|
int16 v[8];
|
|
};
|
|
};
|
|
|
|
struct int16_16 {
|
|
union {
|
|
#if ARM
|
|
svint16_t s;
|
|
#else
|
|
__m256i s;
|
|
#endif
|
|
|
|
int16 v[16];
|
|
};
|
|
};
|
|
|
|
struct int16_32 {
|
|
union {
|
|
#if ARM
|
|
svint16_t s;
|
|
#else
|
|
__m512i s;
|
|
#endif
|
|
|
|
int16 v[32];
|
|
};
|
|
};
|
|
|
|
|
|
inline int16_8 load_int16_8(const int16* mem)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_load_si128((__m128i *) mem);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 init_int16_8(const int16* mem)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_set_epi16(
|
|
mem[0], mem[1], mem[2], mem[3],
|
|
mem[4], mem[5], mem[6], mem[7]
|
|
);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline void unload_int16_8(int16_8 a, int16 *array) { _mm_store_si128((__m128i *) array, a.s); }
|
|
|
|
inline int16_16 load_int16_16(const int16* mem)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_load_si256((__m256i *) mem);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 init_int16_16(const int16* mem)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_set_epi16(
|
|
mem[0], mem[1], mem[2], mem[3], mem[4], mem[5], mem[6], mem[7],
|
|
mem[8], mem[9], mem[10], mem[11], mem[12], mem[13], mem[14], mem[15]
|
|
);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline void unload_int16_16(int16_16 a, int16 *array) { _mm256_store_si256((__m256i *) array, a.s); }
|
|
|
|
inline int16_32 load_int16_32(const int16* mem)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_load_si512((__m512i *) mem);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 init_int16_32(const int16* mem)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_set_epi16(
|
|
mem[0], mem[1], mem[2], mem[3], mem[4], mem[5], mem[6], mem[7],
|
|
mem[8], mem[9], mem[10], mem[11], mem[12], mem[13], mem[14], mem[15],
|
|
mem[16], mem[17], mem[18], mem[19], mem[20], mem[21], mem[22], mem[23],
|
|
mem[24], mem[25], mem[26], mem[27], mem[28], mem[29], mem[30], mem[31]
|
|
);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline void unload_int16_32(int16_32 a, int16 *array) { _mm512_storeu_epi16(array, a.s); }
|
|
|
|
inline int16_8 init_zero_int16_8()
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_setzero_si128();
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 init_zero_int16_16()
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_setzero_si256();
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 init_zero_int16_32()
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_setzero_si512();
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 init_value_int16_8(int16 value)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_set1_epi16(value);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 init_value_int16_16(int16 value)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_set1_epi16(value);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 init_value_int16_32(int16 value)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_set1_epi16(value);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator+(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_add_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator+(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_add_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator+(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_add_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator-(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_sub_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator-(int16_8 a) { return init_zero_int16_8() - a; }
|
|
|
|
inline int16_16 operator-(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_sub_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator-(int16_16 a) { return init_zero_int16_16() - a; }
|
|
|
|
inline int16_32 operator-(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_sub_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator-(int16_32 a) { return init_zero_int16_32() - a; }
|
|
|
|
inline int16_8 operator*(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_mul_epi32(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator*(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_mul_epi32(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator*(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_mul_epi32(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator^(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_xor_si128(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator^(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_xor_si256(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator^(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_xor_si512(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 &operator-=(int16_8 &a, int16_8 b)
|
|
{
|
|
a = a - b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_16 &operator-=(int16_16 &a, int16_16 b)
|
|
{
|
|
a = a - b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_32 &operator-=(int16_32 &a, int16_32 b)
|
|
{
|
|
a = a - b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_8 &operator+=(int16_8 &a, int16_8 b)
|
|
{
|
|
a = a + b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_16 &operator+=(int16_16 &a, int16_16 b)
|
|
{
|
|
a = a + b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_32 &operator+=(int16_32 &a, int16_32 b)
|
|
{
|
|
a = a + b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_8 &operator*=(int16_8 &a, int16_8 b)
|
|
{
|
|
a = a * b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_16 &operator*=(int16_16 &a, int16_16 b)
|
|
{
|
|
a = a * b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_32 &operator*=(int16_32 &a, int16_32 b)
|
|
{
|
|
a = a * b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_8 &operator^=(int16_8 &a, int16_8 b)
|
|
{
|
|
a = a ^ b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_16 &operator^=(int16_16 &a, int16_16 b)
|
|
{
|
|
a = a ^ b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_32 &operator^=(int16_32 &a, int16_32 b)
|
|
{
|
|
a = a ^ b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_8 operator<(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_cmplt_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator<(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_xor_si256(_mm256_cmpgt_epi16(a.s, b.s), _mm256_set1_epi16(-1));
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator<(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_mask_blend_epi16(_mm512_cmplt_epi16_mask(a.s, b.s), a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator<=(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_andnot_si128(_mm_cmplt_epi16(b.s, a.s), _mm_set1_epi16(-1));
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator<=(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_andnot_si256(_mm256_cmpgt_epi16(a.s, b.s), _mm256_set1_epi16(-1));
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator<=(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
__mmask32 mask = _mm512_cmp_epi16_mask(a.s, b.s, _MM_CMPINT_LE);
|
|
simd.s = _mm512_mask_blend_epi16(mask, b.s, a.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator>(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_cmpgt_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator>(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_cmpgt_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator>(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_mask_blend_epi16(_mm512_cmpgt_epi16_mask(a.s, b.s), a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator>=(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_andnot_si128(_mm_cmplt_epi16(a.s, b.s), _mm_set1_epi16(-1));
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator>=(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_andnot_si256(_mm256_cmpgt_epi16(b.s, a.s), _mm256_set1_epi16(-1));
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator>=(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_mask_blend_epi16(_mm512_cmpge_epi16_mask(a.s, b.s), a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator==(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_cmpeq_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator==(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_cmpeq_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator==(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_mask_blend_epi16(_mm512_cmpeq_epi16_mask(a.s, b.s), a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator!=(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_andnot_si128(_mm_cmpeq_epi16(a.s, b.s), _mm_set1_epi16(-1));
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator!=(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_mask_blend_epi16(_mm256_cmp_epi16_mask(a.s, b.s, _MM_CMPINT_NE), a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator!=(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_mask_blend_epi16(_mm512_cmp_epi16_mask(a.s, b.s, _MM_CMPINT_NE), a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator&(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_and_si128(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator&(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_and_si256(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator&(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_and_si512(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 operator|(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_or_si128(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 operator|(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_or_si256(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 operator|(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_or_si512(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 &operator&=(int16_8 &a, int16_8 b)
|
|
{
|
|
a = a & b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_16 &operator&=(int16_16 &a, int16_16 b)
|
|
{
|
|
a = a & b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_32 &operator&=(int16_32 &a, int16_32 b)
|
|
{
|
|
a = a & b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_8 &operator|=(int16_8 &a, int16_8 b)
|
|
{
|
|
a = a | b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_16 &operator|=(int16_16 &a, int16_16 b)
|
|
{
|
|
a = a | b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_32 &operator|=(int16_32 &a, int16_32 b)
|
|
{
|
|
a = a | b;
|
|
|
|
return a;
|
|
}
|
|
|
|
inline int16_8 abs(int16_8 a)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_abs_epi16(a.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 abs(int16_16 a)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_abs_epi16(a.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 abs(int16_32 a)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_abs_epi16(a.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 simd_min(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_min_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 simd_min(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_min_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 simd_min(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_min_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 simd_max(int16_8 a, int16_8 b)
|
|
{
|
|
int16_8 simd;
|
|
simd.s = _mm_max_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_16 simd_max(int16_16 a, int16_16 b)
|
|
{
|
|
int16_16 simd;
|
|
simd.s = _mm256_max_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_32 simd_max(int16_32 a, int16_32 b)
|
|
{
|
|
int16_32 simd;
|
|
simd.s = _mm512_max_epi16(a.s, b.s);
|
|
|
|
return simd;
|
|
}
|
|
|
|
inline int16_8 clamp(int16_8 min_value, int16_8 a, int16_8 max_value)
|
|
{
|
|
return simd_min(simd_max(a, min_value), max_value);
|
|
}
|
|
|
|
inline int16_16 clamp(int16_16 min_value, int16_16 a, int16_16 max_value)
|
|
{
|
|
return simd_min(simd_max(a, min_value), max_value);
|
|
}
|
|
|
|
inline int16_32 clamp(int16_32 min_value, int16_32 a, int16_32 max_value)
|
|
{
|
|
return simd_min(simd_max(a, min_value), max_value);
|
|
}
|
|
|
|
inline int32 which_true(int16_8 a)
|
|
{
|
|
return _mm_movemask_epi8(a.s);
|
|
}
|
|
|
|
inline int32 which_true(int16_16 a)
|
|
{
|
|
return _mm256_movemask_epi8(a.s);
|
|
}
|
|
|
|
inline int32 which_true(int16_32 a)
|
|
{
|
|
return _mm512_movepi16_mask(a.s);
|
|
}
|
|
|
|
inline bool any_true(int16_8 a)
|
|
{
|
|
bool is_any_true = _mm_movemask_epi8(a.s) > 0;
|
|
|
|
return is_any_true;
|
|
}
|
|
|
|
inline bool any_true(int16_16 a)
|
|
{
|
|
bool is_any_true = _mm256_movemask_epi8(a.s) > 0;
|
|
|
|
return is_any_true;
|
|
}
|
|
|
|
inline bool any_true(int16_32 a)
|
|
{
|
|
bool is_any_true = _mm512_movepi16_mask(a.s) > 0;
|
|
|
|
return is_any_true;
|
|
}
|
|
|
|
inline bool all_true(int16_8 a)
|
|
{
|
|
bool is_true = _mm_movemask_epi8(a.s) == 15;
|
|
|
|
return is_true;
|
|
}
|
|
|
|
inline bool all_true(int16_16 a)
|
|
{
|
|
bool is_true = _mm256_movemask_epi8(a.s) == 255;
|
|
|
|
return is_true;
|
|
}
|
|
|
|
inline bool all_true(int16_32 a)
|
|
{
|
|
bool is_true = _mm512_movepi16_mask(a.s) == 65535;
|
|
|
|
return is_true;
|
|
}
|
|
|
|
inline bool all_false(int16_8 a)
|
|
{
|
|
bool is_false = _mm_movemask_epi8(a.s) == 0;
|
|
|
|
return is_false;
|
|
}
|
|
|
|
inline bool all_false(int16_16 a)
|
|
{
|
|
bool is_false = _mm256_movemask_epi8(a.s) == 0;
|
|
|
|
return is_false;
|
|
}
|
|
|
|
inline bool all_false(int16_32 a)
|
|
{
|
|
// @todo This can be optimized (requires also changes in the comparison functions return)
|
|
bool is_false = _mm512_movepi16_mask(a.s) == 0;
|
|
|
|
return is_false;
|
|
}
|
|
|
|
// @todo from down here we can optimize some of the code by NOT using the wrappers
|
|
// the code is self contained and we could use te intrinsic functions directly
|
|
|
|
inline
|
|
void simd_mult(const int16* a, f32 b, int16* result, int32 size, int32 steps)
|
|
{
|
|
int32 i = 0;
|
|
|
|
if (steps == 16) {
|
|
__m512i a_16;
|
|
__m512 af_lo, af_hi;
|
|
__m512 b_16 = _mm512_set1_ps(b);
|
|
__m512 result_lo, result_hi;
|
|
__m512i result_16;
|
|
|
|
for (; i <= size - steps; i += steps) {
|
|
a_16 = _mm512_loadu_si512((__m512i*) a);
|
|
|
|
af_lo = _mm512_cvtepi32_ps(_mm512_cvtepi16_epi32(_mm512_extracti64x4_epi64(a_16, 0)));
|
|
af_hi = _mm512_cvtepi32_ps(_mm512_cvtepi16_epi32(_mm512_extracti64x4_epi64(a_16, 1)));
|
|
|
|
result_lo = _mm512_mul_ps(af_lo, b_16);
|
|
result_hi = _mm512_mul_ps(af_hi, b_16);
|
|
|
|
result_16 = _mm512_packs_epi32(_mm512_cvtps_epi32(result_lo), _mm512_cvtps_epi32(result_hi));
|
|
_mm512_storeu_si512((__m512i*) result, result_16);
|
|
|
|
a += steps;
|
|
result += steps;
|
|
}
|
|
} else if (steps == 8) {
|
|
__m256i a_8;
|
|
__m256 af_lo, af_hi;
|
|
__m256 b_8 = _mm256_set1_ps(b);
|
|
__m256 result_lo, result_hi;
|
|
__m256i result_8;
|
|
|
|
for (; i <= size - steps; i += steps) {
|
|
a_8 = _mm256_loadu_si256((__m256i*) a);
|
|
|
|
af_lo = _mm256_cvtepi32_ps(_mm256_cvtepi16_epi32(_mm256_extracti128_si256(a_8, 0)));
|
|
af_hi = _mm256_cvtepi32_ps(_mm256_cvtepi16_epi32(_mm256_extracti128_si256(a_8, 1)));
|
|
|
|
result_lo = _mm256_mul_ps(af_lo, b_8);
|
|
result_hi = _mm256_mul_ps(af_hi, b_8);
|
|
|
|
result_8 = _mm256_packs_epi32(_mm256_cvtps_epi32(result_lo), _mm256_cvtps_epi32(result_hi));
|
|
_mm256_storeu_si256((__m256i*) result, result_8);
|
|
|
|
a += steps;
|
|
result += steps;
|
|
}
|
|
} else if (steps == 4) {
|
|
__m128i a_4;
|
|
__m128 af_lo, af_hi;
|
|
__m128 b_4 = _mm_set1_ps(b);
|
|
__m128 result_lo, result_hi;
|
|
__m128i result_4;
|
|
|
|
for (; i <= size - steps; i += steps) {
|
|
a_4 = _mm_loadu_si128((__m128i*) a);
|
|
|
|
af_lo = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(a_4));
|
|
af_hi = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(_mm_srli_si128(a_4, 8)));
|
|
|
|
result_lo = _mm_mul_ps(af_lo, b_4);
|
|
result_hi = _mm_mul_ps(af_hi, b_4);
|
|
|
|
result_4 = _mm_packs_epi32(_mm_cvtps_epi32(result_lo), _mm_cvtps_epi32(result_hi));
|
|
_mm_storeu_si128((__m128i*) result, result_4);
|
|
|
|
a += steps;
|
|
result += steps;
|
|
}
|
|
}
|
|
|
|
// Handle any remaining elements
|
|
for (; i < size; ++i) {
|
|
*result = (int16) ((f32) (*a) * b);
|
|
++a;
|
|
++result;
|
|
}
|
|
}
|
|
|
|
#endif |