blob: f89eaa87e9a27c41a749681c924a2b131730fdcf [file] [log] [blame]
/*
* Copyright (c) 2021, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 3-Clause Clear License
* and the Alliance for Open Media Patent License 1.0. If the BSD 3-Clause Clear
* License was not distributed with this source code in the LICENSE file, you
* can obtain it at aomedia.org/license/software-license/bsd-3-c-c/. If the
* Alliance for Open Media Patent License 1.0 was not distributed with this
* source code in the PATENTS file, you can obtain it at
* aomedia.org/license/patent-license/.
*/
#ifndef AOM_AOM_DSP_X86_MEM_SSE2_H_
#define AOM_AOM_DSP_X86_MEM_SSE2_H_
#include <emmintrin.h> // SSE2
#include <string.h>
#include "config/aom_config.h"
#include "aom/aom_integer.h"
static INLINE uint16_t loadu_uint16(const void *src) {
uint16_t v;
memcpy(&v, src, sizeof(v));
return v;
}
static INLINE uint32_t loadu_uint32(const void *src) {
uint32_t v;
memcpy(&v, src, sizeof(v));
return v;
}
static INLINE uint64_t loadu_uint64(const void *src) {
uint64_t v;
memcpy(&v, src, sizeof(v));
return v;
}
static INLINE void _mm_storeh_epi64(__m128i *const d, const __m128i s) {
_mm_storeh_pi((__m64 *)d, _mm_castsi128_ps(s));
}
static INLINE __m128i loadh_epi64(const void *const src, const __m128i s) {
return _mm_castps_si128(
_mm_loadh_pi(_mm_castsi128_ps(s), (const __m64 *)src));
}
static INLINE __m128i load_8bit_4x4_to_1_reg_sse2(const void *const src,
const int byte_stride) {
return _mm_setr_epi32(loadu_uint32((int8_t *)src + 0 * byte_stride),
loadu_uint32((int8_t *)src + 1 * byte_stride),
loadu_uint32((int8_t *)src + 2 * byte_stride),
loadu_uint32((int8_t *)src + 3 * byte_stride));
}
static INLINE __m128i load_8bit_8x2_to_1_reg_sse2(const void *const src,
const int byte_stride) {
__m128i dst;
dst = _mm_loadl_epi64((__m128i *)((int8_t *)src + 0 * byte_stride));
dst = loadh_epi64((int8_t *)src + 1 * byte_stride, dst);
return dst;
}
static INLINE void store_8bit_8x4_from_16x2(const __m128i *const s,
uint8_t *const d,
const ptrdiff_t stride) {
_mm_storel_epi64((__m128i *)(d + 0 * stride), s[0]);
_mm_storeh_epi64((__m128i *)(d + 1 * stride), s[0]);
_mm_storel_epi64((__m128i *)(d + 2 * stride), s[1]);
_mm_storeh_epi64((__m128i *)(d + 3 * stride), s[1]);
}
static INLINE void store_8bit_4x4(const __m128i *const s, uint8_t *const d,
const ptrdiff_t stride) {
*(int *)(d + 0 * stride) = _mm_cvtsi128_si32(s[0]);
*(int *)(d + 1 * stride) = _mm_cvtsi128_si32(s[1]);
*(int *)(d + 2 * stride) = _mm_cvtsi128_si32(s[2]);
*(int *)(d + 3 * stride) = _mm_cvtsi128_si32(s[3]);
}
static INLINE void store_8bit_4x4_sse2(const __m128i s, uint8_t *const d,
const ptrdiff_t stride) {
__m128i ss[4];
ss[0] = s;
ss[1] = _mm_srli_si128(s, 4);
ss[2] = _mm_srli_si128(s, 8);
ss[3] = _mm_srli_si128(s, 12);
store_8bit_4x4(ss, d, stride);
}
static INLINE void load_8bit_8x4(const uint8_t *const s, const ptrdiff_t stride,
__m128i *const d) {
d[0] = _mm_loadl_epi64((const __m128i *)(s + 0 * stride));
d[1] = _mm_loadl_epi64((const __m128i *)(s + 1 * stride));
d[2] = _mm_loadl_epi64((const __m128i *)(s + 2 * stride));
d[3] = _mm_loadl_epi64((const __m128i *)(s + 3 * stride));
}
static INLINE void loadu_8bit_16x4(const uint8_t *const s,
const ptrdiff_t stride, __m128i *const d) {
d[0] = _mm_loadu_si128((const __m128i *)(s + 0 * stride));
d[1] = _mm_loadu_si128((const __m128i *)(s + 1 * stride));
d[2] = _mm_loadu_si128((const __m128i *)(s + 2 * stride));
d[3] = _mm_loadu_si128((const __m128i *)(s + 3 * stride));
}
static INLINE void load_8bit_8x8(const uint8_t *const s, const ptrdiff_t stride,
__m128i *const d) {
load_8bit_8x4(s + 0 * stride, stride, &d[0]);
load_8bit_8x4(s + 4 * stride, stride, &d[4]);
}
static INLINE void storeu_8bit_16x4(const __m128i *const s, uint8_t *const d,
const ptrdiff_t stride) {
_mm_storeu_si128((__m128i *)(d + 0 * stride), s[0]);
_mm_storeu_si128((__m128i *)(d + 1 * stride), s[1]);
_mm_storeu_si128((__m128i *)(d + 2 * stride), s[2]);
_mm_storeu_si128((__m128i *)(d + 3 * stride), s[3]);
}
#endif // AOM_AOM_DSP_X86_MEM_SSE2_H_