/*
* Copyright (c) 2020, Alliance for Open Media. All rights reserved.
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. 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 www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <immintrin.h>
// AVX2
#include "config/aom_dsp_rtcd.h"
#include "aom_dsp/aom_filter.h"
#include "aom_dsp/x86/synonyms.h"
typedef void (*high_variance_fn_t)(
const uint16_t *src,
int src_stride,
const uint16_t *ref,
int ref_stride,
uint32_t *sse,
int *sum);
static uint32_t aom_highbd_var_filter_block2d_bil_avx2(
const uint8_t *src_ptr8,
unsigned int src_pixels_per_line,
int pixel_step,
unsigned int output_height,
unsigned int output_width,
const uint32_t xoffset,
const uint32_t yoffset,
const uint8_t *dst_ptr8,
int dst_stride, uint32_t *sse) {
const __m256i filter1 =
_mm256_set1_epi32((
int)(bilinear_filters_2t[xoffset][1] << 16) |
bilinear_filters_2t[xoffset][0]);
const __m256i filter2 =
_mm256_set1_epi32((
int)(bilinear_filters_2t[yoffset][1] << 16) |
bilinear_filters_2t[yoffset][0]);
const __m256i one = _mm256_set1_epi16(1);
const int bitshift = 0x40;
(
void)pixel_step;
unsigned int i, j, prev = 0, curr = 2;
uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src_ptr8);
uint16_t *dst_ptr = CONVERT_TO_SHORTPTR(dst_ptr8);
uint16_t *src_ptr_ref = src_ptr;
uint16_t *dst_ptr_ref = dst_ptr;
int64_t sum_long = 0;
uint64_t sse_long = 0;
unsigned int rshift = 0, inc = 1;
__m256i rbias = _mm256_set1_epi32(bitshift);
__m256i opointer[8];
unsigned int range;
if (xoffset == 0) {
if (yoffset == 0) {
// xoffset==0 && yoffset==0
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
__m256i V_S_SRC = _mm256_loadu_si256((
const __m256i *)src_ptr);
src_ptr += src_pixels_per_line;
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
else if (yoffset == 4) {
// xoffset==0 && yoffset==4
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
opointer[0] = _mm256_loadu_si256((
const __m256i *)src_ptr);
src_ptr += src_pixels_per_line;
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
prev = curr;
curr = (curr == 0) ? 1 : 0;
opointer[curr] = _mm256_loadu_si256((
const __m256i *)src_ptr);
src_ptr += src_pixels_per_line;
__m256i V_S_SRC = _mm256_avg_epu16(opointer[curr], opointer[prev]);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
else {
// xoffset==0 && yoffset==1,2,3,5,6,7
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
opointer[0] = _mm256_loadu_si256((
const __m256i *)src_ptr);
src_ptr += src_pixels_per_line;
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
prev = curr;
curr = (curr == 0) ? 1 : 0;
opointer[curr] = _mm256_loadu_si256((
const __m256i *)src_ptr);
src_ptr += src_pixels_per_line;
__m256i V_S_M1 =
_mm256_unpacklo_epi16(opointer[prev], opointer[curr]);
__m256i V_S_M2 =
_mm256_unpackhi_epi16(opointer[prev], opointer[curr]);
__m256i V_S_MAD1 = _mm256_madd_epi16(V_S_M1, filter2);
__m256i V_S_MAD2 = _mm256_madd_epi16(V_S_M2, filter2);
__m256i V_S_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD1, rbias), 7);
__m256i V_S_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD2, rbias), 7);
__m256i V_S_SRC = _mm256_packus_epi32(V_S_S1, V_S_S2);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
}
else if (xoffset == 4) {
if (yoffset == 0) {
// xoffset==4 && yoffset==0
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
__m256i V_H_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_H_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
opointer[0] = _mm256_avg_epu16(V_H_D1, V_H_D2);
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
prev = curr;
curr = (curr == 0) ? 1 : 0;
__m256i V_V_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_V_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
opointer[curr] = _mm256_avg_epu16(V_V_D1, V_V_D2);
__m256i V_S_M1 =
_mm256_unpacklo_epi16(opointer[prev], opointer[curr]);
__m256i V_S_M2 =
_mm256_unpackhi_epi16(opointer[prev], opointer[curr]);
__m256i V_S_MAD1 = _mm256_madd_epi16(V_S_M1, filter2);
__m256i V_S_MAD2 = _mm256_madd_epi16(V_S_M2, filter2);
__m256i V_S_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD1, rbias), 7);
__m256i V_S_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD2, rbias), 7);
__m256i V_S_SRC = _mm256_packus_epi32(V_S_S1, V_S_S2);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
else if (yoffset == 4) {
// xoffset==4 && yoffset==4
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
__m256i V_H_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_H_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
opointer[0] = _mm256_avg_epu16(V_H_D1, V_H_D2);
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
prev = curr;
curr = (curr == 0) ? 1 : 0;
__m256i V_V_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_V_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
opointer[curr] = _mm256_avg_epu16(V_V_D1, V_V_D2);
__m256i V_S_SRC = _mm256_avg_epu16(opointer[curr], opointer[prev]);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
else {
// xoffset==4 && yoffset==1,2,3,5,6,7
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
__m256i V_H_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_H_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
opointer[0] = _mm256_avg_epu16(V_H_D1, V_H_D2);
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
prev = curr;
curr = (curr == 0) ? 1 : 0;
__m256i V_V_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_V_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
opointer[curr] = _mm256_avg_epu16(V_V_D1, V_V_D2);
__m256i V_S_M1 =
_mm256_unpacklo_epi16(opointer[prev], opointer[curr]);
__m256i V_S_M2 =
_mm256_unpackhi_epi16(opointer[prev], opointer[curr]);
__m256i V_S_MAD1 = _mm256_madd_epi16(V_S_M1, filter2);
__m256i V_S_MAD2 = _mm256_madd_epi16(V_S_M2, filter2);
__m256i V_S_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD1, rbias), 7);
__m256i V_S_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD2, rbias), 7);
__m256i V_S_SRC = _mm256_packus_epi32(V_S_S1, V_S_S2);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
}
else if (yoffset == 0) {
// xoffset==1,2,3,5,6,7 && yoffset==0
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
__m256i V_V_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_V_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
__m256i V_V_M1 = _mm256_unpacklo_epi16(V_V_D1, V_V_D2);
__m256i V_V_M2 = _mm256_unpackhi_epi16(V_V_D1, V_V_D2);
__m256i V_V_MAD1 = _mm256_madd_epi16(V_V_M1, filter1);
__m256i V_V_MAD2 = _mm256_madd_epi16(V_V_M2, filter1);
__m256i V_V_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_V_MAD1, rbias), 7);
__m256i V_V_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_V_MAD2, rbias), 7);
opointer[curr] = _mm256_packus_epi32(V_V_S1, V_V_S2);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(opointer[curr], V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
else if (yoffset == 4) {
// xoffset==1,2,3,5,6,7 && yoffset==4
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
__m256i V_H_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_H_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
__m256i V_H_M1 = _mm256_unpacklo_epi16(V_H_D1, V_H_D2);
__m256i V_H_M2 = _mm256_unpackhi_epi16(V_H_D1, V_H_D2);
__m256i V_H_MAD1 = _mm256_madd_epi16(V_H_M1, filter1);
__m256i V_H_MAD2 = _mm256_madd_epi16(V_H_M2, filter1);
__m256i V_H_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_H_MAD1, rbias), 7);
__m256i V_H_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_H_MAD2, rbias), 7);
opointer[0] = _mm256_packus_epi32(V_H_S1, V_H_S2);
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < 16 / inc; ++i) {
prev = curr;
curr = (curr == 0) ? 1 : 0;
__m256i V_V_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_V_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
__m256i V_V_M1 = _mm256_unpacklo_epi16(V_V_D1, V_V_D2);
__m256i V_V_M2 = _mm256_unpackhi_epi16(V_V_D1, V_V_D2);
__m256i V_V_MAD1 = _mm256_madd_epi16(V_V_M1, filter1);
__m256i V_V_MAD2 = _mm256_madd_epi16(V_V_M2, filter1);
__m256i V_V_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_V_MAD1, rbias), 7);
__m256i V_V_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_V_MAD2, rbias), 7);
opointer[curr] = _mm256_packus_epi32(V_V_S1, V_V_S2);
__m256i V_S_SRC = _mm256_avg_epu16(opointer[prev], opointer[curr]);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
else {
// xoffset==1,2,3,5,6,7 && yoffset==1,2,3,5,6,7
range = output_width / 16;
if (output_height == 8) inc = 2;
if (output_height == 4) inc = 4;
unsigned int nloop = 16 / inc;
for (j = 0; j < range * output_height * inc / 16; j++) {
if (j % (output_height * inc / 16) == 0) {
src_ptr = src_ptr_ref;
src_ptr_ref += 16;
dst_ptr = dst_ptr_ref;
dst_ptr_ref += 16;
__m256i V_H_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_H_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
__m256i V_H_M1 = _mm256_unpacklo_epi16(V_H_D1, V_H_D2);
__m256i V_H_M2 = _mm256_unpackhi_epi16(V_H_D1, V_H_D2);
__m256i V_H_MAD1 = _mm256_madd_epi16(V_H_M1, filter1);
__m256i V_H_MAD2 = _mm256_madd_epi16(V_H_M2, filter1);
__m256i V_H_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_H_MAD1, rbias), 7);
__m256i V_H_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_H_MAD2, rbias), 7);
opointer[0] = _mm256_packus_epi32(V_H_S1, V_H_S2);
curr = 0;
}
__m256i sum1 = _mm256_setzero_si256();
__m256i sse1 = _mm256_setzero_si256();
for (i = 0; i < nloop; ++i) {
prev = curr;
curr = !curr;
__m256i V_V_D1 = _mm256_loadu_si256((
const __m256i *)src_ptr);
__m256i V_V_D2 = _mm256_loadu_si256((
const __m256i *)(src_ptr + 1));
src_ptr += src_pixels_per_line;
__m256i V_V_M1 = _mm256_unpacklo_epi16(V_V_D1, V_V_D2);
__m256i V_V_M2 = _mm256_unpackhi_epi16(V_V_D1, V_V_D2);
__m256i V_V_MAD1 = _mm256_madd_epi16(V_V_M1, filter1);
__m256i V_V_MAD2 = _mm256_madd_epi16(V_V_M2, filter1);
__m256i V_V_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_V_MAD1, rbias), 7);
__m256i V_V_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_V_MAD2, rbias), 7);
opointer[curr] = _mm256_packus_epi32(V_V_S1, V_V_S2);
__m256i V_S_M1 = _mm256_unpacklo_epi16(opointer[prev], opointer[curr]);
__m256i V_S_M2 = _mm256_unpackhi_epi16(opointer[prev], opointer[curr]);
__m256i V_S_MAD1 = _mm256_madd_epi16(V_S_M1, filter2);
__m256i V_S_MAD2 = _mm256_madd_epi16(V_S_M2, filter2);
__m256i V_S_S1 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD1, rbias), 7);
__m256i V_S_S2 =
_mm256_srli_epi32(_mm256_add_epi32(V_S_MAD2, rbias), 7);
__m256i V_S_SRC = _mm256_packus_epi32(V_S_S1, V_S_S2);
__m256i V_D_DST = _mm256_loadu_si256((
const __m256i *)dst_ptr);
dst_ptr += dst_stride;
__m256i V_R_SUB = _mm256_sub_epi16(V_S_SRC, V_D_DST);
__m256i V_R_MAD = _mm256_madd_epi16(V_R_SUB, V_R_SUB);
sum1 = _mm256_add_epi16(sum1, V_R_SUB);
sse1 = _mm256_add_epi32(sse1, V_R_MAD);
}
__m256i v_sum0 = _mm256_madd_epi16(sum1, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, sse1);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, sse1);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
sum_long += _mm_extract_epi32(v_d, 0);
sse_long += _mm_extract_epi32(v_d, 1);
}
rshift = get_msb(output_height) + get_msb(output_width);
}
*sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 4);
int sum = (
int)ROUND_POWER_OF_TWO(sum_long, 2);
int32_t var = *sse - (uint32_t)(((int64_t)sum * sum) >> rshift);
return (var > 0) ? var : 0;
}
static void highbd_calc8x8var_avx2(
const uint16_t *src,
int src_stride,
const uint16_t *ref,
int ref_stride,
uint32_t *sse,
int *sum) {
__m256i v_sum_d = _mm256_setzero_si256();
__m256i v_sse_d = _mm256_setzero_si256();
for (
int i = 0; i < 8; i += 2) {
const __m128i v_p_a0 = _mm_loadu_si128((
const __m128i *)src);
const __m128i v_p_a1 = _mm_loadu_si128((
const __m128i *)(src + src_stride));
const __m128i v_p_b0 = _mm_loadu_si128((
const __m128i *)ref);
const __m128i v_p_b1 = _mm_loadu_si128((
const __m128i *)(ref + ref_stride));
__m256i v_p_a = _mm256_castsi128_si256(v_p_a0);
__m256i v_p_b = _mm256_castsi128_si256(v_p_b0);
v_p_a = _mm256_inserti128_si256(v_p_a, v_p_a1, 1);
v_p_b = _mm256_inserti128_si256(v_p_b, v_p_b1, 1);
const __m256i v_diff = _mm256_sub_epi16(v_p_a, v_p_b);
const __m256i v_sqrdiff = _mm256_madd_epi16(v_diff, v_diff);
v_sum_d = _mm256_add_epi16(v_sum_d, v_diff);
v_sse_d = _mm256_add_epi32(v_sse_d, v_sqrdiff);
src += src_stride * 2;
ref += ref_stride * 2;
}
__m256i v_sum00 = _mm256_cvtepi16_epi32(_mm256_castsi256_si128(v_sum_d));
__m256i v_sum01 = _mm256_cvtepi16_epi32(_mm256_extracti128_si256(v_sum_d, 1));
__m256i v_sum0 = _mm256_add_epi32(v_sum00, v_sum01);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, v_sse_d);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, v_sse_d);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
*sum = _mm_extract_epi32(v_d, 0);
*sse = _mm_extract_epi32(v_d, 1);
}
static void highbd_calc16x16var_avx2(
const uint16_t *src,
int src_stride,
const uint16_t *ref,
int ref_stride,
uint32_t *sse,
int *sum) {
__m256i v_sum_d = _mm256_setzero_si256();
__m256i v_sse_d = _mm256_setzero_si256();
const __m256i one = _mm256_set1_epi16(1);
for (
int i = 0; i < 16; ++i) {
const __m256i v_p_a = _mm256_loadu_si256((
const __m256i *)src);
const __m256i v_p_b = _mm256_loadu_si256((
const __m256i *)ref);
const __m256i v_diff = _mm256_sub_epi16(v_p_a, v_p_b);
const __m256i v_sqrdiff = _mm256_madd_epi16(v_diff, v_diff);
v_sum_d = _mm256_add_epi16(v_sum_d, v_diff);
v_sse_d = _mm256_add_epi32(v_sse_d, v_sqrdiff);
src += src_stride;
ref += ref_stride;
}
__m256i v_sum0 = _mm256_madd_epi16(v_sum_d, one);
__m256i v_d_l = _mm256_unpacklo_epi32(v_sum0, v_sse_d);
__m256i v_d_h = _mm256_unpackhi_epi32(v_sum0, v_sse_d);
__m256i v_d_lh = _mm256_add_epi32(v_d_l, v_d_h);
const __m128i v_d0_d = _mm256_castsi256_si128(v_d_lh);
const __m128i v_d1_d = _mm256_extracti128_si256(v_d_lh, 1);
__m128i v_d = _mm_add_epi32(v_d0_d, v_d1_d);
v_d = _mm_add_epi32(v_d, _mm_srli_si128(v_d, 8));
*sum = _mm_extract_epi32(v_d, 0);
*sse = _mm_extract_epi32(v_d, 1);
}
static void highbd_10_variance_avx2(
const uint16_t *src,
int src_stride,
const uint16_t *ref,
int ref_stride,
int w,
int h, uint32_t *sse,
int *sum,
high_variance_fn_t var_fn,
int block_size) {
int i, j;
uint64_t sse_long = 0;
int32_t sum_long = 0;
for (i = 0; i < h; i += block_size) {
for (j = 0; j < w; j += block_size) {
unsigned int sse0;
int sum0;
var_fn(src + src_stride * i + j, src_stride, ref + ref_stride * i + j,
ref_stride, &sse0, &sum0);
sse_long += sse0;
sum_long += sum0;
}
}
*sum = ROUND_POWER_OF_TWO(sum_long, 2);
*sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 4);
}
#define VAR_FN(w, h, block_size, shift) \
uint32_t aom_highbd_10_variance
##w
##x
##h
##_avx2( \
const uint8_t *src8,
int src_stride,
const uint8_t *ref8, \
int ref_stride, uint32_t *sse) { \
int sum; \
int64_t var; \
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
highbd_10_variance_avx2(src, src_stride, ref, ref_stride, w, h, sse, &sum, \
highbd_calc
##block_size
##x
##block_size
##var_avx2, \
block_size); \
var = (int64_t)(*sse) - (((int64_t)sum * sum) >> shift); \
return (var >= 0) ? (uint32_t)var : 0; \
}
VAR_FN(128, 128, 16, 14)
VAR_FN(128, 64, 16, 13)
VAR_FN(64, 128, 16, 13)
VAR_FN(64, 64, 16, 12)
VAR_FN(64, 32, 16, 11)
VAR_FN(32, 64, 16, 11)
VAR_FN(32, 32, 16, 10)
VAR_FN(32, 16, 16, 9)
VAR_FN(16, 32, 16, 9)
VAR_FN(16, 16, 16, 8)
VAR_FN(16, 8, 8, 7)
VAR_FN(8, 16, 8, 7)
VAR_FN(8, 8, 8, 6)
#if !CONFIG_REALTIME_ONLY
VAR_FN(16, 64, 16, 10)
VAR_FN(32, 8, 8, 8)
VAR_FN(64, 16, 16, 10)
VAR_FN(8, 32, 8, 8)
#endif // !CONFIG_REALTIME_ONLY
#undef VAR_FN
unsigned int aom_highbd_10_mse16x16_avx2(
const uint8_t *src8,
int src_stride,
const uint8_t *ref8,
int ref_stride,
unsigned int *sse) {
int sum;
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
highbd_10_variance_avx2(src, src_stride, ref, ref_stride, 16, 16, sse, &sum,
highbd_calc16x16var_avx2, 16);
return *sse;
}
#define SSE2_HEIGHT(H) \
uint32_t aom_highbd_10_sub_pixel_variance8x
##H
##_sse2( \
const uint8_t *src8,
int src_stride,
int x_offset,
int y_offset, \
const uint8_t *dst8,
int dst_stride, uint32_t *sse_ptr);
SSE2_HEIGHT(8)
SSE2_HEIGHT(16)
#undef SSE2_HEIGHT
#define HIGHBD_SUBPIX_VAR(W, H) \
uint32_t aom_highbd_10_sub_pixel_variance
##W
##x
##H
##_avx2( \
const uint8_t *src,
int src_stride,
int xoffset,
int yoffset, \
const uint8_t *dst,
int dst_stride, uint32_t *sse) { \
if (W == 8 && H == 16) \
return aom_highbd_10_sub_pixel_variance8x16_sse2( \
src, src_stride, xoffset, yoffset, dst, dst_stride, sse); \
else if (W == 8 && H == 8) \
return aom_highbd_10_sub_pixel_variance8x8_sse2( \
src, src_stride, xoffset, yoffset, dst, dst_stride, sse); \
else \
return aom_highbd_var_filter_block2d_bil_avx2( \
src, src_stride, 1, H, W, xoffset, yoffset, dst, dst_stride, sse); \
}
HIGHBD_SUBPIX_VAR(128, 128)
HIGHBD_SUBPIX_VAR(128, 64)
HIGHBD_SUBPIX_VAR(64, 128)
HIGHBD_SUBPIX_VAR(64, 64)
HIGHBD_SUBPIX_VAR(64, 32)
HIGHBD_SUBPIX_VAR(32, 64)
HIGHBD_SUBPIX_VAR(32, 32)
HIGHBD_SUBPIX_VAR(32, 16)
HIGHBD_SUBPIX_VAR(16, 32)
HIGHBD_SUBPIX_VAR(16, 16)
HIGHBD_SUBPIX_VAR(16, 8)
HIGHBD_SUBPIX_VAR(8, 16)
HIGHBD_SUBPIX_VAR(8, 8)
#undef HIGHBD_SUBPIX_VAR
static uint64_t mse_4xh_16bit_highbd_avx2(uint16_t *dst,
int dstride,
uint16_t *src,
int sstride,
int h) {
uint64_t sum = 0;
__m128i reg0_4x16, reg1_4x16, reg2_4x16, reg3_4x16;
__m256i src0_8x16, src1_8x16, src_16x16;
__m256i dst0_8x16, dst1_8x16, dst_16x16;
__m256i res0_4x64, res1_4x64, res2_4x64, res3_4x64;
__m256i sub_result;
const __m256i zeros = _mm256_broadcastsi128_si256(_mm_setzero_si128());
__m256i square_result = _mm256_broadcastsi128_si256(_mm_setzero_si128());
for (
int i = 0; i < h; i += 4) {
reg0_4x16 = _mm_loadl_epi64((__m128i
const *)(&dst[(i + 0) * dstride]));
reg1_4x16 = _mm_loadl_epi64((__m128i
const *)(&dst[(i + 1) * dstride]));
reg2_4x16 = _mm_loadl_epi64((__m128i
const *)(&dst[(i + 2) * dstride]));
reg3_4x16 = _mm_loadl_epi64((__m128i
const *)(&dst[(i + 3) * dstride]));
dst0_8x16 =
_mm256_castsi128_si256(_mm_unpacklo_epi64(reg0_4x16, reg1_4x16));
dst1_8x16 =
_mm256_castsi128_si256(_mm_unpacklo_epi64(reg2_4x16, reg3_4x16));
dst_16x16 = _mm256_permute2x128_si256(dst0_8x16, dst1_8x16, 0x20);
reg0_4x16 = _mm_loadl_epi64((__m128i
const *)(&src[(i + 0) * sstride]));
reg1_4x16 = _mm_loadl_epi64((__m128i
const *)(&src[(i + 1) * sstride]));
reg2_4x16 = _mm_loadl_epi64((__m128i
const *)(&src[(i + 2) * sstride]));
reg3_4x16 = _mm_loadl_epi64((__m128i
const *)(&src[(i + 3) * sstride]));
src0_8x16 =
_mm256_castsi128_si256(_mm_unpacklo_epi64(reg0_4x16, reg1_4x16));
src1_8x16 =
_mm256_castsi128_si256(_mm_unpacklo_epi64(reg2_4x16, reg3_4x16));
src_16x16 = _mm256_permute2x128_si256(src0_8x16, src1_8x16, 0x20);
sub_result = _mm256_abs_epi16(_mm256_sub_epi16(src_16x16, dst_16x16));
src_16x16 = _mm256_unpacklo_epi16(sub_result, zeros);
dst_16x16 = _mm256_unpackhi_epi16(sub_result, zeros);
src_16x16 = _mm256_madd_epi16(src_16x16, src_16x16);
dst_16x16 = _mm256_madd_epi16(dst_16x16, dst_16x16);
res0_4x64 = _mm256_unpacklo_epi32(src_16x16, zeros);
res1_4x64 = _mm256_unpackhi_epi32(src_16x16, zeros);
res2_4x64 = _mm256_unpacklo_epi32(dst_16x16, zeros);
res3_4x64 = _mm256_unpackhi_epi32(dst_16x16, zeros);
square_result = _mm256_add_epi64(
square_result,
_mm256_add_epi64(
_mm256_add_epi64(_mm256_add_epi64(res0_4x64, res1_4x64), res2_4x64),
res3_4x64));
}
const __m128i sum_2x64 =
_mm_add_epi64(_mm256_castsi256_si128(square_result),
_mm256_extracti128_si256(square_result, 1));
const __m128i sum_1x64 = _mm_add_epi64(sum_2x64, _mm_srli_si128(sum_2x64, 8));
xx_storel_64(&sum, sum_1x64);
return sum;
}
static uint64_t mse_8xh_16bit_highbd_avx2(uint16_t *dst,
int dstride,
uint16_t *src,
int sstride,
int h) {
uint64_t sum = 0;
__m256i src0_8x16, src1_8x16, src_16x16;
__m256i dst0_8x16, dst1_8x16, dst_16x16;
__m256i res0_4x64, res1_4x64, res2_4x64, res3_4x64;
__m256i sub_result;
const __m256i zeros = _mm256_broadcastsi128_si256(_mm_setzero_si128());
__m256i square_result = _mm256_broadcastsi128_si256(_mm_setzero_si128());
for (
int i = 0; i < h; i += 2) {
dst0_8x16 =
_mm256_castsi128_si256(_mm_loadu_si128((__m128i *)&dst[i * dstride]));
dst1_8x16 = _mm256_castsi128_si256(
_mm_loadu_si128((__m128i *)&dst[(i + 1) * dstride]));
dst_16x16 = _mm256_permute2x128_si256(dst0_8x16, dst1_8x16, 0x20);
src0_8x16 =
_mm256_castsi128_si256(_mm_loadu_si128((__m128i *)&src[i * sstride]));
src1_8x16 = _mm256_castsi128_si256(
_mm_loadu_si128((__m128i *)&src[(i + 1) * sstride]));
src_16x16 = _mm256_permute2x128_si256(src0_8x16, src1_8x16, 0x20);
sub_result = _mm256_abs_epi16(_mm256_sub_epi16(src_16x16, dst_16x16));
src_16x16 = _mm256_unpacklo_epi16(sub_result, zeros);
dst_16x16 = _mm256_unpackhi_epi16(sub_result, zeros);
src_16x16 = _mm256_madd_epi16(src_16x16, src_16x16);
dst_16x16 = _mm256_madd_epi16(dst_16x16, dst_16x16);
res0_4x64 = _mm256_unpacklo_epi32(src_16x16, zeros);
res1_4x64 = _mm256_unpackhi_epi32(src_16x16, zeros);
res2_4x64 = _mm256_unpacklo_epi32(dst_16x16, zeros);
res3_4x64 = _mm256_unpackhi_epi32(dst_16x16, zeros);
square_result = _mm256_add_epi64(
square_result,
_mm256_add_epi64(
_mm256_add_epi64(_mm256_add_epi64(res0_4x64, res1_4x64), res2_4x64),
res3_4x64));
}
const __m128i sum_2x64 =
_mm_add_epi64(_mm256_castsi256_si128(square_result),
_mm256_extracti128_si256(square_result, 1));
const __m128i sum_1x64 = _mm_add_epi64(sum_2x64, _mm_srli_si128(sum_2x64, 8));
xx_storel_64(&sum, sum_1x64);
return sum;
}
uint64_t aom_mse_wxh_16bit_highbd_avx2(uint16_t *dst,
int dstride,
uint16_t *src,
int sstride,
int w,
int h) {
assert((w == 8 || w == 4) && (h == 8 || h == 4) &&
"w=8/4 and h=8/4 must satisfy");
switch (w) {
case 4:
return mse_4xh_16bit_highbd_avx2(dst, dstride, src, sstride, h);
case 8:
return mse_8xh_16bit_highbd_avx2(dst, dstride, src, sstride, h);
default: assert(0 &&
"unsupported width");
return -1;
}
}
