mirror of
https://github.com/ultravideo/uvg266.git
synced 2024-11-23 18:14:06 +00:00
25.6 working optimation, ~50% faster than original
This commit is contained in:
Reima Hyvönen
parent
9dfd72628a
commit
17babfffa4
37
src/inter.c
37
src/inter.c
|
|
@ -464,46 +464,21 @@ void kvz_inter_recon_bipred(const encoder_state_t * const state,
|
|||
hi_prec_buf_t* high_precision_rec1 = 0;
|
||||
if (hi_prec_chroma_rec0) high_precision_rec0 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
|
||||
if (hi_prec_chroma_rec1) high_precision_rec1 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
|
||||
|
||||
|
||||
//Reconstruct both predictors
|
||||
inter_recon_unipred(state, ref1, xpos, ypos, width, height, mv_param[0], lcu, high_precision_rec0);
|
||||
if (!hi_prec_luma_rec0){
|
||||
memcpy(temp_lcu_y, lcu->rec.y, sizeof(kvz_pixel) * 64 * 64);
|
||||
memcpy(temp_lcu_y, lcu->rec.y, sizeof(kvz_pixel) * 64 * 64); // copy to temp_lcu_y
|
||||
}
|
||||
if (!hi_prec_chroma_rec0){
|
||||
memcpy(temp_lcu_u, lcu->rec.u, sizeof(kvz_pixel) * 32 * 32);
|
||||
memcpy(temp_lcu_v, lcu->rec.v, sizeof(kvz_pixel) * 32 * 32);
|
||||
memcpy(temp_lcu_u, lcu->rec.u, sizeof(kvz_pixel) * 32 * 32); // copy to temp_lcu_u
|
||||
memcpy(temp_lcu_v, lcu->rec.v, sizeof(kvz_pixel) * 32 * 32); // copy to temp_lcu_v
|
||||
}
|
||||
inter_recon_unipred(state, ref2, xpos, ypos, width, height, mv_param[1], lcu, high_precision_rec1);
|
||||
|
||||
kvz_inter_recon_bipred_test(hi_prec_luma_rec0, hi_prec_luma_rec1, hi_prec_chroma_rec0, hi_prec_chroma_rec1, height, width, ypos, xpos, high_precision_rec0, high_precision_rec1, lcu);
|
||||
|
||||
/*
|
||||
// After reconstruction, merge the predictors by taking an average of each pixel
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (temp_x = 0; temp_x < width; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
int16_t sample0_y = (hi_prec_luma_rec0 ? high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_y = (hi_prec_luma_rec1 ? high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_y + sample1_y + offset) >> shift);
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
for (temp_y = 0; temp_y < height >> 1; ++temp_y) {
|
||||
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
for (temp_x = 0; temp_x < width >> 1; ++temp_x) {
|
||||
int x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
int16_t sample0_u = (hi_prec_chroma_rec0 ? high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_u = (hi_prec_chroma_rec1 ? high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_u + sample1_u + offset) >> shift);
|
||||
|
||||
int16_t sample0_v = (hi_prec_chroma_rec0 ? high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_v = (hi_prec_chroma_rec1 ? high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_v + sample1_v + offset) >> shift);
|
||||
}
|
||||
}
|
||||
*/
|
||||
kvz_inter_recon_bipred_generic(hi_prec_luma_rec0, hi_prec_luma_rec1, hi_prec_chroma_rec0, hi_prec_chroma_rec1, height, width, ypos, xpos, high_precision_rec0, high_precision_rec1, lcu, temp_lcu_y, temp_lcu_u, temp_lcu_v);
|
||||
|
||||
if (high_precision_rec0 != 0) kvz_hi_prec_buf_t_free(high_precision_rec0);
|
||||
if (high_precision_rec1 != 0) kvz_hi_prec_buf_t_free(high_precision_rec1);
|
||||
|
|
|
|||
|
|
@ -25,6 +25,9 @@
|
|||
|
||||
#if COMPILE_INTEL_AVX2
|
||||
#include <immintrin.h>
|
||||
#include <emmintrin.h>
|
||||
#include <mmintrin.h>
|
||||
#include <xmmintrin.h>
|
||||
#include <string.h>
|
||||
|
||||
#include "kvazaar.h"
|
||||
|
|
@ -714,6 +717,403 @@ static unsigned pixels_calc_ssd_avx2(const kvz_pixel *const ref, const kvz_pixel
|
|||
}
|
||||
}
|
||||
|
||||
|
||||
static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
||||
const int hi_prec_luma_rec1,
|
||||
const int hi_prec_chroma_rec0,
|
||||
const int hi_prec_chroma_rec1,
|
||||
const int height,
|
||||
const int width,
|
||||
const int ypos,
|
||||
const int xpos,
|
||||
const hi_prec_buf_t*high_precision_rec0,
|
||||
const hi_prec_buf_t*high_precision_rec1,
|
||||
lcu_t* lcu,
|
||||
kvz_pixel temp_lcu_y[LCU_WIDTH*LCU_WIDTH],
|
||||
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C],
|
||||
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C]) {
|
||||
|
||||
int y_in_lcu;
|
||||
int x_in_lcu;
|
||||
|
||||
int shift = 15 - KVZ_BIT_DEPTH;
|
||||
int offset = 1 << (shift-1);
|
||||
int shift_left = 14 - KVZ_BIT_DEPTH;
|
||||
|
||||
__m256i offset_epi32 = _mm256_set1_epi32(offset);
|
||||
__m256i temp_epi32;
|
||||
__m256i temp_epi16;
|
||||
|
||||
__m256i temp_epi8;
|
||||
__m256i temp_zeros_256 = _mm256_setzero_si256();
|
||||
__m256i temp_y_epi32, temp_u_epi32, temp_v_epi32;
|
||||
|
||||
__m256i temp_epi32_u = _mm256_setzero_si256();
|
||||
__m256i temp_epi32_v = _mm256_setzero_si256();
|
||||
|
||||
__m256i sample0_epi32;
|
||||
__m256i sample1_epi32;
|
||||
|
||||
__m256i temp_epi16_u = _mm256_setzero_si256();
|
||||
__m256i temp_epi16_v = _mm256_setzero_si256();
|
||||
|
||||
__m256i final_epi8_256 = _mm256_setzero_si256();
|
||||
|
||||
__m128i offset_4 = _mm_set1_epi32(offset);
|
||||
__m128i sample_epi32;
|
||||
__m128i sample0_y_epi32, sample1_y_epi32, sample0_u_epi32, sample1_u_epi32, sample0_v_epi32, sample1_v_epi32;
|
||||
__m128i temp_zeros_128 = _mm_setzero_si128();
|
||||
|
||||
__m256i idx = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
|
||||
|
||||
__m128i final_epi8_128;
|
||||
|
||||
|
||||
switch (width)
|
||||
{
|
||||
|
||||
case 4:
|
||||
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos) & ((LCU_WIDTH)-1));
|
||||
|
||||
|
||||
|
||||
sample0_y_epi32 = hi_prec_luma_rec0 ? _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i*)&(high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm_slli_epi32(_mm_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), shift_left);
|
||||
|
||||
sample1_y_epi32 = hi_prec_luma_rec1 ? _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i*)&(high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm_slli_epi32(_mm_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), shift_left);
|
||||
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
sample_epi32 = _mm_add_epi32(_mm_add_epi32(sample0_y_epi32, sample1_y_epi32), offset_4);
|
||||
sample_epi32 = _mm_srai_epi32(sample_epi32, shift);
|
||||
|
||||
final_epi8_128 = _mm_packus_epi16(_mm_packus_epi32(sample_epi32, temp_zeros_128), temp_zeros_128);
|
||||
|
||||
int8_t*temp_int_y = (int8_t*)&final_epi8_128;
|
||||
|
||||
for (int i = 0; i < 4; i++) {
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu + i] = temp_int_y[i];
|
||||
}
|
||||
|
||||
|
||||
if (temp_y < height >> 1) {
|
||||
y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
x_in_lcu = (((xpos >> 1)) & (LCU_WIDTH_C - 1));
|
||||
|
||||
sample0_u_epi32 = hi_prec_chroma_rec0 ? _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i*)&(high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm_slli_epi32(_mm_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))), shift_left);
|
||||
|
||||
sample1_u_epi32 = hi_prec_chroma_rec1 ? _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i*)&(high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm_slli_epi32(_mm_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
sample_epi32 = _mm_add_epi32(_mm_add_epi32(sample0_u_epi32, sample1_u_epi32), offset_4);
|
||||
sample_epi32 = _mm_srai_epi32(sample_epi32, shift);
|
||||
|
||||
__m128i temp_u = _mm_packus_epi16(_mm_packus_epi32(sample_epi32, temp_zeros_128), temp_zeros_128);
|
||||
int8_t*temp_int_u = (int8_t*)&temp_u;
|
||||
|
||||
sample0_v_epi32 = hi_prec_chroma_rec0 ? _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i*)&(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm_slli_epi32(_mm_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))), shift_left);
|
||||
|
||||
sample1_v_epi32 = hi_prec_chroma_rec1 ? _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i*)&(high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm_slli_epi32(_mm_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
sample_epi32 = _mm_add_epi32(_mm_add_epi32(sample0_v_epi32, sample1_v_epi32), offset_4);
|
||||
sample_epi32 = _mm_srai_epi32(sample_epi32, shift);
|
||||
|
||||
__m128i temp_v = _mm_packus_epi16(_mm_packus_epi32(sample_epi32, temp_zeros_128), temp_zeros_128);
|
||||
int8_t*temp_int_v = (int8_t*)&temp_v;
|
||||
|
||||
for (int i = 0; i < 2; i++) {
|
||||
lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu + i] = temp_int_u[i];
|
||||
lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu + i] = temp_int_v[i];
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
break;
|
||||
|
||||
default:
|
||||
|
||||
int start_point = 0;
|
||||
int start_point_uv = 0;
|
||||
|
||||
for (int temp_y = 0; temp_y < height; temp_y += 1) {
|
||||
temp_epi32 = _mm256_setzero_si256();
|
||||
temp_epi16 = _mm256_setzero_si256();
|
||||
int temp = 0;
|
||||
int temp_uv = 0;
|
||||
|
||||
|
||||
for (int temp_x = 0; temp_x < width; temp_x += 8) {
|
||||
|
||||
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
|
||||
// Load total of 8 elements from memory to vector and convert all to 32-bit
|
||||
sample0_epi32 = hi_prec_luma_rec0 ? (_mm256_cvtepu16_epi32(_mm_load_si128((__m128i*) &(high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu])))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), shift_left);
|
||||
|
||||
sample1_epi32 = hi_prec_luma_rec1 ? (_mm256_cvtepu16_epi32(_mm_load_si128((__m128i*) &(high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu])))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepu8_epi32((_mm_loadl_epi64((__m128i*) &(lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_y_epi32 = _mm256_add_epi32(sample0_epi32, sample1_epi32);
|
||||
temp_y_epi32 = _mm256_add_epi32(temp_y_epi32, offset_epi32);
|
||||
temp_y_epi32 = _mm256_srai_epi32(temp_y_epi32, shift);
|
||||
|
||||
switch (width)
|
||||
{
|
||||
case 8:
|
||||
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_y_epi32, temp_zeros_256), temp_zeros_256);
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
final_epi8_128 = _mm_loadu_si128((__m128i*)&temp_epi8);
|
||||
|
||||
// Store 64-bits from vector to memory
|
||||
_mm_storel_epi64((__m128i*)&(lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu]), final_epi8_128);
|
||||
|
||||
break;
|
||||
|
||||
case 16:
|
||||
|
||||
if (temp == 0) {
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
|
||||
// Save starting point to memory
|
||||
start_point = (y_in_lcu)* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
|
||||
else if (temp == 1) {
|
||||
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_epi32, temp_y_epi32), temp_zeros_256);
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
// Fill 128 bit vector with packed data and store it to memory
|
||||
__m128i final_epi8_16 = _mm_loadu_si128((__m128i*)&temp_epi8);
|
||||
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.y[start_point]), final_epi8_16);
|
||||
|
||||
temp = 0;
|
||||
}
|
||||
|
||||
|
||||
break;
|
||||
|
||||
default:
|
||||
if (temp == 0) {
|
||||
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
|
||||
start_point = y_in_lcu* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
|
||||
else if (temp == 1) {
|
||||
|
||||
// Convert packed 16-bit integers to packed 8-bit integers and store result to vector
|
||||
temp_epi16 = _mm256_packus_epi32(temp_epi32, temp_y_epi32);
|
||||
temp++;
|
||||
}
|
||||
|
||||
else if (temp == 2) {
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
}
|
||||
|
||||
else {
|
||||
|
||||
// Convert packed 32-bit integers to packed 8-bit integers and store result to vector
|
||||
temp_epi8 = _mm256_packus_epi16(temp_epi16, _mm256_packus_epi32(temp_epi32, temp_y_epi32));
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
final_epi8_256 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.y[start_point]), final_epi8_256);
|
||||
temp = 0;
|
||||
}
|
||||
}
|
||||
|
||||
if (temp_x < width >> 1 && temp_y < height >> 1) {
|
||||
y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
|
||||
sample0_epi32 = hi_prec_chroma_rec0 ? (_mm256_cvtepu16_epi32(_mm_load_si128((__m128i*) &(high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi32 = hi_prec_chroma_rec1 ? (_mm256_cvtepu16_epi32(_mm_load_si128((__m128i*) &(high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi32 = _mm256_add_epi32(sample0_epi32, sample1_epi32);
|
||||
temp_u_epi32 = _mm256_add_epi32(temp_u_epi32, offset_epi32);
|
||||
temp_u_epi32 = _mm256_srai_epi16(temp_u_epi32, shift);
|
||||
|
||||
sample0_epi32 = hi_prec_chroma_rec0 ? (_mm256_cvtepu16_epi32(_mm_load_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi32 = hi_prec_chroma_rec1 ? (_mm256_cvtepu16_epi32(_mm_load_si128((__m128i*) &(high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi32 = _mm256_add_epi32(sample0_epi32, sample1_epi32);
|
||||
temp_v_epi32 = _mm256_add_epi32(temp_v_epi32, offset_epi32);
|
||||
temp_v_epi32 = _mm256_srai_epi32(temp_v_epi32, shift);
|
||||
|
||||
|
||||
switch (width) {
|
||||
|
||||
case 8:
|
||||
|
||||
__m256i temp_epi8u = _mm256_packus_epi16(_mm256_packus_epi32(temp_u_epi32, temp_zeros_256), temp_zeros_256);
|
||||
|
||||
int8_t *temp_int_8_u = (int8_t*)&temp_epi8u;
|
||||
|
||||
__m256i temp_epi8v = _mm256_packus_epi16(_mm256_packus_epi32(temp_v_epi32, temp_zeros_256), temp_zeros_256);
|
||||
|
||||
int8_t *temp_int_8_v = (int8_t*)&temp_epi8v;
|
||||
|
||||
|
||||
|
||||
for (int i = 0; i < 4; i++) {
|
||||
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + i] = temp_int_8_u[i];
|
||||
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + i] = temp_int_8_v[i];
|
||||
}
|
||||
|
||||
break;
|
||||
|
||||
case 16:
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_u_epi32, temp_zeros_256), temp_zeros_256);
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
final_epi8_128 = _mm_loadu_si128((__m128i*)&temp_epi8);
|
||||
|
||||
// Store 64-bit integer into memory
|
||||
_mm_storel_epi64((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), final_epi8_128);
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_v_epi32, temp_zeros_256), temp_zeros_256);
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
final_epi8_128 = _mm_loadu_si128((__m128i*)&temp_epi8);
|
||||
|
||||
// Store 64-bit integer into memory
|
||||
_mm_storel_epi64((__m128i*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), final_epi8_128);
|
||||
|
||||
break;
|
||||
|
||||
case 32:
|
||||
|
||||
if (temp_uv == 0) {
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
|
||||
// Save starting point to memory
|
||||
start_point_uv = (y_in_lcu)* LCU_WIDTH_C + x_in_lcu;
|
||||
|
||||
temp_uv++;
|
||||
}
|
||||
|
||||
else{
|
||||
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
__m256i temp_epi8_u = _mm256_packus_epi16(_mm256_packus_epi32(temp_epi32_u, temp_u_epi32), temp_zeros_256);
|
||||
__m256i temp_epi8_v = _mm256_packus_epi16(_mm256_packus_epi32(temp_epi32_v, temp_v_epi32), temp_zeros_256);
|
||||
|
||||
temp_epi8_u = _mm256_permutevar8x32_epi32(temp_epi8_u, idx);
|
||||
temp_epi8_v = _mm256_permutevar8x32_epi32(temp_epi8_v, idx);
|
||||
|
||||
// Fill 128 bit vector with packed data and store it to memory
|
||||
__m128i final_epi8_u = _mm_loadu_si128((__m128i*)&temp_epi8_u);
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.u[start_point_uv]), final_epi8_u);
|
||||
|
||||
// Fill 128 bit vector with packed data and store it to memory
|
||||
__m128i final_epi8_v = _mm_loadu_si128((__m128i*)&temp_epi8_v);
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.v[start_point_uv]), final_epi8_v);
|
||||
|
||||
temp_uv = 0;
|
||||
}
|
||||
break;
|
||||
|
||||
default:
|
||||
if (temp_uv == 0) {
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
|
||||
// Save starting point to memory
|
||||
start_point_uv = (y_in_lcu)* LCU_WIDTH_C + x_in_lcu;
|
||||
|
||||
temp_uv++;
|
||||
}
|
||||
|
||||
else if (temp_uv == 1) {
|
||||
|
||||
// Convert packed 16-bit integers to packed 8-bit integers and store result to vector
|
||||
temp_epi16_u = _mm256_packus_epi32(temp_epi32_u, temp_u_epi32);
|
||||
temp_epi16_v = _mm256_packus_epi32(temp_epi32_v, temp_v_epi32);
|
||||
temp_uv++;
|
||||
}
|
||||
|
||||
else if (temp_uv == 2) {
|
||||
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
temp_uv++;
|
||||
}
|
||||
|
||||
else {
|
||||
// Pack 32 bit to 8 bit
|
||||
__m256i temp_epi8_u = _mm256_packus_epi16(temp_epi16_u, _mm256_packus_epi32(temp_epi32_u, temp_u_epi32));
|
||||
__m256i temp_epi8_v = _mm256_packus_epi16(temp_epi16_v, _mm256_packus_epi32(temp_epi32_v, temp_v_epi32));
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
final_epi8_256 = _mm256_permutevar8x32_epi32(temp_epi8_u, idx);
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.u[start_point_uv]), final_epi8_256);
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
final_epi8_256 = _mm256_permutevar8x32_epi32(temp_epi8_v, idx);
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), final_epi8_256);
|
||||
|
||||
temp_uv = 0;
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
#endif //COMPILE_INTEL_AVX2
|
||||
|
||||
|
||||
|
|
@ -746,6 +1146,8 @@ int kvz_strategy_register_picture_avx2(void* opaque, uint8_t bitdepth)
|
|||
success &= kvz_strategyselector_register(opaque, "satd_any_size_quad", "avx2", 40, &satd_any_size_quad_avx2);
|
||||
|
||||
success &= kvz_strategyselector_register(opaque, "pixels_calc_ssd", "avx2", 40, &pixels_calc_ssd_avx2);
|
||||
success &= kvz_strategyselector_register(opaque, "inter_recon_bipred", "avx2", 40, &inter_recon_bipred_avx2);
|
||||
|
||||
}
|
||||
#endif
|
||||
return success;
|
||||
|
|
|
|||
|
|
@ -539,53 +539,79 @@ static void inter_recon_bipred_generic(const int hi_prec_luma_rec0,
|
|||
const int hi_prec_luma_rec1,
|
||||
const int hi_prec_chroma_rec0,
|
||||
const int hi_prec_chroma_rec1,
|
||||
int height,
|
||||
int width,
|
||||
int ypos,
|
||||
int xpos,
|
||||
int32_t height,
|
||||
int32_t width,
|
||||
int32_t ypos,
|
||||
int32_t xpos,
|
||||
const hi_prec_buf_t*high_precision_rec0,
|
||||
const hi_prec_buf_t*high_precision_rec1,
|
||||
lcu_t* lcu) {
|
||||
lcu_t* lcu,
|
||||
kvz_pixel temp_lcu_y[LCU_WIDTH*LCU_WIDTH],
|
||||
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C],
|
||||
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C]) {
|
||||
|
||||
kvz_pixel temp_lcu_y[LCU_WIDTH*LCU_WIDTH];
|
||||
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C];
|
||||
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C];
|
||||
|
||||
int temp_x, temp_y;
|
||||
int shift = 15 - KVZ_BIT_DEPTH;
|
||||
int offset = 1 << (shift - 1);
|
||||
int y_in_lcu1;
|
||||
int y_in_lcu2;
|
||||
|
||||
int y_in_lcu;
|
||||
int x_in_lcu;
|
||||
|
||||
|
||||
//After reconstruction, merge the predictors by taking an average of each pixel
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
y_in_lcu1 = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
|
||||
for (temp_x = 0; temp_x < width; ++temp_x) {
|
||||
|
||||
int x_in_lcu1 = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
for (int temp_x = 0; temp_x < width; ++temp_x) {
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
|
||||
int16_t sample0_y = (hi_prec_luma_rec0 ? high_precision_rec0->y[y_in_lcu1 * LCU_WIDTH + x_in_lcu1] : (temp_lcu_y[y_in_lcu1 * LCU_WIDTH + x_in_lcu1] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_y = (hi_prec_luma_rec1 ? high_precision_rec1->y[y_in_lcu1 * LCU_WIDTH + x_in_lcu1] : (lcu->rec.y[y_in_lcu1 * LCU_WIDTH + x_in_lcu1] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.y[y_in_lcu1 * LCU_WIDTH + x_in_lcu1] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_y + sample1_y + offset) >> shift);
|
||||
int16_t sample0_y = (hi_prec_luma_rec0 ? high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_y = (hi_prec_luma_rec1 ? high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_y + sample1_y + offset) >> shift);
|
||||
|
||||
if (temp_x < width >> 1 && temp_y < height >> 1) {
|
||||
|
||||
y_in_lcu2 = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
int x_in_lcu2 = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
|
||||
int16_t sample0_u = (hi_prec_chroma_rec0 ? high_precision_rec0->u[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] : (temp_lcu_u[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_u = (hi_prec_chroma_rec1 ? high_precision_rec1->u[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] : (lcu->rec.u[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.u[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_u + sample1_u + offset) >> shift);
|
||||
int16_t sample0_u = (hi_prec_chroma_rec0 ? high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_u = (hi_prec_chroma_rec1 ? high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_u + sample1_u + offset) >> shift);
|
||||
|
||||
int16_t sample0_v = (hi_prec_chroma_rec0 ? high_precision_rec0->v[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] : (temp_lcu_v[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_v = (hi_prec_chroma_rec1 ? high_precision_rec1->v[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] : (lcu->rec.v[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.v[y_in_lcu2 * LCU_WIDTH_C + x_in_lcu2] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_v + sample1_v + offset) >> shift);
|
||||
int16_t sample0_v = (hi_prec_chroma_rec0 ? high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_v = (hi_prec_chroma_rec1 ? high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_v + sample1_v + offset) >> shift);
|
||||
}
|
||||
}
|
||||
}
|
||||
/*
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (int temp_x = 0; temp_x < width; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
int16_t sample0_y = (hi_prec_luma_rec0 ? high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_y = (hi_prec_luma_rec1 ? high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_y + sample1_y + offset) >> shift);
|
||||
}
|
||||
|
||||
}
|
||||
for (int temp_y = 0; temp_y < height >> 1; ++temp_y) {
|
||||
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
for (int temp_x = 0; temp_x < width >> 1; ++temp_x) {
|
||||
int x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
int16_t sample0_u = (hi_prec_chroma_rec0 ? high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_u = (hi_prec_chroma_rec1 ? high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_u + sample1_u + offset) >> shift);
|
||||
|
||||
int16_t sample0_v = (hi_prec_chroma_rec0 ? high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_v = (hi_prec_chroma_rec1 ? high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_v + sample1_v + offset) >> shift);
|
||||
|
||||
|
||||
|
||||
}
|
||||
}*/
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -61,7 +61,7 @@ cost_pixel_any_size_multi_func * kvz_satd_any_size_quad = 0;
|
|||
|
||||
pixels_calc_ssd_func * kvz_pixels_calc_ssd = 0;
|
||||
|
||||
inter_recon_bipred_func * kvz_inter_recon_bipred_test = 0;
|
||||
inter_recon_bipred_func * kvz_inter_recon_bipred_generic = 0;
|
||||
|
||||
|
||||
int kvz_strategy_register_picture(void* opaque, uint8_t bitdepth) {
|
||||
|
|
|
|||
|
|
@ -124,7 +124,10 @@ typedef void (inter_recon_bipred_func)(const int hi_prec_luma_rec0,
|
|||
int xpos,
|
||||
const hi_prec_buf_t*high_precision_rec0,
|
||||
const hi_prec_buf_t*high_precision_rec1,
|
||||
lcu_t* lcu);
|
||||
lcu_t* lcu,
|
||||
kvz_pixel temp_lcu_y[LCU_WIDTH*LCU_WIDTH],
|
||||
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C],
|
||||
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C]);
|
||||
|
||||
|
||||
|
||||
|
|
@ -160,7 +163,7 @@ extern cost_pixel_any_size_multi_func *kvz_satd_any_size_quad;
|
|||
|
||||
extern pixels_calc_ssd_func *kvz_pixels_calc_ssd;
|
||||
|
||||
extern inter_recon_bipred_func * kvz_inter_recon_bipred_test;
|
||||
extern inter_recon_bipred_func * kvz_inter_recon_bipred_generic;
|
||||
|
||||
int kvz_strategy_register_picture(void* opaque, uint8_t bitdepth);
|
||||
cost_pixel_nxn_func * kvz_pixels_get_satd_func(unsigned n);
|
||||
|
|
@ -192,7 +195,8 @@ cost_pixel_nxn_multi_func * kvz_pixels_get_sad_dual_func(unsigned n);
|
|||
{"satd_32x32_dual", (void**) &kvz_satd_32x32_dual}, \
|
||||
{"satd_64x64_dual", (void**) &kvz_satd_64x64_dual}, \
|
||||
{"satd_any_size_quad", (void**) &kvz_satd_any_size_quad}, \
|
||||
{"inter_recon_bipred", (void**) &kvz_inter_recon_bipred_test}, \
|
||||
{"pixels_calc_ssd", (void**) &kvz_pixels_calc_ssd}, \
|
||||
{"inter_recon_bipred", (void**) &kvz_inter_recon_bipred_generic}, \
|
||||
|
||||
|
||||
|
||||
|
|
|
|||
1456
src/strategies/x86_asm/x86inc.asm
Normal file
1456
src/strategies/x86_asm/x86inc.asm
Normal file
File diff suppressed because it is too large
Load diff
218
tests/bipred_generic_tests.c
Normal file
218
tests/bipred_generic_tests.c
Normal file
|
|
@ -0,0 +1,218 @@
|
|||
/*****************************************************************************
|
||||
* This file is part of Kvazaar HEVC encoder.
|
||||
*
|
||||
* Copyright (C) 2017 Tampere University of Technology and others (see
|
||||
* COPYING file).
|
||||
*
|
||||
* Kvazaar is free software: you can redistribute it and/or modify
|
||||
* it under the terms of the GNU Lesser General Public License version 2.1 as
|
||||
* published by the Free Software Foundation.
|
||||
*
|
||||
* Kvazaar is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
||||
* Lesser General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with Kvazaar. If not, see <http://www.gnu.org/licenses/>.
|
||||
****************************************************************************/
|
||||
|
||||
#include "greatest/greatest.h"
|
||||
|
||||
#include "test_strategies.h"
|
||||
#include "strategies/generic/picture-generic.h"
|
||||
#include <string.h>
|
||||
#include <stdlib.h>
|
||||
|
||||
|
||||
static lcu_t expected_test_result;
|
||||
static lcu_t result;
|
||||
|
||||
static lcu_t lcu1;
|
||||
|
||||
int temp1, temp2, temp3, temp4;
|
||||
|
||||
int16_t mv_param[2][2] = { { 7,7 },{ 7,7 } };
|
||||
int width = 4;
|
||||
int height = 4;
|
||||
int xpos = 0;
|
||||
int ypos = 0;
|
||||
|
||||
kvz_pixel temp_lcu_y[LCU_WIDTH*LCU_WIDTH];
|
||||
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C];
|
||||
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C];
|
||||
|
||||
int hi_prec_luma_rec0;
|
||||
int hi_prec_luma_rec1;
|
||||
int hi_prec_chroma_rec0;
|
||||
int hi_prec_chroma_rec1;
|
||||
|
||||
hi_prec_buf_t* high_precision_rec0 = 0;
|
||||
hi_prec_buf_t* high_precision_rec1 = 0;
|
||||
|
||||
int temp_x, temp_y;
|
||||
|
||||
|
||||
|
||||
static void setup()
|
||||
{
|
||||
|
||||
memset(lcu1.rec.y, 0, sizeof(kvz_pixel) * 64 * 64);
|
||||
memset(lcu1.rec.u, 0, sizeof(kvz_pixel) * 32 * 32);
|
||||
memset(lcu1.rec.v, 0, sizeof(kvz_pixel) * 32 * 32);
|
||||
|
||||
for (int i = 0; i < LCU_WIDTH*LCU_WIDTH; i++) {
|
||||
temp_lcu_y[i] = rand() %256;
|
||||
lcu1.rec.y[i] = rand() % 256;
|
||||
}
|
||||
|
||||
for (int i = 0; i < LCU_WIDTH_C*LCU_WIDTH_C; i++) {
|
||||
temp_lcu_u[i] = rand() % 256;
|
||||
temp_lcu_v[i] = rand() % 256;
|
||||
lcu1.rec.v[i] = rand() % 256;
|
||||
lcu1.rec.u[i] = rand() % 256;
|
||||
}
|
||||
|
||||
|
||||
|
||||
memset(expected_test_result.rec.y, 0, sizeof(kvz_pixel) * 64 * 64);
|
||||
memset(expected_test_result.rec.u, 0, sizeof(kvz_pixel) * 32 * 32);
|
||||
memset(expected_test_result.rec.v, 0, sizeof(kvz_pixel) * 32 * 32);
|
||||
|
||||
memcpy(expected_test_result.rec.y, lcu1.rec.y, sizeof(kvz_pixel) * 64 * 64);
|
||||
memcpy(expected_test_result.rec.u, lcu1.rec.u, sizeof(kvz_pixel) * 32 * 32);
|
||||
memcpy(expected_test_result.rec.v, lcu1.rec.v, sizeof(kvz_pixel) * 32 * 32);
|
||||
|
||||
// Setup is not optimized working function from picture-generic.c.
|
||||
|
||||
|
||||
int shift = 15 - KVZ_BIT_DEPTH;
|
||||
int offset = 1 << (shift - 1);
|
||||
|
||||
hi_prec_luma_rec0 = 0;// mv_param[0][0] & 3 || mv_param[0][1] & 3;
|
||||
hi_prec_luma_rec1 = 0;// mv_param[1][0] & 3 || mv_param[1][1] & 3;
|
||||
|
||||
hi_prec_chroma_rec0 = 0; // mv_param[0][0] & 7 || mv_param[0][1] & 7;
|
||||
hi_prec_chroma_rec1 =0; // mv_param[1][0] & 7 || mv_param[1][1] & 7;
|
||||
|
||||
if (hi_prec_chroma_rec0) high_precision_rec0 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
|
||||
if (hi_prec_chroma_rec1) high_precision_rec1 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
|
||||
|
||||
|
||||
|
||||
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (temp_x = 0; temp_x < width; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
int16_t sample0_y = (hi_prec_luma_rec0 ? high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_y = (hi_prec_luma_rec1 ? high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu] : (expected_test_result.rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
expected_test_result.rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_y + sample1_y + offset) >> shift);
|
||||
}
|
||||
|
||||
}
|
||||
for (temp_y = 0; temp_y < height >> 1; ++temp_y) {
|
||||
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
for (temp_x = 0; temp_x < width >> 1; ++temp_x) {
|
||||
int x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
int16_t sample0_u = (hi_prec_chroma_rec0 ? high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_u = (hi_prec_chroma_rec1 ? high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (expected_test_result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
expected_test_result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_u + sample1_u + offset) >> shift);
|
||||
|
||||
int16_t sample0_v = (hi_prec_chroma_rec0 ? high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
int16_t sample1_v = (hi_prec_chroma_rec1 ? high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] : (expected_test_result.rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] << (14 - KVZ_BIT_DEPTH)));
|
||||
expected_test_result.rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu] = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_v + sample1_v + offset) >> shift);
|
||||
|
||||
|
||||
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
TEST test_inter_recon_bipred_generic()
|
||||
{
|
||||
|
||||
memset(result.rec.y, 0, sizeof(kvz_pixel) * 64 * 64);
|
||||
memset(result.rec.u, 0, sizeof(kvz_pixel) * 32 * 32);
|
||||
memset(result.rec.v, 0, sizeof(kvz_pixel) * 32 * 32);
|
||||
|
||||
|
||||
memcpy(result.rec.y, lcu1.rec.y, sizeof(kvz_pixel) * 64 * 64);
|
||||
memcpy(result.rec.u, lcu1.rec.u, sizeof(kvz_pixel) * 32 * 32);
|
||||
memcpy(result.rec.v, lcu1.rec.v, sizeof(kvz_pixel) * 32 * 32);
|
||||
|
||||
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (temp_x = 0; temp_x < width; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
printf("%d ", (expected_test_result.rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu]));
|
||||
}
|
||||
}
|
||||
printf("\n");
|
||||
/*
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
for (temp_x = 0; temp_x < width >> 1; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH_C)-1));
|
||||
printf("%d ", (expected_test_result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]));
|
||||
}
|
||||
}
|
||||
printf("\n");*/
|
||||
|
||||
|
||||
kvz_inter_recon_bipred_generic(hi_prec_luma_rec0, hi_prec_luma_rec1, hi_prec_chroma_rec0, hi_prec_chroma_rec1, width, height, xpos, ypos, high_precision_rec0, high_precision_rec1, &result, temp_lcu_y, temp_lcu_u, temp_lcu_v);
|
||||
|
||||
/*
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
for (temp_x = 0; temp_x < width >> 1; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH_C)-1));
|
||||
printf("%d ", (result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]));
|
||||
}
|
||||
}
|
||||
printf("\n");*/
|
||||
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (temp_x = 0; temp_x < width; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
printf("%d ", (result.rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu]));
|
||||
}
|
||||
}
|
||||
printf("\n");
|
||||
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (temp_x = 0; temp_x < width; ++temp_x) {
|
||||
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
ASSERT_EQ_FMT(expected_test_result.rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu], result.rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu], "%d");
|
||||
}
|
||||
}
|
||||
|
||||
for (temp_y = 0; temp_y < height >> 1; ++temp_y) {
|
||||
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
for (temp_x = 0; temp_x < width >> 1; ++temp_x) {
|
||||
int x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
ASSERT_EQ_FMT(expected_test_result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu], result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu], "%d");
|
||||
ASSERT_EQ_FMT(expected_test_result.rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu], result.rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu], "%d");
|
||||
}
|
||||
}
|
||||
|
||||
PASS();
|
||||
}
|
||||
|
||||
SUITE(bipred_generic_tests)
|
||||
{
|
||||
setup();
|
||||
|
||||
for (volatile int i = 0; i < strategies.count; ++i) {
|
||||
if (strcmp(strategies.strategies[i].type, "inter_recon_bipred") != 0) {
|
||||
continue;
|
||||
}
|
||||
|
||||
kvz_inter_recon_bipred_generic = strategies.strategies[i].fptr;
|
||||
RUN_TEST(test_inter_recon_bipred_generic);
|
||||
}
|
||||
}
|
||||
|
|
@ -32,6 +32,7 @@ extern SUITE(dct_tests);
|
|||
|
||||
extern SUITE(coeff_sum_tests);
|
||||
extern SUITE(mv_cand_tests);
|
||||
extern SUITE(bipred_generic_tests);
|
||||
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
|
|
@ -57,5 +58,7 @@ int main(int argc, char **argv)
|
|||
|
||||
RUN_SUITE(mv_cand_tests);
|
||||
|
||||
RUN_SUITE(bipred_generic_tests);
|
||||
|
||||
GREATEST_MAIN_END();
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in a new issue