hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-24 02:24:07 +00:00
Code Issues Actions Projects Releases Wiki Activity

Switched some variables to different location, altered inter_recon_bipred_avx2 function

Browse source
This commit is contained in:
Reima Hyvönen 2018-08-02 16:08:59 +03:00
parent 9ba85ab7b0
commit ed50d71fde
2 changed files with 20 additions and 22 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
src/strategies/avx2/picture-avx2.c
View file

@ -875,23 +875,6 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
kvz_pixel* temp_lcu_v) kvz_pixel* temp_lcu_v)
{ {
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 temp_epi16_y;
__m256i temp_epi8;
__m256i temp_y_epi16;
__m256i sample0_epi16;
__m256i sample1_epi16;
int start_point = 0;
int start_point_uv = 0;
if (hi_prec_luma_rec0 == 0 && hi_prec_luma_rec1 == 0 && hi_prec_chroma_rec0 == 0 && hi_prec_chroma_rec1 == 0) if (hi_prec_luma_rec0 == 0 && hi_prec_luma_rec1 == 0 && hi_prec_chroma_rec0 == 0 && hi_prec_chroma_rec1 == 0)
{ {
inter_recon_bipred_no_mov_avx2(height, width, ypos, xpos, high_precision_rec0, high_precision_rec1, lcu, temp_lcu_y, temp_lcu_u, temp_lcu_v); inter_recon_bipred_no_mov_avx2(height, width, ypos, xpos, high_precision_rec0, high_precision_rec1, lcu, temp_lcu_y, temp_lcu_u, temp_lcu_v);
@ -899,6 +882,17 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
else else
{ {
int y_in_lcu, x_in_lcu;
int shift = 15 - KVZ_BIT_DEPTH;
int offset = 1 << (shift - 1);
int shift_left = 14 - KVZ_BIT_DEPTH;
__m256i temp_epi16_y, temp_epi8, temp_y_epi16, sample0_epi16, sample1_epi16;
int start_point = 0;
int start_point_uv = 0;
for (int temp_y = 0; temp_y < height; temp_y += 1) { for (int temp_y = 0; temp_y < height; temp_y += 1) {
temp_epi16_y = _mm256_setzero_si256(); temp_epi16_y = _mm256_setzero_si256();
int temp = 0; int temp = 0;
@ -931,7 +925,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
case 4: case 4:
temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16); temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu] = _mm256_extract_epi32(temp_epi8, 0);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 0] = _mm256_extract_epi8(temp_epi8, 0);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 1] = _mm256_extract_epi8(temp_epi8, 1);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 2] = _mm256_extract_epi8(temp_epi8, 2);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 3] = _mm256_extract_epi8(temp_epi8, 3);
break; break;

8
tests/inter_recon_bipred_tests.c
View file

@ -89,11 +89,11 @@ static void setup()
int shift = 15 - KVZ_BIT_DEPTH; int shift = 15 - KVZ_BIT_DEPTH;
int offset = 1 << (shift - 1); int offset = 1 << (shift - 1);
hi_prec_luma_rec0 = 0; //mv_param[0][0] & 3 || mv_param[0][1] & 3; hi_prec_luma_rec0 = 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_luma_rec1 = 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_rec0 = 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; hi_prec_chroma_rec1 = 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_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); if (hi_prec_chroma_rec1) high_precision_rec1 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);