mirror of
https://github.com/ultravideo/uvg266.git
synced 2024-11-23 18:14:06 +00:00
Trying to find the bug in luma
This commit is contained in:
Reima Hyvönen
parent
2f5f81bac3
commit
381e786e10
|
|
@ -887,6 +887,320 @@ static void inter_recon_bipred_no_mov_avx2(
|
|||
|
||||
}
|
||||
|
||||
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,
|
||||
kvz_pixel* temp_lcu_u,
|
||||
kvz_pixel* temp_lcu_v)
|
||||
{
|
||||
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);
|
||||
}
|
||||
|
||||
else
|
||||
{
|
||||
|
||||
int y_in_lcu, x_in_lcu;
|
||||
int shift = 15 - KVZ_BIT_DEPTH;
|
||||
int offset = 1 << (shift - 1);
|
||||
int8_t shift_left = 14 - KVZ_BIT_DEPTH;
|
||||
|
||||
__m256i temp_epi32_y, temp_epi8, temp_y_epi32, sample0_epi32, sample1_epi32, temp_epi16;
|
||||
|
||||
int32_t * pointer = 0;
|
||||
|
||||
__m256i offset_epi32 = _mm256_set1_epi32(offset);
|
||||
/*
|
||||
printf("%d ", hi_prec_luma_rec0);
|
||||
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)));
|
||||
uint8_t luku = (kvz_pixel)kvz_fast_clip_32bit_to_pixel((sample0_y + sample1_y + offset) >> shift);
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] = luku;
|
||||
}
|
||||
}
|
||||
*/
|
||||
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
temp_epi32_y = _mm256_setzero_si256();
|
||||
|
||||
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
|
||||
|
||||
for (int temp_x = 0; temp_x < width; temp_x += 8) {
|
||||
x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
|
||||
|
||||
switch (width)
|
||||
{
|
||||
|
||||
case 4:
|
||||
|
||||
// Load total of 4 elements from memory to vector
|
||||
|
||||
sample0_epi32 = hi_prec_luma_rec0 ? _mm256_cvtepi16_epi32(_mm_loadl_epi64((__m128i*) &(high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepi8_epi32(_mm_cvtsi32_si128(*(int32_t*)&(temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu]))), 14 - KVZ_BIT_DEPTH);
|
||||
|
||||
sample1_epi32 = hi_prec_luma_rec1 ? _mm256_cvtepi16_epi32(_mm_loadl_epi64((__m128i*) &(high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepi8_epi32(_mm_cvtsi32_si128(*(int32_t*) &(lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu]))), 14 - KVZ_BIT_DEPTH);
|
||||
|
||||
|
||||
// (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);
|
||||
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi16 = _mm256_packs_epi32(temp_y_epi32, temp_y_epi32);
|
||||
temp_epi16 = _mm256_permute4x64_epi64(temp_epi16, _MM_SHUFFLE(3, 1, 2, 0));
|
||||
temp_epi8 = _mm256_packus_epi16(temp_epi16, temp_epi16);
|
||||
|
||||
|
||||
pointer = (int32_t*)&(lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu]);
|
||||
*pointer = _mm_cvtsi128_si32(_mm256_castsi256_si128(temp_epi8));
|
||||
break;
|
||||
|
||||
default:
|
||||
|
||||
// Load total of 8 elements from memory to vector
|
||||
sample0_epi32 = hi_prec_luma_rec0 ? _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*) &(high_precision_rec0->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepi8_epi32((_mm_loadl_epi64((__m128i*) &(temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), 14 - KVZ_BIT_DEPTH);
|
||||
|
||||
sample1_epi32 = hi_prec_luma_rec1 ? _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*) &(high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm256_slli_epi32(_mm256_cvtepi8_epi32((_mm_loadl_epi64((__m128i*) &(lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), 14 - KVZ_BIT_DEPTH);
|
||||
|
||||
|
||||
// (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);
|
||||
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi16 = _mm256_packs_epi32(temp_y_epi32, temp_y_epi32);
|
||||
temp_epi16 = _mm256_permute4x64_epi64(temp_epi16, _MM_SHUFFLE(3, 1, 2, 0));
|
||||
temp_epi8 = _mm256_packus_epi16(temp_epi16, temp_epi16);
|
||||
|
||||
|
||||
// Store 64-bits from vector to memory
|
||||
_mm_storel_epi64((__m128i*)&(lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
__m256i sample0_epi16, sample1_epi16;
|
||||
int temp_uv = 0;
|
||||
|
||||
int start_point_uv = 0;
|
||||
|
||||
__m256i temp_u_epi16 = _mm256_setzero_si256();
|
||||
__m256i temp_v_epi16 = _mm256_setzero_si256();
|
||||
|
||||
__m256i offset_epi16 = _mm256_set1_epi16(offset);
|
||||
__m256i temp_epi16_u = _mm256_setzero_si256();
|
||||
__m256i temp_epi16_v = _mm256_setzero_si256();
|
||||
|
||||
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+=16) {
|
||||
int x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));
|
||||
switch (width) {
|
||||
|
||||
case 8:
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadl_epi64((__m128i*) &(high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_cvtsi32_si128(*(int32_t*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_castsi128_si256(_mm_loadl_epi64((__m128i*) &(high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_cvtsi32_si128(*(int32_t*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_adds_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_u_epi16, temp_u_epi16);
|
||||
|
||||
pointer = (int32_t*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]);
|
||||
*pointer = _mm_cvtsi128_si32(_mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_v_epi16, temp_v_epi16);
|
||||
|
||||
pointer = (int32_t*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]);
|
||||
*pointer = _mm_cvtsi128_si32(_mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
break;
|
||||
|
||||
case 16:
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_adds_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_u_epi16, temp_u_epi16);
|
||||
|
||||
// Store 64-bit integer into memory
|
||||
_mm_storel_epi64((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_v_epi16, temp_v_epi16);
|
||||
|
||||
// Store 64-bit integer into memory
|
||||
_mm_storel_epi64((__m128i*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
break;
|
||||
|
||||
case 32:
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_adds_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_adds_epu16(temp_v_epi16, offset_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
|
||||
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_u_epi16, temp_u_epi16), _MM_SHUFFLE(3, 1, 2, 0));
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_v_epi16, temp_v_epi16), _MM_SHUFFLE(3, 1, 2, 0));
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
break;
|
||||
|
||||
default:
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec0->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec1->u[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
sample1_epi16 = hi_prec_chroma_rec1 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec1->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_v_epi16 = _mm256_adds_epi16(temp_v_epi16, offset_epi16);
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
|
||||
if (temp_uv == 0) {
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi16_u = temp_u_epi16;
|
||||
temp_epi16_v = temp_v_epi16;
|
||||
|
||||
// Save starting point to memory
|
||||
start_point_uv = (y_in_lcu)* LCU_WIDTH_C + x_in_lcu;
|
||||
|
||||
temp_uv++;
|
||||
}
|
||||
|
||||
else {
|
||||
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_u, temp_u_epi16), _MM_SHUFFLE(3, 1, 2, 0));
|
||||
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.u[start_point_uv]), temp_epi8);
|
||||
|
||||
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_v, temp_v_epi16), _MM_SHUFFLE(3, 1, 2, 0));
|
||||
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), temp_epi8);
|
||||
|
||||
temp_uv = 0;
|
||||
|
||||
}
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
||||
const int hi_prec_luma_rec1,
|
||||
const int hi_prec_chroma_rec0,
|
||||
|
|
@ -919,8 +1233,6 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
__m256i temp_u_epi16 = _mm256_setzero_si256();
|
||||
__m256i temp_v_epi16 = _mm256_setzero_si256();
|
||||
__m256i max_values = _mm256_setzero_si256();
|
||||
__m256i compare_epi16 = _mm256_setzero_si256();
|
||||
__m256i and_epi16 = _mm256_set1_epi16(-1);
|
||||
|
||||
int start_point = 0;
|
||||
int start_point_uv = 0;
|
||||
|
|
@ -960,14 +1272,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_y_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_y_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_add_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_y_epi16), max_values);
|
||||
temp_y_epi16 = _mm256_adds_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_srli_epi16(temp_y_epi16, shift);
|
||||
temp_y_epi16 = _mm256_and_si256(temp_y_epi16, _mm256_subs_epi16(and_epi16, compare_epi16));
|
||||
|
||||
// Pack the bits from 16-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16);
|
||||
|
|
@ -989,17 +1298,13 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
sample1_epi16 = hi_prec_luma_rec1 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec1->y[y_in_lcu * LCU_WIDTH + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16((_mm_loadl_epi64((__m128i*) &(lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu])))), shift_left);
|
||||
|
||||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_y_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_add_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_y_epi16), max_values);
|
||||
|
||||
temp_y_epi16 = _mm256_adds_epu16(temp_y_epi16, offset_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_srli_epi16(temp_y_epi16, shift);
|
||||
temp_y_epi16 = _mm256_and_si256(temp_y_epi16, _mm256_subs_epi16(and_epi16, compare_epi16));
|
||||
|
||||
|
||||
// Pack the bits from 16-bit to 8-bit
|
||||
|
|
@ -1023,14 +1328,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_y_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_add_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_y_epi16), max_values);
|
||||
temp_y_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_adds_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_srli_epi16(temp_y_epi16, shift);
|
||||
temp_y_epi16 = _mm256_and_si256(temp_y_epi16, _mm256_subs_epi16(and_epi16, compare_epi16));
|
||||
|
||||
// Pack the bits from 16-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16);
|
||||
|
|
@ -1054,14 +1356,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_y_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_y_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_add_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_y_epi16), max_values);
|
||||
temp_y_epi16 = _mm256_adds_epi16(temp_y_epi16, offset_epi16);
|
||||
|
||||
temp_y_epi16 = _mm256_srli_epi16(temp_y_epi16, shift);
|
||||
temp_y_epi16 = _mm256_and_si256(temp_y_epi16, _mm256_subs_epi16(and_epi16, compare_epi16));
|
||||
|
||||
// Pack the bits from 16-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16);
|
||||
|
|
@ -1103,15 +1402,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_add_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_u_epi16), max_values);
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
temp_u_epi16 = _mm256_and_si256(temp_u_epi16, _mm256_subs_epu16(and_epi16, compare_epi16));
|
||||
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
|
@ -1122,15 +1417,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_v_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_add_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_v_epi16), max_values);
|
||||
temp_v_epi16 = _mm256_adds_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
temp_v_epi16 = _mm256_and_si256(temp_v_epi16, _mm256_subs_epu16(and_epi16, compare_epi16));
|
||||
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_u_epi16, temp_u_epi16);
|
||||
|
||||
|
|
@ -1155,15 +1446,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_add_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_u_epi16), max_values);
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
temp_u_epi16 = _mm256_and_si256(temp_u_epi16, _mm256_subs_epu16(and_epi16, compare_epi16));
|
||||
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
|
@ -1174,14 +1461,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_v_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_add_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_v_epi16), max_values);
|
||||
temp_v_epi16 = _mm256_adds_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
temp_v_epi16 = _mm256_and_si256(temp_v_epi16, _mm256_subs_epu16(and_epi16, compare_epi16));
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_u_epi16, temp_u_epi16);
|
||||
|
||||
|
|
@ -1207,15 +1491,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_add_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_u_epi16), max_values);
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
temp_u_epi16 = _mm256_and_si256(temp_u_epi16, _mm256_subs_epu16(and_epi16, compare_epi16));
|
||||
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_castsi128_si256(_mm_loadu_si128((__m128i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))) :
|
||||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
|
@ -1226,14 +1506,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
max_values = _mm256_max_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_v_epi16 = _mm256_adds_epu16(sample0_epi16, sample1_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_add_epi16(temp_v_epi16, offset_epi16);
|
||||
|
||||
compare_epi16 = _mm256_cmpeq_epi16(_mm256_max_epu16(max_values, temp_v_epi16), max_values);
|
||||
temp_v_epi16 = _mm256_adds_epu16(temp_v_epi16, offset_epi16);
|
||||
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
temp_v_epi16 = _mm256_and_si256(temp_v_epi16, _mm256_subs_epu16(and_epi16, compare_epi16));
|
||||
|
||||
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_u_epi16, temp_u_epi16), _MM_SHUFFLE(3, 1, 2, 0));
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
|
@ -1252,8 +1529,8 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_u_epi16 = _mm256_add_epi16(temp_u_epi16, offset_epi16);
|
||||
temp_u_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_u_epi16 = _mm256_adds_epi16(temp_u_epi16, offset_epi16);
|
||||
temp_u_epi16 = _mm256_srli_epi16(temp_u_epi16, shift);
|
||||
|
||||
sample0_epi16 = hi_prec_chroma_rec0 ? _mm256_loadu_si256((__m256i*) &(high_precision_rec0->v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])) :
|
||||
|
|
@ -1263,8 +1540,8 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
_mm256_slli_epi16(_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))), shift_left);
|
||||
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi16 = _mm256_add_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_v_epi16 = _mm256_add_epi16(temp_v_epi16, offset_epi16);
|
||||
temp_v_epi16 = _mm256_adds_epi16(sample0_epi16, sample1_epi16);
|
||||
temp_v_epi16 = _mm256_adds_epi16(temp_v_epi16, offset_epi16);
|
||||
temp_v_epi16 = _mm256_srli_epi16(temp_v_epi16, shift);
|
||||
|
||||
if (temp_uv == 0) {
|
||||
|
|
@ -1302,7 +1579,7 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
|
||||
}
|
||||
}
|
||||
|
||||
*/
|
||||
#endif //COMPILE_INTEL_AVX2
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -32,9 +32,9 @@ static lcu_t lcu1;
|
|||
|
||||
int temp1, temp2, temp3, temp4;
|
||||
|
||||
int16_t mv_param[2][2] = { { 7,7 },{ 7,7 } };
|
||||
int width = 8;
|
||||
int height = 8;
|
||||
int16_t mv_param[2][2] = { { 3,3 },{ 3,3 } };
|
||||
int width = 16;
|
||||
int height = 16;
|
||||
int xpos = 0;
|
||||
int ypos = 0;
|
||||
|
||||
|
|
@ -125,27 +125,10 @@ TEST test_inter_recon_bipred()
|
|||
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 += 1) {
|
||||
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 >> 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));
|
||||
printf("%d ", expected_test_result.rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]);
|
||||
}
|
||||
}
|
||||
printf("\n");*/
|
||||
|
||||
|
||||
kvz_inter_recon_bipred_blend(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 + temp_y) & ((LCU_WIDTH)-1));
|
||||
for (temp_x = 0; temp_x < width; temp_x += 1) {
|
||||
|
|
@ -154,8 +137,8 @@ TEST test_inter_recon_bipred()
|
|||
}
|
||||
}
|
||||
printf("\n");
|
||||
*/
|
||||
|
||||
|
||||
/*
|
||||
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) {
|
||||
|
|
@ -164,7 +147,7 @@ TEST test_inter_recon_bipred()
|
|||
}
|
||||
}
|
||||
printf("\n");
|
||||
|
||||
*/
|
||||
|
||||
for (temp_y = 0; temp_y < height; ++temp_y) {
|
||||
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
|
|
|
|||
Loading…
Reference in a new issue