mirror of
https://github.com/ultravideo/uvg266.git
synced 2024-11-23 18:14:06 +00:00
Toimiva ratkaisu
This commit is contained in:
Reima Hyvönen
parent
17babfffa4
commit
ea83ae45f0
|
|
@ -21,9 +21,6 @@ Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "kvazaar_cli", "kvazaar_cli\
|
|||
EndProjectSection
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
Debug|Win32 = Debug|Win32
|
||||
Debug|x64 = Debug|x64
|
||||
|
|
@ -70,4 +67,28 @@ Global
|
|||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
GlobalSection(Performance) = preSolution
|
||||
HasPerformanceSessions = true
|
||||
EndGlobalSection
|
||||
EndGlobal
|
||||
|
|
|
|||
|
|
@ -98,6 +98,7 @@
|
|||
</ProjectReference>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="..\..\tests\bipred_generic_tests.c" />
|
||||
<ClCompile Include="..\..\tests\coeff_sum_tests.c" />
|
||||
<ClCompile Include="..\..\tests\dct_tests.c" />
|
||||
<ClCompile Include="..\..\tests\test_strategies.c" />
|
||||
|
|
|
|||
|
|
@ -42,6 +42,9 @@
|
|||
<ClCompile Include="..\..\tests\coeff_sum_tests.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\tests\bipred_generic_tests.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClInclude Include="..\..\tests\sad_tests.h">
|
||||
|
|
|
|||
|
|
@ -731,387 +731,545 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
|
|||
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_v[LCU_WIDTH_C*LCU_WIDTH_C])
|
||||
{
|
||||
|
||||
int y_in_lcu;
|
||||
int x_in_lcu;
|
||||
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;
|
||||
int shift = 15 - KVZ_BIT_DEPTH;
|
||||
int offset = 1 << (shift-1);
|
||||
int shift_left = 14 - KVZ_BIT_DEPTH;
|
||||
__m256i temp_epi32;
|
||||
__m256i temp_epi16;
|
||||
|
||||
__m256i offset_epi32 = _mm256_set1_epi32(offset);
|
||||
__m256i temp_epi32;
|
||||
__m256i temp_epi16;
|
||||
__m256i temp_epi8;
|
||||
__m256i temp_y_epi32, temp_u_epi32, temp_v_epi32;
|
||||
|
||||
__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 temp_epi32_u = _mm256_setzero_si256();
|
||||
__m256i temp_epi32_v = _mm256_setzero_si256();
|
||||
__m256i sample0_epi32;
|
||||
__m256i sample1_epi32;
|
||||
|
||||
__m256i sample0_epi32;
|
||||
__m256i sample1_epi32;
|
||||
__m256i temp_epi8_u, temp_epi8_v;
|
||||
|
||||
__m256i temp_epi16_u = _mm256_setzero_si256();
|
||||
__m256i temp_epi16_v = _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;
|
||||
|
||||
__m256i final_epi8_256 = _mm256_setzero_si256();
|
||||
__m128i final_epi8_128;
|
||||
|
||||
__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);
|
||||
switch (width)
|
||||
{
|
||||
|
||||
__m128i final_epi8_128;
|
||||
case 4:
|
||||
|
||||
x_in_lcu = ((xpos) & ((LCU_WIDTH)-1));
|
||||
|
||||
switch (width)
|
||||
{
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
|
||||
case 4:
|
||||
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);
|
||||
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
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);
|
||||
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos) & ((LCU_WIDTH)-1));
|
||||
|
||||
// (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, sample_epi32), sample_epi32);
|
||||
|
||||
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);
|
||||
int8_t*temp_int_y = (int8_t*)&final_epi8_128;
|
||||
|
||||
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);
|
||||
for (int i = 0; i < 4; i++) {
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu + i] = temp_int_y[i];
|
||||
}
|
||||
|
||||
//lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu] = _mm_cvtsi128_si32(final_epi8_128);
|
||||
|
||||
// (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);
|
||||
if (temp_y < height >> 1) {
|
||||
|
||||
int8_t*temp_int_y = (int8_t*)&final_epi8_128;
|
||||
y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
|
||||
x_in_lcu = (((xpos >> 1)) & (LCU_WIDTH_C - 1));
|
||||
|
||||
for (int i = 0; i < 4; i++) {
|
||||
lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu + i] = temp_int_y[i];
|
||||
}
|
||||
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);
|
||||
|
||||
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));
|
||||
// (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);
|
||||
|
||||
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);
|
||||
__m128i temp_u = _mm_packus_epi16(_mm_packus_epi32(sample_epi32, sample_epi32), sample_epi32);
|
||||
int8_t*temp_int_u = (int8_t*)&temp_u;
|
||||
|
||||
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);
|
||||
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 + 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);
|
||||
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);
|
||||
|
||||
__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;
|
||||
// (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);
|
||||
|
||||
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);
|
||||
__m128i temp_v = _mm_packus_epi16(_mm_packus_epi32(sample_epi32, sample_epi32), sample_epi32);
|
||||
int8_t*temp_int_v = (int8_t*)&temp_v;
|
||||
|
||||
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);
|
||||
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;
|
||||
|
||||
// (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;
|
||||
default:
|
||||
|
||||
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];
|
||||
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)
|
||||
{
|
||||
|
||||
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;
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
|
||||
}
|
||||
break;
|
||||
for (int temp_x = 0; temp_x < width; temp_x += 16) {
|
||||
|
||||
default:
|
||||
x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
|
||||
int start_point = 0;
|
||||
int start_point_uv = 0;
|
||||
sample0_epi32 = (_mm256_cvtepu8_epi16((_mm_loadu_si128((__m128i*) &(temp_lcu_y[y_in_lcu * LCU_WIDTH + x_in_lcu])))));
|
||||
|
||||
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;
|
||||
sample1_epi32 = (_mm256_cvtepu8_epi16((_mm_loadu_si128((__m128i*) &(lcu->rec.y[y_in_lcu * LCU_WIDTH + x_in_lcu])))));
|
||||
|
||||
temp_y_epi32 = _mm256_avg_epu16(sample0_epi32, sample1_epi32);
|
||||
|
||||
for (int temp_x = 0; temp_x < width; temp_x += 8) {
|
||||
switch (width)
|
||||
{
|
||||
case 8:
|
||||
|
||||
// Pack the bits from 16-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(temp_y_epi32, temp_y_epi32);
|
||||
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
// 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));
|
||||
|
||||
// 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);
|
||||
break;
|
||||
|
||||
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);
|
||||
case 16:
|
||||
|
||||
// (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_epi8 = _mm256_packus_epi16(temp_y_epi32, temp_y_epi32);
|
||||
|
||||
switch (width)
|
||||
{
|
||||
case 8:
|
||||
temp_epi8 = _mm256_permute4x64_epi64(temp_epi8, 216);
|
||||
|
||||
// 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 128-bit to memory
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu]), _mm256_castsi256_si128(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;
|
||||
|
||||
break;
|
||||
default:
|
||||
if (temp == 0) {
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp = 1;
|
||||
start_point = y_in_lcu* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
|
||||
case 16:
|
||||
else {
|
||||
|
||||
if (temp == 0) {
|
||||
// Convert packed 16-bit integers to packed 8-bit integers and store result to vector
|
||||
temp_epi8 = _mm256_packus_epi16(temp_epi32, temp_y_epi32);
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
// Arrange the vector to right order before inserting it
|
||||
temp_epi8 = _mm256_permute4x64_epi64(temp_epi8, 216);
|
||||
|
||||
// Save starting point to memory
|
||||
start_point = (y_in_lcu)* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
|
||||
else if (temp == 1) {
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.y[start_point]), temp_epi8);
|
||||
temp = 0;
|
||||
}
|
||||
}
|
||||
|
||||
// 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);
|
||||
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));
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
sample0_epi32 = (_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(temp_lcu_u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))));
|
||||
|
||||
// Fill 128 bit vector with packed data and store it to memory
|
||||
__m128i final_epi8_16 = _mm_loadu_si128((__m128i*)&temp_epi8);
|
||||
sample1_epi32 = (_mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.u[y_in_lcu * LCU_WIDTH_C + x_in_lcu]))));
|
||||
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.y[start_point]), final_epi8_16);
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_u_epi32 = _mm256_avg_epu16(sample0_epi32, sample1_epi32);
|
||||
|
||||
temp = 0;
|
||||
}
|
||||
sample0_epi32 = _mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(temp_lcu_v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])));
|
||||
|
||||
sample1_epi32 = _mm256_cvtepu8_epi16(_mm_loadu_si128((__m128i*) &(lcu->rec.v[y_in_lcu * LCU_WIDTH_C + x_in_lcu])));
|
||||
|
||||
break;
|
||||
// (sample1 + sample2 + offset)>>shift
|
||||
temp_v_epi32 = _mm256_avg_epu16(sample0_epi32, sample1_epi32);
|
||||
|
||||
default:
|
||||
if (temp == 0) {
|
||||
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
switch (width)
|
||||
{
|
||||
case 8:
|
||||
|
||||
start_point = y_in_lcu* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
temp_epi8_u = _mm256_packus_epi16(temp_u_epi32, temp_u_epi32);
|
||||
temp_epi8_v = _mm256_packus_epi16(temp_v_epi32, temp_v_epi32);
|
||||
|
||||
else if (temp == 1) {
|
||||
int8_t*temp_int_u = (int8_t*)&temp_epi8_u;
|
||||
int8_t*temp_int_v = (int8_t*)&temp_epi8_v;
|
||||
|
||||
// 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++;
|
||||
}
|
||||
for (int i = 0; i < 4; 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];
|
||||
}
|
||||
|
||||
else if (temp == 2) {
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
}
|
||||
break;
|
||||
|
||||
else {
|
||||
case 16:
|
||||
temp_epi8 = _mm256_packus_epi16(temp_u_epi32, temp_u_epi32);
|
||||
|
||||
// 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));
|
||||
// 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));
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
final_epi8_256 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
temp_epi8 = _mm256_packus_epi16(temp_v_epi32, temp_v_epi32);
|
||||
|
||||
// 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));
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.y[start_point]), final_epi8_256);
|
||||
temp = 0;
|
||||
}
|
||||
}
|
||||
break;
|
||||
|
||||
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));
|
||||
case 32:
|
||||
|
||||
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);
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi8_u = _mm256_packus_epi16(temp_u_epi32, temp_u_epi32);
|
||||
temp_epi8_v = _mm256_packus_epi16(temp_v_epi32, temp_v_epi32);
|
||||
|
||||
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);
|
||||
temp_epi8_u = _mm256_permute4x64_epi64(temp_epi8_u, 216);
|
||||
temp_epi8_v = _mm256_permute4x64_epi64(temp_epi8_v, 216);
|
||||
|
||||
// (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);
|
||||
// Fill 128 bit vector with packed data and store it to memory
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8_u));
|
||||
|
||||
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);
|
||||
// Fill 128 bit vector with packed data and store it to memory
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8_v));
|
||||
|
||||
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);
|
||||
break;
|
||||
|
||||
default:
|
||||
if (temp_uv == 0) {
|
||||
|
||||
switch (width) {
|
||||
// Store to temporary vector
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
|
||||
case 8:
|
||||
// Save starting point to memory
|
||||
start_point_uv = (y_in_lcu)* LCU_WIDTH_C + x_in_lcu;
|
||||
|
||||
__m256i temp_epi8u = _mm256_packus_epi16(_mm256_packus_epi32(temp_u_epi32, temp_zeros_256), temp_zeros_256);
|
||||
temp_uv = 1;
|
||||
}
|
||||
|
||||
int8_t *temp_int_8_u = (int8_t*)&temp_epi8u;
|
||||
else {
|
||||
// Pack 32 bit to 8 bit
|
||||
temp_epi8_u = _mm256_packus_epi16(temp_epi32_u, temp_u_epi32);
|
||||
temp_epi8_v = _mm256_packus_epi16(temp_epi32_v, temp_v_epi32);
|
||||
|
||||
__m256i temp_epi8v = _mm256_packus_epi16(_mm256_packus_epi32(temp_v_epi32, temp_zeros_256), temp_zeros_256);
|
||||
// Arrange the vector to right order before inserting it
|
||||
temp_epi8_u = _mm256_permute4x64_epi64(temp_epi8_u, 216);
|
||||
|
||||
int8_t *temp_int_8_v = (int8_t*)&temp_epi8v;
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.u[start_point_uv]), temp_epi8_u);
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
temp_epi8_v = _mm256_permute4x64_epi64(temp_epi8_v, 216);
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), temp_epi8_v);
|
||||
temp_uv = 0;
|
||||
}
|
||||
}
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
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];
|
||||
}
|
||||
else
|
||||
{
|
||||
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;
|
||||
__m256i idx = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
|
||||
__m256i offset_epi32 = _mm256_set1_epi32(offset);
|
||||
__m256i zeros_256 = _mm256_setzero_si256();
|
||||
__m256i temp_epi16_u = _mm256_setzero_si256();
|
||||
__m256i temp_epi16_v = _mm256_setzero_si256();
|
||||
|
||||
break;
|
||||
|
||||
case 16:
|
||||
for (int temp_x = 0; temp_x < width; temp_x += 8) {
|
||||
|
||||
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);
|
||||
y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
|
||||
|
||||
// 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);
|
||||
// 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);
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_v_epi32, temp_zeros_256), temp_zeros_256);
|
||||
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);
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
final_epi8_128 = _mm_loadu_si128((__m128i*)&temp_epi8);
|
||||
// (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);
|
||||
|
||||
// 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);
|
||||
switch (width)
|
||||
{
|
||||
case 8:
|
||||
|
||||
break;
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_y_epi32, zeros_256), zeros_256);
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
case 32:
|
||||
// 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));
|
||||
|
||||
if (temp_uv == 0) {
|
||||
break;
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
case 16:
|
||||
|
||||
// Save starting point to memory
|
||||
start_point_uv = (y_in_lcu)* LCU_WIDTH_C + x_in_lcu;
|
||||
if (temp == 0) {
|
||||
|
||||
temp_uv++;
|
||||
}
|
||||
// Store to temporary vector
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
|
||||
else{
|
||||
// Save starting point to memory
|
||||
start_point = (y_in_lcu)* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
|
||||
// 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);
|
||||
else if (temp == 1) {
|
||||
|
||||
temp_epi8_u = _mm256_permutevar8x32_epi32(temp_epi8_u, idx);
|
||||
temp_epi8_v = _mm256_permutevar8x32_epi32(temp_epi8_v, idx);
|
||||
// Pack the bits from 32-bit to 8-bit
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_epi32, temp_y_epi32), 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_u = _mm_loadu_si128((__m128i*)&temp_epi8_u);
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.u[start_point_uv]), final_epi8_u);
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.y[start_point]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
// 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 = 0;
|
||||
}
|
||||
|
||||
temp_uv = 0;
|
||||
}
|
||||
break;
|
||||
|
||||
default:
|
||||
if (temp_uv == 0) {
|
||||
break;
|
||||
|
||||
// Store to temporary vector
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
default:
|
||||
if (temp == 0) {
|
||||
|
||||
// Save starting point to memory
|
||||
start_point_uv = (y_in_lcu)* LCU_WIDTH_C + x_in_lcu;
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
|
||||
temp_uv++;
|
||||
}
|
||||
start_point = y_in_lcu* LCU_WIDTH + x_in_lcu;
|
||||
}
|
||||
|
||||
else if (temp_uv == 1) {
|
||||
else if (temp == 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++;
|
||||
}
|
||||
// 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_uv == 2) {
|
||||
else if (temp == 2) {
|
||||
temp_epi32 = temp_y_epi32;
|
||||
temp++;
|
||||
}
|
||||
|
||||
temp_epi32_u = temp_u_epi32;
|
||||
temp_epi32_v = temp_v_epi32;
|
||||
temp_uv++;
|
||||
}
|
||||
else {
|
||||
|
||||
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));
|
||||
// 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_u, idx);
|
||||
// Arrange the vector to right order before inserting it
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, 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.y[start_point]), temp_epi8);
|
||||
temp = 0;
|
||||
}
|
||||
}
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), final_epi8_256);
|
||||
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));
|
||||
|
||||
temp_uv = 0;
|
||||
}
|
||||
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_epi8_u = _mm256_packus_epi16(_mm256_packus_epi32(temp_u_epi32, zeros_256), zeros_256);
|
||||
|
||||
int8_t *temp_int_8_u = (int8_t*)&temp_epi8_u;
|
||||
|
||||
__m256i temp_epi8_v = _mm256_packus_epi16(_mm256_packus_epi32(temp_v_epi32, zeros_256), zeros_256);
|
||||
int8_t *temp_int_8_v = (int8_t*)&temp_epi8_v;
|
||||
|
||||
|
||||
|
||||
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, zeros_256), zeros_256);
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
// 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(_mm256_packus_epi32(temp_v_epi32, zeros_256), zeros_256);
|
||||
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
// 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:
|
||||
|
||||
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
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_epi32_u, temp_u_epi32), zeros_256);
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.u[start_point_uv]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(_mm256_packus_epi32(temp_epi32_v, temp_v_epi32), zeros_256);
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
_mm_storeu_si128((__m128i*)&(lcu->rec.v[start_point_uv]), _mm256_castsi256_si128(temp_epi8));
|
||||
|
||||
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
|
||||
temp_epi8 = _mm256_packus_epi16(temp_epi16_u, _mm256_packus_epi32(temp_epi32_u, temp_u_epi32));
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.u[start_point_uv]), temp_epi8);
|
||||
|
||||
temp_epi8 = _mm256_packus_epi16(temp_epi16_v, _mm256_packus_epi32(temp_epi32_v, temp_v_epi32));
|
||||
|
||||
// Arrange the vector to right order before inserting it
|
||||
temp_epi8 = _mm256_permutevar8x32_epi32(temp_epi8, idx);
|
||||
|
||||
// Store 256-bits of integer data into memory
|
||||
_mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), temp_epi8);
|
||||
|
||||
temp_uv = 0;
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif //COMPILE_INTEL_AVX2
|
||||
|
|
|
|||
|
|
@ -555,7 +555,30 @@ static void inter_recon_bipred_generic(const int hi_prec_luma_rec0,
|
|||
|
||||
int y_in_lcu;
|
||||
int x_in_lcu;
|
||||
/*
|
||||
y_in_lcu = ((ypos) & ((LCU_WIDTH)-1));
|
||||
x_in_lcu = ((xpos) & ((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)));
|
||||
|
||||
printf("%d ", sample0_y);
|
||||
printf("\n");
|
||||
|
||||
printf("%d ", sample1_y);
|
||||
printf("\n");
|
||||
|
||||
printf("%d ", shift);
|
||||
printf("\n");
|
||||
|
||||
printf("%d ", offset);
|
||||
printf("\n");
|
||||
|
||||
printf("%d ", (sample0_y+sample1_y+64) >> 7);
|
||||
printf("\n");
|
||||
|
||||
printf("%d ", (20+15+1) >> 1);
|
||||
printf("\n");*/
|
||||
|
||||
//After reconstruction, merge the predictors by taking an average of each pixel
|
||||
for (int temp_y = 0; temp_y < height; ++temp_y) {
|
||||
|
|
|
|||
|
|
@ -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 = 4;
|
||||
int height = 4;
|
||||
int16_t mv_param[2][2] = { { 3,3 },{ 3,3 } };
|
||||
int width = 32;
|
||||
int height = 32;
|
||||
int xpos = 0;
|
||||
int ypos = 0;
|
||||
|
||||
|
|
@ -89,11 +89,11 @@ static void setup()
|
|||
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_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;
|
||||
hi_prec_chroma_rec0 = mv_param[0][0] & 7 || mv_param[0][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_rec1) high_precision_rec1 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
|
||||
|
|
|
|||
Loading…
Reference in a new issue