Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Do not use AVX2 instructions if the CPU doesn't support it #67

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Do not use AVX2 instructions if the CPU doesn't support it #67

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
eb2f70b
First check CPUID, and then use xgetbv
edwintorok Oct 16, 2019
0dc9806
Use non-AVX2 codepaths on non-AVX2 CPUs
edwintorok Oct 16, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
50 changes: 25 additions & 25 deletions Source/Lib/Codec/EbEncDecProcess.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -463,7 +463,7 @@ void perform_coding_loop(
switch (tx_size) {
case TX_32X32:
if (!(is_encode_pass && context_ptr->skip_eob_zero_mode_ep && *eob == 0)) {
residual_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][32 >> 3](
residual_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][32 >> 3](
input_buffer,
input_stride,
pred_buffer,
Expand Down Expand Up @@ -510,7 +510,7 @@ void perform_coding_loop(
}
if (context_ptr->spatial_sse_full_loop || (is_encode_pass && do_recon)) {
// Hsan: both pred and rec samples are needed @ MD and EP to perform the eob zero mode decision
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][4](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][4](
pred_buffer,
pred_stride,
recon_buffer,
Expand All @@ -530,7 +530,7 @@ void perform_coding_loop(

case TX_16X16:
if (!(is_encode_pass && context_ptr->skip_eob_zero_mode_ep && *eob == 0)) {
residual_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][16 >> 3](
residual_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][16 >> 3](
input_buffer,
input_stride,
pred_buffer,
Expand Down Expand Up @@ -570,7 +570,7 @@ void perform_coding_loop(
if (context_ptr->spatial_sse_full_loop || (is_encode_pass && do_recon)) {

// Hsan: both pred and rec samples are needed @ MD and EP to perform the eob zero mode decision
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][2](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][2](
pred_buffer,
pred_stride,
recon_buffer,
Expand All @@ -592,7 +592,7 @@ void perform_coding_loop(

case TX_8X8:
if (!(is_encode_pass && context_ptr->skip_eob_zero_mode_ep && *eob == 0)) {
residual_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][8 >> 3](
residual_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][8 >> 3](
input_buffer,
input_stride,
pred_buffer,
Expand Down Expand Up @@ -632,7 +632,7 @@ void perform_coding_loop(
if (context_ptr->spatial_sse_full_loop || (is_encode_pass && do_recon)) {

// Hsan: both pred and rec samples are needed @ MD and EP to perform the eob zero mode decision
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][1](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][1](
pred_buffer,
pred_stride,
recon_buffer,
Expand All @@ -655,7 +655,7 @@ void perform_coding_loop(
default:
assert(tx_size == TX_4X4);
if (!(is_encode_pass && context_ptr->skip_eob_zero_mode_ep && *eob == 0)) {
residual_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][4 >> 3](
residual_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][4 >> 3](
input_buffer,
input_stride,
pred_buffer,
Expand Down Expand Up @@ -701,7 +701,7 @@ void perform_coding_loop(
}
if (context_ptr->spatial_sse_full_loop || (is_encode_pass && do_recon)) {
// Hsan: both pred and rec samples are needed @ MD and EP to perform the eob zero mode decision
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][0](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][0](
pred_buffer,
pred_stride,
recon_buffer,
Expand Down Expand Up @@ -761,7 +761,7 @@ void perform_inv_trans_add(
switch (tx_size) {
case TX_32X32:

pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][4](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][4](
pred_buffer,
pred_stride,
recon_buffer,
Expand All @@ -781,7 +781,7 @@ void perform_inv_trans_add(

case TX_16X16:

pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][2](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][2](
pred_buffer,
pred_stride,
recon_buffer,
Expand All @@ -802,7 +802,7 @@ void perform_inv_trans_add(
break;

case TX_8X8:
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][1](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][1](
pred_buffer,
pred_stride,
recon_buffer,
Expand All @@ -824,7 +824,7 @@ void perform_inv_trans_add(

default:
assert(tx_size == TX_4X4);
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][0](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][0](
pred_buffer,
pred_stride,
recon_buffer,
Expand Down Expand Up @@ -926,15 +926,15 @@ static void perform_dist_rate_calc(
int tu_size = 1 << (2 + tx_size);

if (context_ptr->spatial_sse_full_loop) {
tufull_distortion[DIST_CALC_RESIDUAL] = (int)spatialfull_distortion_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][tx_size](
tufull_distortion[DIST_CALC_RESIDUAL] = (int)spatialfull_distortion_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][tx_size](
input_buffer,
input_stride,
recon_buffer,
recon_stride,
tu_size,
tu_size);

tufull_distortion[DIST_CALC_PREDICTION] = (int)spatialfull_distortion_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][tx_size](
tufull_distortion[DIST_CALC_PREDICTION] = (int)spatialfull_distortion_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][tx_size](
input_buffer,
input_stride,
pred_buffer,
Expand All @@ -948,7 +948,7 @@ static void perform_dist_rate_calc(
else {
const int shift = tx_size == TX_32X32 ? 0 : 2;
uint64_t tufull_distortionTemp[DIST_CALC_TOTAL];
full_distortion_intrinsic_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][*eob != 0][0][tu_size >> 3](
full_distortion_intrinsic_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][*eob != 0][0][tu_size >> 3](
trans_coeff_buffer,
tu_size,
recon_coeff_buffer,
Expand Down Expand Up @@ -987,7 +987,7 @@ static void perform_dist_rate_calc(
tufull_distortion[DIST_CALC_RESIDUAL] = (tufull_distortion[DIST_CALC_PREDICTION]);
*tu_coeff_bits = 0;
if (*eob) {
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][tu_size == 32 ? 4 : tx_size](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][tu_size == 32 ? 4 : tx_size](
pred_buffer,
pred_stride,
recon_buffer,
Expand Down Expand Up @@ -1245,7 +1245,7 @@ static void perform_dist_rate_calc(
candidate_buffer->candidate_ptr->eob[0][tu_index] = 0;
int arr_index = tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1;
if (arr_index >= 0 && arr_index < 9)
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][arr_index](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][arr_index](
&(candidate_buffer->prediction_ptr->buffer_y[context_ptr->block_origin_index + pred_recon_tu_origin_index]),
candidate_buffer->prediction_ptr->stride_y,
&(candidate_buffer->recon_ptr->buffer_y[context_ptr->block_origin_index + pred_recon_tu_origin_index]),
Expand All @@ -1259,7 +1259,7 @@ static void perform_dist_rate_calc(
candidate_buffer->candidate_ptr->eob[1][0] = 0;
int arr_index = tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1;
if (arr_index >= 0 && arr_index < 9)
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][arr_index](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][arr_index](
&(candidate_buffer->prediction_ptr->buffer_cb[context_ptr->block_chroma_origin_index]),
candidate_buffer->prediction_ptr->stride_cb,
&(candidate_buffer->recon_ptr->buffer_cb[context_ptr->block_chroma_origin_index]),
Expand All @@ -1272,7 +1272,7 @@ static void perform_dist_rate_calc(
candidate_buffer->candidate_ptr->eob[2][0] = 0;
int arr_index = tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1;
if (arr_index >= 0 && arr_index < 9)
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][arr_index](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][arr_index](
&(candidate_buffer->prediction_ptr->buffer_cr[context_ptr->block_chroma_origin_index]),
candidate_buffer->prediction_ptr->stride_cr,
&(candidate_buffer->recon_ptr->buffer_cr[context_ptr->block_chroma_origin_index]),
Expand Down Expand Up @@ -1914,7 +1914,7 @@ static void perform_dist_rate_calc(
{
(void)sb_ptr;
if ((context_ptr->ep_block_stats_ptr->sq_size >> 3) < 9) {
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][context_ptr->ep_block_stats_ptr->sq_size >> 3](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][context_ptr->ep_block_stats_ptr->sq_size >> 3](
&(recon_src_ptr->buffer_y[context_ptr->ep_block_stats_ptr->origin_x + context_ptr->ep_block_stats_ptr->origin_y * recon_src_ptr->stride_y]),
recon_src_ptr->stride_y,
&(recon_dst_ptr->buffer_y[context_ptr->ep_block_stats_ptr->origin_x + context_ptr->ep_block_stats_ptr->origin_y * recon_dst_ptr->stride_y]),
Expand All @@ -1926,15 +1926,15 @@ static void perform_dist_rate_calc(
uint16_t chromaorigin_x = context_ptr->ep_block_stats_ptr->origin_x >> 1;
uint16_t chromaorigin_y = context_ptr->ep_block_stats_ptr->origin_y >> 1;

pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][context_ptr->ep_block_stats_ptr->sq_size_uv >> 3](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][context_ptr->ep_block_stats_ptr->sq_size_uv >> 3](
&(recon_src_ptr->buffer_cb[chromaorigin_x + chromaorigin_y * recon_src_ptr->stride_cb]),
recon_src_ptr->stride_cb,
&(recon_dst_ptr->buffer_cb[chromaorigin_x + chromaorigin_y * recon_dst_ptr->stride_cb]),
recon_dst_ptr->stride_cb,
context_ptr->ep_block_stats_ptr->sq_size_uv,
context_ptr->ep_block_stats_ptr->sq_size_uv);

pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][context_ptr->ep_block_stats_ptr->sq_size_uv >> 3](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][context_ptr->ep_block_stats_ptr->sq_size_uv >> 3](
&(recon_src_ptr->buffer_cr[chromaorigin_x + chromaorigin_y * recon_src_ptr->stride_cr]),
recon_src_ptr->stride_cr,
&(recon_dst_ptr->buffer_cr[chromaorigin_x + chromaorigin_y * recon_dst_ptr->stride_cr]),
Expand Down Expand Up @@ -4215,7 +4215,7 @@ static void perform_dist_rate_calc(
uint32_t pred_tu_origin_index = ((tu_index % 2) * tu_size[context_ptr->ep_block_stats_ptr->tx_size]) + ((tu_index > 1) * tu_size[context_ptr->ep_block_stats_ptr->tx_size] * context_ptr->prediction_buffer->stride_y);
context_ptr->block_ptr->eob[0][tu_index] = 0;
if (do_recon)
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][tu_size[context_ptr->ep_block_stats_ptr->tx_size] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][tu_size[context_ptr->ep_block_stats_ptr->tx_size] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size](
&(context_ptr->prediction_buffer->buffer_y[context_ptr->block_origin_index + pred_tu_origin_index]),
context_ptr->prediction_buffer->stride_y,
&(context_ptr->recon_buffer->buffer_y[cuOriginReconIndex + pred_recon_tu_origin_index]),
Expand All @@ -4227,7 +4227,7 @@ static void perform_dist_rate_calc(
if (context_ptr->block_ptr->eob[1][0]) {
context_ptr->block_ptr->eob[1][0] = 0;
if (do_recon)
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1](
&((context_ptr->prediction_buffer->buffer_cb)[context_ptr->block_chroma_origin_index]),
context_ptr->prediction_buffer->stride_cb,
&(context_ptr->recon_buffer->buffer_cb[cuChromaOriginReconIndex]),
Expand All @@ -4238,7 +4238,7 @@ static void perform_dist_rate_calc(
if (context_ptr->block_ptr->eob[2][0]) {
context_ptr->block_ptr->eob[2][0] = 0;
if (do_recon)
pic_copy_kernel_func_ptr_array[(ASM_TYPES & PREAVX2_MASK) && 1][tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1](
pic_copy_kernel_func_ptr_array[(ASM_TYPES & AVX2_MASK) && 1][tu_size[context_ptr->ep_block_stats_ptr->tx_size_uv] == 32 ? 4 : context_ptr->ep_block_stats_ptr->tx_size - 1](
&((context_ptr->prediction_buffer->buffer_cr)[context_ptr->block_chroma_origin_index]),
context_ptr->prediction_buffer->stride_cr,
&(context_ptr->recon_buffer->buffer_cr[cuChromaOriginReconIndex]),
Expand Down
5 changes: 3 additions & 2 deletions Source/Lib/Codec/EbEncHandle.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -239,12 +239,13 @@ static int32_t can_use_intel_avx512()
| (1 << 30) // AVX-512BW
| (1 << 31); // AVX-512VL

if (!check4th_gen_intel_core_features())
return 0;

// ensure OS supports ZMM registers (and YMM, and XMM)
if (!check_xcr0_zmm())
return 0;

if (!check4th_gen_intel_core_features())
return 0;

run_cpuid(7, 0, abcd);
if ((abcd[1] & avx512_ebx_mask) != avx512_ebx_mask)
Expand Down
Loading