forked from HandBrake/HandBrake
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
contrib: backported commit from svt-av1 master to fix performance reg…
- Loading branch information
Showing
1 changed file
with
139 additions
and
0 deletions.
There are no files selected for viewing
139 changes: 139 additions & 0 deletions
139
contrib/svt-av1/A01-fix-performance-regression-for-systems-with-multiple.patch
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,139 @@ | ||
| From 4579ddcfe28d56aca15cd6bb932ed35f59e9fca7 Mon Sep 17 00:00:00 2001 | ||
| From: Artem Galin <[email protected]> | ||
| Date: Tue, 2 Apr 2024 23:40:35 +0000 | ||
| Subject: [PATCH] Fix performance regression for systems with multiple | ||
| processor groups | ||
|
|
||
| --- | ||
| Source/Lib/Common/Codec/EbThreads.h | 24 +++++++++++++----------- | ||
| Source/Lib/Decoder/Codec/EbDecHandle.c | 3 ++- | ||
| Source/Lib/Decoder/Codec/EbDecProcess.c | 1 + | ||
| Source/Lib/Encoder/Globals/EbEncHandle.c | 21 ++++++++++++++------- | ||
| 4 files changed, 30 insertions(+), 19 deletions(-) | ||
|
|
||
| diff --git a/Source/Lib/Common/Codec/EbThreads.h b/Source/Lib/Common/Codec/EbThreads.h | ||
| index b6c69ba9..5d79a957 100644 | ||
| --- a/Source/Lib/Common/Codec/EbThreads.h | ||
| +++ b/Source/Lib/Common/Codec/EbThreads.h | ||
| @@ -56,17 +56,19 @@ extern EbErrorType svt_block_on_mutex(EbHandle mutex_handle); | ||
| extern EbErrorType svt_destroy_mutex(EbHandle mutex_handle); | ||
| #ifdef _WIN32 | ||
|
|
||
| -#define EB_CREATE_THREAD(pointer, thread_function, thread_context) \ | ||
| - do { \ | ||
| - pointer = svt_create_thread(thread_function, thread_context); \ | ||
| - EB_ADD_MEM(pointer, 1, EB_THREAD); \ | ||
| - if (num_groups == 1) \ | ||
| - SetThreadAffinityMask(pointer, svt_aom_group_affinity.Mask); \ | ||
| - else if (num_groups == 2 && alternate_groups) { \ | ||
| - svt_aom_group_affinity.Group = 1 - svt_aom_group_affinity.Group; \ | ||
| - SetThreadGroupAffinity(pointer, &svt_aom_group_affinity, NULL); \ | ||
| - } else if (num_groups == 2 && !alternate_groups) \ | ||
| - SetThreadGroupAffinity(pointer, &svt_aom_group_affinity, NULL); \ | ||
| +#define EB_CREATE_THREAD(pointer, thread_function, thread_context) \ | ||
| + do { \ | ||
| + pointer = svt_create_thread(thread_function, thread_context); \ | ||
| + EB_ADD_MEM(pointer, 1, EB_THREAD); \ | ||
| + if (svt_aom_group_affinity_enabled) { \ | ||
| + if (num_groups == 1) \ | ||
| + SetThreadAffinityMask(pointer, svt_aom_group_affinity.Mask); \ | ||
| + else if (num_groups == 2 && alternate_groups) { \ | ||
| + svt_aom_group_affinity.Group = 1 - svt_aom_group_affinity.Group; \ | ||
| + SetThreadGroupAffinity(pointer, &svt_aom_group_affinity, NULL); \ | ||
| + } else if (num_groups == 2 && !alternate_groups) \ | ||
| + SetThreadGroupAffinity(pointer, &svt_aom_group_affinity, NULL); \ | ||
| + } \ | ||
| } while (0) | ||
|
|
||
| #else | ||
| diff --git a/Source/Lib/Decoder/Codec/EbDecHandle.c b/Source/Lib/Decoder/Codec/EbDecHandle.c | ||
| index 643b4c0b..74c901dd 100644 | ||
| --- a/Source/Lib/Decoder/Codec/EbDecHandle.c | ||
| +++ b/Source/Lib/Decoder/Codec/EbDecHandle.c | ||
| @@ -50,7 +50,8 @@ | ||
| #ifdef _WIN32 | ||
| uint8_t num_groups = 0; | ||
| GROUP_AFFINITY svt_aom_group_affinity; | ||
| -Bool alternate_groups = 0; | ||
| +Bool alternate_groups = 0; | ||
| +int8_t svt_aom_group_affinity_enabled = 0; | ||
| #elif defined(__linux__) | ||
| cpu_set_t svt_aom_group_affinity; | ||
| #endif | ||
| diff --git a/Source/Lib/Decoder/Codec/EbDecProcess.c b/Source/Lib/Decoder/Codec/EbDecProcess.c | ||
| index dbca6ef2..72b1a6f5 100644 | ||
| --- a/Source/Lib/Decoder/Codec/EbDecProcess.c | ||
| +++ b/Source/Lib/Decoder/Codec/EbDecProcess.c | ||
| @@ -45,6 +45,7 @@ | ||
| extern uint8_t num_groups; | ||
| extern GROUP_AFFINITY svt_aom_group_affinity; | ||
| extern Bool alternate_groups; | ||
| +extern uint8_t svt_aom_group_affinity_enabled; | ||
| #elif defined(__linux__) | ||
| extern cpu_set_t svt_aom_group_affinity; | ||
| #endif | ||
| diff --git a/Source/Lib/Encoder/Globals/EbEncHandle.c b/Source/Lib/Encoder/Globals/EbEncHandle.c | ||
| index be262d38..34566723 100644 | ||
| --- a/Source/Lib/Encoder/Globals/EbEncHandle.c | ||
| +++ b/Source/Lib/Encoder/Globals/EbEncHandle.c | ||
| @@ -109,6 +109,7 @@ | ||
| **************************************/ | ||
| static uint8_t num_groups = 0; | ||
| #ifdef _WIN32 | ||
| +static uint8_t svt_aom_group_affinity_enabled = 0; | ||
| static GROUP_AFFINITY svt_aom_group_affinity; | ||
| static Bool alternate_groups = 0; | ||
| #elif defined(__linux__) | ||
| @@ -175,9 +176,7 @@ static const char *get_asm_level_name_str(EbCpuFlags cpu_flags) { | ||
| //Get Number of logical processors | ||
| static uint32_t get_num_processors() { | ||
| #ifdef _WIN32 | ||
| - SYSTEM_INFO sysinfo; | ||
| - GetSystemInfo(&sysinfo); | ||
| - return num_groups == 1 ? sysinfo.dwNumberOfProcessors : sysinfo.dwNumberOfProcessors << 1; | ||
| + return GetActiveProcessorCount(ALL_PROCESSOR_GROUPS); | ||
| #else | ||
| return sysconf(_SC_NPROCESSORS_ONLN); | ||
| #endif | ||
| @@ -248,6 +247,7 @@ uint64_t get_affinity_mask(uint32_t lpnum) { | ||
| void svt_set_thread_management_parameters(EbSvtAv1EncConfiguration *config_ptr) | ||
| { | ||
| #ifdef _WIN32 | ||
| + svt_aom_group_affinity_enabled = 1; | ||
| const uint32_t num_logical_processors = get_num_processors(); | ||
| // For system with a single processor group(no more than 64 logic processors all together) | ||
| // Affinity of the thread can be set to one or more logical processors | ||
| @@ -262,8 +262,9 @@ void svt_set_thread_management_parameters(EbSvtAv1EncConfiguration *config_ptr) | ||
| svt_aom_group_affinity.Group = config_ptr->target_socket; | ||
| } | ||
| else { | ||
| - const uint32_t num_lp_per_group = num_logical_processors / num_groups; | ||
| if (config_ptr->target_socket == -1) { | ||
| + // target socket is not set, use current group | ||
| + const uint32_t num_lp_per_group = GetActiveProcessorCount(svt_aom_group_affinity.Group); | ||
| if (config_ptr->logical_processors > num_lp_per_group) { | ||
| alternate_groups = TRUE; | ||
| SVT_WARN("-lp(logical processors) setting is ignored. Run on both sockets. \n"); | ||
| @@ -272,10 +273,16 @@ void svt_set_thread_management_parameters(EbSvtAv1EncConfiguration *config_ptr) | ||
| svt_aom_group_affinity.Mask = get_affinity_mask(config_ptr->logical_processors); | ||
| } | ||
| else { | ||
| - const uint32_t lps = | ||
| + // run on target socket only | ||
| + if (config_ptr->target_socket < num_groups) { | ||
| + const uint32_t num_lp_per_group = GetActiveProcessorCount(config_ptr->target_socket); | ||
| + const uint32_t lps = | ||
| config_ptr->logical_processors < num_lp_per_group ? config_ptr->logical_processors : num_lp_per_group; | ||
| - svt_aom_group_affinity.Mask = get_affinity_mask(lps); | ||
| - svt_aom_group_affinity.Group = config_ptr->target_socket; | ||
| + svt_aom_group_affinity.Mask = get_affinity_mask(lps); | ||
| + svt_aom_group_affinity.Group = config_ptr->target_socket; | ||
| + } | ||
| + else | ||
| + SVT_WARN("target socket setting is ignored. \n"); | ||
| } | ||
| } | ||
| } | ||
| -- | ||
| 2.25.1 | ||
|
|