Sole tbb executor #17851

anton-potapov · 2020-07-15T13:33:57Z

This Merge Request adds TBB based executor into GAPI

Only the executor itself, which is not yet used
The executor uses priority queue based logic for ready to execution tasks

TBB executor description

TBB executor engine operate with graph of tasks represented via cv::gimpl::parallel::tile_node instances.

Dependencies

For every task (tile_node ) there might be a number of input dependencies (i.e. tasks that should be executed before). This number is represented twice in the tile_node structure. They are dependencies and dependency_count. First one describes total number (i.e. input edges in the graph before execution) and does not change during execution, while second one describes current number of unresolved input dependencies. Once dependency_count drops to zero, task is ready for execution as all it's dependencies are satisfied

Dependents

Every task that directly depends on the current one is listed in the tile_node::dependees member. The engine decrees dependency_count of these tasks right after the completion of the current task.

Sync Vs. Async payload

Task's body in enclosed in either task_body or async_task_body members depending on task's nature (and async flag). While in both cases the engine calls passed in function there are two differences:

async one accept the callback (from the engine), which should be called when asynchronous activity started by this function completes
While for synchronous payloads engine treat the task completed after the function returns, for asynchronous ones task treated completed once the callback is called.

Execution and Priorities

Once all input dependencies for the task are satisfied (i.e. it's dependency_count drops to zero) it is treated as ready for execution and is put into the "ready" queue. The queue (tbb::concurrent_priority_queue) sorts all task according to their priorities.

Some TBB tasking Likbez

TBB uses tbb::task derivatives to describe a piece of work. They can be linked together into a tree describing tasks dependencies. Every TBB task has a refcount field describing unsatisfied dependencies. Once it drops to zero task is ready for execution (i.e. can be spawned). Dependency tasks are called child tasks (i.e. parent depends on it's childs). Every child task on creation increases it's parent's refcount. After child task is executed TBB task scheduler automatically decrease parent's refcount. Root of tasks tree is used only to call it's tbb::task::wait_for_all method to wait for the CPU work to complete. wait_for_all waits for root refcount value to become 1 as a sign that all work is done.

Mapping to TBB

The executor engine creates a root task (derived from tbb::task) - a root of the task's tree. Once new tile_node item is pushed to the ready queue, new child task of root's is created (Effectively increasing root dependency count). Every child task body does the following :

Pops ready for execution tile_node instance from the priority queue
Execute it's body
Decrease dependency_count for every dependent tile_node instance (via tile_node::dependees member)
Per every tile_node with dependency_count dropped to zero, spawn extra (tbb) child task.

Asynchronicity work and TBB active waiting

While waiting for the root task, TBB does not put to sleep the calling thread. Instead, it keeps searching for spawned (or enqueued) TBB tasks to do, assuming it somehow related with the root task.

However, in case of async work (like GPU or non TBB CPU work) there may be no such (tbb) tasks to do. In this case calling thread wasting CPU time, constantly searching for tasks or spin waiting. To overcome drawback, more elaborated scheme then plain calling to tbb::task::wait_for_all is needed.

The scheme

Master thread (i.e. one that enters cv::gimpl::parallel::execute) after graph execution ignition does the following loop:

if there are ready to go TBB tasks, call root->wait_for_all() effectively participating in those tasks execution
(wait_for_all() will return once all (soawned) childes of root complete)
if no such tasks exist, but there active async tasks, then wait on the conditional variable until :
- there is new TBB work to do (i.e. new tasks are spawned)
- last async task completes (and it's callback is called)

In order to track number of active asynchronous tasks count filed of exec_ctx::async_tasks is used.

Graph re-execution

The engine allows correct re-execution of already completed (tile_node based) graph instance. To allow it dependency_count of every executed tile_node is reset back to dependencies immediately after execution.

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

I agree to contribute to the project under OpenCV (BSD) License.
To the best of my knowledge, the proposed patch is not based on a code under GPL or other license that is incompatible with OpenCV
The PR is proposed to proper branch
There is reference to original bug report and related work
There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
The feature is well documented and sample code can be built with the project CMake

force_builders=ARMv7,Custom
build_image:Custom=centos:7
buildworker:Custom=linux-1
build_gapi_standalone:Custom=ade-0.1.1f
build_gapi_standalone:Linux x64=ade-0.1.1f
build_gapi_standalone:Win64=ade-0.1.1f
build_gapi_standalone:Mac=ade-0.1.1f
build_gapi_standalone:Linux x64 Debug=ade-0.1.1f

dmatveev · 2020-07-27T18:08:45Z

@anton-potapov please invite other reviewers from the team first

asmorkalov · 2020-07-29T06:17:40Z

@anton-potapov your turn

alalek · 2020-08-18T10:38:56Z

modules/gapi/src/executor/gapi_itt.hpp

+#include <type_traits>
+#include <memory>
+
+#ifdef OPENCV_WITH_ITT


This macro is not propagated to other modules except opencv_core

Hmmm.... So what is the right way to check for ITT here then ?

@alalek ping

The right way is reusing OpenCV's trace.hpp functionality for profiling purposes.

added a FIXME

alalek · 2020-08-18T10:41:36Z

modules/gapi/src/executor/gapi_itt.hpp

+
+    #define ITT_AUTO_TRACE_GUARD_IMPL_(LINE, h)        ITT_NAMED_TRACE_GUARD(itt_trace_guard_##LINE , h)
+    #define ITT_AUTO_TRACE_GUARD_IMPL(LINE, h)         ITT_AUTO_TRACE_GUARD_IMPL_(LINE, h)
+    #define ITT_AUTO_TRACE_GUARD(h)                    ITT_AUTO_TRACE_GUARD_IMPL(__LINE__ , h)


Be careful with definition macros with ITT_ prefixes.
Consider using of GAPI_ITT_ prefix or similar instead.

changed prefix to GAPI_ITT

alalek · 2020-08-18T11:02:01Z

modules/gapi/src/executor/gapi_tbb_executor.cpp

+      return (use_tbb_scheduler_bypass::yes ==  reuse_current_task) ? (recycle_as_continuation(), this) : nullptr;
+   }
+   ~functor_task(){
+      assert_graph_is_running(parent());


Be careful with exceptions in destructors.

Details: https://en.cppreference.com/w/cpp/error/uncaught_exception

thanks for the warning, but this intentionally is not a exception, but and assert either CV_ one or a hand-crafted one

Changed to be enabled in debug only

alalek · 2020-08-18T11:03:10Z

modules/gapi/src/executor/gapi_tbb_executor.cpp

+             if ((active_async_tasks == 0) || (wake_master == wake_tbb_master::yes) )//was the last or there is the new TBB tasks to execute
+             {
+                 ITT_AUTO_TRACE_GUARD(ittTbbUnlockMasterThread);
+                //Wile decrement of async_tasks_t::count is atomic it might be done after waiting thread checked it value but _before_ it actually start waiting on the condition variable.


broken indentation

+a too long line

Typo in while

dmatveev · 2020-09-03T14:46:18Z

modules/gapi/CMakeLists.txt

@@ -94,6 +94,7 @@ set(gapi_srcs

    # Executor
    src/executor/gexecutor.cpp
+    src/executor/gapi_tbb_executor.cpp


Everything here is gapi, so I'd recommend renaming it to gtbbexecutor

dmatveev · 2020-09-03T14:46:36Z

modules/gapi/CMakeLists.txt

+if(HAVE_TBB)
+  if(TARGET opencv_test_gapi)


dmatveev · 2020-09-03T14:47:19Z

modules/gapi/src/executor/gapi_itt.hpp

+//
+// Copyright (C) 2020 Intel Corporation
+
+#ifndef OPENCV_GAPI_GAPI_ITT_HPP


Should ITT be part of the /executor? maybe move it to backends/common?

added a FIXME

dmatveev · 2020-09-03T14:47:33Z

modules/gapi/src/executor/gapi_itt.hpp

+
+    #define ITT_AUTO_TRACE_GUARD_IMPL_(LINE, h)        ITT_NAMED_TRACE_GUARD(itt_trace_guard_##LINE , h)
+    #define ITT_AUTO_TRACE_GUARD_IMPL(LINE, h)         ITT_AUTO_TRACE_GUARD_IMPL_(LINE, h)
+    #define ITT_AUTO_TRACE_GUARD(h)                    ITT_AUTO_TRACE_GUARD_IMPL(__LINE__ , h)


dmatveev · 2020-09-03T14:48:02Z

modules/gapi/src/executor/gapi_itt.hpp

+    }
+}  //namespace util
+
+namespace gimpl { namespace parallel {


Though it has nothing with parallelism, right? :)

You are right :)

added a FIXME

dmatveev · 2020-09-03T14:49:06Z

modules/gapi/src/executor/gapi_tbb_executor.cpp

+
+}}}  // namespace cv::gimpl::parallel
+
+void cv::gimpl::parallel::execute(tbb::concurrent_priority_queue<tile_node* , tile_node_indirect_priority_comparator> & q)


Normally we don't have lines long like this

dmatveev · 2020-09-03T14:49:19Z

modules/gapi/src/executor/gapi_tbb_executor.cpp

+    //Specify tbb::task_group_context::concurrent_wait in the traits to ask TBB scheduler not to change ref_count of the task we wait on (root) when wait is complete.
+    //As the traits is last argument explicitly specify (default) value for first argument


And especially like this

dmatveev · 2020-09-03T14:50:00Z

modules/gapi/src/executor/gapi_tbb_executor.cpp

+             if ((active_async_tasks == 0) || (wake_master == wake_tbb_master::yes) )//was the last or there is the new TBB tasks to execute
+             {
+                 ITT_AUTO_TRACE_GUARD(ittTbbUnlockMasterThread);
+                //Wile decrement of async_tasks_t::count is atomic it might be done after waiting thread checked it value but _before_ it actually start waiting on the condition variable.


+a too long line

AsyaPronina

Implementation is very cool, thanks a lot for this!!

Comments are decorative, part of comments are just questions to understand code idea more deeply.

The main comment about code style - unaligned spaces, could you please align all indents in expressions, functions signatures and other statements? I haven't caught all unaligned spaces in comments because started to write feedback for them quite late..

The second comment about code style which is also was caught by Dmitry Matveev is too long comments and too long signatures/statements. Could you please break long comments, signatures and statements to multiple lines?

The third proposal is just about code documentation - may be we can split code by phases or highlight logical entities with comments? For example, highlight discussed logical entities with "Creating G-API static graph" (it is actually phase in the test, but still), "Creating TBB dynamic graph", "Main execution entry point for the each TBB task", "Execution of real tile node's body", "Spawning tasks for dynamic TBB graph", and so on?

modules/gapi/src/executor/gapi_tbb_executor.cpp

AsyaPronina · 2020-09-10T13:21:02Z

modules/gapi/src/executor/gapi_tbb_executor.cpp

+             if ((active_async_tasks == 0) || (wake_master == wake_tbb_master::yes) )//was the last or there is the new TBB tasks to execute
+             {
+                 ITT_AUTO_TRACE_GUARD(ittTbbUnlockMasterThread);
+                //Wile decrement of async_tasks_t::count is atomic it might be done after waiting thread checked it value but _before_ it actually start waiting on the condition variable.


Typo in while

modules/gapi/src/executor/gapi_tbb_executor.cpp

modules/gapi/src/executor/gapi_tbb_executor.hpp

dmatveev · 2020-10-05T08:59:28Z

Note there are failing tests:
https://pullrequest.opencv.org/buildbot/builders/precommit_custom_linux/builds/4182

modules/gapi/src/executor/gapi_tbb_executor.cpp

- the sole executor - unit tests for it - no usage in the GAPI at the momnet

- introduced new overload of execute to explicitly accept tbb::arena argument - added more basic tests - moved arena creation code into tests -

- fixed compie errors & warnings

- split all-in-one execute() function into logicaly independant parts

- used util::variant in in the tile_node

- moved copy_through_move to separate header - rearranged details staff in proper namespaces - moved all implementation into detail namespace

- fixed build error with TBB 4.4. - fixed build warnings

- aligned strings width - fixed spaces in expressions - fixed english grammar - minor improvements

- added more comments - minor improvements

- changed ITT_ prefix for macroses to GAPI_ITT

- no more "unused" warning for GAPI_DbgAssert - changed local assert macro to man onto GAPI_DbgAssert

- file renamings - changed local assert macro to man onto GAPI_DbgAsse

- test file renamed - add more comments

OrestChura

Great! I have some little suggestions, mostly typo-corrections

modules/gapi/include/opencv2/gapi/util/copy_through_move.hpp

modules/gapi/include/opencv2/gapi/own/assert.hpp

modules/gapi/src/executor/gapi_itt.hpp

modules/gapi/src/executor/gtbbexecutor.cpp

modules/gapi/test/executor/gtbbexecutor_internal_tests.cpp

- minor clenups and cosmetic changes

- changed spaces and curly braces alignment

- minor cleanups

anton-potapov · 2020-11-27T07:34:37Z

@alalek could you please merge it ?

alalek

Is there some performance test to check benefits?

Approve from G-API team (@dmatveev @rgarnov ) is required for merge.
I'm not able to check in detail G-API code internals (or it will took enormous amount of time which I don't have).

alalek · 2020-11-27T08:28:03Z

modules/gapi/include/opencv2/gapi/own/assert.hpp

@@ -2,16 +2,23 @@
 // It is subject to the license terms in the LICENSE file found in the top-level directory
 // of this distribution and at https://opencv.org/license.html.
 //
-// Copyright (C) 2018 Intel Corporation
+// Copyright (C) 2020 Intel Corporation


2018-2020 ?

alalek · 2020-11-27T08:31:48Z

modules/gapi/include/opencv2/gapi/own/assert.hpp

@@ -33,7 +40,7 @@ namespace detail


 #ifdef NDEBUG
-#  define GAPI_DbgAssert(expr)
+#  define GAPI_DbgAssert(expr) cv::util::suppress_unused_warning(expr)


AFAIK, expression is still should be computed by compiler (and then dead code is eliminated - but not all).

As suppress_unused_warning(expr) is the function call, it's actual argument i.e. expr is calculated, then however it's value is discarded as bound to unnamed function argument.

The suppress_unused_warning defined as:

template<typename T> void suppress_unused_warning( const T& ) {}

All functions and methods are not "pure" in C++ standard ("pure" is GCC extension).
So some non-inline .type() calls will be evaluated if they are part of this expr.

Please follow the classic C assert() implementation from assert.h / cassert headers (it is "dbg" assert in OpenCV):
https://en.cppreference.com/w/cpp/error/assert

Having the assert implemented like this does not prevent "unused" warnings (and that is why it was not enough to use CV_DbgAssert as is and I had to re-invent it. )

it seems that sizeof() is the solution here. As it is silence the warning and does not evaluate it's argument

it seems that sizeof() is the solution here.

But not for lambdas :) - fixed to use constexpr and bool operations short-circuit evaluation

alalek · 2020-11-27T08:36:34Z

modules/gapi/src/executor/gtbbexecutor.cpp

+
+enum class is_tbb_work_present {
+   no,
+   yes


By coding style constants / enums should be upper case.
NO is a macro on some platforms (Apple).

Consider using bool type instead. Enum doesn't look as necessary here.

using meaningful names instead of faceless true / false , so bool is not an option :)

will rename to UPPERCASE

anton-potapov · 2020-11-27T10:17:43Z

Is there some performance test to check benefits?

Not yet, as it is first part of the puzzle- the executor itself. Actual usage/integration is to come up

OrestChura

Great! Thanks!

dmatveev

OK with the TBB part, not very certain about macro changes, though

dmatveev · 2020-11-27T10:29:30Z

modules/gapi/include/opencv2/gapi/own/assert.hpp

+#  define GAPI_DbgAssert CV_DbgAssert
+#else
+#  define GAPI_DbgAssert(expr) cv::util::suppress_unused_warning(expr)
+#endif


So what's the reason for all these changes?

the reason is - compiler warnings on release builds on unused variables in asserts

So, this unused variable is still going to be evaluated. Why we need that?

It seems to me that actually evaluate the expression is the only reliable way to actually "use" variables it made of (and there fore silence those warnings).

And yes compiling optimization of dead code removal is crucial here

"unused"

problem is in caller code. Not in assertion macro.

Just replace to C-like assert() and you will receive the same "unused" warning in release builds.

Workaround for such cases:

-auto result = t->decrement_ref_count(); +auto result = t->decrement_ref_count(); (void)result; ASSERT(result >= 1);

Yes, almost every problem can be workarounded :) ... but why if there is a proper solution to it ?

if there is a proper solution

It still evaluates expression.
More complex and common cases looks like ASSERT(node->isValid());. Compiler's DCE can't remove that completely.

Assertion macro must not evaluate passed expression (should be empty) in case of "noop".
Do not confuse developers with new specific behavior.

Last version of the patch does not evaluate the expression (and not cause "unused" warning)

OK, it is a tricky way, but looks like it works.

AsyaPronina · 2020-11-29T05:05:47Z

modules/gapi/src/executor/gtbbexecutor.cpp

+               // First participate in execution of TBB graph till there are no more ready tasks.
+               ctx.root->wait_for_all();
+
+               if (!async_work_done()) { // Wait on the conditional variable iff there is active async work


iff -> if

https://en.wikipedia.org/wiki/If_and_only_if

AsyaPronina

Thanks a lot for the efforts!

alalek · 2020-11-29T23:41:26Z

modules/gapi/include/opencv2/gapi/util/copy_through_move.hpp

+{
+namespace util
+{
+    //This is a tool to move initialize captures of a lambda in C++11


Check gasync.cpp for similar stuff. We really don't want to have duplicated code.

If you do not mind, I will unify them in a separate PR.

alalek · 2020-11-29T23:54:03Z

modules/gapi/src/executor/gtbbexecutor.cpp

+#include <opencv2/gapi/util/copy_through_move.hpp>
+#include "logger.hpp" // GAPI_LOG
+
+#include <tbb/task.h>


BTW,

note: ‘#pragma message: TBB Warning: tbb/task.h is deprecated. For details, please see Deprecated Features appendix in the TBB reference manual.’

TBB (ver 2020.3 interface 11103)

thanks, I know this

alalek · 2020-11-30T00:17:13Z

modules/gapi/include/opencv2/gapi/own/assert.hpp

+#  define GAPI_DbgAssert CV_DbgAssert
+#else
+#  define GAPI_DbgAssert(expr) cv::util::suppress_unused_warning(expr)
+#endif


OK, it is a tricky way, but looks like it works.

alalek · 2020-11-30T00:24:38Z

modules/gapi/src/executor/gapi_itt.hpp

+#endif
+
+#include <opencv2/gapi/util/compiler_hints.hpp>
+#include <iostream>


iostream

Do we really need that here? (consider using <ostream> if std::ostream is required only without cout/cerr)

good catch - will change

- minor cleanups

anton-potapov · 2020-11-30T13:28:07Z

Great!

* TBB executor for GAPI - the sole executor - unit tests for it - no usage in the GAPI at the momnet * TBB executor for GAPI - introduced new overload of execute to explicitly accept tbb::arena argument - added more basic tests - moved arena creation code into tests - * TBB executor for GAPI - fixed compie errors & warnings * TBB executor for GAPI - split all-in-one execute() function into logicaly independant parts * TBB executor for GAPI - used util::variant in in the tile_node * TBB executor for GAPI - moved copy_through_move to separate header - rearranged details staff in proper namespaces - moved all implementation into detail namespace * TBB executor for GAPI - fixed build error with TBB 4.4. - fixed build warnings * TBB executor for GAPI - aligned strings width - fixed spaces in expressions - fixed english grammar - minor improvements * TBB executor for GAPI - added more comments - minor improvements * TBB executor for GAPI - changed ITT_ prefix for macroses to GAPI_ITT * TBB executor for GAPI - no more "unused" warning for GAPI_DbgAssert - changed local assert macro to man onto GAPI_DbgAssert * TBB executor for GAPI - file renamings - changed local assert macro to man onto GAPI_DbgAsse * TBB executor for GAPI - test file renamed - add more comments * TBB executor for GAPI - minor clenups and cosmetic changes * TBB executor for GAPI - minor clenups and cosmetic changes * TBB executor for GAPI - changed spaces and curly braces alignment * TBB executor for GAPI - minor cleanups * TBB executor for GAPI - minor cleanups

anton-potapov added the category: g-api / gapi label Jul 15, 2020

anton-potapov force-pushed the sole_tbb_executor branch from e348b3d to b612d42 Compare July 24, 2020 11:41

asmorkalov requested a review from dmatveev July 27, 2020 08:51

asmorkalov assigned dmatveev Jul 27, 2020

asmorkalov mentioned this pull request Jul 29, 2020

TBB executor for GAPI #17170

Closed

6 tasks

alalek reviewed Aug 18, 2020

View reviewed changes

anton-potapov requested review from AsyaPronina and OrestChura September 1, 2020 09:54

dmatveev reviewed Sep 3, 2020

View reviewed changes

AsyaPronina reviewed Sep 10, 2020

View reviewed changes

anton-potapov force-pushed the sole_tbb_executor branch 2 times, most recently from a97e530 to a02c2fc Compare October 13, 2020 09:38

AsyaPronina reviewed Oct 15, 2020

View reviewed changes

modules/gapi/src/executor/gapi_tbb_executor.cpp Outdated Show resolved Hide resolved

modules/gapi/src/executor/gapi_tbb_executor.cpp Outdated Show resolved Hide resolved

anton-potapov force-pushed the sole_tbb_executor branch from a02c2fc to 61a4933 Compare October 16, 2020 09:52

anton-potapov added 13 commits November 3, 2020 11:45

TBB executor for GAPI

43a3cd4

- the sole executor - unit tests for it - no usage in the GAPI at the momnet

TBB executor for GAPI

3e927a8

- introduced new overload of execute to explicitly accept tbb::arena argument - added more basic tests - moved arena creation code into tests -

TBB executor for GAPI

d5175c2

- fixed compie errors & warnings

TBB executor for GAPI

14a444a

- split all-in-one execute() function into logicaly independant parts

TBB executor for GAPI

d82ef0a

- used util::variant in in the tile_node

TBB executor for GAPI

7a16ecd

- moved copy_through_move to separate header - rearranged details staff in proper namespaces - moved all implementation into detail namespace

TBB executor for GAPI

af432b1

- fixed build error with TBB 4.4. - fixed build warnings

TBB executor for GAPI

8b300d7

- aligned strings width - fixed spaces in expressions - fixed english grammar - minor improvements

TBB executor for GAPI

1957dcb

- added more comments - minor improvements

TBB executor for GAPI

589c917

- changed ITT_ prefix for macroses to GAPI_ITT

TBB executor for GAPI

496185d

- no more "unused" warning for GAPI_DbgAssert - changed local assert macro to man onto GAPI_DbgAssert

TBB executor for GAPI

f52bfe2

- file renamings - changed local assert macro to man onto GAPI_DbgAsse

TBB executor for GAPI

a217a97

- test file renamed - add more comments

anton-potapov force-pushed the sole_tbb_executor branch from e96bf3f to a217a97 Compare November 5, 2020 11:39

OrestChura reviewed Nov 9, 2020

View reviewed changes

anton-potapov added 2 commits November 11, 2020 11:44

TBB executor for GAPI

7d1b310

- minor clenups and cosmetic changes

TBB executor for GAPI

598371d

- minor clenups and cosmetic changes

anton-potapov force-pushed the sole_tbb_executor branch from fcc2a44 to 598371d Compare November 12, 2020 10:17

TBB executor for GAPI

cd73aac

- changed spaces and curly braces alignment

anton-potapov force-pushed the sole_tbb_executor branch from 6b9e13b to cd73aac Compare November 18, 2020 09:53

TBB executor for GAPI

742dc63

- minor cleanups

alalek reviewed Nov 27, 2020

View reviewed changes

OrestChura approved these changes Nov 27, 2020

View reviewed changes

dmatveev approved these changes Nov 27, 2020

View reviewed changes

anton-potapov force-pushed the sole_tbb_executor branch 2 times, most recently from 63da882 to 35cf42a Compare November 27, 2020 21:07

AsyaPronina reviewed Nov 29, 2020

View reviewed changes

AsyaPronina approved these changes Nov 29, 2020

View reviewed changes

alalek reviewed Nov 30, 2020

View reviewed changes

TBB executor for GAPI

920d8d3

- minor cleanups

anton-potapov force-pushed the sole_tbb_executor branch from 35cf42a to 920d8d3 Compare November 30, 2020 07:07

This was referenced Nov 30, 2020

GAPI: reuse copy_through_move_t in the gasync.cpp file anton-potapov/opencv#1

Open

GAPI: Reuse move through copy #18967

Merged

anton-potapov unassigned dmatveev Nov 30, 2020

alalek assigned dmatveev Nov 30, 2020

alalek merged commit 95ce8f4 into opencv:master Nov 30, 2020

alalek mentioned this pull request Nov 30, 2020

G-API: sporadic test failures TBBExecutor.AsyncBasic #18974

Closed

alalek mentioned this pull request Apr 9, 2021

(5.x) Merge 4.x #19885

Merged


		}}} // namespace cv::gimpl::parallel

		void cv::gimpl::parallel::execute(tbb::concurrent_priority_queue<tile_node* , tile_node_indirect_priority_comparator> & q)

		//Specify tbb::task_group_context::concurrent_wait in the traits to ask TBB scheduler not to change ref_count of the task we wait on (root) when wait is complete.
		//As the traits is last argument explicitly specify (default) value for first argument

Uh oh!

Sole tbb executor #17851

Sole tbb executor #17851

Uh oh!

Conversation

anton-potapov commented Jul 15, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

This Merge Request adds TBB based executor into GAPI

TBB executor description

Dependencies

Dependents

Sync Vs. Async payload

Execution and Priorities

Some TBB tasking Likbez

Mapping to TBB

Asynchronicity work and TBB active waiting

The scheme

Graph re-execution

Pull Request Readiness Checklist

Uh oh!

dmatveev commented Jul 27, 2020

Uh oh!

asmorkalov commented Jul 29, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

anton-potapov commented Jul 15, 2020 •

edited

Loading