Add gpu timer and use seconds in benchmark by yhmtsai · Pull Request #669 · ginkgo-project/ginkgo

yhmtsai · 2020-11-24T09:45:20Z

This PR adds gpu timer (based on event of cuda/hip) in benchmark.
--gpu_timer=true/false to choose which timer to use.
The gpu timer requires the first and the last op are on GPU.

UPDATED This PR also changes the timing result from nanoseconds to seconds

I only add it for major time measurement.
For the component time (detail), I keep original code because they usually contain some cpu operations.

Slaedr · 2020-11-24T14:04:29Z

+        tic_called_ = true;
+    }
+
+    std::size_t toc()


Why is it nice to have the return type as size_t? I see that Cuda and Hip timers actually return float which are then cast to size_t. Maybe the return type can just be kept as double?

It is based on the chrono library.
In chrono, we can count the number of nanosecond, so I use size_t to represent it.

Okay. But it does not seem like good practice to cast a float to size_t... though the other way round (or better yet to double) is not too bad. [Side note: cppreference says nanosecond is a signed 64 bit type, not size_t].

I don't think it is a good idea to have toc() return something because it is not obvious what it returns and in any case, the return value is not really useful. It could be any of these return values:

The number of times the time has already been recorded

the average runtime up until this point

the sum of the runtime until this point

the runtime of the latest measurement (you have chosen this one)

Additionally, it is not clear that the unit returned is nanoseconds. All in all, I would recommend to have this function not return anything.
If you want to know the latest result, I would prefer an additional function (std::chrono does the same).

pratikvn

LGTM! Cleans up a lot of the benchmarking code!

pratikvn · 2020-11-27T10:37:27Z

-                std::chrono::duration_cast<std::chrono::nanoseconds>(g_tac -
-                                                                     g_tic) /
-                FLAGS_repetitions;
+                generate_timer->get_total_time() / FLAGS_repetitions;


I guess you can use the get_average_time here as well ?

The timer is out of the repetition loop, so total_time = average_time here .
To get the correct average time, need total_time/FLAGS_repetitions or average_time/FLAGS_repetitions.
total_time is less confusing than average_time?

pratikvn · 2020-11-27T10:38:25Z

+#include "hip/base/device_guard.hip.hpp"
+
+
+#endif  // HAS_CUDA


Suggested change

#endif // HAS_CUDA

#endif // HAS_HIP

pratikvn · 2020-11-27T10:40:00Z

+public:
+    void tic()
+    {
+        assert(tic_called_ == false);


Not sure if this is useful, as if someone puts a #define NDEBUG at some point, it gets optimized out when in Release. Maybe you can define your own simple GKO_ASSERT macro ?

I guess the point would be to avoid this check in release, so as to disturb the timing as little as possible? In case someone suspects issues, they can run in debug and check. Admittedly, the extra assert would not usually matter for the timing, but still, you never know how it might be used later.

pratikvn · 2020-11-27T10:45:22Z

+
+    std::size_t get_total_time() { return total_duration_ns_; }
+
+    std::size_t get_tictoc_num() { return duration_ns_.size(); }


Maybe you can rename this to num_repetitions or something like that ?

I agree, that would be more self-explanatory.

thoasm

When using the script, you currently can't measure with GPU time.

thoasm · 2020-11-27T16:31:55Z

-                FLAGS_repetitions;
-            add_or_set_member(this_precond_data["apply"], "time",
-                              apply_time.count(), allocator);
+            auto apply_time = apply_timer->get_total_time() / FLAGS_repetitions;


Suggested change

auto apply_time = apply_timer->get_total_time() / FLAGS_repetitions;

auto apply_time = apply_timer->get_average_time();

Since this is outside the loop and the timer doesn't know about FLAGS_repetitions, wouldn't this give a wrong result? (equivalent to get_total_time()).

The apply_timer does know how often it was started and stopped, therefore, it is independent of the FLAGS_repetitions (and in my opinion safer) when using get_average_time().

It's independent only if you put it inside the for loop, AFAIK it is not here. See the context of the line highlighted:
https://github.com/ginkgo-project/ginkgo/pull/669/files#diff-d573d9686e53b5ce4b41de8061f46d1a30693d2dfa59795bec0680f6bdd4e4dcR178-R183

You also get less timing overhead if you put it outside the for loops when you can, as you will synchronize only once instead of at every iteration.

We put the timer out of for-loop, so the apply_timer can not know the number of for-loop. As terry said, I keep the current timer for less overhead. If we need to refill/reset the x,b like spmv or solver, we can put the timer inside the for-loop

thoasm · 2020-11-27T17:13:16Z

+        tic_called_ = true;
+    }
+
+    std::size_t toc()


I don't think it is a good idea to have toc() return something because it is not obvious what it returns and in any case, the return value is not really useful. It could be any of these return values:

The number of times the time has already been recorded

the average runtime up until this point

the sum of the runtime until this point

the runtime of the latest measurement (you have chosen this one)

Additionally, it is not clear that the unit returned is nanoseconds. All in all, I would recommend to have this function not return anything.
If you want to know the latest result, I would prefer an additional function (std::chrono does the same).

thoasm · 2020-11-27T17:14:22Z

+
+    std::size_t get_total_time() { return total_duration_ns_; }
+
+    std::size_t get_tictoc_num() { return duration_ns_.size(); }


I agree, that would be more self-explanatory.

codecov · 2020-11-28T00:38:59Z

Codecov Report

Merging #669 (a149032) into develop (2a951ac) will increase coverage by 0.02%.
The diff coverage is n/a.

@@             Coverage Diff             @@
##           develop     #669      +/-   ##
===========================================
+ Coverage    92.87%   92.89%   +0.02%     
===========================================
  Files          333      333              
  Lines        24266    24265       -1     
===========================================
+ Hits         22537    22541       +4     
+ Misses        1729     1724       -5

Impacted Files	Coverage Δ
omp/reorder/rcm_kernels.cpp	`98.13% <0.00%> (+3.07%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2a951ac...a149032. Read the comment docs.

Suggested changes have been made (but not yet fully reviewed by me).

Slaedr

Looks great! I would still prefer that toc etc. return double, but int64_t is also reasonable so I'll leave it to you.

tcojean

LGTM. Please add documentations though for the the different classes and public API. Also, please add some documentation to BENCHMARKING.md.

tcojean · 2020-11-30T16:30:01Z

-                FLAGS_repetitions;
-            add_or_set_member(this_precond_data["apply"], "time",
-                              apply_time.count(), allocator);
+            auto apply_time = apply_timer->get_total_time() / FLAGS_repetitions;


Since this is outside the loop and the timer doesn't know about FLAGS_repetitions, wouldn't this give a wrong result? (equivalent to get_total_time()).

tcojean · 2020-11-30T16:37:08Z

+            "executor is cuda or hip");
+
+
+class Timer {


Should we use the gko namespace?

Should I put it in gko:: namespace? I put the timer only in benchmark, so I am not sure whether it is suitable with gko namespace

thoasm

I agree with @tcojean: I would like to have more documentation.

thoasm · 2020-12-01T00:53:00Z

+        }
+#endif  // HAS_HIP
+    }
+    // Not use gpu_timer or not cuda/hip executor


nit:

Suggested change

// Not use gpu_timer or not cuda/hip executor

// No cuda/hip executor available or no gpu_timer used

Slaedr · 2020-12-01T11:42:50Z

+        auto duration_time =
+            std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start_)
+                .count();
+        return static_cast<std::int64_t>(duration_time);


I guess a cast is not required here anymore.

Suggested change

return static_cast<std::int64_t>(duration_time);

return duration_time;

yhmtsai · 2020-12-07T21:18:03Z

format!

tcojean

LGTM.

- fix gpu_timer in script - use int64_t - add const to function - update documentation - add get_latest_time - rename get_average_time -> compute_average_time Co-authored-by: Aditya Kashi <aditya.kashi@kit.edu> Co-authored-by: Pratik Nayak <pratikvn@protonmail.com> Co-authored-by: Terry Cojean <terry.cojean@kit.edu> Co-authored-by: Thomas Grützmacher <thomas.gruetzmacher@kit.edu>

Slaedr · 2020-12-15T08:33:29Z

+    {
+        exec_->synchronize();
+        auto stop = std::chrono::steady_clock::now();
+        auto duration_time = get_duration_in_seconds(stop - start_);


We could just use the standard way of doing this as shown here: https://en.cppreference.com/w/cpp/chrono/steady_clock/now
We don't really need an entire function and file to do this.

Suggested change

auto duration_time = get_duration_in_seconds(stop - start_);

std::chrono::duration<double> duration_time = stop - start_;

thanks for the example, I did not know that before.

Slaedr · 2020-12-15T08:34:52Z

+        exec_->synchronize();
+        auto stop = std::chrono::steady_clock::now();
+        auto duration_time = get_duration_in_seconds(stop - start_);
+        return duration_time;


Suggested change

return duration_time;

return duration_time.count();

Co-authored-by: Aditya Kashi <aditya.kashi@kit.edu>

sonarqubecloud · 2020-12-17T19:48:59Z

Kudos, SonarCloud Quality Gate passed!

0 Bugs
0 Vulnerabilities
0 Security Hotspots
10 Code Smells

0.0% Coverage
0.0% Duplication

Ginkgo release 1.4.0 The Ginkgo team is proud to announce the new Ginkgo minor release 1.4.0. This release brings most of the Ginkgo functionality to the Intel DPC++ ecosystem which enables Intel-GPU and CPU execution. The only Ginkgo features which have not been ported yet are some preconditioners. Ginkgo's mixed-precision support is greatly enhanced thanks to: 1. The new Accessor concept, which allows writing kernels featuring on-the-fly memory compression, among other features. The accessor can be used as header-only, see the [accessor BLAS benchmarks repository](https://github.com/ginkgo-project/accessor-BLAS/tree/develop) as a usage example. 2. All LinOps now transparently support mixed-precision execution. By default, this is done through a temporary copy which may have a performance impact but already allows mixed-precision research. Native mixed-precision ELL kernels are implemented which do not see this cost. The accessor is also leveraged in a new CB-GMRES solver which allows for performance improvements by compressing the Krylov basis vectors. Many other features have been added to Ginkgo, such as reordering support, a new IDR solver, Incomplete Cholesky preconditioner, matrix assembly support (only CPU for now), machine topology information, and more! Supported systems and requirements: + For all platforms, cmake 3.13+ + C++14 compliant compiler + Linux and MacOS + gcc: 5.3+, 6.3+, 7.3+, all versions after 8.1+ + clang: 3.9+ + Intel compiler: 2018+ + Apple LLVM: 8.0+ + CUDA module: CUDA 9.0+ + HIP module: ROCm 3.5+ + DPC++ module: Intel OneAPI 2021.3. Set the CXX compiler to `dpcpp`. + Windows + MinGW and Cygwin: gcc 5.3+, 6.3+, 7.3+, all versions after 8.1+ + Microsoft Visual Studio: VS 2019 + CUDA module: CUDA 9.0+, Microsoft Visual Studio + OpenMP module: MinGW or Cygwin. Algorithm and important feature additions: + Add a new DPC++ Executor for SYCL execution and other base utilities [#648](#648), [#661](#661), [#757](#757), [#832](#832) + Port matrix formats, solvers and related kernels to DPC++. For some kernels, also make use of a shared kernel implementation for all executors (except Reference). [#710](#710), [#799](#799), [#779](#779), [#733](#733), [#844](#844), [#843](#843), [#789](#789), [#845](#845), [#849](#849), [#855](#855), [#856](#856) + Add accessors which allow multi-precision kernels, among other things. [#643](#643), [#708](#708) + Add support for mixed precision operations through apply in all LinOps. [#677](#677) + Add incomplete Cholesky factorizations and preconditioners as well as some improvements to ILU. [#672](#672), [#837](#837), [#846](#846) + Add an AMGX implementation and kernels on all devices but DPC++. [#528](#528), [#695](#695), [#860](#860) + Add a new mixed-precision capability solver, Compressed Basis GMRES (CB-GMRES). [#693](#693), [#763](#763) + Add the IDR(s) solver. [#620](#620) + Add a new fixed-size block CSR matrix format (for the Reference executor). [#671](#671), [#730](#730) + Add native mixed-precision support to the ELL format. [#717](#717), [#780](#780) + Add Reverse Cuthill-McKee reordering [#500](#500), [#649](#649) + Add matrix assembly support on CPUs. [#644](#644) + Extends ISAI from triangular to general and spd matrices. [#690](#690) Other additions: + Add the possibility to apply real matrices to complex vectors. [#655](#655), [#658](#658) + Add functions to compute the absolute of a matrix format. [#636](#636) + Add symmetric permutation and improve existing permutations. [#684](#684), [#657](#657), [#663](#663) + Add a MachineTopology class with HWLOC support [#554](#554), [#697](#697) + Add an implicit residual norm criterion. [#702](#702), [#818](#818), [#850](#850) + Row-major accessor is generalized to more than 2 dimensions and a new "block column-major" accessor has been added. [#707](#707) + Add an heat equation example. [#698](#698), [#706](#706) + Add ccache support in CMake and CI. [#725](#725), [#739](#739) + Allow tuning and benchmarking variables non intrusively. [#692](#692) + Add triangular solver benchmark [#664](#664) + Add benchmarks for BLAS operations [#772](#772), [#829](#829) + Add support for different precisions and consistent index types in benchmarks. [#675](#675), [#828](#828) + Add a Github bot system to facilitate development and PR management. [#667](#667), [#674](#674), [#689](#689), [#853](#853) + Add Intel (DPC++) CI support and enable CI on HPC systems. [#736](#736), [#751](#751), [#781](#781) + Add ssh debugging for Github Actions CI. [#749](#749) + Add pipeline segmentation for better CI speed. [#737](#737) Changes: + Add a Scalar Jacobi specialization and kernels. [#808](#808), [#834](#834), [#854](#854) + Add implicit residual log for solvers and benchmarks. [#714](#714) + Change handling of the conjugate in the dense dot product. [#755](#755) + Improved Dense stride handling. [#774](#774) + Multiple improvements to the OpenMP kernels performance, including COO, an exclusive prefix sum, and more. [#703](#703), [#765](#765), [#740](#740) + Allow specialization of submatrix and other dense creation functions in solvers. [#718](#718) + Improved Identity constructor and treatment of rectangular matrices. [#646](#646) + Allow CUDA/HIP executors to select allocation mode. [#758](#758) + Check if executors share the same memory. [#670](#670) + Improve test install and smoke testing support. [#721](#721) + Update the JOSS paper citation and add publications in the documentation. [#629](#629), [#724](#724) + Improve the version output. [#806](#806) + Add some utilities for dim and span. [#821](#821) + Improved solver and preconditioner benchmarks. [#660](#660) + Improve benchmark timing and output. [#669](#669), [#791](#791), [#801](#801), [#812](#812) Fixes: + Sorting fix for the Jacobi preconditioner. [#659](#659) + Also log the first residual norm in CGS [#735](#735) + Fix BiCG and HIP CSR to work with complex matrices. [#651](#651) + Fix Coo SpMV on strided vectors. [#807](#807) + Fix segfault of extract_diagonal, add short-and-fat test. [#769](#769) + Fix device_reset issue by moving counter/mutex to device. [#810](#810) + Fix `EnableLogging` superclass. [#841](#841) + Support ROCm 4.1.x and breaking HIP_PLATFORM changes. [#726](#726) + Decreased test size for a few device tests. [#742](#742) + Fix multiple issues with our CMake HIP and RPATH setup. [#712](#712), [#745](#745), [#709](#709) + Cleanup our CMake installation step. [#713](#713) + Various simplification and fixes to the Windows CMake setup. [#720](#720), [#785](#785) + Simplify third-party integration. [#786](#786) + Improve Ginkgo device arch flags management. [#696](#696) + Other fixes and improvements to the CMake setup. [#685](#685), [#792](#792), [#705](#705), [#836](#836) + Clarification of dense norm documentation [#784](#784) + Various development tools fixes and improvements [#738](#738), [#830](#830), [#840](#840) + Make multiple operators/constructors explicit. [#650](#650), [#761](#761) + Fix some issues, memory leaks and warnings found by MSVC. [#666](#666), [#731](#731) + Improved solver memory estimates and consistent iteration counts [#691](#691) + Various logger improvements and fixes [#728](#728), [#743](#743), [#754](#754) + Fix for ForwardIterator requirements in iterator_factory. [#665](#665) + Various benchmark fixes. [#647](#647), [#673](#673), [#722](#722) + Various CI fixes and improvements. [#642](#642), [#641](#641), [#795](#795), [#783](#783), [#793](#793), [#852](#852) Related PR: #857

Release 1.4.0 to master The Ginkgo team is proud to announce the new Ginkgo minor release 1.4.0. This release brings most of the Ginkgo functionality to the Intel DPC++ ecosystem which enables Intel-GPU and CPU execution. The only Ginkgo features which have not been ported yet are some preconditioners. Ginkgo's mixed-precision support is greatly enhanced thanks to: 1. The new Accessor concept, which allows writing kernels featuring on-the-fly memory compression, among other features. The accessor can be used as header-only, see the [accessor BLAS benchmarks repository](https://github.com/ginkgo-project/accessor-BLAS/tree/develop) as a usage example. 2. All LinOps now transparently support mixed-precision execution. By default, this is done through a temporary copy which may have a performance impact but already allows mixed-precision research. Native mixed-precision ELL kernels are implemented which do not see this cost. The accessor is also leveraged in a new CB-GMRES solver which allows for performance improvements by compressing the Krylov basis vectors. Many other features have been added to Ginkgo, such as reordering support, a new IDR solver, Incomplete Cholesky preconditioner, matrix assembly support (only CPU for now), machine topology information, and more! Supported systems and requirements: + For all platforms, cmake 3.13+ + C++14 compliant compiler + Linux and MacOS + gcc: 5.3+, 6.3+, 7.3+, all versions after 8.1+ + clang: 3.9+ + Intel compiler: 2018+ + Apple LLVM: 8.0+ + CUDA module: CUDA 9.0+ + HIP module: ROCm 3.5+ + DPC++ module: Intel OneAPI 2021.3. Set the CXX compiler to `dpcpp`. + Windows + MinGW and Cygwin: gcc 5.3+, 6.3+, 7.3+, all versions after 8.1+ + Microsoft Visual Studio: VS 2019 + CUDA module: CUDA 9.0+, Microsoft Visual Studio + OpenMP module: MinGW or Cygwin. Algorithm and important feature additions: + Add a new DPC++ Executor for SYCL execution and other base utilities [#648](#648), [#661](#661), [#757](#757), [#832](#832) + Port matrix formats, solvers and related kernels to DPC++. For some kernels, also make use of a shared kernel implementation for all executors (except Reference). [#710](#710), [#799](#799), [#779](#779), [#733](#733), [#844](#844), [#843](#843), [#789](#789), [#845](#845), [#849](#849), [#855](#855), [#856](#856) + Add accessors which allow multi-precision kernels, among other things. [#643](#643), [#708](#708) + Add support for mixed precision operations through apply in all LinOps. [#677](#677) + Add incomplete Cholesky factorizations and preconditioners as well as some improvements to ILU. [#672](#672), [#837](#837), [#846](#846) + Add an AMGX implementation and kernels on all devices but DPC++. [#528](#528), [#695](#695), [#860](#860) + Add a new mixed-precision capability solver, Compressed Basis GMRES (CB-GMRES). [#693](#693), [#763](#763) + Add the IDR(s) solver. [#620](#620) + Add a new fixed-size block CSR matrix format (for the Reference executor). [#671](#671), [#730](#730) + Add native mixed-precision support to the ELL format. [#717](#717), [#780](#780) + Add Reverse Cuthill-McKee reordering [#500](#500), [#649](#649) + Add matrix assembly support on CPUs. [#644](#644) + Extends ISAI from triangular to general and spd matrices. [#690](#690) Other additions: + Add the possibility to apply real matrices to complex vectors. [#655](#655), [#658](#658) + Add functions to compute the absolute of a matrix format. [#636](#636) + Add symmetric permutation and improve existing permutations. [#684](#684), [#657](#657), [#663](#663) + Add a MachineTopology class with HWLOC support [#554](#554), [#697](#697) + Add an implicit residual norm criterion. [#702](#702), [#818](#818), [#850](#850) + Row-major accessor is generalized to more than 2 dimensions and a new "block column-major" accessor has been added. [#707](#707) + Add an heat equation example. [#698](#698), [#706](#706) + Add ccache support in CMake and CI. [#725](#725), [#739](#739) + Allow tuning and benchmarking variables non intrusively. [#692](#692) + Add triangular solver benchmark [#664](#664) + Add benchmarks for BLAS operations [#772](#772), [#829](#829) + Add support for different precisions and consistent index types in benchmarks. [#675](#675), [#828](#828) + Add a Github bot system to facilitate development and PR management. [#667](#667), [#674](#674), [#689](#689), [#853](#853) + Add Intel (DPC++) CI support and enable CI on HPC systems. [#736](#736), [#751](#751), [#781](#781) + Add ssh debugging for Github Actions CI. [#749](#749) + Add pipeline segmentation for better CI speed. [#737](#737) Changes: + Add a Scalar Jacobi specialization and kernels. [#808](#808), [#834](#834), [#854](#854) + Add implicit residual log for solvers and benchmarks. [#714](#714) + Change handling of the conjugate in the dense dot product. [#755](#755) + Improved Dense stride handling. [#774](#774) + Multiple improvements to the OpenMP kernels performance, including COO, an exclusive prefix sum, and more. [#703](#703), [#765](#765), [#740](#740) + Allow specialization of submatrix and other dense creation functions in solvers. [#718](#718) + Improved Identity constructor and treatment of rectangular matrices. [#646](#646) + Allow CUDA/HIP executors to select allocation mode. [#758](#758) + Check if executors share the same memory. [#670](#670) + Improve test install and smoke testing support. [#721](#721) + Update the JOSS paper citation and add publications in the documentation. [#629](#629), [#724](#724) + Improve the version output. [#806](#806) + Add some utilities for dim and span. [#821](#821) + Improved solver and preconditioner benchmarks. [#660](#660) + Improve benchmark timing and output. [#669](#669), [#791](#791), [#801](#801), [#812](#812) Fixes: + Sorting fix for the Jacobi preconditioner. [#659](#659) + Also log the first residual norm in CGS [#735](#735) + Fix BiCG and HIP CSR to work with complex matrices. [#651](#651) + Fix Coo SpMV on strided vectors. [#807](#807) + Fix segfault of extract_diagonal, add short-and-fat test. [#769](#769) + Fix device_reset issue by moving counter/mutex to device. [#810](#810) + Fix `EnableLogging` superclass. [#841](#841) + Support ROCm 4.1.x and breaking HIP_PLATFORM changes. [#726](#726) + Decreased test size for a few device tests. [#742](#742) + Fix multiple issues with our CMake HIP and RPATH setup. [#712](#712), [#745](#745), [#709](#709) + Cleanup our CMake installation step. [#713](#713) + Various simplification and fixes to the Windows CMake setup. [#720](#720), [#785](#785) + Simplify third-party integration. [#786](#786) + Improve Ginkgo device arch flags management. [#696](#696) + Other fixes and improvements to the CMake setup. [#685](#685), [#792](#792), [#705](#705), [#836](#836) + Clarification of dense norm documentation [#784](#784) + Various development tools fixes and improvements [#738](#738), [#830](#830), [#840](#840) + Make multiple operators/constructors explicit. [#650](#650), [#761](#761) + Fix some issues, memory leaks and warnings found by MSVC. [#666](#666), [#731](#731) + Improved solver memory estimates and consistent iteration counts [#691](#691) + Various logger improvements and fixes [#728](#728), [#743](#743), [#754](#754) + Fix for ForwardIterator requirements in iterator_factory. [#665](#665) + Various benchmark fixes. [#647](#647), [#673](#673), [#722](#722) + Various CI fixes and improvements. [#642](#642), [#641](#641), [#795](#795), [#783](#783), [#793](#793), [#852](#852) Related PR: #866

yhmtsai added reg:benchmarking This is related to benchmarking. 1:ST:ready-for-review This PR is ready for review labels Nov 24, 2020

yhmtsai requested review from Slaedr, fritzgoebel, hartwiganzt, pratikvn, tcojean, thoasm and upsj November 24, 2020 09:45

yhmtsai self-assigned this Nov 24, 2020

Slaedr reviewed Nov 24, 2020

View reviewed changes

Comment thread benchmark/utils/timer.hpp Outdated

Slaedr reviewed Nov 24, 2020

View reviewed changes

Comment thread benchmark/utils/timer.hpp Outdated

Slaedr reviewed Nov 24, 2020

View reviewed changes

Comment thread benchmark/utils/timer.hpp Outdated

pratikvn approved these changes Nov 27, 2020

View reviewed changes

thoasm previously requested changes Nov 27, 2020

View reviewed changes

yhmtsai force-pushed the gpu_timer branch from d0d6b18 to cf24880 Compare November 27, 2020 20:50

thoasm self-requested a review November 30, 2020 04:39

Slaedr approved these changes Nov 30, 2020

View reviewed changes

tcojean approved these changes Nov 30, 2020

View reviewed changes

thoasm reviewed Dec 1, 2020

View reviewed changes

Slaedr reviewed Dec 1, 2020

View reviewed changes

yhmtsai force-pushed the gpu_timer branch 2 times, most recently from 01581b1 to 3ecb031 Compare December 7, 2020 21:15

tcojean approved these changes Dec 11, 2020

View reviewed changes

yhmtsai added 2 commits December 11, 2020 15:43

timer for benchmark

fe0ee42

add timer gflags and use timer in benchmark

f9085b2

yhmtsai force-pushed the gpu_timer branch from cc4321c to 93af397 Compare December 11, 2020 14:43

use second and double precision of timing result

52a9d81

yhmtsai force-pushed the gpu_timer branch from d049a09 to 52a9d81 Compare December 14, 2020 20:01

Slaedr reviewed Dec 15, 2020

View reviewed changes

yhmtsai changed the title ~~Add gpu timer for benchmark~~ Add gpu timer and use seconds in benchmark Dec 15, 2020

yhmtsai force-pushed the gpu_timer branch from e28b43c to f4bebad Compare December 17, 2020 13:24

yhmtsai added 1:ST:ready-to-merge This PR is ready to merge. and removed 1:ST:ready-for-review This PR is ready for review labels Dec 17, 2020

use standard way to output duration in seconds

a149032

Co-authored-by: Aditya Kashi <aditya.kashi@kit.edu>

yhmtsai force-pushed the gpu_timer branch from f4bebad to a149032 Compare December 17, 2020 13:42

yhmtsai merged commit 297d732 into develop Dec 17, 2020

yhmtsai deleted the gpu_timer branch December 17, 2020 22:51

tcojean mentioned this pull request Jun 23, 2021

Benchmarks auto repetitions #791

Merged

2 tasks

yhmtsai mentioned this pull request Mar 29, 2023

Add support for device timers in summary hook #1313

Merged


		std::size_t get_total_time() { return total_duration_ns_; }

		std::size_t get_tictoc_num() { return duration_ns_.size(); }

	auto apply_time = apply_timer->get_total_time() / FLAGS_repetitions;
	auto apply_time = apply_timer->get_average_time();

	// Not use gpu_timer or not cuda/hip executor
	// No cuda/hip executor available or no gpu_timer used

	return static_cast<std::int64_t>(duration_time);
	return duration_time;

	auto duration_time = get_duration_in_seconds(stop - start_);
	std::chrono::duration<double> duration_time = stop - start_;

Conversation

yhmtsai commented Nov 24, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

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!

Uh oh!

Uh oh!

Uh oh!

pratikvn left a comment

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!

thoasm left a comment

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!

tcojean Dec 3, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

codecov Bot commented Nov 28, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Slaedr left a comment

Choose a reason for hiding this comment

Uh oh!

tcojean left a comment

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!

thoasm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yhmtsai commented Nov 24, 2020 •

edited

Loading

tcojean Dec 3, 2020 •

edited

Loading

codecov Bot commented Nov 28, 2020 •

edited

Loading