Extract allocations from OpenMP kernels #740

upsj · 2021-04-09T16:43:54Z

I'll just copy over the explanation from the source code:

The dependency on OpenMP needs to be public because well, it's a long story:
std::shared_ptr decides whether we need to increment/decrement the reference counters atomically by calling the pthread function __gthread_active_p, i.e. checking whether there are multiple active threads. Linking against OpenMP also links against pthread on Linux. Due to issues with static initialization order for pthread_key_create, if some piece of code doesn't link against pthread, it may happen (I didn't want to dive too deep into that pit) that ginkgo_omp assumes that a piece of code is being executed sequentially, causing race conditions in the reference counters. See easybuilders/easybuild-easyconfigs#12563

Extract allocations from Jacobi OpenMP kernels
Extract allocations from CSR SpGEMM OpenMP kernels (this means switching the algorithm to a multiway merge-based implementation with a binary heap) based on Add custom SpGEMM implementation #626
Add a bit of explanation to the SpGEMM implementation
~~Extract allocations from RCM OpenMP kernels~~ I think this is a bit too much to ask for now
Prevent any potential problems with std::shared_ptr refcount atomics due to pthread linking
Add more tests for SpGEMM

Fixes #732

tcojean · 2021-04-09T17:07:55Z

I'm not sure what the impact would be of letting OpenMP creep up everywhere, how about making only -pthread public instead?

Some more considerations. Compare:

My understanding is that, either we :

Support OpenMP 5.1 only, and then we can use C++11 features safely
Support lesser standards, but we need to ban std::shared_ptr (due to underlying use of atomics) inside OpenMP blocks, and possibly review the RCM code as well.

upsj · 2021-04-09T17:38:31Z

So like adding

find_package(Threads REQUIRED)
target_link_libraries(ginkgo_omp PUBLIC Threads::Threads)

CMake should remove duplicate linker flags, right? Or in the worst case, it shouldn't matter if we link against pthread twice?

upsj · 2021-04-09T17:41:47Z

The issue with OpenMP 5.1 would be that this problem isn't caused by C++/OpenMP interaction, but by dynamic linking issues.

tcojean · 2021-04-09T17:43:39Z

Sure, technically the root of the problem seems to be static variable initialization, but we still use C++ features which technically are UB unless we have a strict requirement for OpenMP 5.1. So either we don't require anything but things could break on various settings (I wonder how Power PC/IBM compilers would fare), or we do something about this as well.

thoasm

LGTM!

omp/matrix/csr_kernels.cpp

pratikvn

LGTM!

omp/matrix/csr_kernels.cpp

upsj · 2021-04-20T08:21:55Z

Before we can merge this, there is still the issue with clang-cuda and HIP dependencies on pthread: https://gitlab.com/ginkgo-project/ginkgo-public-ci/-/jobs/1167965921

This can sometimes affect std::shared_ptr refcount atomics

Co-authored-by: Thomas Grützmacher <thomas.gruetzmacher@kit.edu> Co-authored-by: Pratik Nayak <pratik.nayak@kit.edu>

sonarqubecloud · 2021-05-04T10:23:18Z

Kudos, SonarCloud Quality Gate passed!

0 Bugs
0 Vulnerabilities
0 Security Hotspots
15 Code Smells

0.0% Coverage
0.5% Duplication

codecov · 2021-05-04T10:30:24Z

Codecov Report

Merging #740 (e18a183) into develop (7c7fc42) will increase coverage by 1.24%.
The diff coverage is 96.33%.

@@             Coverage Diff             @@
##           develop     #740      +/-   ##
===========================================
+ Coverage    92.81%   94.05%   +1.24%     
===========================================
  Files          392      392              
  Lines        30613    30647      +34     
===========================================
+ Hits         28412    28825     +413     
+ Misses        2201     1822     -379

Impacted Files	Coverage Δ
omp/matrix/csr_kernels.cpp	`94.77% <94.28%> (-1.10%)`	⬇️
omp/preconditioner/jacobi_kernels.cpp	`94.79% <100.00%> (+0.15%)`	⬆️
omp/test/matrix/csr_kernels.cpp	`100.00% <100.00%> (ø)`
devices/machine_topology.cpp	`82.71% <0.00%> (-2.47%)`	⬇️
omp/reorder/rcm_kernels.cpp	`97.53% <0.00%> (-0.61%)`	⬇️
reference/factorization/par_ilut_kernels.cpp	`100.00% <0.00%> (+0.49%)`	⬆️
core/test/utils/assertions.hpp	`72.84% <0.00%> (+1.35%)`	⬆️
omp/test/factorization/par_ilut_kernels.cpp	`98.90% <0.00%> (+98.90%)`	⬆️
omp/factorization/par_ilut_kernels.cpp	`100.00% <0.00%> (+100.00%)`	⬆️

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 7c7fc42...e18a183. Read the comment docs.

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

upsj added the is:bug Something looks wrong. label Apr 9, 2021

upsj requested review from pratikvn, Slaedr, thoasm, yhmtsai, tcojean and fritzgoebel April 9, 2021 16:43

upsj self-assigned this Apr 9, 2021

ginkgo-bot added mod:openmp This is related to the OpenMP module. reg:build This is related to the build system. labels Apr 9, 2021

upsj added the 1:ST:ready-for-review This PR is ready for review label Apr 9, 2021

upsj force-pushed the fix_omp_link_issue branch from 6cedf36 to da3e4ca Compare April 10, 2021 00:48

upsj added 1:ST:WIP This PR is a work in progress. Not ready for review. and removed 1:ST:ready-for-review This PR is ready for review labels Apr 10, 2021

upsj changed the title ~~Fix race conditions in std::shared_ptr refcounts~~ Extract allocations from OpenMP kernels Apr 10, 2021

upsj force-pushed the fix_omp_link_issue branch from da3e4ca to 25b94ce Compare April 10, 2021 01:09

upsj force-pushed the fix_omp_link_issue branch from 25b94ce to ed57244 Compare April 19, 2021 09:54

upsj added 1:ST:ready-for-review This PR is ready for review and removed 1:ST:WIP This PR is a work in progress. Not ready for review. labels Apr 19, 2021

thoasm approved these changes Apr 20, 2021

View reviewed changes

pratikvn approved these changes Apr 20, 2021

View reviewed changes

omp/matrix/csr_kernels.cpp Show resolved Hide resolved

upsj added 1:ST:do-not-merge Please do not merge PR this yet. and removed 1:ST:ready-for-review This PR is ready for review labels Apr 20, 2021

upsj added 2 commits May 4, 2021 10:05

fix linking issues with OpenMP and pthread

c67ce57

This can sometimes affect std::shared_ptr refcount atomics

extract allocations from Jacobi OpenMP loops

51909e1

upsj added 4 commits May 4, 2021 10:05

extract allocations from CSR OpenMP

25dc0ce

add SpGEMM documentation

22f54d4

add more OpenMP CSR SpGEMM tests

226a0e4

remove hip-nvcc from clang-cuda builds

d64bbd0

upsj force-pushed the fix_omp_link_issue branch from ed57244 to d64bbd0 Compare May 4, 2021 08:08

upsj added 1:ST:ready-to-merge This PR is ready to merge. and removed 1:ST:do-not-merge Please do not merge PR this yet. labels May 4, 2021

review update

e18a183

Co-authored-by: Thomas Grützmacher <thomas.gruetzmacher@kit.edu> Co-authored-by: Pratik Nayak <pratik.nayak@kit.edu>

upsj merged commit bd33473 into develop May 4, 2021

upsj deleted the fix_omp_link_issue branch May 4, 2021 12:41

This was referenced Mar 18, 2022

Adds support for MPI distributed matrix #971

Merged

Thread-safety of Ginkgo #996

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Extract allocations from OpenMP kernels #740

Extract allocations from OpenMP kernels #740

upsj commented Apr 9, 2021 •

edited

Loading

tcojean commented Apr 9, 2021

upsj commented Apr 9, 2021

upsj commented Apr 9, 2021 •

edited

Loading

tcojean commented Apr 9, 2021

thoasm left a comment

pratikvn left a comment

upsj commented Apr 20, 2021

sonarqubecloud bot commented May 4, 2021

codecov bot commented May 4, 2021 •

edited

Loading

Extract allocations from OpenMP kernels #740

Extract allocations from OpenMP kernels #740

Conversation

upsj commented Apr 9, 2021 • edited Loading

tcojean commented Apr 9, 2021

upsj commented Apr 9, 2021

upsj commented Apr 9, 2021 • edited Loading

tcojean commented Apr 9, 2021

thoasm left a comment

Choose a reason for hiding this comment

pratikvn left a comment

Choose a reason for hiding this comment

upsj commented Apr 20, 2021

sonarqubecloud bot commented May 4, 2021

codecov bot commented May 4, 2021 • edited Loading

Codecov Report

upsj commented Apr 9, 2021 •

edited

Loading

upsj commented Apr 9, 2021 •

edited

Loading

codecov bot commented May 4, 2021 •

edited

Loading