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.
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
Check if executors share the same memory #670
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
how about we still isolate the reference executor?
If we optimize omp executor with some additional information like csr in cuda/hip,
considering omp/reference as the same memory will make the conversion failed.
Awesome idea, this might also help later on with the MPI executor. Two comments before I review it in more detail:
|
I don't think this is such a big issue. If the algorithm needs to clone anyway,
Indeed, it does in two ways: 1) temporary clone is called after this PR only when there is no direct access to memory, and 2) we do not ensure anymore that LinOps have exactly the same executors when calling a function like
Sure, I don't see how else we would use |
50361e6
to
6acf8eb
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! But I have one question: This PR changes the behaviour of make_temporary_clone
and make_temporary_clone
seems to be part of the public interface.
Doesn't that mean we would be possibly breaking existing code ?
std::for_each(device_type_.begin(), device_type_.end(), | ||
[](char &c) { c = std::tolower(c); }); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
:)
I don't think changing the temporary clone interface is interface breaking, since the point of this function is to do a copy when you need it. This still happens now, all that changes is that you get less copies when the data is already sitting on memory that you can access. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Wow, this is an intricate piece of work.
My understanding of why we need a two-level dispatch:
Because we do not want to template Executor
on the concrete executor type, we have an ExecutorBase
template with CRTP, and this is a friend
of Executor
for all CRTP values. The verify_memory_to
functions, which do the actual work, depend on being overloaded on the concrete executor types. This means calling one of them requires knowledge of the concrete type. Therefore, Executor
cannot directly call verify_memory_to
, so it calls verify_memory_from
. This second function is not overloaded but is implemented in ExecutorBase
and thus has access to its concrete type, and can call verify_memory_to
of its argument, passing its concrete self as the argument in turn.
Is that correct? Does this get affected when actual memory space classes are introduced, or has the plan for that changed?
I have a minor reservation which is a point that @pratikvn raised, about consistency in using pointers and references for verify_memory_to
and verify_memory_from
.
@Slaedr yes you are pretty much exactly correct about that design, this is how it works. The idea of this double dispatch with the CRTP enabled class in the middle is to allow to evaluate to the actual executor type so that we can call the proper About the memory space classes, either this will stay with memory spaces instead of executors, or, more likely, we build the memory spaces properly so that they are shared when they should be, and anyway we don't need all of this anymore. About using pointers, yes I will change to that, this is a leftover of the previous design which I started changing in some recent commits, and had not done so for the deepest levels. |
So all this is a only temporary measure until the full memory spaces are implemented? How much of this will be retained? |
That is hard to say. First of all, the memory space is definitely interface breaking so it could be in a while still, whereas this is not. Removing this and replacing by memory spaces will be interface breaking as well, but not much more since memory spaces are interface breaking to begin with. Second, it depends as I said on how the memory space implementation is done. You could keep this together with memory space, if we somehow want to keep a CudaMemory and HipMemory for example, you would still need to say Cuda == HIP when HIP is ran on CUDA and both have the same device id. If it's done in another way, where HIP executor gets a CudaMemory space while using a CUDA backend, then in this case you don't need the verify memory overload. |
78f8a62
to
c9d73fd
Compare
Error: The following files need to be formatted:
You can find a formatting patch under Artifacts here or run |
Error: The following files need to be formatted:
You can find a formatting patch under Artifacts here or run |
c9d73fd
to
4f0ec9c
Compare
rebase! |
Error: The following files need to be formatted:
You can find a formatting patch under Artifacts here or run |
format! |
Codecov Report
@@ Coverage Diff @@
## develop #670 +/- ##
===========================================
- Coverage 93.00% 92.89% -0.11%
===========================================
Files 332 333 +1
Lines 24187 24265 +78
===========================================
+ Hits 22495 22541 +46
- Misses 1692 1724 +32
Continue to review full report at Codecov.
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! Just one minor question.
Co-authored-with: Terry Cojean <terry.cojean@kit.edu>
+ Do not use `operator==`, but a funciton `memory_accessible` instead. + Make DPC++ host and CPU be memory compatible. + Use pointers for the interface instead of references. + Ensure DPC++ tests always work. + Fix some typos. Co-authored-by: Yuhsiang M. Tsai <yhmtsai@gmail.com> Co-authored-by: Tobias Ribizel <ribizel@kit.edu> Co-authored-by: Pratik Nayak <pratikvn@protonmail.com> Co-authored-by: Aditya Kashi <aditya.kashi@kit.edu>
Co-authored-by: tcojean <tcojean@users.noreply.github.com>
727584b
to
bd4936f
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, some minor comments.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the comments, I will fix the issues pointed out.
Some code style issues. Co-authored-by: Thomas Grützmacher <thomas.gruetzmacher@kit.edu>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Thanks for handling this issue and add a lot of tests.
I have one question on the test.
we will make the reference executor isolated, right?
Could you remind me why we need the reference executor to have a different memory space from the omp executor? Do we document this somewhere? |
9dd05db
to
80297f3
Compare
we have some specific data in different executor, like csr srow. Another point is from testing or design. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! Just a minor question
80297f3
to
018244f
Compare
Kudos, SonarCloud Quality Gate passed! |
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
This PR contains only the first commit of #652.
This implements a new executor function
memory_accessible
between executors which verifies whether they share the same memory. Currently, this functionality is used in thetemporary_clone
function to save copies.This currently means that: