Format conversion, SpMV and write improvements #905

upsj · 2021-10-15T15:26:17Z

This PR fixes a few issues found while developing #904:

Fix edge case behavior of Csr SpMV sparselib, fall back to classical if necessary instead of throwing NotImplemented exception.
Fix edge case behavior of Dense GEMM for empty matrices
Simplify and unify format conversions from and to Dense/Csr
Re-use existing storage in conversions to Dense/Ell if the dimensions match
Initialize arrays in Csr to valid values in all constructors
Use a more modern CMake CUDA include path variable, remove unecessary uses of CUDA includes
Add a few missing conversions
Fix Sellp SpMV for non-default slice sizes
Guard against empty CUDA kernel launches
Remove unnecessary uses of dim3 for 1D grids

A few words on the format conversions:

All conversions from Csr to Ell/Sellp/Hybrid read non-coalescing and write coalescing memory, because the output uses more storage (padding) and the formats use transposed storage (Csr: row-major, Ell, Sellp: column-major)
All conversions from Dense to other formats read coalescing and write potentially non-coalescing memory, because except for dense matrices, we read much more memory than we write

tcojean · 2021-10-21T15:29:35Z

As part of this, should we also set classical to be the default strategy anyway, seeing that we cannot rely on it in some cases (esp. mixed precision use cases)?

Slaedr · 2021-10-22T08:52:19Z

I'm not sure it's a good idea to initialize the arrays in all constructors. I think we should have an option of allocating without initializing. Though that kind of goes against the <insert correct C++ principle here>, for a HPC code I think we should have that option. Depending on the size of the matrix (and given a good system state), allocation could be instantaneous in comparison to initialization.

pratikvn

Thanks for adding this. It mostly looks good. Some small comments and questions.

common/unified/matrix/coo_kernels.cpp

hip/matrix/dense_kernels.hip.cpp

include/ginkgo/core/matrix/csr.hpp

reference/matrix/dense_kernels.cpp

reference/matrix/hybrid_kernels.cpp

core/matrix/sellp.cpp

cuda/matrix/csr_kernels.cu

common/unified/matrix/dense_kernels.cpp

common/unified/matrix/hybrid_kernels.cpp

common/unified/matrix/coo_kernels.cpp

upsj · 2022-01-30T22:27:09Z

@pratikvn some of your suggestions involve functional changes not immediately related to my changes - would you be okay with addressing them in another PR instead of extending this one?

* remove unnecessary kernel already present in format_conversion * fix include structure * fix Hybrid::resize * remove unnecessary kernel launch guards * add HIP kernel launch guards * simplify 2D kernel launch guards * add tests for format conversion components Co-authored-by: Yuhsiang Tsai <[email protected]> Co-authored-by: Pratik Nayak <[email protected]> Co-authored-by: Thomas Grützmacher <[email protected]>

Co-authored-by: Pratik Nayak <[email protected]>

* avoid unnecessary copy in cross-executor convert_to(Dense) * improve naming in format conversion tests * make format conversion tests more resilient Co-authored-by: Yuhsiang Tsai <[email protected]>

ginkgo-bot · 2022-01-31T17:13:50Z

Note: This PR changes the Ginkgo ABI:

Functions changes summary: 196 Removed, 1424 Changed (12021 filtered out), 8292 Added functions
Variables changes summary: 0 Removed, 0 Changed, 0 Added variable

For details check the full ABI diff under Artifacts here

pratikvn

LGTM!

pratikvn · 2022-01-31T18:04:05Z

common/unified/base/kernel_launch_reduction.hpp

+#define GKO_KERNEL_REDUCE_SUM(ValueType)               \
+    [] GKO_KERNEL(auto a, auto b) { return a + b; },   \
+        [] GKO_KERNEL(auto a) { return a; }, ValueType \
+    {}
+#define GKO_KERNEL_REDUCE_MAX(ValueType)                     \
+    [] GKO_KERNEL(auto a, auto b) { return a > b ? a : b; }, \
+        [] GKO_KERNEL(auto a) { return a; }, ValueType       \
+    {}


pratikvn · 2022-01-31T18:09:02Z

core/matrix/fbcsr.cpp

-    auto tmp = SparsityCsr<ValueType, IndexType>::create(
-        exec,
+    result->set_size(
        gko::dim<2>{static_cast<size_type>(this->get_num_block_rows()),
-                    static_cast<size_type>(this->get_num_block_cols())},
-        this->get_num_stored_blocks());
-
-    tmp->col_idxs_ = this->col_idxs_;
-    tmp->row_ptrs_ = this->row_ptrs_;
-    tmp->value_ = Array<ValueType>(exec, {one<ValueType>()});
-    tmp->move_to(result);
+                    static_cast<size_type>(this->get_num_block_cols())});
+    result->col_idxs_ = this->col_idxs_;
+    result->row_ptrs_ = this->row_ptrs_;
+    result->value_ =
+        Array<ValueType>(result->get_executor(), {one<ValueType>()});


The problem I see is that one would expect the size of the matrix and its sparsity pattern to be the same, which will not be the case here. But anyway, I guess we will postpone this discussion to a later point.

sonarcloud · 2022-01-31T22:12:53Z

SonarCloud Quality Gate failed.

1 Bug
0 Vulnerabilities
0 Security Hotspots
137 Code Smells

54.1% Coverage
17.4% Duplication

codecov · 2022-01-31T22:15:24Z

Codecov Report

Merging #905 (a7379ad) into develop (f25b795) will decrease coverage by 0.59%.
The diff coverage is 62.33%.

@@             Coverage Diff             @@
##           develop     #905      +/-   ##
===========================================
- Coverage    93.53%   92.93%   -0.60%     
===========================================
  Files          474      476       +2     
  Lines        38999    39029      +30     
===========================================
- Hits         36477    36273     -204     
- Misses        2522     2756     +234

Impacted Files	Coverage Δ
.../unified/components/device_matrix_data_kernels.cpp	`100.00% <ø> (+70.58%)`	⬆️
common/unified/solver/cgs_kernels.cpp	`95.45% <ø> (ø)`
common/unified/solver/fcg_kernels.cpp	`93.33% <ø> (ø)`
core/device_hooks/common_kernels.inc.cpp	`0.00% <0.00%> (ø)`
include/ginkgo/core/matrix/coo.hpp	`91.48% <ø> (ø)`
include/ginkgo/core/matrix/dense.hpp	`95.58% <ø> (ø)`
include/ginkgo/core/matrix/ell.hpp	`94.59% <ø> (ø)`
include/ginkgo/core/matrix/sparsity_csr.hpp	`93.10% <ø> (ø)`
omp/components/device_matrix_data_kernels.cpp	`100.00% <ø> (ø)`
omp/matrix/coo_kernels.cpp	`88.88% <ø> (-2.92%)`	⬇️
... and 87 more

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 f25b795...a7379ad. Read the comment docs.

Advertise release 1.5.0 and last changes + Add changelog, + Update third party libraries + A small fix to a CMake file See PR: #1195 The Ginkgo team is proud to announce the new Ginkgo minor release 1.5.0. This release brings many important new features such as: - MPI-based multi-node support for all matrix formats and most solvers; - full DPC++/SYCL support, - functionality and interface for GPU-resident sparse direct solvers, - an interface for wrapping solvers with scaling and reordering applied, - a new algebraic Multigrid solver/preconditioner, - improved mixed-precision support, - support for device matrix assembly, and much more. If you face an issue, please first check our [known issues page](https://github.com/ginkgo-project/ginkgo/wiki/Known-Issues) and the [open issues list](https://github.com/ginkgo-project/ginkgo/issues) and if you do not find a solution, feel free to [open a new issue](https://github.com/ginkgo-project/ginkgo/issues/new/choose) or ask a question using the [github discussions](https://github.com/ginkgo-project/ginkgo/discussions). Supported systems and requirements: + For all platforms, CMake 3.13+ + C++14 compliant compiler + Linux and macOS + GCC: 5.5+ + clang: 3.9+ + Intel compiler: 2018+ + Apple LLVM: 8.0+ + NVHPC: 22.7+ + Cray Compiler: 14.0.1+ + CUDA module: CUDA 9.2+ or NVHPC 22.7+ + HIP module: ROCm 4.0+ + DPC++ module: Intel OneAPI 2021.3 with oneMKL and oneDPL. Set the CXX compiler to `dpcpp`. + Windows + MinGW and Cygwin: GCC 5.5+ + Microsoft Visual Studio: VS 2019 + CUDA module: CUDA 9.2+, Microsoft Visual Studio + OpenMP module: MinGW or Cygwin. Algorithm and important feature additions: + Add MPI-based multi-node for all matrix formats and solvers (except GMRES and IDR). ([#676](#676), [#908](#908), [#909](#909), [#932](#932), [#951](#951), [#961](#961), [#971](#971), [#976](#976), [#985](#985), [#1007](#1007), [#1030](#1030), [#1054](#1054), [#1100](#1100), [#1148](#1148)) + Porting the remaining algorithms (preconditioners like ISAI, Jacobi, Multigrid, ParILU(T) and ParIC(T)) to DPC++/SYCL, update to SYCL 2020, and improve support and performance ([#896](#896), [#924](#924), [#928](#928), [#929](#929), [#933](#933), [#943](#943), [#960](#960), [#1057](#1057), [#1110](#1110), [#1142](#1142)) + Add a Sparse Direct interface supporting GPU-resident numerical LU factorization, symbolic Cholesky factorization, improved triangular solvers, and more ([#957](#957), [#1058](#1058), [#1072](#1072), [#1082](#1082)) + Add a ScaleReordered interface that can wrap solvers and automatically apply reorderings and scalings ([#1059](#1059)) + Add a Multigrid solver and improve the aggregation based PGM coarsening scheme ([#542](#542), [#913](#913), [#980](#980), [#982](#982), [#986](#986)) + Add infrastructure for unified, lambda-based, backend agnostic, kernels and utilize it for some simple kernels ([#833](#833), [#910](#910), [#926](#926)) + Merge different CUDA, HIP, DPC++ and OpenMP tests under a common interface ([#904](#904), [#973](#973), [#1044](#1044), [#1117](#1117)) + Add a device_matrix_data type for device-side matrix assembly ([#886](#886), [#963](#963), [#965](#965)) + Add support for mixed real/complex BLAS operations ([#864](#864)) + Add a FFT LinOp for all but DPC++/SYCL ([#701](#701)) + Add FBCSR support for NVIDIA and AMD GPUs and CPUs with OpenMP ([#775](#775)) + Add CSR scaling ([#848](#848)) + Add array::const_view and equivalent to create constant matrices from non-const data ([#890](#890)) + Add a RowGatherer LinOp supporting mixed precision to gather dense matrix rows ([#901](#901)) + Add mixed precision SparsityCsr SpMV support ([#970](#970)) + Allow creating CSR submatrix including from (possibly discontinuous) index sets ([#885](#885), [#964](#964)) + Add a scaled identity addition (M <- aI + bM) feature interface and impls for Csr and Dense ([#942](#942)) Deprecations and important changes: + Deprecate AmgxPgm in favor of the new Pgm name. ([#1149](#1149)). + Deprecate specialized residual norm classes in favor of a common `ResidualNorm` class ([#1101](#1101)) + Deprecate CamelCase non-polymorphic types in favor of snake_case versions (like array, machine_topology, uninitialized_array, index_set) ([#1031](#1031), [#1052](#1052)) + Bug fix: restrict gko::share to rvalue references (*possible interface break*) ([#1020](#1020)) + Bug fix: when using cuSPARSE's triangular solvers, specifying the factory parameter `num_rhs` is now required when solving for more than one right-hand side, otherwise an exception is thrown ([#1184](#1184)). + Drop official support for old CUDA < 9.2 ([#887](#887)) Improved performance additions: + Reuse tmp storage in reductions in solvers and add a mutable workspace to all solvers ([#1013](#1013), [#1028](#1028)) + Add HIP unsafe atomic option for AMD ([#1091](#1091)) + Prefer vendor implementations for Dense dot, conj_dot and norm2 when available ([#967](#967)). + Tuned OpenMP SellP, COO, and ELL SpMV kernels for a small number of RHS ([#809](#809)) Fixes: + Fix various compilation warnings ([#1076](#1076), [#1183](#1183), [#1189](#1189)) + Fix issues with hwloc-related tests ([#1074](#1074)) + Fix include headers for GCC 12 ([#1071](#1071)) + Fix for simple-solver-logging example ([#1066](#1066)) + Fix for potential memory leak in Logger ([#1056](#1056)) + Fix logging of mixin classes ([#1037](#1037)) + Improve value semantics for LinOp types, like moved-from state in cross-executor copy/clones ([#753](#753)) + Fix some matrix SpMV and conversion corner cases ([#905](#905), [#978](#978)) + Fix uninitialized data ([#958](#958)) + Fix CUDA version requirement for cusparseSpSM ([#953](#953)) + Fix several issues within bash-script ([#1016](#1016)) + Fixes for `NVHPC` compiler support ([#1194](#1194)) Other additions: + Simplify and properly name GMRES kernels ([#861](#861)) + Improve pkg-config support for non-CMake libraries ([#923](#923), [#1109](#1109)) + Improve gdb pretty printer ([#987](#987), [#1114](#1114)) + Add a logger highlighting inefficient allocation and copy patterns ([#1035](#1035)) + Improved and optimized test random matrix generation ([#954](#954), [#1032](#1032)) + Better CSR strategy defaults ([#969](#969)) + Add `move_from` to `PolymorphicObject` ([#997](#997)) + Remove unnecessary device_guard usage ([#956](#956)) + Improvements to the generic accessor for mixed-precision ([#727](#727)) + Add a naive lower triangular solver implementation for CUDA ([#764](#764)) + Add support for int64 indices from CUDA 11 onward with SpMV and SpGEMM ([#897](#897)) + Add a L1 norm implementation ([#900](#900)) + Add reduce_add for arrays ([#831](#831)) + Add utility to simplify Dense View creation from an existing Dense vector ([#1136](#1136)). + Add a custom transpose implementation for Fbcsr and Csr transpose for unsupported vendor types ([#1123](#1123)) + Make IDR random initilization deterministic ([#1116](#1116)) + Move the algorithm choice for triangular solvers from Csr::strategy_type to a factory parameter ([#1088](#1088)) + Update CUDA archCoresPerSM ([#1175](#1116)) + Add kernels for Csr sparsity pattern lookup ([#994](#994)) + Differentiate between structural and numerical zeros in Ell/Sellp ([#1027](#1027)) + Add a binary IO format for matrix data ([#984](#984)) + Add a tuple zip_iterator implementation ([#966](#966)) + Simplify kernel stubs and declarations ([#888](#888)) + Simplify GKO_REGISTER_OPERATION with lambdas ([#859](#859)) + Simplify copy to device in tests and examples ([#863](#863)) + More verbose output to array assertions ([#858](#858)) + Allow parallel compilation for Jacobi kernels ([#871](#871)) + Change clang-format pointer alignment to left ([#872](#872)) + Various improvements and fixes to the benchmarking framework ([#750](#750), [#759](#759), [#870](#870), [#911](#911), [#1033](#1033), [#1137](#1137)) + Various documentation improvements ([#892](#892), [#921](#921), [#950](#950), [#977](#977), [#1021](#1021), [#1068](#1068), [#1069](#1069), [#1080](#1080), [#1081](#1081), [#1108](#1108), [#1153](#1153), [#1154](#1154)) + Various CI improvements ([#868](#868), [#874](#874), [#884](#884), [#889](#889), [#899](#899), [#903](#903), [#922](#922), [#925](#925), [#930](#930), [#936](#936), [#937](#937), [#958](#958), [#882](#882), [#1011](#1011), [#1015](#1015), [#989](#989), [#1039](#1039), [#1042](#1042), [#1067](#1067), [#1073](#1073), [#1075](#1075), [#1083](#1083), [#1084](#1084), [#1085](#1085), [#1139](#1139), [#1178](#1178), [#1187](#1187))

upsj added the 1:ST:ready-for-review This PR is ready for review label Oct 15, 2021

upsj added this to the Ginkgo 1.5.0 milestone Oct 15, 2021

upsj requested a review from a team October 15, 2021 15:26

upsj self-assigned this Oct 15, 2021

upsj force-pushed the format_conversion_improvements branch from 41991f1 to 83cdb35 Compare October 18, 2021 13:30

upsj mentioned this pull request Oct 18, 2021

Add device_matrix_data and device-side matrix::read #886

Merged

9 tasks

upsj added 1:ST:need-feedback The PR is somewhat ready but feedback on a blocking topic is required before a proper review. and removed 1:ST:ready-for-review This PR is ready for review labels Oct 20, 2021

upsj mentioned this pull request Oct 25, 2021

Add a class for partitions of intervals #909

Merged

upsj force-pushed the format_conversion_improvements branch 2 times, most recently from e3900d8 to 630d3ea Compare November 3, 2021 10:55

upsj mentioned this pull request Nov 4, 2021

CUDA >= 11.0 support for int64 indices for spmv and spgemm #897

Merged

pratikvn reviewed Nov 10, 2021

View reviewed changes

common/unified/matrix/dense_kernels.cpp Outdated Show resolved Hide resolved

pratikvn reviewed Nov 10, 2021

View reviewed changes

common/unified/matrix/hybrid_kernels.cpp Show resolved Hide resolved

upsj commented Nov 10, 2021

View reviewed changes

common/unified/matrix/coo_kernels.cpp Outdated Show resolved Hide resolved

upsj force-pushed the format_conversion_improvements branch from 630d3ea to a100f16 Compare November 22, 2021 12:26

upsj force-pushed the format_conversion_improvements branch 3 times, most recently from aeb1210 to 9a33a05 Compare December 9, 2021 17:01

pratikvn mentioned this pull request Dec 10, 2021

Improving Csr::strategy_type #320

Open

upsj removed the 1:ST:need-feedback The PR is somewhat ready but feedback on a blocking topic is required before a proper review. label Dec 17, 2021

upsj added 9 commits January 30, 2022 23:08

fix sellp SpMV on GPUs

57c56f9

fix const-correctness for Hybrid strategy

eae6e48

zero Fbcsr and Sellp pointer arrays

b2dcca0

work around cuSPARSE issues with empty matrices

72f0bbe

add Fbcsr conversion to Dense

77422ee

guard empty CUDA kernel launches

31ae3e6

improve sorting test output

17891e5

remove unnecessary dim3 usages

66b5ed2

add conversions to Fbcsr

f732c3c

upsj force-pushed the format_conversion_improvements branch from 600a554 to 493daef Compare January 30, 2022 22:10

upsj force-pushed the format_conversion_improvements branch from 200ac00 to 863316d Compare January 31, 2022 10:00

upsj and others added 6 commits January 31, 2022 14:28

simplify cuSPARSE type handling

b4cfc27

Co-authored-by: Pratik Nayak <[email protected]>

fix accessor compilation issues

84ebf9d

review updates

b982ec6

* avoid unnecessary copy in cross-executor convert_to(Dense) * improve naming in format conversion tests * make format conversion tests more resilient Co-authored-by: Yuhsiang Tsai <[email protected]>

add shortcuts for common reduction boilerplate

2d27767

fix CUDA version detection for triangular solve

f98170c

upsj force-pushed the format_conversion_improvements branch from 12d0c0a to f98170c Compare January 31, 2022 13:29

upsj added the 1:ST:run-full-test label Jan 31, 2022

fix dpcpp compilation

a7379ad

pratikvn approved these changes Jan 31, 2022

View reviewed changes

upsj 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 Feb 1, 2022

upsj merged commit 0436f3e into develop Feb 1, 2022

upsj deleted the format_conversion_improvements branch February 1, 2022 08:45

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Format conversion, SpMV and write improvements #905

Format conversion, SpMV and write improvements #905

upsj commented Oct 15, 2021 •

edited

Loading

tcojean commented Oct 21, 2021

Slaedr commented Oct 22, 2021

pratikvn left a comment

upsj commented Jan 30, 2022

ginkgo-bot commented Jan 31, 2022

pratikvn left a comment

pratikvn Jan 31, 2022

pratikvn Jan 31, 2022

sonarcloud bot commented Jan 31, 2022

codecov bot commented Jan 31, 2022

Format conversion, SpMV and write improvements #905

Format conversion, SpMV and write improvements #905

Conversation

upsj commented Oct 15, 2021 • edited Loading

tcojean commented Oct 21, 2021

Slaedr commented Oct 22, 2021

pratikvn left a comment

Choose a reason for hiding this comment

upsj commented Jan 30, 2022

ginkgo-bot commented Jan 31, 2022

pratikvn left a comment

Choose a reason for hiding this comment

pratikvn Jan 31, 2022

Choose a reason for hiding this comment

pratikvn Jan 31, 2022

Choose a reason for hiding this comment

sonarcloud bot commented Jan 31, 2022

codecov bot commented Jan 31, 2022

Codecov Report

upsj commented Oct 15, 2021 •

edited

Loading