BSR block SpGEMM implementation #1099

mzuzek · 2021-09-08T18:25:29Z

This PR introduces implementation for block version of sparse BSR matrix-matrix multiplication (block spgemm).

Minimal changes to baseline CSR spgemm implementation are intended for clarity and correctness verifiability, so that this PR can provide correct implementation checkpoint. For the follow up with related refactoring and performance tuning see PR #1215.

📌 Files changed with respect to original sparse GEMM implementation

Status

Algorithm coverage:

Implementation

mzuzek · 2021-09-08T18:25:51Z

Default algorithm implementation details (FY21 block spgemm status)

Symbolic version is reused from non-block spgemm without modification.
Numeric phase was implemented as KokkosBSPGEMM in sparse/impl/KokkosSparse_bspgemm_impl_kkmem.hpp with MultiCoreTag (SPGEMM_KK) and MultiCoreTag4 (SPGEMM_KK_LP) functors.
Accumulator is work-in-progress: MultiCoreTag operator uses HashingIndex which is essentially HashmapAccumulator reduced to indexing part used by spgemm (HashmapAccumulator works on scalars) while MultiCoreTag4 operator inlines the indexing like it did in non-block version. I'll try to refactor/unify it once I dig into GPU operators and understand all indexing/accumulation scenarios. This plan includes detailed review of related memory estimations and allocations.
I used KokkosBatched::SerialGemmInternal<KokkosBatched::Algo::Gemm::Unblocked> from sparse/impl/KokkosBatched_Gemm_Serial_Internal.hpp for block accumulation/combination (getting C = A * B and C += A * B with beta 0 and 1 respectively).
Adjusted workload estimations (and algorithm selection) for block CPU operators, which use: suggested_vector_size, suggested_team_size and team_row_chunk_size.
Unit tests are based on new overload of kk_generate_sparse_matrix() in common/KokkosKernels_IOUtils.hpp that takes block size and generates random BSR matrix (also fixed hangs on large relative bandwith to be able to generate denser test matrices).
There are some extra utilities for unit testing in unit_test/sparse/Test_Sparse_bspgemm.hpp, including << operator that renders BSR matrix on stdout.

mzuzek · 2021-09-08T18:26:00Z

Preliminary CPU performance

I connected MKL to benchmark block spgemm performance on CPU. For multiplication of random matrix by itself on i9-9900K (8 cores/16 threads)/32GB RAM/Ubuntu/gcc-8.4, I got some rough timings (no repetition, accuracy around 0,5~1s):

Note: Google Test filter = *block_spgemm*
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from threads
[ RUN      ] threads.sparse_block_spgemm_double_int_int_TestExecSpace

Testing BSPGEMM: nrows=300000   block:2x2       nnz=8000000(0.009%)     bandwidth=100000(33.3%)        variance=10 (0.0%)      schedule=dynamic
 nnz_in=7800201
nnz_out=202326755
SPGEMM_DEBUG (reference result): 17.0
SPGEMM_MKL              spgemm_time:    5.5     output_check_time:      0.6
SPGEMM_KK               spgemm_time:    1.7     output_check_time:      0.5
SPGEMM_KK_LP            spgemm_time:    1.9     output_check_time:      0.5

Testing BSPGEMM: nrows=300000   block:3x3       nnz=6000000(0.007%)     bandwidth=100000(33.3%)        variance=10 (0.0%)      schedule=dynamic
 nnz_in=6000201
nnz_out=119850101
SPGEMM_DEBUG (reference result): 23.0
SPGEMM_MKL              spgemm_time:    10.7    output_check_time:      6.7
SPGEMM_KK               spgemm_time:    4.8     output_check_time:      0.7
SPGEMM_KK_LP            spgemm_time:    2.2     output_check_time:      0.7
[       OK ] threads.sparse_block_spgemm_double_int_int_TestExecSpace (80583 ms)
[----------] 1 test from threads (80587 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (80589 ms total)
[  PASSED  ] 1 test.

Notes:

nnz above means number of non-zero blocks.
Tested on Threads Kokkos execution space (because of silly issue with OpenMP running on single core...)
In non-block version, mkl_sparse_spmm() is called twice: in symbolic and numeric phases. I drafted a new interface that avoids this duplicated allocations and calculation to get more accurate insight into performance baseline. This optimization would need some discussion to be included in the actual implementation (i.e. outside `the unit test), since it doesn't immediately fit the interface forcing separation of symbolic and numeric phases.

kokkos-devops-admin · 2021-09-08T18:38:54Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
WARNING: NO REVIEWERS HAVE BEEN REQUESTED FOR THIS PULL REQUEST!

kokkos-devops-admin · 2021-09-08T19:01:03Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
NO REVIEWS HAVE BEEN PERFORMED ON THIS PULL REQUEST!

kokkos-devops-admin · 2021-10-11T10:04:15Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
NO REVIEWS HAVE BEEN PERFORMED ON THIS PULL REQUEST!

kokkos-devops-admin · 2021-11-24T13:58:31Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
NO INSPECTION HAS BEEN PERFORMED ON THIS PULL REQUEST! - This PR must be inspected by setting label 'AT: PRE-TEST INSPECTED'.

kokkos-devops-admin · 2021-12-06T16:49:38Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
NO INSPECTION HAS BEEN PERFORMED ON THIS PULL REQUEST! - This PR must be inspected by setting label 'AT: PRE-TEST INSPECTED'.

mzuzek · 2021-12-06T18:37:02Z

@lucbv @srajama1 @brian-kelley @ppebay @fnrizzi @uhetmaniuk

I've redone this PR from scratch, restructuring and cleaning its content to get BSR spgemm with minimal changes introduced to baseline CSR implementation. Those BSR changes over CRS implementation can be inspected here (also pinned this link in the PR description for convenience). To accommodate for this approach, I took some irregular implementation decisions like disabling large parts of unused code with #if 0 or mirroring whole HashmapAccumulator with BlockHashmapAccumulator in a separate header - all this to be cleaned up on the follow up PR #1215.

My motivation behind proposing this minimal changes approach and the follow-up PR for refactoring and performance tuning was to secure a clear correct implementation checkpoint - with the following considerations taken into account:

spgemm code is very easy to break due too its combined complexity and extent (bugs can be easily introduced even with smallest changes) ;
implementation correctness can be very hard to verify in review (bugs can be extremely hard to spot on source code diffs);
significant part of execution paths is not covered [yet] by the unit test (good chance for bugs not to be detected automatically either).

kokkos-devops-admin · 2021-12-29T12:52:27Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
NO INSPECTION HAS BEEN PERFORMED ON THIS PULL REQUEST! - This PR must be inspected by setting label 'AT: PRE-TEST INSPECTED'.

kokkos-devops-admin · 2022-04-22T14:19:27Z

Status Flag 'Pre-Test Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
NO INSPECTION HAS BEEN PERFORMED ON THIS PULL REQUEST! - This PR must be inspected by setting label 'AT: PRE-TEST INSPECTED'.

mzuzek · 2022-04-22T16:24:01Z

Hi @lucbv! Would you allow a Jenkins re-test ?

I just rebased this PR on current develop in hope to have [unrelated] CI failures solved by kokkos/kokkos#4959 now.

kokkos-devops-admin · 2022-04-26T19:23:43Z

Status Flag 'Pull Request AutoTester' - User Requested Retest - Label AT: RETEST will be reset after testing.

kokkos-devops-admin · 2022-04-26T19:23:49Z

Status Flag 'Pre-Test Inspection' - SUCCESS: The last commit to this Pull Request has been INSPECTED by label AT: PRE-TEST INSPECTED! Autotester is Removing Label; This inspection will remain valid until a new commit to source branch is performed.

kokkos-devops-admin · 2022-04-26T19:26:06Z

Status Flag 'Pull Request AutoTester' - Testing Jenkins Projects:

Pull Request Auto Testing STARTING (click to expand)

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA9_GCC720_Light_Tpls_GCC720_GCC740

Build Num: 180
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA10_Tpls_CUDA10_LayoutRight

Build Num: 173
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_GCC720

Build Num: 983
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_GCC720_Light_LayoutRight

Build Num: 627
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_GCC720

Build Num: 971
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_INTEL18

Build Num: 958
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_CLANG1001

Build Num: 363
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Using Repos:

Repo: KOKKOSKERNELS (fnrizzi/kokkos-kernels)

Pull Request Author: MikolajZuzek

kokkos-devops-admin · 2022-04-28T21:52:55Z

Status Flag 'Pull Request AutoTester' - User Requested Retest - Label AT: RETEST will be reset after testing.

kokkos-devops-admin · 2022-04-28T21:57:04Z

Status Flag 'Pull Request AutoTester' - Testing Jenkins Projects:

Pull Request Auto Testing STARTING (click to expand)

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA9_GCC720_Light_Tpls_GCC720_GCC740

Build Num: 181
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA10_Tpls_CUDA10_LayoutRight

Build Num: 174
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_GCC720

Build Num: 984
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_GCC720_Light_LayoutRight

Build Num: 628
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_GCC720

Build Num: 972
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_INTEL18

Build Num: 959
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_CLANG1001

Build Num: 364
Status: STARTED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Using Repos:

Repo: KOKKOSKERNELS (fnrizzi/kokkos-kernels)

Pull Request Author: MikolajZuzek

kokkos-devops-admin · 2022-04-28T23:29:06Z

Status Flag 'Pull Request AutoTester' - Jenkins Testing: all Jobs PASSED

Pull Request Auto Testing has PASSED (click to expand)

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA9_GCC720_Light_Tpls_GCC720_GCC740

Build Num: 181
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA10_Tpls_CUDA10_LayoutRight

Build Num: 174
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_GCC720

Build Num: 984
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_GCC720_Light_LayoutRight

Build Num: 628
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_GCC720

Build Num: 972
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_INTEL18

Build Num: 959
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

Build Information

Test Name: KokkosKernels_PullRequest_CLANG1001

Build Num: 364
Status: PASSED

Jenkins Parameters

Parameter Name	Value
KOKKOSKERNELS_SOURCE_BRANCH	29-bspgemm-implementation
KOKKOSKERNELS_SOURCE_REPO	https://github.com/fnrizzi/kokkos-kernels
KOKKOSKERNELS_SOURCE_SHA	`b712a56`
KOKKOSKERNELS_TARGET_BRANCH	develop
KOKKOSKERNELS_TARGET_REPO	https://github.com/kokkos/kokkos-kernels
KOKKOSKERNELS_TARGET_SHA	`4057d74`
PR_LABELS	AT: RETEST
PULLREQUESTNUM	1099
TEST_REPO_ALIAS	KOKKOSKERNELS

kokkos-devops-admin · 2022-04-28T23:29:23Z

Status Flag 'Pre-Merge Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging
THE LAST COMMIT TO THIS PULL REQUEST HAS NOT BEEN REVIEWED YET!

kokkos-devops-admin · 2022-04-28T23:29:30Z

All Jobs Finished; status = PASSED, However Inspection must be performed before merge can occur...

lucbv

I am okay with the changes made

kokkos-devops-admin · 2022-04-29T15:12:46Z

Status Flag 'Pre-Merge Inspection' - SUCCESS: The last commit to this Pull Request has been INSPECTED AND APPROVED by [ lucbv ]!

kokkos-devops-admin · 2022-04-29T15:12:53Z

Status Flag 'Pull Request AutoTester' - Pull Request MUST BE MERGED MANUALLY BY Project Team - This Repo does not support Automerge

mzuzek · 2022-04-30T10:37:59Z

@lucbv @brian-kelley Thank you very much for your thorough reviews and great comments!

mzuzek mentioned this pull request Sep 8, 2021

#29 Block sparse GEMM implementation fnrizzi/kokkos-kernels#32

Closed

lucbv requested review from srajama1, brian-kelley, lucbv and uhetmaniuk and removed request for srajama1 September 8, 2021 18:58

mzuzek force-pushed the 29-bspgemm-implementation branch from eae9f61 to 85b19e3 Compare October 11, 2021 09:59

mzuzek force-pushed the 29-bspgemm-implementation branch 3 times, most recently from 13d919b to 159d08c Compare November 24, 2021 13:57

mzuzek force-pushed the 29-bspgemm-implementation branch 2 times, most recently from 993e92c to 5bfef1c Compare November 24, 2021 15:08

mzuzek changed the title ~~Block sparse matrix-matrix implementation~~ WIP Block SpGEMM implementation Nov 25, 2021

mzuzek changed the title ~~WIP Block SpGEMM implementation~~ WIP: Block SpGEMM implementation Nov 25, 2021

mzuzek force-pushed the 29-bspgemm-implementation branch from 5bfef1c to 221fc89 Compare December 6, 2021 16:44

mzuzek mentioned this pull request Dec 6, 2021

[WIP] Block SpGEMM performance tuning #1215

Closed

9 tasks

mzuzek marked this pull request as ready for review December 6, 2021 18:37

mzuzek changed the title ~~WIP: Block SpGEMM implementation~~ BSR block SpGEMM implementation Dec 6, 2021

mzuzek mentioned this pull request Dec 21, 2021

Prune extra dependencies #1241

Merged

mzuzek force-pushed the 29-bspgemm-implementation branch from 221fc89 to 2549b48 Compare December 29, 2021 12:46

Mikołaj Zuzek added 4 commits April 22, 2022 16:10

clean up unused code

682a175

Clean SpGEMM code not used in block version

e380d1b

Disable unavailable TPL implementations

dcc1600

add some explanation to CPU functor in speed method

b712a56

mzuzek force-pushed the 29-bspgemm-implementation branch from 090f472 to b712a56 Compare April 22, 2022 14:15

brian-kelley added AT: RETEST Have this PR retested. AT: PRE-TEST INSPECTED Mark this PR as approved for testing. labels Apr 26, 2022

kokkos-devops-admin removed the AT: PRE-TEST INSPECTED Mark this PR as approved for testing. label Apr 26, 2022

kokkos-devops-admin removed the AT: RETEST Have this PR retested. label Apr 28, 2022

lucbv added the AT: PRE-TEST INSPECTED Mark this PR as approved for testing. label Apr 29, 2022

lucbv approved these changes Apr 29, 2022

View reviewed changes

lucbv merged commit db72c4b into kokkos:develop Apr 29, 2022

mzuzek deleted the 29-bspgemm-implementation branch April 30, 2022 10:38

ndellingwood mentioned this pull request May 2, 2022

BlockSPGEMM build error #1395

Closed

ndellingwood mentioned this pull request May 18, 2022

Nightly test failures in Cuda with rdc+uvm builds: cuda.sparse_block_spgemm tests #1413

Closed

This was referenced May 26, 2022

Add MKL implementation for BSR SpGEMM #1423

Open

Add BSR block support to SpGEMM perf_test #1424

Open

BSR block SpGEMM implementation #1099

BSR block SpGEMM implementation #1099

Conversation

mzuzek commented Sep 8, 2021 • edited Loading

Status

Algorithm coverage:

Implementation

mzuzek commented Sep 8, 2021 • edited Loading

Default algorithm implementation details (FY21 block spgemm status)

mzuzek commented Sep 8, 2021

Preliminary CPU performance

kokkos-devops-admin commented Sep 8, 2021

kokkos-devops-admin commented Sep 8, 2021

kokkos-devops-admin commented Oct 11, 2021

kokkos-devops-admin commented Nov 24, 2021

kokkos-devops-admin commented Dec 6, 2021

mzuzek commented Dec 6, 2021

kokkos-devops-admin commented Dec 29, 2021

kokkos-devops-admin commented Apr 22, 2022

mzuzek commented Apr 22, 2022

kokkos-devops-admin commented Apr 26, 2022

kokkos-devops-admin commented Apr 26, 2022

kokkos-devops-admin commented Apr 26, 2022

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA9_GCC720_Light_Tpls_GCC720_GCC740

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA10_Tpls_CUDA10_LayoutRight

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_GCC720

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_GCC720_Light_LayoutRight

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_GCC720

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_INTEL18

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_CLANG1001

Jenkins Parameters

Using Repos:

kokkos-devops-admin commented Apr 28, 2022

kokkos-devops-admin commented Apr 28, 2022

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA9_GCC720_Light_Tpls_GCC720_GCC740

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA10_Tpls_CUDA10_LayoutRight

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_GCC720

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_GCC720_Light_LayoutRight

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_GCC720

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_INTEL18

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_CLANG1001

Jenkins Parameters

Using Repos:

kokkos-devops-admin commented Apr 28, 2022

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA9_GCC720_Light_Tpls_GCC720_GCC740

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_Tpls_CUDA10_Tpls_CUDA10_LayoutRight

Jenkins Parameters

Build Information

Test Name: KokkosKernels_PullRequest_GCC720

Jenkins Parameters

Build Information

mzuzek commented Sep 8, 2021 •

edited

Loading

mzuzek commented Sep 8, 2021 •

edited

Loading