ROCm / hipBLASLt

hipBLASLt is a library that provides general matrix-matrix operations with a flexible API and extends functionalities beyond a traditional BLAS library
https://rocm.docs.amd.com/projects/hipBLASLt/en/latest/index.html
MIT License
63 stars 88 forks source link

Tune Aquavanjaram942 Grid sizes for BBS NN #1097

Closed pkamd closed 2 months ago

pkamd commented 2 months ago

Updated 8 grid points to address perf tuning request. Ensure changes doesn't degrade existing grid point performance.

jichangjichang commented 2 months ago

If this PR is necessary for 6.3, please remember file PR to release-staging/rocm-rel-6.3 as well.

nakajee commented 2 months ago

Please run hipblaslt-test on your local node (with the change) and paste the result here if you put noCI label.

pkamd commented 2 months ago

All hipBLASLt tests passed -

[----------] 1 test from ExtOpTest/ExtOpAMaxWithScaleUnsupportedDatatypeTest
[ RUN      ] ExtOpTest/ExtOpAMaxWithScaleUnsupportedDatatypeTest.amaxWithScaleFailureUnsupportedDatatype/0
[       OK ] ExtOpTest/ExtOpAMaxWithScaleUnsupportedDatatypeTest.amaxWithScaleFailureUnsupportedDatatype/0 (0 ms)
[----------] 1 test from ExtOpTest/ExtOpAMaxWithScaleUnsupportedDatatypeTest (0 ms total)

[----------] Global test environment tear-down
[==========] 48206 tests from 13 test suites ran. (2234494 ms total)
[  PASSED  ] 48206 tests.
hipBLASLt version: 1000

command line: ./hipblaslt-test
nakajee commented 2 months ago

Don't we need to cover HHS?

pkamd commented 2 months ago

Don't we need to cover HHS?

Yes the request is for both HHS and BBS. Planning to address it as a separate PR.