Skip to content

FlagBLAS v0.3.0

Latest
Latest

Choose a tag to compare

View all tags
@wbavon wbavon released this 28 Sep 09:23
v0.3.0
f0a8666
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

FlagBLAS v0.3.0

Part of FlagOS 2.2.

Changes since v0.2.0 (FlagOS 2.1).

76 pull requests and 29 additional commits.

Source: v0.3.0, commit f0a86661aa44.

New Features

  • Add run_test.py for benchmark and accuracy check and fix ci test (#5) by @bin913
  • add new krenel shape size dispatch with db (#7) by @bin913
  • Add run_op.sh in backend_test.yaml for cicd (#8) by @bin913
  • Add static dispatch (#9) by @bin913
  • add statatic dispatch for hgemm and fix copyright for code_utils.py (#13) by @bin913
  • add dgemm,cgemm,zgemm (#14) by @bin913
  • Add Group gemm for nvidia backend (#18) by @bin913
  • add weekly_full_test (#19) by @bin913
  • add pytest-md-report in pyproject (#20) by @bin913
  • Add Ascend 910C backend CI support (#25) by @bin913
  • Add Hygon DCU backend CI support (#26) by @bin913
  • Add new l2 and accuracy test with multi_backend support (#29) by @bin913
  • Add level2 test files (#34) by @bin913
  • Add triangular matrix benchmark files (#45) by @bin913
  • Add Givens rotation benchmark files (#46) by @bin913
  • Add Hygon/hipBLAS support to banded Level-2 benchmarks (#47) by @bin913
  • Add Hygon/hipBLAS support to symmetric/Hermitian Level-2 benchmarks (#48) by @bin913
  • Add Hygon/hipBLAS support to triangular Level-2 benchmarks (#49) by @bin913
  • Add Level-2 rank-1 update benchmark files (#50) by @bin913
  • Add Level-2 rank-2 update benchmark files (#51) by @bin913
  • Add Hermitian packed matrix-vector benchmark files (#52) by @bin913
  • Add ger (general rank-1 update) benchmark file (#56) by @bin913
  • Add trmm (triangular matrix-matrix) benchmark file (#57) by @bin913
  • Add SbmvBenchmark core shapes to core_shapes.yaml (#64) by @bin913
  • feat(ascend): update CHEMV tile layout (#101) by @AriesMTR
  • feat(ascend): add TBMV support (#102) by @AriesMTR
  • feat(ascend): add TBSV support (#104) by @AriesMTR
  • feat(ascend): add TPSV support (#105) by @AriesMTR
  • feat(ascend): update GEMV support (#107) by @AriesMTR
  • Add iluvatar gemm (#113) by @bin913
21 more changes

Performance

  • perf(ascend): optimize TPSV, TBSV, SYR, SYR2, HER, HER2, HPMV and GER (#124) by @uesrshihe
  • perf(thead): extend L2 operator performance support (#133) by @AriesMTR
  • Optimize fp64 scalar packing and GEMV small-N paths (3f4a800aee58; commit)

Documentation

  • chore: add Apache 2.0 copyright header to all source files (#12) by @tengqm

CI/Infrastructure

  • [packaging] Add Debian + RPM packaging (#1) by @shiptux
  • ci: build packages on packaging changes, not on every source change (#130) by @shiptux

Testing

  • tests: remove platform-only L2 skips (#132) by @AriesMTR
  • test(her2): document Ascend input construction (2bd13d2f6e94; commit)

Bug Fixes

  • Fix run test (#10) by @bin913
  • fix code style (#15) by @bin913
  • Fix benchmark logger handler check and add Level-2 benchmark shapes (#40) by @bin913
  • fix ger per-op markers for tools/run_test.py (#66) by @bin913
  • fix fp8_gemv markers for tools/run_test.py (#67) by @bin913
  • fix l2 benchmark op_name to match pytest markers (#68) by @bin913
  • fix(tools): continue pytest run despite collection errors (#81) by @bin913
  • fix pyproject.toml (#89) by @bin913
  • Fix nvidia ci fail (#94) by @bin913
  • fix(ci): install nvidia flagtree with uv pip so triton is available (#96) by @bin913
  • fix(ascend): remove redundant operator export lists (#103) by @AriesMTR
  • fix(ascend): prepare GBMV reference band storage on CPU (#109) by @AriesMTR
  • fix pyproject.toml for thead (#121) by @bin913
  • fix: remove unuse level1 init (3d72e3b9e165; commit)
  • fix code-format for gemm ops (06117ee099a1; commit)
  • fix operators.yaml for hgemm, bfgemm, fp8gemm (af942e013b44; commit)

Other

  • Sgemm hgemm bfgemm (#17) by @bin913
  • dubug weekly-full-test.yaml (#21) by @bin913
  • Remove redundant validate_results and refactor swap benchmark (#41) by @bin913
  • Restore Hygon/hipBLAS benchmark support in test_gemv_perf.py (#58) by @bin913
  • Restore OOM skip handling in test_amin_perf.py (#59) by @bin913
  • Register missing BLAS ops in operators.yaml and remove orphan fp8gemm entry (#62) by @bin913
  • Promote all L2/L3 operators to stable stage (#65) by @bin913
  • Update Level 2 operator support (#70) by @AriesMTR
  • Update Level 2 benchmarks (#71) by @AriesMTR
  • Update SYMV and HEMV support (#72) by @AriesMTR
  • Update SPMV and HPMV support (#73) by @AriesMTR
  • Update GER, SYR, and HER support (#74) by @AriesMTR
  • Update SPR and HPR support (#75) by @AriesMTR
  • Update SYR2 and HER2 support (#76) by @AriesMTR
  • Update GBMV, HBMV, and SBMV support (#79) by @AriesMTR
  • Update TRMV, TBMV, and TPMV support (#82) by @AriesMTR
  • Update TPSV support (#83) by @AriesMTR
  • Update TRSV support (#85) by @AriesMTR
  • Update TBSV support (#87) by @AriesMTR
  • Vectorize CGBMV band loop with BAND_TILE tiling (9bb8761896d9; commit)
  • opt sgemm_nn_kernel (138f674a96f0; commit)
  • opt sgemm kernel for thin shape M=64 or N=64 (34608bbacac3; commit)
  • opt tune_configs.yaml for sgemm_nn model_shape (480401d6d941; commit)
  • Normalize L1 test and benchmark shapes (d580c0d82c94; commit)
  • Merge remote-tracking branch 'origin/master' into feat/ascend-l2-triangular (32b91926f629; commit)
  • Merge pull request #93 from flagos-ai/feat/ascend-l2-triangular (2bba707ab3de; commit)
  • Merge pull request #97 from flagos-ai/feat/ascend-her (36e22bac201b; commit)
  • Merge pull request #98 from flagos-ai/feat/ascend-her2 (cd89939d40eb; commit)
  • Merge pull request #99 from flagos-ai/feat/ascend-syr (55f5e780ce9a; commit)
  • Merge pull request #100 from flagos-ai/feat/ascend-syr2 (b075f6808822; commit)

Contributors

Thanks to the contributors to the pull requests included in this release.

Contributors

tengqm, bin913, and 3 other contributors
Assets 2
Loading