This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch nightly
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 81f7da8a7f [Relax][PyTorch] Support prod, std and var ops for
ExportedProgram importer (#17785)
add 3f16ec25ac [Relax][PyTorch] Add support for where, cumprod and
reciprocal ops (#17788)
add aaf185bd4b [Relax][PyTorch] Support softshrink op for ExportedProgram
(#17786)
add b0ccfb39c5 [CUTLASS] Add blockwise scale gemm/bmm kernels (#17789)
add 2c80e5c7a4 [Relax][PyTorch] Add support for index_select (#17790)
No new revisions were added by this update.
Summary of changes:
3rdparty/cutlass | 2 +-
3rdparty/cutlass_fpA_intB_gemm | 2 +-
cmake/modules/contrib/CUTLASS.cmake | 6 +-
.../frontend/torch/base_fx_graph_translator.py | 56 ++++
.../frontend/torch/exported_program_translator.py | 4 +-
python/tvm/relax/frontend/torch/fx_translator.py | 13 +-
.../cutlass/blockwise_scaled_gemm_runner.cuh | 228 +++++++++++++
.../contrib/cutlass/fp8_blockwise_scaled_gemm.cu | 164 ++++++++++
src/runtime/contrib/cutlass/group_gemm_runner.cuh | 14 +-
tests/python/contrib/test_cutlass_gemm.py | 352 +++++++++++++++++++++
tests/python/relax/test_from_exported_to_cuda.py | 12 +
.../relax/test_frontend_from_exported_program.py | 51 +++
tests/python/relax/test_frontend_from_fx.py | 65 ++++
13 files changed, 952 insertions(+), 17 deletions(-)
create mode 100644 src/runtime/contrib/cutlass/blockwise_scaled_gemm_runner.cuh
create mode 100644 src/runtime/contrib/cutlass/fp8_blockwise_scaled_gemm.cu
create mode 100644 tests/python/contrib/test_cutlass_gemm.py
