This pull request introduces a number of changes across the python/bitblas package in order to improve the functionality of the BitBlas library. The changes include updates to the Rasterization and TensorCoreExtraConfig classes, modifications to the fast_decode_impl method, and the addition of the MatmulWithSplitK class.
Updates to Rasterization and TensorCoreExtraConfig classes:
python/bitblas/gpu/intrin/lop3.py: Reformatted the arguments in the get_fast_decode_intrin calls within the fast_decode_impl method for better readability. [1][2]
Addition of MatmulWithSplitK class:
python/bitblas/ops/general_matmul_splitk.py: Added a new file implementing the MatmulWithSplitK class, which extends the functionality of the Matmul class with the ability to split the K dimension.
python/bitblas/ops/general_matmul.py: Removed the OPExecutorCPU class and added a condition to check if fast decoding is supported in the __initialize_fast_decoding method. [1][2]
This pull request introduces a number of changes across the
python/bitblaspackage in order to improve the functionality of the BitBlas library. The changes include updates to theRasterizationandTensorCoreExtraConfigclasses, modifications to thefast_decode_implmethod, and the addition of theMatmulWithSplitKclass.Updates to
RasterizationandTensorCoreExtraConfigclasses:python/bitblas/base/roller/__init__.py: Imported newRasterizationclasses.python/bitblas/base/roller/hint.py: Added a new methodtensorcore_legalizationto theTensorCoreExtraConfigclass.Modifications to
fast_decode_implmethod:python/bitblas/gpu/intrin/lop3.py: Reformatted the arguments in theget_fast_decode_intrincalls within thefast_decode_implmethod for better readability. [1] [2]Addition of
MatmulWithSplitKclass:python/bitblas/ops/general_matmul_splitk.py: Added a new file implementing theMatmulWithSplitKclass, which extends the functionality of theMatmulclass with the ability to split the K dimension.Other important changes:
3rdparty/tvm: Updated the subproject commit.python/bitblas/base/roller/policy/tensorcore.py: Added a call totensorcore_legalizationin the_scoremethod.python/bitblas/module/__init__.py: Changed the default value offast_decodingfromTruetoNonein the__init__method.python/bitblas/ops/general_matmul.py: Removed theOPExecutorCPUclass and added a condition to check if fast decoding is supported in the__initialize_fast_decodingmethod. [1] [2]