This pull request includes a number of changes across several files in the codebase. The main themes of these changes are updates to versioning, modifications to methods and classes, and the addition of new methods and classes.
Updates to versioning:
VERSION: The version number was incremented from 0.0.1.dev7 to 0.0.1.dev8.
python/bitblas/__init__.py: The version number was also updated here to match the change in the VERSION file.
integration/BitNet/utils_quant.py: In the forward_fp32_simulated method, a print statement was removed. In the forward method, a commented-out return statement was added. [1][2]
python/bitblas/gpu/intrin/lop3.py: Added new intrin registrations: LOP3_FAST_DECODE_INT1_TO_INT8_TO_FP16_L8_INTRIN and LOP3_FAST_DECODE_INT1_TO_INT8_TO_FP16_L8_SCALE_INTRIN.
This pull request includes a number of changes across several files in the codebase. The main themes of these changes are updates to versioning, modifications to methods and classes, and the addition of new methods and classes.
Updates to versioning:
VERSION
: The version number was incremented from 0.0.1.dev7 to 0.0.1.dev8.python/bitblas/__init__.py
: The version number was also updated here to match the change in theVERSION
file.3rdparty/tvm
: The subproject commit was updated.Modifications to methods and classes:
integration/BitNet/utils_quant.py
: In theforward_fp32_simulated
method, a print statement was removed. In theforward
method, a commented-out return statement was added. [1] [2]python/bitblas/base/roller/hint.py
: In theTensorCoreExtraConfig
class, a new methodtensorcore_legalization
was added.python/bitblas/base/roller/policy/tensorcore.py
: In the_score
method, conditions were added to check if the output data type is float32 and to set the shared scope to "shared.dyn" if it is. [1] [2]Addition of new methods and classes:
python/bitblas/base/roller/__init__.py
: Imported new classes:NoRasterization
,Rasterization2DRow
,Rasterization2DColumn
.python/bitblas/gpu/intrin/lop3.py
: Added new intrin registrations:LOP3_FAST_DECODE_INT1_TO_INT8_TO_FP16_L8_INTRIN
andLOP3_FAST_DECODE_INT1_TO_INT8_TO_FP16_L8_SCALE_INTRIN
.python/bitblas/gpu/matmul_analysis.py
: Added a new methodcollect_vars_from_expr
to collect variables from a given expression.