MCPcopy Create free account
hub / github.com/NVIDIA/cutlass / _valid_tile_description

Method _valid_tile_description

python/cutlass_cppgen/op/conv.py:372–397  ·  view source on GitHub ↗

Checks whether the provided tile description is valid for the given compute capability. At present, this checks the following: - Does the tile description use a number of stages supported by the compute capability in question? - Does the tile size requested fit with

(self, td: TileDescription)

Source from the content-addressed store, hash-verified

370 raise Exception(msg)
371
372 def _valid_tile_description(self, td: TileDescription) -> tuple:
373 """
374 Checks whether the provided tile description is valid for the given compute capability. At present,
375 this checks the following:
376
377 - Does the tile description use a number of stages supported by the compute capability in question?
378 - Does the tile size requested fit within shared memory?
379 - Are cluster dimensions outside the valid range requested for a given architecture (e.g.,
380 more non-unit cluster dimensions for pre-SM90 architectures)?
381 - Is the kernel schedule being used supported on the architecture in question?
382
383 :param td: tile description to validate
384 :type td: cutlass_cppgen.backend.TileDescription
385 :return: tuple in which the first element is a bool indicating that the tile description is valid
386 and the second element is a string providing an optional error message.
387 :rtype: tuple
388 """
389 valid, msg = check.valid_stage_count(self.cc, self.current_cc, td)
390 if not valid:
391 return (valid, msg)
392
393 valid, msg = check.valid_cluster_shape(self.current_cc, td.cluster_shape)
394 if not valid:
395 return (valid, msg)
396
397 return valid, msg
398
399 def tile_descriptions(self) -> list:
400 """

Callers 2

tile_descriptionMethod · 0.95
constructMethod · 0.95

Calls

no outgoing calls

Tested by

no test coverage detected