提交 533f8738 authored 作者: Frederic Bastien's avatar Frederic Bastien

fix white space.

上级 f7064e29
...@@ -676,7 +676,7 @@ def local_fill_to_alloc(node): ...@@ -676,7 +676,7 @@ def local_fill_to_alloc(node):
else: else:
# we are broadcasting both v and r, # we are broadcasting both v and r,
# the output shape must be computed # the output shape must be computed
# #
# TODO: implement this case (including a test!) # TODO: implement this case (including a test!)
# #
# I think the strategy should be to extend the shorter shape vector # I think the strategy should be to extend the shorter shape vector
...@@ -2972,10 +2972,10 @@ def local_elemwise_fusion_op(OP, max_input_fct=lambda node: 1024): ...@@ -2972,10 +2972,10 @@ def local_elemwise_fusion_op(OP, max_input_fct=lambda node: 1024):
GPU kernel currently have a limit of 256 bytes for GPU kernel currently have a limit of 256 bytes for
the size of all parameter passed to it. As currently the size of all parameter passed to it. As currently
we pass many information only by parameter, we must we pass many information only by parameter, we must
limit how many op we fuse together to don't bust that limit how many op we fuse together to don't bust that
256 limit. 256 limit.
On the CPU we limit to 1024 input variable to the On the CPU we limit to 1024 input variable to the
resulting fused op. This is big enought that resulting fused op. This is big enought that
if we hit it, I'm not sure it will affect performance. if we hit it, I'm not sure it will affect performance.
""" """
......
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论