提交 0f8d0341 authored 作者: Arnaud Bergeron's avatar Arnaud Bergeron

Make sure we don't try to optimize nodes that have a subclass of ours when doing fusion.

上级 e7feb8e3
......@@ -4609,7 +4609,7 @@ def local_elemwise_fusion_op(OP, max_input_fct=lambda node: 1024):
# worthwhile if the summation axis doesn't line up with a
# contiguous dimension)
if not isinstance(node.op, OP):
if type(node.op) is not OP:
return False
inputs = [] # inputs of the new Elemwise op.
s_inputs = [] # inputs of the new scalar op used by the Composite.
......
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