提交 690b8a44 authored 作者: sentient07's avatar sentient07

Trying more opts and moved FusionOptimizer up

上级 a2fc1b7a
...@@ -1289,6 +1289,8 @@ class LocalOptGroup(LocalOptimizer): ...@@ -1289,6 +1289,8 @@ class LocalOptGroup(LocalOptimizer):
apply_mult_opts(opt_list, new_node, True) apply_mult_opts(opt_list, new_node, True)
return repl return repl
opts = self.track_map.get(type(node.op), []) opts = self.track_map.get(type(node.op), [])
opts += self.track_map.get(node.op, [])
opts += self.track_map.get(None, [])
return apply_mult_opts(opts, node, self.apply_all_opts) return apply_mult_opts(opts, node, self.apply_all_opts)
def print_summary(self, stream=sys.stdout, level=0, depth=-1): def print_summary(self, stream=sys.stdout, level=0, depth=-1):
......
...@@ -730,7 +730,11 @@ gpu_local_elemwise_fusion = tensor.opt.local_elemwise_fusion_op( ...@@ -730,7 +730,11 @@ gpu_local_elemwise_fusion = tensor.opt.local_elemwise_fusion_op(
GpuElemwise, GpuElemwise,
max_inputs_to_GpuElemwise) max_inputs_to_GpuElemwise)
optdb.register('gpua_elemwise_fusion', optdb.register('gpua_elemwise_fusion',
tensor.opt.FusionOptimizer(gpu_local_elemwise_fusion), 71.00, # 48.5 move to gpu
# 48.6 specialize
# 49 cpu fusion
# 49.5 add destroy handler
tensor.opt.FusionOptimizer(gpu_local_elemwise_fusion), 49,
'fast_run', 'fusion', 'local_elemwise_fusion', 'gpuarray') 'fast_run', 'fusion', 'local_elemwise_fusion', 'gpuarray')
inplace_gpu_elemwise_opt = tensor.opt.inplace_elemwise_optimizer_op( inplace_gpu_elemwise_opt = tensor.opt.inplace_elemwise_optimizer_op(
......
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