lixiaoquan commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415261490
##########
File path: include/tvm/relay/op_attr_types.h
##########
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
https://github.com/apache/incubator-tvm/blob/6e1cd8256e461deca51e45e7aae33088e8f5b966/python/tvm/relay/op/_transform.py#L179
This line uses a range() and I think it causes variable index.
I tried to allocate a data shape tensor by tvm.te.placeholder(), but it
can't be passed to tvm.lower() when this shape func is lowered.
But data shape tensor allocated in this line will be stored in CachedNode
object and sent to tvm.lower(), that's why I have to pass data shape tensor
(https://github.com/apache/incubator-tvm/blob/2c1ca60ea38587401a20f11c5e64f452b79fa777/src/relay/backend/compile_engine.cc#L348)
Please correct me if I missunderstood your suggestion about allocating a
buffer.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
