IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Avoid dynamic rank in Snippets (#20387)

Browse Source 
* Avoid dynamic rank in Snippets

* Update src/common/snippets/src/pass/collapse_subgraph.cpp

Co-authored-by: Alexandra Sidorova <alexa.sanechka@ya.ru>

---------

Co-authored-by: Alexandra Sidorova <alexa.sanechka@ya.ru>
This commit is contained in:
Evgenya Nugmanova 2023-10-12 17:54:52 +04:00 committed by GitHub
parent 377e927149
commit 83c593be3b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/common/snippets/src/pass/collapse_subgraph.cpp
View File

@ -179,7 +179,10 @@ auto is_supported_op(const std::shared_ptr<const Node> &n) -> bool {
auto has_supported_in_out(const std::shared_ptr<const Node> &n) -> bool {
auto supported = [&n](descriptor::Tensor& t) -> bool {
// Todo: int32 isn't supported in general because i32 emitters are required for bit-exact i32 calculations in some cases
// TODO [122585] Need to add dynamic rank support
if (t.get_partial_shape().rank().is_dynamic())
return false;
// TODO [105804] int32 isn't supported in general because i32 emitters are required for bit-exact i32 calculations in some cases
// So i32 is supported exclusively for transposes and broadcast
return TokenizeSnippets::get_supported_element_types().count(t.get_element_type()) != 0 ||
(t.get_element_type() == ov::element::i32 &&