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

[LPT] Move FQ disabled for non-per-tensor FQs (#7581)

Browse Source
This commit is contained in:
Vladimir Zinoviev 2021-09-22 01:15:36 +03:00 committed by GitHub
parent be8600af38
commit 1aa6db4aaf
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 15 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
inference-engine/src/low_precision_transformations/src/move_fake_quantize.cpp
View File

@ -39,12 +39,26 @@ MoveFakeQuantize::MoveFakeQuantize(const Params& params) : LayerTransformation(p
output_low, output_low,
output_high }); output_high });
ngraph::graph_rewrite_callback callback = [this](pattern::Matcher& m) { ngraph::graph_rewrite_callback callback = [=](pattern::Matcher& m) {
auto op = m.get_match_root(); auto op = m.get_match_root();
if (transformation_callback(op)) { if (transformation_callback(op)) {
return false; return false;
} }
// workaround: only per-tensor quantization is allowed
const auto& pattern_map = m.get_pattern_value_map();
const auto is_scalar = [&](const std::shared_ptr<ngraph::Node>& wrapped_constant) {
return NetworkHelper::isScalarLike(
as_type_ptr<opset1::Constant>(pattern_map.at(wrapped_constant).get_node_shared_ptr()));
};
if (!is_scalar(input_low) ||
!is_scalar(input_high) ||
!is_scalar(output_low) ||
!is_scalar(output_high)) {
return false;
}
return transform(*context, m); return transform(*context, m);
}; };