Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

[FEATURE] Fusing Leaky Relu operator in Fully Connected #19971

Merged
merged 1 commit into from
Apr 22, 2021
Merged

[FEATURE] Fusing Leaky Relu operator in Fully Connected #19971

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion src/operator/subgraph/mkldnn/mkldnn_fc-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,8 @@ static inline bool SupportMKLDNNFCEltwiseFusion(const std::string op_name) {
op_name == "sqrt" ||
op_name == "exp" ||
op_name == "abs" ||
op_name == "clip") {
op_name == "clip" ||
op_name == "LeakyReLU") {
return true;
} else {
return false;
Expand Down
16 changes: 14 additions & 2 deletions src/operator/subgraph/mkldnn/mkldnn_fc.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -286,8 +286,16 @@ void SgMKLDNNFCOp::Forward(const OpContext &ctx,
if (fuse_requantize || mkldnn_param.enable_float_output) {
float tmp_scale_ = 1.0f;
if (fuse_requantize) {
tmp_scale_ =
GetQuantizeScale(output.dtype(), cached_min_output_, cached_max_output_) / data_scale_;
if (mkldnn_param.with_eltwise) {
tmp_scale_ = 1.0 / data_scale_;
full_param_.eltwise_param.scale =
GetQuantizeScale(output.dtype(), cached_min_output_, cached_max_output_);
} else {
tmp_scale_ =
GetQuantizeScale(output.dtype(),
cached_min_output_,
cached_max_output_) / data_scale_;
}
} else {
tmp_scale_ = 1.0 / data_scale_;
}
Expand Down Expand Up @@ -405,6 +413,10 @@ static void SgMKLDNNFCParamParser(nnvm::NodeAttrs *attrs) {
if (op_name == "Activation") {
const ActivationParam act_param = nnvm::get<ActivationParam>(node->attrs.parsed);
full_param.eltwise_param.alg = GetMKLDNNActAlgo(act_param);
} else if (op_name == "LeakyReLU") {
const auto act_param = nnvm::get<LeakyReLUParam>(node->attrs.parsed);
full_param.eltwise_param.alpha = act_param.slope;
full_param.eltwise_param.alg = GetMKLDNNActAlgo(act_param);
} else if (op_name == "clip") {
const ClipParam clip_param = nnvm::get<ClipParam>(node->attrs.parsed);
full_param.eltwise_param.alg = mkldnn::algorithm::eltwise_bounded_relu;
Expand Down
10 changes: 10 additions & 0 deletions src/operator/subgraph/mkldnn/mkldnn_fc_property.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,16 @@ class SgMKLDNNFCSelector : public SubgraphSelector {
return true;
}
}
if (new_node.op() == Op::Get("LeakyReLU")) {
const LeakyReLUParam &param = nnvm::get<LeakyReLUParam>(new_node.attrs.parsed);
if (param.act_type == leakyrelu::kLeakyReLU ||
param.act_type == leakyrelu::kELU ||
param.act_type == leakyrelu::kGELU) {
matched_list_.push_back(&new_node);
status_ = kSuccess;
return true;
}
Comment on lines +107 to +113
Copy link
Member

@szha szha Apr 19, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The full set of supported activation types are {'elu', 'gelu', 'leaky', 'prelu', 'rrelu', 'selu'}. What about the rest of the activation types? Will they be supported by onednn in the future?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Those were activation types that can be easily fused with FC in this PR. We plan to enable other activation fuses in the next PR as they need different approach.

}
if (!quantized_ && (new_node.op() == Op::Get("square") ||
new_node.op() == Op::Get("sqrt") ||
new_node.op() == Op::Get("exp"))) {
Expand Down
4 changes: 3 additions & 1 deletion tests/python/mkl/test_subgraph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@
}

DATA_SHAPE=[(64, 4, 10, 10), (4, 3, 24, 24), (1, 16, 32, 32)]
fc_post_ops_list=['relu', 'sigmoid', 'tanh', 'softrelu',
fc_post_ops_list=['relu', 'sigmoid', 'tanh', 'softrelu', 'gelu',
'square', 'square_root', 'abs', 'exp', 'bounded_relu']

def check_qsym_calibrated(qsym, out_type, name='conv'):
Expand Down Expand Up @@ -664,6 +664,8 @@ def fc_eltwise(no_bias, data_shape, flatten=True, alg='relu'):
no_bias=no_bias, flatten=flatten)
if alg in ['relu', 'sigmoid', 'tanh', 'softrelu']:
sym = mx.symbol.Activation(data=fc, name='act', act_type=alg)
elif alg == "gelu":
sym = mx.symbol.LeakyReLU(data=fc, act_type='gelu')
elif alg == 'square':
sym = mx.symbol.square(data=fc, name='square')
elif alg == 'square_root':
Expand Down