Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Metal] fix Conv2d image #8496

Merged
merged 4 commits into from
Mar 4, 2022
Merged

[Metal] fix Conv2d image #8496

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
ed7dc35
fix
xiaoxiaohehe001 Jan 29, 2022
c3d7f75
fix
xiaoxiaohehe001 Feb 11, 2022
f5c2ea9
Merge branch 'develop' of /~https://github.com/PaddlePaddle/Paddle-Lite…
xiaoxiaohehe001 Feb 15, 2022
a23d90c
fix_conv2d
xiaoxiaohehe001 Feb 22, 2022
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
18 changes: 11 additions & 7 deletions lite/kernels/metal/image_op/conv2d_image_compute.mm
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,19 +110,22 @@
if (metal_context_->use_mps()) {
int input_c = static_cast<int>(input_buffer_->tensor_dim_[1]);
int output_c = static_cast<int>(output_buffer_->tensor_dim_[1]);
// intput & output C channel must >=3
// input channel must >=3
// attention: should be >=4, texture data layout is RGBA
if (input_c >= 3 && output_c >= 3) {
should_use_mps = true;
if (is_depthwise_) {
if (input_c >= 3 && output_c >= 3) {
should_use_mps = true;
}
} else {
if (input_c >= 3) {
should_use_mps = true;
}
}
}
}
if (IsWinoGrad(function_name_) || IsQuadruple(function_name_)) {
should_use_mps = false;
}
if (!is_depthwise_ && param.groups > 1) {
should_use_mps = false;
}
if (param.bias) {
if (!canMPSAddByChannel()) {
should_use_mps = false;
Expand Down Expand Up @@ -472,7 +475,7 @@
auto filter_h = static_cast<int>(param.filter->dims()[2]);
auto filter_w = static_cast<int>(param.filter->dims()[3]);
auto input_c = static_cast<int>(input_buffer_->tensor_dim_[1]);
auto output_c = static_cast<int>(output_buffer_->tensor_dim_[1]);
auto output_c = fmax(4, static_cast<int>(output_buffer_->tensor_dim_[1]));
MPSCNNConvolutionDescriptor* description = nil;
if (is_depthwise_) {
description = [MPSCNNDepthWiseConvolutionDescriptor
Expand All @@ -491,6 +494,7 @@
description.strideInPixelsY = param.strides[1];
description.dilationRateX = (*param.dilations)[0];
description.dilationRateY = (*param.dilations)[1];
if (!is_depthwise_) description.groups = param.groups;
// active function
switch (param.activation_param.active_type) {
case lite_api::ActivationType::kRelu: {
Expand Down