Skip to content

[BYOC][TRT]Fix groups cannot divide output channel count error for deconv when groups>1 #7595

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

Merged
masahi merged 3 commits into from
Mar 5, 2021
Merged

[BYOC][TRT]Fix groups cannot divide output channel count error for deconv when groups>1 #7595

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
aca8874
trt num_outputs
Mar 4, 2021
9a44284
asdf
Mar 4, 2021
ae56593
fix lint
Mar 5, 2021
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
12 changes: 6 additions & 6 deletions src/runtime/contrib/tensorrt/tensorrt_ops.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -309,8 +309,8 @@ class Conv3DOpConverter : public TensorRTOpConverter {
bool use_asymmetric_padding;
GetPadding3D(str_padding, &use_asymmetric_padding, &prepadding, &postpadding);

// Could use attrs->channels.as<IntImmNode>()->value
const int num_outputs = weight_shape[0];
const int num_outputs =
std::stoi(params->node.GetAttr<std::vector<std::string>>("channels")[0]);
const auto kernel_size = nvinfer1::Dims3(weight_shape[2], weight_shape[3], weight_shape[4]);
nvinfer1::Weights bias{nvinfer1::DataType::kFLOAT, nullptr, 0};
auto conv_layer = params->network->addConvolutionNd(*input_tensor, num_outputs, kernel_size,
Expand Down Expand Up @@ -788,8 +788,8 @@ class Conv2DTransposeOpConverter : public TensorRTOpConverter {
}
#endif

// Could use conv2d_attr->channels.as<IntImmNode>()->value
const int num_outputs = weight_shape[1];
const int num_outputs =
std::stoi(params->node.GetAttr<std::vector<std::string>>("channels")[0]);
const auto kernel_size = nvinfer1::DimsHW(weight_shape[2], weight_shape[3]);
nvinfer1::Weights bias{nvinfer1::DataType::kFLOAT, nullptr, 0};
auto deconv_layer = params->network->addDeconvolution(*input_tensor, num_outputs, kernel_size,
Expand Down Expand Up @@ -846,8 +846,8 @@ class Conv3DTransposeOpConverter : public TensorRTOpConverter {
bool use_asymmetric_padding;
GetPadding3D(str_padding, &use_asymmetric_padding, &prepadding, &postpadding);

// Could use attrs->channels.as<IntImmNode>()->value
const int num_outputs = weight_shape[1];
const int num_outputs =
std::stoi(params->node.GetAttr<std::vector<std::string>>("channels")[0]);
const auto kernel_size = nvinfer1::Dims3(weight_shape[2], weight_shape[3], weight_shape[4]);
nvinfer1::Weights bias{nvinfer1::DataType::kFLOAT, nullptr, 0};
auto deconv_layer = params->network->addDeconvolutionNd(*input_tensor, num_outputs, kernel_size,
Expand Down