Thanks to visit codestin.com
Credit goes to github.com

Skip to content

dnn (opencl): fix incorrect steps and elemsize when dtype changes #24968

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
merged 1 commit into from
Feb 6, 2024
Merged

dnn (opencl): fix incorrect steps and elemsize when dtype changes #24968

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
42 changes: 31 additions & 11 deletions modules/dnn/src/layers/nary_eltwise_layers.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,11 +47,12 @@ class NaryEltwiseHelper CV_FINAL
std::vector<char*> ptrs;
std::vector<std::vector<int>> shapes;
std::vector<std::vector<size_t>> steps;
std::vector<size_t> elemsize;

NaryEltwiseHelper() {
}

void helperInit(const std::vector<Mat>& inputs, const std::vector<Mat>& outputs)
void init(const std::vector<Mat>& inputs, const std::vector<Mat>& outputs)
{
narrays = 0;
max_ndims = 0;
Expand All @@ -61,6 +62,7 @@ class NaryEltwiseHelper CV_FINAL
ptrs.clear();
shapes.clear();
steps.clear();
elemsize.clear();

ninputs = inputs.size();
narrays = ninputs + 1;
Expand Down Expand Up @@ -95,15 +97,33 @@ class NaryEltwiseHelper CV_FINAL
}
orig_shapes.push_back(_size);
orig_steps.push_back(_step);

int esz = i == 0 ? outputs[0].elemSize() : inputs[i - 1].elemSize();
elemsize.push_back(esz);
}
}

void reInit(size_t newElemSize) {
std::vector<size_t> newElemSizes(elemsize.size(), newElemSize);
reInit(newElemSizes);
}

void reInit(std::vector<size_t> newElemSizes) {
for (size_t array_index = 0; array_index < orig_steps.size(); array_index++) {
auto &step = orig_steps[array_index];
int esz = elemsize[array_index];
int new_esz = newElemSizes[array_index];
for (size_t step_index = 0; step_index < step.size(); step_index++) {
step[step_index] = static_cast<size_t>(step[step_index] / esz * new_esz);
}
elemsize[array_index] = newElemSizes[array_index];
}
prepare_for_broadcast_op();
}

// use FP32 as default type in finalized() function
template <typename T>
bool prepare_for_broadcast_op()
{
int i, j, k;
std::vector<size_t> elemsize(this->all_ndims.size(), sizeof(T));

// step 1.
// * make all inputs and the output max_ndims-dimensional.
Expand Down Expand Up @@ -313,8 +333,8 @@ class NaryEltwiseLayerImpl CV_FINAL : public NaryEltwiseLayer
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

helper.helperInit(inputs, outputs);
CV_Assert(helper.prepare_for_broadcast_op<float>());
helper.init(inputs, outputs);
CV_Assert(helper.prepare_for_broadcast_op());
}

bool getMemoryShapes(const std::vector<MatShape> &inputs,
Expand Down Expand Up @@ -579,6 +599,7 @@ class NaryEltwiseLayerImpl CV_FINAL : public NaryEltwiseLayer

if (inputs_arr.depth() == CV_16F)
{
helper.reInit(sizeof(float));
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
Expand Down Expand Up @@ -733,14 +754,13 @@ class NaryEltwiseLayerImpl CV_FINAL : public NaryEltwiseLayer
switch (type)
{
case CV_8U:
// TODO: integrate with type inference
helper.reInit(sizeof(uint8_t));
opDispatch<uint8_t>(std::forward<Args>(args)...);
helper.prepare_for_broadcast_op<uint8_t>();
/*
recompute broadcasted shapes
because default type is FP32 which is calculated in finalize() function
*/
break;
case CV_32S:
// TODO: integrate with type inference
helper.reInit(sizeof(int32_t));
opDispatch<int32_t>(std::forward<Args>(args)...);
break;
case CV_32F:
Expand Down