/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Copyright (C) 2017, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "../precomp.hpp"
#include "layers_common.hpp"
#include "../op_cuda.hpp"
#include "../op_halide.hpp"
#include "../op_inf_engine.hpp"
#include "../ie_ngraph.hpp"
#include "../op_vkcom.hpp"
#include "../op_webnn.hpp"
#include <opencv2/core/utils/configuration.private.hpp>
#include <opencv2/core/utils/logger.hpp>
#include "opencv2/core/hal/hal.hpp"
#include "opencv2/core/hal/intrin.hpp"
#include <iostream>
#include <numeric>
#ifdef HAVE_OPENCL
#include "opencl_kernels_dnn.hpp"
using namespace cv::dnn::ocl4dnn;
#endif
#ifdef HAVE_TENGINE
#include "../tengine4dnn/include/tengine_graph_convolution.hpp"
#endif
#ifdef HAVE_CUDA
#include "../cuda4dnn/primitives/convolution.hpp"
#include "../cuda4dnn/primitives/transpose_convolution.hpp"
using namespace cv::dnn::cuda4dnn;
#endif
namespace cv
{
namespace dnn
{
class BaseConvolutionLayerImpl : public ConvolutionLayer
{
public:
bool fusedWeights, fusedBias;
std::vector<double> weightsMultipliers;
#ifdef HAVE_WEBNN
int groups;
#endif
BaseConvolutionLayerImpl(const LayerParams ¶ms)
{
setParamsFrom(params);
getConvolutionKernelParams(params, kernel_size, pads_begin, pads_end, strides, dilations, padMode, adjust_pads);
numOutput = params.get<int>("num_output");
int ngroups = params.get<int>("group", 1);
#ifdef HAVE_WEBNN
groups = ngroups;
#endif
CV_Assert(numOutput % ngroups == 0);
if (kernel_size.size() == 2) {
kernel = Size(kernel_size[1], kernel_size[0]);
stride = Size(strides[1], strides[0]);
for (int i = 0; i < pads_begin.size(); i++) {
if (pads_begin[i] != pads_end[i])
CV_Error(Error::StsNotImplemented, "Unsupported asymmetric padding in convolution layer");
}
pad = Size(pads_begin[1], pads_begin[0]);
dilation = Size(dilations[1], dilations[0]);
adjustPad.height = adjust_pads[0];
adjustPad.width = adjust_pads[1];
}
for (int i = 0; i < adjust_pads.size(); i++) {
CV_Assert(adjust_pads[i] < strides[i]);
}
fusedWeights = false;
fusedBias = false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
CV_Assert((inputs.size() > outputs.size() && blobs.empty()) ||
(!inputs.empty() && (blobs.size() == 1 || blobs.size() == 2)));
MatSize weightShape = blobs.empty() ? inputs[1].size : blobs[0].size;
CV_Assert(inputs[0].dims == outputs[0].dims);
if (weightShape.dims() == 3)
{
kernel_size.assign(1, kernel_size[0]);
strides.assign(1, strides[0]);
dilations.assign(1, dilations[0]);
pads_begin.assign(1, pads_begin[0]);
pads_end.assign(1, pads_end[0]);
}
CV_Assert(weightShape.dims() == kernel_size.size() + 2);
for (int i = 0; i < kernel_size.size(); i++) {
CV_Assert(weightShape[i + 2] == kernel_size[i]);
}
const Mat &input = inputs[0];
CV_Assert(((input.dims == 3 && kernel_size.size() == 1) || input.dims == 4 || input.dims == 5) && (input.type() == CV_32F || input.type() == CV_16S));
for (size_t i = 0; i < outputs.size(); i++)
{
CV_Assert(inputs[i].type() == input.type());
CV_Assert(((input.dims == 3 && kernel_size.size() == 1) || inputs[i].dims == 4 || inputs[i].dims == 5) && inputs[i].size[1] == input.size[1]);
for (int j = 0; j < inputs[i].dims; j++) {
CV_Assert(inputs[i].size[j] == input.size[j]);
}
}
std::vector<int> inpShape;
std::vector<int> outShape;
for (int i = 2; i < inputs[0].dims; i++) {
inpShape.push_back(inputs[0].size[i]);
outShape.push_back(outputs[0].size[i]);
}
getConvPoolPaddings(inpShape, kernel_size, strides, padMode, pads_begin, pads_end);
if (pads_begin.size() == 2) {
for (int i = 0; i < pads_begin.size(); i++) {
if (pads_begin[i] != pads_end[i])
CV_Error(Error::StsNotImplemented, "Unsupported asymmetric padding in convolution layer");
}
pad = Size(pads_begin[1], pads_begin[0]);
}
fusedWeights = false;
fusedBias = false;
}
bool hasBias() const
{
return blobs.size() >= 2;
}
virtual MatShape computeColRowShape(const MatShape &inpShape, const MatShape &outShape) const = 0;
bool is1x1() const
{
return (kernel.height == 1 && kernel.width == 1) &&
(stride.height == 1 && stride.width == 1) &&
(dilation.height == 1 && dilation.width == 1);
}
virtual bool tryFuse(Ptr<Layer>& top) CV_OVERRIDE
{
Ptr<BlankLayer> blank_layer = top.dynamicCast<BlankLayer>();
if (blank_layer)
return true;
Mat w, b;
top->getScaleShift(w, b);
if (!w.empty() || !b.empty())
{
fuseWeights(w, b);
fusedWeights = fusedWeights || !w.empty();
fusedBias = fusedBias || (hasBias() && !w.empty()) || !b.empty();
return true;
}
return false;
}
virtual void fuseWeights(const Mat& w_, const Mat& b_) = 0;
virtual void applyHalideScheduler(Ptr<BackendNode>& node,
const std::vector<Mat*> &inputs,
const std::vector<Mat> &outputs,
int targetId) const CV_OVERRIDE
{
#ifdef HAVE_HALIDE
if (targetId != DNN_TARGET_CPU)
{
Layer::applyHalideScheduler(node, inputs, outputs, targetId);
return;
}
Halide::Var x("x"), y("y"), c("c"), n("n"), tile("tile"), yi("yi"), yo("yo"), co("co"), ci("ci");
Halide::Func& top = node.dynamicCast<HalideBackendNode>()->funcs[1];
Halide::Func& padded_input = node.dynamicCast<HalideBackendNode>()->funcs[0];
int outW, outH, outC, outN;
getCanonicalSize(outputs[0].size, &outW, &outH, &outC, &outN);
if (outW == 1 || outH <= 2)
return;
if (is1x1() || outC <= 16)
top.reorder(x, c, y)
.split(y, yo, yi, 2)
.fuse(yo, n, tile)
.parallel(tile)
.unroll(yi)
.vectorize(x, outW >= 16 ? 16 : outW);
else
top.reorder(x, c, y)
.split(y, yo, yi, 2)
.split(c, co, ci, 16)
.fuse(yo, co, tile).fuse(n, tile, tile)
.parallel(tile)
.unroll(yi)
.vectorize(x, outW >= 16 ? 16 : outW);
padded_input.compute_at(top, yi);
#endif // HAVE_HALIDE
}
};
#define IS_POWER_LAYER(layer) \
(!layer.empty() && !layer->type.compare("Power"))
//TODO: simultaneously convolution and bias addition for cache optimization
class ConvolutionLayerImpl CV_FINAL : public BaseConvolutionLayerImpl
{
public:
enum { VEC_ALIGN = 8, DFT_TYPE = CV_32F };
Mat weightsMat;
std::vector<float> biasvec;
std::vector<float> reluslope;
Ptr<ActivationLayer> activ;
#ifdef HAVE_OPENCL
Ptr<OCL4DNNConvSpatial<float> > convolutionOp;
std::vector<UMat> umat_blobs;
bool newActiv;
ocl4dnnFusedActiv_t activType;
float power;
#endif
#ifdef HAVE_TENGINE
teng_graph_t tengine_graph;
#endif
#ifdef HAVE_CUDA
cuda4dnn::ConvolutionConfiguration::FusionMode cudaFusionMode;
cuda4dnn::ConvolutionConfiguration::ActivationType cudaActType;
float cuda_relu_slope, cuda_crelu_floor, cuda_crelu_ceil;
float cuda_power_exp, cuda_power_scale, cuda_power_shift;
#endif
ConvolutionLayerImpl(const LayerParams ¶ms) : BaseConvolutionLayerImpl(params)
{
#ifdef HAVE_OPENCL
newActiv = false;
activType = OCL4DNN_CONV_FUSED_ACTIV_NONE;
power = 0.f;
#endif
#ifdef HAVE_CUDA
cudaFusionMode = cuda4dnn::ConvolutionConfiguration::FusionMode::NONE;
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::IDENTITY;
#endif
#ifdef HAVE_TENGINE
tengine_graph=NULL;
#endif
}
#ifdef HAVE_TENGINE
~ConvolutionLayerImpl()
{
if(NULL != tengine_graph )
{
tengine_release(tengine_graph);
}
}
#endif
MatShape computeColRowShape(const MatShape &inpShape, const MatShape &outShape) const CV_OVERRIDE
{
CV_Assert(!blobs.empty());
int dims = inpShape.size();
int inpD = dims == 5 ? inpShape[2] : 1;
int inpH = inpShape[dims - 2];
int inpW = inpShape.back();
int inpGroupCn = blobs[0].size[1];
int ksize = inpGroupCn * std::accumulate(kernel_size.begin(), kernel_size.end(),
1, std::multiplies<size_t>());
return shape(inpD * inpH * inpW, ksize);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
size_t ksize = kernel_size.size();
#ifdef HAVE_CUDA
if (backendId == DNN_BACKEND_CUDA)
{
/* only 1d, 2d and 3d convolutions supported */
if (ksize > 0 && ksize <= 3)
return true;
return false;
}
#endif
#ifdef HAVE_INF_ENGINE
if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
{
bool isArmTarget = preferableTarget == DNN_TARGET_CPU && isArmComputePlugin();
if (isArmTarget && blobs.empty())
return false;
if (ksize == 1)
return isArmTarget;
if (ksize == 3)
return preferableTarget != DNN_TARGET_MYRIAD && !isArmTarget;
bool isMyriad = preferableTarget == DNN_TARGET_MYRIAD || preferableTarget == DNN_TARGET_HDDL;
if (!isMyriad && blobs.empty())
return false;
return (!isMyriad || dilation.width == dilation.height);
}
#endif
if (backendId == DNN_BACKEND_OPENCV)
return ksize >= 1 && ksize <= 3;
#ifdef HAVE_HALIDE
if (backendId == DNN_BACKEND_HALIDE)
return ksize == 2 && !blobs.empty();
#endif
#ifdef HAVE_VULKAN
if (backendId == DNN_BACKEND_VKCOM)
return ksize == 2;
#endif
#ifdef HAVE_WEBNN
if (backendId == DNN_BACKEND_WEBNN)
{
if (ksize != 2)
{
CV_LOG_WARNING(NULL, "WebNN only supports Conv2d.");
return false;
}
return true;
}
#endif
return false;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE
{
CV_Assert(!blobs.empty() || inputs.size() > 1);
const int* weightShape = blobs.empty() ? &inputs[1][0] : blobs[0].size.p;
CV_Assert(!hasBias() || blobs[1].total() == (size_t)weightShape[0]);
internals.clear();
CV_Assert(inputs.size() != 0);
std::vector<int> inpShape(inputs[0].begin() + 2, inputs[0].end());
int outCn = weightShape[0];
std::vector<int> outShape;
outShape.push_back(inputs[0][0]);
outShape.push_back(outCn);
int inpCn = inputs[0][1];
if (padMode.empty())
{
for (int i = 0; i < inpShape.size(); i++)
outShape.push_back((inpShape[i] + pads_begin[i] + pads_end[i] - dilations[i] * (kernel_size[i] - 1) - 1) / strides[i] + 1);
}
else
{
getConvPoolOutParams(inpShape, kernel_size, strides, padMode, dilations, outShape);
}
int ngroups = inpCn / weightShape[1];
if (ngroups == 0 || ngroups * weightShape[1] != inpCn)
CV_Error(Error::StsError, format("Number of input channels should "
"be multiple of %d but got %d", weightShape[1], inpCn));
CV_Assert(ngroups > 0 && inpCn % ngroups == 0 && outCn % ngroups == 0);
outputs.resize(1, outShape);
return false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
BaseConvolutionLayerImpl::finalize(inputs_arr, outputs_arr);
std::vector<Mat> inputs;
inputs_arr.getMatVector(inputs);
// prepare weightsMat where each row is aligned and has enough zero padding on the right to
// use vectorized (i.e. with intrinsics) loops without tail processing
if (!blobs.empty())
{
Mat wm = blobs[0].reshape(1, numOutput);
if ((wm.step1() % VEC_ALIGN != 0) ||
!isAligned<VEC_ALIGN * sizeof(float)>(wm.data)
)
{
int newcols = (int)alignSize(wm.step1(), VEC_ALIGN);
Mat wm_buffer = Mat(numOutput, newcols, wm.type());
Mat wm_padding = wm_buffer.colRange(wm.cols, newcols);
wm_padding.setTo(Scalar::all(0.));
Mat wm_aligned = wm_buffer.colRange(0, wm.cols);
wm.copyTo(wm_aligned);
wm = wm_aligned;
}
weightsMat = wm;
}
else
{
// initialized in .forward()
weightsMat.release();
}
weightsMultipliers.assign(numOutput, 1.0);
Mat biasMat = hasBias() ? blobs[1].reshape(1, numOutput) : Mat();
biasvec.resize(numOutput+2);
if( biasMat.empty() )
{
for(int i = 0; i < numOutput; i++ )
biasvec[i] = 0.f;
}
else
{
for(int i = 0; i < numOutput; i++ )
biasvec[i] = biasMat.at<float>(i);
}
#ifdef HAVE_TENGINE
if(NULL != tengine_graph )
{
tengine_release(tengine_graph);
tengine_graph = NULL ;
}
#endif
#ifdef HAVE_OPENCL
convolutionOp.release();
#endif
}
bool setActivation(const Ptr<ActivationLayer>& layer) CV_OVERRIDE
{
if ((!activ.empty() && !layer.empty()) || blobs.empty())
return false;
activ = layer;
if (activ.empty())
reluslope.clear();
#ifdef HAVE_OPENCL
newActiv = true;
activType = OCL4DNN_CONV_FUSED_ACTIV_NONE;
if (IS_DNN_OPENCL_TARGET(preferableTarget))
{
Ptr<PowerLayer> activ_power = activ.dynamicCast<PowerLayer>();
if (!activ_power.empty())
{
if (activ_power->scale != 1.0f) // not supported well by implementation, #17964
{
// FIXIT no way to check number of blobs (like, eltwise input)
CV_LOG_DEBUG(NULL, "DNN/OpenCL: can't configure Power activation (scale != 1.0f)");
activ.release();
newActiv = false;
return false;
}
if (activ_power->scale != 1.f || activ_power->shift != 0.f)
{
const int outCh = blobs[0].size[0];
fuseWeights(Mat(1, outCh, CV_32F, Scalar(activ_power->scale)),
Mat(1, outCh, CV_32F, Scalar(activ_power->shift)));
}
power = activ_power->power;
activType = OCL4DNN_CONV_FUSED_ACTIV_POWER;
}
Ptr<TanHLayer> activ_tanh = activ.dynamicCast<TanHLayer>();
if (!activ_tanh.empty())
{
activType = OCL4DNN_CONV_FUSED_ACTIV_TANH;
}
}
#endif
#ifdef HAVE_CUDA
if (activ.empty())
{
/* setActivation was called with empty argument => reset all fusions */
cudaFusionMode = cuda4dnn::ConvolutionConfiguration::FusionMode::NONE;
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::IDENTITY;
}
if(IS_DNN_CUDA_TARGET(preferableTarget))
{
CV_Assert(cudaFusionMode == ConvolutionConfiguration::FusionMode::NONE ||
cudaFusionMode == ConvolutionConfiguration::FusionMode::ELTWISE_SUM);
Ptr<ReLULayer> activ_relu = activ.dynamicCast<ReLULayer>();
if(!activ_relu.empty())
{
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::RELU;
cuda_relu_slope = activ_relu->negativeSlope;
}
Ptr<ReLU6Layer> activ_relu6 = activ.dynamicCast<ReLU6Layer>();
if(!activ_relu6.empty())
{
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::CLIPPED_RELU;
cuda_crelu_floor = activ_relu6->minValue;
cuda_crelu_ceil = activ_relu6->maxValue;
}
Ptr<PowerLayer> activ_power = activ.dynamicCast<PowerLayer>();
if (!activ_power.empty())
{
cuda_power_scale = activ_power->scale;
cuda_power_shift = activ_power->shift;
cuda_power_exp = activ_power->power;
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::POWER;
}
Ptr<TanHLayer> activ_tanh = activ.dynamicCast<TanHLayer>();
if(!activ_tanh.empty())
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::TANH;
Ptr<SigmoidLayer> activ_sigmoid = activ.dynamicCast<SigmoidLayer>();
if(!activ_sigmoid.empty())
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::SIGMOID;
Ptr<SwishLayer> activ_swish = activ.dynamicCast<SwishLayer>();
if(!activ_swish.empty())
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::SWISH;
Ptr<MishLayer> activ_mish = activ.dynamicCast<MishLayer>();
if(!activ_mish.empty())
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::MISH;
if (cudaActType == cuda4dnn::ConvolutionConfiguration::ActivationType::IDENTITY)
{
/* no activation fused */
activ.reset();
}
else
{
/* activation was fused */
if (cudaFusionMode == ConvolutionConfiguration::FusionMode::NONE) /* no previous fusion */
cudaFusionMode = ConvolutionConfiguration::FusionMode::ACTIVATION; /* now activation */
else if (cudaFusionMode == ConvolutionConfiguration::FusionMode::ELTWISE_SUM) /* previously eltwise was fused */
cudaFusionMode = ConvolutionConfiguration::FusionMode::ELTWISE_SUM_THEN_ACTIVATION; /* now activation on eltwise output */
}
}
#endif
return !activ.empty();
}
virtual bool tryFuse(Ptr<Layer>& top) CV_OVERRIDE
{
#ifdef HAVE_CUDA
if(IS_DNN_CUDA_TARGET(preferableTarget))
{
Ptr<EltwiseLayer> eltwise = top.dynamicCast<EltwiseLayer>();
if (!eltwise.empty()) // && eltwise->op == EltwiseLayer::SUM && eltwise->coeffs.empty())
{
/* we also need to check that the eltwise input does not require shortcut mechanism
* it's difficult to verify it here but we hope that `fuseLayers` has done the check already
*/
if (cudaFusionMode == ConvolutionConfiguration::FusionMode::NONE)
{
/* no previous fusion */
cudaFusionMode = ConvolutionConfiguration::FusionMode::ELTWISE_SUM; /* now eltwise */
return true;
}
else if(cudaFusionMode == ConvolutionConfiguration::FusionMode::ACTIVATION)
{
/* previously an activation was fused */
cudaFusionMode = ConvolutionConfiguration::FusionMode::ACTIVATION_THEN_ELTWISE_SUM;
return true;
}
return false;
}
}
#endif
return BaseConvolutionLayerImpl::tryFuse(top);
}
void fuseWeights(const Mat& w_, const Mat& b_) CV_OVERRIDE
{
// Convolution weights have OIHW data layout. Parameters fusion in case of
// (conv(I) + b1 ) * w + b2
// means to replace convolution's weights to [w*conv(I)] and bias to [b1 * w + b2]
const int outCn = weightsMat.size[0];
Mat w = w_.total() == 1 ? Mat(1, outCn, CV_32F, Scalar(w_.at<float>(0))) : w_;
Mat b = b_.total() == 1 ? Mat(1, outCn, CV_32F, Scalar(b_.at<float>(0))) : b_;
CV_Assert_N(!weightsMat.empty(), biasvec.size() == outCn + 2,
w.empty() || outCn == w.total(), b.empty() || outCn == b.total());
if (!w.empty())
{
// Keep origin weights unchanged.
if (weightsMat.data == blobs[0].data)
weightsMat = weightsMat.clone();
Mat originWeights = blobs[0].reshape(1, outCn);
for (int i = 0; i < outCn; ++i)
{
double wi = w.at<float>(i);
weightsMultipliers[i] *= wi;
cv::multiply(originWeights.row(i), weightsMultipliers[i], weightsMat.row(i));
biasvec[i] *= wi;
}
}
if (!b.empty())
{
for (int i = 0; i < outCn; ++i)
biasvec[i] += b.at<float>(i);
}
biasvec[outCn] = biasvec[outCn+1] = biasvec[outCn-1];
}
virtual Ptr<BackendNode> initVkCom(const std::vector<Ptr<BackendWrapper> > &inputs) CV_OVERRIDE
{
#ifdef HAVE_VULKAN
CV_Assert(!blobs.empty());
int out_channel = blobs[0].size[0];
bool has_bias = hasBias() || fusedBias;
int filter_size[2] = {kernel.height, kernel.width};
int pad_size[2] = {pad.height, pad.width};
int stride_size[2] = {stride.height, stride.width};
int dilation_size[2] = {dilation.height, dilation.width};
int activation = 0;
vkcom::Tensor input_tensor = VkComTensor(inputs[0]);
int in_channel = input_tensor.dimSize(1);
int group = in_channel / blobs[0].size[1];
// TODO: support group > 1
if (group != 1)
return Ptr<BackendNode>();
int padding_mode;
if (padMode.empty())
{
padding_mode = vkcom::kPaddingModeCaffe;
}
else if (padMode == "VALID")
{
padding_mode = vkcom::kPaddingModeValid;
}
else if (padMode == "SAME")
{
padding_mode = vkcom::kPaddingModeSame;
}
else
CV_Error(Error::StsError, "Unsupported padding mode " + padMode);
std::shared_ptr<vkcom::OpBase> op(new vkcom::OpConv(out_channel, has_bias,
filter_size, pad_size,
stride_size, dilation_size,
activation, group,
padding_mode));
std::vector<Ptr<BackendWrapper> > blobsWrapper;
if (fusedWeights)
{
Mat wm;
weightsMat.copyTo(wm); // to handle the case of isContinuous() == false
wm = wm.reshape(1, blobs[0].dims, blobs[0].size);
blobsWrapper.push_back(Ptr<BackendWrapper>(new VkComBackendWrapper(wm)));
}
else
{
blobsWrapper.push_back(Ptr<BackendWrapper>(new VkComBackendWrapper(blobs[0])));
}
if (has_bias)
{
Mat biasesMat({out_channel}, CV_32F, &biasvec[0]);
blobsWrapper.push_back(Ptr<BackendWrapper>(new VkComBackendWrapper(biasesMat)));
}
return Ptr<BackendNode>(new VkComBackendNode(inputs, op, blobsWrapper));
#endif // HAVE_VULKAN
return Ptr<BackendNode>();
}
virtual Ptr<BackendNode> initHalide(const std::vector<Ptr<BackendWrapper> > &inputs) CV_OVERRIDE
{
#ifdef HAVE_HALIDE
CV_Assert(!blobs.empty());
Halide::Buffer<float> inputBuffer = halideBuffer(inputs[0]);
const int inpCn = inputBuffer.channels();
const int outCn = blobs[0].size[0];
const int inpGroupCn = blobs[0].size[1];
const int group = inpCn / inpGroupCn;
const int outGroupCn = outCn / group;
Halide::Buffer<float> weights = wrapToHalideBuffer(blobs[0]);
Halide::Var x("x"), y("y"), c("c"), n("n");
Halide::Func top = (name.empty() ? Halide::Func() : Halide::Func(name));
Halide::Func padded_input(name + "_constant_exterior");
if (pad.width || pad.height)
{
Halide::Func bounded =
Halide::BoundaryConditions::constant_exterior(inputBuffer, 0);
padded_input(x, y, c, n) = bounded(x, y, c, n);
}
else
{
padded_input(x, y, c, n) = inputBuffer(x, y, c, n);
}
Halide::RDom r(0, kernel.width, 0, kernel.height, 0, inpGroupCn);
Halide::Expr kx = x * stride.width - pad.width + r.x * dilation.width;
Halide::Expr ky = y * stride.height - pad.height + r.y * dilation.height;
Halide::Expr kc = r.z;
for (int i = 1; i < group; ++i)
{
kc = select(c < outGroupCn * i, kc, inpGroupCn * i + r.z);
}
Halide::Expr topExpr = sum(padded_input(kx, ky, kc, n) *
weights(r.x, r.y, r.z, c));
if (hasBias())
{
Halide::Buffer<float> bias = wrapToHalideBuffer(blobs[1], {outCn});
topExpr += bias(c);
}
top(x, y, c, n) = topExpr;
return Ptr<BackendNode>(new HalideBackendNode({ padded_input, top }));
#endif // HAVE_HALIDE
return Ptr<BackendNode>();
}
#ifdef HAVE_DNN_NGRAPH
virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper> > &inputs,
const std::vector<Ptr<BackendNode> >& nodes) CV_OVERRIDE
{
CV_Assert(!blobs.empty());
CV_Assert_N(inputs.size() >= 1, nodes.size() >= 1);
auto& ieInpNode = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
std::vector<size_t> dims = ieInpNode->get_shape();
CV_Check(dims.size(), dims.size() >= 3 && dims.size() <= 5, "");
std::shared_ptr<ngraph::Node> ieWeights = nodes.size() > 1 ? nodes[1].dynamicCast<InfEngineNgraphNode>()->node : nullptr;
if (nodes.size() > 1)
CV_Assert(ieWeights); // dynamic_cast should not fail
const int inpCn = dims[1];
const int inpGroupCn = nodes.size() > 1 ? ieWeights->get_shape()[1] : blobs[0].size[1];
const int group = inpCn / inpGroupCn;
std::vector<size_t> kernel_shape;
if (group != 1)
{
kernel_shape.push_back(group);
}
kernel_shape.push_back(numOutput / group);
kernel_shape.push_back(inpCn / group);
std::copy(kernel_size.begin(), kernel_size.end(), back_inserter(kernel_shape));
if (nodes.size() == 1)
{
ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, kernel_shape, blobs[0].data);
if (fusedWeights)
{
if (weightsMat.isContinuous())
{
ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, kernel_shape, weightsMat.data);
}
else
{
Mat newWeights;
Mat cvWeights = weightsMat.colRange(0, blobs[0].total() / numOutput);
cvWeights.copyTo(newWeights);
ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, kernel_shape, newWeights.data);
}
}
}
else
{
auto shape = std::make_shared<ngraph::op::Constant>(ngraph::element::i64,
ngraph::Shape{kernel_shape.size()}, std::vector<int64_t>(kernel_shape.begin(), kernel_shape.end()));
ieWeights = std::make_shared<ngraph::op::v1::Reshape>(ieWeights, shape, true);
}
ngraph::op::PadType pad_type = ngraph::op::PadType::EXPLICIT;
if (!padMode.empty())
pad_type = padMode == "VALID" ? ngraph::op::PadType::VALID : ngraph::op::PadType::SAME_UPPER;
std::shared_ptr<ngraph::Node> conv_node;
if (group != 1) {
conv_node = std::make_shared<ngraph::op::v1::GroupConvolution>(
ieInpNode, ieWeights,
ngraph::Strides(strides),
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(pads_begin.begin(), pads_begin.end())),
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(pads_end.begin(), pads_end.end())),
ngraph::Strides(dilations),
pad_type);
} else {
conv_node = std::make_shared<ngraph::op::v1::Convolution>(
ieInpNode, ieWeights,
ngraph::Strides(strides),
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(pads_begin.begin(), pads_begin.end())),
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(pads_end.begin(), pads_end.end())),
ngraph::Strides(dilations),
pad_type);
}
if (hasBias() || fusedBias || nodes.size() == 3)
{
std::vector<size_t> shape(conv_node->get_shape().size(), 1);
shape[1] = conv_node->get_shape()[1];
std::shared_ptr<ngraph::Node> bias;
if (nodes.size() == 3)
{
auto bias_shape = std::make_shared<ngraph::op::Constant>(ngraph::element::i64,
ngraph::Shape{shape.size()}, std::vector<int64_t>(shape.begin(), shape.end()));
bias = std::make_shared<ngraph::op::v1::Reshape>(nodes[2].dynamicCast<InfEngineNgraphNode>()->node, bias_shape, true);
}
else
{
bias = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, ngraph::Shape(shape), biasvec.data());
}
auto conv_bias = std::make_shared<ngraph::op::v1::Add>(conv_node, bias, ngraph::op::AutoBroadcastType::NUMPY);
return Ptr<BackendNode>(new InfEngineNgraphNode(conv_bias));
}
return Ptr<BackendNode>(new InfEngineNgraphNode(conv_node));
}
#endif // HAVE_DNN_NGRAPH
#ifdef HAVE_WEBNN
virtual Ptr<BackendNode> initWebnn(const std::vector<Ptr<BackendWrapper> >& inputs, const std::vector<Ptr<BackendNode> >& nodes) CV_OVERRIDE
{
CV_Assert(!blobs.empty());
CV_Assert_N(inputs.size() >= 1, nodes.size() >= 1);
Ptr<WebnnBackendNode> node = nodes[0].dynamicCast<WebnnBackendNode>();
auto& webnnInpOperand = node->operand;
auto& webnnGraphBuilder = node->net->builder;
ml::Operand webnnWeights = nodes.size() > 1 ? nodes[1].dynamicCast<WebnnBackendNode>()->operand : nullptr;
if (nodes.size() > 1)
CV_Assert(webnnWeights);
const int inpCn = weightsMat.total()/(kernel_size[0]*kernel_size[1]*numOutput);
const int group = groups;
const int inpGroupCn = inpCn / group;
std::vector<int32_t> kernel_shape;
if (group != 1)
{
kernel_shape.push_back(group);
}
kernel_shape.push_back(numOutput / group);
kernel_shape.push_back(inpGroupCn);
std::copy(kernel_size.begin(), kernel_size.end(), back_inserter(kernel_shape));
if (nodes.size() == 1)
{
webnnWeights = webnn::BuildConstant(webnnGraphBuilder, webnn::getShape(blobs[0]), blobs[0].data, blobs[0].total()*blobs[0].elemSize(), ml::OperandType::Float32);
if (fusedWeights)
{
if (weightsMat.isContinuous())
{
webnnWeights = webnn::BuildConstant(webnnGraphBuilder, webnn::getShape(weightsMat), weightsMat.data, weightsMat.total()*weightsMat.elemSize(), ml::OperandType::Float32);
}
else
{
Mat newWeights;
Mat cvWeights = weightsMat.colRange(0, blobs[0].total() / numOutput);
cvWeights.copyTo(newWeights);
webnnWeights = webnn::BuildConstant(webnnGraphBuilder, webnn::getShape(newWeights), newWeights.data, newWeights.total()*newWeights.elemSize(), ml::OperandType::Float32);
}
}
}
else
{
webnnWeights = webnnGraphBuilder.Reshape(webnnWeights, kernel_shape.data(), kernel_shape.size());
}
ml::AutoPad pad_type = ml::AutoPad::Explicit;
if (!padMode.empty())
pad_type = padMode == "VALID" ? ml::AutoPad::Explicit : ml::AutoPad::SameUpper;
ml::Conv2dOptions options = {};
options.groups = group;
options.autoPad = pad_type;
std::vector<int32_t> Strides(strides.begin(), strides.end());
if (!Strides.empty())
{
options.stridesCount = Strides.size();
options.strides = Strides.data();
}
std::vector<int32_t> Padding;
if (padMode.empty())
{
Padding = {static_cast<int32_t>(pads_begin[0]),
static_cast<int32_t>(pads_end[0]),
static_cast<int32_t>(pads_begin[1]),
static_cast<int32_t>(pads_end[1])};
}
else if (padMode == "VALID")
{
Padding = {0, 0, 0, 0};
}
if (!Padding.empty())
{
options.paddingCount = Padding.size();
options.padding = Padding.data();
}
std::vector<int32_t> Dilations(dilations.begin(), dilations.end());
if (!Dilations.empty())
{
options.dilationsCount = Dilations.size();
options.dilations = Dilations.data();
}
ml::Operand operand = webnnGraphBuilder.Conv2d(webnnInpOperand, webnnWeights, &options);
// ml::Operand result = operand;
if (hasBias() || fusedBias || nodes.size() == 3)
{
ml::Operand webnnBias = nullptr;
if (nodes.size() == 3)
{
std::vector<int32_t> bias_shape = {1, numOutput, 1, 1};
webnnBias = webnnGraphBuilder.Reshape(nodes[2].dynamicCast<WebnnBackendNode>()->operand, bias_shape.data(), bias_shape.size());
}
else
{
webnnBias = webnn::BuildConstant(webnnGraphBuilder, {1, numOutput, 1, 1}, biasvec.data(), (numOutput) * sizeof(float), ml::OperandType::Float32);
}
operand = webnnGraphBuilder.Add(operand, webnnBias);
}
return Ptr<BackendNode>(new WebnnBackendNode(operand));
}
#endif // HAVE_WEBNN
class ParallelConv : public cv::ParallelLoopBody
{
public:
enum { BLK_SIZE = 32, BLK_SIZE_CN = 64 };
const Mat* input_;
const Mat* weights_;
Mat* output_;
int outShape[4]; // used only for conv2d
std::vector<size_t> kernel_size, pads_begin, pads_end, strides, dilations;
int ngroups_, nstripes_;
std::vector<int> ofstab_;
const std::vector<float>* biasvec_;
const std::vector<float>* reluslope_;
const ActivationLayer* activ_;
bool is1x1_;
bool useAVX;
bool useAVX2;
bool useAVX512;
bool useRVV;
int blk_size_cn;
ParallelConv()
: input_(0), weights_(0), output_(0), ngroups_(0), nstripes_(0),
biasvec_(0), reluslope_(0), activ_(0), is1x1_(false), useAVX(false), useAVX2(false), useAVX512(false), useRVV(false)
, blk_size_cn(0)
{}
static void run( const Mat& input, Mat& output, const Mat& weights,
const std::vector<float>& biasvec,
const std::vector<float>& reluslope,
const std::vector<size_t>& kernel_size, const std::vector<size_t>& strides,
const std::vector<size_t>& pads_begin, const std::vector<size_t>& pads_end,
const std::vector<size_t>& dilations,
const ActivationLayer* activ, int ngroups, int nstripes )
{
size_t karea = std::accumulate(kernel_size.begin(), kernel_size.end(),
1, std::multiplies<size_t>());
bool isConv1D = input.dims == 3;
bool isConv2D = input.dims == 4;
bool isConv3D = input.dims == 5;
CV_CheckEQ(static_cast<int>(kernel_size.size()), input.dims - 2, "");
CV_Assert_N(input.dims == output.dims,
input.size[0] == output.size[0],
weights.rows == output.size[1],
weights.cols == (input.size[1]/ngroups)*karea,
input.type() == output.type(),
input.type() == weights.type(),
input.type() == CV_32FC1,
input.isContinuous(),
output.isContinuous(),
biasvec.size() == (size_t)output.size[1]+2);
CV_Check(weights.step1(), weights.step1() % VEC_ALIGN == 0, "");
CV_CheckType(weights.type(), CV_32FC1, "");
ParallelConv p;
p.input_ = &input;
p.weights_ = &weights;
p.output_ = &output;
int max_ind = isConv1D? 3: 4;
for( int i = 0; i < max_ind; i++ ) p.outShape[i] = output.size[i];
p.outShape[1] /= ngroups;
p.kernel_size = kernel_size; p.strides = strides; p.dilations = dilations;
p.pads_begin = pads_begin; p.pads_end = pads_end;
p.ngroups_ = ngroups;
p.nstripes_ = nstripes;
int inpCnAll = input.size[1];
int depth = (input.dims == 5) ? input.size[2] : 1;
int width = input.size[input.dims - 1];
int height = isConv1D? 1 : input.size[input.dims - 2];
int inpCn = inpCnAll / ngroups;
p.is1x1_ = (isConv2D && kernel_size[0] == 1 && kernel_size[1] == 1 &&
pads_begin[0] == 0 && pads_begin[1] == 0) ||
(isConv1D && pads_begin[0] == 0 && kernel_size[0] == 1);
p.useAVX = checkHardwareSupport(CPU_AVX) && isConv2D;
p.useAVX2 = checkHardwareSupport(CPU_AVX2) && isConv2D;
p.useAVX512 = CV_CPU_HAS_SUPPORT_AVX512_SKX && isConv2D;
p.useRVV = checkHardwareSupport(CPU_RVV) && isConv2D;
int kernel_d = isConv3D? kernel_size[0] : 1;
int kernel_h = isConv1D? 1 : kernel_size[kernel_size.size() - 2];
int kernel_w = kernel_size.back();
int blk_size_cn0 = cvCeil(800./(kernel_w*kernel_h));
int ncn = 16;
while (ncn*2 < blk_size_cn0 && ncn < inpCn)
ncn *= 2;
ncn = std::min(ncn, inpCn);
p.blk_size_cn = ncn;
int dil_d = isConv3D? dilations[0] : 1;
int dil_h = isConv1D? 1 : dilations[dilations.size() - 2];
int dil_w = dilations.back();
p.ofstab_.resize(karea * ncn);
int* ofstab = &p.ofstab_[0];
if (isConv1D)
{
for( int k = 0; k < ncn; k++ )
for( int k_c = 0; k_c < kernel_w; k_c++ )
ofstab[k*kernel_w + k_c] = k*width + k_c*dil_w;
}
else if (isConv2D)
{
for( int k = 0; k < ncn; k++ )
for( int k_r = 0; k_r < kernel_h; k_r++ )
for( int k_c = 0; k_c < kernel_w; k_c++ )
ofstab[(k*kernel_h + k_r)*kernel_w + k_c] =
(k*height + k_r*dil_h)*width + k_c*dil_w;
}
else
{
for( int k = 0; k < ncn; k++ )
for (int k_d = 0; k_d < kernel_d; k_d++)
for( int k_r = 0; k_r < kernel_h; k_r++ )
for( int k_c = 0; k_c < kernel_w; k_c++ )
ofstab[(k*kernel_d*kernel_h + k_d*kernel_h + k_r)*kernel_w + k_c] =
(k*depth*height + k_d*dil_d*height + k_r*dil_h)*width + k_c*dil_w;
}
p.biasvec_ = &biasvec;
p.reluslope_ = &reluslope;
p.activ_ = p.reluslope_->empty() ? activ : 0;
parallel_for_(Range(0, nstripes), p, nstripes);
}
virtual void operator ()(const Range &r0) const CV_OVERRIDE
{
const int valign = ConvolutionLayerImpl::VEC_ALIGN;
int ngroups = ngroups_, batchSize = input_->size[0]*ngroups;
bool isConv1D = input_->dims == 3;
bool isConv2D = input_->dims == 4;
bool isConv3D = input_->dims == 5;
int outW = output_->size[output_->dims - 1];
int outH = isConv1D? 1 : output_->size[output_->dims - 2];
int outCn = output_->size[1]/ngroups;
int depth = isConv3D? input_->size[2] : 1;
int height = isConv1D? 1 : input_->size[input_->dims - 2];
int width = input_->size[input_->dims - 1];
int inpCn = input_->size[1]/ngroups;
const int nstripes = nstripes_;
int kernel_d = isConv3D? kernel_size[0] : 1;
int kernel_h = isConv1D? 1 : kernel_size[kernel_size.size() - 2];
int kernel_w = kernel_size.back();
int karea = kernel_w*kernel_h*kernel_d;
int pad_d = isConv3D? pads_begin[0] : 0;
int pad_t = isConv1D? 0 : pads_begin[pads_begin.size() - 2];
int pad_l = pads_begin.back();
int stride_d = isConv3D? strides[0] : 0;
int stride_h = isConv1D? 0 : strides[strides.size() - 2];
int stride_w = strides.back();
int dilation_d = isConv3D? dilations[0] : 1;
int dilation_h = isConv1D? 1 : dilations[dilations.size() - 2];
int dilation_w = dilations.back();
int i, j, k, d;
int inpPlaneSize = (int)input_->total(2);
int outPlaneSize = (int)output_->total(2);
bool is1x1 = is1x1_;
int stripesPerSample;
int stripeSize;
Range r = r0;
bool depthWiseConvolution = !is1x1 && isConv2D && ngroups > 1 && inpCn == 1 &&
outCn == 1 && kernel_d == 1 && dilation_d == 1 && stride_d == 0 && pad_d == 0 &&
width >= 16 + dilation_w*(kernel_w - 1);
// for now only 3x3 depth-wise convolutions are supported
depthWiseConvolution = depthWiseConvolution && kernel_w == 3 && kernel_h == 3 &&
// computing at most 1 pixel from each side can involve padding
max(stride_w, dilation_w) >= pad_l && max(stride_h, dilation_h) >= pad_t &&
pad_l <= 1 && pad_t <= 1;
if( !depthWiseConvolution && nstripes >= batchSize*2 )
{
stripesPerSample = nstripes/batchSize;
stripeSize = (int)alignSize((outPlaneSize + stripesPerSample - 1)/stripesPerSample, valign);
stripeSize = std::min(stripeSize, outPlaneSize);
}
else
{
stripesPerSample = 1;
int samplesPerStripe = std::max((batchSize + nstripes - 1)/nstripes, 1);
r.start *= samplesPerStripe;
r.end *= samplesPerStripe;
stripeSize = outPlaneSize;
}
const float* data_inp0_ = input_->ptr<float>();
const int* ofstab = &ofstab_[0];
const float* wptr_orig_ = weights_->ptr<float>();
size_t wstep = weights_->step1();
const float* biasptr_ = &biasvec_->at(0);
const float* reluptr_ = reluslope_->empty() ? 0 : &reluslope_->at(0);
float* data_out0_ = output_->ptr<float>();
AutoBuffer<float> rowbuf0_;
float* rowbuf0 = 0;
bool use_rowbuf = !depthWiseConvolution;
int blk_size = depthWiseConvolution ? outPlaneSize : min((int)BLK_SIZE, stripeSize);
// im2row buffer is not used for depth-wise convolution
if(use_rowbuf)
{
size_t rowbufsz = alignSize(karea*blk_size_cn, valign)*min((int)BLK_SIZE, blk_size);
//printf("karea=%d, blk_size_cn=%d, rowbufsz=%d, stripeSize=%d\n", karea, blk_size_cn, (int)rowbufsz, stripeSize);
rowbuf0_.allocate(rowbufsz + valign);
rowbuf0 = alignPtr(rowbuf0_.data(), (int)(valign*sizeof(float)));
// we clear the buffer once; ultimately, it lets us to avoid
// tail processing after running the unrolled/vectorized loop.
// the main idea is to make sure that the tail (a.k.a. padding) of each row
// (i.e. the elements with indices between vsz=karea*ncn and vsz_a)
// does not contain NaNs or Infs. Because the padding in the weights
// matrix is explicitly initialized with 0's, we handle all other
// cases nicely, i.e. we can skip expliciting re-initialization
// of the padding - we just retain elements from the previous iteration
// of the loop over channels (cn0).
memset(rowbuf0, 0, rowbufsz*sizeof(rowbuf0[0]) );
}
for( int stripe = r.start; stripe < r.end; stripe++ )
{
int subsampleIdx = stripe/stripesPerSample;
if( subsampleIdx >= batchSize )
break;
int stripeStart = (int)((stripe - subsampleIdx*stripesPerSample)*stripeSize);
int stripeEnd = (int)std::min(stripeStart + stripeSize, outPlaneSize);
const float* data_inp0 = data_inp0_ + subsampleIdx*inpPlaneSize*inpCn;
float* data_out0 = data_out0_ + subsampleIdx*outPlaneSize*outCn;
int startOutCn = (subsampleIdx % ngroups)*outCn;
const float* wptr_orig = wptr_orig_ + wstep*startOutCn;
const float* biasptr = biasptr_ + startOutCn;
for( int cn0 = 0; cn0 < inpCn; cn0 += blk_size_cn )
{
int cn1 = std::min(cn0 + blk_size_cn, inpCn);
int ncn = cn1 - cn0, vsz = karea*ncn;
int vsz_a = (int)alignSize(vsz, valign);
const float* wptr = wptr_orig + cn0*karea;
// we apply [Channels][P]ReLU (if any) during the final pass only.
const float* relu = cn1 == inpCn && reluptr_ ? reluptr_ + startOutCn : 0;
for( int ofs0 = stripeStart; ofs0 < stripeEnd; ofs0 += blk_size )
{
int ofs, ofs1 = std::min(ofs0 + blk_size, stripeEnd);
int bsz = ofs1 - ofs0;
int out_d = ofs0 / (outH * outW);
int out_i = (ofs0 - out_d * outH * outW) / outW;
int out_j = ofs0 % outW;
if (depthWiseConvolution)
{
CV_Assert(out_i == 0 && out_j == 0);
int in_d = out_d * stride_d - pad_d;
const float* inptr_ = data_inp0 + (cn0*depth*height + in_d*height)*width;
float* outptr_ = data_out0 + ofs0;
#if CV_TRY_AVX2
if(useAVX2)
opt_AVX2::fastDepthwiseConv(wptr, kernel_h, kernel_w,
stride_h, stride_w, dilation_h, dilation_w, pad_t, pad_l,
biasptr, relu, inptr_, height, width, outptr_, out_d, outH, outW);
else
#endif
#if CV_TRY_AVX
if(useAVX)
opt_AVX::fastDepthwiseConv(wptr, kernel_h, kernel_w,
stride_h, stride_w, dilation_h, dilation_w, pad_t, pad_l,
biasptr, relu, inptr_, height, width, outptr_, out_d, outH, outW);
else
#endif
#if CV_TRY_RVV
if(useRVV)
opt_RVV::fastDepthwiseConv(wptr, kernel_h, kernel_w,
stride_h, stride_w, dilation_h, dilation_w, pad_t, pad_l,
biasptr, relu, inptr_, height, width, outptr_, out_d, outH, outW);
else
#endif
{
const float w00_ = wptr[0], w01_ = wptr[1], w02_ = wptr[2],
w10 = wptr[3], w11 = wptr[4], w12 = wptr[5],
w20_ = wptr[6], w21_ = wptr[7], w22_ = wptr[8];
int outW1 = min(outW, (width - dilation_w*(kernel_w - 1) + pad_l)/stride_w);
float relu_coeff = relu ? relu[out_d] : 1.f, bias = biasptr[out_d];
for (int out_i = 0; out_i < outH; out_i++)
{
int in_i = out_i * stride_h - pad_t, out_j = 0;
const float* imgptr0 = inptr_ + in_i*width;
const float* imgptr1 = imgptr0 + dilation_h*width;
const float* imgptr2 = imgptr0 + (dilation_h*2)*width;
float out, w00 = w00_, w01 = w01_, w02 = w02_;
float w20 = w20_, w21 = w21_, w22 = w22_;
if (in_i < 0)
{
w00 = w01 = w02 = 0.f;
imgptr0 = imgptr1;
}
else if (in_i + dilation_h*(kernel_h-1) >= height)
{
w20 = w21 = w22 = 0.f;
imgptr2 = imgptr1;
}
float* outptr = outptr_ + out_i*outW;
if (pad_l > 0)
{
out = imgptr0[0]*w01 + imgptr0[dilation_w]*w02 +
imgptr1[0]*w11 + imgptr1[dilation_w]*w12 +
imgptr2[0]*w21 + imgptr2[dilation_w]*w22 + bias;
if (relu)
out = out > 0.f ? out : out*relu_coeff;
outptr[0] = out;
out_j = 1;
}
#if CV_SIMD
// maybe with AVX or AVX512 strided depthwise convolution
// can be accelerated with vector code, but with 4xfloat vectors
// it's hardly the case
if( stride_w == 1 )
{
const int VECSZ = v_float32::nlanes;
const int out_delta = VECSZ/stride_w;
v_float32 vw00 = vx_setall_f32(w00), vw01 = vx_setall_f32(w01), vw02 = vx_setall_f32(w02),
vw10 = vx_setall_f32(w10), vw11 = vx_setall_f32(w11), vw12 = vx_setall_f32(w12),
vw20 = vx_setall_f32(w20), vw21 = vx_setall_f32(w21), vw22 = vx_setall_f32(w22);
v_float32 z = vx_setzero_f32(), vbias = vx_setall_f32(bias), vrc = vx_setall_f32(relu_coeff);
for( ; out_j < outW1; out_j += out_delta )
{
if (out_j + out_delta > outW1)
{
if (out_j <= pad_l)
break;
out_j = outW1 - out_delta;
}
int in_j = out_j * stride_w - pad_l;
v_float32 v00 = vx_load(imgptr0 + in_j),
v01 = vx_load(imgptr0 + in_j + dilation_w),
v02 = vx_load(imgptr0 + in_j + dilation_w*2),
v10 = vx_load(imgptr1 + in_j),
v11 = vx_load(imgptr1 + in_j + dilation_w),
v12 = vx_load(imgptr1 + in_j + dilation_w*2),
v20 = vx_load(imgptr2 + in_j),
v21 = vx_load(imgptr2 + in_j + dilation_w),
v22 = vx_load(imgptr2 + in_j + dilation_w*2);
v_float32 vout = v00*vw00 + v01*vw01 + v02*vw02 +
v10*vw10 + v11*vw11 + v12*vw12 +
v20*vw20 + v21*vw21 + v22*vw22 + vbias;
if (relu)
vout = v_select(vout > z, vout, vout*vrc);
v_store(outptr + out_j, vout);
}
}
#endif
for (; out_j < outW1; out_j++)
{
int in_j = out_j * stride_w - pad_l;
out = imgptr0[in_j]*w00 + imgptr0[in_j + dilation_w]*w01 + imgptr0[in_j + dilation_w*2]*w02 +
imgptr1[in_j]*w10 + imgptr1[in_j + dilation_w]*w11 + imgptr1[in_j + dilation_w*2]*w12 +
imgptr2[in_j]*w20 + imgptr2[in_j + dilation_w]*w21 + imgptr2[in_j + dilation_w*2]*w22 + bias;
if (relu)
out = out > 0.f ? out : out*relu_coeff;
outptr[out_j] = out;
}
for (; out_j < outW; out_j++ )
{
int in_j0 = out_j * stride_w - pad_l, in_j1 = in_j0 + dilation_w, in_j2 = in_j0 + dilation_w*2;
float s0 = 1.f, s1 = 1.f, s2 = 1.f;
if (in_j0 >= width)
{
in_j0 = 0;
s0 = 0.f;
}
if (in_j1 >= width)
{
in_j1 = 0;
s1 = 0.f;
}
if (in_j2 >= width)
{
in_j2 = 0;
s2 = 0.f;
}
out = imgptr0[in_j0]*w00*s0 + imgptr0[in_j1]*w01*s1 + imgptr0[in_j2]*w02*s2 +
imgptr1[in_j0]*w10*s0 + imgptr1[in_j1]*w11*s1 + imgptr1[in_j2]*w12*s2 +
imgptr2[in_j0]*w20*s0 + imgptr2[in_j1]*w21*s1 + imgptr2[in_j2]*w22*s2 + bias;
if (relu)
out = out > 0.f ? out : out*relu_coeff;
outptr[out_j] = out;
}
}
}
continue;
}
// do im2row for a part of input tensor
float* rowbuf = rowbuf0;
if (isConv1D)
{
for( ofs = ofs0; ofs < ofs1; out_j = 0, ++out_i )
{
int delta = std::min(ofs1 - ofs, outW - out_j);
int out_j1 = out_j + delta;
int in_j = out_j * stride_w - pad_l;
const float* imgptr = data_inp0 + cn0*width + in_j;
ofs += delta;
// do im2row for a part of input tensor
if( is1x1 )
{
for( ; out_j < out_j1; out_j++, rowbuf += vsz_a, imgptr += stride_w )
{
for( k = 0; k < vsz; k++ )
rowbuf[k] = imgptr[k*inpPlaneSize];
}
}
else
{
for( ; out_j < out_j1; out_j++, rowbuf += vsz_a, imgptr += stride_w, in_j += stride_w )
{
// this condition should be true for most of the tensor elements, i.e.
// most of the time the kernel aperture is inside the tensor X-Y plane.
if( out_j + 2 <= out_j1 && 0 <= in_j && in_j + stride_w*2 <= width - (kernel_w-1)*dilation_w )
{
for( k = 0; k < vsz; k++ )
{
int k1 = ofstab[k];
float v0 = imgptr[k1];
float v1 = imgptr[k1 + stride_w];
rowbuf[k] = v0;
rowbuf[k+vsz_a] = v1;
}
out_j++;
rowbuf += vsz_a;
imgptr += stride_w;
in_j += stride_w;
}
else
{
int i0 = std::max(0, (-in_j + dilation_w-1)/dilation_w);
int i1 = std::min(kernel_w, (width - in_j + dilation_w-1)/dilation_w);
// here some non-continuous sub-row of the row will not be
// filled from the tensor; we need to make sure that the uncovered
// elements are explicitly set to 0's. the easiest way is to
// set all the elements to 0's before the loop.
memset(rowbuf, 0, vsz*sizeof(rowbuf[0]));
for( k = 0; k < ncn; k++ )
{
for( i = i0; i < i1; i++ )
{
int imgofs = k*width + i*dilation_w;
rowbuf[k*kernel_w + i] = imgptr[imgofs];
}
}
}
}
}
}
}
else if (isConv2D)
{
if( is1x1 && stride_w == 1 && stride_h == 1 )
{
const float* imgptr = data_inp0 + (cn0*height + out_i)*width + out_j;
for( int j = 0; j < bsz; j++, rowbuf += vsz_a )
{
if( j + 4 <= bsz )
{
k = 0;
#if CV_SIMD128
for( ; k <= vsz - 4; k += 4 )
{
const float* inp = imgptr + j + k*inpPlaneSize;
v_float32x4 p0 = v_load(inp), p1 = v_load(inp + inpPlaneSize);
v_float32x4 p2 = v_load(inp + inpPlaneSize*2), p3 = v_load(inp + inpPlaneSize*3);
v_float32x4 r0, r1, r2, r3;
v_transpose4x4(p0, p1, p2, p3, r0, r1, r2, r3);
v_store(rowbuf + k, r0);
v_store(rowbuf + k + vsz_a, r1);
v_store(rowbuf + k + vsz_a*2, r2);
v_store(rowbuf + k + vsz_a*3, r3);
}
#endif
for( ; k < vsz; k++ )
{
const float* inp = imgptr + j + k*inpPlaneSize;
float v0 = inp[0], v1 = inp[1], v2 = inp[2], v3 = inp[3];
rowbuf[k] = v0;
rowbuf[k + vsz_a] = v1;
rowbuf[k + vsz_a*2] = v2;
rowbuf[k + vsz_a*3] = v3;
}
j += 3;
rowbuf += vsz_a*3;
}
else
{
for( k = 0; k < vsz; k++ )
{
rowbuf[k] = imgptr[j + k*inpPlaneSize];
}
}
}
}
else
for( ofs = ofs0; ofs < ofs1; out_j = 0, ++out_i )
{
int delta = std::min(ofs1 - ofs, outW - out_j);
int out_j1 = out_j + delta;
int in_i = out_i * stride_h - pad_t;
int in_j = out_j * stride_w - pad_l;
const float* imgptr = data_inp0 + (cn0*height + in_i)*width + in_j;
ofs += delta;
// do im2row for a part of input tensor
if( is1x1 )
{
for( ; out_j < out_j1; out_j++, rowbuf += vsz_a, imgptr += stride_w )
{
for( k = 0; k < vsz; k++ )
rowbuf[k] = imgptr[k*inpPlaneSize];
}
}
else
{
bool ok_i = 0 <= in_i && in_i < height - (kernel_h-1)*dilation_h;
int i0 = std::max(0, (-in_i + dilation_h-1)/dilation_h);
int i1 = std::min(kernel_h, (height - in_i + dilation_h-1)/dilation_h);
for( ; out_j < out_j1; out_j++, rowbuf += vsz_a, imgptr += stride_w, in_j += stride_w )
{
// this condition should be true for most of the tensor elements, i.e.
// most of the time the kernel aperture is inside the tensor X-Y plane.
if( ok_i && out_j + 2 <= out_j1 && 0 <= in_j && in_j + stride_w*2 <= width - (kernel_w-1)*dilation_w )
{
for( k = 0; k < vsz; k++ )
{
int k1 = ofstab[k];
float v0 = imgptr[k1];
float v1 = imgptr[k1 + stride_w];
rowbuf[k] = v0;
rowbuf[k+vsz_a] = v1;
}
out_j++;
rowbuf += vsz_a;
imgptr += stride_w;
in_j += stride_w;
}
else
{
int j0 = std::max(0, (-in_j + dilation_w-1)/dilation_w);
int j1 = std::min(kernel_w, (width - in_j + dilation_w-1)/dilation_w);
// here some non-continuous sub-row of the row will not be
// filled from the tensor; we need to make sure that the uncovered
// elements are explicitly set to 0's. the easiest way is to
// set all the elements to 0's before the loop.
memset(rowbuf, 0, vsz*sizeof(rowbuf[0]));
for( k = 0; k < ncn; k++ )
{
for( i = i0; i < i1; i++ )
{
for( j = j0; j < j1; j++ )
{
int imgofs = k*(width*height) + i*(dilation_h*width) + j*dilation_w;
rowbuf[(k*kernel_h + i)*kernel_w + j] = imgptr[imgofs];
}
}
}
}
}
}
}
}
else
{
for( ofs = ofs0; ofs < ofs1; out_d += (out_i + 1) / outH, out_i = (out_i + 1) % outH, out_j = 0 )
{
int delta = std::min(ofs1 - ofs, outW - out_j);
int out_j1 = out_j + delta;
int in_d = out_d * stride_d - pad_d;
int in_i = out_i * stride_h - pad_t;
int in_j = out_j * stride_w - pad_l;
const float* imgptr = data_inp0 + (cn0*depth*height + in_d*height + in_i)*width + in_j;
ofs += delta;
int d0 = std::max(0, (-in_d + dilation_d - 1) / dilation_d);
int d1 = std::min(kernel_d, (depth - in_d + dilation_d - 1) / dilation_d);
int i0 = std::max(0, (-in_i + dilation_h-1)/dilation_h);
int i1 = std::min(kernel_h, (height - in_i + dilation_h-1)/dilation_h);
for( ; out_j < out_j1; out_j++, rowbuf += vsz_a, imgptr += stride_w, in_j += stride_w )
{
int j0 = std::max(0, (-in_j + dilation_w-1)/dilation_w);
int j1 = std::min(kernel_w, (width - in_j + dilation_w-1)/dilation_w);
// here some non-continuous sub-row of the row will not be
// filled from the tensor; we need to make sure that the uncovered
// elements are explicitly set to 0's. the easiest way is to
// set all the elements to 0's before the loop.
memset(rowbuf, 0, vsz*sizeof(rowbuf[0]));
for( k = 0; k < ncn; k++ )
{
for ( d = d0; d < d1; d++)
{
for( i = i0; i < i1; i++ )
{
for( j = j0; j < j1; j++ )
{
int imgofs = k*(depth*width*height) + d*dilation_d*width*height + i*(dilation_h*width) + j*dilation_w;
rowbuf[(k*kernel_d*kernel_h + d*kernel_h + i)*kernel_w + j] = imgptr[imgofs];
}
}
}
}
}
}
}
// now compute dot product of the weights
// and im2row-transformed part of the tensor
#if CV_TRY_AVX512_SKX
/* AVX512 convolution requires an alignment of 16, and ROI is only there for larger vector sizes */
if(useAVX512)
opt_AVX512_SKX::fastConv(wptr, wstep, biasptr, rowbuf0, data_out0 + ofs0,
outShape, bsz, vsz, vsz_a, relu, cn0 == 0);
else
#endif
#if CV_TRY_AVX2
if(useAVX2)
opt_AVX2::fastConv(wptr, wstep, biasptr, rowbuf0, data_out0 + ofs0,
outShape, bsz, vsz, vsz_a, relu, cn0 == 0);
else
#endif
#if CV_TRY_AVX
if(useAVX)
opt_AVX::fastConv(wptr, wstep, biasptr, rowbuf0, data_out0 + ofs0,
outShape, bsz, vsz, vsz_a, relu, cn0 == 0);
else
#endif
#if CV_TRY_RVV
if(useRVV)
opt_RVV::fastConv(wptr, wstep, biasptr, rowbuf0, data_out0 + ofs0,
outShape, bsz, vsz, vsz_a, relu, cn0 == 0);
else
#endif
for( int i = 0; i < outCn; i += 2 )
{
const float* wptr0 = wptr + i*wstep;
const float* wptr1 = wptr0 + wstep;
float* outptr0 = data_out0 + ofs0 + i*outPlaneSize;
float* outptr1 = outptr0 + outPlaneSize;
float bias0 = biasptr[i], bias1 = biasptr[i+1];
float r0 = 1.f, r1 = 1.f;
if( i+1 >= outCn )
{
wptr1 = wptr0;
outptr1 = outptr0;
bias1 = bias0;
}
if( relu )
{
r0 = relu[i]; r1 = relu[i+1];
if( i+1 >= outCn )
r1 = r0;
}
int j = 0;
#if CV_SIMD128
v_float32x4 vr0 = v_setall_f32(r0), vr1 = v_setall_f32(r1), z = v_setzero_f32();
for( ; j <= bsz - 4; j += 4 )
{
const float* rptr = rowbuf0 + j*vsz_a;
v_float32x4 s0, s1;
if( cn0 == 0 )
{
s0 = v_setall_f32(bias0);
s1 = v_setall_f32(bias1);
}
else
{
s0 = v_load(outptr0 + j);
s1 = v_load(outptr1 + j);
}
v_float32x4 vs00 = v_setzero_f32(), vs01 = v_setzero_f32(),
vs02 = v_setzero_f32(), vs03 = v_setzero_f32(),
vs10 = v_setzero_f32(), vs11 = v_setzero_f32(),
vs12 = v_setzero_f32(), vs13 = v_setzero_f32();
for( k = 0; k < vsz; k += 4, rptr += 4 )
{
v_float32x4 w0 = v_load_aligned(wptr0 + k);
v_float32x4 w1 = v_load_aligned(wptr1 + k);
v_float32x4 r0 = v_load_aligned(rptr);
v_float32x4 r1 = v_load_aligned(rptr + vsz_a);
v_float32x4 r2 = v_load_aligned(rptr + vsz_a*2);
v_float32x4 r3 = v_load_aligned(rptr + vsz_a*3);
vs00 = v_fma(w0, r0, vs00);
vs01 = v_fma(w0, r1, vs01);
vs02 = v_fma(w0, r2, vs02);
vs03 = v_fma(w0, r3, vs03);
vs10 = v_fma(w1, r0, vs10);
vs11 = v_fma(w1, r1, vs11);
vs12 = v_fma(w1, r2, vs12);
vs13 = v_fma(w1, r3, vs13);
}
s0 += v_reduce_sum4(vs00, vs01, vs02, vs03);
s1 += v_reduce_sum4(vs10, vs11, vs12, vs13);
if( relu )
{
s0 = v_select(s0 > z, s0, s0*vr0);
s1 = v_select(s1 > z, s1, s1*vr1);
}
v_store(outptr0 + j, s0);
v_store(outptr1 + j, s1);
}
#endif
for( ; j < bsz; j++ )
{
const float* rptr = rowbuf0 + j*vsz_a;
float s00, s10;
if( cn0 == 0 )
{
s00 = bias0;
s10 = bias1;
}
else
{
s00 = outptr0[j];
s10 = outptr1[j];
}
for( k = 0; k < vsz; k++ )
{
float r0 = rptr[k];
s00 += wptr0[k]*r0;
s10 += wptr1[k]*r0;
}
if( relu )
{
s00 = s00 > 0.f ? s00 : s00*r0;
s10 = s10 > 0.f ? s10 : s10*r1;
}
outptr0[j] = s00;
outptr1[j] = s10;
}
}
}
}
if( activ_ )
activ_->forwardSlice(data_out0 + stripeStart, data_out0 + stripeStart,
(int)(stripeEnd - stripeStart),
outPlaneSize, startOutCn, startOutCn + outCn);
}
}
};
#ifdef HAVE_OPENCL
bool forward_ocl(InputArrayOfArrays inps, OutputArrayOfArrays outs, OutputArrayOfArrays internals)
{
if (kernel_size.size() != 2)
{
// no OpenCL optimizations, see .supportedBacked()
return false;
}
std::vector<UMat> inputs;
std::vector<UMat> outputs;
bool use_half = (inps.depth() == CV_16S);
inps.getUMatVector(inputs);
outs.getUMatVector(outputs);
CV_Assert(outputs.size() == 1);
for (int i = 0; i < inputs.size(); ++i)
CV_Assert(inputs[i].u != outputs[0].u);
if (blobs.empty())
{
size_t n = inputs.size() - 1;
umat_blobs.resize(n);
for (size_t i = 0; i < n; i++)
{
inputs[i + 1].copyTo(umat_blobs[i]);
}
inputs.resize(1);
}
if (umat_blobs.empty())
{
size_t n = blobs.size();
umat_blobs.resize(n);
for (size_t i = 0; i < n; i++)
{
if (use_half)
convertFp16(blobs[i], umat_blobs[i]);
else
blobs[i].copyTo(umat_blobs[i]);
}
}
if (convolutionOp.empty() || blobs.empty())
{
OCL4DNNConvConfig config;
config.in_shape = shape(inputs[0]);
config.out_shape = shape(outputs[0]);
config.kernel = kernel;
config.pad = pad;
config.stride = stride;
config.dilation = dilation;
if (inputs[0].dims != 4 && inputs[0].dims != umat_blobs[0].dims)
{
static bool bypassCheck = utils::getConfigurationParameterBool("OPENCV_OCL4DNN_CONVOLUTION_IGNORE_INPUT_DIMS_4_CHECK", false);
if (!bypassCheck)
{
CV_LOG_ERROR(NULL, "DNN/OpenCL: Unsupported configuration: inputs[0].dims=" << inputs[0].dims << " umat_blobs[0].dims=" << umat_blobs[0].dims
<< ". Consider reporting complete reproducer to https://github.com/opencv/opencv/issues/20833."
<< " You can skip this check temporary through OPENCV_OCL4DNN_CONVOLUTION_IGNORE_INPUT_DIMS_4_CHECK=1"
);
return false;
}
}
config.group = inputs[0].size[1] / umat_blobs[0].size[1];
if (config.group < 1) // config.group == 0 causes div by zero in ocl4dnn code
{
CV_LOG_WARNING(NULL, "DNN/OpenCL: Unsupported config.group=" << config.group
<< ". Consider reporting complete reproducer to https://github.com/opencv/opencv/issues/20833"
);
return false;
}
config.bias_term = umat_blobs.size() == 2;
config.use_half = use_half;
convolutionOp = Ptr<OCL4DNNConvSpatial<float> >(new OCL4DNNConvSpatial<float>(config));
}
int outCn = umat_blobs[0].size[0];
reluslope.clear();
if( activ )
{
Ptr<ReLULayer> activ_relu = activ.dynamicCast<ReLULayer>();
if( !activ_relu.empty() )
{
reluslope.assign(outCn+2, activ_relu->negativeSlope);
activType = OCL4DNN_CONV_FUSED_ACTIV_RELU;
}
Ptr<ReLU6Layer> activ_relu6 = activ.dynamicCast<ReLU6Layer>();
if( !activ_relu6.empty() )
{
reluslope.resize(2);
reluslope[0] = activ_relu6->minValue;
reluslope[1] = activ_relu6->maxValue;
activType = OCL4DNN_CONV_FUSED_ACTIV_RELU6;
}
Ptr<ChannelsPReLULayer> activ_chprelu = activ.dynamicCast<ChannelsPReLULayer>();
if( !activ_chprelu.empty() )
{
const Mat& m = activ_chprelu->blobs[0];
CV_Assert(m.isContinuous() && m.type() == CV_32F && (int)m.total() == outCn);
const float* mdata = m.ptr<float>();
reluslope.resize(outCn+2);
std::copy(mdata, mdata + outCn, reluslope.begin());
reluslope[outCn] = reluslope[outCn+1] = reluslope[outCn-1];
activType = OCL4DNN_CONV_FUSED_ACTIV_PRELU;
}
}
if (fusedWeights)
{
if (use_half)
convertFp16(weightsMat, umat_blobs[0]);
else
weightsMat.copyTo(umat_blobs[0]);
fusedWeights = false;
}
if (fusedBias)
{
if ( umat_blobs.size() < 2 )
umat_blobs.resize(2);
if (use_half)
convertFp16(Mat(biasvec, true), umat_blobs[1]);
else
Mat(biasvec, true).copyTo(umat_blobs[1]);
convolutionOp->setBias(true);
fusedBias = false;
}
if ( newActiv )
{
if ( activType == OCL4DNN_CONV_FUSED_ACTIV_RELU )
{
CV_Assert(!reluslope.empty());
convolutionOp->setActivReLU(true, reluslope[0]);
}
else if ( activType == OCL4DNN_CONV_FUSED_ACTIV_PRELU)
{
CV_Assert(!reluslope.empty());
convolutionOp->setActivPReLU(true, reluslope);
}
else if ( activType == OCL4DNN_CONV_FUSED_ACTIV_POWER)
{
convolutionOp->setActivPower(true, power);
}
else if ( activType == OCL4DNN_CONV_FUSED_ACTIV_TANH)
{
convolutionOp->setActivTanh(true);
}
else if ( activType == OCL4DNN_CONV_FUSED_ACTIV_RELU6)
{
convolutionOp->setActivReLU6(true, reluslope[0], reluslope[1]);
}
else
{
convolutionOp->setActivReLU(false, 0);
convolutionOp->setActivPReLU(false, reluslope);
convolutionOp->setActivPower(false, 1.f);
convolutionOp->setActivTanh(false);
convolutionOp->setActivReLU6(false, 0, 0);
}
newActiv = false;
}
UMat& inpMat = inputs[0];
UMat& outMat = outputs[0];
int batch_size = inpMat.size[0];
return convolutionOp->Forward(inpMat,
inputs.size() == 2 ? inputs[1] : UMat(),
umat_blobs[0],
umat_blobs.size() > 1 ? umat_blobs[1] : UMat(),
outMat,
batch_size);
}
#endif
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
int outCn = blobs.empty() ? inputs[1].size[0] : blobs[0].size[0];
// Need to align non-const blobs
if (blobs.empty())
{
Mat wm = inputs[1].reshape(1, outCn);
if (wm.data != weightsMat.data)
{
int newcols = (int)alignSize(wm.step1(), VEC_ALIGN);
Mat wm_buffer = Mat(numOutput, newcols, wm.type());
Mat wm_padding = wm_buffer.colRange(wm.cols, newcols);
wm_padding.setTo(Scalar::all(0.));
weightsMat = wm_buffer.colRange(0, wm.cols);
wm.copyTo((const Mat&)weightsMat);
if (inputs.size() > 2)
{
Mat biasMat = inputs[2].reshape(1, outCn);
biasMat.col(0).copyTo(biasvec);
}
biasvec.resize(outCn + 2, 0);
}
}
/*if (inputs[0].dims > 3) {
printf("conv %s: input (%d x %d x %d x %d), kernel (%d x %d), pad (%d x %d), stride (%d x %d), dilation (%d x %d)\n",
name.c_str(), inputs[0].size[0], inputs[0].size[1], inputs[0].size[2], inputs[0].size[3],
kernel.width, kernel.height, pad.width, pad.height,
stride.width, stride.height, dilation.width, dilation.height);
}
else {
printf("conv %s: input (%d x %d x %d), kernel (%d x %d), pad (%d x %d), stride (%d x %d), dilation (%d x %d)\n",
name.c_str(), inputs[0].size[0], inputs[0].size[1], inputs[0].size[2],
kernel.width, kernel.height, pad.width, pad.height,
stride.width, stride.height, dilation.width, dilation.height);
}*/
int inpGroupCn = blobs.empty() ? inputs[1].size[1] : blobs[0].size[1];
CV_Assert_N(inputs.size() >= (size_t)1, inputs[0].size[1] % inpGroupCn == 0,
outputs.size() == 1, inputs[0].data != outputs[0].data);
int ngroups = inputs[0].size[1] / inpGroupCn;
CV_Assert(outputs[0].size[1] % ngroups == 0);
reluslope.clear();
if( activ )
{
Ptr<ReLULayer> activ_relu = activ.dynamicCast<ReLULayer>();
if( !activ_relu.empty() )
{
reluslope.assign(outCn+2, activ_relu->negativeSlope);
}
Ptr<ChannelsPReLULayer> activ_chprelu = activ.dynamicCast<ChannelsPReLULayer>();
if( !activ_chprelu.empty() )
{
const Mat& m = activ_chprelu->blobs[0];
CV_Assert(m.isContinuous() && m.type() == CV_32F && (int)m.total() == outCn);
const float* mdata = m.ptr<float>();
reluslope.resize(outCn+2);
std::copy(mdata, mdata + outCn, reluslope.begin());
reluslope[outCn] = reluslope[outCn+1] = reluslope[outCn-1];
}
}
#ifdef HAVE_TENGINE
bool tengine_ret = false; ;
std::vector<Mat> teng_in, teng_out;
inputs_arr.getMatVector(teng_in);
outputs_arr.getMatVector(teng_out);
int inch = teng_in[0].size[1]; // inch
int in_h = teng_in[0].size[2]; // in_h
int in_w = teng_in[0].size[3]; // in_w
int out_b = teng_out[0].size[0]; // out batch size
int outch = teng_out[0].size[1]; // outch
int out_h = teng_out[0].size[2]; // out_h
int out_w = teng_out[0].size[3]; // out_w
float *input_ = teng_in[0].ptr<float>();
float *output_ = teng_out[0].ptr<float>();
float *kernel_ = weightsMat.ptr<float>();
float *teg_bias = &biasvec[0];
int nstripes = std::max(getNumThreads(), 1);
/* tengine_init will run when first time. */
if(NULL == tengine_graph)
{
tengine_graph = tengine_init(name.c_str(), input_, inch, ngroups, in_h, in_w,
output_, out_b, outch, out_h, out_w,
kernel_, kernel_size.size(), kernel.height, kernel.width,
teg_bias, stride.height, stride.width,
pad.height, pad.width, dilation.height, dilation.width,
weightsMat.step1(), padMode, tengine_graph, nstripes);
/*printf("Init(%s): input=%p(%d %d %d %d ),output=%p(%d %d %d %d ),kernel=%p(%ld %d %d ), bias=%p ,"
"stride(%d %d), pad(%d %d), dilation(%d %d) ,weightsMat=%ld, padMode=%s ,tengine_graph = %p \n",
name.c_str(),input_, inch, ngroups, in_h, in_w,
output_, out_b, outch, out_h, out_w,
kernel_, kernel_size.size(), kernel.height, kernel.width,
teg_bias, stride.height, stride.width,
pad.height, pad.width, dilation.height, dilation.width,
weightsMat.step1(), padMode.c_str() ,tengine_graph);*/
}
if(NULL != tengine_graph)
{
tengine_ret = tengine_forward(tengine_graph);
}
/* activation */
if((true == tengine_ret) && activ )
{
int out_cstep = out_h * out_w; // out_cstep
ActivationLayer* activ_ = activ.get();
activ_->forwardSlice(output_, output_, out_cstep, out_cstep, 0, outch);
}
if(false == tengine_ret)
#endif
{
int nstripes = std::max(getNumThreads(), 1);
ParallelConv::run(inputs[0], outputs[0], weightsMat, biasvec, reluslope,
kernel_size, strides, pads_begin, pads_end, dilations, activ.get(), ngroups, nstripes);
}
}
#ifdef HAVE_CUDA
Ptr<BackendNode> initCUDA(
void *context_,
const std::vector<Ptr<BackendWrapper>>& inputs,
const std::vector<Ptr<BackendWrapper>>& outputs
) override
{
auto context = reinterpret_cast<csl::CSLContext*>(context_);
CV_Assert(inputs.size() == 1 || inputs.size() == 2);
auto input_wrapper = inputs[0].dynamicCast<CUDABackendWrapper>();
auto input_shape = input_wrapper->getShape();
CV_Assert(outputs.size() == 1);
auto output_wrapper = outputs[0].dynamicCast<CUDABackendWrapper>();
auto output_shape = output_wrapper->getShape();
CV_Assert(!blobs.empty());
const auto output_feature_maps = blobs[0].size[0];
const auto input_feature_maps = input_shape[1];
const auto input_feature_maps_per_group = blobs[0].size[1];
const auto groups = input_feature_maps / input_feature_maps_per_group;
ConvolutionConfiguration config;
if (input_shape.size() == 3)
{
// Conv1D
// We add an extra dim for input and output tensors, because CuDNN doesn't support convolution with 3D tensors
input_shape.insert(std::end(input_shape) - 1, 1);
output_shape.insert(std::end(output_shape) - 1, 1);
// Do the similar thing for the other parameters
pads_begin.insert(std::begin(pads_begin), 0);
pads_end.insert(std::begin(pads_end), 0);
strides.insert(std::begin(strides), 1);
dilations.insert(std::begin(dilations), 1);
kernel_size.insert(std::begin(kernel_size), 1);
}
config.kernel_size.assign(std::begin(kernel_size), std::end(kernel_size));
config.dilations.assign(std::begin(dilations), std::end(dilations));
config.strides.assign(std::begin(strides), std::end(strides));
if (padMode.empty())
{
config.padMode = ConvolutionConfiguration::PaddingMode::MANUAL;
config.pads_begin.assign(std::begin(pads_begin), std::end(pads_begin));
config.pads_end.assign(std::begin(pads_end), std::end(pads_end));
}
else if (padMode == "VALID")
{
config.padMode = ConvolutionConfiguration::PaddingMode::VALID;
}
else if (padMode == "SAME")
{
config.padMode = ConvolutionConfiguration::PaddingMode::SAME;
}
else
{
CV_Error(Error::StsNotImplemented, padMode + " padding mode not supported by ConvolutionLayer");
}
config.input_shape.assign(std::begin(input_shape), std::end(input_shape));
config.output_shape.assign(std::begin(output_shape), std::end(output_shape));
config.groups = groups;
config.fusion_mode = cudaFusionMode;
config.activation_type = cudaActType;
config.relu_negative_slope = cuda_relu_slope;
config.crelu_floor = cuda_crelu_floor;
config.crelu_ceil = cuda_crelu_ceil;
config.power_exp = cuda_power_exp;
config.power_scale = cuda_power_scale;
config.power_shift = cuda_power_shift;
Mat filtersMat = fusedWeights ? weightsMat : blobs[0];
Mat biasMat = (hasBias() || fusedBias) ? Mat(output_feature_maps, 1, CV_32F, biasvec.data()) : Mat();
if (countNonZero(biasMat) == 0)
biasMat = Mat();
return make_cuda_node<cuda4dnn::ConvolutionOp>(
preferableTarget, std::move(context->stream), std::move(context->cudnn_handle), config, filtersMat, biasMat);
}
#endif
virtual bool tryQuantize(const std::vector<std::vector<float> > &scales,
const std::vector<std::vector<int> > &zeropoints, LayerParams& params) CV_OVERRIDE
{
// References - https://arxiv.org/pdf/1712.05877.pdf
// Quantized convolution with variable weights is not supported.
if (blobs.empty())
return false;
float inputScale = scales[0][0], outputScale = scales[1][0];
int inputZp = zeropoints[0][0];
params.set("input_zeropoint", inputZp);
Mat weightsQuantized(weightsMat.rows, weightsMat.cols, CV_8S);
Mat biasQuantized(1, numOutput, CV_32S);
Mat outputMultiplier(1, numOutput, CV_32F);
double realMin, realMax, weightsScale;
for( int i = 0; i < numOutput; i++ )
{
// Quantize weights
cv::minMaxIdx(weightsMat.row(i), &realMin, &realMax);
realMin = std::min(realMin, 0.0);
realMax = std::max(realMax, 0.0);
weightsScale = (realMax == realMin) ? 1.0 : std::max(-realMin, realMax)/127;
weightsMat.row(i).convertTo(weightsQuantized.row(i), CV_8S, 1.f/weightsScale);
// Quantize biases
float biasScale = inputScale * weightsScale;
biasQuantized.at<int>(i) = (int)std::round(biasvec[i]/biasScale) - inputZp*(cv::sum(weightsQuantized.row(i))[0]);
// Store multiplier
outputMultiplier.at<float>(i) = biasScale / outputScale;
}
params.blobs.clear();
params.blobs.push_back(weightsQuantized.reshape(1, shape(blobs[0])));
params.blobs.push_back(biasQuantized);
params.blobs.push_back(outputMultiplier);
return true;
}
virtual int64 getFLOPS(const std::vector<MatShape> &inputs,
const std::vector<MatShape> &outputs) const CV_OVERRIDE
{
CV_Assert(inputs.size() == outputs.size() || inputs.size() == outputs.size() + blobs.size());
int64 flops = 0;
int karea = std::accumulate(kernel_size.begin(), kernel_size.end(), 1, std::multiplies<size_t>());
for (int i = 0; i < outputs.size(); i++)
{
flops += total(outputs[i])*(CV_BIG_INT(2)*karea*inputs[i][1] + 1);
}
return flops;
}
};
class DeConvolutionLayerImpl CV_FINAL : public BaseConvolutionLayerImpl
{
public:
Mat weightsMat, biasesMat;
UMat umat_weights;
UMat umat_biases;
DeConvolutionLayerImpl(const LayerParams& params) : BaseConvolutionLayerImpl(params) {}
MatShape computeColRowShape(const MatShape &inpShape, const MatShape &outShape) const CV_OVERRIDE
{
int dims = inpShape.size();
int inpCn = inpShape[1];
int inpD = dims == 5 ? inpShape[2] : 1;
int inpH = inpShape[dims - 2];
int inpW = inpShape.back();
int outCn = outShape[1];
int ngroups = inpCn / blobs[0].size[0];
int outGroupCn = outCn / ngroups;
int ksize = outGroupCn * std::accumulate(kernel_size.begin(), kernel_size.end(),
1, std::multiplies<size_t>());
return shape(ksize, inpD * inpH * inpW);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
if (backendId == DNN_BACKEND_CUDA)
{
/* only deconvolution 2d and 3d supported */
if (kernel_size.size() == 2 || kernel_size.size() == 3)
return true;
return false;
}
#ifdef HAVE_INF_ENGINE
const int outGroupCn = blobs[0].size[1]; // Weights are in IOHW or IODHW layout
const int group = numOutput / outGroupCn;
if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) {
return group == 1;
}
#endif // HAVE_INF_ENGINE
{
return backendId == DNN_BACKEND_CUDA ||
(kernel_size.size() == 2 && (backendId == DNN_BACKEND_OPENCV || backendId == DNN_BACKEND_HALIDE));
}
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE
{
CV_Assert(!hasBias() || blobs[1].total() == (size_t)numOutput);
CV_Assert(inputs.size() != 0);
int outCn = numOutput;
std::vector<int> outShape;
outShape.push_back(inputs[0][0]); // batch
outShape.push_back(outCn);
if (padMode.empty())
{
for (int i = 0; i < kernel_size.size(); i++)
outShape.push_back(strides[i] * (inputs[0][2 + i] - 1) + kernel_size[i] - pads_begin[i] - pads_end[i] + adjust_pads[i]);
}
else if (padMode == "VALID")
{
for (int i = 0; i < kernel_size.size(); i++)
outShape.push_back(strides[i] * (inputs[0][2 + i] - 1) + kernel_size[i] + adjust_pads[i]);
}
else if (padMode == "SAME")
{
for (int i = 0; i < kernel_size.size(); i++)
outShape.push_back(strides[i] * (inputs[0][2 + i] - 1) + 1 + adjust_pads[i]);
}
else
CV_Error(Error::StsError, "Unsupported padding mode " + padMode);
CV_Assert(outCn % blobs[0].size[1] == 0);
int ngroups = outCn / blobs[0].size[1];
int inpCn = inputs[0][1];
CV_Assert(inpCn % ngroups == 0 && outCn % ngroups == 0);
CV_Assert(blobs[0].size[0] == inpCn);
outputs.resize(1, outShape);
if (!is1x1())
internals.push_back(computeColRowShape(inputs[0], outputs[0]));
return false;
}
void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
BaseConvolutionLayerImpl::finalize(inputs_arr, outputs_arr);
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
std::vector<int> inpShape;
std::vector<int> outShape;
for (int i = 2; i < inputs[0].dims; i++) {
inpShape.push_back(inputs[0].size[i]);
outShape.push_back(outputs[0].size[i]);
}
getConvPoolPaddings(outShape, kernel_size, strides, padMode, pads_begin, pads_end);
if (pads_begin.size() == 2) {
for (int i = 0; i < pads_begin.size(); i++) {
if (pads_begin[i] != pads_end[i])
CV_Error(Error::StsNotImplemented, "Unsupported asymmetric padding in deconvolution layer");
}
pad = Size(pads_begin[1], pads_begin[0]);
}
weightsMultipliers.assign(numOutput, 1.0);
if (weightsMat.empty())
{
transpose(blobs[0].reshape(1, blobs[0].size[0]), weightsMat);
biasesMat = hasBias() ? blobs[1].reshape(1, numOutput)
: Mat::zeros(numOutput, 1, CV_32F);
}
}
void fuseWeights(const Mat& w_, const Mat& b_) CV_OVERRIDE
{
Mat w = w_.total() == 1 ? Mat(1, numOutput, CV_32F, Scalar(w_.at<float>(0))) : w_;
Mat b = b_.total() == 1 ? Mat(1, numOutput, CV_32F, Scalar(b_.at<float>(0))) : b_;
CV_Assert_N(!weightsMat.empty(),
w.empty() || numOutput == w.total(),
b.empty() || numOutput == b.total());
if (!w.empty())
{
transpose(blobs[0].reshape(1, blobs[0].size[0]), weightsMat);
weightsMat = weightsMat.reshape(1, numOutput);
for (int i = 0; i < numOutput; ++i)
{
double wi = w.at<float>(i);
weightsMultipliers[i] *= wi;
cv::multiply(weightsMat.row(i), weightsMultipliers[i], weightsMat.row(i));
biasesMat.at<float>(i) *= wi;
}
weightsMat = weightsMat.reshape(1, weightsMat.total() / blobs[0].size[0]);
}
if (!b.empty())
{
cv::add(biasesMat, b.reshape(1, numOutput), biasesMat);
}
}
class MatMulInvoker : public ParallelLoopBody
{
public:
MatMulInvoker(const Mat& a, const Mat& b, Mat& c, int nstripes)
{
a_ = &a;
b_ = &b;
c_ = &c;
nstripes_ = nstripes;
useAVX = checkHardwareSupport(CPU_AVX);
useAVX2 = checkHardwareSupport(CPU_AVX2);
useAVX512 = CV_CPU_HAS_SUPPORT_AVX512_SKX;
useRVV = checkHardwareSupport(CPU_RVV);
}
void operator()(const Range& range_) const CV_OVERRIDE
{
int stripeSize = (int)alignSize((b_->cols + nstripes_ - 1)/nstripes_, 16);
Range range(range_.start*stripeSize, std::min(range_.end*stripeSize, b_->cols));
int mmax = a_->rows;
int nmax = range.end - range.start;
int kmax = a_->cols;
int m, n, k;
const float* aptr = a_->ptr<float>();
const float* bptr = b_->ptr<float>() + range.start;
float* cptr = c_->ptr<float>() + range.start;
size_t astep = a_->step1();
size_t bstep = b_->step1();
size_t cstep = c_->step1();
#if CV_TRY_AVX512_SKX
if( useAVX512 )
opt_AVX512_SKX::fastGEMM( aptr, astep, bptr, bstep, cptr, cstep, mmax, kmax, nmax );
else
#endif
#if CV_TRY_AVX2
if( useAVX2 )
opt_AVX2::fastGEMM( aptr, astep, bptr, bstep, cptr, cstep, mmax, kmax, nmax );
else
#endif
#if CV_TRY_AVX
if( useAVX )
opt_AVX::fastGEMM( aptr, astep, bptr, bstep, cptr, cstep, mmax, kmax, nmax );
else
#endif
#if CV_TRY_RVV
if( useRVV ) {
opt_RVV::fastGEMM( aptr, astep, bptr, bstep, cptr, cstep, mmax, kmax, nmax );
}
else
#endif
for( m = 0; m < mmax; m += 2 )
{
float* dst0 = cptr + cstep*m;
float* dst1 = cptr + cstep*std::min(m+1, mmax-1);
const float* aptr0 = aptr + astep*m;
const float* aptr1 = aptr + astep*std::min(m+1, mmax-1);
for( n = 0; n < nmax; n++ )
{
dst0[n] = 0.f;
dst1[n] = 0.f;
}
for( k = 0; k < kmax; k += 4 )
{
float alpha00 = aptr0[k];
float alpha01 = aptr1[k];
float alpha10 = 0.f, alpha11 = 0.f;
float alpha20 = 0.f, alpha21 = 0.f;
float alpha30 = 0.f, alpha31 = 0.f;
const float* bptr0 = bptr + k*bstep;
const float* bptr1 = bptr0;
const float* bptr2 = bptr0;
const float* bptr3 = bptr0;
if( k+1 < kmax )
{
alpha10 = aptr0[k+1];
alpha11 = aptr1[k+1];
bptr1 = bptr0 + bstep;
if( k+2 < kmax )
{
alpha20 = aptr0[k+2];
alpha21 = aptr1[k+2];
bptr2 = bptr1 + bstep;
if( k+3 < kmax )
{
alpha30 = aptr0[k+3];
alpha31 = aptr1[k+3];
bptr3 = bptr2 + bstep;
}
}
}
n = 0;
#if CV_SIMD128
v_float32x4 a00 = v_setall_f32(alpha00);
v_float32x4 a01 = v_setall_f32(alpha01);
v_float32x4 a10 = v_setall_f32(alpha10);
v_float32x4 a11 = v_setall_f32(alpha11);
v_float32x4 a20 = v_setall_f32(alpha20);
v_float32x4 a21 = v_setall_f32(alpha21);
v_float32x4 a30 = v_setall_f32(alpha30);
v_float32x4 a31 = v_setall_f32(alpha31);
for( ; n <= nmax - 4; n += 4 )
{
v_float32x4 d0 = v_load(dst0 + n);
v_float32x4 d1 = v_load(dst1 + n);
v_float32x4 b0 = v_load(bptr0 + n);
v_float32x4 b1 = v_load(bptr1 + n);
v_float32x4 b2 = v_load(bptr2 + n);
v_float32x4 b3 = v_load(bptr3 + n);
// TODO try to improve pipeline width
d0 = v_fma(b0, a00, d0);
d1 = v_fma(b0, a01, d1);
d0 = v_fma(b1, a10, d0);
d1 = v_fma(b1, a11, d1);
d0 = v_fma(b2, a20, d0);
d1 = v_fma(b2, a21, d1);
d0 = v_fma(b3, a30, d0);
d1 = v_fma(b3, a31, d1);
v_store(dst0 + n, d0);
v_store(dst1 + n, d1);
}
#endif
for( ; n < nmax; n++ )
{
float b0 = bptr0[n];
float b1 = bptr1[n];
float b2 = bptr2[n];
float b3 = bptr3[n];
float d0 = dst0[n] + alpha00*b0 + alpha10*b1 + alpha20*b2 + alpha30*b3;
float d1 = dst1[n] + alpha01*b0 + alpha11*b1 + alpha21*b2 + alpha31*b3;
dst0[n] = d0;
dst1[n] = d1;
}
}
}
}
const Mat *a_, *b_;
Mat* c_;
int nstripes_;
bool useAVX;
bool useAVX2;
bool useAVX512;
bool useRVV;
};
class Col2ImInvoker : public cv::ParallelLoopBody
{
public:
const float* data_col;
const float* biasvec;
int channels, height, width;
int kernel_h, kernel_w;
int pad_h, pad_w;
int stride_h, stride_w;
float* data_im;
int height_col, width_col;
int nstripes;
bool is1x1;
Col2ImInvoker()
: data_col(0), biasvec(0), channels(0), height(0), width(0),
kernel_h(0), kernel_w(0), pad_h(0), pad_w(0), stride_h(0), stride_w(0), data_im(0),
height_col(0), width_col(0), nstripes(0), is1x1(0)
{}
static void run(const float* data_col,
int channels, int height, int width,
int kernel_h, int kernel_w,
int pad_h, int pad_w,
int stride_h, int stride_w,
int height_col, int width_col,
float* data_im,
const float* biasvec,
bool is1x1)
{
const int nstripes = getNumThreads();
Col2ImInvoker t;
t.data_col = data_col;
t.data_im = data_im;
t.channels = channels; t.height = height; t.width = width;
t.kernel_h = kernel_h; t.kernel_w = kernel_w;
t.pad_h = pad_h; t.pad_w = pad_w;
t.stride_h = stride_h; t.stride_w = stride_w;
t.height_col = height_col;
t.width_col = width_col;
t.nstripes = nstripes;
t.is1x1 = is1x1;
t.biasvec = biasvec;
parallel_for_(Range(0, nstripes), t, nstripes);
}
virtual void operator ()(const Range &r) const CV_OVERRIDE
{
const float* data_col_ = data_col;
float* data_im_ = data_im;
int coeff_h = (1 - stride_h * kernel_w * height_col) * width_col;
int coeff_w = (1 - stride_w * height_col * width_col);
size_t total = (size_t)channels * height * width;
size_t stripeSize = (total + nstripes - 1)/nstripes;
size_t startIndex = r.start*stripeSize;
size_t endIndex = std::min(r.end*stripeSize, total);
int w = (int)(startIndex % width + pad_w);
int h = (int)((startIndex / width) % height + pad_h);
int c = (int)(startIndex / (width * height));
int h_col_start = (h < kernel_h) ? 0 : (h - kernel_h) / stride_h + 1;
int h_col_end = std::min(h / stride_h + 1, height_col);
int plane_size_col = height_col * width_col;
int offset = (c * kernel_h * kernel_w + h * kernel_w + w) * plane_size_col;
bool is1x1_ = is1x1;
const float* biasvec_ = biasvec;
for (size_t index = startIndex; index < endIndex; index++)
{
// compute the start and end of the output
int w_col_start = (w < kernel_w) ? 0 : (w - kernel_w) / stride_w + 1;
int w_col_end = std::min(w / stride_w + 1, width_col);
float val;
if( is1x1_ )
val = data_im_[index];
else
{
val = 0.f;
for (int h_col = h_col_start; h_col < h_col_end; ++h_col) {
for (int w_col = w_col_start; w_col < w_col_end; ++w_col) {
val += data_col_[offset + h_col * coeff_h + w_col * coeff_w];
}
}
}
data_im_[index] = val + biasvec_[c];
offset += plane_size_col;
if( ++w >= width + pad_w )
{
w = (int)((index + 1)% width + pad_w);
h = (int)(((index + 1) / width) % height + pad_h);
c = (int)((index + 1) / (width * height));
h_col_start = (h < kernel_h) ? 0 : (h - kernel_h) / stride_h + 1;
h_col_end = std::min(h / stride_h + 1, height_col);
offset = (c * kernel_h * kernel_w + h * kernel_w + w) * plane_size_col;
}
}
}
};
#ifdef HAVE_OPENCL
bool forward_ocl(InputArrayOfArrays inputs_, OutputArrayOfArrays outputs_, OutputArrayOfArrays internals_)
{
std::vector<UMat> inputs;
std::vector<UMat> outputs;
std::vector<UMat> internals;
if (inputs_.depth() == CV_16S)
return false;
inputs_.getUMatVector(inputs);
outputs_.getUMatVector(outputs);
internals_.getUMatVector(internals);
int outCn = numOutput;
int inpCn = inputs[0].size[1];
if (is1x1())
return false;
if (umat_weights.empty())
{
if (fusedWeights)
weightsMat.copyTo(umat_weights);
else
transpose(blobs[0].reshape(1, inpCn), umat_weights);
if (fusedBias)
biasesMat.copyTo(umat_biases);
else
{
if (hasBias())
blobs[1].reshape(1, outCn).copyTo(umat_biases);
else
umat_biases = UMat::zeros(outCn, 1, CV_32F);
}
}
String buildopt = format("-DT=%s ", ocl::typeToStr(inputs[0].type()));
buildopt += format("-DPAD_H=%d -DPAD_W=%d -DKERNEL_H=%d -DKERNEL_W=%d -DSTRIDE_H=%d -DSTRIDE_W=%d ",
pad.height, pad.width, kernel.height, kernel.width, stride.height, stride.width);
for (size_t ii = 0; ii < outputs.size(); ii++)
{
int ngroups = outCn / blobs[0].size[1];
int inpGroupCn = inpCn / ngroups;
int outGroupCn = blobs[0].size[1];
const UMat& inp = inputs[ii];
UMat& out = outputs[ii];
int numImg = inp.size[0];
int inpH = inp.size[2], inpW = inp.size[3];
int outH = out.size[2], outW = out.size[3];
MatShape inpshape = shape(numImg*inpCn, inpH*inpW);
MatShape outshape = shape(numImg*outCn, outH*outW);
UMat convBlob = inputs[ii].reshape(1, inpshape.size(), &inpshape[0]);
UMat decnBlob = out.reshape(1, outshape.size(), &outshape[0]);
int rows = internals[0].rows / ngroups;
for (int n = 0; n < numImg; n++)
{
for (int g = 0; g < ngroups; g++)
{
UMat colMat = internals[0].rowRange(_Range(g * rows, rows));
UMat convMat = convBlob.rowRange(_Range((g + n * ngroups) * inpGroupCn, inpGroupCn));
UMat wghtMat = umat_weights.colRange(_Range(g * inpGroupCn, inpGroupCn));
gemm(wghtMat, convMat, 1, noArray(), 0, colMat, 0);
}
for (int g = 0; g < ngroups; g++)
{
int total = outGroupCn * decnBlob.cols;
int index = 0;
int height_col = inpH;
int width_col = inpW;
int coeff_h = (1 - stride.height * kernel.width * height_col) * width_col;
int coeff_w = (1 - stride.width * height_col * width_col);
ocl::Kernel k("col2im", ocl::dnn::col2im_oclsrc, buildopt);
k.set(index++, total);
k.set(index++, ocl::KernelArg::PtrReadOnly(internals[0]));
k.set(index++, (int)(g * rows * internals[0].cols));
k.set(index++, outGroupCn);
k.set(index++, outH);
k.set(index++, outW);
k.set(index++, height_col);
k.set(index++, width_col);
k.set(index++, coeff_h);
k.set(index++, coeff_w);
k.set(index++, ocl::KernelArg::PtrReadOnly(umat_biases));
k.set(index++, (int)(g * outGroupCn * umat_biases.cols));
k.set(index++, ocl::KernelArg::PtrWriteOnly(decnBlob));
k.set(index++, (int)((g + n * ngroups) * outGroupCn * decnBlob.cols));
size_t global[] = { (size_t)total };
bool ret = k.run(1, global, NULL, false);
if (!ret)
return false;
}
}
}
return true;
}
#endif
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr));
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs, internals;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
internals_arr.getMatVector(internals);
int outCn = numOutput;
int inpCn = inputs[0].size[1];
bool is1x1flag = is1x1();
int nstripes = getNumThreads();
if( weightsMat.empty() )
{
transpose(blobs[0].reshape(1, inpCn), weightsMat);
biasesMat = hasBias() ? blobs[1].reshape(1, outCn) : Mat::zeros(outCn, 1, CV_32F);
}
for (size_t ii = 0; ii < outputs.size(); ii++)
{
int ngroups = outCn / blobs[0].size[1];
int inpGroupCn = inpCn / ngroups;
int outGroupCn = blobs[0].size[1];
const Mat& inp = inputs[ii];
Mat& out = outputs[ii];
int numImg = inp.size[0];
int inpH = inp.size[2], inpW = inp.size[3];
int outH = out.size[2], outW = out.size[3];
Mat convBlob = inputs[ii].reshape(1, numImg*inpCn);
Mat decnBlob = out.reshape(1, numImg*outCn);
for (int n = 0; n < numImg; n++)
{
for (int g = 0; g < ngroups; g++)
{
Mat dstMat = decnBlob.rowRange(_Range((g + n * ngroups) * outGroupCn, outGroupCn));
Mat &colMat = is1x1flag ? dstMat : internals[0];
Mat convMat = convBlob.rowRange(_Range((g + n * ngroups) * inpGroupCn, inpGroupCn));
Mat wghtMat = weightsMat.colRange(_Range(g * inpGroupCn, inpGroupCn));
Mat curBiasMat = biasesMat.rowRange(_Range(g * outGroupCn, outGroupCn));
//gemm(wghtMat, convMat, 1, colMat, 0, colMat, 0);
MatMulInvoker mminvoker(wghtMat, convMat, colMat, nstripes);
parallel_for_(Range(0, nstripes), mminvoker, nstripes);
Col2ImInvoker::run(colMat.ptr<float>(), outGroupCn, outH, outW,
kernel.height, kernel.width, pad.height, pad.width,
stride.height, stride.width, inpH, inpW, dstMat.ptr<float>(),
curBiasMat.ptr<float>(), is1x1flag);
}
}
}
}
#ifdef HAVE_CUDA
Ptr<BackendNode> initCUDA(
void *context_,
const std::vector<Ptr<BackendWrapper>>& inputs,
const std::vector<Ptr<BackendWrapper>>& outputs
) override
{
CV_Assert(!blobs.empty());
auto context = reinterpret_cast<csl::CSLContext*>(context_);
CV_Assert(inputs.size() == 1);
auto input_wrapper = inputs[0].dynamicCast<CUDABackendWrapper>();
auto input_shape = input_wrapper->getShape();
CV_Assert(outputs.size() == 1);
auto output_wrapper = outputs[0].dynamicCast<CUDABackendWrapper>();
auto output_shape = output_wrapper->getShape();
const auto output_feature_maps = numOutput;
const auto output_feature_maps_per_group = blobs[0].size[1];
const auto groups = output_feature_maps / output_feature_maps_per_group;
TransposeConvolutionConfiguration config;
config.kernel_size.assign(std::begin(kernel_size), std::end(kernel_size));
config.dilations.assign(std::begin(dilations), std::end(dilations));
config.strides.assign(std::begin(strides), std::end(strides));
if (padMode.empty())
{
config.padMode = TransposeConvolutionConfiguration::PaddingMode::MANUAL;
config.pads_begin.assign(std::begin(pads_begin), std::end(pads_begin));
config.pads_end.assign(std::begin(pads_end), std::end(pads_end));
}
else if (padMode == "VALID")
{
config.padMode = TransposeConvolutionConfiguration::PaddingMode::VALID;
}
else if (padMode == "SAME")
{
config.padMode = TransposeConvolutionConfiguration::PaddingMode::SAME;
}
else
{
CV_Error(Error::StsNotImplemented, padMode + " padding mode not supported by DeconvolutionLayer");
}
config.input_shape.assign(std::begin(input_shape), std::end(input_shape));
config.output_shape.assign(std::begin(output_shape), std::end(output_shape));
config.groups = groups;
CV_Assert(blobs.size() >= 1);
Mat filtersMat = fusedWeights ? weightsMat.t() : blobs[0];
Mat biasMat = (hasBias() || fusedBias) ? biasesMat : Mat();
if (countNonZero(biasMat) == 0)
biasMat = Mat();
return make_cuda_node<cuda4dnn::TransposeConvolutionOp>(
preferableTarget, std::move(context->stream), std::move(context->cudnn_handle), config, filtersMat, biasMat);
}
#endif
virtual Ptr<BackendNode> initHalide(const std::vector<Ptr<BackendWrapper> > &inputs) CV_OVERRIDE
{
#ifdef HAVE_HALIDE
CV_Assert(!blobs.empty());
Halide::Buffer<float> inputBuffer = halideBuffer(inputs[0]);
int inW, inH, inC, inN;
getCanonicalSize(inputBuffer, &inW, &inH, &inC, &inN);
const int outGroupCn = blobs[0].size[1];
const int group = numOutput / outGroupCn;
const int inpGroupCn = blobs[0].size[0] / group;
Halide::Var x("x"), y("y"), c("c"), n("n");
Halide::Func top = (name.empty() ? Halide::Func() : Halide::Func(name));
Halide::Func padded_input(name + "_constant_exterior");
auto weights = wrapToHalideBuffer(blobs[0]);
Halide::Func dilated_input("dilated_input");
dilated_input(x, y, c, n) = 0.0f;
Halide::RDom r1(0, inW, 0, inH);
dilated_input(r1.x * stride.width, r1.y * stride.height, c, n) =
inputBuffer(r1.x, r1.y, c, n);
dilated_input.compute_root();
Halide::Func bounded =
Halide::BoundaryConditions::constant_exterior(dilated_input, 0,
0, (inW - 1) * stride.width + 1,
0, (inH - 1) * stride.height + 1,
0, inC, 0, inN);
padded_input(x, y, c, n) = bounded(x, y, c, n);
Halide::RDom r(0, kernel.width, 0, kernel.height, 0, inpGroupCn);
Halide::Expr kx = x + pad.width - r.x;
Halide::Expr ky = y + pad.height - r.y;
Halide::Expr kInC = r.z;
Halide::Expr kOutC = c;
for (int i = 1; i < group; ++i)
{
kInC = select(c < outGroupCn * i, kInC, inpGroupCn * i + r.z);
kOutC = select(c < outGroupCn * i, kOutC, c - outGroupCn * i);
}
Halide::Expr topExpr = sum(padded_input(kx, ky, kInC, n) *
weights(r.x, r.y, kOutC, kInC));
if (hasBias())
{
auto bias = wrapToHalideBuffer(blobs[1], {numOutput});
topExpr += bias(c);
}
top(x, y, c, n) = topExpr;
return Ptr<BackendNode>(new HalideBackendNode({ padded_input, top }));
#endif // HAVE_HALIDE
return Ptr<BackendNode>();
}
#ifdef HAVE_DNN_NGRAPH
virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper> > &inputs,
const std::vector<Ptr<BackendNode> >& nodes) CV_OVERRIDE
{
CV_Assert(!blobs.empty());
const int outGroupCn = blobs[0].size[1];
const int group = numOutput / outGroupCn;
CV_Assert(group == 1);
auto& ieInpNode = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
std::vector<size_t> kernel_shape = getShape<size_t>(blobs[0]);
auto ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, kernel_shape, blobs[0].data);
if (fusedWeights)
{
Mat newWeights;
transpose(weightsMat, newWeights);
ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, kernel_shape, newWeights.data);
}
std::vector<size_t> paddings_end;
if (padMode == "SAME")
{
for (int i = 0; i < pads_begin.size(); i++) {
paddings_end.push_back(kernel_size[i] - pads_begin[i] - 1 - adjust_pads[i]);
}
adjust_pads = std::vector<size_t>(pads_begin.size(), 0);
} else {
paddings_end = pads_end;
}
ngraph::op::PadType pad_type = padMode == "VALID" ? ngraph::op::PadType::VALID : ngraph::op::PadType::EXPLICIT;
auto deconv = std::make_shared<ngraph::op::v1::ConvolutionBackpropData>(
ieInpNode,
ieWeights,
ngraph::Strides(strides),
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(pads_begin.begin(), pads_begin.end())),
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(paddings_end.begin(), paddings_end.end())),
ngraph::Strides(dilations),
pad_type,
ngraph::CoordinateDiff(std::vector<std::ptrdiff_t>(adjust_pads.begin(), adjust_pads.end())));
if (hasBias() || fusedBias)
{
std::vector<size_t> shape(deconv->get_shape().size(), 1);
shape[1] = numOutput;
auto bias = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, ngraph::Shape(shape), blobs[1].data);
auto deconv_bias = std::make_shared<ngraph::op::v1::Add>(deconv, bias, ngraph::op::AutoBroadcastType::NUMPY);
return Ptr<BackendNode>(new InfEngineNgraphNode(deconv_bias));
}
return Ptr<BackendNode>(new InfEngineNgraphNode(deconv));
}
#endif // HAVE_DNN_NGRAPH
virtual int64 getFLOPS(const std::vector<MatShape> &inputs,
const std::vector<MatShape> &outputs) const CV_OVERRIDE
{
CV_Assert(inputs.size() == outputs.size());
float flops = 0;
int outChannels = blobs[0].size[0];
size_t karea = std::accumulate(kernel_size.begin(), kernel_size.end(),
1, std::multiplies<size_t>());
for (int i = 0; i < inputs.size(); i++)
{
flops += CV_BIG_INT(2)*outChannels*karea*total(inputs[i]);
}
return flops;
}
};
Ptr<BaseConvolutionLayer> ConvolutionLayer::create(const LayerParams ¶ms)
{
Ptr<ConvolutionLayerImpl> l(new ConvolutionLayerImpl(params));
return l;
}
Ptr<BaseConvolutionLayer> DeconvolutionLayer::create(const LayerParams ¶ms)
{
return Ptr<BaseConvolutionLayer>(new DeConvolutionLayerImpl(params));
}
}
}