mirror of https://github.com/alibaba/MNN.git
184 lines
8.6 KiB
C++
184 lines
8.6 KiB
C++
//
|
|
// DeconvExecution.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on 2019/02/28.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include "backend/opencl/execution/image/DeconvExecution.hpp"
|
|
#include "core/ConvolutionCommon.hpp"
|
|
|
|
namespace MNN {
|
|
namespace OpenCL {
|
|
|
|
DeconvExecution::DeconvExecution(const std::vector<Tensor *> &inputs, const MNN::Op *op, Backend *backend)
|
|
: ConvCommonExecution(op->main_as_Convolution2D(), backend), CommonExecution(backend, op) {
|
|
mOpenCLBackend = static_cast<OpenCLBackend *>(backend);
|
|
const auto *conv2dParams = op->main_as_Convolution2D();
|
|
const auto *conv2dCommonParams = conv2dParams->common();
|
|
mResource->mConv2dCommonParams = conv2dCommonParams;
|
|
mResource->mStrides = {conv2dCommonParams->strideY(), conv2dCommonParams->strideX()};
|
|
mResource->mDilations = {conv2dCommonParams->dilateY(), conv2dCommonParams->dilateX()};
|
|
int kernelWidth = conv2dCommonParams->kernelX();
|
|
int kernelHeight = conv2dCommonParams->kernelY();
|
|
|
|
int outputChannel = conv2dCommonParams->outputCount();
|
|
|
|
const float* filterDataPtr = nullptr;
|
|
int weightSize = 0;
|
|
std::shared_ptr<ConvolutionCommon::Int8Common> quanCommon;
|
|
ConvolutionCommon::getConvParameters(&quanCommon, backend, op, &filterDataPtr, &weightSize);
|
|
|
|
int inputChannel = weightSize / (kernelWidth * kernelHeight * outputChannel);
|
|
std::vector<int> filterShape{outputChannel, inputChannel, kernelHeight, kernelWidth};
|
|
std::vector<int> filterImageShape{(int)inputChannel, (int)UP_DIV(outputChannel, 4) * kernelWidth * kernelHeight};
|
|
std::vector<float> filterDataPtrTransformed;
|
|
filterDataPtrTransformed.resize(weightSize);
|
|
IOHW2OIHW<float, int>(filterDataPtr, filterDataPtrTransformed.data(), outputChannel, inputChannel, kernelHeight,
|
|
kernelWidth);
|
|
|
|
std::shared_ptr<Tensor> filterBuffer(
|
|
Tensor::createDevice<float>({outputChannel, inputChannel, kernelHeight, kernelWidth}));
|
|
|
|
size_t buffer_size = filterBuffer->elementSize() * sizeof(float);
|
|
cl::Buffer filterBufferCL(mOpenCLBackend->getOpenCLRuntime()->context(), CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR, buffer_size);
|
|
filterBuffer->buffer().device = (uint64_t)(&filterBufferCL);
|
|
cl_int error;
|
|
auto ptrCL = mOpenCLBackend->getOpenCLRuntime()->commandQueue().enqueueMapBuffer(filterBufferCL, true, CL_MAP_WRITE, 0, buffer_size, nullptr, nullptr, &error);
|
|
if(ptrCL != nullptr && error == CL_SUCCESS){
|
|
::memcpy(ptrCL, filterDataPtrTransformed.data(), filterBuffer->size());
|
|
}else{
|
|
MNN_ERROR("Map error ptrCL == nullptr \n");
|
|
}
|
|
mOpenCLBackend->getOpenCLRuntime()->commandQueue().enqueueUnmapMemObject(filterBufferCL, ptrCL);
|
|
|
|
mResource->mFilter.reset(Tensor::createDevice<float>({1, filterImageShape[1], 1, 4 * filterImageShape[0]}));
|
|
mOpenCLBackend->onAcquireBuffer(mResource->mFilter.get(), Backend::STATIC);
|
|
MNN::OpenCL::ImageBufferConvertor imageBufferConvertor{mOpenCLBackend->getOpenCLRuntime()};
|
|
|
|
std::string buildOption = "-DBUFFER_INP_FP32";
|
|
imageBufferConvertor.convertBufferToImage(filterBuffer.get(), MNN::OpenCL::CONV2D_FILTER, mResource->mFilter.get(), mOpenCLBackend->getPrecision(), false, buildOption);
|
|
|
|
mResource->mBuildOptions.emplace("-DBIAS");
|
|
if (conv2dCommonParams->relu() == true) {
|
|
mResource->mBuildOptions.emplace("-DRELU");
|
|
} else if (conv2dCommonParams->relu6() == true) {
|
|
mResource->mBuildOptions.emplace("-DRELU6");
|
|
}
|
|
}
|
|
|
|
DeconvExecution::~DeconvExecution() {
|
|
// Do nothing
|
|
}
|
|
|
|
DeconvExecution::DeconvExecution(std::shared_ptr<ConvResource> resource, const MNN::Op* op, Backend *backend)
|
|
: ConvCommonExecution(backend), CommonExecution(backend, op) {
|
|
mResource = resource;
|
|
const auto *conv2dParams = op->main_as_Convolution2D();
|
|
const auto *conv2dCommonParams = conv2dParams->common();
|
|
mResource->mConv2dParams = conv2dParams;
|
|
mResource->mConv2dCommonParams = conv2dCommonParams;
|
|
}
|
|
|
|
bool DeconvExecution::onClone(Backend* bn, const Op* op, Execution** dst) {
|
|
if (!mValid) {
|
|
return false;
|
|
}
|
|
if (nullptr == dst) {
|
|
return true;
|
|
}
|
|
*dst = new DeconvExecution(mResource, op, bn);
|
|
return true;
|
|
}
|
|
|
|
ErrorCode DeconvExecution::onEncode(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
|
|
mUnits.resize(1);
|
|
auto &unit = mUnits[0];
|
|
auto output = outputs[0];
|
|
auto input = inputs[0];
|
|
|
|
std::vector<int> inputShape = tensorShapeFormat(input);
|
|
std::vector<int> outputShape = tensorShapeFormat(output);
|
|
|
|
const int outputBatch = outputShape.at(0);
|
|
const int outputHeight = outputShape.at(1);
|
|
const int outputWidth = outputShape.at(2);
|
|
const int outputChannels = outputShape.at(3);
|
|
|
|
const int inputChannels = inputShape.at(3);
|
|
|
|
const int outputChannelBlocks = UP_DIV(outputChannels, 4);
|
|
const int strideHeight = mResource->mStrides[0];
|
|
const int strideWidth = mResource->mStrides[1];
|
|
|
|
auto pad = ConvolutionCommon::convolutionTransposePad(input, output, mResource->mConv2dCommonParams);
|
|
const int paddingHeight = pad.second;
|
|
const int paddingWidth = pad.first;
|
|
|
|
auto ky = mResource->mConv2dCommonParams->kernelY();
|
|
auto kx = mResource->mConv2dCommonParams->kernelX();
|
|
auto kernelSize = kx * ky;
|
|
|
|
const int transPadH = ky - 1 - pad.second;
|
|
const int transPadW = kx - 1 - pad.first;
|
|
|
|
const int alignHeight = mResource->mStrides[0] - 1 - transPadH;
|
|
const int alignWidth = mResource->mStrides[1] - 1 - transPadW;
|
|
|
|
auto runtime = mOpenCLBackend->getOpenCLRuntime();
|
|
unit.kernel = runtime->buildKernel("deconv_2d", "deconv_2d", mResource->mBuildOptions, mOpenCLBackend->getPrecision());
|
|
auto maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(unit.kernel));
|
|
mGWS = {static_cast<uint32_t>(outputChannelBlocks), static_cast<uint32_t>(outputWidth),
|
|
static_cast<uint32_t>(outputHeight * outputBatch)};
|
|
|
|
int inputImageShape[2] = {inputShape.at(1), inputShape.at(2)};
|
|
int outputImageShape[2] = {outputHeight, outputWidth};
|
|
int strideShape[2] = {strideHeight, strideWidth};
|
|
int paddingShape[2] = {transPadH, transPadW};
|
|
int alignShape[2] = {alignHeight, alignWidth};
|
|
int kernelShape[2] = {ky, kx};
|
|
|
|
uint32_t idx = 0;
|
|
unit.kernel->get().setArg(idx++, mGWS[0]);
|
|
unit.kernel->get().setArg(idx++, mGWS[1]);
|
|
unit.kernel->get().setArg(idx++, mGWS[2]);
|
|
unit.kernel->get().setArg(idx++, openCLImage(input));
|
|
unit.kernel->get().setArg(idx++, openCLImage(mResource->mFilter.get()));
|
|
unit.kernel->get().setArg(idx++, openCLImage(mResource->mBias.get()));
|
|
unit.kernel->get().setArg(idx++, openCLImage(output));
|
|
unit.kernel->get().setArg(idx++, sizeof(inputImageShape), inputImageShape);
|
|
unit.kernel->get().setArg(idx++, sizeof(outputImageShape), outputImageShape);
|
|
unit.kernel->get().setArg(idx++, sizeof(strideShape), strideShape);
|
|
unit.kernel->get().setArg(idx++, sizeof(alignShape), alignShape);
|
|
unit.kernel->get().setArg(idx++, sizeof(paddingShape), paddingShape);
|
|
unit.kernel->get().setArg(idx++, sizeof(kernelShape), kernelShape);
|
|
unit.kernel->get().setArg(idx++, static_cast<int32_t>(kernelSize));
|
|
unit.kernel->get().setArg(idx++, static_cast<int32_t>(UP_DIV(inputChannels, 4)));
|
|
unit.kernel->get().setArg(idx++, static_cast<int32_t>(outputChannelBlocks));
|
|
|
|
std::string name = "deconv2d";
|
|
std::string info = std::to_string(inputChannels) + "_" + std::to_string(outputChannels) + "_" + std::to_string(ky) + "_" + std::to_string(kx) + "_" + std::to_string(strideHeight) + "_" + std::to_string(strideWidth);
|
|
mLWS = localWS3DDefault(mGWS, maxWorkGroupSize, mOpenCLBackend->getOpenCLRuntime(), name + info, unit.kernel, mOpenCLBackend->getCLTuneLevel(), "deconv_2d").first;
|
|
mOpenCLBackend->recordKernel3d(unit.kernel, mGWS, mLWS);
|
|
unit.globalWorkSize = {mGWS[0], mGWS[1], mGWS[2]};
|
|
unit.localWorkSize = {mLWS[0], mLWS[1], mLWS[2]};
|
|
return NO_ERROR;
|
|
}
|
|
|
|
class DeconvolutionCreator : public OpenCLBackend::Creator {
|
|
public:
|
|
virtual ~DeconvolutionCreator() = default;
|
|
virtual Execution *onCreate(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs,
|
|
const MNN::Op *op, Backend *backend) const override {
|
|
if(inputs.size() != 1){
|
|
return nullptr;
|
|
}
|
|
return new DeconvExecution(inputs, op, backend);
|
|
}
|
|
};
|
|
|
|
REGISTER_OPENCL_OP_CREATOR(DeconvolutionCreator, OpType_Deconvolution, IMAGE);
|
|
} // namespace OpenCL
|
|
} // namespace MNN
|