MNN/source/backend/cpu/compute/ConvolutionFloatFactory.cpp

//
//  ConvolutionFloatFactory.cpp
//  MNN
//
//  Created by MNN on 2018/07/16.
//  Copyright © 2018, Alibaba Group Holding Limited
//

#include "ConvolutionFloatFactory.h"
#include "CPUConvolutionDepthwise.hpp"
#include "ConvOpt.h"
#include "Convolution1x1Strassen.hpp"
#include "Convolution3x3.hpp"
#include "ConvolutionGroup.hpp"
#include "ConvolutionIntFactory.hpp"
#include "ConvolutionTiledExecutor.hpp"
#include "ConvolutionWinograd.hpp"
#include "Macro.h"
namespace MNN {

static Execution* _createUnit(const Tensor* input, const Tensor* output, Backend* backend,
                              const Convolution2DCommon* common, const float* originWeight, size_t originWeightSize,
                              const float* bias, size_t biasSize) {
    auto layer   = common;
    bool fastWay = layer->kernelY() == 1 && layer->kernelX() == 1;
    if (fastWay) {
        return new Convolution1x1Strassen(common, backend, originWeight, originWeightSize, bias, biasSize);
    }
    if (!ConvolutionWinograd::canUseWinograd(common)) {
        return new ConvolutionTiledExecutor(common, backend, originWeight, originWeightSize, bias, biasSize);
    }
    auto cpuBackend = (CPUBackend*)backend;
    if (cpuBackend->memoryMode() == BackendConfig::Memory_Low) {
        return new ConvolutionTiledExecutor(common, backend, originWeight, originWeightSize, bias, biasSize);
    }
    auto unit = ConvolutionWinograd::bestWinogradUnit(common, input, output, cpuBackend->threadNumber());
    if (unit <= 1) {
        return new ConvolutionTiledExecutor(common, backend, originWeight, originWeightSize, bias, biasSize);
    }
    // MNN_PRINT("ic=%d, channel=%d, kx=%d, unit=%d\n", input->channel(), output->channel(), common->kernelX(), unit);
    if (common->kernelY() == 3 && common->kernelX() == 3 && unit <= 4) {
        return new Convolution3x3(common, backend, originWeight, originWeightSize, bias, biasSize);
    }
    return new ConvolutionWinograd(common, input, output, backend, originWeight, originWeightSize, bias, biasSize,
                                   unit);
}

Execution* ConvolutionFloatFactory::create(const std::vector<Tensor*>& inputs, const std::vector<Tensor*>& outputs,
                                           const MNN::Op* op, Backend* backend) {
    auto conv2d               = op->main_as_Convolution2D();
    const float* originWeight = nullptr;
    size_t originWeightSize   = 0;
    std::shared_ptr<ConvolutionIntFactory::Int8Common> quanCommon;
    if (nullptr != conv2d->quanParameter()) {
        quanCommon = ConvolutionIntFactory::load(conv2d->quanParameter());
        if (nullptr == quanCommon) {
            MNN_ERROR("Memory not Enough, can't extract IDST Convolution: %s \n", op->name()->c_str());
            return nullptr;
        }
        if (quanCommon->weightFloat.get() == nullptr) {
            return ConvolutionIntFactory::create(inputs[0], outputs[0], op, backend, quanCommon.get());
        }
        // Back to float
        originWeight     = quanCommon->weightFloat.get();
        originWeightSize = quanCommon->weightFloat.size();
    }
    auto common = conv2d->common();
    if (nullptr == originWeight) {
        originWeight     = op->main_as_Convolution2D()->weight()->data();
        originWeightSize = op->main_as_Convolution2D()->weight()->size();
    }

    if (1 == common->group()) {
        return _createUnit(inputs[0], outputs[0], backend, common, originWeight, originWeightSize,
                           conv2d->bias()->data(), conv2d->bias()->size());
    }
    // Split
    std::vector<std::shared_ptr<Execution>> subConvolution;
    auto group            = common->group();
    auto groupOutputCount = common->outputCount() / group;
    auto groupWeightSize  = originWeightSize / group;
    std::shared_ptr<Tensor> emptyInput(Tensor::createDevice<float>(inputs[0]->shape(), Tensor::CAFFE));
    std::shared_ptr<Tensor> emptyOutput(Tensor::createDevice<float>(outputs[0]->shape(), Tensor::CAFFE));
    emptyInput->setLength(1, inputs[0]->channel() / group);
    emptyOutput->setLength(1, outputs[0]->channel() / group);
    for (int i = 0; i < group; ++i) {
        auto newConvolution =
            _createUnit(emptyInput.get(), emptyOutput.get(), backend, common, originWeight + groupWeightSize * i,
                        groupWeightSize, conv2d->bias()->data() + groupOutputCount * i, groupOutputCount);
        subConvolution.push_back(std::shared_ptr<Execution>(newConvolution));
    }
    return new ConvolutionGroup(backend, subConvolution);
}

} // namespace MNN