MNN/source/backend/opencl/execution/ReluExecution.cpp

106 lines
4.6 KiB
C++

//
// ReluExecution.cpp
// MNN
//
// Created by MNN on 2019/02/28.
// Copyright © 2018, Alibaba Group Holding Limited
//
#include "ReluExecution.hpp"
#include "TensorUtils.hpp"
#include "UnaryExecution.hpp"
namespace MNN {
namespace OpenCL {
ReluExecution::ReluExecution(const std::vector<Tensor *> &inputs, const MNN::Op *op, Backend *backend)
: CommonExecution(backend) {
auto mOpenCLBackend = static_cast<OpenCLBackend *>(backend);
auto mPreluParamPtr = op->main_as_PRelu();
int preluSize = mPreluParamPtr->slopeCount();
const float *preluDataPtr = mPreluParamPtr->slope()->data();
auto preluSizeAlign = UP_DIV(preluSize, 4) * 4;
cl::Buffer preluBuffer(mOpenCLBackend->getOpenCLRuntime()->context(), CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR,
preluSizeAlign * sizeof(float));
auto preluDataPtrCL = mOpenCLBackend->getOpenCLRuntime()->commandQueue().enqueueMapBuffer(
preluBuffer, true, CL_MAP_WRITE, 0, preluSizeAlign * sizeof(float));
if(preluDataPtrCL != nullptr){
::memset(preluDataPtrCL, 0, sizeof(float) * preluSizeAlign);
::memcpy(preluDataPtrCL, preluDataPtr, preluSize * sizeof(float));
}else{
MNN_ERROR("Map error preluDataPtrCL == nullptr \n");
}
mOpenCLBackend->getOpenCLRuntime()->commandQueue().enqueueUnmapMemObject(preluBuffer, preluDataPtrCL);
mPreluParam.reset(Tensor::createDevice<float>({1, 1, 1, preluSize}));
mOpenCLBackend->onAcquireBuffer(mPreluParam.get(), Backend::STATIC);
copyBufferToImage(mOpenCLBackend->getOpenCLRuntime(), preluBuffer, openCLImage(mPreluParam.get()),
UP_DIV(preluSize, 4), 1);
}
ReluExecution::~ReluExecution() {
backend()->onReleaseBuffer(mPreluParam.get(), Backend::STATIC);
}
ErrorCode ReluExecution::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
mUnits.resize(1);
auto nhwc = tensorShapeFormat(outputs[0]);
int nhwcArray[] = {nhwc[0], nhwc[1], nhwc[2], UP_DIV(nhwc[3], 4)};
auto imageWidth = nhwc[2] * UP_DIV(nhwc[3], 4);
auto imageHeight = nhwc[0] * nhwc[1];
int reluImageWH[] = {1, 1};
int reluStride[] = {0, 0, 0, 1};
cl::NDRange localSize = {16, 16};
cl::NDRange globalSize = {(uint32_t)UP_DIV(imageWidth, 16) * 16, (uint32_t)UP_DIV(imageHeight, 16) * 16};
auto runTime = ((OpenCLBackend *)backend())->getOpenCLRuntime();
mUnits[0].kernel = runTime->buildKernel("binary", "binary", {"-DOPERATOR=select(in0*in1,in0,in0>=(FLOAT4)0)"});
mUnits[0].kernel.setArg(0, openCLImage(inputs[0]));
mUnits[0].kernel.setArg(1, openCLImage(mPreluParam.get()));
mUnits[0].kernel.setArg(2, openCLImage(outputs[0]));
mUnits[0].kernel.setArg(3, nhwcArray);
mUnits[0].kernel.setArg(4, reluImageWH);
mUnits[0].kernel.setArg(5, reluStride);
mUnits[0].globalWorkSize = globalSize;
mUnits[0].localWorkSize = localSize;
return NO_ERROR;
}
class ReluCreator : public OpenCLBackend::Creator {
public:
virtual Execution *onCreate(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs,
const MNN::Op *op, Backend *backend) const override {
if (op->type() == OpType_ReLU6) {
return new UnaryExecution("clamp(in,(float4)0,(float4)6)", backend);
}
if (op->type() == OpType_ReLU) {
if (op->main_as_Relu()->slope() == 0.0f) {
return new UnaryExecution("fmax(in,(float4)0)", backend);
}
auto slope = op->main_as_Relu()->slope();
char slopeCStr[30] = {};
sprintf(slopeCStr, "%.8f", slope);
std::string slopeStr = slopeCStr;
return new UnaryExecution("select(" + slopeStr + "f*in,in,in>=(float4)0)", backend);
}
if (op->type() == OpType_PReLU) {
if (op->main_as_PRelu()->slopeCount() == 1) {
auto slope = op->main_as_PRelu()->slope()->data()[0];
char slopeCStr[30] = {};
sprintf(slopeCStr, "%.8f", slope);
std::string slopeStr = slopeCStr;
return new UnaryExecution("select(" + slopeStr + "f*in,in,in>=(float4)0)", backend);
}
// FUNC_PRINT(1);
return new ReluExecution(inputs, op, backend);
}
return nullptr;
}
};
OpenCLCreatorRegister<ReluCreator> __Relu_op(OpType_ReLU);
OpenCLCreatorRegister<ReluCreator> __PRelu_op(OpType_PReLU);
OpenCLCreatorRegister<ReluCreator> __Relu6_op(OpType_ReLU6);
} // namespace OpenCL
} // namespace MNN