root
/
MNN
mirror of https://github.com/alibaba/MNN.git
1
0
Fork 0
Code Issues Actions 7 Packages Projects Releases Wiki Activity
MNN/source/backend/opencl/core/ImageBufferConvertor.cpp

683 lines
32 KiB
C++
Raw Blame History

//
// ImageBufferConvertor.cpp
// MNN
//
// Created by MNN on 2019/02/28.
// Copyright © 2018, Alibaba Group Holding Limited
//
#include "backend/opencl/core/ImageBufferConvertor.hpp"
namespace MNN {
namespace OpenCL {
static void AddBuildOptionOfDataTypeForImage(const Tensor *input, const Tensor *output, std::set<std::string> &buildOptions, int input_precision, int output_precision, bool toDevice, bool toHost){
if(input->getType().code == halide_type_int) {
buildOptions.emplace("-DINPUT_TYPE_I=int");
buildOptions.emplace("-DINPUT_TYPE_I4=int4");
if(input->getType().bits == 8){
buildOptions.emplace("-DINPUT_TYPE=char");
buildOptions.emplace("-DINPUT_TYPE4=char4");
buildOptions.emplace("-DRI_DATA=read_imagei");
} else if(input->getType().bits == 32){
buildOptions.emplace("-DINPUT_TYPE=int");
buildOptions.emplace("-DINPUT_TYPE4=int4");
buildOptions.emplace("-DRI_DATA=read_imagei");
} else {
MNN_PRINT("opencl input datatype not support, bit:%d\n", input->getType().bits);
MNN_ASSERT(false);
}
} else if(input->getType().code == halide_type_uint){
buildOptions.emplace("-DINPUT_TYPE_I=uint");
buildOptions.emplace("-DINPUT_TYPE_I4=uint4");
if(input->getType().bits == 8){
buildOptions.emplace("-DINPUT_TYPE=uchar");
buildOptions.emplace("-DINPUT_TYPE4=uchar4");
buildOptions.emplace("-DRI_DATA=read_imageui");
} else if(input->getType().bits == 32){
buildOptions.emplace("-DINPUT_TYPE=uint");
buildOptions.emplace("-DINPUT_TYPE4=uint4");
buildOptions.emplace("-DRI_DATA=read_imageui");
} else {
MNN_PRINT("opencl input datatype not support, bit:%d\n", input->getType().bits);
MNN_ASSERT(false);
}
} else {
if(input_precision != BackendConfig::Precision_High && toHost){
buildOptions.emplace("-DINPUT_TYPE_I=half");
buildOptions.emplace("-DINPUT_TYPE_I4=half4");
buildOptions.emplace("-DINPUT_TYPE=half");
buildOptions.emplace("-DINPUT_TYPE4=half4");
buildOptions.emplace("-DRI_DATA=read_imageh");
}else{
buildOptions.emplace("-DINPUT_TYPE_I=float");
buildOptions.emplace("-DINPUT_TYPE_I4=float4");
buildOptions.emplace("-DINPUT_TYPE=float");
buildOptions.emplace("-DINPUT_TYPE4=float4");
buildOptions.emplace("-DRI_DATA=read_imagef");
}
}
if(output->getType().code == halide_type_int) {
buildOptions.emplace("-DOUTPUT_TYPE_I=int");
buildOptions.emplace("-DOUTPUT_TYPE_I4=int4");
buildOptions.emplace("-DCONVERT_OUTPUT_I4=convert_int4");
if(output->getType().bits == 8){
buildOptions.emplace("-DOUTPUT_TYPE=char");
buildOptions.emplace("-DOUTPUT_TYPE4=char4");
buildOptions.emplace("-DCONVERT_OUTPUT4=convert_char4");
buildOptions.emplace("-DWI_DATA=write_imagei");
} else if(output->getType().bits == 32){
buildOptions.emplace("-DOUTPUT_TYPE=int");
buildOptions.emplace("-DOUTPUT_TYPE4=int4");
buildOptions.emplace("-DCONVERT_OUTPUT4=convert_int4");
buildOptions.emplace("-DWI_DATA=write_imagei");
} else {
MNN_PRINT("opencl input datatype not support, bit:%d\n", output->getType().bits);
MNN_ASSERT(false);
}
} else if(output->getType().code == halide_type_uint){
buildOptions.emplace("-DOUTPUT_TYPE_I=uint");
buildOptions.emplace("-DOUTPUT_TYPE_I4=uint4");
buildOptions.emplace("-DCONVERT_OUTPUT_I4=convert_uint4");
if(output->getType().bits == 8){
buildOptions.emplace("-DOUTPUT_TYPE=uchar");
buildOptions.emplace("-DOUTPUT_TYPE4=uchar4");
buildOptions.emplace("-DCONVERT_OUTPUT4=convert_uchar4");
buildOptions.emplace("-DWI_DATA=write_imageui");
} else if(output->getType().bits == 32){
buildOptions.emplace("-DOUTPUT_TYPE=uint");
buildOptions.emplace("-DOUTPUT_TYPE4=uint4");
buildOptions.emplace("-DCONVERT_OUTPUT4=convert_uint4");
buildOptions.emplace("-DWI_DATA=write_imageui");
} else {
MNN_PRINT("opencl input datatype not support, bit:%d\n", output->getType().bits);
MNN_ASSERT(false);
}
} else {
if(output_precision != BackendConfig::Precision_High && toDevice){
buildOptions.emplace("-DOUTPUT_TYPE_I=half");
buildOptions.emplace("-DOUTPUT_TYPE_I4=half4");
buildOptions.emplace("-DCONVERT_OUTPUT_I4=convert_half4");
buildOptions.emplace("-DOUTPUT_TYPE=half");
buildOptions.emplace("-DOUTPUT_TYPE4=half4");
buildOptions.emplace("-DCONVERT_OUTPUT4=convert_half4");
buildOptions.emplace("-DWI_DATA=write_imageh");
}else{
buildOptions.emplace("-DOUTPUT_TYPE_I=float");
buildOptions.emplace("-DOUTPUT_TYPE_I4=float4");
buildOptions.emplace("-DCONVERT_OUTPUT_I4=convert_float4");
buildOptions.emplace("-DOUTPUT_TYPE=float");
buildOptions.emplace("-DOUTPUT_TYPE4=float4");
buildOptions.emplace("-DCONVERT_OUTPUT4=convert_float4");
buildOptions.emplace("-DWI_DATA=write_imagef");
}
}
}
bool convertNCHWBufferToImage(const Tensor *input, Tensor *output,
OpenCLRuntime *runtime, int precision, bool needWait, bool svmFlag) {
std::vector<int> outputShape = tensorShapeFormat(input);
uint32_t outputGlobalWorkSize[2] = {static_cast<uint32_t>(UP_DIV(outputShape[3], 4) * outputShape[2]),
static_cast<uint32_t>(outputShape[0] * outputShape[1])};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, precision, precision, true, false);
auto bufferToImageKernelW = runtime->buildKernelWithCache("buffer_to_image", "nchw_buffer_to_image", buildOptions, precision);
auto bufferToImageKernel = bufferToImageKernelW->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= bufferToImageKernel.setArg(idx++, outputGlobalWorkSize[0]);
ret |= bufferToImageKernel.setArg(idx++, outputGlobalWorkSize[1]);
#ifdef MNN_OPENCL_SVM_ENABLE
if(svmFlag == true)
{
ret |= clSetKernelArgSVMPointer(bufferToImageKernel.get(), idx++, (const void *)input->deviceId());
}
else
#endif
{
ret |= bufferToImageKernel.setArg(idx++, openCLBuffer(input));
}
ret |= bufferToImageKernel.setArg(idx++, static_cast<uint32_t>(outputShape[1]));
ret |= bufferToImageKernel.setArg(idx++, static_cast<uint32_t>(outputShape[2]));
ret |= bufferToImageKernel.setArg(idx++, static_cast<uint32_t>(outputShape[3]));
ret |= bufferToImageKernel.setArg(idx++, openCLImage(output));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertNCHWBufferToImage");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(bufferToImageKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(outputGlobalWorkSize[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(bufferToImageKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "nchw_buffer_to_image");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"inputFormatTransform", event});
#endif
return true;
}
bool convertNHWCBufferToImage(const Tensor *input, Tensor *output,
OpenCLRuntime *runtime, int precision, bool needWait, bool svmFlag) {
std::vector<int> outputShape = tensorShapeFormat(input);
uint32_t outputGlobalWorkSize[2] = {static_cast<uint32_t>(UP_DIV(outputShape[3], 4) * outputShape[2]),
static_cast<uint32_t>(outputShape[0] * outputShape[1])};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, precision, precision, true, false);
auto bufferToImageKernelW = runtime->buildKernelWithCache("buffer_to_image", "nhwc_buffer_to_image", buildOptions, precision);
auto bufferToImageKernel = bufferToImageKernelW->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= bufferToImageKernel.setArg(idx++, outputGlobalWorkSize[0]);
ret |= bufferToImageKernel.setArg(idx++, outputGlobalWorkSize[1]);
#ifdef MNN_OPENCL_SVM_ENABLE
if(svmFlag == true) {
ret |= clSetKernelArgSVMPointer(bufferToImageKernel.get(), idx++, (const void *)input->deviceId());
}
else
#endif
{
ret |= bufferToImageKernel.setArg(idx++, openCLBuffer(input));
}
ret |= bufferToImageKernel.setArg(idx++, static_cast<uint32_t>(outputShape[1]));
ret |= bufferToImageKernel.setArg(idx++, static_cast<uint32_t>(outputShape[2]));
ret |= bufferToImageKernel.setArg(idx++, static_cast<uint32_t>(outputShape[3]));
ret |= bufferToImageKernel.setArg(idx++, openCLImage(output));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertNHWCBufferToImage");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(bufferToImageKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(outputGlobalWorkSize[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(bufferToImageKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "nhwc_buffer_to_image");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"inputFormatTransform", event});
#endif
return true;
}
bool convertImageToNCHWBuffer(const Tensor *input, Tensor *output,
OpenCLRuntime *runtime, int precision, bool needWait, bool svmFlag) {
std::vector<int> inputShape = tensorShapeFormat(input);
uint32_t in_gws[2] = {static_cast<uint32_t>(UP_DIV(inputShape[3], 4) * inputShape[2]),
static_cast<uint32_t>(inputShape[0] * inputShape[1])};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, precision, precision, false, true);
auto imageToBufferKernelW = runtime->buildKernelWithCache("buffer_to_image", "image_to_nchw_buffer", buildOptions, precision);
auto imageToBufferKernel = imageToBufferKernelW->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= imageToBufferKernel.setArg(idx++, in_gws[0]);
ret |= imageToBufferKernel.setArg(idx++, in_gws[1]);
#ifdef MNN_OPENCL_SVM_ENABLE
if(svmFlag == true)
{
ret |= clSetKernelArgSVMPointer(imageToBufferKernel.get(), idx++, (const void *)output->deviceId());
}
else
#endif
{
ret |= imageToBufferKernel.setArg(idx++, openCLBuffer(output));
}
ret |= imageToBufferKernel.setArg(idx++, static_cast<uint32_t>(inputShape[1]));
ret |= imageToBufferKernel.setArg(idx++, static_cast<uint32_t>(inputShape[2]));
ret |= imageToBufferKernel.setArg(idx++, static_cast<uint32_t>(inputShape[3]));
ret |= imageToBufferKernel.setArg(idx++, openCLImage(input));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertImageToNCHWBuffer");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(imageToBufferKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(in_gws[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(imageToBufferKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "image_to_nchw_buffer");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"outputFormatTransform", event});
#endif
return true;
}
bool convertNC4HW4BufferToImage(const Tensor *input, Tensor *output,
OpenCLRuntime *runtime, int precision, bool needWait, bool svmFlag) {
uint32_t outputGlobalWorkSize[2] = {static_cast<uint32_t>(UP_DIV(input->channel(), 4) * input->width()),
static_cast<uint32_t>(input->batch() * input->height())};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, precision, precision, true, false);
auto bufferToImageKernelW = runtime->buildKernelWithCache("buffer_to_image", "nc4hw4_buffer_to_image", buildOptions, precision);
auto bufferToImageKernel = bufferToImageKernelW->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
int outputImageShape[2] = {input->height(), input->width()};
ret |= bufferToImageKernel.setArg(idx++, outputGlobalWorkSize[0]);
ret |= bufferToImageKernel.setArg(idx++, outputGlobalWorkSize[1]);
#ifdef MNN_OPENCL_SVM_ENABLE
if(svmFlag == true)
{
ret |= clSetKernelArgSVMPointer(bufferToImageKernel.get(), idx++, (const void *)input->deviceId());
}
else
#endif
{
ret |= bufferToImageKernel.setArg(idx++, openCLBuffer(input));
}
ret |= bufferToImageKernel.setArg(idx++, sizeof(outputImageShape), outputImageShape);
ret |= bufferToImageKernel.setArg(idx++, input->batch());
ret |= bufferToImageKernel.setArg(idx++, openCLImage(output));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertNC4HW4BufferToImage");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(bufferToImageKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(outputGlobalWorkSize[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(bufferToImageKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "nc4hw4_buffer_to_image");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"inputFormatTransform", event});
#endif
return true;
}
/**
* @brief convert image to nc/4hwc%4 buffer.
* @param input input tensor.
* @param output output tensor.
* @param bufferToImageKernel opencl kernel reference.
* @param runtime opencl runtime instance pointer.
* @param needWait whether need wait opencl complete before return or not, default false.
* @return true if success, false otherwise.
*/
bool convertImageToNC4HW4Buffer(const Tensor *input, Tensor *output,
OpenCLRuntime *runtime, int precision, bool needWait, bool svmFlag) {
auto inputShape = tensorShapeFormat(input);
uint32_t in_gws[2] = {static_cast<uint32_t>(UP_DIV(inputShape.at(3), 4) * inputShape.at(2)),
static_cast<uint32_t>(inputShape.at(0) * inputShape.at(1))};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, precision, precision, false, true);
auto imageToBufferKernelW = runtime->buildKernelWithCache("buffer_to_image", "image_to_nc4hw4_buffer", buildOptions, precision);
auto imageToBufferKernel = imageToBufferKernelW->get();
uint32_t idx = 0;
int outputImageShape[2] = {inputShape.at(1), inputShape.at(2)};
cl_int ret = CL_SUCCESS;
ret |= imageToBufferKernel.setArg(idx++, in_gws[0]);
ret |= imageToBufferKernel.setArg(idx++, in_gws[1]);
#ifdef MNN_OPENCL_SVM_ENABLE
if(svmFlag == true)
{
ret |= clSetKernelArgSVMPointer(imageToBufferKernel.get(), idx++, (const void *)output->deviceId());
}
else
#endif
{
ret |= imageToBufferKernel.setArg(idx++, openCLBuffer(output));
}
ret |= imageToBufferKernel.setArg(idx++, sizeof(outputImageShape), outputImageShape);
ret |= imageToBufferKernel.setArg(idx++, input->batch());
ret |= imageToBufferKernel.setArg(idx++, openCLImage(input));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertImageToNC4HW4Buffer");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(imageToBufferKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(in_gws[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(imageToBufferKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "image_to_nc4hw4_buffer");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"outputFormatTransform", event});
#endif
return true;
}
bool convertImageToNHWCBuffer(const Tensor *input, Tensor *output,
OpenCLRuntime *runtime, int precision, bool needWait, bool svmFlag) {
std::vector<int> inputShape = tensorShapeFormat(input);
uint32_t in_gws[2] = {static_cast<uint32_t>(UP_DIV(inputShape[3], 4) * inputShape[2]),
static_cast<uint32_t>(inputShape[0] * inputShape[1])};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, precision, precision, false, true);
auto imageToBufferKernelW = runtime->buildKernelWithCache("buffer_to_image", "image_to_nhwc_buffer", buildOptions, precision);
auto imageToBufferKernel = imageToBufferKernelW->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= imageToBufferKernel.setArg(idx++, in_gws[0]);
ret |= imageToBufferKernel.setArg(idx++, in_gws[1]);
#ifdef MNN_OPENCL_SVM_ENABLE
if(svmFlag == true)
{
ret |= clSetKernelArgSVMPointer(imageToBufferKernel.get(), idx++, (const void *)output->deviceId());
}
else
#endif
{
ret |= imageToBufferKernel.setArg(idx++, openCLBuffer(output));
}
ret |= imageToBufferKernel.setArg(idx++, static_cast<uint32_t>(inputShape[1]));
ret |= imageToBufferKernel.setArg(idx++, static_cast<uint32_t>(inputShape[2]));
ret |= imageToBufferKernel.setArg(idx++, static_cast<uint32_t>(inputShape[3]));
ret |= imageToBufferKernel.setArg(idx++, openCLImage(input));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertImageToNHWCBuffer");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(imageToBufferKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(in_gws[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(imageToBufferKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "image_to_nhwc_buffer");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"outputFormatTransform", event});
#endif
return true;
}
bool convertImageToImage(Tensor *input, Tensor *output, OpenCLRuntime *runtime, int input_precision, int output_precision, int backend_precison, bool needWait){
std::vector<int> inputShape = tensorShapeFormat(input);
uint32_t in_gws[2] = {static_cast<uint32_t>(UP_DIV(inputShape[3], 4) * inputShape[2]), static_cast<uint32_t>(inputShape[0] * inputShape[1])};
std::set<std::string> buildOptions;
AddBuildOptionOfDataTypeForImage(input, output, buildOptions, input_precision, output_precision, false, true);
auto imageToBufferKernelW = runtime->buildKernelWithCache("buffer_to_image", "image_to_image", buildOptions, backend_precison);
auto imageToBufferKernel = imageToBufferKernelW->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= imageToBufferKernel.setArg(idx++, in_gws[0]);
ret |= imageToBufferKernel.setArg(idx++, in_gws[1]);
ret |= imageToBufferKernel.setArg(idx++, openCLImage(output));
ret |= imageToBufferKernel.setArg(idx++, openCLImage(input));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertImageToImage");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(imageToBufferKernelW));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(in_gws[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(imageToBufferKernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "image_to_image");
if (true == needWait) {
event.wait();
}
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"convertImageToImage", event});
#endif
return true;
}
bool ImageBufferConvertor::convertImageToBuffer(const Tensor *image, const OpenCLBufferFormat type, Tensor *buffer, int precision,
bool needWait, bool svmFlag) {
#ifdef LOG_VERBOSE
MNN_PRINT("start convertImageToBuffer !\n");
#endif
auto formattedBufferShape = tensorShapeFormat(image);
auto runtime = mOpenCLRuntime;
std::string kernelName;
if (type == NHWC_BUFFER) {
kernelName = "image_to_nhwc_buffer";
} else if (type == NCHW_BUFFER) {
kernelName = "image_to_nchw_buffer";
} else if (type == CONV2D_FILTER) {
kernelName = "conv2d_filter_image_to_buffer";
} else if (type == ARGUMENT) {
kernelName = "arg_image_to_buffer";
} else {
MNN_PRINT("not support such type !!! \n");
}
if (mImageToBufferKernel.get() == nullptr || mImageToBufferKernelName != kernelName) {
mImageToBufferKernelName = kernelName;
std::set<std::string> buildOptions;
mImageToBufferKernel = runtime->buildKernelWithCache("buffer_to_image", kernelName, buildOptions, precision, image, buffer);
}
auto kernel = mImageToBufferKernel->get();
std::vector<size_t> gws;
getImageShape(formattedBufferShape, type, &gws);
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= kernel.setArg(idx++, gws[0]);
ret |= kernel.setArg(idx++, gws[1]);
ret |= kernel.setArg(idx++, openCLBuffer(buffer));
if (type == CONV2D_FILTER) {
const int channelHeightWidthSumSize =
buffer->buffer().dim[1].extent * buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
const int heightWidthSumSize = buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
int kernelShape[2] = {buffer->buffer().dim[2].extent, buffer->buffer().dim[3].extent};
ret |= kernel.setArg(idx++, static_cast<uint32_t>(buffer->buffer().dim[0].extent));
ret |= kernel.setArg(idx++, sizeof(kernelShape), kernelShape);
ret |= kernel.setArg(idx++, static_cast<uint32_t>(channelHeightWidthSumSize));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(heightWidthSumSize));
} else if (type == ARGUMENT) {
ret |= kernel.setArg(idx++, static_cast<uint32_t>(buffer->buffer().dim[0].extent));
} else {
ret |= kernel.setArg(idx++, static_cast<uint32_t>(formattedBufferShape[1]));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(formattedBufferShape[2]));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(formattedBufferShape[3]));
}
ret |= kernel.setArg(idx++, openCLImage(image));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertImageToBuffer");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(mImageToBufferKernel));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(gws[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(kernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "convertImageToBuffer");
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"convertBufferToImage", event});
#endif
if (needWait) {
event.wait();
}
#ifdef LOG_VERBOSE
MNN_PRINT("end convertImageToBuffer !\n");
#endif
return true;
}
bool ImageBufferConvertor::convertBufferToImage(const Tensor *buffer, const OpenCLBufferFormat type, Tensor *image, int precision, bool needWait, const std::string &buildOption) {
#ifdef LOG_VERBOSE
MNN_PRINT("start convertBufferToImage !\n");
#endif
auto formattedBufferShape = tensorShapeFormat(buffer);
std::vector<size_t> imageShape;
getImageShape(formattedBufferShape, type, &imageShape);
uint32_t gws[2] = {static_cast<uint32_t>(imageShape[0]), static_cast<uint32_t>(imageShape[1])};
auto runtime = mOpenCLRuntime;
std::string kernelName;
switch (type) {
case CONV2D_FILTER:
kernelName = "conv2d_filter_buffer_to_image";
break;
case CONV2D1x1_OPT_FILTER:
kernelName = "conv2d1x1_opt_filter_buffer_to_image";
break;
case DW_CONV2D_FILTER:
kernelName = "dw_filter_buffer_to_image";
break;
case NHWC_BUFFER:
kernelName = "nhwc_buffer_to_image";
break;
case NCHW_BUFFER:
kernelName = "nchw_buffer_to_image";
break;
case ARGUMENT:
kernelName = "arg_buffer_to_image";
break;
default:
break;
}
if (mBufferToImageKernel.get() == nullptr || mBufferToImageKernelName != kernelName) {
mBufferToImageKernelName = kernelName;
std::set<std::string> buildOptions;
buildOptions.emplace(buildOption);
mBufferToImageKernel = runtime->buildKernelWithCache("buffer_to_image", kernelName, buildOptions, precision, buffer, image);
}
auto kernel = mBufferToImageKernel->get();
uint32_t idx = 0;
cl_int ret = CL_SUCCESS;
ret |= kernel.setArg(idx++, gws[0]);
ret |= kernel.setArg(idx++, gws[1]);
ret |= kernel.setArg(idx++, openCLBuffer(buffer));
if (type == CONV2D_FILTER) {
const int channelHeightWidthSumSize =
buffer->buffer().dim[1].extent * buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
const int heightWidthSumSize = buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
int kernelShape[2] = {buffer->buffer().dim[2].extent, buffer->buffer().dim[3].extent};
ret |= kernel.setArg(idx++, static_cast<uint32_t>(buffer->buffer().dim[0].extent));
ret |= kernel.setArg(idx++, sizeof(kernelShape),kernelShape);
ret |= kernel.setArg(idx++, static_cast<uint32_t>(channelHeightWidthSumSize));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(heightWidthSumSize));
} else if (type == DW_CONV2D_FILTER) {
const int heightWidthSumSize = buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
int kernelShape[4] = {buffer->buffer().dim[0].extent, buffer->buffer().dim[1].extent, buffer->buffer().dim[2].extent, buffer->buffer().dim[3].extent};
ret |= kernel.setArg(idx++, sizeof(kernelShape),kernelShape);
ret |= kernel.setArg(idx++, static_cast<uint32_t>(heightWidthSumSize));
} else if (type == ARGUMENT) {
ret |= kernel.setArg(idx++, static_cast<uint32_t>(buffer->buffer().dim[0].extent));
} else if(type == CONV2D1x1_OPT_FILTER){
const int channelHeightWidthSumSize =
buffer->buffer().dim[1].extent * buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
const int heightWidthSumSize = buffer->buffer().dim[2].extent * buffer->buffer().dim[3].extent;
int kernelShape[2] = {buffer->buffer().dim[2].extent, buffer->buffer().dim[3].extent};
ret |= kernel.setArg(idx++, static_cast<uint32_t>(buffer->buffer().dim[1].extent));
ret |= kernel.setArg(idx++, sizeof(kernelShape),kernelShape);
ret |= kernel.setArg(idx++, static_cast<uint32_t>(channelHeightWidthSumSize));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(heightWidthSumSize));
}else {
ret |= kernel.setArg(idx++, static_cast<uint32_t>(formattedBufferShape[1]));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(formattedBufferShape[2]));
ret |= kernel.setArg(idx++, static_cast<uint32_t>(formattedBufferShape[3]));
}
ret |= kernel.setArg(idx++, openCLImage(image));
MNN_CHECK_CL_SUCCESS(ret, "setArg convertBufferToImage");
const uint32_t maxWorkGroupSize = static_cast<uint32_t>(runtime->getMaxWorkGroupSize(mBufferToImageKernel));
const std::vector<uint32_t> lws = {16, std::max((uint32_t)1, maxWorkGroupSize / 16)};
cl::Event event;
cl_int res;
std::vector<uint32_t> roundUpGroupWorkSize(lws.size());
for (size_t i = 0; i < lws.size(); ++i) {
roundUpGroupWorkSize[i] = ROUND_UP(gws[i], lws[i]);
}
res = runtime->commandQueue().enqueueNDRangeKernel(kernel, cl::NullRange,
cl::NDRange(roundUpGroupWorkSize[0], roundUpGroupWorkSize[1]),
cl::NDRange(lws[0], lws[1]), nullptr, &event);
MNN_CHECK_CL_SUCCESS(res, "convertBufferToImage");
#ifdef ENABLE_OPENCL_TIME_PROFILER
runtime->pushEvent({"convertBufferToImage", event});
#endif
if (needWait) {
event.wait();
}
#ifdef LOG_VERBOSE
MNN_PRINT("end convertBufferToImage !\n");
#endif
return true;
}
} // namespace OpenCL
} // namespace MNN
Reference in New Issue View Git Blame Copy Permalink