wehub-resource-sync
82 lines
2.9 KiB
C++
82 lines
2.9 KiB
C++
//
|
|
// RasterGrad.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on 2022/10/09.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include "OpGrad.hpp"
|
|
#include "core/TensorUtils.hpp"
|
|
using namespace std;
|
|
namespace MNN {
|
|
using namespace MNN::Express;
|
|
|
|
class RasterGrad : public OpGrad {
|
|
public:
|
|
virtual std::vector<Express::VARP> onGrad(Express::EXPRP expr,
|
|
const std::vector<Express::VARP>& backwardOutput) override {
|
|
auto inputs = expr->inputs();
|
|
std::vector<VARP> result(inputs.size(), nullptr);
|
|
auto rasterInfo = expr->get()->main_as_Extra();
|
|
const int32_t* regionData = nullptr;
|
|
const int REGION_LENGTH = 11;
|
|
std::vector<int32_t> regionDataHolder;
|
|
if (nullptr != rasterInfo && nullptr != rasterInfo->attr()) {
|
|
for (int i=0; i<rasterInfo->attr()->size(); ++i) {
|
|
auto attr = rasterInfo->attr()->GetAs<Attribute>(i);
|
|
if (attr->key()->str() == "region") {
|
|
regionData = attr->list()->i()->data();
|
|
MNN_ASSERT(inputs.size() * REGION_LENGTH == attr->list()->i()->size());
|
|
break;
|
|
}
|
|
}
|
|
} else {
|
|
regionDataHolder.resize(inputs.size() * REGION_LENGTH);
|
|
regionData = regionDataHolder.data();
|
|
auto outputTensor = Variable::create(expr)->getTensor();
|
|
auto des = TensorUtils::getDescribe(outputTensor);
|
|
MNN_ASSERT(des->regions.size() == inputs.size());
|
|
for (int i=0; i<inputs.size(); ++i) {
|
|
auto& r = des->regions[i];
|
|
auto dstPtr = regionDataHolder.data() + REGION_LENGTH * i;
|
|
dstPtr[0] = r.src.offset;
|
|
dstPtr[1] = r.src.stride[0];
|
|
dstPtr[2] = r.src.stride[1];
|
|
dstPtr[3] = r.src.stride[2];
|
|
|
|
dstPtr[4] = r.dst.offset;
|
|
dstPtr[5] = r.dst.stride[0];
|
|
dstPtr[6] = r.dst.stride[1];
|
|
dstPtr[7] = r.dst.stride[2];
|
|
|
|
dstPtr[8] = r.size[0];
|
|
dstPtr[9] = r.size[1];
|
|
dstPtr[10] = r.size[2];
|
|
}
|
|
}
|
|
for (int i=0; i<inputs.size(); ++i) {
|
|
auto regionInfo = regionData + REGION_LENGTH * i;
|
|
auto info = inputs[i]->getInfo();
|
|
auto shape = info->dim;
|
|
std::vector<int> curRegion(REGION_LENGTH);
|
|
::memcpy(curRegion.data(), regionInfo, REGION_LENGTH * sizeof(int));
|
|
::memcpy(curRegion.data(), regionInfo + 4, 4 * sizeof(int));
|
|
::memcpy(curRegion.data() + 4, regionInfo, 4 * sizeof(int));
|
|
auto grad = _RasterRaw({backwardOutput[0]}, curRegion, shape, info->type, info->order);
|
|
result[i] = grad;
|
|
}
|
|
return result;
|
|
}
|
|
};
|
|
|
|
static void _create() {
|
|
static RasterGrad _c;
|
|
OpGrad::insert(OpType_Raster, &_c);
|
|
|
|
}
|
|
|
|
REGISTER_GRAD(RasterGrad_cpp, _create);
|
|
};
|
|
|