HcalObjects/src/HcalQIECodersGPU.cc

0001 #include "CondFormats/HcalObjects/interface/HcalQIECodersGPU.h"
0002 #include "FWCore/Utilities/interface/typelookup.h"
0003 #include "HeterogeneousCore/CUDAUtilities/interface/copyAsync.h"
0004
0005 HcalQIECodersGPU::HcalQIECodersGPU(HcalQIEData const& qiedata)
0006     : totalChannels_{qiedata.getAllContainers()[0].second.size() + qiedata.getAllContainers()[1].second.size()},
0007       offsets_(totalChannels_ * numValuesPerChannel),
0008       slopes_(totalChannels_ * numValuesPerChannel) {
0009   auto const containers = qiedata.getAllContainers();
0010
0011   // fill in hb
0012   auto const& barrelValues = containers[0].second;
0013   for (uint64_t i = 0; i < barrelValues.size(); ++i) {
0014     for (uint32_t k = 0; k < 4; k++)
0015       for (uint32_t l = 0; l < 4; l++) {
0016         auto const linear = k * 4 + l;
0017         offsets_[i * numValuesPerChannel + linear] = barrelValues[i].offset(k, l);
0018         slopes_[i * numValuesPerChannel + linear] = barrelValues[i].slope(k, l);
0019       }
0020   }
0021
0022   // fill in he
0023   auto const& endcapValues = containers[1].second;
0024   auto const offset = barrelValues.size();
0025   for (uint64_t i = 0; i < endcapValues.size(); ++i) {
0026     auto const off = (i + offset) * numValuesPerChannel;
0027     for (uint32_t k = 0; k < 4; k++)
0028       for (uint32_t l = 0; l < 4; l++) {
0029         auto const linear = k * 4u + l;
0030         offsets_[off + linear] = endcapValues[i].offset(k, l);
0031         slopes_[off + linear] = endcapValues[i].slope(k, l);
0032       }
0033   }
0034 }
0035
0036 HcalQIECodersGPU::Product const& HcalQIECodersGPU::getProduct(cudaStream_t stream) const {
0037   auto const& product =
0038       product_.dataForCurrentDeviceAsync(stream, [this](HcalQIECodersGPU::Product& product, cudaStream_t stream) {
0039         // allocate
0040         product.offsets = cms::cuda::make_device_unique<float[]>(offsets_.size(), stream);
0041         product.slopes = cms::cuda::make_device_unique<float[]>(slopes_.size(), stream);
0042
0043         // transfer
0044         cms::cuda::copyAsync(product.offsets, offsets_, stream);
0045         cms::cuda::copyAsync(product.slopes, slopes_, stream);
0046       });
0047
0048   return product;
0049 }
0050
0051 TYPELOOKUP_DATA_REG(HcalQIECodersGPU);