HcalObjects/src/HcalPedestalWidthsGPU.cc

0001 #include "CondFormats/HcalObjects/interface/HcalPedestalWidths.h"
0002 #include "CondFormats/HcalObjects/interface/HcalPedestalWidthsGPU.h"
0003 #include "FWCore/Utilities/interface/typelookup.h"
0004 #include "HeterogeneousCore/CUDAUtilities/interface/copyAsync.h"
0005
0006 // FIXME: add proper getters to conditions
0007 HcalPedestalWidthsGPU::HcalPedestalWidthsGPU(HcalPedestalWidths const& pedestals)
0008     : unitIsADC_{pedestals.isADC()},
0009       totalChannels_{pedestals.getAllContainers()[0].second.size() + pedestals.getAllContainers()[1].second.size()},
0010       sigma00_(totalChannels_),
0011       sigma01_(totalChannels_),
0012       sigma02_(totalChannels_),
0013       sigma03_(totalChannels_),
0014       sigma10_(totalChannels_),
0015       sigma11_(totalChannels_),
0016       sigma12_(totalChannels_),
0017       sigma13_(totalChannels_),
0018       sigma20_(totalChannels_),
0019       sigma21_(totalChannels_),
0020       sigma22_(totalChannels_),
0021       sigma23_(totalChannels_),
0022       sigma30_(totalChannels_),
0023       sigma31_(totalChannels_),
0024       sigma32_(totalChannels_),
0025       sigma33_(totalChannels_) {
0026   auto const containers = pedestals.getAllContainers();
0027
0028   // fill in hb
0029   auto const& barrelValues = containers[0].second;
0030   for (uint64_t i = 0; i < barrelValues.size(); ++i) {
0031     sigma00_[i] = *(barrelValues[i].getValues() /* + 0 */);
0032     sigma01_[i] = *(barrelValues[i].getValues() + 1);
0033     sigma02_[i] = *(barrelValues[i].getValues() + 2);
0034     sigma03_[i] = *(barrelValues[i].getValues() + 3);
0035     sigma10_[i] = *(barrelValues[i].getValues() + 3);
0036     sigma11_[i] = *(barrelValues[i].getValues() + 5);
0037     sigma12_[i] = *(barrelValues[i].getValues() + 6);
0038     sigma13_[i] = *(barrelValues[i].getValues() + 7);
0039     sigma20_[i] = *(barrelValues[i].getValues() + 8);
0040     sigma21_[i] = *(barrelValues[i].getValues() + 9);
0041     sigma22_[i] = *(barrelValues[i].getValues() + 10);
0042     sigma23_[i] = *(barrelValues[i].getValues() + 11);
0043     sigma30_[i] = *(barrelValues[i].getValues() + 12);
0044     sigma31_[i] = *(barrelValues[i].getValues() + 13);
0045     sigma32_[i] = *(barrelValues[i].getValues() + 14);
0046     sigma33_[i] = *(barrelValues[i].getValues() + 15);
0047   }
0048
0049   // fill in he
0050   auto const& endcapValues = containers[1].second;
0051   auto const offset = barrelValues.size();
0052   for (uint64_t i = 0; i < endcapValues.size(); ++i) {
0053     sigma00_[i + offset] = *(endcapValues[i].getValues() /* + 0 */);
0054     sigma01_[i + offset] = *(endcapValues[i].getValues() + 1);
0055     sigma02_[i + offset] = *(endcapValues[i].getValues() + 2);
0056     sigma03_[i + offset] = *(endcapValues[i].getValues() + 3);
0057     sigma10_[i + offset] = *(endcapValues[i].getValues() + 3);
0058     sigma11_[i + offset] = *(endcapValues[i].getValues() + 5);
0059     sigma12_[i + offset] = *(endcapValues[i].getValues() + 6);
0060     sigma13_[i + offset] = *(endcapValues[i].getValues() + 7);
0061     sigma20_[i + offset] = *(endcapValues[i].getValues() + 8);
0062     sigma21_[i + offset] = *(endcapValues[i].getValues() + 9);
0063     sigma22_[i + offset] = *(endcapValues[i].getValues() + 10);
0064     sigma23_[i + offset] = *(endcapValues[i].getValues() + 11);
0065     sigma30_[i + offset] = *(endcapValues[i].getValues() + 12);
0066     sigma31_[i + offset] = *(endcapValues[i].getValues() + 13);
0067     sigma32_[i + offset] = *(endcapValues[i].getValues() + 14);
0068     sigma33_[i + offset] = *(endcapValues[i].getValues() + 15);
0069   }
0070 }
0071
0072 HcalPedestalWidthsGPU::Product const& HcalPedestalWidthsGPU::getProduct(cudaStream_t stream) const {
0073   auto const& product =
0074       product_.dataForCurrentDeviceAsync(stream, [this](HcalPedestalWidthsGPU::Product& product, cudaStream_t stream) {
0075         // allocate
0076         product.sigma00 = cms::cuda::make_device_unique<float[]>(sigma00_.size(), stream);
0077         product.sigma01 = cms::cuda::make_device_unique<float[]>(sigma01_.size(), stream);
0078         product.sigma02 = cms::cuda::make_device_unique<float[]>(sigma02_.size(), stream);
0079         product.sigma03 = cms::cuda::make_device_unique<float[]>(sigma03_.size(), stream);
0080
0081         product.sigma10 = cms::cuda::make_device_unique<float[]>(sigma10_.size(), stream);
0082         product.sigma11 = cms::cuda::make_device_unique<float[]>(sigma11_.size(), stream);
0083         product.sigma12 = cms::cuda::make_device_unique<float[]>(sigma12_.size(), stream);
0084         product.sigma13 = cms::cuda::make_device_unique<float[]>(sigma13_.size(), stream);
0085
0086         product.sigma20 = cms::cuda::make_device_unique<float[]>(sigma20_.size(), stream);
0087         product.sigma21 = cms::cuda::make_device_unique<float[]>(sigma21_.size(), stream);
0088         product.sigma22 = cms::cuda::make_device_unique<float[]>(sigma22_.size(), stream);
0089         product.sigma23 = cms::cuda::make_device_unique<float[]>(sigma23_.size(), stream);
0090
0091         product.sigma30 = cms::cuda::make_device_unique<float[]>(sigma30_.size(), stream);
0092         product.sigma31 = cms::cuda::make_device_unique<float[]>(sigma31_.size(), stream);
0093         product.sigma32 = cms::cuda::make_device_unique<float[]>(sigma32_.size(), stream);
0094         product.sigma33 = cms::cuda::make_device_unique<float[]>(sigma33_.size(), stream);
0095
0096         // transfer
0097         cms::cuda::copyAsync(product.sigma00, sigma00_, stream);
0098         cms::cuda::copyAsync(product.sigma01, sigma01_, stream);
0099         cms::cuda::copyAsync(product.sigma02, sigma02_, stream);
0100         cms::cuda::copyAsync(product.sigma03, sigma03_, stream);
0101
0102         cms::cuda::copyAsync(product.sigma10, sigma10_, stream);
0103         cms::cuda::copyAsync(product.sigma11, sigma11_, stream);
0104         cms::cuda::copyAsync(product.sigma12, sigma12_, stream);
0105         cms::cuda::copyAsync(product.sigma13, sigma13_, stream);
0106
0107         cms::cuda::copyAsync(product.sigma20, sigma20_, stream);
0108         cms::cuda::copyAsync(product.sigma21, sigma21_, stream);
0109         cms::cuda::copyAsync(product.sigma22, sigma22_, stream);
0110         cms::cuda::copyAsync(product.sigma23, sigma23_, stream);
0111
0112         cms::cuda::copyAsync(product.sigma30, sigma30_, stream);
0113         cms::cuda::copyAsync(product.sigma31, sigma31_, stream);
0114         cms::cuda::copyAsync(product.sigma32, sigma32_, stream);
0115         cms::cuda::copyAsync(product.sigma33, sigma33_, stream);
0116       });
0117
0118   return product;
0119 }
0120
0121 TYPELOOKUP_DATA_REG(HcalPedestalWidthsGPU);