Moved more things out of the alpaka directory

RecoTracker/LST/plugins/alpaka/LSTModulesDevESProducer.cc

@@ -8,8 +8,8 @@
 #include "RecoTracker/Record/interface/TrackerRecoGeometryRecord.h"
 
 // LST includes
-#include "RecoTracker/LSTCore/interface/alpaka/Module.h"
-#include "RecoTracker/LSTCore/interface/alpaka/LST.h"
+#include "RecoTracker/LSTCore/interface/Module.h"
+#include "RecoTracker/LSTCore/interface/LST.h"
 
 namespace ALPAKA_ACCELERATOR_NAMESPACE {
 

RecoTracker/LST/plugins/alpaka/LSTProducer.cc

@@ -19,7 +19,7 @@
 
 #include "RecoTracker/Record/interface/TrackerRecoGeometryRecord.h"
 
-#include "RecoTracker/LSTCore/interface/alpaka/LST.h"
+#include "RecoTracker/LSTCore/interface/LST.h"
 
 namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
@@ -91,7 +91,7 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
     const bool verbose_, nopLSDupClean_, tcpLSTriplets_;
     edm::EDPutTokenT<LSTOutput> lstOutputToken_;
 
-    SDL::LST<Acc3D> lst_;
+    SDL::LST<Device> lst_;
   };
 
 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE

RecoTracker/LST/src/ES_ModulesDev.cc

@@ -0,0 +1,5 @@
+#include "RecoTracker/LSTCore/interface/LSTESData.h"
+#include "HeterogeneousCore/AlpakaInterface/interface/config.h"
+#include "FWCore/Utilities/interface/typelookup.h"
+
+TYPELOOKUP_DATA_REG(SDL::LSTESData<alpaka_common::DevHost>);

RecoTracker/LST/src/alpaka/ES_ModulesDev.cc

@@ -1,7 +1,4 @@
-#include "RecoTracker/LSTCore/interface/alpaka/LST.h"
+#include "RecoTracker/LSTCore/interface/LSTESData.h"
 #include "HeterogeneousCore/AlpakaCore/interface/alpaka/typelookup.h"
 
-// Temporary hack: The DevHost instantiation is needed in both CPU and GPU plugins,
-// whereas the (non-host-)Device instantiation only in the GPU plugin
-TYPELOOKUP_DATA_REG(ALPAKA_ACCELERATOR_NAMESPACE::SDL::LSTESData<alpaka_common::DevHost>);
-TYPELOOKUP_ALPAKA_TEMPLATED_DATA_REG(ALPAKA_ACCELERATOR_NAMESPACE::SDL::LSTESData);
+TYPELOOKUP_ALPAKA_TEMPLATED_DATA_REG(SDL::LSTESData);

RecoTracker/LSTCore/interface/Constants.h

@@ -1,9 +1,36 @@
 #ifndef RecoTracker_LSTCore_interface_Constants_h
 #define RecoTracker_LSTCore_interface_Constants_h
 
+#include "HeterogeneousCore/AlpakaInterface/interface/config.h"
+
+#ifdef CACHE_ALLOC
+#include "HeterogeneousCore/AlpakaInterface/interface/CachedBufAlloc.h"
+#endif
 
 namespace SDL {
 
+  // Buffer type for allocations where auto type can't be used.
+  template <typename TDev, typename TData>
+  using Buf = alpaka::Buf<TDev, TData, alpaka_common::Dim1D, alpaka_common::Idx>;
+
+  // Allocation wrapper function to make integration of the caching allocator easier and reduce code boilerplate.
+  template <typename T, typename TAcc, typename TSize, typename TQueue>
+  ALPAKA_FN_HOST ALPAKA_FN_INLINE Buf<alpaka::Dev<TAcc>, T> allocBufWrapper(TAcc const& devAccIn,
+                                                                            TSize nElements,
+                                                                            TQueue queue) {
+#ifdef CACHE_ALLOC
+    return cms::alpakatools::allocCachedBuf<T, alpaka_common::Idx>(devAccIn, queue, alpaka_common::Vec1D(static_cast<alpaka_common::Idx>(nElements)));
+#else
+    return alpaka::allocBuf<T, alpaka_common::Idx>(devAccIn, alpaka_common::Vec1D(static_cast<alpaka_common::Idx>(nElements)));
+#endif
+  }
+
+  // Second allocation wrapper function when queue is not given. Reduces code boilerplate.
+  template <typename T, typename TAcc, typename TSize>
+  ALPAKA_FN_HOST ALPAKA_FN_INLINE Buf<alpaka::Dev<TAcc>, T> allocBufWrapper(TAcc const& devAccIn, TSize nElements) {
+    return alpaka::allocBuf<T, alpaka_common::Idx>(devAccIn, alpaka_common::Vec1D(static_cast<alpaka_common::Idx>(nElements)));
+  }
+
 // If a compile time flag does not define PT_CUT, default to 0.8 (GeV)
 #ifndef PT_CUT
   constexpr float PT_CUT = 0.8f;

RecoTracker/LSTCore/interface/alpaka/EndcapGeometryBuffer.h → RecoTracker/LSTCore/interface/EndcapGeometryBuffer.h

@@ -1,5 +1,5 @@
-#ifndef RecoTracker_LSTCore_interface_alpaka_EndcapGeometryBuffers_h
-#define RecoTracker_LSTCore_interface_alpaka_EndcapGeometryBuffers_h
+#ifndef RecoTracker_LSTCore_interface_EndcapGeometryBuffers_h
+#define RecoTracker_LSTCore_interface_EndcapGeometryBuffers_h
 
 #include <map>
 #include <iostream>
@@ -9,11 +9,9 @@
 #include <vector>
 #include <stdexcept>
 
-#include "HeterogeneousCore/AlpakaInterface/interface/host.h"
+#include "RecoTracker/LSTCore/interface/Constants.h"
 
-#include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
-
-namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
+namespace SDL {
 
   struct EndcapGeometryDev {
     const unsigned int* geoMapDetId;

RecoTracker/LSTCore/interface/alpaka/LST.h → RecoTracker/LSTCore/interface/LST.h

@@ -1,29 +1,30 @@
-#ifndef RecoTracker_LSTCore_interface_alpaka_LST_h
-#define RecoTracker_LSTCore_interface_alpaka_LST_h
+#ifndef RecoTracker_LSTCore_interface_LST_h
+#define RecoTracker_LSTCore_interface_LST_h
 
-#include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
-#include "RecoTracker/LSTCore/interface/alpaka/LSTESData.h"
+#include "RecoTracker/LSTCore/interface/Constants.h"
+#include "RecoTracker/LSTCore/interface/LSTESData.h"
 
 #include <cstdlib>
 #include <numeric>
 #include <mutex>
 #include <alpaka/alpaka.hpp>
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
+namespace SDL {
   template <typename>
   class Event;
 
   template <typename>
   class LST;
 
-  template <>
-  class LST<Acc3D> {
+  template <typename TDev>
+  class LST {
   public:
     LST() = default;
 
-    void run(Queue& queue,
+    template <typename TQueue>
+    void run(TQueue& queue,
              bool verbose,
-             const LSTESData<Device>* deviceESData,
+             const LSTESData<TDev>* deviceESData,
              const std::vector<float> see_px,
              const std::vector<float> see_py,
              const std::vector<float> see_pz,
@@ -71,7 +72,7 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
                       const std::vector<float> ph2_y,
                       const std::vector<float> ph2_z);
 
-    void getOutput(SDL::Event<Acc3D>& event);
+    void getOutput(SDL::Event<TDev>& event);
     std::vector<unsigned int> getHitIdxs(const short trackCandidateType,
                                          const unsigned int TCIdx,
                                          const unsigned int* TCHitIndices,

RecoTracker/LSTCore/interface/alpaka/LSTESData.h → RecoTracker/LSTCore/interface/LSTESData.h

@@ -1,17 +1,17 @@
-#ifndef RecoTracker_LSTCore_interface_alpaka_LSTESData_h
-#define RecoTracker_LSTCore_interface_alpaka_LSTESData_h
+#ifndef RecoTracker_LSTCore_interface_LSTESData_h
+#define RecoTracker_LSTCore_interface_LSTESData_h
 
-#include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
-#include "RecoTracker/LSTCore/interface/alpaka/EndcapGeometryBuffer.h"
-#include "RecoTracker/LSTCore/interface/alpaka/Module.h"
+#include "RecoTracker/LSTCore/interface/Constants.h"
+#include "RecoTracker/LSTCore/interface/EndcapGeometryBuffer.h"
+#include "RecoTracker/LSTCore/interface/Module.h"
 #include "RecoTracker/LSTCore/interface/PixelMap.h"
 
 #include "HeterogeneousCore/AlpakaInterface/interface/CopyToDevice.h"
 
 #include <filesystem>
 #include <memory>
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
+namespace SDL {
 
   template <typename TDev>
   struct LSTESData {
@@ -21,15 +21,15 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
     unsigned int nEndCapMap;
     std::shared_ptr<const modulesBuffer<TDev>> modulesBuffers;
     std::shared_ptr<const EndcapGeometryBuffer<TDev>> endcapGeometryBuffers;
-    std::shared_ptr<const ::SDL::pixelMap> pixelMapping;
+    std::shared_ptr<const pixelMap> pixelMapping;
 
     LSTESData(uint16_t const& nModulesIn,
               uint16_t const& nLowerModulesIn,
               unsigned int const& nPixelsIn,
               unsigned int const& nEndCapMapIn,
               std::shared_ptr<const modulesBuffer<TDev>> const& modulesBuffersIn,
               std::shared_ptr<const EndcapGeometryBuffer<TDev>> const& endcapGeometryBuffersIn,
-              std::shared_ptr<const ::SDL::pixelMap> const& pixelMappingIn)
+              std::shared_ptr<const pixelMap> const& pixelMappingIn)
         : nModules(nModulesIn),
           nLowerModules(nLowerModulesIn),
           nPixels(nPixelsIn),
@@ -45,18 +45,18 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
 
 namespace cms::alpakatools {
   template <>
-  struct CopyToDevice<ALPAKA_ACCELERATOR_NAMESPACE::SDL::LSTESData<alpaka_common::DevHost>> {
+  struct CopyToDevice<SDL::LSTESData<alpaka_common::DevHost>> {
     template <typename TQueue>
-    static ALPAKA_ACCELERATOR_NAMESPACE::SDL::LSTESData<alpaka::Dev<TQueue>> copyAsync(TQueue& queue,
-                                                         ALPAKA_ACCELERATOR_NAMESPACE::SDL::LSTESData<alpaka_common::DevHost> const& srcData) {
-      auto deviceModulesBuffers = std::make_shared<ALPAKA_ACCELERATOR_NAMESPACE::SDL::modulesBuffer<alpaka::Dev<TQueue>>>(
+    static SDL::LSTESData<alpaka::Dev<TQueue>> copyAsync(TQueue& queue,
+                                                         SDL::LSTESData<alpaka_common::DevHost> const& srcData) {
+      auto deviceModulesBuffers = std::make_shared<SDL::modulesBuffer<alpaka::Dev<TQueue>>>(
           alpaka::getDev(queue), srcData.nModules, srcData.nPixels);
       deviceModulesBuffers->copyFromSrc(queue, *srcData.modulesBuffers);
       auto deviceEndcapGeometryBuffers =
-          std::make_shared<ALPAKA_ACCELERATOR_NAMESPACE::SDL::EndcapGeometryBuffer<alpaka::Dev<TQueue>>>(alpaka::getDev(queue), srcData.nEndCapMap);
+          std::make_shared<SDL::EndcapGeometryBuffer<alpaka::Dev<TQueue>>>(alpaka::getDev(queue), srcData.nEndCapMap);
       deviceEndcapGeometryBuffers->copyFromSrc(queue, *srcData.endcapGeometryBuffers);
 
-      return ALPAKA_ACCELERATOR_NAMESPACE::SDL::LSTESData<alpaka::Dev<TQueue>>(srcData.nModules,
+      return SDL::LSTESData<alpaka::Dev<TQueue>>(srcData.nModules,
                                                  srcData.nLowerModules,
                                                  srcData.nPixels,
                                                  srcData.nEndCapMap,

RecoTracker/LSTCore/interface/alpaka/Module.h → RecoTracker/LSTCore/interface/Module.h

@@ -1,11 +1,9 @@
-#ifndef RecoTracker_LSTCore_interface_alpaka_Module_h
-#define RecoTracker_LSTCore_interface_alpaka_Module_h
+#ifndef RecoTracker_LSTCore_interface_Module_h
+#define RecoTracker_LSTCore_interface_Module_h
 
-#include <alpaka/alpaka.hpp>
+#include "RecoTracker/LSTCore/interface/Constants.h"
 
-#include "RecoTracker/LSTCore/interface/alpaka/Constants.h"
-
-namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
+namespace SDL {
   enum SubDet { InnerPixel = 0, Barrel = 5, Endcap = 4 };
 
   enum Side { NegZ = 1, PosZ = 2, Center = 3 };
@@ -286,7 +284,7 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
 
     modulesBuffer(TDev const& dev, unsigned int nMod, unsigned int nPixs)
         : detIds_buf(allocBufWrapper<unsigned int>(dev, nMod)),
-          moduleMap_buf(allocBufWrapper<uint16_t>(dev, nMod * ::SDL::MAX_CONNECTED_MODULES)),
+          moduleMap_buf(allocBufWrapper<uint16_t>(dev, nMod * MAX_CONNECTED_MODULES)),
           mapdetId_buf(allocBufWrapper<unsigned int>(dev, nMod)),
           mapIdx_buf(allocBufWrapper<uint16_t>(dev, nMod)),
           nConnectedModules_buf(allocBufWrapper<uint16_t>(dev, nMod)),

RecoTracker/LSTCore/interface/alpaka/Constants.h

@@ -1,20 +1,13 @@
 #ifndef RecoTracker_LSTCore_interface_alpaka_Constants_h
 #define RecoTracker_LSTCore_interface_alpaka_Constants_h
 
-#include <alpaka/alpaka.hpp>
-
 #include "RecoTracker/LSTCore/interface/Constants.h"
-#include "HeterogeneousCore/AlpakaInterface/interface/config.h"
-
-#ifdef CACHE_ALLOC
-#include "HeterogeneousCore/AlpakaInterface/interface/CachedBufAlloc.h"
-#endif
 
 #ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
 #include <cuda_fp16.h>
 #endif
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
+namespace SDL {
 // Half precision wrapper functions.
 #if defined(FP16_Base)
 #define __F2H __float2half
@@ -26,7 +19,7 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
   typedef float FPX;
 #endif
 
-  alpaka_common::Vec3D const elementsPerThread(alpaka_common::Vec3D::all(static_cast<Idx>(1)));
+  alpaka_common::Vec3D const elementsPerThread(alpaka_common::Vec3D::all(static_cast<alpaka_common::Idx>(1)));
 
 // Needed for files that are compiled by g++ to not throw an error.
 // uint4 is defined only for CUDA, so we will have to revisit this soon when running on other backends.
@@ -39,53 +32,31 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
   };
 #endif
 
-  // Buffer type for allocations where auto type can't be used.
-  template <typename TDev, typename TData>
-  using Buf = alpaka::Buf<TDev, TData, Dim1D, Idx>;
-
-  // Allocation wrapper function to make integration of the caching allocator easier and reduce code boilerplate.
-  template <typename T, typename TAcc, typename TSize, typename TQueue>
-  ALPAKA_FN_HOST ALPAKA_FN_INLINE Buf<alpaka::Dev<TAcc>, T> allocBufWrapper(TAcc const& devAccIn,
-                                                                            TSize nElements,
-                                                                            TQueue queue) {
-#ifdef CACHE_ALLOC
-    return cms::alpakatools::allocCachedBuf<T, Idx>(devAccIn, queue, Vec1D(static_cast<Idx>(nElements)));
-#else
-    return alpaka::allocBuf<T, Idx>(devAccIn, Vec1D(static_cast<Idx>(nElements)));
-#endif
-  }
-
-  // Second allocation wrapper function when queue is not given. Reduces code boilerplate.
-  template <typename T, typename TAcc, typename TSize>
-  ALPAKA_FN_HOST ALPAKA_FN_INLINE Buf<alpaka::Dev<TAcc>, T> allocBufWrapper(TAcc const& devAccIn, TSize nElements) {
-    return alpaka::allocBuf<T, Idx>(devAccIn, Vec1D(static_cast<Idx>(nElements)));
-  }
-
   // Wrapper function to reduce code boilerplate for defining grid/block sizes.
   ALPAKA_FN_HOST ALPAKA_FN_INLINE alpaka_common::Vec3D createVec(int x, int y, int z) {
-    return alpaka_common::Vec3D(static_cast<Idx>(x), static_cast<Idx>(y), static_cast<Idx>(z));
+    return alpaka_common::Vec3D(static_cast<alpaka_common::Idx>(x), static_cast<alpaka_common::Idx>(y), static_cast<alpaka_common::Idx>(z));
   }
 
   // Adjust grid and block sizes based on backend configuration
   template <typename Vec>
-  ALPAKA_FN_HOST ALPAKA_FN_INLINE WorkDiv3D createWorkDiv(const Vec& blocksPerGrid,
+  ALPAKA_FN_HOST ALPAKA_FN_INLINE alpaka_common::WorkDiv3D createWorkDiv(const Vec& blocksPerGrid,
                                                         const Vec& threadsPerBlock,
                                                         const Vec& elementsPerThreadArg) {
     Vec adjustedBlocks = blocksPerGrid;
     Vec adjustedThreads = threadsPerBlock;
 
     // Serial execution, so all launch parameters set to 1.
 #if defined(ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED)
-    adjustedBlocks = Vec::all(static_cast<Idx>(1));
-    adjustedThreads = Vec::all(static_cast<Idx>(1));
+    adjustedBlocks = Vec::all(static_cast<alpaka_common::Idx>(1));
+    adjustedThreads = Vec::all(static_cast<alpaka_common::Idx>(1));
 #endif
 
     // Threads enabled, set number of blocks to 1.
 #if defined(ALPAKA_ACC_CPU_B_SEQ_T_THREADS_ENABLED)
-    adjustedBlocks = Vec::all(static_cast<Idx>(1));
+    adjustedBlocks = Vec::all(static_cast<alpaka_common::Idx>(1));
 #endif
 
-    return WorkDiv3D(adjustedBlocks, adjustedThreads, elementsPerThreadArg);
+    return alpaka_common::WorkDiv3D(adjustedBlocks, adjustedThreads, elementsPerThreadArg);
   }
 
   // 15 MeV constant from the approximate Bethe-Bloch formula
@@ -100,11 +71,7 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float k2Rinv1GeVf = (2.99792458e-3 * 3.8) / 2;
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float kR1GeVf = 1. / (2.99792458e-3 * 3.8);
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float sinAlphaMax = 0.95;
-#ifdef PT_CUT
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float ptCut = PT_CUT;
-#else
-  ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float ptCut = ::SDL::PT_CUT;
-#endif
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float deltaZLum = 15.0;
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float pixelPSZpitch = 0.15;
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float stripPSZpitch = 2.4;
@@ -117,7 +84,7 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE::SDL {
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float SDL_INF = 123456789.0;
 }  //namespace SDL
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE::T5DNN {
+namespace T5DNN {
   // Working points matching LST fake rate (43.9%) or signal acceptance (82.0%)
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float LSTWP1 = 0.3418833f;  // 94.0% TPR, 43.9% FPR
   ALPAKA_STATIC_ACC_MEM_GLOBAL constexpr float LSTWP2 = 0.6177366f;  // 82.0% TPR, 20.0% FPR