make code checks happy

cms-sw · May 10, 2022 · 9402cb7 · 9402cb7
1 parent 521d4c0
commit 9402cb7
Show file tree

Hide file tree

Showing 4 changed files with 48 additions and 52 deletions.
diff --git a/CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDA.h b/CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDA.h
@@ -10,7 +10,7 @@ class SiPixelDigisCUDA {
 public:
   using StoreType = uint16_t;
   SiPixelDigisCUDA() = default;
-  /*inline*/ SiPixelDigisCUDA(size_t maxFedWords, cudaStream_t stream);
+  SiPixelDigisCUDA(size_t maxFedWords, cudaStream_t stream);
   ~SiPixelDigisCUDA() = default;
 
   SiPixelDigisCUDA(const SiPixelDigisCUDA &) = delete;
@@ -41,6 +41,4 @@ class SiPixelDigisCUDA {
   uint32_t nDigis_h = 0;
 };
 
-// #include "SiPixelDigisCUDAImpl.h"
-
 #endif  // CUDADataFormats_SiPixelDigi_interface_SiPixelDigisCUDA_h
diff --git a/CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDAImpl.h b/CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDAImpl.h
diff --git a/CUDADataFormats/SiPixelDigi/src/SiPixelDigisCUDA.cc b/CUDADataFormats/SiPixelDigi/src/SiPixelDigisCUDA.cc
@@ -1,2 +1,29 @@
+#include <cassert>
+
 #include "CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDA.h"
-#include "CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDAImpl.h"
+#include "HeterogeneousCore/CUDAUtilities/interface/cudaMemoryPool.h"
+#include "HeterogeneousCore/CUDAUtilities/interface/cudaCheck.h"
+
+SiPixelDigisCUDA::SiPixelDigisCUDA(size_t maxFedWords, cudaStream_t stream)
+    : m_store(memoryPool::cuda::makeBuffer<SiPixelDigisCUDA::StoreType>(
+          SiPixelDigisCUDASOAView::roundFor128ByteAlignment(maxFedWords) *
+              static_cast<int>(SiPixelDigisCUDASOAView::StorageLocation::kMAX),
+          stream,
+          memoryPool::onDevice)),
+      m_view(m_store, maxFedWords, SiPixelDigisCUDASOAView::StorageLocation::kMAX) {
+  assert(maxFedWords != 0);
+}
+
+memoryPool::Buffer<SiPixelDigisCUDA::StoreType> SiPixelDigisCUDA::copyAllToHostAsync(cudaStream_t stream) const {
+  auto ret = memoryPool::cuda::makeBuffer<StoreType>(
+      m_view.roundFor128ByteAlignment(nDigis()) * static_cast<int>(SiPixelDigisCUDASOAView::StorageLocationHost::kMAX),
+      stream,
+      memoryPool::onHost);
+  cudaCheck(cudaMemcpyAsync(ret.get(),
+                            m_view.clus(),
+                            m_view.roundFor128ByteAlignment(nDigis()) * sizeof(SiPixelDigisCUDA::StoreType) *
+                                static_cast<int>(SiPixelDigisCUDASOAView::StorageLocationHost::kMAX),
+                            cudaMemcpyDeviceToHost,
+                            stream));
+  return ret;
+}
diff --git a/CUDADataFormats/TrackingRecHit/interface/TrackingRecHit2DHeterogeneous.h b/CUDADataFormats/TrackingRecHit/interface/TrackingRecHit2DHeterogeneous.h
@@ -36,22 +36,22 @@ class TrackingRecHit2DHeterogeneous {
 
   TrackingRecHit2DHeterogeneous() = default;
 
-  /*inline*/ TrackingRecHit2DHeterogeneous(uint32_t nHits,
-                                           bool isPhase2,
-                                           int32_t offsetBPIX2,
-                                           pixelCPEforGPU::ParamsOnGPU const* cpeParams,
-                                           uint32_t const* hitsModuleStart,
-                                           memoryPool::Where where,
-                                           cudaStream_t stream,
-                                           TrackingRecHit2DHeterogeneous const* input = nullptr);
+  TrackingRecHit2DHeterogeneous(uint32_t nHits,
+                                bool isPhase2,
+                                int32_t offsetBPIX2,
+                                pixelCPEforGPU::ParamsOnGPU const* cpeParams,
+                                uint32_t const* hitsModuleStart,
+                                memoryPool::Where where,
+                                cudaStream_t stream,
+                                TrackingRecHit2DHeterogeneous const* input = nullptr);
 
   // used on CPU only
-  /*inline*/ TrackingRecHit2DHeterogeneous(float* store32,
-                                           uint16_t* store16,
-                                           uint32_t* modules,
-                                           int nHits,
-                                           memoryPool::Where where = memoryPool::onCPU,
-                                           cudaStream_t stream = nullptr);
+  TrackingRecHit2DHeterogeneous(float* store32,
+                                uint16_t* store16,
+                                uint32_t* modules,
+                                int nHits,
+                                memoryPool::Where where = memoryPool::onCPU,
+                                cudaStream_t stream = nullptr);
   ~TrackingRecHit2DHeterogeneous() = default;
 
   TrackingRecHit2DHeterogeneous(const TrackingRecHit2DHeterogeneous&) = delete;
@@ -72,15 +72,15 @@ class TrackingRecHit2DHeterogeneous {
   auto phiBinnerStorage() { return m_phiBinnerStorage; }
   auto iphi() { return m_iphi; }
 
-  /*inline*/ Buffer<float> localCoordToHostAsync(cudaStream_t stream) const;
+  Buffer<float> localCoordToHostAsync(cudaStream_t stream) const;
 
-  /*inline*/ Buffer<uint32_t> hitsModuleStartToHostAsync(cudaStream_t stream) const;
+  Buffer<uint32_t> hitsModuleStartToHostAsync(cudaStream_t stream) const;
 
-  /*inline*/ Buffer<uint16_t> store16ToHostAsync(cudaStream_t stream) const;
-  /*inline*/ Buffer<float> store32ToHostAsync(cudaStream_t stream) const;
+  Buffer<uint16_t> store16ToHostAsync(cudaStream_t stream) const;
+  Buffer<float> store32ToHostAsync(cudaStream_t stream) const;
 
   // needed for Host
-  /*inline*/ void copyFromGPU(TrackingRecHit2DHeterogeneous const* input, cudaStream_t stream);
+  void copyFromGPU(TrackingRecHit2DHeterogeneous const* input, cudaStream_t stream);
 
 private:
   static constexpr uint32_t n16 = 4;                 // number of elements in m_store16