Update alpakaconfig (cms-patatrack#70)

* Use alpaka ENABLED flag instead of PRESENT * Rename alpaka acc namespace to avoid conflict with cmssw * Update platform aliases * Move definition of accelerator namespace * Update namespace names in benchmarking scripts * Update cuda alpakacore * Update hip alpakacore * Update package version
sbaldu · Jan 8, 2025 · e3c5036 · e3c5036
1 parent d8e476d
commit e3c5036
Show file tree

Hide file tree

Showing 14 changed files with 69 additions and 63 deletions.
diff --git a/CLUEstering/alpaka/AlpakaCore/CachedBufAlloc.h b/CLUEstering/alpaka/AlpakaCore/CachedBufAlloc.h
@@ -91,7 +91,7 @@ namespace clue {
         auto deleter = [alloc = &allocator](TElem* ptr) { alloc->free(ptr); };
 
         return alpaka::BufCudaRt<TElem, TDim, TIdx>(
-            dev, reinterpret_cast<TElem*>(memPtr), std::move(deleter), pitchBytes, extent);
+            dev, reinterpret_cast<TElem*>(memPtr), std::move(deleter), extent, pitchBytes);
       }
     };
 
@@ -153,7 +153,7 @@ namespace clue {
         auto deleter = [alloc = &allocator](TElem* ptr) { alloc->free(ptr); };
 
         return alpaka::BufHipRt<TElem, TDim, TIdx>(
-            dev, reinterpret_cast<TElem*>(memPtr), std::move(deleter), pitchBytes, extent);
+            dev, reinterpret_cast<TElem*>(memPtr), std::move(deleter), extent, pitchBytes);
       }
     };
 

diff --git a/CLUEstering/alpaka/AlpakaCore/CachingAllocator.h b/CLUEstering/alpaka/AlpakaCore/CachingAllocator.h
@@ -334,9 +334,10 @@ namespace clue {
         return alpaka::allocBuf<std::byte, size_t>(device_, bytes);
       } else if constexpr (std::is_same_v<Device, alpaka::DevCpu>) {
         // allocate pinned host memory accessible by the queue's platform
-        return alpaka::allocMappedBuf<alpaka::Platform<alpaka::Dev<Queue>>,
-                                      std::byte,
-                                      size_t>(device_, bytes);
+        using Platform = alpaka::Platform<alpaka::Dev<Queue>>;
+        return alpaka::
+            allocMappedBuf<alpaka::Platform<alpaka::Dev<Queue>>, std::byte, size_t>(
+                device_, ::clue::platform<Platform>(), bytes);
       } else {
         // unsupported combination
         static_assert(std::is_same_v<Device, alpaka::Dev<Queue>> or

diff --git a/CLUEstering/alpaka/AlpakaCore/alpakaConfig.h b/CLUEstering/alpaka/AlpakaCore/alpakaConfig.h
@@ -2,6 +2,7 @@
 #pragma once
 
 #include "alpakaFwd.h"
+#include <alpaka/alpaka.hpp>
 
 namespace alpaka_common {
 
@@ -30,15 +31,15 @@ namespace alpaka_common {
 
   // host types
   using DevHost = alpaka::DevCpu;
-  using PltfHost = alpaka::PltfCpu;
+  using PlatformHost = alpaka::PlatformCpu;
 
 }  // namespace alpaka_common
 
-#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
+#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
 namespace alpaka_cuda_async {
   using namespace alpaka_common;
 
-  using Platform = alpaka::PltfCudaRt;
+  using Platform = alpaka::PlatformCudaRt;
   using Device = alpaka::DevCudaRt;
   using Queue = alpaka::QueueCudaRtNonBlocking;
   using Event = alpaka::EventCudaRt;
@@ -49,19 +50,16 @@ namespace alpaka_cuda_async {
   using Acc2D = Acc<Dim2D>;
   using Acc3D = Acc<Dim3D>;
 
+#define ALPAKA_ACCELERATOR_NAMESPACE_CLUE alpaka_cuda_async
 }  // namespace alpaka_cuda_async
 
-#endif  // ALPAKA_ACC_GPU_CUDA_PRESENT
-
-#ifdef ALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND
-#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_cuda_async
 #endif  // ALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
+#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
 namespace alpaka_rocm_async {
   using namespace alpaka_common;
 
-  using Platform = alpaka::PltfHipRt;
+  using Platform = alpaka::PlatformHipRt;
   using Device = alpaka::DevHipRt;
   using Queue = alpaka::QueueHipRtNonBlocking;
   using Event = alpaka::EventHipRt;
@@ -72,19 +70,16 @@ namespace alpaka_rocm_async {
   using Acc2D = Acc<Dim2D>;
   using Acc3D = Acc<Dim3D>;
 
+#define ALPAKA_ACCELERATOR_NAMESPACE_CLUE alpaka_rocm_async
 }  // namespace alpaka_rocm_async
 
-#endif  // ALPAKA_ACC_GPU_HIP_PRESENT
-
-#ifdef ALPAKA_ACC_GPU_HIP_ASYNC_BACKEND
-#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_rocm_async
-#endif  // ALPAKA_ACC_GPU_HIP_ASYNC_BACKEND
+#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
 
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
 namespace alpaka_serial_sync {
   using namespace alpaka_common;
 
-  using Platform = alpaka::PltfCpu;
+  using Platform = alpaka::PlatformCpu;
   using Device = alpaka::DevCpu;
   using Queue = alpaka::QueueCpuBlocking;
   using Event = alpaka::EventCpu;
@@ -95,19 +90,16 @@ namespace alpaka_serial_sync {
   using Acc2D = Acc<Dim2D>;
   using Acc3D = Acc<Dim3D>;
 
+#define ALPAKA_ACCELERATOR_NAMESPACE_CLUE alpaka_serial_sync
 }  // namespace alpaka_serial_sync
 
-#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
-
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
-#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_serial_sync
-#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
+#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
 
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
 namespace alpaka_tbb_async {
   using namespace alpaka_common;
 
-  using Platform = alpaka::PltfCpu;
+  using Platform = alpaka::PlatformCpu;
   using Device = alpaka::DevCpu;
   using Queue = alpaka::QueueCpuNonBlocking;
   using Event = alpaka::EventCpu;
@@ -118,19 +110,16 @@ namespace alpaka_tbb_async {
   using Acc2D = Acc<Dim2D>;
   using Acc3D = Acc<Dim3D>;
 
+#define ALPAKA_ACCELERATOR_NAMESPACE_CLUE alpaka_tbb_async
 }  // namespace alpaka_tbb_async
 
-#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
+#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
 
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND
-#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_tbb_async
-#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND
-
-#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_PRESENT
+#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLED
 namespace alpaka_omp2_async {
   using namespace alpaka_common;
 
-  using Platform = alpaka::PltfCpu;
+  using Platform = alpaka::PlatformCpu;
   using Device = alpaka::DevCpu;
   using Queue = alpaka::QueueCpuBlocking;
   using Event = alpaka::EventCpu;
@@ -141,10 +130,7 @@ namespace alpaka_omp2_async {
   using Acc2D = Acc<Dim2D>;
   using Acc3D = Acc<Dim3D>;
 
+#define ALPAKA_ACCELERATOR_NAMESPACE_CLUE alpaka_omp2_async
 }  // namespace alpaka_omp2_async
 
-#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_PRESENT
-
-#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ASYNC_BACKEND
-#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_omp2_async
-#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ASYNC_BACKEND
+#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLED
diff --git a/CLUEstering/alpaka/AlpakaCore/alpakaDevices.h b/CLUEstering/alpaka/AlpakaCore/alpakaDevices.h
@@ -10,14 +10,29 @@
 #include "getDeviceIndex.h"
 
 namespace clue {
+  // returns the alpaka accelerator platform
+  template <typename TPlatform, typename = std::enable_if_t<alpaka::isPlatform<TPlatform>>>
+  inline TPlatform const& platform() {
+    // initialise the platform the first time that this function is called
+    static const auto platform = TPlatform{};
+    return platform;
+  }
+
+  // return the alpaka accelerator devices for the given platform
+  template <typename TPlatform, typename = std::enable_if_t<alpaka::isPlatform<TPlatform>>>
+  inline std::vector<alpaka::Dev<TPlatform>> const& devices() {
+    // enumerate all devices the first time that this function is called
+    static const auto devices = alpaka::getDevs(platform<TPlatform>());
+    return devices;
+  }
 
   // alpaka host device
   inline const alpaka_common::DevHost host =
       alpaka::getDevByIdx(alpaka::PlatformCpu{}, 0u);
 
   // alpaka accelerator devices
-  template <typename TPlatform>
-  inline std::vector<alpaka::Dev<TPlatform>> devices;
+  /* template <typename TPlatform> */
+  /* inline std::vector<alpaka::Dev<TPlatform>> devices; */
 
   template <typename TPlatform>
   std::vector<alpaka::Dev<TPlatform>> enumerate() {
@@ -27,10 +42,10 @@ namespace clue {
     using Platform = TPlatform;
 
     std::vector<Device> devices;
-    uint32_t n = alpaka::getDevCount<Platform>();
+    uint32_t n = alpaka::getDevCount(Platform{});
     devices.reserve(n);
     for (uint32_t i = 0; i < n; ++i) {
-      devices.push_back(alpaka::getDevByIdx<Platform>(i));
+      devices.push_back(alpaka::getDevByIdx(Platform{}, i));
       assert(getDeviceIndex(devices.back()) == static_cast<int>(i));
     }
     return devices;

diff --git a/CLUEstering/alpaka/AlpakaCore/getDeviceCachingAllocator.h b/CLUEstering/alpaka/AlpakaCore/getDeviceCachingAllocator.h
@@ -57,7 +57,7 @@ namespace clue {
 
     size_t const index = getDeviceIndex(device);
 
-    std::vector<TDevice> devs = alpaka::getDevs<alpaka::Platform<TDevice>>();
+    std::vector<TDevice> devs = alpaka::getDevs(alpaka::Platform<TDevice>{});
 
     assert(index < clue::enumerate<alpaka::Platform<TDevice>>().size());
 

diff --git a/CLUEstering/alpaka/CLUE/CLUEAlgoAlpaka.h b/CLUEstering/alpaka/CLUE/CLUEAlgoAlpaka.h
@@ -22,7 +22,7 @@
 
 using clue::VecArray;
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
 
   template <uint8_t Ndim>
   class CLUEAlgoAlpaka {
@@ -247,4 +247,4 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
     return {h_points.m_clusterIndex, h_points.m_isSeed};
   }
-}  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+}  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
diff --git a/CLUEstering/alpaka/CLUE/CLUEAlpakaKernels.h b/CLUEstering/alpaka/CLUE/CLUEAlpakaKernels.h
@@ -13,7 +13,7 @@
 
 using clue::VecArray;
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
 
   constexpr int32_t max_followers{100};
   constexpr int32_t reserve{1000000};
@@ -349,4 +349,4 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
       });
     }
   };
-}  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+}  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
diff --git a/CLUEstering/alpaka/CLUE/Run.h b/CLUEstering/alpaka/CLUE/Run.h
@@ -5,7 +5,7 @@
 
 #include "CLUEAlgoAlpaka.h"
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
 
   template <uint8_t Ndim, typename Kernel>
   std::vector<std::vector<int>> run(float dc,
@@ -26,4 +26,4 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
     return algo.make_clusters(h_points, d_points, kernel, queue_, block_size);
   }
 
-};  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+};  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
diff --git a/CLUEstering/alpaka/DataFormats/alpaka/AlpakaVecArray.h b/CLUEstering/alpaka/DataFormats/alpaka/AlpakaVecArray.h
@@ -5,6 +5,8 @@
 // Author: Felice Pantaleo, CERN
 //
 
+#include <alpaka/alpaka.hpp>
+
 namespace clue {
 
   template <class T, int maxSize>

diff --git a/CLUEstering/alpaka/DataFormats/alpaka/PointsAlpaka.h b/CLUEstering/alpaka/DataFormats/alpaka/PointsAlpaka.h
@@ -11,7 +11,7 @@
 
 using clue::VecArray;
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
 
   template <uint8_t Ndim>
   class PointsAlpaka {
@@ -71,4 +71,4 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
   private:
     clue::device_buffer<Device, PointsAlpakaView> view_dev;
   };
-}  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+}  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
diff --git a/CLUEstering/alpaka/DataFormats/alpaka/TilesAlpaka.h b/CLUEstering/alpaka/DataFormats/alpaka/TilesAlpaka.h
@@ -17,7 +17,7 @@ using clue::VecArray;
 constexpr uint32_t max_tile_depth{1 << 10};
 constexpr uint32_t max_n_tiles{1 << 15};
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
 
   template <uint8_t Ndim>
   class CoordinateExtremes {
@@ -136,4 +136,4 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
     float tile_size[Ndim];
     VecArray<VecArray<uint32_t, max_tile_depth>, max_n_tiles> m_tiles;
   };
-}  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+}  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
diff --git a/benchmark/dataset_size/main.cc b/benchmark/dataset_size/main.cc
@@ -83,7 +83,7 @@ void to_csv(const TimeMeasures& measures, const std::string& filename) {
   file.close();
 }
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
   void run(const std::string& input_file) {
     const auto dev_acc = alpaka::getDevByIdx(alpaka::Platform<Acc1D>{}, 0u);
     Queue queue_(dev_acc);
@@ -99,9 +99,9 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
     auto result =
         algo.make_clusters(h_points, d_points, FlatKernel{.5f}, queue_, block_size);
   }
-};  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+};  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
 
-namespace cluetest = ALPAKA_ACCELERATOR_NAMESPACE;
+namespace cluetest = ALPAKA_ACCELERATOR_NAMESPACE_CLUE;
 
 int main(int argc, char* argv[]) {
   auto min = std::stoi(argv[1]);

diff --git a/benchmark/profiling/main.cc b/benchmark/profiling/main.cc
@@ -11,7 +11,7 @@
 
 #include "read_csv.hpp"
 
-namespace ALPAKA_ACCELERATOR_NAMESPACE {
+namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE {
   void run(const std::string& input_file) {
     const auto dev_acc = alpaka::getDevByIdx(alpaka::Platform<Acc1D>{}, 0u);
     Queue queue_(dev_acc);
@@ -24,11 +24,13 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
     CLUEAlgoAlpaka<2> algo(dc, rhoc, outlier, pPBin, queue_);
 
     const std::size_t block_size{256};
-	auto result = algo.make_clusters(h_points, d_points, FlatKernel{.5f}, queue_, block_size);
+    auto result =
+        algo.make_clusters(h_points, d_points, FlatKernel{.5f}, queue_, block_size);
   }
-};  // namespace ALPAKA_ACCELERATOR_NAMESPACE
+};  // namespace ALPAKA_ACCELERATOR_NAMESPACE_CLUE
 
 int main(int argc, char* argv[]) {
   auto input_file{std::string(argv[1])};
-  ALPAKA_ACCELERATOR_NAMESPACE::run(input_file);
+  using ALPAKA_ACCELERATOR_NAMESPACE_CLUE;
+  run(input_file);
 }
diff --git a/setup.py b/setup.py
@@ -4,7 +4,7 @@
 from setuptools import setup
 import subprocess
 
-__version__ = "2.3.0"
+__version__ = "2.3.1"
 
 this_directory = Path(__file__).parent
 long_description = (this_directory/'README.md').read_text()