Introduced benchmarks for HIP.

Simply copying the current CUDA benchmark code, with
all its imperfections.

benchmarks/CMakeLists.txt

@@ -50,6 +50,9 @@ add_subdirectory(core)
 if(VECMEM_BUILD_CUDA_LIBRARY)
    add_subdirectory(cuda)
 endif()
+if(VECMEM_BUILD_HIP_LIBRARY)
+   add_subdirectory(hip)
+endif()
 if(VECMEM_BUILD_SYCL_LIBRARY)
    add_subdirectory(sycl)
 endif()

benchmarks/hip/CMakeLists.txt

@@ -0,0 +1,29 @@
+# VecMem project, part of the ACTS project (R&D line)
+#
+# (c) 2024 CERN for the benefit of the ACTS project
+#
+# Mozilla Public License Version 2.0
+
+# Project include(s).
+include( vecmem-compiler-options-cpp )
+include( vecmem-compiler-options-hip )
+
+# Set up the benchmark(s) for the HIP library.
+add_executable( vecmem_benchmark_hip
+    "benchmark_hip.cpp"
+    "benchmark_copy.cpp"
+    "benchmark_edm_copy.cpp" )
+
+target_link_libraries(
+    vecmem_benchmark_hip
+
+    PRIVATE
+    vecmem::core
+    vecmem::hip
+    vecmem_benchmark_common
+    benchmark::benchmark
+    benchmark::benchmark_main
+)
+
+set_target_properties( vecmem_benchmark_hip PROPERTIES
+   FOLDER "vecmem/benchmarks" )

benchmarks/hip/benchmark_copy.cpp

@@ -0,0 +1,166 @@
+/*
+ * VecMem project, part of the ACTS project (R&D line)
+ *
+ * (c) 2022-2024 CERN for the benefit of the ACTS project
+ *
+ * Mozilla Public License Version 2.0
+ */
+
+// VecMem include(s).
+#include <vecmem/memory/hip/device_memory_resource.hpp>
+#include <vecmem/memory/host_memory_resource.hpp>
+#include <vecmem/utils/hip/copy.hpp>
+
+// Common benchmark include(s).
+#include "../common/make_jagged_sizes.hpp"
+#include "../common/make_jagged_vector.hpp"
+
+// Google benchmark include(s).
+#include <benchmark/benchmark.h>
+
+// System include(s).
+#include <numeric>
+#include <vector>
+
+namespace vecmem::hip::benchmark {
+
+/// The (host) memory resource to use in the benchmark(s).
+static vecmem::host_memory_resource host_mr;
+/// The (device) memory resource to use in the benchmark(s).
+static device_memory_resource device_mr;
+/// The copy object to use in the benchmark(s).
+static copy hip_copy;
+
+/// Function benchmarking "unknown" host-to-device jagged vector copies
+void jaggedVectorUnknownHtoDCopy(::benchmark::State& state) {
+
+    // Generate the sizes of the jagged vector/buffer for the test.
+    const std::vector<std::size_t> sizes =
+        vecmem::benchmark::make_jagged_sizes(state.range(0), state.range(1));
+
+    // Set custom "counters" for the benchmark.
+    const std::size_t bytes = std::accumulate(sizes.begin(), sizes.end(),
+                                              static_cast<std::size_t>(0u)) *
+                              sizeof(int);
+    state.counters["Bytes"] = static_cast<double>(bytes);
+    state.counters["Rate"] =
+        ::benchmark::Counter(static_cast<double>(bytes),
+                             ::benchmark::Counter::kIsIterationInvariantRate,
+                             ::benchmark::Counter::kIs1024);
+
+    // Create the "source vector".
+    jagged_vector<int> source =
+        vecmem::benchmark::make_jagged_vector(sizes, host_mr);
+    const data::jagged_vector_data<int> source_data = get_data(source);
+    // Create the "destination buffer".
+    data::jagged_vector_buffer<int> dest(sizes, device_mr, &host_mr);
+    hip_copy.setup(dest)->wait();
+
+    // Perform the copy benchmark.
+    for (auto _ : state) {
+        hip_copy(source_data, dest)->wait();
+    }
+}
+// Set up the benchmark.
+BENCHMARK(jaggedVectorUnknownHtoDCopy)->Ranges({{10, 100000}, {50, 5000}});
+
+/// Function benchmarking "known" host-to-device jagged vector copies
+void jaggedVectorKnownHtoDCopy(::benchmark::State& state) {
+
+    // Generate the sizes of the jagged vector/buffer for the test.
+    const std::vector<std::size_t> sizes =
+        vecmem::benchmark::make_jagged_sizes(state.range(0), state.range(1));
+
+    // Set custom "counters" for the benchmark.
+    const std::size_t bytes = std::accumulate(sizes.begin(), sizes.end(),
+                                              static_cast<std::size_t>(0u)) *
+                              sizeof(int);
+    state.counters["Bytes"] = static_cast<double>(bytes);
+    state.counters["Rate"] =
+        ::benchmark::Counter(static_cast<double>(bytes),
+                             ::benchmark::Counter::kIsIterationInvariantRate,
+                             ::benchmark::Counter::kIs1024);
+
+    // Create the "source vector".
+    jagged_vector<int> source =
+        vecmem::benchmark::make_jagged_vector(sizes, host_mr);
+    const data::jagged_vector_data<int> source_data = get_data(source);
+    // Create the "destination buffer".
+    data::jagged_vector_buffer<int> dest(sizes, device_mr, &host_mr);
+    hip_copy.setup(dest)->wait();
+
+    // Perform the copy benchmark.
+    for (auto _ : state) {
+        hip_copy(source_data, dest, copy::type::host_to_device)->wait();
+    }
+}
+// Set up the benchmark.
+BENCHMARK(jaggedVectorKnownHtoDCopy)->Ranges({{10, 100000}, {50, 5000}});
+
+/// Function benchmarking "unknown" device-to-host jagged vector copies
+void jaggedVectorUnknownDtoHCopy(::benchmark::State& state) {
+
+    // Generate the sizes of the jagged vector/buffer for the test.
+    const std::vector<std::size_t> sizes =
+        vecmem::benchmark::make_jagged_sizes(state.range(0), state.range(1));
+
+    // Set custom "counters" for the benchmark.
+    const std::size_t bytes = std::accumulate(sizes.begin(), sizes.end(),
+                                              static_cast<std::size_t>(0u)) *
+                              sizeof(int);
+    state.counters["Bytes"] = static_cast<double>(bytes);
+    state.counters["Rate"] =
+        ::benchmark::Counter(static_cast<double>(bytes),
+                             ::benchmark::Counter::kIsIterationInvariantRate,
+                             ::benchmark::Counter::kIs1024);
+
+    // Create the "source buffer".
+    data::jagged_vector_buffer<int> source(sizes, device_mr, &host_mr);
+    hip_copy.setup(source)->wait();
+    // Create the "destination vector".
+    jagged_vector<int> dest =
+        vecmem::benchmark::make_jagged_vector(sizes, host_mr);
+    data::jagged_vector_data<int> dest_data = get_data(dest);
+
+    // Perform the copy benchmark.
+    for (auto _ : state) {
+        hip_copy(source, dest_data)->wait();
+    }
+}
+// Set up the benchmark.
+BENCHMARK(jaggedVectorUnknownDtoHCopy)->Ranges({{10, 100000}, {50, 5000}});
+
+/// Function benchmarking "known" device-to-host jagged vector copies
+void jaggedVectorKnownDtoHCopy(::benchmark::State& state) {
+
+    // Generate the sizes of the jagged vector/buffer for the test.
+    const std::vector<std::size_t> sizes =
+        vecmem::benchmark::make_jagged_sizes(state.range(0), state.range(1));
+
+    // Set custom "counters" for the benchmark.
+    const std::size_t bytes = std::accumulate(sizes.begin(), sizes.end(),
+                                              static_cast<std::size_t>(0u)) *
+                              sizeof(int);
+    state.counters["Bytes"] = static_cast<double>(bytes);
+    state.counters["Rate"] =
+        ::benchmark::Counter(static_cast<double>(bytes),
+                             ::benchmark::Counter::kIsIterationInvariantRate,
+                             ::benchmark::Counter::kIs1024);
+
+    // Create the "source buffer".
+    data::jagged_vector_buffer<int> source(sizes, device_mr, &host_mr);
+    hip_copy.setup(source)->wait();
+    // Create the "destination vector".
+    jagged_vector<int> dest =
+        vecmem::benchmark::make_jagged_vector(sizes, host_mr);
+    data::jagged_vector_data<int> dest_data = get_data(dest);
+
+    // Perform the copy benchmark.
+    for (auto _ : state) {
+        hip_copy(source, dest_data, copy::type::device_to_host)->wait();
+    }
+}
+// Set up the benchmark.
+BENCHMARK(jaggedVectorKnownDtoHCopy)->Ranges({{10, 100000}, {50, 5000}});
+
+}  // namespace vecmem::hip::benchmark

benchmarks/hip/benchmark_edm_copy.cpp

@@ -0,0 +1,81 @@
+/*
+ * VecMem project, part of the ACTS project (R&D line)
+ *
+ * (c) 2024 CERN for the benefit of the ACTS project
+ *
+ * Mozilla Public License Version 2.0
+ */
+
+// VecMem include(s).
+#include <vecmem/memory/hip/device_memory_resource.hpp>
+#include <vecmem/memory/hip/host_memory_resource.hpp>
+#include <vecmem/memory/host_memory_resource.hpp>
+#include <vecmem/utils/copy.hpp>
+#include <vecmem/utils/hip/copy.hpp>
+
+// Common benchmark include(s).
+#include "../common/simple_aos_copy_benchmarks.hpp"
+#include "../common/simple_soa_copy_benchmarks.hpp"
+
+// Google benchmark include(s).
+#include <benchmark/benchmark.h>
+
+namespace vecmem::benchmark {
+
+/// Non-pinned host memory resource to use in the benchmark(s).
+static host_memory_resource host_mr;
+/// Pinned host memory resource to use in the benchmark(s).
+static hip::host_memory_resource hip_host_mr;
+
+/// Device memory resource to use in the benchmark(s).
+static hip::device_memory_resource device_mr;
+
+/// The host copy object to use in the benchmark(s).
+static copy host_copy;
+/// The synchronous device copy object to use in the benchmark(s).
+static hip::copy device_copy;
+
+//
+// Helper macro(s) for setting up all the different benchmarks.
+//
+#define CONFIGURE_BENCHMARK(BM) BM->Range(1UL, 1UL << 26)
+
+#define EDM_COPY_BENCHMARKS(TITLE, HOST_MR, DEVICE_MR, HOST_COPY, DEVICE_COPY) \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(                                     \
+        simple_soa_direct_h2d_copy_benchmark, TITLE##_fixed_buffer, HOST_MR,   \
+        DEVICE_MR, DEVICE_COPY, data::buffer_type::fixed_size));               \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(                                     \
+        simple_soa_direct_h2d_copy_benchmark, TITLE##_resizable_buffer,        \
+        HOST_MR, DEVICE_MR, DEVICE_COPY, data::buffer_type::resizable));       \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(                                     \
+        simple_soa_optimal_h2d_copy_benchmark, TITLE##_fixed_buffer, HOST_MR,  \
+        DEVICE_MR, HOST_COPY, DEVICE_COPY, data::buffer_type::fixed_size));    \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(                                     \
+        simple_soa_optimal_h2d_copy_benchmark, TITLE##_resizable_buffer,       \
+        HOST_MR, DEVICE_MR, HOST_COPY, DEVICE_COPY,                            \
+        data::buffer_type::resizable));                                        \
+    CONFIGURE_BENCHMARK(                                                       \
+        BENCHMARK_CAPTURE(simple_soa_direct_d2h_copy_benchmark, TITLE,         \
+                          HOST_MR, DEVICE_MR, DEVICE_COPY));                   \
+    CONFIGURE_BENCHMARK(                                                       \
+        BENCHMARK_CAPTURE(simple_soa_optimal_d2h_copy_benchmark, TITLE,        \
+                          HOST_MR, DEVICE_MR, HOST_COPY, DEVICE_COPY));        \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(                                     \
+        simple_aos_h2d_copy_benchmark, TITLE##_fixed_buffer, HOST_MR,          \
+        DEVICE_MR, DEVICE_COPY, data::buffer_type::fixed_size));               \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(                                     \
+        simple_aos_h2d_copy_benchmark, TITLE##_resizable_buffer, HOST_MR,      \
+        DEVICE_MR, DEVICE_COPY, data::buffer_type::resizable));                \
+    CONFIGURE_BENCHMARK(BENCHMARK_CAPTURE(simple_aos_d2h_copy_benchmark,       \
+                                          TITLE, HOST_MR, DEVICE_MR,           \
+                                          DEVICE_COPY))
+
+//
+// Set up all the different benchmarks.
+//
+EDM_COPY_BENCHMARKS(hip_pageable_sync, host_mr, device_mr, host_copy,
+                    device_copy);
+EDM_COPY_BENCHMARKS(hip_pinned_sync, hip_host_mr, device_mr, host_copy,
+                    device_copy);
+
+}  // namespace vecmem::benchmark