/*
 * Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 *
 * SPDX-License-Identifier: MIT
 */

#include "memcpy_performance_common.hh"

/**
 * @addtogroup memcpy memcpy
 * @{
 * @ingroup PerformanceTestMemory
 */

class Memcpy2DToArrayBenchmark : public Benchmark<Memcpy2DToArrayBenchmark> {
 public:
  void operator()(hipArray_t dst, const void* src, size_t src_pitch, size_t width, size_t height,
                  hipMemcpyKind kind) {
    TIMED_SECTION(kTimerTypeCpu) {
      HIP_CHECK(hipMemcpy2DToArray(dst, 0, 0, src, src_pitch, width, height, kind));
    }
  }
};

static void RunBenchmark(size_t width, size_t height, hipMemcpyKind kind,
                         bool enable_peer_access = false) {
  Memcpy2DToArrayBenchmark benchmark;
  benchmark.AddSectionName("(" + std::to_string(width) + ", " + std::to_string(height) + ")");

  if (kind == hipMemcpyHostToDevice) {
    size_t allocation_size = width * height * sizeof(int);
    LinearAllocGuard<int> host_allocation(LinearAllocs::hipHostMalloc, allocation_size);
    ArrayAllocGuard<int> array_allocation(make_hipExtent(width, height, 0), hipArrayDefault);
    benchmark.Run(array_allocation.ptr(), host_allocation.ptr(), width * sizeof(int),
                  width * sizeof(int), height, hipMemcpyHostToDevice);
  } else {
    // hipMemcpyDeviceToDevice
    int src_device = std::get<0>(GetDeviceIds(enable_peer_access));
    int dst_device = std::get<1>(GetDeviceIds(enable_peer_access));

    LinearAllocGuard2D<int> device_allocation(width, height);
    HIP_CHECK(hipSetDevice(dst_device));
    ArrayAllocGuard<int> array_allocation(make_hipExtent(width, height, 0), hipArrayDefault);
    HIP_CHECK(hipSetDevice(src_device));
    benchmark.Run(array_allocation.ptr(), device_allocation.ptr(), device_allocation.pitch(),
                  device_allocation.width(), device_allocation.height(), hipMemcpyDeviceToDevice);
  }
}

/**
 * Test Description
 * ------------------------
 *  - Executes `hipMemcpy2DToArray` from Host to Device:
 *    -# Allocation size
 *      - Small: 4 KB x 32 B
 *      - Medium: 8 KB x 32 B
 *      - Large: 16 KB x 32 B
 * Test source
 * ------------------------
 * - performance/api/memcpy/hipMemcpy2DToArray.cc
 * Test requirements
 * ------------------------
 *  - HIP_VERSION >= 5.2
 */
HIP_TEST_CASE(Performance_hipMemcpy2DToArray_HostToDevice) {
  CHECK_IMAGE_SUPPORT

  const auto width = GENERATE(4_KB, 8_KB, 16_KB);
  RunBenchmark(width, 32, hipMemcpyHostToDevice);
}

/**
 * Test Description
 * ------------------------
 *  - Executes `hipMemcpy2DToArray` from Device to Device with peer access disabled:
 *    -# Allocation size
 *      - Small: 4 KB x 32 B
 *      - Medium: 8 KB x 32 B
 *      - Large: 16 KB x 32 B
 * Test source
 * ------------------------
 * - performance/api/memcpy/hipMemcpy2DToArray.cc
 * Test requirements
 * ------------------------
 *  - HIP_VERSION >= 5.2
 */
HIP_TEST_CASE(Performance_hipMemcpy2DToArray_DeviceToDevice_DisablePeerAccess) {
  CHECK_IMAGE_SUPPORT

  const auto width = GENERATE(4_KB, 8_KB, 16_KB);
  RunBenchmark(width, 32, hipMemcpyDeviceToDevice);
}

/**
 * Test Description
 * ------------------------
 *  - Executes `hipMemcpy2DToArray` from Device to Device with peer access enabled:
 *    -# Allocation size
 *      - Small: 4 KB x 32 B
 *      - Medium: 8 KB x 32 B
 *      - Large: 16 KB x 32 B
 * Test source
 * ------------------------
 * - performance/api/memcpy/hipMemcpy2DToArray.cc
 * Test requirements
 * ------------------------
 *  - Multi-device
 *  - Device supports Peer-to-Peer access
 *  - HIP_VERSION >= 5.2
 */
HIP_TEST_CASE(Performance_hipMemcpy2DToArray_DeviceToDevice_EnablePeerAccess) {
  CHECK_IMAGE_SUPPORT

  if (HipTest::getDeviceCount() < 2) {
    HIP_SKIP_TEST(HipTest::SkipReason::kFewerThanTwoGpus);
  }
  const auto width = GENERATE(4_KB, 8_KB, 16_KB);
  RunBenchmark(width, 32, hipMemcpyDeviceToDevice, true);
}

/**
 * End doxygen group memcpy.
 * @}
 */