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

#include "arithmetic_common.hh"
#include "atomicDec_negative_kernels_rtc.hh"

#include <hip_test_common.hh>

/**
 * @addtogroup atomicDec atomicDec
 * @{
 * @ingroup AtomicsTest
 */

/**
 * Test Description
 * ------------------------
 *    - Executes a single kernel on a single device wherein all threads will perform an atomic
 * decrement on a target memory location. Each thread will decrement the memory location,
 * storing the return value into a separate output array slot corresponding to it. Once complete,
 * the output array and target memory is validated to contain all the expected values. Several
 * memory access patterns are tested:
 *      -# All threads decrement a single, compile time deducible, memory location
 *      -# Each thread targets an array containing warp_size elements, using tid % warp_size
 *         for indexing
 *      -# Same as the above, but the elements are spread out by L1 cache line size bytes.
 *
 *    - The test is run for:
 *      - All overloads of atomicDec
 *      - hipMalloc, hipMallocManaged, hipHostMalloc and hipHostRegister allocated memory
 *      - Shared memory
 *      - Several grid and block dimension combinations (only one block is used for shared memory).
 * Test source
 * ------------------------
 *    - unit/atomics/atomicDec.cc
 * Test requirements
 * ------------------------
 *    - HIP_VERSION >= 5.2
 */
HIP_TEMPLATE_TEST_CASE(Unit_atomicDec_Positive, unsigned int) {
  int warp_size = 0;
  HIP_CHECK(hipDeviceGetAttribute(&warp_size, hipDeviceAttributeWarpSize, 0));
  const auto cache_line_size = 128u;

  for (auto current = 0; current < cmd_options.iterations; ++current) {
    DYNAMIC_SECTION("Same address " << current) {
      SingleDeviceSingleKernelTest<TestType, AtomicOperation::kDec>(1, sizeof(TestType));
    }

    DYNAMIC_SECTION("Adjacent addresses " << current) {
      SingleDeviceSingleKernelTest<TestType, AtomicOperation::kDec>(warp_size, sizeof(TestType));
    }

    DYNAMIC_SECTION("Scattered addresses " << current) {
      SingleDeviceSingleKernelTest<TestType, AtomicOperation::kDec>(warp_size, cache_line_size);
    }
  }
}

/**
 * Test Description
 * ------------------------
 *    - Executes a kernel two times concurrently on a single device wherein all threads will perform
 * an atomic decrement on a target memory location. Each thread will decrement the memory
 * location, storing the return value into a separate output array slot corresponding to it. Once
 * complete, the output array and target memory is validated to contain all the expected values.
 * Several memory access patterns are tested:
 *      -# All threads decrement a single, compile time deducible, memory location
 *      -# Each thread targets an array containing warp_size elements, using tid % warp_size
 *         for indexing
 *      -# Same as the above, but the elements are spread out by L1 cache line size bytes.
 *
 *    - The test is run for:
 *      - All overloads of atomicDec
 *      - hipMalloc, hipMallocManaged, hipHostMalloc and hipHostRegister allocated memory
 *      - Several grid and block dimension combinations.
 * Test source
 * ------------------------
 *    - unit/atomics/atomicDec.cc
 * Test requirements
 * ------------------------
 *    - HIP_VERSION >= 5.2
 */
HIP_TEMPLATE_TEST_CASE(Unit_atomicDec_Positive_Multi_Kernel, unsigned int) {
  int warp_size = 0;
  HIP_CHECK(hipDeviceGetAttribute(&warp_size, hipDeviceAttributeWarpSize, 0));
  const auto cache_line_size = 128u;

  for (auto current = 0; current < cmd_options.iterations; ++current) {
    DYNAMIC_SECTION("Same address " << current) {
      SingleDeviceMultipleKernelTest<TestType, AtomicOperation::kDec>(2, 1, sizeof(TestType));
    }

    DYNAMIC_SECTION("Adjacent addresses " << current) {
      SingleDeviceMultipleKernelTest<TestType, AtomicOperation::kDec>(2, warp_size,
                                                                      sizeof(TestType));
    }

    DYNAMIC_SECTION("Scattered addresses " << current) {
      SingleDeviceMultipleKernelTest<TestType, AtomicOperation::kDec>(2, warp_size,
                                                                      cache_line_size);
    }
  }
}

/**
 * Test Description
 * ------------------------
 *    - RTCs kernels that pass combinations of arguments of invalid types for all overloads of
 * atomicDec.
 * Test source
 * ------------------------
 *    - unit/atomics/atomicDec.cc
 * Test requirements
 * ------------------------
 *    - HIP_VERSION >= 5.2
 */
HIP_TEST_CASE(Unit_atomicDec_Negative_Parameters_RTC) {
  hiprtcProgram program{};

  const auto program_source = GENERATE(kAtomicDec_uint);
  HIPRTC_CHECK(
      hiprtcCreateProgram(&program, program_source, "atomicDec_negative.cc", 0, nullptr, nullptr));
  hiprtcResult result{hiprtcCompileProgram(program, 0, nullptr)};

  // Get the compile log and count compiler error messages
  size_t log_size{};
  HIPRTC_CHECK(hiprtcGetProgramLogSize(program, &log_size));
  std::string log(log_size, ' ');
  HIPRTC_CHECK(hiprtcGetProgramLog(program, log.data()));
  int error_count{0};

  int expected_error_count{8};
  std::string error_message{"error:"};

  size_t n_pos = log.find(error_message, 0);
  while (n_pos != std::string::npos) {
    ++error_count;
    n_pos = log.find(error_message, n_pos + 1);
  }

  HIPRTC_CHECK(hiprtcDestroyProgram(&program));
  HIPRTC_CHECK_ERROR(result, HIPRTC_ERROR_COMPILATION);
  REQUIRE(error_count == expected_error_count);
}

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