rocm-systems/projects/hip-tests/catch/unit/memory/hipArrayCreate.cc

/*
Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/

/*
hipArrayCreate API test scenarios
1. Negative Scenarios
2. Allocating Small and big chunk data
3. Multithreaded scenario
*/

#include <array>
#include <numeric>
#include <hip_test_common.hh>
#include <hip_array_common.hh>
#include "hipArrayCommon.hh"
#include "DriverContext.hh"

static constexpr size_t NUM_W{4};
static constexpr size_t BIGNUM_W{100};
static constexpr size_t NUM_H{4};
static constexpr size_t BIGNUM_H{100};
static constexpr auto ARRAY_LOOP{100};

/*
 * This API verifies  memory allocations for small and
 * bigger chunks of data.
 * Two scenarios are verified in this API
 * 1. SmallArray: Allocates NUM_W*NUM_H in a loop and
 *    releases the memory.
 * 2. BigArray: Allocates BIGNUM_W*BIGNUM_H in a loop and
 *    releases the memory.
 *
 */

static void ArrayCreate_DiffSizes(int gpu) {
  HIP_CHECK_THREAD(hipSetDevice(gpu));
  // Use of GENERATE in thead function causes random failures with multithread condition.
  std::vector<std::pair<size_t, size_t>> runs{std::make_pair(NUM_W, NUM_H),
                                              std::make_pair(BIGNUM_W, BIGNUM_H)};
  for (const auto& size : runs) {
    std::array<HIP_ARRAY, ARRAY_LOOP> array;
    size_t pavail;
    HIP_CHECK_THREAD(hipMemGetInfo(&pavail, nullptr));
    HIP_ARRAY_DESCRIPTOR desc;
    desc.NumChannels = 1;
    desc.Width = std::get<0>(size);
    desc.Height = std::get<1>(size);
    desc.Format = HIP_AD_FORMAT_FLOAT;

    for (int i = 0; i < ARRAY_LOOP; i++) {
      HIP_CHECK_THREAD(hipArrayCreate(&array[i], &desc));
    }
    for (int i = 0; i < ARRAY_LOOP; i++) {
      HIP_CHECK_THREAD(hipArrayDestroy(array[i]));
    }
  }
}

/* This testcase verifies hipArrayCreate API for small and big chunks data*/
TEST_CASE("Unit_hipArrayCreate_DiffSizes") {
  CHECK_IMAGE_SUPPORT

  ArrayCreate_DiffSizes(0);
  HIP_CHECK_THREAD_FINALIZE();
}

/*
This testcase verifies the hipArrayCreate API in multithreaded
scenario by launching threads in parallel on multiple GPUs
and verifies the hipArrayCreate API with small and big chunks data
*/
TEST_CASE("Unit_hipArrayCreate_MultiThread", "[multigpu]") {
  CHECK_IMAGE_SUPPORT

  std::vector<std::thread> threadlist;
  int devCnt = 0;

  devCnt = HipTest::getDeviceCount();

  for (int i = 0; i < devCnt; i++) {
    threadlist.push_back(std::thread(ArrayCreate_DiffSizes, i));
  }

  for (auto& t : threadlist) {
    t.join();
  }
  HIP_CHECK_THREAD_FINALIZE();
}


// Tests /////////////////////////////////////////

#if HT_AMD
constexpr auto NORMALIZED_COORDINATES = HIP_TRSF_NORMALIZED_COORDINATES;
constexpr auto READ_AS_INTEGER = HIP_TRSF_READ_AS_INTEGER;
#else
// (EXSWCPHIPT-92) HIP equivalents not defined for CUDA backend.
constexpr auto NORMALIZED_COORDINATES = CU_TRSF_NORMALIZED_COORDINATES;
constexpr auto READ_AS_INTEGER = CU_TRSF_READ_AS_INTEGER;
#endif

// Copy data from host to the hipArray_t, accounting 1D or 2D arrays
template <typename T>
void copyToArray(hipArray_t dst, const std::vector<T>& src, const size_t height) {
  const auto sizeInBytes = src.size() * sizeof(T);
  if (height == 0) {
    // FIXME(EXSWCPHIPT-64) remove cast when API is fixed (will require major version change)
    HIP_CHECK(hipMemcpyHtoA(reinterpret_cast<hipArray_t>(dst), 0, src.data(), sizeInBytes));
  } else {
    const auto pitch = sizeInBytes / height;
    hip_Memcpy2D copyParams{};
    copyParams.srcMemoryType = hipMemoryTypeHost;
    copyParams.srcXInBytes = 0;  // x offset
    copyParams.srcY = 0;         // y offset
    copyParams.srcHost = src.data();
    copyParams.srcPitch = pitch;

    copyParams.dstMemoryType = hipMemoryTypeArray;
    copyParams.dstXInBytes = 0;  // x offset
    copyParams.dstY = 0;         // y offset
    copyParams.dstArray = dst;

    copyParams.WidthInBytes = pitch;
    copyParams.Height = height;

    HIP_CHECK(hipMemcpyParam2D(&copyParams));
  }
}

// Test the allocated array by generating a texture from it then reading from that texture.
// Textures are read-only, so write to the array then copy that into normal device memory.
template <typename T>
void testArrayAsTexture(hipArray_t array, const size_t width, const size_t height) {
  using vec_info = vector_info<T>;
  using scalar_type = typename vec_info::type;
  const auto h = height ? height : 1;
  const auto size = sizeof(T) * width * h;

  // set hip array
  std::vector<scalar_type> hostData(width * h * vec_info::size);
  // assigned ascending values to the data array to show indexing is working
  std::iota(std::begin(hostData), std::end(hostData), 0);

  copyToArray(array, hostData, height);

  // create texture
  hipTextureObject_t textObj{};

  HIP_RESOURCE_DESC resDesc{};
  memset(&resDesc, 0, sizeof(HIP_RESOURCE_DESC));
  resDesc.resType = HIP_RESOURCE_TYPE_ARRAY;
  resDesc.res.array.hArray = array;
  resDesc.flags = 0;

  HIP_TEXTURE_DESC texDesc{};
  memset(&texDesc, 0, sizeof(HIP_TEXTURE_DESC));
  // use the actual values in the texture, not normalized data
  texDesc.filterMode = HIP_TR_FILTER_MODE_POINT;
  // Use normalized coordinates and also read the data in the original data type
  texDesc.flags |= NORMALIZED_COORDINATES | READ_AS_INTEGER;

  HIP_CHECK(hipTexObjectCreate(&textObj, &resDesc, &texDesc, nullptr));

  // run kernel
  T* device_data{};
  HIP_CHECK(hipMalloc(&device_data, size));
  readFromTexture<<<dim3(width / BlockSize, height ? height / BlockSize : 1, 1),
                    dim3(BlockSize, height ? BlockSize : 1, 1)>>>(device_data, textObj, width,
                                                                  height, false);
  HIP_CHECK(hipGetLastError());  // check for errors when running the kernel

  // copy data back and then test it
  std::fill(std::begin(hostData), std::end(hostData), 0);
  HIP_CHECK(hipMemcpy(hostData.data(), device_data, size, hipMemcpyDeviceToHost));

#if !__HIP_NO_IMAGE_SUPPORT
  checkDataIsAscending(hostData);
#endif

  // clean up
  HIP_CHECK(hipTexObjectDestroy(textObj));
  HIP_CHECK(hipFree(device_data));
}

// Selection of types chosen since trying all types would be slow to compile
// Test the happy path of the hipArrayCreate
TEMPLATE_TEST_CASE("Unit_hipArrayCreate_happy", "", uint, int, int4, ushort, short2, char, uchar2,
                   char4, float, float2, float4) {
  CHECK_IMAGE_SUPPORT

  using vec_info = vector_info<TestType>;
  DriverContext ctx;

  HIP_ARRAY_DESCRIPTOR desc;
  desc.Format = vec_info::format;
  desc.NumChannels = vec_info::size;
  desc.Width = 1024;
  desc.Height = GENERATE(0, 1024);

  // pointer to the array in device memory
  hipArray_t array{};

  HIP_CHECK(hipArrayCreate(&array, &desc));

  testArrayAsTexture<TestType>(array, desc.Width, desc.Height);

  HIP_CHECK(hipArrayDestroy(array));
}


// Only widths and Heights up to the maxTexture size is supported
TEMPLATE_TEST_CASE("Unit_hipArrayCreate_maxTexture", "", uint, int, int4, ushort, short2, char,
                   uchar2, char4, float, float2, float4) {
  CHECK_IMAGE_SUPPORT

  using vec_info = vector_info<TestType>;
  DriverContext ctx;

  HIP_ARRAY_DESCRIPTOR desc;
  desc.Format = vec_info::format;
  desc.NumChannels = vec_info::size;

  const Sizes sizes(hipArrayDefault);
  const size_t s = 64;

  hipArray_t array{};
  SECTION("Happy") {
    SECTION("1D - Max") {
      desc.Width = sizes.max1D;
      desc.Height = 0;
    }
    SECTION("2D - Max Width") {
      desc.Width = sizes.max2D[0];
      desc.Height = s;
    }
    SECTION("2D - Max Height") {
      desc.Width = s;
      desc.Height = sizes.max2D[1];
    }
    SECTION("2D - Max Width and Height") {
      desc.Width = sizes.max2D[0];
      desc.Height = sizes.max2D[1];
    }
    auto maxArrayCreateError = hipArrayCreate(&array, &desc);
    // this can try to alloc many GB of memory, so out of memory is acceptable
    // return to avoid destroy
    if (maxArrayCreateError == hipErrorOutOfMemory) return;
    HIP_CHECK(maxArrayCreateError);
    HIP_CHECK(hipArrayDestroy(array));
  }
  SECTION("Negative") {
    SECTION("1D - More Than Max") {
      desc.Width = sizes.max1D + 1;
      desc.Height = 0;
    }
    SECTION("2D - More Than Max Width") {
      desc.Width = sizes.max2D[0] + 1;
      desc.Height = s;
    }
    SECTION("2D - More Than Max Height") {
      desc.Width = s;
      desc.Height = sizes.max2D[1] + 1;
    }
    SECTION("2D - More Than Max Width and Height") {
      desc.Width = sizes.max2D[0] + 1;
      desc.Height = sizes.max2D[1] + 1;
    }
    HIP_CHECK_ERROR(hipArrayCreate(&array, &desc), hipErrorInvalidValue);
  }
}

// zero-width array is not supported
TEST_CASE("Unit_hipArrayCreate_ZeroWidth") {
  CHECK_IMAGE_SUPPORT

  DriverContext ctx;
  HIP_ARRAY_DESCRIPTOR desc;
  desc.Format = driverFormats[0];
  desc.NumChannels = 4;
  desc.Width = 0;
  desc.Height = GENERATE(0, 1024);

  // pointer to the array in device memory
  hipArray_t array;
  HIP_CHECK_ERROR(hipArrayCreate(&array, &desc), hipErrorInvalidValue);
}

// HipArrayCreate will return an error when nullptr is used as the array argument
TEST_CASE("Unit_hipArrayCreate_Nullptr") {
  CHECK_IMAGE_SUPPORT

  DriverContext ctx;
  SECTION("Null array") {
    HIP_ARRAY_DESCRIPTOR desc;
    desc.Format = driverFormats[0];
    desc.NumChannels = 4;
    desc.Width = 1024;
    desc.Height = 1024;

    HIP_CHECK_ERROR(hipArrayCreate(nullptr, &desc), hipErrorInvalidValue);
  }
  SECTION("Null Description") {
    hipArray_t array;
    HIP_CHECK_ERROR(hipArrayCreate(&array, nullptr), hipErrorInvalidValue);
  }
}

// Only elements with 1,2, or 4 channels is supported
TEST_CASE("Unit_hipArrayCreate_BadNumberChannelElement") {
  CHECK_IMAGE_SUPPORT

  DriverContext ctx;
  HIP_ARRAY_DESCRIPTOR desc;
  desc.Format = GENERATE(from_range(std::begin(driverFormats), std::end(driverFormats)));
  desc.NumChannels = GENERATE(-1, 0, 3, 5, 8);
  desc.Width = 1024;
  desc.Height = GENERATE(0, 1024);

  hipArray_t array;

  INFO("Format: " << formatToString(desc.Format) << " NumChannels: " << desc.NumChannels
                  << " Height: " << desc.Height)
  HIP_CHECK_ERROR(hipArrayCreate(&array, &desc), hipErrorInvalidValue);
}

// Only certain channel formats are acceptable.
TEST_CASE("Unit_hipArrayCreate_BadChannelFormat") {
  CHECK_IMAGE_SUPPORT

  DriverContext ctx;
  HIP_ARRAY_DESCRIPTOR desc;

  // create a bad format
  desc.Format =
      std::accumulate(std::begin(driverFormats), std::end(driverFormats), driverFormats[0],
                      [](auto i, auto f) { return static_cast<decltype(desc.Format)>(i + f); });
  for (auto&& format : driverFormats) {
    REQUIRE(desc.Format != format);
  }

  desc.NumChannels = 4;
  desc.Width = 1024;
  desc.Height = GENERATE(0, 1024);

  hipArray_t array;

  INFO("Format: " << formatToString(desc.Format) << " Height: " << desc.Height)
  HIP_CHECK_ERROR(hipArrayCreate(&array, &desc), hipErrorInvalidValue);
}