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

#pragma clang diagnostic ignored "-Wunused-parameter"
#include <hip_test_common.hh>
#include <hip_test_checkers.hh>

#if CUDA_VERSION < CUDA_12000

// Texture reference for 3D texture
texture<float, hipTextureType3D, hipReadModeElementType> texf;
texture<int, hipTextureType3D, hipReadModeElementType> texi;
texture<char, hipTextureType3D, hipReadModeElementType> texc;

template <typename T>
__global__ void simpleKernel3DArray(T* outputData, int width, int height, int depth) {
#if !__HIP_NO_IMAGE_SUPPORT
  for (int i = 0; i < depth; i++) {
    for (int j = 0; j < height; j++) {
      for (int k = 0; k < width; k++) {
        if (std::is_same<T, float>::value)
          outputData[i * width * height + j * width + k] = tex3D(texf, k, j, i);
        else if (std::is_same<T, int>::value)
          outputData[i * width * height + j * width + k] = tex3D(texi, k, j, i);
        else if (std::is_same<T, char>::value)
          outputData[i * width * height + j * width + k] = tex3D(texc, k, j, i);
      }
    }
  }
#endif
}

template <typename T>
static void runSimpleTexture3D_Check(int width, int height, int depth,
                                     texture<T, hipTextureType3D, hipReadModeElementType>* tex) {
  unsigned int size = width * height * depth * sizeof(T);
  T* hData = reinterpret_cast<T*>(malloc(size));
  REQUIRE(hData != nullptr);
  memset(hData, 0, size);

  for (int i = 0; i < depth; i++) {
    for (int j = 0; j < height; j++) {
      for (int k = 0; k < width; k++) {
        hData[i * width * height + j * width + k] = i * width * height + j * width + k;
      }
    }
  }

  // Allocate array and copy image data
  hipChannelFormatDesc channelDesc = hipCreateChannelDesc<T>();
  hipArray_t arr;

  HIP_CHECK(
      hipMalloc3DArray(&arr, &channelDesc, make_hipExtent(width, height, depth), hipArrayDefault));
  hipMemcpy3DParms myparms{};
  myparms.srcPos = make_hipPos(0, 0, 0);
  myparms.dstPos = make_hipPos(0, 0, 0);
  myparms.srcPtr = make_hipPitchedPtr(hData, width * sizeof(T), width, height);
  myparms.dstArray = arr;
  myparms.extent = make_hipExtent(width, height, depth);
  myparms.kind = hipMemcpyHostToDevice;

  HIP_CHECK(hipMemcpy3D(&myparms));

  // set texture parameters
  tex->addressMode[0] = hipAddressModeWrap;
  tex->addressMode[1] = hipAddressModeWrap;
  tex->filterMode = hipFilterModePoint;
  tex->normalized = false;

  // Bind the array to the texture
  HIP_CHECK(hipBindTextureToArray(*tex, arr, channelDesc));

  // Allocate device memory for result
  T* dData = nullptr;
  HIP_CHECK(hipMalloc(&dData, size));
  REQUIRE(dData != nullptr);

  hipLaunchKernelGGL(simpleKernel3DArray, dim3(1, 1, 1), dim3(1, 1, 1), 0, 0, dData, width, height,
                     depth);
  HIP_CHECK(hipGetLastError());
  HIP_CHECK(hipDeviceSynchronize());

  // Allocate mem for the result on host side
  T* hOutputData = reinterpret_cast<T*>(malloc(size));
  REQUIRE(hOutputData != nullptr);
  memset(hOutputData, 0, size);

  // copy result from device to host
  HIP_CHECK(hipMemcpy(hOutputData, dData, size, hipMemcpyDeviceToHost));
  HipTest::checkArray(hData, hOutputData, width, height, depth);

  HIP_CHECK(hipFree(dData));
  HIP_CHECK(hipFreeArray(arr));
  free(hData);
  free(hOutputData);
}

HIP_TEST_CASE(Unit_hipSimpleTexture3D_Check_DataTypes) {
  CHECK_IMAGE_SUPPORT


  for (int i = 1; i < 25; i++) {
    runSimpleTexture3D_Check<float>(i, i, i, &texf);
    runSimpleTexture3D_Check<int>(i + 1, i, i, &texi);
    runSimpleTexture3D_Check<char>(i, i + 1, i, &texc);
  }
}


#endif  // CUDA_VERSION < CUDA_12000