TheRock/rocm-systems/projects/clr/opencl/tests/ocltst/module/runtime/OCLPartialWrkgrp.cpp

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

#include "OCLPartialWrkgrp.h"

#include <Timer.h>
#include <assert.h>
#include <stdio.h>
#include <string.h>

#include "CL/cl.h"

static const size_t BufSize = 0x1000;

const static char* strKernel =
    "__kernel void fillX(__global int4* out)                                \n"
    "{                                                                      \n"
    "   int id = get_global_id(0);                                          \n"
    "   out[id].x = id;                                                     \n"
    "}                                                                      \n"
    "                                                                       \n"
    "__kernel void fillXY(__global int4* out)                               \n"
    "{                                                                      \n"
    "   int id = get_global_id(0) + get_global_id(1) * get_global_size(0);  \n"
    "   out[id].x = get_global_id(0);                                       \n"
    "   out[id].y = get_global_id(1);                                       \n"
    "}                                                                      \n"
    "                                                                       \n"
    "__kernel void fillXYZ(__global int4* out)                              \n"
    "{                                                                      \n"
    "   int id = get_global_id(0) + get_global_id(1) * get_global_size(0) + \n"
    "       get_global_id(2) * get_global_size(0) * get_global_size(1);     \n"
    "   out[id].x = get_global_id(0);                                       \n"
    "   out[id].y = get_global_id(1);                                       \n"
    "   out[id].z = get_global_id(2);                                       \n"
    "}                                                                      \n";

OCLPartialWrkgrp::OCLPartialWrkgrp() {
  _numSubTests = 3;
  isOCL2_ = true;
}

OCLPartialWrkgrp::~OCLPartialWrkgrp() {}

void OCLPartialWrkgrp::open(unsigned int test, char* units, double& conversion,
                            unsigned int deviceId) {
  _openTest = test;

  OCLTestImp::open(test, units, conversion, deviceId);
  CHECK_RESULT((error_ != CL_SUCCESS), "Error opening test");

  char version[128];
  _wrapper->clGetDeviceInfo(devices_[deviceId], CL_DEVICE_VERSION, sizeof(version), version, NULL);

  if (_openTest != 0 && strstr(version, "OpenCL 2.0") == NULL) {
    isOCL2_ = false;
    return;
  }

  program_ = _wrapper->clCreateProgramWithSource(context_, 1, &strKernel, NULL, &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateProgramWithSource()  failed");

  switch (_openTest) {
    case 0:
      error_ = _wrapper->clBuildProgram(program_, 1, &devices_[deviceId], NULL, NULL, NULL);
      break;
    case 1:
      error_ = _wrapper->clBuildProgram(program_, 1, &devices_[deviceId],
                                        "-cl-uniform-work-group-size -cl-std=CL2.0", NULL, NULL);
      break;
    case 2:
      error_ =
          _wrapper->clBuildProgram(program_, 1, &devices_[deviceId], "-cl-std=CL2.0", NULL, NULL);
      break;
    default:
      CHECK_RESULT(false, "Invalid test number > _numSubTests");
      return;
  }

  if (error_ != CL_SUCCESS) {
    char programLog[1024];
    _wrapper->clGetProgramBuildInfo(program_, devices_[deviceId], CL_PROGRAM_BUILD_LOG, 1024,
                                    programLog, 0);
    printf("\n%s\n", programLog);
    fflush(stdout);
  }
  CHECK_RESULT((error_ != CL_SUCCESS), "clBuildProgram() failed");

  kernel_ = _wrapper->clCreateKernel(program_, "fillX", &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateKernel() failed");

  cl_mem buffer;
  buffer = _wrapper->clCreateBuffer(context_, CL_MEM_WRITE_ONLY, BufSize * sizeof(cl_int4), NULL,
                                    &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateBuffer() failed");
  buffers_.push_back(buffer);
}

static void CL_CALLBACK notify_callback(const char* errinfo, const void* private_info, size_t cb,
                                        void* user_data) {}

void OCLPartialWrkgrp::run(void) {
  if (!isOCL2_) return;
  unsigned int* values;
  cl_mem buffer = buffers()[0];
  values = reinterpret_cast<unsigned int*>(new cl_int4[BufSize]);

  //
  // Check unaligned workgroup in X dimension
  //

  // Clear destination buffer
  memset(values, 0, BufSize * sizeof(cl_int4));
  error_ = _wrapper->clEnqueueWriteBuffer(cmdQueues_[_deviceId], buffer, CL_TRUE, 0,
                                          BufSize * sizeof(cl_int4), values, 0, NULL, NULL);
  CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueWriteBuffer() failed");

  error_ = _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), &buffer);
  CHECK_RESULT((error_ != CL_SUCCESS), "clSetKernelArg() failed");

  size_t gws[1] = {BufSize - 1};
  size_t lws[1] = {256};
  error_ = _wrapper->clEnqueueNDRangeKernel(cmdQueues_[_deviceId], kernel_, 1, NULL, gws, lws, 0,
                                            NULL, NULL);

  switch (_openTest) {
    case 2:
      if (error_ != CL_SUCCESS) {
        CHECK_RESULT(false, "clEnqueueNDRangeKernel() failed");
        return;
      }
      error_ = _wrapper->clEnqueueReadBuffer(cmdQueues_[_deviceId], buffer, CL_TRUE, 0,
                                             BufSize * sizeof(cl_int4), values, 0, NULL, NULL);
      CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueReadBuffer() failed");

      for (size_t x = 0; x < BufSize; ++x) {
        if (x == (BufSize - 1)) {
          CHECK_RESULT((values[4 * x] != 0), "Comparison failed!");
        } else {
          CHECK_RESULT((values[4 * x] != x), "Comparison failed!");
        }
      }
      break;
    case 1:
    case 0:
      CHECK_RESULT((error_ != CL_INVALID_WORK_GROUP_SIZE),
                   "clEnqueueNDRangeKernel(): "
                   "Expected to fail for non-uniform work group sizes!");
      break;
    default:
      CHECK_RESULT(false, "Invalid test number > _numSubTests");
      return;
  }

  error_ = _wrapper->clReleaseKernel(kernel_);
  CHECK_RESULT_NO_RETURN((error_ != CL_SUCCESS), "clReleaseKernel() failed");

  //
  // Check unaligned workgroup in X and Y dimensions
  //
  kernel_ = _wrapper->clCreateKernel(program_, "fillXY", &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateKernel() failed");

  // Clear destination buffer
  memset(values, 0, BufSize * sizeof(cl_int4));
  error_ = _wrapper->clEnqueueWriteBuffer(cmdQueues_[_deviceId], buffer, CL_TRUE, 0,
                                          BufSize * sizeof(cl_int4), values, 0, NULL, NULL);
  CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueWriteBuffer() failed");

  error_ = _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), &buffer);
  CHECK_RESULT((error_ != CL_SUCCESS), "clSetKernelArg() failed");

  size_t gws2[2] = {0x3f, 0x3f};
  size_t lws2[2] = {16, 16};
  error_ = _wrapper->clEnqueueNDRangeKernel(cmdQueues_[_deviceId], kernel_, 2, NULL, gws2, lws2, 0,
                                            NULL, NULL);

  switch (_openTest) {
    case 2:
      if (error_ != CL_SUCCESS) {
        CHECK_RESULT(false, "clEnqueueNDRangeKernel() failed");
        return;
      }
      error_ = _wrapper->clEnqueueReadBuffer(cmdQueues_[_deviceId], buffer, CL_TRUE, 0,
                                             BufSize * sizeof(cl_int4), values, 0, NULL, NULL);
      CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueReadBuffer() failed");

      for (size_t y = 0; y < 0x40; ++y) {
        for (size_t x = 0; x < 0x3f; ++x) {
          size_t id = x + y * 0x3f;
          if (y == 0x3f) {
            CHECK_RESULT((values[4 * id] != 0), "Comparison failed!");
            CHECK_RESULT((values[4 * id + 1] != 0), "Comparison failed!");
          } else {
            CHECK_RESULT((values[4 * id] != x), "Comparison failed!");
            CHECK_RESULT((values[4 * id + 1] != y), "Comparison failed!");
          }
        }
      }
      break;
    case 1:
    case 0:
      CHECK_RESULT((error_ != CL_INVALID_WORK_GROUP_SIZE),
                   "clEnqueueNDRangeKernel(): "
                   "Expected to fail for non-uniform work group sizes!");
      break;
    default:
      CHECK_RESULT(false, "Invalid test number > _numSubTests");
      return;
  }

  error_ = _wrapper->clReleaseKernel(kernel_);
  CHECK_RESULT_NO_RETURN((error_ != CL_SUCCESS), "clReleaseKernel() failed");

  //
  // Check unaligned workgroup in X, Y and Z dimensions
  //
  kernel_ = _wrapper->clCreateKernel(program_, "fillXYZ", &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateKernel() failed");

  // Clear destination buffer
  memset(values, 0, BufSize * sizeof(cl_int4));
  error_ = _wrapper->clEnqueueWriteBuffer(cmdQueues_[_deviceId], buffer, CL_TRUE, 0,
                                          BufSize * sizeof(cl_int4), values, 0, NULL, NULL);
  CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueWriteBuffer() failed");

  error_ = _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), &buffer);
  CHECK_RESULT((error_ != CL_SUCCESS), "clSetKernelArg() failed");

  size_t gws3[3] = {0xf, 0x10, 0xf};
  size_t lws3[3] = {4, 4, 4};
  error_ = _wrapper->clEnqueueNDRangeKernel(cmdQueues_[_deviceId], kernel_, 3, NULL, gws3, lws3, 0,
                                            NULL, NULL);
  switch (_openTest) {
    case 2:
      if (error_ != CL_SUCCESS) {
        CHECK_RESULT(false, "clEnqueueNDRangeKernel() failed");
        return;
      }
      error_ = _wrapper->clEnqueueReadBuffer(cmdQueues_[_deviceId], buffer, CL_TRUE, 0,
                                             BufSize * sizeof(cl_int4), values, 0, NULL, NULL);
      CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueReadBuffer() failed");

      for (size_t z = 0; z < 0x10; ++z) {
        for (size_t y = 0; y < 0x10; ++y) {
          for (size_t x = 0; x < 0xf; ++x) {
            size_t id = x + y * 0xf + z * 0xf0;
            if (z == 0xf) {
              CHECK_RESULT((values[4 * id] != 0), "Comparison failed!");
              CHECK_RESULT((values[4 * id + 1] != 0), "Comparison failed!");
              CHECK_RESULT((values[4 * id + 2] != 0), "Comparison failed!");
            } else {
              CHECK_RESULT((values[4 * id] != x), "Comparison failed!");
              CHECK_RESULT((values[4 * id + 1] != y), "Comparison failed!");
              CHECK_RESULT((values[4 * id + 2] != z), "Comparison failed!");
            }
          }
        }
      }
      break;
    case 1:
    case 0:
      CHECK_RESULT((error_ != CL_INVALID_WORK_GROUP_SIZE),
                   "clEnqueueNDRangeKernel(): "
                   "Expected fail for non-uniform work group sizes!");
      break;
    default:
      CHECK_RESULT(false, "Invalid test number > _numSubTests");
      return;
  }

  delete[] values;
}

unsigned int OCLPartialWrkgrp::close(void) { return OCLTestImp::close(); }