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

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

#include "OCLMemDependency.h"

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

#include "CL/cl.h"

const static cl_uint Stages = 4;
const static cl_uint ThreadsForCheck = 1 << Stages;

#define KERNEL_CODE(...) #__VA_ARGS__

const static char* strKernel = KERNEL_CODE(
\n __kernel void bitonicSort(__global uint2* keys, uint stage, uint pass) {
  const uint thread = get_global_id(0);

  const uint pairDistance = 1 << (stage - pass);

  /* The purpose of this is to introduce an additional zero at stage - pass
   * bit*/
  const uint leftID = (thread & (pairDistance - 1)) |
                      ((thread & ~(pairDistance - 1)) << 1); /* Is the same as below */

  const uint direction = ((thread >> stage) & 1) == 1 ? 0 : 1;

  const uint rightID = leftID + pairDistance;
  const uint2 left = keys[leftID];
  const uint2 right = keys[rightID];

  const uint2 larger = left.x > right.x ? left : right;
  const uint2 smaller = left.x > right.x ? right : left;

  keys[leftID] = direction ? smaller : larger;
  keys[rightID] = direction ? larger : smaller;
}
\n);

OCLMemDependency::OCLMemDependency() { _numSubTests = 1; }

OCLMemDependency::~OCLMemDependency() {}

void OCLMemDependency::open(unsigned int test, char* units, double& conversion,
                            unsigned int deviceId) {
  OCLTestImp::open(test, units, conversion, deviceId);
  CHECK_RESULT((error_ != CL_SUCCESS), "Error opening test");
  char dbuffer[1024] = {0};
  program_ = _wrapper->clCreateProgramWithSource(context_, 1, &strKernel, NULL, &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateProgramWithSource()  failed");

  error_ = _wrapper->clBuildProgram(program_, 1, &devices_[deviceId], NULL, NULL, NULL);
  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_, "bitonicSort", &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateKernel() failed");

  cl_mem buffer;
  buffer = _wrapper->clCreateBuffer(context_, CL_MEM_READ_WRITE, ThreadsForCheck * sizeof(cl_uint2),
                                    NULL, &error_);
  CHECK_RESULT((error_ != CL_SUCCESS), "clCreateBuffer() failed");
  buffers_.push_back(buffer);
  cl_buffer_region reg = {0, ThreadsForCheck * sizeof(cl_uint2)};
  buffer = _wrapper->clCreateSubBuffer(buffers()[0], CL_MEM_READ_WRITE,
                                       CL_BUFFER_CREATE_TYPE_REGION, &reg, &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 OCLMemDependency::run(void) {
  cl_uint2 values[ThreadsForCheck] = {
      {{3, 0}},  {{1, 5}}, {{4, 6}},  {{2, 4}},  {{0, 3}},   {{5, 10}},  {{15, 7}}, {{13, 8}},
      {{10, 2}}, {{9, 1}}, {{7, 11}}, {{11, 9}}, {{14, 12}}, {{12, 14}}, {{6, 13}}, {{8, 15}}};
  cl_uint2 reference[ThreadsForCheck] = {
      {{0, 3}},  {{1, 5}}, {{3, 0}},  {{2, 4}},  {{4, 6}},   {{5, 10}},  {{6, 13}}, {{8, 15}},
      {{7, 11}}, {{9, 1}}, {{10, 2}}, {{11, 9}}, {{14, 12}}, {{12, 14}}, {{15, 7}}, {{13, 8}}};
  cl_uint2 results[ThreadsForCheck];

  cl_mem buffer = buffers()[0];
  error_ = _wrapper->clEnqueueWriteBuffer(cmdQueues_[_deviceId], buffer, true, 0, sizeof(values),
                                          values, 0, NULL, NULL);
  CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueWriteBuffer() failed");

  size_t gws[1] = {ThreadsForCheck};

  for (unsigned int i = 0; i < Stages; ++i) {
    buffer = buffers()[i % 2];
    error_ = _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), &buffer);
    CHECK_RESULT((error_ != CL_SUCCESS), "clSetKernelArg() failed");
    for (unsigned int j = 0; j < i; ++j) {
      error_ = _wrapper->clSetKernelArg(kernel_, 1, sizeof(unsigned int), &i);
      CHECK_RESULT((error_ != CL_SUCCESS), "clSetKernelArg() failed");
      error_ = _wrapper->clSetKernelArg(kernel_, 2, sizeof(unsigned int), &j);
      CHECK_RESULT((error_ != CL_SUCCESS), "clSetKernelArg() failed");

      error_ = _wrapper->clEnqueueNDRangeKernel(cmdQueues_[_deviceId], kernel_, 1, NULL, gws, NULL,
                                                0, NULL, NULL);
      CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueNDRangeKernel() failed");
    }
  }

  buffer = buffers()[0];
  error_ = _wrapper->clEnqueueReadBuffer(cmdQueues_[_deviceId], buffer, true, 0, sizeof(results),
                                         results, 0, NULL, NULL);
  CHECK_RESULT((error_ != CL_SUCCESS), "clEnqueueReadBuffer() failed");
  for (unsigned int i = 0; i < ThreadsForCheck; ++i) {
    if ((results[i].s[0] != reference[i].s[0]) || (results[i].s[1] != reference[i].s[1])) {
      CHECK_RESULT(true, "Incorrect result for dependency!\n");
    }
  }
}

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