TheRock/rocm-systems/projects/clr/opencl/tests/ocltst/module/perf/OCLPerfGenericBandwidth.cpp

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

#include "OCLPerfGenericBandwidth.h"

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

#include "CL/cl.h"
#include "Timer.h"

// Quiet pesky warnings
#ifdef WIN_OS
#define SNPRINTF sprintf_s
#else
#define SNPRINTF snprintf
#endif

#define NUM_SIZES 4
// 256KB, 1 MB, 4MB, 16 MB
static const unsigned int Sizes[NUM_SIZES] = {262144, 1048576, 4194304, 16777216};

void OCLPerfGenericBandwidth::genShader(unsigned int idx) {
  shader_.clear();
  if (idx == 0) {
    shader_ +=
        "__kernel __attribute__((reqd_work_group_size(64,1,1))) void "
        "_genericReadSpeed(global float *outBuf, global float *inBuf, local "
        "float *inLocal, float c, char useLocal)\n"
        "{\n"
        "    int gid = (int) get_global_id(0);\n"
        "    int lid = (int) get_local_id(0);\n"
        "    float val0 = 0.0f;\n"
        "    float val1 = 0.0f;\n"
        "    float *localLocal;\n"
        "    int hacklid = gid % 64;\n"
        "    if (useLocal)\n"
        "        localLocal = inLocal;\n"
        "    else\n"
        "        localLocal = inBuf;\n"
        "    for (int i = 0; i < (768/64); i++) {\n"
        "        localLocal[hacklid + i*64] = lid;\n"
        "    }\n"
        "    barrier(CLK_LOCAL_MEM_FENCE);\n"
        "#pragma nounroll\n"
        "    for (uint i = 0; i < 32;i++)\n"
        "    {\n"
        "        val0 += localLocal[lid+0];\n"
        "        val1 += localLocal[lid+64];\n"
        "        val0 += localLocal[lid+128];\n"
        "        val1 += localLocal[lid+192];\n"
        "        val0 += localLocal[lid+256];\n"
        "        val1 += localLocal[lid+320];\n"
        "        val0 += localLocal[lid+384];\n"
        "        val1 += localLocal[lid+448];\n"
        "        lid += 1;\n"
        "    }\n"
        "    val0 += val1;\n"
        "    val1 = min(val0,1.0f);\n"
        "    if ((lid + val1) < 0){\n"
        "        outBuf[gid] = val0;\n"
        "    }\n"
        "}\n";
    dataSizeBytes_ = 768 * 4;
  } else {
    shader_ +=
        "__kernel __attribute__((reqd_work_group_size(64,1,1))) void "
        "_genericReadSpeed(global float *outBuf, global float *inBuf, local "
        "float *inLocal, float c, char useLocal)\n"
        "{\n"
        "    uint gid = (uint) get_global_id(0);\n"
        "    int lid = (int) get_local_id(0);\n"
        "    float val0 = 0.0f;\n"
        "    float val1 = 0.0f;\n"
        "    float *localLocal;\n"
        "    uint hacklid = gid % 64;\n"
        "    if (useLocal)\n"
        "        localLocal = inLocal;\n"
        "    else\n"
        "        localLocal = inBuf;\n"
        "    for (int i = 0; i < (256/64); i++) {\n"
        "        localLocal[hacklid + i*64] = lid;\n"
        "    }\n"
        "    barrier(CLK_LOCAL_MEM_FENCE);\n"
        "    #pragma nounroll\n"
        "    for (uint i = 0; i < 32;i++)\n"
        "    {\n"
        "        val0 += localLocal[8*i+0];\n"
        "        val1 += localLocal[8*i+1];\n"
        "        val0 += localLocal[8*i+2];\n"
        "        val1 += localLocal[8*i+3];\n"
        "        val0 += localLocal[8*i+4];\n"
        "        val1 += localLocal[8*i+5];\n"
        "        val0 += localLocal[8*i+6];\n"
        "        val1 += localLocal[8*i+7];\n"
        "    }\n"
        "    val0 += val1;\n"
        "    val1 = min(val0,1.0f);\n"
        "    if ((lid + val1) < 0){\n"
        "        outBuf[gid] = val0;\n"
        "    }\n"
        "}\n";
    dataSizeBytes_ = 256 * 4;
  }
}

OCLPerfGenericBandwidth::OCLPerfGenericBandwidth() { _numSubTests = NUM_SIZES * 4; }

OCLPerfGenericBandwidth::~OCLPerfGenericBandwidth() {}

void OCLPerfGenericBandwidth::setData(cl_mem buffer, float val) {
  float* data = (float*)_wrapper->clEnqueueMapBuffer(
      cmdQueues_[_deviceId], buffer, true, CL_MAP_WRITE, 0, bufSize_, 0, NULL, NULL, &error_);
  for (unsigned int i = 0; i < (bufSize_ >> 2); i++) data[i] = val;
  error_ = _wrapper->clEnqueueUnmapMemObject(cmdQueues_[_deviceId], buffer, data, 0, NULL, NULL);
  _wrapper->clFinish(cmdQueues_[_deviceId]);
}

void OCLPerfGenericBandwidth::checkData(cl_mem buffer) {
  float* data = (float*)_wrapper->clEnqueueMapBuffer(
      cmdQueues_[_deviceId], buffer, true, CL_MAP_READ, 0, bufSize_, 0, NULL, NULL, &error_);
  for (unsigned int i = 0; i < (bufSize_ >> 2); i++) {
    if (data[i] != (float)numReads_) {
      printf("Data validation failed at index %d!\n", i);
      printf("Expected %d %d %d %d\nGot %d %d %d %d\n", numReads_, numReads_, numReads_, numReads_,
             (unsigned int)data[i], (unsigned int)data[i + 1], (unsigned int)data[i + 2],
             (unsigned int)data[i + 3]);
      CHECK_RESULT_NO_RETURN(0, "Data validation failed!\n");
      break;
    }
  }
  error_ = _wrapper->clEnqueueUnmapMemObject(cmdQueues_[_deviceId], buffer, data, 0, NULL, NULL);
  _wrapper->clFinish(cmdQueues_[_deviceId]);
}

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

void OCLPerfGenericBandwidth::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");
  _crcword = 0;
  conversion = 1.0f;

  failed = false;
  kernel_ = 0;
  inBuffer_ = 0;
  outBuffer_ = 0;
  useLDS_ = ((test / NUM_SIZES) % 2) == 0 ? 1 : 0;

  size_t param_size = 0;
  char* strVersion = 0;
  error_ =
      _wrapper->clGetDeviceInfo(devices_[_deviceId], CL_DEVICE_OPENCL_C_VERSION, 0, 0, &param_size);
  CHECK_RESULT(error_ != CL_SUCCESS, "clGetDeviceInfo failed");
  strVersion = new char[param_size];
  error_ = _wrapper->clGetDeviceInfo(devices_[_deviceId], CL_DEVICE_OPENCL_C_VERSION, param_size,
                                     strVersion, 0);
  CHECK_RESULT(error_ != CL_SUCCESS, "clGetDeviceInfo failed");
  if (strVersion[9] < '2') {
    failed = true;
    return;
  }
  delete strVersion;

  numReads_ = 32;
  width_ = Sizes[test % NUM_SIZES];
  shaderIdx_ = test / (NUM_SIZES * 2);

  bufSize_ = width_;

  inBuffer_ = _wrapper->clCreateBuffer(context_, 0, bufSize_, NULL, &error_);
  CHECK_RESULT(inBuffer_ == 0, "clCreateBuffer(inBuffer) failed");

  outBuffer_ = _wrapper->clCreateBuffer(context_, 0, bufSize_, NULL, &error_);
  CHECK_RESULT(outBuffer_ == 0, "clCreateBuffer(outBuffer) failed");

  genShader(shaderIdx_);
  char* tmp = (char*)shader_.c_str();
  program_ = _wrapper->clCreateProgramWithSource(context_, 1, (const char**)&tmp, NULL, &error_);
  CHECK_RESULT(program_ == 0, "clCreateProgramWithSource failed");

  error_ = _wrapper->clBuildProgram(program_, 1, &devices_[deviceId], "-cl-std=CL2.0", NULL, NULL);

  if (error_ != CL_SUCCESS) {
    cl_int intError;
    char log[16384];
    intError = _wrapper->clGetProgramBuildInfo(program_, devices_[deviceId], CL_PROGRAM_BUILD_LOG,
                                               16384 * sizeof(char), log, NULL);
    printf("Build error -> %s\n", log);

    CHECK_RESULT(0, "clBuildProgram failed");
  }
  kernel_ = _wrapper->clCreateKernel(program_, "_genericReadSpeed", &error_);
  CHECK_RESULT(kernel_ == 0, "clCreateKernel failed");

  float foo = 0;
  error_ = _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), (void*)&outBuffer_);
  error_ = _wrapper->clSetKernelArg(kernel_, 1, sizeof(cl_mem), (void*)&inBuffer_);
  error_ = _wrapper->clSetKernelArg(kernel_, 2, 1024 * sizeof(cl_float), (void*)NULL);
  error_ = _wrapper->clSetKernelArg(kernel_, 3, sizeof(cl_float), (void*)&foo);
  error_ = _wrapper->clSetKernelArg(kernel_, 4, sizeof(cl_char), (void*)&useLDS_);

  setData(outBuffer_, 1.2345678f);
}

void OCLPerfGenericBandwidth::run(void) {
  if (failed) return;
  int global = bufSize_ / sizeof(cl_float);
  int local = 64;

  size_t global_work_size[1] = {(size_t)global};
  size_t local_work_size[1] = {(size_t)local};

  CPerfCounter timer;

  timer.Reset();
  timer.Start();
  for (unsigned int i = 0; i < NUM_ITER; i++) {
    error_ = _wrapper->clEnqueueNDRangeKernel(cmdQueues_[_deviceId], kernel_, 1, NULL,
                                              (const size_t*)global_work_size,
                                              (const size_t*)local_work_size, 0, NULL, NULL);

    CHECK_RESULT(error_, "clEnqueueNDRangeKernel failed");
  }
  _wrapper->clFinish(cmdQueues_[_deviceId]);

  timer.Stop();
  double sec = timer.GetElapsedTime();

  char buf[256];
  const char* buf2;
  if (useLDS_)
    buf2 = "LDS";
  else
    buf2 = "global";
  const char* buf3;
  if (shaderIdx_ == 0) {
    buf3 = "reads";
    numReads_ *= 8;
  } else {
    buf3 = "broadcast";
    numReads_ *= 8;
  }
  // LDS bandwidth in GB/s
  // We have one extra write per LDS location to initialize LDS
  double perf = ((double)global * (numReads_ * sizeof(cl_float) + dataSizeBytes_ / 64) * NUM_ITER *
                 (double)(1e-09)) /
                sec;

  _perfInfo = (float)perf;
  SNPRINTF(buf, sizeof(buf), " %6s %9s %8d threads, %3d reads (GB/s) ", buf2, buf3, global,
           numReads_);
  testDescString = buf;
  // checkData(outBuffer_);
}

unsigned int OCLPerfGenericBandwidth::close(void) {
  if (inBuffer_) {
    error_ = _wrapper->clReleaseMemObject(inBuffer_);
    CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseMemObject(inBuffer_) failed");
  }
  if (outBuffer_) {
    error_ = _wrapper->clReleaseMemObject(outBuffer_);
    CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseMemObject(outBuffer_) failed");
  }

  return OCLTestImp::close();
}