roctracer 的应用示例

1,不用 roctracer 的普通场景

mt.cpp

复制代码
/* Copyright (c) 2018-2022 Advanced Micro Devices, Inc.

 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. */

#include <iostream>

// hip header file
#include <hip/hip_runtime.h>

#define HIP_CALL(call)                                                                             \
  do {                                                                                             \
    hipError_t err = call;                                                                         \
    if (err != hipSuccess) {                                                                       \
      fprintf(stderr, "%s\n", hipGetErrorString(err));                                             \
      abort();                                                                                     \
    }                                                                                              \
  } while (0)

#define WIDTH 1024


#define NUM (WIDTH * WIDTH)

#define THREADS_PER_BLOCK_X 4
#define THREADS_PER_BLOCK_Y 4
#define THREADS_PER_BLOCK_Z 1

// Device (Kernel) function, it must be void
__global__ void matrixTranspose(float* out, float* in, const int width) {
  int x = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x;
  int y = hipBlockDim_y * hipBlockIdx_y + hipThreadIdx_y;

  out[y * width + x] = in[x * width + y];
}

// CPU implementation of matrix transpose
void matrixTransposeCPUReference(float* output, float* input, const unsigned int width) {
  for (unsigned int j = 0; j < width; j++) {
    for (unsigned int i = 0; i < width; i++) {
      output[i * width + j] = input[j * width + i];
    }
  }
}

int main() {
  float* Matrix;
  float* TransposeMatrix;
  float* cpuTransposeMatrix;

  float* gpuMatrix;
  float* gpuTransposeMatrix;

  hipDeviceProp_t devProp;
  HIP_CALL(hipGetDeviceProperties(&devProp, 0));

  std::cerr << "Device name " << devProp.name << std::endl;

  int i;
  int errors;

  Matrix = (float*)malloc(NUM * sizeof(float));
  TransposeMatrix = (float*)malloc(NUM * sizeof(float));
  cpuTransposeMatrix = (float*)malloc(NUM * sizeof(float));

  // initialize the input data
  for (i = 0; i < NUM; i++) {
    Matrix[i] = (float)i * 10.0f;
  }

  // allocate the memory on the device side
  HIP_CALL(hipMalloc((void**)&gpuMatrix, NUM * sizeof(float)));
  HIP_CALL(hipMalloc((void**)&gpuTransposeMatrix, NUM * sizeof(float)));

  uint32_t iterations = 100;
  while (iterations-- > 0) {
    std::cerr << "## Iteration (" << iterations << ") #################" << std::endl;

    // Memory transfer from host to device
    HIP_CALL(hipMemcpy(gpuMatrix, Matrix, NUM * sizeof(float), hipMemcpyHostToDevice));

    // Lauching kernel from host
    hipLaunchKernelGGL(
        matrixTranspose, dim3(WIDTH / THREADS_PER_BLOCK_X, WIDTH / THREADS_PER_BLOCK_Y),
        dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y), 0, 0, gpuTransposeMatrix, gpuMatrix, WIDTH);


    HIP_CALL(
        hipMemcpy(TransposeMatrix, gpuTransposeMatrix, NUM * sizeof(float), hipMemcpyDeviceToHost));


    // CPU MatrixTranspose computation
    matrixTransposeCPUReference(cpuTransposeMatrix, Matrix, WIDTH);

    // verify the results
    errors = 0;
    double eps = 1.0E-6;
    for (i = 0; i < NUM; i++) {
      if (std::abs(TransposeMatrix[i] - cpuTransposeMatrix[i]) > eps) {
        errors++;
      }
    }
    if (errors != 0) {
      fprintf(stderr, "FAILED: %d errors\n", errors);
    } else {
      fprintf(stderr, "PASSED!\n");
    }
  }

  // free the resources on device side
  HIP_CALL(hipFree(gpuMatrix));
  HIP_CALL(hipFree(gpuTransposeMatrix));

  // free the resources on host side
  free(Matrix);
  free(TransposeMatrix);
  free(cpuTransposeMatrix);

  return errors;
}

编译:

$ hipcc mt.cpp -o mt

$ ./mt xxx

不会产生文件;

2,加入roctracer的源文件

MatrixTranspose.cpp:

cpp 复制代码
/* Copyright (c) 2018-2022 Advanced Micro Devices, Inc.

 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. */

#include <iostream>

// hip header file
#include <hip/hip_runtime.h>
#include "roctracer_ext.h"
// roctx header file
#include <roctx.h>

#define HIP_CALL(call)                                                                             \
  do {                                                                                             \
    hipError_t err = call;                                                                         \
    if (err != hipSuccess) {                                                                       \
      fprintf(stderr, "%s\n", hipGetErrorString(err));                                             \
      abort();                                                                                     \
    }                                                                                              \
  } while (0)

#define WIDTH 1024


#define NUM (WIDTH * WIDTH)

#define THREADS_PER_BLOCK_X 4
#define THREADS_PER_BLOCK_Y 4
#define THREADS_PER_BLOCK_Z 1

// Device (Kernel) function, it must be void
__global__ void matrixTranspose(float* out, float* in, const int width) {
  int x = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x;
  int y = hipBlockDim_y * hipBlockIdx_y + hipThreadIdx_y;

  out[y * width + x] = in[x * width + y];
}

// CPU implementation of matrix transpose
void matrixTransposeCPUReference(float* output, float* input, const unsigned int width) {
  for (unsigned int j = 0; j < width; j++) {
    for (unsigned int i = 0; i < width; i++) {
      output[i * width + j] = input[j * width + i];
    }
  }
}

int main() {
  float* Matrix;
  float* TransposeMatrix;
  float* cpuTransposeMatrix;

  float* gpuMatrix;
  float* gpuTransposeMatrix;

  hipDeviceProp_t devProp;
  HIP_CALL(hipGetDeviceProperties(&devProp, 0));

  std::cerr << "Device name " << devProp.name << std::endl;

  int i;
  int errors;

  Matrix = (float*)malloc(NUM * sizeof(float));
  TransposeMatrix = (float*)malloc(NUM * sizeof(float));
  cpuTransposeMatrix = (float*)malloc(NUM * sizeof(float));

  // initialize the input data
  for (i = 0; i < NUM; i++) {
    Matrix[i] = (float)i * 10.0f;
  }

  // allocate the memory on the device side
  HIP_CALL(hipMalloc((void**)&gpuMatrix, NUM * sizeof(float)));
  HIP_CALL(hipMalloc((void**)&gpuTransposeMatrix, NUM * sizeof(float)));

  uint32_t iterations = 100;
  while (iterations-- > 0) {
    std::cerr << "## Iteration (" << iterations << ") #################" << std::endl;

    // Memory transfer from host to device
    HIP_CALL(hipMemcpy(gpuMatrix, Matrix, NUM * sizeof(float), hipMemcpyHostToDevice));

    roctxMark("before hipLaunchKernel");
    int rangeId = roctxRangeStart("hipLaunchKernel range");
    roctxRangePush("hipLaunchKernel");
    // Lauching kernel from host
    hipLaunchKernelGGL(
        matrixTranspose, dim3(WIDTH / THREADS_PER_BLOCK_X, WIDTH / THREADS_PER_BLOCK_Y),
        dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y), 0, 0, gpuTransposeMatrix, gpuMatrix, WIDTH);
    roctxMark("after hipLaunchKernel");

    // Memory transfer from device to host
    roctxRangePush("hipMemcpy");

    HIP_CALL(
        hipMemcpy(TransposeMatrix, gpuTransposeMatrix, NUM * sizeof(float), hipMemcpyDeviceToHost));

    roctxRangePop();  // for "hipMemcpy"
    roctxRangePop();  // for "hipLaunchKernel"
    roctxRangeStop(rangeId);

    // CPU MatrixTranspose computation
    matrixTransposeCPUReference(cpuTransposeMatrix, Matrix, WIDTH);

    // verify the results
    errors = 0;
    double eps = 1.0E-6;
    for (i = 0; i < NUM; i++) {
      if (std::abs(TransposeMatrix[i] - cpuTransposeMatrix[i]) > eps) {
        errors++;
      }
    }
    if (errors != 0) {
      fprintf(stderr, "FAILED: %d errors\n", errors);
    } else {
      fprintf(stderr, "PASSED!\n");
    }
  }

  // free the resources on device side
  HIP_CALL(hipFree(gpuMatrix));
  HIP_CALL(hipFree(gpuTransposeMatrix));

  // free the resources on host side
  free(Matrix);
  free(TransposeMatrix);
  free(cpuTransposeMatrix);

  return errors;
}

编译:

只使用hipcc无法直接编译这个源文件

需要指定include 目录和链接库:

cpp 复制代码
$ hipcc ./MatrixTranspose.cpp  -I /opt/rocm/include/roctracer/ -lroctx64

运行:

./a.out

相关推荐
smallcutepanda10 分钟前
0x22 深度优先搜索0x23剪枝0x24迭代加深meet-in-the-middle
算法·深度优先
xuanjiong10 分钟前
纯个人整理,蓝桥杯使用的算法模板day3(完全背包dp问题),手打个人理解注释,超全面,且均已验证成功(附带详细手写“模拟流程图”,全网首个
算法·蓝桥杯·流程图·动态规划
晴空了无痕13 分钟前
群体智能避障革命:RVO算法在Unity中的深度实践与优化
算法·unity·游戏引擎
双叶83617 分钟前
(51单片机)矩阵按键密码锁表白(C语言代码编撰)(矩阵按键教程)(LCD1602浅教程)
c语言·开发语言·c++·算法·游戏·矩阵·51单片机
海棠蚀omo42 分钟前
C++笔记-string(下)
开发语言·c++·笔记
天堂的恶魔9461 小时前
C++ —— 文件操作(流式操作)
c++
apcipot_rain1 小时前
密码学——序列密码 序列线性复杂度 B-M算法 例题演示
算法·密码学
迪小莫学AI2 小时前
LeetCode 1863. 找出所有子集的异或总和再求和
算法·leetcode·深度优先
dora2 小时前
逼格提起来,使用curl发送网络请求
android·c++
yy_xzz2 小时前
基于条码数据生成校验密码的C++实现方案
开发语言·c++