MatrixTransposeEasyTest.cpp

/*
Copyright (c) 2015-present Advanced Micro Devices, Inc. All rights reserved.

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 "msvc_defines.h"
#include <hip/hip_runtime.h>
#include <iostream>
#include <string>
#include <chrono>

#define HIP_CHECK(status)                                                                                     \
    if (status != hipSuccess) {                                                                               \
        fprintf(stderr, "error: '%s'(%d) at %s:%d\n", hipGetErrorString(status), status, __FILE__, __LINE__); \
        exit(0);                                                                                              \
    }

// Device (Kernel) function, it must be void
__global__ void matrixTranspose(float* out, float* in, const int width) {
    int x = blockDim.x * blockIdx.x + threadIdx.x;
    int y = blockDim.y * blockIdx.y + threadIdx.y;
    out[y * width + x] = in[x * width + y];
}
__global__ void matrixMaltiplyHelper(float* A,float* B,float* output,const int a,const int b,const int c) {
    int x = blockDim.x * blockIdx.x + threadIdx.x;
    int y = blockDim.y * blockIdx.y + threadIdx.y;
    float adder=0.0f;
    if (x < a && y < c) {
        for (int i = 0; i < b; i++) {
            adder += A[x*b+i] * B[i*c+y];
        }
        output[x*c+y] = adder;
    }

}
void matrixMaltiply(float* A,float*B,float* output,int a,int b,int c) {//A:a*b B:b*c output:a*c
    float* gpuMatrixA = 0;
    float* gpuMatrixB = 0;
    float* gpuMatrixOutput = 0;
    HIP_CHECK(hipMalloc((void**)&gpuMatrixA, a * b * sizeof(float)));
    HIP_CHECK(hipMalloc((void**)&gpuMatrixB, b * c * sizeof(float)));
    HIP_CHECK(hipMalloc((void**)&gpuMatrixOutput, a * c * sizeof(float)));
    HIP_CHECK(hipMemcpy(gpuMatrixA, A, a * b * sizeof(float), hipMemcpyHostToDevice));
    HIP_CHECK(hipMemcpy(gpuMatrixB, B, b * c * sizeof(float), hipMemcpyHostToDevice));
    uint32_t THREADS_PER_BLOCK_X = 16;
    uint32_t THREADS_PER_BLOCK_Y = 16;
    uint32_t THREADS_PER_BLOCK_Z = 1;
    uint32_t GRID_X = uint32_t(a / THREADS_PER_BLOCK_X)+1;
    uint32_t GRID_Y = uint32_t(c / THREADS_PER_BLOCK_Y)+1;
    dim3 dimGrid(GRID_X, GRID_Y);
    dim3 dimBlock(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y, THREADS_PER_BLOCK_Z);
    matrixMaltiplyHelper <<<dimGrid, dimBlock, 0, 0 >>> (gpuMatrixA, gpuMatrixB, gpuMatrixOutput, a, b, c);
    HIP_CHECK(hipMemcpy(output, gpuMatrixOutput, a * c * sizeof(float),hipMemcpyDeviceToHost));
    HIP_CHECK(hipFree(gpuMatrixA));
    HIP_CHECK(hipFree(gpuMatrixB));
    HIP_CHECK(hipFree(gpuMatrixOutput));
}
void listAllGPUs() {
    int deviceCount = 0;
    hipGetDeviceCount(&deviceCount);  // 获取 GPU 数量

    if (deviceCount == 0) {
        std::cerr << "No AMD GPU detected!" << std::endl;
        return;
    }

    std::cout << "Found " << deviceCount << " GPU(s):" << std::endl;

    for (int i = 0; i < deviceCount; i++) {
        hipDeviceProp_t props;
        hipGetDeviceProperties(&props, i);  // 获取 GPU 属性

        std::cout << "\nGPU #" << i << ":\n"
            << "  Name: " << props.name << "\n"
            << "  PCI Domain/Bus/Device: "
            << props.pciDomainID << ":" << props.pciBusID << ":" << props.pciDeviceID << "\n"
            << "  Compute Units: " << props.multiProcessorCount << "\n"
            << "  GPU Clock: " << props.clockRate / 1000 << " MHz\n"
            << "  Memory Clock: " << props.memoryClockRate / 1000 << " MHz\n"
            << "  Total Memory: " << props.totalGlobalMem / (1024 * 1024) << " MB\n" << std::endl;
    }
}

int main() {
    auto start = std::chrono::high_resolution_clock::now();
    auto end = std::chrono::high_resolution_clock::now();
    auto ns = std::chrono::duration_cast<std::chrono::nanoseconds>(end - start);
    listAllGPUs();
    HIP_CHECK(hipSetDevice(0));
    hipDeviceProp_t devProp;
    HIP_CHECK(hipGetDeviceProperties(&devProp, 0));

    std::cout << "Device name " << std::string(devProp.name) << std::endl;
    
    const uint32_t A = 16;
    const uint32_t B = 20;
    const uint32_t C = 30;
    float* MatrixA = (float*)malloc(A * B * sizeof(float));
    float* MatrixB = (float*)malloc(B * C * sizeof(float));
    float* MatrixOutput = (float*)malloc(A * C * sizeof(float));
    for (uint32_t i = 0; i < A*B; ++i) {
        MatrixA[i] = (float)i * 1.0f;
    }
    for (uint32_t i = 0; i < B * C; ++i) {
        MatrixB[i] = (float)i * 1.0f;
    }
    start = std::chrono::high_resolution_clock::now();
    matrixMaltiply(MatrixA, MatrixB, MatrixOutput, A, B, C);
    end = std::chrono::high_resolution_clock::now();
    ns = std::chrono::duration_cast<std::chrono::nanoseconds>(end - start);
    std::cout << "  耗时: " << ns.count() << " 纳秒\n";

    for (int i = 0; i < A;i++) {
        for (int j=0;j < C;j++){
            std::cout << MatrixOutput[i*C+j] << std::endl;
        }
        std::cout << std::endl;
    }
    
    /*
    const uint32_t WIDTH = 16384;
    const uint32_t NUM = WIDTH * WIDTH;
    float* Matrix = (float*)malloc(NUM * sizeof(float));
    float* TransposeMatrix = (float*)malloc(NUM * sizeof(float));


    

    // Initialize the input data

    for (uint32_t i = 0; i < NUM; ++i) {
        Matrix[i] = (float)i * 10.987654f;
    }

    float* gpuMatrix = 0;
    float* gpuTransposeMatrix = 0;
    // Allocate the memory on device side
    HIP_CHECK(hipMalloc((void**)&gpuMatrix, NUM * sizeof(float)));
    HIP_CHECK(hipMalloc((void**)&gpuTransposeMatrix, NUM * sizeof(float)));
    // Memory transfer from host to device
    HIP_CHECK(hipMemcpy(gpuMatrix, Matrix, NUM * sizeof(float), hipMemcpyHostToDevice));
    for (int i = 1; i < 15; i++) {
        uint32_t THREADS_PER_BLOCK_X = (int)pow(2.0, i);
        uint32_t THREADS_PER_BLOCK_Y = (int)pow(2.0, i);
        uint32_t THREADS_PER_BLOCK_Z = 1;
        uint32_t GRID_X = uint32_t(WIDTH / THREADS_PER_BLOCK_X);

        uint32_t GRID_Y = uint32_t(WIDTH / THREADS_PER_BLOCK_Y);
        dim3 dimGrid(GRID_X, GRID_Y);
        dim3 dimBlock(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y, THREADS_PER_BLOCK_Z);
        // Lauching kernel from host


        start = std::chrono::high_resolution_clock::now();
        matrixTranspose <<<dimGrid, dimBlock, 0, 0 >>> (gpuTransposeMatrix, gpuMatrix, WIDTH);
        end = std::chrono::high_resolution_clock::now();
        ns = std::chrono::duration_cast<std::chrono::nanoseconds>(end - start);
        std::cout << "THREADS_PER_BLOCK_X = " << (int)pow(2.0, i) << "  耗时: " << ns.count() << " 纳秒\n";
    }



    // Memory transfer from device to host
    HIP_CHECK(hipMemcpy(TransposeMatrix, gpuTransposeMatrix, NUM * sizeof(float), hipMemcpyDeviceToHost));
    
    for (int i= 0; i < WIDTH; i++) {
        for (int j = 0; j < WIDTH; j++) {
            std::cout << "Matrix[" << i << "][" << j << "] = " << Matrix[i * WIDTH + j] << "  ";
            std::cout << "TransposeMatrix[" << j << "][" << i << "] = " << TransposeMatrix[j * WIDTH + i] << std::endl;
        }
    }
    
    HIP_CHECK(hipFree(gpuMatrix));
    HIP_CHECK(hipFree(gpuTransposeMatrix));
    // Free the resources on host side
    free(Matrix);
    free(TransposeMatrix);

    //return errors;
    */
    return 0;
}