oneAPI 学习日志 01

graph TD
    A[单源 C++ SYCL 代码] --> B{oneAPI / DPC++ 编译器}
    B --> C[Intel CPU]
    B --> D[Intel Xe GPU]
    B --> E[NVIDIA/AMD GPU via 适配器]
    B --> F[FPGA / 其它加速器]

# 下载并配置 Intel 的 GPG 密钥
wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.key | gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null

# 添加 APT 仓库
echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list

# 更新并安装 oneAPI Toolkit
sudo apt update
sudo apt install intel-oneapi-toolkit

source /opt/intel/oneapi/setvars.sh

# 只能安装最新版本，在写文章的当下2026/07/10，最新版本是 2026.0.1
sudo pacman -Sy intel-oneapi-toolkit
# AUR 有比较旧的版本，如2025.3，可以用你喜欢的AUR helper下载
# paru
paru -Sy intel-oneapi-basekit-2025
# 或者yay
yay -Sy intel-oneapi-basekit-2025

source /opt/intel/oneapi/setvars.sh

#include <sycl/sycl.hpp>
#include <iostream>

int main() {
    // 获取系统中所有的平台
    auto platforms = sycl::platform::get_platforms();

    std::cout << "========================================\n";
    std::cout << "       System SYCL Device List          \n";
    std::cout << "========================================\n";

    for (const auto& platform : platforms) {
        std::cout << "Platform: " << platform.get_info<sycl::info::platform::name>() << "\n";
        
        // 获取平台下的所有设备
        auto devices = platform.get_devices();
        for (const auto& device : devices) {
            std::cout << "  |- Device: " << device.get_info<sycl::info::device::name>() << "\n";
            std::cout << "  |   - Type: ";
            if (device.is_gpu()) std::cout << "GPU\n";
            else if (device.is_cpu()) std::cout << "CPU\n";
            else std::cout << "Other / Accelerator\n";
            
            std::cout << "  |   - Max Compute Units: " 
                      << device.get_info<sycl::info::device::max_compute_units>() << "\n";
            std::cout << "  |   - Global Memory Size: " 
                      << device.get_info<sycl::info::device::global_mem_size>() / (1024 * 1024) << " MB\n";
        }
    }
    std::cout << "========================================\n";
    return 0;
}

#include <sycl/sycl.hpp>
#include <iostream>
#include <vector>

int main() {
    // 1. 创建队列。这里显式指定了 gpu_selector_v，如果找不到 GPU 则会抛出异常
    // 你也可以使用 default_selector_v 让运行时自动挑选设备
    sycl::queue q{sycl::gpu_selector_v};

    std::cout << "Running on device: " 
              << q.get_device().get_info<sycl::info::device::name>() << "\n";

    const size_t N = 1000000; // 向量长度：100万
    
    // 2. 使用 USM 共享内存分配
    // malloc_shared 分配的内存在 Host 和 Device 均可直接读写，运行时自动同步
    int* a = sycl::malloc_shared<int>(N, q);
    int* b = sycl::malloc_shared<int>(N, q);
    int* c = sycl::malloc_shared<int>(N, q);

    // 3. 在 Host 上初始化数据
    for (size_t i = 0; i < N; ++i) {
        a[i] = static_cast<int>(i);
        b[i] = static_cast<int>(i * 2);
    }

    // 4. 提交 Kernel 并行计算
    // parallel_for 会在 GPU 硬件的多计算单元上发射 N 个硬件线程执行 Lambda 表达式
    q.parallel_for(sycl::range<1>(N), [=](sycl::id<1> idx) {
        c[idx] = a[idx] + b[idx];
    }).wait(); // wait() 会阻塞 Host 线程，直到 GPU 计算任务全部完成

    // 5. 验证结果
    bool success = true;
    for (size_t i = 0; i < N; ++i) {
        if (c[i] != static_cast<int>(i * 3)) {
            success = false;
            std::cout << "Error at index " << i << ": Expected " << i * 3 << ", got " << c[i] << "\n";
            break;
        }
    }

    if (success) {
        std::cout << "Verification SUCCESS! 1,000,000 element vector add completed on GPU.\n";
    } else {
        std::cout << "Verification FAILED.\n";
    }

    // 6. 释放 USM 内存
    sycl::free(a, q);
    sycl::free(b, q);
    sycl::free(c, q);

    return 0;
}

icpx -fsycl sycl_info.cpp -o sycl_info
icpx -fsycl vector_add.cpp -o vector_add

icx /Fsycl sycl_info.cpp /Fe:sycl_info.exe
icx /Fsycl vector_add.cpp /Fe:vector_add.exe

========================================
       System SYCL Device List
========================================
Platform: Intel(R) oneAPI Unified Runtime over Level-Zero
  |- Device: Intel(R) Arc(TM) A770 Graphics
  |   - Type: GPU
  |   - Max Compute Units: 512
  |   - Global Memory Size: 15932 MB
Platform: Intel(R) OpenCL
  |- Device: Intel(R) Xeon(R) CPU E5-2696 v3 @ 2.30GHz
  |   - Type: CPU
  |   - Max Compute Units: 36
  |   - Global Memory Size: 49056 MB
========================================