OpenCL内核中赋值给uint32_t*导致程序挂起问题求助

我在尝试用OpenCL从私钥生成公钥时，内核执行出现挂起，已将问题定位到secp256k1_fe_sqr_inner函数（无需关注其具体密码学功能）：

#define uint32_t uint
#define uint64_t ulong

static void secp256k1_fe_sqr_inner(uint32_t *r, const uint32_t *a) {
    uint64_t c, d;
    uint64_t u0, u1, u2, u3, u4, u5, u6, u7, u8;
    uint32_t t9, t0, t1, t2, t3, t4, t5, t6, t7;
    const uint32_t M = 0x3FFFFFFUL, R0 = 0x3D10UL, R1 = 0x400UL;

    d  = (uint64_t)(a[0]*2) * a[9]
       + (uint64_t)(a[1]*2) * a[8]
       + (uint64_t)(a[2]*2) * a[7]
       + (uint64_t)(a[3]*2) * a[6]
       + (uint64_t)(a[4]*2) * a[5];
    t9 = d & M; d >>= 26;
    c  = (uint64_t)a[0] * a[0];
    d += (uint64_t)(a[1]*2) * a[9]
       + (uint64_t)(a[2]*2) * a[8]
       + (uint64_t)(a[3]*2) * a[7]
       + (uint64_t)(a[4]*2) * a[6]
       + (uint64_t)a[5] * a[5];
    u0 = d & M; d >>= 26; c += u0 * R0;
    t0 = c & M; c >>= 26; c += u0 * R1;
    c += (uint64_t)(a[0]*2) * a[1];
    d += (uint64_t)(a[2]*2) * a[9]
       + (uint64_t)(a[3]*2) * a[8]
       + (uint64_t)(a[4]*2) * a[7]
       + (uint64_t)(a[5]*2) * a[6];
    u1 = d & M; d >>= 26; c += u1 * R0;
    t1 = c & M; c >>= 26; c += u1 * R1;
    c += (uint64_t)(a[0]*2) * a[2]
       + (uint64_t)a[1] * a[1];
    d += (uint64_t)(a[3]*2) * a[9]
       + (uint64_t)(a[4]*2) * a[8]
       + (uint64_t)(a[5]*2) * a[7]
       + (uint64_t)a[6] * a[6];
    u2 = d & M; d >>= 26; c += u2 * R0;
    t2 = c & M; c >>= 26; c += u2 * R1;
    c += (uint64_t)(a[0]*2) * a[3]
       + (uint64_t)(a[1]*2) * a[2];
    d += (uint64_t)(a[4]*2) * a[9]
       + (uint64_t)(a[5]*2) * a[8]
       + (uint64_t)(a[6]*2) * a[7];
    u3 = d & M; d >>= 26; c += u3 * R0;
    t3 = c & M; c >>= 26; c += u3 * R1;
    c += (uint64_t)(a[0]*2) * a[4]
       + (uint64_t)(a[1]*2) * a[3]
       + (uint64_t)a[2] * a[2];
    d += (uint64_t)(a[5]*2) * a[9]
       + (uint64_t)(a[6]*2) * a[8]
       + (uint64_t)a[7] * a[7];
    u4 = d & M; d >>= 26; c += u4 * R0;
    t4 = c & M; c >>= 26; c += u4 * R1;
    c += (uint64_t)(a[0]*2) * a[5]
       + (uint64_t)(a[1]*2) * a[4]
       + (uint64_t)(a[2]*2) * a[3];
    d += (uint64_t)(a[6]*2) * a[9]
       + (uint64_t)(a[7]*2) * a[8];
    u5 = d & M; d >>= 26; c += u5 * R0;
    t5 = c & M; c >>= 26; c += u5 * R1;
    c += (uint64_t)(a[0]*2) * a[6]
       + (uint64_t)(a[1]*2) * a[5]
       + (uint64_t)(a[2]*2) * a[4]
       + (uint64_t)a[3] * a[3];
    d += (uint64_t)(a[7]*2) * a[9]
       + (uint64_t)a[8] * a[8];
    u6 = d & M; d >>= 26; c += u6 * R0;
    t6 = c & M; c >>= 26; c += u6 * R1;
    c += (uint64_t)(a[0]*2) * a[7]
       + (uint64_t)(a[1]*2) * a[6]
       + (uint64_t)(a[2]*2) * a[5]
       + (uint64_t)(a[3]*2) * a[4];
    d += (uint64_t)(a[8]*2) * a[9];
    u7 = d & M; d >>= 26; c += u7 * R0;
    t7 = c & M; c >>= 26; c += u7 * R1;
    c += (uint64_t)(a[0]*2) * a[8]
       + (uint64_t)(a[1]*2) * a[7]
       + (uint64_t)(a[2]*2) * a[6]
       + (uint64_t)(a[3]*2) * a[5]
       + (uint64_t)a[4] * a[4];
    d += (uint64_t)a[9] * a[9];
    u8 = d & M; d >>= 26; c += u8 * R0;
    // --> 挂起点
    r[3] = t3;
    r[4] = t4;
    r[5] = t5;
    r[6] = t6;
    r[7] = t7;
    r[8] = c & M; c >>= 26; c += u8 * R1;
    c   += d * R0 + t9;
    r[9] = c & (M >> 4); c >>= 22; c += d * (R1 << 4);
    d    = c * (R0 >> 4) + t0;
    r[0] = d & M; d >>= 26;
    d   += c * (R1 >> 4) + t1;
    r[1] = d & M; d >>= 26;
    d   += t2;
    r[2] = d;
}

r是一个大小为10的uint32_t数组，程序执行到r[3] = t3时出现挂起。内核编译通过并能启动，但触发挂起后，GPU-Z显示GPU无负载、温度处于空闲状态，仅风扇转速异常偏高。

我尝试过创建局部数组uint32_t temp[10]，按照原逻辑给temp赋值后执行r = temp，这种方式能避免挂起，但计算结果完全错误——无论输入是什么，输出都相同。

开发环境 #

• 编译工具：Visual Studio 2022
• OpenCL依赖：CUDA SDK 12.2自带的OpenCL库，同时安装Intel oneAPI用于CPU端运行OpenCL代码
• 硬件：Windows 10设备，Intel处理器 + RTX 3080移动GPU
• 编译选项：clBuildProgram中设置-cl-std=CL1.2

内容的提问来源于stack exchange，提问作者MrPuzzler