OpenCL中的NDRangeKernel函数中的局部大小错误

当我通过乘以2来增加局部大小值时，虽然globalsize/localsize是一个整数，但出现错误。我无法弄清楚问题所在。OpenCL中的NDRangeKernel函数中的局部大小错误

Kernel: 
    __kernel void add(__global float *a, 
        __global float *b, 
        __global float *answer, 
        __local float *shared, 
        __local float *result) 
    { 
     int gid = get_global_id(0); 
     int lid = get_local_id(0); 
     int lsize = get_local_size(0); 

     float tempa, tempb; 
     shared[lid] = a[gid]; 
     shared[lid + lsize] = b[gid]; 
     barrier(CLK_LOCAL_MEM_FENCE); 

     for(int k = 0; k < lsize; k++){ 
     tempa = shared[lid + k]; 
     tempb = shared[lid + lsize + k]; 
     result[lid + k] = tempa + tempb; 
     } 
     barrier(CLK_LOCAL_MEM_FENCE); 

     answer[gid] = result[lid]; 
    }

可以说本地大小是2^n。如果该程序起作用，否则它崩溃。用C 主机代码：

size_t global_work_size = n; //n = 1000000 
size_t local_work_size = 32; 
size_t sharedSize = (2 * local_work_size) * sizeof(float); 
size_t resultSize = local_work_size * sizeof(float); 

err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &cmDevBufInA); //HERE 
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &cmDevBufInB); 
err |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &cmDevBufOut); 
err |= clSetKernelArg(kernel, 3, sharedSize, NULL); 
err |= clSetKernelArg(kernel, 4, resultSize, NULL); 
assert(err == CL_SUCCESS); 

//EXECUTION AND READ 
cl_event calculation; 


err = clEnqueueNDRangeKernel(cmd_queue, kernel, 1, NULL, &global_work_size, &local_work_size,0, NULL, &calculation); 
assert(err == CL_SUCCESS); 
clFinish(cmd_queue);

来源

2012-03-28 Shnkc

当你说“本地大小”时，你不明白你指的是什么。你是指变量'local_work_size'？ – 2012-03-30 07:35:06

是的。分别是global_work_size是全局大小。抱歉不匹配。 – Shnkc 2012-03-30 21:52:44

什么的 '盖子+ LSIZE + K' 的最大值 tempb =共享[盖子+ LSIZE + K];

盖子= 31 LSIZE = 32 K = 32

但共享被分配作为

为size_t sharedSize =（2 * local_work_size）*的sizeof（浮点）;

来源

2012-04-01 02:39:24

OpenCL中的NDRangeKernel函数中的局部大小错误

回答

相关问题