Create local array dynamic inside OpenCL kernel

问题:

I have a OpenCL kernel that needs to process a array as multiple arrays where each sub-array sum is saved in a local cache array.

For example, imagine the fowling array:

[[1, 2, 3, 4], [10, 30, 1, 23]]

• Each work-group gets a array (in the exemple we have 2 work-groups);

• Each work-item process two array indexes (for example multiply the value index the local_id), where the work-item result is saved in a work-group shared array.

  __kernel void test(__global int **values, __global int *result, const int array_size){
      __local int cache[array_size];
  
      // initialise
      if (get_local_id(0) == 0){
          for (int i = 0; i < array_size; i++)
              cache[i] = 0;
      }
  
      barrier (CLK_LOCAL_MEM_FENCE);
  
      if(get_global_id(0) < 4){
          for (int i = 0; i<2; i++)
              cache[get_local_id(0)] += values[get_group_id(0)][i] * 
                                                           get_local_id(0);
      }
  
      barrier (CLK_LOCAL_MEM_FENCE);
  
      if(get_local_id(0) == 0){
          for (int i = 0; i<array_size; i++)
              result[get_group_id(0)] += cache[i];
      }
  }
  

The problem is that I can not define the cache array size by using a kernel parameter, but i need to in order to have a dynamic kernel.

How can I create it dynamically? like malloc function in c...

Or the only solution available is to send a temp array to my kernel function?

回答1:

This can be achieved by adding __local array as a kernel parameter:

__kernel void test(__global int **values, __global int *result, 
    const int array_size, __local int * cache)

and providing desired size of the kernel parameter:

clSetKernelArg(kernel, 3, array_size*sizeof(int), NULL);

The local memory will be allocated upon the kernel invocation. Note, that extra checks may be necessary to ensure that required local memory size does not exceed the device limit.