3 Replies Latest reply on Dec 29, 2014 9:09 AM by boxerab

    Question about global pointers

    boxerab

      The AMD optimization guide states:

       

      --------------------------------------------------------------------------------------------------------

      Avoid writing code with dynamic pointer assignment on the GPU. For

      example:

      kernel void dyn_assign(global int* a, global int* b, global int* c)

      {

      global int* d;

      size_t idx = get_global_id(0);

      if (idx & 1) {

      d = b;

      } else {

      d = c;

      }

      a[idx] = d[idx];

      }

      ------------------------------------------------------------------------------------------------------

       

      Now, in my case, I have the following code:

       

       

      void KERNEL run(GLOBAL const uint* restrict idata, GLOBAL uchar* restrict odata) {

         ....

       

          idata +=  get_local_id(0);

          odata += get_local_id(0)

       

         .....

      }

       

      Is this a case of dynamic pointer assignment, since I am modifying the global pointers idata and odata?

       

      i.e. should I use an index into idata, instead of modifying it?

       

      Thanks.

        • Re: Question about global pointers
          dipak

          Hi,

          If you read that section carefully, it says that compiler must be aware of the base pointer of the data buffer (see bold line below) in order to perform certain optimizations.

          This is inefficient because the GPU compiler must know the base pointer that every load comes from and in this situation, the compiler cannot determine what ‘d’ points to. So, both B and C are assigned to the same GPU resource, removing the ability to do certain optimizations.

          I guess, in your case, it is not the same as above.

           

          Regards,

            • Re: Question about global pointers
              nibal

              Hmmm, I don't see much of a difference though. Both idata & odata are unknown to the compiler. Furthermore, idata is declared constant pointer and then changed :-(. Referencing through indexes, would at least cure constant aspect.

              PS: Haven't read optimization guide, yet.

                • Re: Question about global pointers
                  boxerab

                  Thanks, guys. @nibal: the const modifier in this case indicates that what

                  is pointed to is constant, i.e. read only. The pointer itself can change.

                   

                  The following incantation:

                   

                  global uchar* const idata

                   

                  would indicate that the pointer itself is constant.

                   

                   

                  Aaron