Search:

Type: Posts; User: jbasic

Page 1 of 2 1 2

Search: Search took 0.00 seconds.

  1. Replies
    9
    Views
    5,297

    Re: Invalid command queue / access violation

    Thanks for the ideas guys. Build goes completely successful. I always Finish after each OpenCL action, for debugging purposes. david.garcia, as others read correctly, queue goes invalid while...
  2. Replies
    9
    Views
    5,297

    Invalid command queue / access violation

    For the perfectly correct kernel (written by nVidia), I'm getting the "invalid command queue" error while waiting kernel to finish. The same code run on CPU returns "access violation writing at...
  3. Re: Max __constant variables defined in program source

    My constants are defined in global program source e.g.: __constant int something = 1;
    It sucks if I'm limited to use only CL_DEVICE_MAX_CONSTANT_ARGS of these from kernels, because they are not in...
  4. Replies
    3
    Views
    2,281

    Re: clSetKernelArg performance

    Ok, so global (and constant memory) is persistent across kernel invocations.

    But is this the case with simple arguments (that are copied to private memory) also ?
  5. Replies
    3
    Views
    2,281

    clSetKernelArg performance

    Since I simulate things, basically I loop running same kernels over and over. Large amount of those kernels don't change their argument values. For now I always set all the arguments before...
  6. Re: Max __constant variables defined in program source

    coleb, true, I pasted wrong enum.
    matrem, I know constant variables reside in global memory and are cached, I just though maybe there was another limitation. Thanks.
  7. Max __constant variables defined in program source

    CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE can be used to get the max number of arguments declared with the __constant qualifier in a
    kernel. Also, max size of a constant buffer can be retrieved...

    But...
  8. Replies
    8
    Views
    5,469

    Re: Accesing both Amd/nVidia OpenCL

    Now that both nVidia and AMD implementation export clean undecorated (cdecl calling convention) functions, it could be possible with dynamic loading of wanted implementation library. Take a look at...
  9. Replies
    2
    Views
    2,735

    Re: Dynamically allocated shared memory

    You set your local variable as a kernel argument:

    __kernel void Something(..., __local float *var, ...)
    And then from host allocate the local memory with clSetKernelArg that has last parameter...
  10. Re: CL_DEVICE_MAX_WORK_ITEM_SIZES for CL_DEVICE_TYPE_CPU

    Hmm, this is interesting...
    But why does my crappy P4 2.8 CPU (ATI OpenCL implementation) report:

    CL_DEVICE_MAX_COMPUTE_UNITS: 1
    CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS: 3...
  11. Replies
    9
    Views
    4,175

    Re: LLVM compilation failure -- why?

    The error I can see here is the declaration one of your parameters: const uint *filter.
    Pointer argument always must point to address space global, local, or constant. So try with: constant uint...
  12. Replies
    2
    Views
    3,130

    Re: OpenCL function syntax

    Unfortunately variable references, like in your example, are not available. You must use pointers:


    void doSomething(float4 *a, float4 *b)
    {
    *a = ...
    (*b).x = ...
    }
    Note that you must...
  13. Replies
    4
    Views
    3,474

    Re: passing constant value to kernel

    __constant variables are allocated in global memory, and the constant memory space is cached. Accessing private memory is faster and should remain as only option where to store function arguments (as...
  14. Re: Local memory access question in the specification

    Host cannot read nor write local memory. From host you can only allocate amount of local memory you need in work-groups.
    Of course if you try to allocate some local memory and device runs out of...
  15. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    Yes, it's is reasonable, nVidia's example confirms it.
  16. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    Thanks, I see what you mean... Calling clCreateKernel for the *same* kernel *multiple* times, gives you *different* cl_kernel instances. Now with different instances, you can set args without...
  17. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    I know, you're not following me. To repeat myself one more time, nVidia's example acts contradictory to what we said.


    Contradiction: reduceKernel[i] = clCreateKernel(cpProgram, "reduce",...
  18. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    So clSetKernelArg followed by enqueueing the kernel on specific devices tells OpenCL which device should get the data set by clSetKernelArg... interesting, thought unlogical and never mentioned in...
  19. Replies
    3
    Views
    4,199

    Re: Constant Memory latency

    Paul,

    nVidia's OpenCL best practices guide 3.2.5 Constant Memory is saying:

    Also I recommend reading Dr.Dobb's "CUDA, Supercomputing for the Masses" whole article,...
  20. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    I've encountered on one more obstacle in specifications, even when trying to implement "secure" data transfer method with multiple buffers.
    Since clCreateKernel returns one object for all devices...
  21. Re: Who will write the platform -> driver interface specs ?

    I agree, but I'm afraid the process will take some time.
    Dynamic loading with dlopen on Linux, that is LoadLibrary on Windows could be a temporary solution, but is a very slack one.
  22. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    So the proper way that devices will get the right read-only buffer data would be to use clEnqueueWrite and clFinish for all command queues (all devices) using the same cl_mem object.

    On more...
  23. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    Since I manage quasi-splitting (offsetting) by myself, it's not problem for me to enqueue read multiple times - for each command queue one read: the part only A was writing, and then the part B was...
  24. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    Thanks for the hints.
    I was thinking the same offset trick. I just hope OpenCL is smart enough to manage 1 read/write buffer on all devices. I promise every device will write on their own buffer...
  25. Replies
    19
    Views
    6,369

    Re: Running the same kernel on multiple devices

    I was afraid of that. I assume this also applies for SLI and Crossfire networked devices?

    For my problem, logical solution requires the readability of whole memory inside kernel. It's not fun to...
Results 1 to 25 of 35
Page 1 of 2 1 2