Search:

Type: Posts; User: duanmu

Search: Search took 0.00 seconds.

  1. Replies
    4
    Views
    3,888

    Re: passing constant value to kernel

    I understand your argument. However, at least in NVIDIA's implementation, constant memory is as fast as private memory as long as it is cached. So I wonder why the standard did not allow it since it...
  2. Replies
    4
    Views
    3,888

    Re: passing constant value to kernel

    Suppose we have a kernel like this


    __kernel void foo(const int val, ...)
    { ... }

    In this example, the same value of "val" is passed from the host program to each work item. But it is...
  3. Replies
    4
    Views
    3,888

    passing constant value to kernel

    The OpenCL 1.0 does not allow to pass a number, e.g. __constant int val, to kernel from the constant memory. Only pointers to constant memory space may be passed. As a result, one has to pass the...
  4. Replies
    5
    Views
    3,392

    Re: specify work group sizes

    It makes some sense. Alternatively, the standard could allow the kernel to be queued even with local_work_size=NULL as long as global_work_size is a multiple of reqd_work_group_size and makes...
  5. Replies
    5
    Views
    3,392

    Re: specify work group sizes

    Actually the NULL value won't work, at least with NVIDIA'S implementation, which is also implied by the OpenCL specification 1.0.
  6. Replies
    5
    Views
    3,392

    specify work group sizes

    One can specify work group sizes with kernel attribute reqd_work_group_size and/or with the local_work_size argument of clEnqueueNDRangeKernel(). If reqd_work_group_size is specified, then...
  7. Replies
    99
    Views
    68,185

    Re: OpenCL C++ Bindings

    Great job! However, Program::getInfo() does not work (e.g., CL_PROGRAM_BINARY_SIZES) if the context has multiple devices.
Results 1 to 7 of 8