Search:

Type: Posts; User: Dithermaster

Page 1 of 11 1 2 3 4

Search: Search took 0.01 seconds.

  1. Replies
    4
    Views
    293

    Seems like you'll need to use the cast to int, do...

    Seems like you'll need to use the cast to int, do xor, cast to ptr route.
  2. Replies
    4
    Views
    293

    Why not share the error the compiler threw? It...

    Why not share the error the compiler threw? It would be helpful. Was it that "1" isn't the same type as "p"? If so, perhaps `*(p ^ (__global uint *)1)` would work. If not, try casting the pointer to...
  3. It is defined by hardware, and reflected to...

    It is defined by hardware, and reflected to application by driver. Some hardware could trade cache size for local memory size.

    You can find the values for your device by running clInfo application.
  4. Some of NVIDIA's post-1.2 features are indicated...

    Some of NVIDIA's post-1.2 features are indicated by clGetDeviceInfo / CL_DEVICE_EXTENSIONS strings, such as cl_nv_copy_opts which was used when their driver was at 1.1 to indicate a 1.2 feature (a...
  5. You counts seems right. However, unless the...

    You counts seems right. However, unless the contents buffer are changing for every use, there is no need to transfer them over and over, just leave them on the device and use them repeatedly. Since...
  6. tdchen, I don't know. It sounds like a timeline...

    tdchen, I don't know. It sounds like a timeline profiling tool would be helpful.
  7. The maximum size of all parameters has to be less...

    The maximum size of all parameters has to be less than 1K or 2K (I'm not going to check the spec, but it's something like that). The setup for such a transfer costs more than the transfer itself. I'd...
  8. With CL_TRUE it must block; with CL_FALSE it is...

    With CL_TRUE it must block; with CL_FALSE it is allowed to not block. But it might still block. It depends on your platform and device, which you did not mention.
  9. My mental model is that clSetKernelArg just...

    My mental model is that clSetKernelArg just changes data on host side, and clEnqueueNDRangeKernel is what sends it to the GPU queue. My assumption is that a copy of the parameters are made during...
  10. You can try to create a CL/GL shared context and...

    You can try to create a CL/GL shared context and if it works they are the same device. But if they are not we have seen crashes. So we at least don't try across non-matching vendors. The way we do...
  11. Expect it to be from non-main-thread. Or, to...

    Expect it to be from non-main-thread. Or, to think of it another way, driver actually has no way to call you from main thread, if you never call it from main thread.
  12. Replies
    4
    Views
    474

    It also depends on if they use the same resources...

    It also depends on if they use the same resources (for example, if kernelA writes to a buffer than kernelB takes, the runtime should not run them in parallel). A given vendor might make you jump...
  13. You might try older NVIDIA tools. They've been...

    You might try older NVIDIA tools. They've been slowly removing OpenCL support in newer versions of the tools.
  14. > 1)clSetKernelArg also transfer data object to...

    > 1)clSetKernelArg also transfer data object to device, right?
    No, it doesn't cause data tranfers. It's just used to pass the handle to the cl_mem object.

    > 2)Does OpenGL have a function like...
  15. > 1) When the data Object was really transfered...

    > 1) When the data Object was really transfered between main momory and opencl devices

    During clEnqueueRead/Write or clEnqueueMap/Unmap operations.

    > 2)How to deal with the GPU used by the OS?...
  16. As long as the jobs aren't too small it won't run...

    As long as the jobs aren't too small it won't run slower. Jobs are subdivided by the runtime into what the hardware can do, so once it is doing that larger jobs run at the same speed (ignoring the...
  17. Divide your work up into smaller bites and feed...

    Divide your work up into smaller bites and feed them to the GPUs at the rate they can eat them. In more detail: use OpenCL Events with each clEnqueueNDRangeItem. Enqueue 3 jobs to each GPU. As jobs...
  18. The documentation...

    The documentation (https://www.khronos.org/registry/OpenCL/sdk/1.0/docs/man/xhtml/clEnqueueReadBuffer.html) says that clEnqueueReadBuffers can return CL_INVALID_VALUE "if the region being read...
  19. Replies
    5
    Views
    572

    Did you make sure your platform & device support...

    Did you make sure your platform & device support pipes before calling that API? Maybe it's an OpenCL 1.2 device, which doesn't support pipes.
  20. 1) Both Intel and AMD have CPU drivers that work...

    1) Both Intel and AMD have CPU drivers that work on both Intel and AMD CPUs. Try them both to see which one you like better.
    2) Every vendor supplies an OpenCL.dll but if you already have one they...
  21. Replies
    1
    Views
    700

    It depends on the particular OpenCL runtime, but...

    It depends on the particular OpenCL runtime, but yes, in general, when you run an OpenCL kernel on CPU, it will run on all cores, and in some cases will also take advantage of vector instructions...
  22. Replies
    4
    Views
    3,922

    We use an ancient AMD SDK for just that reason:...

    We use an ancient AMD SDK for just that reason: it keeps us OpenCL 1.1/1.2 "clean" (no 2.x APIs because they are not in the header) and the .lib works with old and new OpenCL.dll instances. You only...
  23. Replies
    1
    Views
    744

    The Khronos C++ header works with Xcode, but it's...

    The Khronos C++ header works with Xcode, but it's only a C++ wrapper around the OpenCL host API. To use C++ kernels, you need OpenCL 2.x, which Apple does not support.
  24. When you say "if I run my kernel code in a loop...

    When you say "if I run my kernel code in a loop until 1/60 seconds have elapsed" what you are really doing is _enqueueing_ as many kernels as you can in 1/60 of a second. Then you try to read the...
  25. Replies
    8
    Views
    938

    Why do you think it would be costly?

    Why do you think it would be costly?
Results 1 to 25 of 273
Page 1 of 11 1 2 3 4
Proudly hosted by Digital Ocean