Search:

Type: Posts; User: mangpo

Search: Search took 0.00 seconds.

  1. Replies
    1
    Views
    694

    Change arguments before start running

    I'm curious about the behavior of opencl when I do something like this.

    In my program, I have one opencl kernel, and I want to call it twice with different arguments. I use non-blocking read and...
  2. Re: OpenCL doesn't work with multiple host threads?

    I figure it out. There is actually a problem with incompatibility between the library that my group implemented and AMD/intel SDK. We suspected that they use new and malloc/ delete/free...
  3. OpenCL doesn't work with multiple host threads?

    I write an OpenCL program for running on CPUs, so basically the host and the device are essentially the same. When I run my program using 1 CPU thread, everything works fine, but when I use more than...
  4. Replies
    7
    Views
    2,006

    Re: No speed up from using 2 GPUs

    That works! Now it runs faster on 2 GPUs. Thank you so much.
  5. Replies
    7
    Views
    2,006

    Re: No speed up from using 2 GPUs

    Okay. I'll try that, and I'll let you know the result.
  6. Replies
    7
    Views
    2,006

    Re: No speed up from using 2 GPUs

    I did use clGetEventProfilingInfo() to measure those time. Only total time that I used gettimeofday.

    Here are my code.
    1 command queue on 1 device:


    ciErrNum = clGetDeviceIDs(cpPlatform,...
  7. Re: Is write/read buffer a work on cpu or gpu side?

    Can you give me examples?
  8. Is write/read buffer a work on cpu or gpu side?

    When I execute clEnqueueReadBuffer/clEnqueueWriteBuffer, does CPU or GPU handle the work (i.e. which device perform the copy when it is necessary)?

    If CPU handle that job, what is it doing when I...
  9. Replies
    7
    Views
    2,006

    Re: No speed up from using 2 GPUs

    For overall time, I use gettimeofday(&time, NULL). I put make a call before creating command queue and after reading output buffers are done. For the kernel runtime on gpu, I use:
    ...
  10. Replies
    1
    Views
    1,616

    Re: if and while loops better on gpu or cpu

    I think it depends. If all workthreads' control flows are the same, then running on gpu can be faster. For example, if the number of iteration of the loop is the same acroos all the workthread. If...
  11. Replies
    7
    Views
    2,006

    No speed up from using 2 GPUs

    I just got 2 new GPUs yesterday. They are both NVIDIA C2070. I wrote a simple program to compare the runtime of using 1 GPU and 2 GPUs. Surprisingly, 2 GPUs don't give me any speedup. Basically, I...
  12. Replies
    1
    Views
    787

    Tool for viewing threads' information

    A few topic in this forum and CUDA forum discuss about CPU busy wait after clEnqueueNDRangeKernel call. People said the busy wait happen occurs on some devices. Someone said OpenCL creates a...
  13. Re: Non-blocking write buffer problem with multiple contexts

    It's fixed. The bug is on my side. Nothing to do with gpu. Thank you so much for helping anyway.
  14. Re: What's going to happen if not releasing kernels

    Leak until my application is terminated or until I reboot my machine?
  15. Re: What's going to happen if not releasing kernels

    Will using smart pointers increase the runtime of the application?
  16. Re: Non-blocking write buffer problem with multiple contexts

    Yeah, sorry I didn't make it clear. By program, I mean application. One application has one context and one command queue (but it is multi-threading), and only one thread (gpu manager thread) invokes...
  17. Re: What's going to happen if not releasing kernels

    Okay. Thanks. I'm just wondering if I really need to because it requires a decent amount of work to find the right place to release kernels. (I'm implementing a compiler that generate OpenCL if you...
  18. Re: Non-blocking write buffer problem with multiple contexts

    Thank you so much for responding.


    Thanks for clarification, and sorry for misusing the term.


    Yes, I do know that. However, each program has its own context, and the programs are...
  19. What's going to happen if not releasing kernels

    I wonder if there is something bad about not releasing cl_program and cl_kernel on a gpu device. At the end of my program I do release cl_command_queue and cl_context, but I don't release those...
  20. Non-blocking write buffer problem with multiple contexts

    I encounter the problem with non-blocking clEnqueueWriteBuffer when I use multiple contexts concurrently. Within a program I run, there is one in-order-execution cl_command_queue and one cl_context....
Results 1 to 20 of 20