Search:

Type: Posts; User: arthur.sw

Search: Search took 0.00 seconds.

  1. Replies
    4
    Views
    1,489

    Re: Register and local mem problems

    I just had the following answer from a professional:

    "The available number of registers is always a huge problem, most times a
    simple splitting of the algorithm into multiple kernels is the...
  2. Replies
    4
    Views
    1,489

    Re: Register and local mem problems

    Ok !

    Yes, I must initialize border data as well.
    I tried this global to local mem copy and unfortunately it did not decrease the number of register (I have currently 23 and it must go down to...
  3. Replies
    4
    Views
    1,489

    Re: Register and local mem problems

    Any idea Mr David Garcia ;-) ?
  4. Replies
    4
    Views
    1,489

    Register and local mem problems

    Hello !

    I am working on an optical flow algorithm, I developed a working 2D version and now I have troubles with the 3D version... I work with blocks of 8*8*8 pixels since I have...
  5. Replies
    8
    Views
    2,698

    Re: Block edges updates

    Yes that's right, this is the same problem. I realized it when I woke up this morning ;-)
  6. Replies
    8
    Views
    2,698

    Re: Block edges updates

    ahah, thank you :-)

    For my loop problem, I guess there is no other way than executing the kernel n times (instead of having the for loop inside the kernel).

    It works but it really slows down...
  7. Replies
    8
    Views
    2,698

    Re: Block edges updates

    Thank you very much for this clear answer :-) !
    If I understand correctly, this problem can not be fixed with barrier or mem_fence functions since it would only synchronize memory within blocks ?
    ...
  8. Replies
    0
    Views
    1,265

    Re: block edges

    Thank you very much for this clear answer :-) !
    If I understand correctly, this problem can not be fixed with barrier or mem_fence functions since it would only synchronize memory within blocks ?
  9. Replies
    8
    Views
    2,698

    Re: Block edges updates

    Well, I have 3 problems, but what I want to know is: is my code correct ? Why does the "edge update" differ from the oclMedianFilter example ?

    On of my problem is that with my previous code, I get...
  10. Replies
    8
    Views
    2,698

    Block edges updates

    Hello,

    I am working on an algorithm where each pixel needs its four neighbors.

    So for each 16*16 block of threads, I want to copy the global memory into 18*18 shared memory blocks (This way...
  11. Replies
    4
    Views
    1,127

    Re: for loop problem

    Forget the last post, I had an error.
  12. Replies
    4
    Views
    1,127

    Re: for loop problem

    On the other algorithm, my code ends by those lines:



    vLocal[li*lWidth+lj] = uLocal[li*lWidth+lj] + LAMBDA*THETA*dI1 * (rou < -LTdI2)
    - LAMBDA*THETA*dI1 * (rou > LTdI2)...
  13. Replies
    4
    Views
    1,127

    Re: for loop problem

    Thanks for your answer Chai,

    I am aware of this compiler optimization, but in my case the whole loop is skipped, it is not even executed once ! (if L=1 it works, if L>1 it is skipped)

    Moreover,...
  14. Replies
    4
    Views
    1,127

    for loop problem

    Hello,

    The following algorithm is skipped when L>1 (the number of iteration), except when the INIT part is outside the for loop :



    __kernel void test(__global float* f, __global float* u,...
Results 1 to 14 of 22