Databases
StackOverflow2013
Postsdbo
PO
 

Note that there are some explanatory texts on larger screens.

plurals
  1. PO
    primarykey
    data
    text
    <p>Well, here are some thoughs:</p> <p>You perform ~640*480 iterations of <code>numpy.dot</code>, which itself processes 6*6*32 elements. Parallelizing dot-product barely worth it: 192 parallel threads is not enough for GPU, and reduction on CUDA is additional troubles. So, IMO, the best way to parallelize you task is to assign one element of output array to each thread.</p> <p>Now about memory: output array will be in global memory, there is not much choice. For input data, <code>A</code> looks quite good for texture memory, since adjacent threads access adjacent elements. Alternatively, you can manually "cache" it in shared memory, but in this case it does not look much advantageous over simply using texture. For <code>B</code>, shared memory is not good, since it would cause bank conflicts, since when you calculate dot-product, all threads in half-warp access the same B's element (you can start summation from different elements in different threads, but that's (again) doesn't look promising). So the choice is either texture or constant. I vote for constant, since (a) constant memory is suited for data which is accessed by all threads on the device, (b) you won't pollute texture cache.</p> <p>The above is just my guesses, and to actually achieve good performance you better try out different variants...</p> <p><strong>Update regarding your naive implementation</strong> </p> <pre><code>for (int Yi = 0; Yi &lt; Ydims[0]; Yi++ ) </code></pre> <p>Here, you do aceess to a global memory on each iteration. That's a <strong>huge</strong> performance killer. Since you have 3 dimensions, you better replace your <code>int *Ydims</code> with <code>int3 Ydims</code> (same for <code>Xdims</code> and <code>outdims</code>).</p> <pre><code>out[out_indx] += X[X_indx]*Y[Y_indx]; </code></pre> <p>Again, a very bad idea. Create a register variable and do all operations with it. Write to a global array only once at the end of a kernel.</p> <p>These optimizations are first thing you should do. Second thing is to make you <code>X</code> and <code>Y</code> 3D textures, so access to them will be cached. I guess, after this CUDA would outperform CPU.</p> <p>For further optimisations, you'd better read <a href="http://developer.download.nvidia.com/compute/cuda/4_0/toolkit/docs/CUDA_C_Best_Practices_Guide.pdf" rel="nofollow">CUDA C Best Practices Guide</a>. It's must read, and you would get much better idea of how to write efficient GPU code (right now your implementation is toooo naive)</p>
    singulars
    1. This table or related slice is empty.
    1. POGeneralized sliding-window computation on the GPU
      singulars
      1. PTQuestion
    1. PTAnswer
    1. USaland
    1. USaland
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. VO
      singulars
      1. PO
      1. This table or related slice is empty.
      1. VTUpMod
    1. COThanks! Tried your suggestion and mapped each output pixel to a single thread. Haven't attempted to do any memory optimization. The results are mixed so far.
      singulars
      1. PO
      1. USBrianTheLion
    2. COWow, awesome help! From what I can tell, kernel parameters are stored in local memory and local memory is off-chip. Is there any way I can get outdims, Xdims, and Ydims to on-chip memory?
      singulars
      1. PO
      1. USBrianTheLion
    3. CO@BrianTheLion Nope, the kernel parameters are stored in on-chip shared memory, which is usually nearly as fast as registers. You might be confusing OpenCL'ish local memory, which is the same as CUDA'ish shared, and CUDA'ish local, which is actually just a part of off-chip global memory.
      singulars
      1. PO
      1. USaland
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload