|
2b6e8416
|
2011-06-29T23:38:16
|
|
Use a buffer of up to 512 * 4 integers when retrieving work from the GPU.
This allows each local thread id to have one slot to put any positive results into, thus making overlapping results far less likely.
Thus races will be much rarer, allowing more threads.
It should also pick up blocks close to each other more reliably and hopefully decrease the number of rejects and opencl errors.
Do the search over the buffer entirely in a separate thread to allow the GPU to stay as busy as possible.
Detach threads from themselves to prevent unlucky even where dereferencing occurs by freeing the data that stores the thread info.
|
|
|
2dbb3944
|
2011-06-27T22:05:03
|
|
Base was being set wrongly meaning we were repeating searches and the rate was actually lower than displayed :(
Tweak Ma with new changes.
Change default vectors to 2 since it's faster than 4 even when 4 is reported as preferred.
|
|
|
623b9b9f
|
2011-06-27T12:45:03
|
|
Patch bitalign separately from bfi_int.
Recover from failing to patch for bfi int.
|
|
|
4257deaf
|
2011-06-23T23:14:47
|
|
Convert abcd... to an array.
|
|
|
75cf5ccd
|
2011-06-23T23:04:34
|
|
Replace Ws with an array.
|
|
|
19eea906
|
2011-06-23T17:50:37
|
|
Implement code detecting max work size and optimal vector width.
Use this to patch the kernel to suit the idea values for the card.
Then use these values when invoking the kernel.
|
|
|
f54d2cc0
|
2011-06-22T23:07:30
|
|
Make poclbm use 4 vectors and decrease worksize to keep pipelines fullish.
Make it possible to have 0 CPU threads and update docs.
Fix counter with no cpu threads.
|
|
|
b4d2733c
|
2011-06-22T16:47:34
|
|
Convert to poclbm kernel.
|
|
|
8253f141
|
2011-06-23T23:38:04
|
|
Use some line breaks in the kernel.
|