|
86d5377e
|
2012-08-23T10:48:15
|
|
Use atomic ops to never miss a nonce on opencl kernels, including nonce==0, also allowing us to make the output buffer smaller.
|
|
|
1b5c676d
|
2012-07-21T11:00:36
|
|
Use 256 output slots for kernels to allow 1 for each worksize.
|
|
|
02e126f4
|
2011-08-23T10:28:30
|
|
The worksize was unintentionally changed back to 4k by mistake, this caused a slowdown.
|
|
|
bd79a61c
|
2011-08-19T17:20:49
|
|
Move poclbm to new branch optimisation as well.
|
|
|
cf54f9b8
|
2011-08-17T16:07:15
|
|
Move to 256 sized buffers and don't risk overwrite by using only 127 mask.
|
|
|
0f782ba6
|
2011-08-17T15:47:18
|
|
Update poclbm kernel to FF sized mask and only check that range.
|
|
|
eea05c05
|
2011-07-15T13:04:25
|
|
Update kernel with a shorter output path, and use 4k output buffer to match OS page sizes.
|
|
|
cb13e2cf
|
2011-07-05T19:47:03
|
|
Make it possible to build without opencl for cpu mining only.
|
|
|
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.
|
|
|
e1dd27c5
|
2011-06-29T11:19:43
|
|
Ensure that we don't overflow due to 32 bit limitations.
|
|
|
a45c54aa
|
2011-06-27T11:31:05
|
|
Make postcalc_hash asynchronous as well.
|
|
|
f117675a
|
2011-06-22T10:15:23
|
|
Optimise work loop to make cl calls asynchronous where possible.
|
|
|
6b77d850
|
2011-06-17T14:00:41
|
|
Fixes.
|
|
|
dde70397
|
2011-06-14T10:32:54
|
|
Merge gpumining from oclmine. Unstable.
|