|
28880d0d
|
2011-08-13T20:54:20
|
|
Move the non cl_ variables into the cgpu info struct to allow creating a new cl state on reinit, preserving known GPU variables.
Create a new context from scratch in initCQ in case something was corrupted to maximise our chance of succesfully creating a new worker thread.
|
|
|
071a0ad2
|
2011-08-12T13:00:25
|
|
Preinitialise the devices only once on startup.
|
|
|
8f186e61
|
2011-07-30T16:59:54
|
|
Restart threads by abstracting out the clcontext initialisation and using that instead of probing all cards.
|
|
|
4365896b
|
2011-07-29T10:17:36
|
|
Release the command queue created after we've copied the binary data.
|
|
|
283d5d23
|
2011-07-29T10:09:24
|
|
Create a command queue from the program created from source which allows us to flush the command queue in the hope it will not generate a zero sized binary any more.
|
|
|
2e37e337
|
2011-07-24T10:58:03
|
|
Out of order command queue may fail on osx. Try without if it fails.
|
|
|
4cd12aa8
|
2011-07-24T09:04:56
|
|
Fix harmless warning.
|
|
|
a9e1a255
|
2011-07-23T15:15:46
|
|
Make it possible to select the choice of kernel on the command line.
|
|
|
116a9dc0
|
2011-07-23T14:17:25
|
|
Update phatk kernel to one with new parameters for slightly less overhead again.
Make the queue kernel parameters call a function pointer to select phatk or poclbm.
|
|
|
1c67f606
|
2011-07-21T10:07:29
|
|
Sometimes the cl compiler generates zero sized binaries and only a reboot seems to fix it.
|
|
|
7b13812e
|
2011-07-21T09:58:28
|
|
Kernels are safely flushed in a way that allows out of order execution to work.
|
|
|
a7707a26
|
2011-07-18T10:42:24
|
|
Rename the poclbm file to ensure a new binary is built since.
|
|
|
eea05c05
|
2011-07-15T13:04:25
|
|
Update kernel with a shorter output path, and use 4k output buffer to match OS page sizes.
|
|
|
857902a1
|
2011-07-12T22:23:03
|
|
Commit a new phatk kernel renamed to force new binary building and add proper support in makefiles.
|
|
|
0c910673
|
2011-07-10T00:30:12
|
|
Set max preferred size to 256 to prevent lying cards from crashing when no worksize is set.
|
|
|
826cc480
|
2011-07-08T11:58:04
|
|
Opcode should be ULL.
|
|
efebee5a
|
2011-07-06T16:47:29
|
|
Fix the case where there are no GPUs, and exit if they give errors.
If there are no GPUs, set nDevs to 0 not -1 (status is set to an
unhelpful -1001 here on my laptop, so we can't rely on a particular
status value).
Also, if nDevs is -1, exit rather than screwing up later.
|
|
a636a674
|
2011-07-05T21:31:41
|
|
Merge branch 'cgminer' of git://github.com/ckolivas/cgminer into cgminer
|
|
|
52d6e7ca
|
2011-07-05T21:31:24
|
|
Fixed fread issues under Windows.
|
|
|
a93b22c6
|
2011-07-05T17:34:54
|
|
Make it possible to build without GPU mining by picking up HAVE_OPENCL from config.h.
|
|
|
5d301c8b
|
2011-07-02T10:22:09
|
|
Make a binary load failure build from source.
|
|
|
13b43cfa
|
2011-07-03T00:28:51
|
|
Update copyright and authors.
|
|
|
821da37c
|
2011-07-04T13:49:28
|
|
Add hardware name to binary kernel name allowing for unique kernels for different cards on the same machine.
|
|
|
594b38b8
|
2011-07-02T13:46:17
|
|
Fix redefinition of gnu source.
|
|
|
ec831917
|
2011-06-25T04:43:37
|
|
Build on windows using mingw32.
|
|
|
3aa5be4f
|
2011-07-01T01:14:43
|
|
Reinstate binary kernel loading with fixes.
Build binaries with unique filenames from the kernel generated and save them.
Try to load this cached binary if it matches on next kernel instantiation.
This speeds up start-up dramatically, and has a unique kernel binary for different kernel configurations.
|
|
|
a095f0fa
|
2011-06-30T14:30:10
|
|
Broke source generated program. Fix.
|
|
|
4d730577
|
2011-06-30T10:36:19
|
|
Build binaries with unique filenames from the kernel generated and save them.
Try to load this cached binary if it matches on next kernel instantiation.
This speeds up start-up dramatically, and has a unique kernel binary for different kernel configurations.
|
|
|
973b2199
|
2011-06-30T08:58:07
|
|
Tidy.
|
|
|
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.
|
|
|
60f0bb19
|
2011-06-30T15:47:17
|
|
Temporarily back out binary building till it's working more reliably.
|
|
|
401586f9
|
2011-06-25T18:06:38
|
|
Only try to patch GPU referenced.
|
|
|
656b485d
|
2011-06-25T18:58:59
|
|
Make the worksize and vector width configurable.
|
|
|
fdb46f2d
|
2011-06-26T00:06:19
|
|
32bit fixes.
|
|
|
6374e0fa
|
2011-06-28T21:11:04
|
|
Import the phatk kernel. Enable it only for hardware with amd media ops for now since it crashes nvidia et. al.
Fallback to the poclbm kernel for the rest. Try harder to avoid stale blocks around longpoll detecting new blocks.
|
|
|
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.
|
|
|
1dfbe603
|
2011-06-29T09:24:38
|
|
Put sanity limit on work size since some nvidia fail :(
|
|
|
623b9b9f
|
2011-06-27T12:45:03
|
|
Patch bitalign separately from bfi_int.
Recover from failing to patch for bfi int.
|
|
|
c5666051
|
2011-06-27T16:41:39
|
|
Tab dainbramage.
|
|
|
11c88185
|
2011-06-27T16:08:46
|
|
32 bit only builds one elf, not an elf in an elf, so account for it to be able to bfi int patch properly.
|
|
|
06f39506
|
2011-06-26T08:49:50
|
|
Fix typo which prevented BFI INT patch working on multi-GPUs.
|
|
|
30e38e2e
|
2011-06-26T00:59:07
|
|
Typo i - gpu
|
|
|
ac4ab6af
|
2011-06-25T12:32:30
|
|
Fix mutli-gpu initialisation when BFI_INT patching.
|
|
|
64f268c8
|
2011-06-24T17:00:54
|
|
Display correct worksize.
|
|
|
808d5b07
|
2011-06-24T11:17:58
|
|
Make output debug.
|
|
|
feb8cfc8
|
2011-06-24T09:51:54
|
|
applog fixes.
|
|
|
d28f526b
|
2011-06-23T17:56:11
|
|
Need memmem in advance() to work properly for patching.
|
|
|
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.
|
|
|
33352213
|
2011-06-23T16:39:56
|
|
Report preferred vector width.
|
|
|
f05270b8
|
2011-06-22T01:19:19
|
|
Optimise loop and make debug debug only.
|
|
|
91e5cef3
|
2011-06-22T00:13:46
|
|
Actually get first BFI_INT patch working.
|
|
|
c548dea8
|
2011-06-20T16:53:13
|
|
First BFI_INT patch changes.
|
|
|
6b77d850
|
2011-06-17T14:00:41
|
|
Fixes.
|
|
|
a40003ab
|
2011-06-21T09:46:25
|
|
More bfi_int patching.
|
|
|
79fec01a
|
2011-06-22T12:27:57
|
|
Remove the input buffer and just pass args to the kernel as per plugin design.
|
|
|
afc83cb2
|
2011-06-23T21:10:19
|
|
Need gnu_source for memmem to work properly.
|
|
|
b4d2733c
|
2011-06-22T16:47:34
|
|
Convert to poclbm kernel.
|
|
|
407df0f4
|
2011-06-23T12:35:22
|
|
Use correct boolean testing and long enough string for cl_amd_media ops testing and kernel changing.
|
|
|
c2e41655
|
2011-06-23T12:04:53
|
|
Detect platforms that support cl_amd_media_ops and decide whether to BFI_INT patch or not.
Remove unused kernels.
|
|
|
4cd5f47e
|
2011-06-23T21:09:22
|
|
Revert "Multiple compiler warning fixes."
This reverts commit a5cbfbde2610e9f60e14b41a4e0595bcb34c772a.
Broke.
|
|
|
88761e6c
|
2011-06-23T21:04:29
|
|
Multiple compiler warning fixes.
|
|
|
910e6943
|
2011-06-19T22:21:51
|
|
Increase baseline threads to 1<<22. Make total counter regularly update every 5 seconds.
Only write the blank buffer when it needs to be blanked.
|
|
|
26546ad5
|
2011-06-24T09:17:09
|
|
Make the optimisations per-gpu card and update code to work properly with multiple cards.
|
|
|
dde70397
|
2011-06-14T10:32:54
|
|
Merge gpumining from oclmine. Unstable.
|