|
Updated git tree:
Make the thread restart do a pthread_join after disabling the device, only re-enabling it if we succeed in restarting the thread.
Do this from a separate thread so as to not block any other code.
This will allow cgminer to continue even if one GPU hangs.
Try to do every curses manipulation under the curses lock. Hopefully this will prevent the crash on window resizing.
Only use the sockoptfunction if the version of curl is recent enough. This should make it possible to build on machines with older libcurls installed.
|
|
|
|
|
Can't do "make install" at the moment with cgminer, just run it from its own directory. Presumably if you've installed a newer curl, you need to do ldconfig to let the system know it's there as well.
|
|
|
|
|
Need to do ./autogen.sh again
|
|
|
|
RPM for EL6, still broken on EL5 because of curl :-( ck, -O2 -g is not helping either, it still doesn't segfault under gdb, but reliably segfaults when run from the terminal. Look, I managed to make it to dump core with -O0, then loaded in gdb and did bt full. Does it help? Core was generated by `/opt/cgminer/bin/cgminer --cpu-threads 4 --gpu-threads 0 --url xxx'.
Program terminated with signal 11, Segmentation fault.
(gdb) bt full
#0 memset () at ../sysdeps/x86_64/memset.S:1016
No locals.
#1 0x00007ff9732ad842 in wredrawln () from /lib/libncurses.so.5
No symbol table info available.
#2 0x000000000040ab0a in watchdog_thread (userdata=0x0) at main.c:2991
y = 17
logx = 131
i = 4
x = 131
logy = 17
now = {tv_sec = 1311508655, tv_usec = 958182}
interval = 2
rotate_tv = {tv_sec = 1311508643, tv_usec = 956086}
zero_tv = {tv_sec = 0, tv_usec = 0}
#3 0x00007ff9736e09ca in start_thread (arg=<value optimized out>) at pthread_create.c:300
__res = <value optimized out>
pd = 0x7ff968a9a700
unwind_buf = {cancel_jmp_buf = {{jmp_buf = {140709179533056, -1219236616075443200, 0, 0, 4, 0, 1217862590511481856,
1217801615496348672}, mask_was_saved = 0}}, priv = {pad = {0x0, 0x0, 0x0, 0x0}, data = {prev = 0x0,
cleanup = 0x0, canceltype = 0}}}
not_first_call = <value optimized out>
robust = <value optimized out>
freesize = <value optimized out>
__PRETTY_FUNCTION__ = "start_thread"
#4 0x00007ff972ff570d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112
No locals.
#5 0x0000000000000000 in ?? ()
No symbol table info available. I'm looking, and it ain't making much sense. It's crashing on the redraw window function within libncurses for reasons that aren't obvious... I'll try doing that window redraw somewhere else in the code. |
|
|
|
|
In your distribution's repository, not on your computer...
|
|
|
|
|
The file you're missing is:
/usr/share/aclocal/libcurl.m4
on my distro that's in:
libcurl4-openssl-dev
So, unless you need to refresh your configure after installing a new package, you need to find how to search for the libcurl.m4 file in your distribution's repository.
|
|
|
|
1st up, you're missing libcurl dev
2nd, your GCC is so old it doesn't support "-march=native" so remove that
your proposal is I remove gcc? LOL no, remove the "-march=native" part of the CFLAGS. |
|
|
|
|
1st up, you're missing libcurl dev
2nd, your GCC is so old it doesn't support "-march=native" so remove that
|
|
|
|
|
Yes, preventative... nod nod nod...
Nod furiously.
|
|
|
|
|
I wasn't trying to start a distro flamewar... I was just pointing out "Much of the world." is an exaggeration to say the least...
|
|
|
|
but I don't really understand if anyone is interested in RHEL5 builds other than me.
Much of the world. If this doesn't work on Centos, it's worthless. Kinda like coming up with windows only mining software  To the author, this is what was required to build poclbm under centos: http://forum.bitcoin.org/?topic=7534.0RedHat and Centos uses older (proven stable) versions of their software. Unfortunatly I, like zaytsev, am stuck using poclbm atm. * Viceroy sadly drops his head and walks away I think you grossly overestimate how many people are using centos  Anyway I've committed a fix to the git tree which prevents using just the sockopt function if the version of curl is too old. This should fix it building on older curls. |
|
|
|
|
Crossfire has been known to slow down cards on windows.
|
|
|
|
|
That's a function from a newer curl. RHEL5 (and others) has a curl that's too old so you need to find a way to update curl - either through non-standard packages or some kind of upgrade or manual install.
|
|
|
|
New release: Version 1.4.1 Source: http://ck.kolivas.org/apps/cgminer/cgminer-1.4.1.tar.bz2Linux x86_64 dynamic binary: http://ck.kolivas.org/apps/cgminer/cgminer-1.4.1-x86_64-built.tar.bz2Win 32 binary: http://ck.kolivas.org/apps/cgminer/cgminer-1.4.1-win32.zipI thought it was time for a mostly bugfix release. The only new feature in this version is the new trickle function which trickles work automatically to a backup pool if the primary pool is still responding but slow to ensure the GPU is always doing work even if it hasn't quite reached the stage where cgminer decides to switch pools. This happens by default with any pool management strategy choice. The rest of the changes are bugfixes to things reported mostly on this forum (thanks for your reports). The rest of the changelog follows: Do away with GET for dealing with longpoll forever. POST is the one that works everywhere, not the other way around. - Detect when the primary pool is lagging and start queueing requests on backup pools if possible before needing to roll work. - Load balancing puts more into the current pool if there are disabled pools. Fix. - Disable a GPU device should the thread fail to init. - Out of order command queue may fail on osx. Try without if it fails. - Fix possible dereference on blank inputs during input_pool. - Defines missing would segfault on --help when no sse mining is built in. - Revert "Free up resources/stale compilers." - didn't help. - Only try to print the status of active devices or it would crash. - Some hardware might benefit from the less OPS so there's no harm in leaving kernel changes that do that apart from readability of the code. |
|
|
|
|
re: POST vs GET on longpoll
I am absolutely getting sick of this freaking problem of how to do a longpoll. I originally converted to GET because someone suggested that's what the deepbit unofficial guide suggested was the way to do a longpoll. Then I discovered lots of other longpolls didn't work because MANY operating systems could easily lose their longpoll connection through routers and so on because they didn't issue any sort of keepalive for up to 2 hours by default (7200 seconds). So with help I converted it to use keepalives much more frequently. In the meantime, deepbit longpolls were BROKEN by the use of GET. Now someone posts me a fix for deepbit longpolls, using POST again. As far as I can see, there is NO ADVANTAGE to using GET and unless someone can positively prove me otherwise, I will be converting every longpoll to POST once more and putting this issue to bed forever.
|
|
|
|
Would it be possible to add a multipool strategy that's similar to load balance, but with the first one having priory? So the second one is just active when the first one for some reason doesn't have enough work, for example because of latency.
That's basically failover that you're describing... That's the default policy. Or do you mean micro delays instead of macro ones? Yes, to even out (very) short performance drops. Updated git tree: Now the default policy will detect when the primary pool is lagging and use a backup pool if there's one available. |
|
|
|
how does one specify phatk110714.cl or poclbm110717.cl ?
One does not. The poclbm kernel is far far behind the phatk kernel in speed and any hardware that can run the phatk kernel is given it. The poclbm kernel is reserved for nvidia cards and ATI 4x cards only. The phatk kernel, even without bitalign patching, crashes on these lesser mining cards. Phoenix with phatk worked on my 4850, and that was faster than poclbm. You can now specify which kernel you want with the -k option. |
|
|
|
when solo-mining, i get what happened to the 78 requested work items? i know up to 2*NB (due to 2 threads) might be thrown away, but at that time NB was 24. still missing 30 "shares". another thing: is it possible to set a flag for solo mining so the miner will report rejected shares as accepted? i really like the overview, but with rejected shares everything says "0" :< Not every work item has a solution. Thus you will not get always get the same number of submissions as queries. Conversely some work items have more than one solution. This leads to "efficiency" of greater than 100% (when not solo mining). As for the flag idea... |
|
|
|
ck, I am still trying to track down the annoying window resize segfault. I compiled the git version like this: CFLAGS="-O0 -g -Wall -march=native" ./configure --prefix=/opt/cgminer and started like this gdb /opt/cgminer/bin/cgminer then I did "set args" and "run" and it does NOT segfault. If I run it like this from the command line it segfaults upon resize. Is there any other way to collect the information that would help you to fix the issue? Leave the -O2 -g on and try again (in gdb). |
|
|
|
[2011-07-24 02:22:31] Failed to init GPU thread 0
Aha. It never even started and you hit another bug trying to reference a thread that wasn't there. Luckily I have an idea as to why it didn't start. Perhaps osx doesn't like out of order execution enabled. Now updated in git, hopefully fixing both.
Latest git works (doesn't crash) but gets about 20% less MH/s than 1.3.1 (~200 MH/s vs ~250MH/s) on my HD 5870/OS X Lion. Great, glad it works, and thanks for testing! Give it time. The statistics now start out at time = 0, and it can take a while before you get the real value (10+ mins). Perhaps I should go back to resetting the statistics like I used to, but that made for some nice funky high values initially. What do people think? |
|
|
|
|
Show Posts
