OLD: BFGMiner 3.10.0: modular ASIC+FPGA, GBT+Strtm, RPC, Mac/Lnx/W64, AntU1, DRB - page 155.

somedood

newbie

Activity: 14

Merit: 0

Quote from: Luke-Jr on June 14, 2013, 11:18:51 PM

Windows 64-bit 3.1.0 is broken for GPU mining.
Either turn it off with the -G option or replace the EXE with this fixed one until 3.1.1.

Thanks now it works again. One problem though:
Intensity settings > 0 don't seem to have a lot of effect (compared to 3.0.2)
Intensity settings < 0 reduce GPU usage properly but consume huge amounts of CPU (2 cores 100%)
Dynamic works great (low CPU usage, good hashing performance)

Tested with Catalys 13.4 on Win 7 x64 and

Code:

[2013-06-15 17:51:41] CL Platform 0 vendor: Advanced Micro Devices, Inc.
[2013-06-15 17:51:41] CL Platform 0 name: ATI Stream
[2013-06-15 17:51:41] CL Platform 0 version: OpenCL 1.0 ATI-Stream-v2.1 (145)
[2013-06-15 17:51:41] Platform 0 devices: 1
[2013-06-15 17:51:41] 0 Cypress
[2013-06-15 17:51:41] CL Platform 1 vendor: Advanced Micro Devices, Inc.
[2013-06-15 17:51:41] CL Platform 1 name: AMD Accelerated Parallel Processing
[2013-06-15 17:51:41] CL Platform 1 version: OpenCL 1.2 AMD-APP (1124.2)
[2013-06-15 17:51:41] Platform 1 devices: 1
[2013-06-15 17:51:41] 0 Cypress
[2013-06-15 17:51:41] GPU 0 ATI Radeon HD 5800 Series hardware monitoring enabled
[2013-06-15 17:51:41] 1 GPU devices max detected

Maybe it is related to the debug build?

EDIT: when looking at bfg's threads there are the two gpu threads that seem to use all the cycles:
atiocl64.dll!clGetSamplerInfo+0x36a80
They also exist for dynamic intensity but only with 0.5% CPU instead of double digits.

twobitcoins

full member

Activity: 144

Merit: 100

Theoretical bug in libblkmaker:

parse_txn populates txn->hash but is missing a call to "my_flip(txn->hash, 32);".

Luke-Jr

legendary

Activity: 2576

Merit: 1186

Windows 64-bit 3.1.0 is broken for GPU mining.
Either turn it off with the -G option or replace the EXE with this fixed one until 3.1.1.

Thanks to unit3 for the initial bug report (though neither of us imagined it'd crash Win64!) and TelVoSRS for finishing the bisect first

Edit: Details: I changed the OpenCL driver to use the common open_bitstream function to find its kernels, to allow installing them to the standard location for such data files on GNU/Linux. However, I forgot to include the fpgautils.h headers which contained the definition of how to call the function. When the compiler doesn't know, it makes an educated guess based on the syntax used. For Win64, that guess happened to be wrong, and resulted in the build crashing when it attempted to open an OpenCL kernel. When unit3 went to update the Ubuntu PPA, it refused to build because it knew the educated-guess function was a bad idea and broke IA-64. Nobody really uses IA-64, so I simply fixed it and he finished the build. It wasn't until later that the bisection results identified the change to blame for the crash that I realized what happened there - in fact, I'd say I probably wouldn't have figured it out without hours (if not days) of debugging had Launchpad and unit3 not caught the problem earlier.

ChipGeek

full member

Activity: 198

Merit: 100

Windows 64 bit version of bfgminer 3.1.0 never starts for me at all - it crashes back to command line immediately.
BFL's Easyminer (running bfgminer) seems to work fine.

gveltre

newbie

Activity: 17

Merit: 0

Quote from: Epoch on June 14, 2013, 09:22:25 AM

I've had numerous crashes with 3.1.0 over the past 24 hours across 3 separate clusters. I need to babysit this one too much, so it has turned out to be impractical for me. This is the 32-bit Windows version running on both 32-bit and 64-bit Win7 instances.

I've since reverted to the previous experimental version bfgminer-w32-320e7ed (reports itself as 3.0.99) which is much more stable.

Yes, my 32bit version running on win7 has crashed about 3 times in the last 24 hours. I couldn't get the 64bit version to run.

Epoch

legendary

Activity: 922

Merit: 1003

I've had numerous crashes with 3.1.0 over the past 24 hours across 3 separate clusters. I need to babysit this one too much, so it has turned out to be impractical for me. This is the 32-bit Windows version running on both 32-bit and 64-bit Win7 instances.

I've since reverted to the previous experimental version bfgminer-w32-320e7ed (reports itself as 3.0.99) which is much more stable.

SirMintALot

full member

Activity: 140

Merit: 100

In POS we trust

Something is wrong with the w64 build. It crashes after starting on my AMD A10-5800K and my AMD E2-1800. The w32 build and the old 3.0.99 w64 build are without problems.

Luke-Jr

legendary

Activity: 2576

Merit: 1186

Quote from: twobitcoins on June 13, 2013, 01:28:28 PM

The tag for 3.1.0 is missing on github. Thanks!

Fixed

twobitcoins

full member

Activity: 144

Merit: 100

The tag for 3.1.0 is missing on github. Thanks!

Luke-Jr

legendary

Activity: 2576

Merit: 1186

Quote from: gveltre on June 13, 2013, 01:14:46 PM

version 3.1.0 crashes when running on my Win7 64bit. Had to go back to 3.0.3. It seems to work good.

Can you meet me on IRC some time to try to debug this regression?

Quote from: gveltre on June 13, 2013, 01:14:46 PM

I would like to run BFGMiner 3.1.0 on my linux computer. I am not that familiar with linux. I have tried to compile it on my
linux mint debian edition but get an error when using make. "No targets specified and no makefile found. Stop." Any
help would be greatly appreciated.

Please follow the compile instructions in README.

gveltre

newbie

Activity: 17

Merit: 0

Hi,

Just a few issues. version 3.1.0 crashes when running on my Win7 64bit. Had to go back to 3.0.3. It seems to work good.

I would like to run BFGMiner 3.1.0 on my linux computer. I am not that familiar with linux. I have tried to compile it on my
linux mint debian edition but get an error when using make. "No targets specified and no makefile found. Stop." Any
help would be greatly appreciated.

Thanks!

af_newbie

legendary

Activity: 2702

Merit: 1468

After about 1000 accepts, 3.1.0 shows about the same HW error count as 3.0.2. (Running Erupters on Win 7x64)

Initially, it was higher, it settled down after a while.

af_newbie

legendary

Activity: 2702

Merit: 1468

Running 3.1.0 with a bunch of Erupters.

The good news is that the switch from 3.0.2 to 3.1.0 was painless.

The bad news is 3.1.0 reports more hardware errors in MY setup than 3.0.2 (Win 7, x64).

Still few percentage lower than cgmienr 3.1.1 but higher than bfgminer 3.0.2.

I'll monitor it a bit longer to see if it settles down...

evilscoop

sr. member

Activity: 350

Merit: 250

sweet, lets hope this works with the cairnsmore....unlike cgminer....

edit ---
all working fine for cm1 on rpi Cheesy

*happy dance*

Luke-Jr

legendary

Activity: 2576

Merit: 1186

NEW VERSION 3.1.0, JUNE 13 2013

This release includes numerous merges of contributions by Con & Kano, from cgminer 2.11.0 through 3.2.1. Major merges are noted in the human readable changelog with "cg". Notable exclusions are the libusb rewrites (overall a bad idea), "hotplug" support (not implemented correctly), and "cURL-less" stratum (breaks proxy support, and the underlying bugs they hoped to workaround with it are already fixed in BFGMiner). BFGMiner also continues to use the original, time-tested driver for Butterfly Labs devices, including ASICs, and does not include cgminer's ASIC-specific rewritten "BFLSC" driver in any form.

Additionally, there have been many internal code cleanups. Please note that as uthash is now available for most operating systems, it is no longer bundled with BFGMiner. You will need to install the uthash-dev package to compile BFGMiner 3.1+.

Human readable changelog:

BitForce: Support for new long-board job-per-chip protocol.
Reformatted TUI summaries, including percentage of shares rejected or stale.
Icarus-compatible devices: Autodetect work_division by default (eg, for Block Erupter USB ASICs).
Avalon: Driver is now compiled in by default, including in official binaries
Avalon: Will only configure with specific -S avalon:/dev/ttyUSBn option (note the driver name must be included), since there is no reliable probe for it.
Avalon: Full data for 4-module devices in RPC. cg
OpenCL: Support for reading temperature from free software radeon drivers via libsensors.
Fixes for writing configuration files when --device and/or --remove-disabled are specified.
Faster startup now when pools are slow, connecting to the first pool available! cg
Adding a pool now from the menu will return quickly while it leaves probing it to the background. cg
Scrypt: Fixed the failure to create high thread concurrencies. cg
Scrypt: Intensities above 13 will not lead to garbage being generated on GCN (7xxx) hardware now. cg
Scrypt: Decreased default GPU threads to 1. Intensities above 13 cannot handle more. cg
Scrypt: Changed the default choice of thread concurrency if no options are passed as a hint, based on amount of available GPU RAM. cg
Scrypt: Extensive updates to the SCRYPT-README file in line with changes to the code and new information. cg
Scrypt: Slightly larger buffer which may lead to less hardware errors on very fast GPUs. cg
CPU mining: Numerous improvements to portability. Should also work on Windows now, at least.

Full changelog

Bugfix: openwrt: Expect fixed bitstream path for input
Improve Makefile organization and fix "make install"
icarus: Upgrade work division autodetection probe to be faster and also detect 8 core
Calculate rejection percentage based on weighed shares, not absolute counts
Count weighed discarded (stale) shares per cgpu
Bugfix: Cleanly count discarded (stale) shares without overlapping devices/sshare locks within clear_stratum_shares
configure: Enable Avalon support by default now that it behaves reasonably
avalon: Since detection is not really implemented, only probe devices when the driver is specified to -S by name (eg, "avalon:/dev/ttyUSB0")
Bugfix: bitforce_queue: Never try to reinitialize a slave processor
Bugfix: bitforce_queue: Use work_list_del everywhere we remove from work_list to ensure work gets freed properly when done
Reduce HW error logging to debug level, so it doesn't clutter the TUI log by default
DevAPI: When a device has more than 26 processors, represent them as aa-zz
bitforce: bitforce: Fix TUI display of slave processors on parallelized boards
bitforce: Only display temperature in TUI for the first chip on parallelized boards
Bugfix: bitforce: Set temperature for all processors on the board
Bugfix: bitforce_queue: Initialize proc->device_data to board data for parallelized chip processors
Bugfix: bitforce_queue: Defer nonce count check until after thiswork has been identified (or handled as NULL)
avalon: avalon_clear_readbuf can simply wait for a single read timeout on Windows to avoid select
avalon: Simplify avalon_get_result by using avalon_gets
avalon: Go back to good old serial timeouts for gets, since select() is for sockets (only, on Windows)
Updated api.c to return the hashrate with 3 decimal places
Change hashrate display to never display 3 fractional digits, because it looks a bit ugly
bitforce: Credit hashrate to the correct chip on parallelized devices
Re-set work thr_id on share submissions, to ensure the result gets credited to the correct device in case of shared job queues (as in BitForce long boards)
bitforce: Turn parallelization into separate logical processors for more details on each, including working with XLink
bitforce_queue: Implement job sanity checks using new "ZqX" for devices using parallelization
bitforce_queue: Minimal support for parallelization
Add --device-protocol-dump option to debuglog low-level bitforce protocol details
When shutting down, set work restart flag (and trigger notifier) to help mining threads escape to their main minerloop (and check for shutdown)
Document and check for uthash version 1.9.2+
Bugfix: Don't report failure for graceful mining thread shutdown
Name devices in mining thread failures
Warn about killing mining threads
Bugfix: Wake up mining threads when asking them to shutdown
Disable pthread cancel within curses locking
Shorten the avalon statline to fit in the curses interface and show the lowest speed fan cooling the asic devices.
Change switch_compact function name to switch_logsize to be used for other changes.
Only adjust cursor positions with curses locked.
devs display - fix GPU duplicate bug
basic copyright statement in API.java
Change the --device parameter parsing and configuration to accept ranges and comma separated values.
Modify scrypt kernel message.
Check for pool_disabled in wait_lp_current
Check for pool enabled in cnx_needed.
Add README.ASIC to debian packaging and make-release
Document avalon options in ASIC-README
Create README.ASIC with basic summary of supported ASIC devices.
Do avalon driver detection last as it will try to claim any similar device and they are not reliably detected.
Set the fanspeed to the nominal chosen for GPUs.
Clamp initial GPU fanspeed to within user specified range.
Avalon fan factor is already multiplied into the info values.
Get rid of zeros which corrupt display.
Logic fail on minimum fanspeed reporting.
Provide a workaround for fan0 sensor not being used on avalon and pad fan RPM with zeros.
Add ambient temp and lowest fan RPM information to avalon statline.
Display max temperature and fanspeed data for avalon.
Set devices to disabled after they exit the hashing loops to prevent the watchdog thread from trying to act on them.
Scanhash functions perform driver shutdown so don't repeat it.
Change the opencl shutdown sequence.
Send the shutdown message to threads and do the thread shutdown functions before more forcefully sending pthread_cancel to threads.
Icarus report data direction with comms errors
Execute driver shutdown sequence during kill_work.
Provide an nusleep equivalent function to nmsleep.
Set avalon_info to device data void struct.
Make submit_nonce return a bool for whether it's a valid share or not.
Do a non-blocking read of anything in the avalon buffer after opening the device.
Assign the avalon info data to the device_data in cgpu_info.
Rename cgpu_data to use new device_data
miner.h remove unused device_file and add device_data
Must unlock curses as well in logwin_update.
icarus report usb write error information
Make mining threads report out during work submission.
submit_work_async is no longer used directly by driver code.
Create a logwin_update function which mandatorily updates the logwin and use it when input is expected to prevent display refresh delays.
All stratum calls to recv_line are serialised from the one place so there is no need to use locking around recv().
Only allow the mining thread to be cancelled when it is not within driver code, making for cleaner shutdown and allowing us to pthread_join the miner threads on kill_work().
Implement pthread_testcancel replacement for BIONIC
Attribute whatever stats we can get on untracked stratum shares based on current pool diff.
Downgrade OpenCL headers to 1.0, which work fine for our purposes and are more compatible
icarus: If work_division autodetect fails, just use the old default of 2
avalonhost-raminst script to help with installing to RAM on Avalon-host routers
Attempt to probe /dev/cu.usb* for fallback "-S all"
openwrt: Download uthash dependency
Bugfix: openwrt: Always build with libsensors support disabled
configure: Check for uthash headers
Bugfix: ztex: Only destroy libztex device after the last handle to it has been released
ztex: Remove libztex slave device interface, simply passing fpgaNum to selectFpga
Bugfix: cpu: Fix yasm and sse2 detection
cpu: Check for SSE2 support independently from yasm
Bugfix: cpu: Make sure to link libsse2cpuminer.a before x86_32/libx8632.a
Bugfix: cpu: Only build libsse2cpuminer iff yasm is available and targetting x86_32
Bugfix: Free work only after deleting it from list
Remove embedded uthash.h and utlist.h from Makefile
windows-build.txt: Update for system uthash
Remove embedded uthash (and add dependency on system uthash)
Replace elist.h with utlist.h
Bugfix: Fix build with CPU mining and *without* yasm
cpu: Be explicit about size of sha256_init address
cpu: Add --algo fastauto (new default) to detect a usable algorithm without taking over a minute
cpu: Default to --algo auto
cpu: Support all platform-applicable assembly algorithms, even if used CFLAGS don't support them
Ubuntu: Updated changelog, added scrypt support.
cpu: Set fixed symbol names for stuff shared with assembly
cpu: Create Mach-O asm binaries on Darwin-based systems
Bugfix: cpu: Use COFF yasm binfmt on Cygwin
Bugfix: cpu: Get correct nonce from data, where the CPU sub-drivers leave it
Remove redundant "Reject ratio" in exit-time summary
Apply "R+S(%)" formatting to long-form statistics
Group stale shares in with rejects (but still distinctly counted) and make the percentage be (reject+stale)/total
Include rejected shares as a percentage
Move Utility and Best Share to status line
Remove LW from status line, since it is basically useless
ztex: Clean up a lot of ugly casting
Bugfix: Correctly avoid SIGPIPE on Mac
Make set_work_target a function to set a specified char as target for use elsewhere.
Minor typo.
Support more shares to be returned for scrypt mining.
Set all stratum sockets to nonblocking to avoid trying to use MSG_DONTWAIT on windows.
Only use MSG_NOSIGNAL for !win32 since it doesn't exist on windows.
Use MSG_NOSIGNAL on stratum send()
Set TCP_NODELAY for !linux for raw sockets.
Use TCP_NODELAY with raw sockets if !opt_delaynet
Recheck select succeeds on EWOULDBLOCK for stratum.
Don't use TCP_NODELAY if opt_delaynet is enabled with stratum.
Fix warnings in avalon driver.
correct applog typing
Simplify the many lines passed as API data in the avalon driver now that the API does not need persistent storage for the name.
Duplicate the name string always in api_add_data_full to not need persistent storage for names passed to it.
Add extra matching work count data in API for Avalon with 4 modules.
Clean up summary slightly better on exit.
opencl: Disable using binary kernels on Apple by default
Use sock_blocks in api.c
Fix build and distdir.
compile on win32
Update README.scrypt with improved hashrates for 7970.
Use copy_time helper throughout miner.c
Provide wrappers for commonly used timer routines with API stats.
Use flip32 function instead of open coding it in gen_stratum_work.
Move util.c exports to util.h
Replace gettimeofday usage with cgtime
Adopt gettimeofday wrapper from cgminer (cgtime) that is always called with tz set to NULL and increases the resolution on windows.
Add high resolution to nmsleep wrapper on windows.
Bugfix: Export stats_lock for deviceapi
Set default ocl work size for scrypt to 256.
fliter out the wrong result from adjust fan code
Set last device valid work on adding device.
Make scrypt submission use the submit_nonce code, with nonces matching endianness.
Increment hardware error count from the one site.
compile avalon driver on win32 and win64
build out of source dir
Rename scrypt regenhash function for consistency.
Add Mac FAQ.
Further driver FAQs.
Check for work restart after disable in the hash queued work loop since it may be a long time before we re-enable a device.
Unconditionally test for many wrong results on avalon and reset to avoid passing a corrupt avalon result to temperature code.
Only reset an avalon device with no results when there are no results consecutively.
More FAQs.
Avoid applog in recalloc_sock.
Avoid applog under cg_wlock.
Put spacing around locking code for clarity.
Avoid applog under pool_lock.
Avoid more recursive locks.
Avoid applog while ch_lock is held.
Avoid recursive locks in fill_queue.
Variable is already initialised in global scope.
More GPU FAQs.
More README faqs.
Yet more README faqs.
Add more FAQs to README.
Wrap result wrong tests in avalon scanhash in unlikely() and only consider a hash count of zero wrong if a restart wasn't issued.
avalon: if result_wrong >= get_work_count jump out the read loop
Fix warning on 32bit.
fix the fan control on max temp2/3
for some reason network down. one simple bfgminer command: "bfgminer -o 127.0.0.1:8888 -O fa:ke --avalon-options 115200:32:10:50:256" can idle the avalon for safe power and protect chip
if hash_count == 0; reinit avalon, fix the 0MHS bug use the max value of temp1 and temp2 for fan control
Reinstate the matching_work_count per subdevice on avalon based on the work subid.
Rationalise and simplify the share diff and block solve detection to a common site.
subid field for devices that do not yet support the distinct device/processor interface
Make the avalon array size a macro.
Use replacement of work items in the avalon buffer as needed instead of flushing them.
Reinstate wrong work count to reset avalon regardless and display number of wrong results.
select() on serial usb in avalon does not work properly with zero timeout.
Use no timeout on further reads in avalon_gets
Do sequential reads in avalon_get_reset to cope with partial reads.
Show read discrepancy in avalon_get_reset.
Reuse avalon_get_work_count variable.
Check for AVA_GETS_RESTART when deciding if avalon has messed up.
Make the detection of all wrong results on avalon much more conservative to avoid false positives on work restarts.
Show error codes on select and read fail in avalon.
If we get a restart message in avalon_gets still check if there's a receive message to parse first without a timeout before returning AVA_GETS_RESTART.
avalon_gets is always called from the one call site so inline it.
The read_count is unused by the avalon get result code and no longer required for avalon reset so simplify code removing it.
Use a separate avalon_get_reset function for resetting avalon instead of using avalon_get_result.
The current hash count returned by avalon scanhash is just an obfuscated utility counter so make it explicit.
Check for a restart before a timeout in message parsing code in avalon.
We should check for a restart message before checking for a timeout in avalon scanhash.
Store the subid for the work item in avalon.
Fix record_temp_fan function in avalon driver.
Remove inappropriate memset of struct avalon result which was corrupting fan values.
Only do_avalon_close once on multiple errors.
Reset the result_wrong count on block change in avalon scanhash to prevent false positives for all nonces failed.
Small timeouts on select() instead of instant timeout increase reliability of socket reads and writes.
Rotate the avalon work array and free work on AVA_SEND_BUFFER_EMPTY as well.
Only get extra work in fill_queue if we don't have any unqueued work in the list.
Don't get any work if our queue is already full in avalon_fill.
Free avalon->works in the event we call avalon_prepare on failure to initialise.
Fix warnings.
Create an array of 4 lots of work for avalon and cycle through them.
Remove unused per unit matching work count for avalon.
Rename the confusing avalon_info pointer.
Simplify avalon scanhash code using the new find_queued_work_bymidstate function.
Members of cgpu_info for avalon are not meant to be in the union.
Use correct struct device_drv for avalon_drv.
Check enough work is queued before queueing more in avalon_fill.
Actually put the work in the avalon queue.
Rename avalon_api to avalon_drv.
First draft of port of avalon driver to new cgminer queued infrastructure.
Minor README updates.
README.GPU: Properly warn about overclocking damage
Add example 7970 tuning for scrypt in readme.
Update driver recommendations.
Add extensive GPU FAQs for the flood of new Scrypt miners.
Bugfix: Expect bitstreams and kernels to be in the srcdir, not build dir
cpu: Prefer sse4_64 algorithm if supported
cpu: sse2_32: Force bare symbols for cross-asm/C symbols
Compile CPU mining for win32 and win64
configure: Check for pthread in -lwinpthread
Use has_pth flag instead of trying to mess with pthread internals
configure: Explicitly check for nanosleep
configure: Include "no" response in BFG_PTHREAD_FLAG_CHECK
miner.h missing extern
Update links and recommended SDKs.
Bugfix: README.GPU: Fix some typos
Update README to match changes to display.
Remove increasingly irrelevant discarded work from status lines.
Remove increasingly irrelevant GW value from status.
README.GPU: Correct terminology
Update README about intensity.
Bugfix: Restore always autodetecting stratum as non-scrypt
icarus: Replace default of 2 work_division/fpga_count with autodetection of 1, 2, or 4
Update scrypt readme with newer information and to match changes in code.
Set default GPU threads to 1 for scrypt.
Connect backup stratum pools if the primary pool cannot deliver work.
Use a new algorithm for choosing a thread concurrency when none or no shader value is specified for scrypt.
Do not round up the bufsize to the maximum allocable with scrypt.
Remove the rounding-up of the scrypt padbuffer which was not effectual and counter-productive on devices with lots of ram, limiting thread concurrencies and intensities.
Make pool adding while running asynchronous, using the pool test thread functionality.
Only curl easy cleanup a stratum curl if it exists.
Add intermediate variants of cglocks that can be up or downgraded to read or write locks and use them for stratum work generation.
Move the stratum data to be protected under a new cg_lock data_lock.
Convert the ch_lock to cg_lock.
Convert the control_lock to a cg_lock.
Remove unused qd_lock.
Implement cg_lock write biased rwlocks.
Don't start testing any pools with the watchpool thread if any of the test threads are still active.
Set sockd to false should curl setup fail on stratum.
Reopen the socket whenever we're retrying stratum.
Set pool died on failed testing to allow idle flag and time to be set.
Remove unused pthread_t typedefs from struct pool.
Perform pool_resus on all pools that are found alive with the test pool threads.
Use pool_unworkable in select_balanced as well.
Differentiate pool_unusable from pool_unworkable.
Keep a connection open on higher priority stratum pools to fail back to them.
Set the wrong bool in pool_active
Only bypass unusable pools if they're not the selected ones.
Find the first usable pool in preference to the current pool in select_pool for work.
Add a pool_unusable function which checks if a pool is stratum but not active to use within switch_pools.
API no longer ignore send() status
API make the main socket non-static
Start the stratum thread only if we successfully init and authorise it, otherwise unset the init flag.
Make the initialisation of the stratum thread more robust allowing the watchpool thread safe access to it after the stratum thread is started.
Shorten the time before keepalive probes are sent out and how frequently they're sent with stratum curls.
Display select return value on select fail in stratum thread.
Clear the socket of anything in the receive buffer if we're going to retry connecting.
Perform pool resus on pools that were not set as the initial pool at startup.
Allow pools to be resuscitated on first startup by the watchpool thread.
Check all pools simultaneously at startup switching to the first alive one to speed up startup.
Close any sockets opened if we fail to initiate stratum but have opened the socket.
API use control_lock when switching pools
Clear last pool work on switching pools if the current pool supports local work generation or we are in failover only mode.
make rw locks: mining_thr_lock and devices_lock
work queues - remove new but unnecessary functions
functions for handling work queues
find_work() to find work in devices work queue
Add a get_queued function for devices to use to retrieve work items from the queued hashtable.
Add the choice of hash loop to the device driver, defaulting to hash_sole_work if none is specified.
Add a driver specific flush_work for queued devices that may have work items already queued to abort working on them on the device and discard them.
Flush queued work on a restart from the hash database and discard the work structs.
Create a central point for removal of work items completed by queued device drivers.
Create a fill_queue function that creates hashtables of as many work items as is required by the device driver till it flags the queue full.
Create the hash queued work variant for use with devices that are fast enough to require a queue.
Make sure to do full avalon_init if the device_fd is invalid.
Document extra zero byte in avalon_reset.
Microoptimise likely paths in avalon_gets.
Make sure to set timeout to 100ms instead of 1ms in avalon read loop for select.
Make sure to get time of first response in avalon read loop.
Use select for a reliable timeout in avalon read and don't read 1 byte at a time, optimising read loop.
We should break on the loop on a work restart in avalon, and only consider all errors if we actually have gotten some results.
Avalon init on comms error as well.
Reinit avalon device in case of FPGA controller mess up.
Increase reliability of avalon startup by only opening and resetting the device once, looking for the id sequence offset by one byte as well, and still ignoring if it's wrong, assuming it is an avalon.
Nest avalon_decode functions to avoid doing unnecessary lookups once we have found the nonce.
Use htole32 wrapper for nonce encoding in avalon.
Remove unused rev8 function from avalon driver.
Remove const qualifier from driver structs
rename device_api -> device_drv and all related api -> drv
rename get_proc_by_id() to get_devices()
Wrap access to devices array under a mutex
Provide wrappers for grabbing of thr value under the mining_thr_lock.
mutex all access to mining_thr
Split thr_info array into control_thr and mining_thr pointers so more mining threads can be added later
Update the hashmeter at most 5 times per second.
Speed up watchdog interval and therefore display updates to 2 seconds.
Add README.GPU to EXTRA_DIST.
Split out the GPU specific information from the README into a README.GPU file.
Update docs and reorder README to show executive summary near top.
Add more FAQs about crossfire.
Convert error getting device IDs in ocl code to info log level only since multiple platforms may be installed and the error is harmless there.
Unnecessary extra array in ocl code.
Cope with the highest opencl platform not having usable devices.
Update kernel file names signifying changes.
Use constants from the array of __constants throughout the diablo kernel.
Create a __constant array for use within diablo kernel.
Use global constant arrays for all other constants used in scrypt kernel.
Use global __constants for sha functions in scrypt kernel.
Use constants for endian swap macros.
Revise scrypt kernel copyright notice.
Separate out additions in scrypt kernel.
Reuse some Vals[] variables that can be assigned to constants earlier in the poclbm kernel, making for fewer ops.
Put all constants used in poclbm kernel into __const memory array to speed up concurrent reads on the wavefront.
opencl: Support for reading temperature from free software radeon drivers via libsensors
make-release: Include all submodules, recursively
Remove API.java example (no copyright license)
Minimally fix "make install" to ignore bitstream sources
Add submodule for ZtexBTCMiner (aka ztex_ufm1_*)
Add submodule for X6000_ztex_comm4 (aka x6500-overclocker-0402)
ztex: Use standard file header/comment formatting, and update license to GPLv3
bitforce: Allow a longer timeout (1250ms) for ZCX (Device Information) commands
Bugfix: pdcurses doesn't like changing logwin size without clearing it also, so do that in display options
-S all: Start QueryDosDevices probe with 256-byte buffer
Use common code to ensure the prefix of -S *:all remains in all implementations of it (/dev glob was removing prefixes)
bitforce_queue: Ensure comma following nonce count is there, to error cleanly
bitforce: Report queue send failures, and count as hw errors
Bugfix: bitforce_queue: Don't try to send ready-to-queue work to device, when there is no ready-to-queue work
Bugfix: bitforce: Clear want_to_send_queue flag when flushing queue, since we can't send an empty queue
bitforce: Include new total queued count in flush debugging
Bugfix: bitforce_queue: Implement a minimum wait time of 10ms
README: Document serial device format for Mac OS X
Bugfix: cairnsmore1: Enable building with libudev autodetection even if only icarus drivers are enabled
Bugfix: sysfs autodetect: Continue searching even after finding one tty (fixes multiple ttys per device, such as some Cairnsmore1s)
Bugfix: ztex: Avoid destroying libztex device in case other processors are still in use (fixes crash when 1.15y becomes unavailable)
Update windows-build.txt
ccan: Add missing copyright comment headers
Remove obsolete mknsis.sh
Add missing copyright sections to files that may need them
Standard copyright format (including year) for adl_functions.h
Bugfix: When disabling device, ensure its prev_work (if any) gets freed properly
Check stratum socket exists and is writable before even considering whether the sessionid is the same or not
Bugfix: Check that the stratum_share struct for a failed submission is still in the submission hashtable before trying to delete it
README: Add missing documentation for CPU algorithms cryptopp_asm32, sse2_32, and altivec_4way
Bugfix: icarus: Check work restart before timeout
Bugfix: icarus: Debuglog the correct read timeout (and omit from work restart since there's no trivial way to get it)
README: Update links
Bugfix: cpu: Fix warning on Win64
Bugfix: avalon: Strict formatting
Bugfix: Cleanup trivial warnings
Bugfix: bitforce: Seek to end of nonce counter to find nonces, in case there are more than 9
Bugfix: Build hexdump.c into the project normally like everything else
Bugfix: Really fix device entries in saved config file
Update the write config to properly record device entries and remove disabled option.
avalon: Really fix applog formatting
va_copy is meant to be matched by a va_end in log_generic.
Further fix distdir for hexdump.c
Fulltest is true if value is <= target.
Fix warning with no curses built in.
Bugfix: configure: Check NEED_FPGAUTILS correctly
configure: Better grammar for --enable-cpumining help
Bugfix: Check for SSE 4.1 support before building sse4_64 asm CPU miner (uses MOVNTDQA instruction)
Bugfix: elist: Use uintptr_t for member offset
Bugfix: opencl/adl: Fix format string
Bugfix: opencl: Correct usage of formatted prints
Increase fd limits as much as possible at startup
Bugfix: bitforce: bulk queue: Cleanly retry for high temperature recovery
Fixed deps for raring, which has newer libudev1.
bitforce: debuglog actual result data
Bugfix: Missing 'else' can result in null pointer dereference in race
Minor grammo in avalon driver.
Make avalon temperature reading LOG_INFO level.
Fix the problem of seting up termio of ttyUSB0 for icarus. the CSIZE is the mask of CS2/4/8
bufsize is an unsigned integer, make it so for debug.
Bugfix: bitforce: Include get_api_stats in BQUEUE mode
Bugfix: Always compile support for commandline --temp-target and --temp-cutoff, and write them in the config for all devices
Bugfix: Ensure cURL timers get set correctly in submission thread
Bugfix: modminer: Remove unused parameter to sprintf
Bugfix: modminer: Use correct format for bytes left in bitstream upload message
Bugfix: Access strategy name string directly instead of accidentally
Add printf-format syntax checks to more functions that should use it
AUTHORS: Add more contributors
Support configure flag --with-system-libblkmaker to allow building without the bundled copy
Bugfix: Use HTTP/1.1 compatible product token for User-Agent header

shawtux

newbie

Activity: 57

Merit: 0

Quote from: evilscoop on June 12, 2013, 09:05:34 AM

are you building for pi ?

I have latest bfgminer compiled on pi if you need it

thanks! i had a missing lib so it didnt work.
the only issue is that Av. Hash rate is double to the real hash. all other graphs appear normal.

Epoch

legendary

Activity: 922

Merit: 1003

Quote from: Epoch on June 04, 2013, 01:21:17 PM

Quote from: Luke-Jr on May 31, 2013, 11:19:41 PM

Quote from: Luke-Jr on May 31, 2013, 11:29:56 AM

Good news: I think I have stumbled upon the stratum-went-down-related crash people have been reporting while testing BFGMiner in Valgrind.
This should give me enough information to get it fixed in the next release.

Ok, I believe this problem should be fixed in the latest git code.

For Windows testers... Win32 and Win64 binaries are built (drop them in a directory with 3.0.2 DLLs)

As one of the people experiencing these 'crash' issues, I just wanted to mention that I've been running this experimental version for about 3 days straight now. 3 separate clusters. No issues so far. Still too early to conclude anything with certainty, but it *seems* to have solved the issues I've been having with anything post 2.10.x.

Just an update to my last post, I've been running 320e7ed on 3 separate Win7 PCs and have experienced a few crashes on each since that time. Just now there was a hiccup on bitminter and 2 of my bfgminer instances crashes simultaneously (one with 14 BFL Singles; another with 16 BFL Singles). A 3rd instance (with 2 BFL Singles) survived the event. I've seen this behavior before; bfgminer seems more stable with fewer devices.

So although I feel this latest experimental version is an improvement over its predecessors in terms of stability (crashes are not as frequent), it still has some issues.

kingcoin

sr. member

Activity: 262

Merit: 250

Quote from: evilscoop on June 12, 2013, 09:05:34 AM

are you building for pi ?

No, on Altera SoC ARM.

evilscoop

sr. member

Activity: 350

Merit: 250

are you building for pi ?

I have latest bfgminer compiled on pi if you need it

kingcoin

sr. member

Activity: 262

Merit: 250

Quote from: Luke-Jr on June 11, 2013, 11:07:30 PM

Is there a location where it will look for cross compiled versions of its libraries? I have this on by build host:[/quote]Try

Code:

arm-linux-gnueabihf-emerge uthash

...assuming you're using crossdev.
[/quote]

Unfortunately I don't have crossdev set up for this target. I have for others, but for arm-linux-gnueabihf I only have the gcc family of tools.

[EDIT] Sorry about the messed up quoting

Topic: OLD: BFGMiner 3.10.0: modular ASIC+FPGA, GBT+Strtm, RPC, Mac/Lnx/W64, AntU1, DRB - page 155. (Read 1193223 times)