New version: 3.9.0, 23rd December 2013New stable version that coincidentally includes a new driver, hence the updated minor version.
Human readable changelog:- Driver for drillbit ASICs.
- Fixes for various KnC hardware errors, with improvements to hashrate. Note this is not a comprehensive fix for the hardware errors specific to rEligius - you will find a substantial drop in hardware errors if you start cgminer with the quiet and text only options (-q -T). An updated binary is here:
http://ck.kolivas.org/apps/cgminer/kncminer/- Updated bi*fury driver with support for the latest firmware. This includes dynamic clocking based on temperature which tries to maintain a constant temperature set intiially to 82 degrees but adjustable with --bxf-temp-target .
- Much more API output for bxf devices.
- Less spewing of errors when bxf devices are removed/die
- Updates to hashfast driver code
- Fixes for working with proxies that use small nonce2 sizes
Full changelog:- drillbit asic - enable in api.c
- Fix trivial warnings in knc driver.
- Reinstate work utility based hashmeter for knc.
- drillbit format %z not valid on windows
- drillbit more formatting changes
- usbutils remove old code added back
- Memset the spi tx buffer under lock in knc driver.
- drillbit fix temp display to fit in standard space
- Drillbit formatting
- drillbit - use one drvlog and display dname before add_cgpu
- Keep orginal naming for the bitfury driver
- knc: Bugfix - good shares wrongly reported as HW errors. Root cause of the
problem: several work items were assigned the same work_id in the active works
queue of the knc driver. Thus when good nonce report arrived from the FPGA,
wrong work item was picked up from the queue, and submit_nonce evaluated that
as an error. Fix: Limit the work_id counter update rate. Update it only to the
number of works actually consumed by the FPGA, not to the number of works
send.
- Store per-chip submit information for bxf device and show them in the API.
- Check for removed bxf devices before trying to update work or send messages.
- api.c no decref if not json
- Minimise risk of nonce2 overflow with small nonce2 lengths by always encoding
the work little endian, and increasing the maximum size of nonce2 to 8 bytes.
- Change default hashfast timeout to 500ms.
- Ensure we can look up the work item in the hashfast driver or print out an
error if we don't.
- Drillbit source formatting - reindent and retabify
- Add ASIC count, temperature status to drillbit API output (closes #1)
- Many warning fixes
- knc: Do not include variable "last minute" data into the "last hour" per-core
stats
- knc: Make per-core statistics available through API
- Implement command line control of the bxf target temperature.
- Add a simple PID-like controller to bi*fury devices to dynamically alter the
clock setting to maintain a nominal target temperature set to 82 degrees.
- Add data to BXF API output.
- Add support for newer protocol bi*fury commands job, clock and hwerror,
setting clock to default 54 value, turning parsing into a compact macro.
- Look for the thermal overload flag in the gwq status message in the hashfast
driver and send it a shutdown followed by an attempted reset.
- Log message fixups
- Fix for "Timing out unresponsive ASIC" for pools which send early reconnect
requests, and then take a short time to send work (ie BTCGuild)
- Shorten initial config line, win32/pdcurses doesn't like long lines during
early logging
- Pull back the very long timeouts set in fe478953cf50
- Fix bug where work restart during results scan could lead to bad device state
- Align device status lines same regardless of number of temp status or >10
ASICs
- Tag log lines from brand new devices as DRB-1 until they are initialised
- Tag log lines as 'DRB0' rather than 'DRB 0', same as other places in cgminer
- Print a summary of the device settings at level NOTICE during initialisation
- Allow choosing device settings based on 'short' product names shown in status
line
- Allow per-device settings to use "DRBnn" as an identifier instead
- Issue an ASIC restart during a work_restart, removes spurious timeout messages
from ASICs and probably some rejected shares
- Check all results against all work instead of just taking the first match
(avoids some rejected submissions to the pool, ASIC can produce multiple
candidate results.)
- Fix memory leak caused by unnecesarily copied work
- Fix bug with find_settings not returning default value
- Set timeouts on write, set very long timeouts
- Merge drillbit driver
