[argyllcms] Re: Hello I have a problem with "colprof -qh"

Martin Weberg wrote:

Hi,

this is a follow-up/confirmation to:
http://www.freelists.org/post/argyllcms/Hello-I-have-a-problem-with-colprof-qh

Mac OS X 10.4 PPC, 4.5GB RAM
ArgyllCMS 1.0.3 binary

colprof -v -qh
[...]
Rev cache RAM = 2787 Mbytes
There is 1 rev cache instance with 2787 Mbytes limit
43%colprof(398) malloc: *** vm_allocate(size=8421376) failed (error code=3)
colprof(398) malloc: *** error: can't allocate region
colprof(398) malloc: *** set a breakpoint in szone_error to debug
colprof: Error - rspl malloc failed - reverse cell simplexes - base simplex

I then tried:
export ARGYLL_REV_CACHE_MULT=0.7
giving:
Rev cache RAM = 1950 Mbytes
There is 1 rev cache instance with 1950 Mbytes limit
43%colprof(904) malloc: *** vm_allocate(size=8421376) failed (error code=3)
colprof(904) malloc: *** error: can't allocate region
colprof(904) malloc: *** set a breakpoint in szone_error to debug
colprof: Error - rspl malloc failed - reverse cell simplexes - base simplex

Then:
export ARGYLL_REV_CACHE_MULT=0.5
finally:
profile check complete, peak err = 0.942062, avg err = 0.189789

It works.
My next GaMapICC release will contain code to determine the optimal ARGYLL_REV_CACHE_MULT for collink dynamically based on the output of "time -lp collink ...".
There are correlations between cache multiplier, execution time, probability for reverse cell cache warnings and malloc errors and the number of page faults reported by time -lp (see attachment):
- too low cache multiplier settings cause collink reverse cell cache warnings and /slightly/ increased execution times,
- too high values cause disk swapping, page faults, /dramatically/ increasing execution times and finally malloc errors. 

Good ARGYLL_REV_CACHE_MULT settings for collink work for colproof too.
On my MacBook Pro with 4GB RAM I typically end up with cache multiplyers below 0.4. 

Klaus

Other related posts: