Hi Stephen, Do you know if this issue is on SolidAngle's radar for fixing in the MtoA package soon? For high resolution source images this overhead can be quite significant particularly if it happens on a farm per frame.
On Mon, Jun 4, 2018 at 10:33 AM, Will Anielewicz <[email protected]> wrote: > Hi Stephen, > > Are you mentioning the difference between maketx in MtoA as opposed to > Arnold5.1.0.1 > because we may see unexpected results when using these options > > maketx -v --runstats --oiio --threads 7 --compression zip --format exr -d > half > > thanks > > On Mon, Jun 4, 2018 at 9:36 AM, Stephen Blair <[email protected]> > wrote: > >> The maketx that comes with Arnold 5.1.0.1 uses OpenImageIO 1.7.17 >> >> MtoA includes an older maketx that has been customized (to work with >> syncolor, iirc) >> >> On Mon, Jun 4, 2018 at 12:28 PM, Will Anielewicz < >> [email protected]> wrote: >> >>> Hi Thiago, >>> >>> Thanks for the suggestion of -threads 1 ( and yes it did make the hash >>> processing take 13s on my workstation) >>> Maya about Arnold is Arnold Core 5.1.0.1 which I think would include >>> the "newer" maketx >>> which is exhibiting the slow processing >>> my maketx is reporting >>> OpenImageIO 1.7.7 >>> Does that match the version you are using? >>> >>> On Mon, Jun 4, 2018 at 8:08 AM, Thiago Ize <[email protected]> >>> wrote: >>> >>>> Thanks for the repro! I was able to get similarly poor hashing times >>>> using the same older maketx that you reported in the other email list. >>>> Changing to "--threads 1" fixes the slowdown (hashing takes 5.8s and the >>>> total maketx time is 33s). Using a newer maketx (as supplied from at least >>>> Arnold 5.1.0) gives a big speedup in "--threads 1" mode, and hashing is now >>>> expectedly faster with "--threads 7" (the entire maketx run takes less than >>>> 13s on my laptop). >>>> >>>> So either switch to single threaded maketx, or better yet, upgrade your >>>> maketx to that supplied in a newer Arnold. >>>> >>>> On Mon, Jun 4, 2018 at 8:36 AM, Will Anielewicz < >>>> [email protected]> wrote: >>>> >>>>> This is a cross posting with 3d-pro. Larry Gritz suggested I post this >>>>> here as well. >>>>> We are using many openexr 8k textures for some CG production. >>>>> Several of them can take many minutes to convert to .tx >>>>> It seems to be correlated to image complexity. >>>>> >>>>> Example openexr image on googledrive: >>>>> https://drive.google.com/open?id=118ucgUMlUjxuDg0z_kLoDpMEA8Cu6F7c >>>>> >>>>> >>>>> Here are some runstats >>>>> >>>>> >>>>> Notice this stat in particular >>>>> >>>>> SHA-1 hash 4m 52.17s (492.6 MB) >>>>> >>>>> >>>>> maketx -v --runstats --oiio --threads 7 --compression zip --format exr >>>>> -d half testEXR_Mari_16bit_8K.exr >>>>> >>>>> I have a 16bit float exr test image which is taking 5 minutes on my >>>>> workstation >>>>> >>>>> prep 0.00s (10.6 MB) >>>>> Reading file: testEXR_Mari_16bit_8K.exr >>>>> read "testEXR_Mari_16bit_8K.exr" 0.00s (10.7 MB) >>>>> misc2 16.32s (274.7 MB) >>>>> misc3 0.00s (274.7 MB) >>>>> resize & data convert 0.00s (274.7 MB) >>>>> SHA-1: 2F36DD66063B77836B264CEBEA56B1A9B5064FA8 >>>>> SHA-1 hash 4m 52.17s (492.6 MB) >>>>> AverageColor: 0.531764,0.510344,0.514165,1 >>>>> misc4 0.00s (492.6 MB) >>>>> Writing file: testEXR_Mari_16bit_8K.898d81fa.temp.tx >>>>> Filter "box" >>>>> Top level is 8192x8192 >>>>> Mipmapping... >>>>> 4096x4096 (1018.4 MB) >>>>> 2048x2048 (1018.4 MB) >>>>> 1024x1024 (778.4 MB) >>>>> 512x512 (778.4 MB) >>>>> 256x256 (762.4 MB) >>>>> 128x128 (762.4 MB) >>>>> 64x64 (762.4 MB) >>>>> 32x32 (762.4 MB) >>>>> 16x16 (762.4 MB) >>>>> 8x8 (762.4 MB) >>>>> 4x4 (762.4 MB) >>>>> 2x2 (762.4 MB) >>>>> 1x1 (762.4 MB) >>>>> Wrote file: testEXR_Mari_16bit_8K.898d81fa.temp.tx (762.4 MB) >>>>> maketx run time (seconds): 333.31 >>>>> file read: 0.00 >>>>> file write: 10.36 >>>>> initial resize: 0.00 >>>>> hash: 292.17 >>>>> mip computation: 14.45 >>>>> color convert: 0.00 >>>>> unaccounted: 16.33 ( 0.00 16.32 0.00 0.00) >>>>> maketx peak memory used: 1018.4 MB >>>>> >>>>> workstation specs >>>>> >>>>> CPU op-mode(s): 32-bit, 64-bit >>>>> Byte Order: Little Endian >>>>> CPU(s): 12 >>>>> On-line CPU(s) list: 0-11 >>>>> Thread(s) per core: 2 >>>>> Core(s) per socket: 6 >>>>> Socket(s): 1 >>>>> NUMA node(s): 1 >>>>> Vendor ID: GenuineIntel >>>>> CPU family: 6 >>>>> Model: 63 >>>>> Model name: Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz >>>>> >>>>> >>>>> >> >> >
_______________________________________________ Oiio-dev mailing list [email protected] http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
