Yes please, send me the patch and I'll give it a shot. On Tue, Jun 14, 2016 at 6:11 PM, Larry Gritz <[email protected]> wrote:
> Oops, yes, some of those stats print the "top 3" in various categories but > skip any that were broken or invalid. If any of the top 3 are broken, fewer > than 3 will print! > > Proposed fix is here: https://github.com/OpenImageIO/oiio/pull/1433 > > I also incorporated your other suggestion of explicitly listing any > broken/invalid files. > > Are you able to apply that patch on your end and try it out to let me know > if it works well for your case and prints helpful output? > > On Jun 14, 2016, at 2:16 AM, Søren Ragsdale <[email protected]> wrote: > > I've got some renders which print texture statistics normally. (There are > textures in the render.) The introduction of a hair system (adding more > textures) causes "top files by bytes read" and "top files by I/O time" to > stop printing, even though "slowest I/O rates" still prints: > > OpenImageIO ImageCache statistics (shared) ver 1.6.14 >> Options: max_memory_MB=8000.0 max_open_files=100 autotile=64 >> autoscanline=0 automip=1 forcefloat=0 accept_untiled=1 >> accept_unmipped=1 read_before_insert=0 deduplicate=1 >> unassociatedalpha=0 failure_retries=0 >> Images : 127 unique >> ImageInputs : 127 created, 95 current, 95 peak >> Total size of all images referenced : 21.7 GB >> Read from disk : 2.8 GB >> Find file time : 31.0s >> File I/O time : 30m 39.4s (38.3s average per thread) >> File open time only : 10.4s >> Tiles: 88598 created, 16324 current, 24896 peak >> total tile requests : 13153664 >> micro-cache misses : 4444498 (33.789%) >> main cache misses : 18525 (0.140835%) >> Peak cache memory : 115.0 MB >> Find tile time : 27.8s >> 32 were constant-valued in all pixels >> >> * Top files by bytes read:* >> * Top files by I/O time:* >> * Files with slowest I/O rates:* >> * 1 0.01 MB/s (0.00MB/0.58s) 8x 8x4.u8 >> /jobs/CFW/3D/ITEM/character/deer/body/TEXS/MSK/default_alt0/T_character_deer-body_default_alt0_MSK_2_lodDefault_v004/T_character_deer-body_default_alt0_MSK_2_lodDefault.1007.tex >> MIP-COUNT [1,1,1]* >> * 2 0.38 MB/s (0.29MB/0.78s) 8192x8192x1.u8 >> /jobs/CFW/3D/ITEM/character/deer/body/TEXS/SPCR/default_alt0/T_character_deer-body_default_alt0_SPCR_1_lodDefault_v006/T_character_deer-body_default_alt0_SPCR_1_lodDefault.1007.tex >> MIP-COUNT [0,20,18,23,8,3,1,1,1,1,1,1,1]* >> * 3 0.70 MB/s (0.21MB/0.29s) 8192x8192x1.u8 >> /jobs/CFW/3D/ITEM/character/deer/body/TEXS/SPCR/default_alt0/T_character_deer-body_default_alt0_SPCR_1_lodDefault_v006/T_character_deer-body_default_alt0_SPCR_1_lodDefault.1006.tex >> MIP-COUNT [0,0,20,17,9,4,1,1,1,1,1,1,1]* (fastest was 26.7 MB/s) > > > > I've run this through gdb, and it looks like it's failing in this way. The > top-ranked files are either broken or have an invalid spec, so they don't > print: > > 1711 for (int i = 0; i < std::min<int> (topN, files.size()); ++i) { >> (gdb) >> 1712 if (files[i]->broken() || !files[i]->validspec()) >> (gdb) >> 1713 continue; >> (gdb) > > > > I think that OIIO should print which textures are broken or invalid (which > is something we should know), and it should also print the top three > textures which are valid and not-broken. > _______________________________________________ > Oiio-dev mailing list > [email protected] > http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org > > > -- > Larry Gritz > [email protected] > > > > _______________________________________________ > Oiio-dev mailing list > [email protected] > http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org > >
_______________________________________________ Oiio-dev mailing list [email protected] http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
