More from my protracted examination of tweaking my builds. I've now updated everything at http://www.linuxfromscratch.org/~ken/tuning/
Specifically: tuning-3-mtune-not-march.txt: ---------------------------- I took my build with -march=cpu and the various cheap hardening flags (tuning-2) and replaced -march=native with -mtune=native. Then I stripped out the cheap hardening. Some of the more simple compiles are a little quicker like this, but runtimes are generally slower. The one exception is 'sox' (not in BLFS) which runs faster using -mtune and faster still with -mtune and cheap hardening. A few of the packages I build take significantly longer when not using -march=natve. My current view is that IFF you can use -march=native on modernish machines, then adding the Cheap Hardening options produces faster programs than a default unhardened build. Reasons NOT to use the -march=native flag are also mentioned. tuning-notes-2A.txt: ------------------- I'm still hoping to get around to trying -O3 with more packages. Today I've been using the completed system to play with compiling rustc-1.35.0 and firefox-67 to that end (had to make some changes to my scripts to allow -Copt-level=3 to be passed to rust). From what this produced, I've updated notes 30 and 37 - I am now dubious that using -Copt-level=3 will make much differnece to these two packages. desktop-runtime-comparisons.ods ------------------------------- I've added an extra test to this LibreOffice spreadsheet (ImageMagick using LibRaw to convert some example camera raw files - which are not described as tiff variations - to png), because I noticed that the LibRaw compilation was faster when -mtune was used instead of -march and I wondered if the runtime was also faster. Turns out the runtime was slower, as expected. The spreadsheet was getting unwieldy, so I've revised my descriptions of the different builds and also tweaked the percentage calculations to properly record negative (i.e. faster) mean values. Maybe I can now get on to the sort of tuning options I really want to test ;-) ĸen -- This is magic for grown-ups; it has to be hard because we know there's no such thing as a free goblin. -- Pratchett, Stewart & Cohen - The Science of Discworld II -- http://lists.linuxfromscratch.org/listinfo/blfs-dev FAQ: http://www.linuxfromscratch.org/blfs/faq.html Unsubscribe: See the above information page
