... to be clear, I meant optimizing C/C++ to get better assembler output. as I understood it, you wanted to check the number of CPU cycles each assembler instruction does to see if the code can be tweaked to use more-efficient/less instructions.
On Wed, Oct 10, 2012 at 9:37 AM, Campbell Barton <[email protected]> wrote: > Hi Dwight, > > Not as far as I know, the way I've used assembly in blender is just to > check code is correctly optimized out, or to verify that some change > is actually reducing the compiled output as intended. > > We have a lot of room for improvement by making parts of the code > threaded and improving the algorithms too, so if your looking to > optimize in general there are likely higher level areas that can be > improved. > > But if you like to check this low level stuff - it could be helpful, > just need to figure out the best areas to check/optimize... > > Some suggestions off hand where some low level magic might help... > > - frequent called math functions (math_*.c), matrix multiply, > normalize, color conversions from float to bytes for eg, (any pixel > operations really),... > - the new compositor, (COM_*.cpp files) speedups with common > operations here too? > - frequent called code in bmesh (perhaps bmesh_iterators.c or > BLI_mempool.c's iterator) > - low level library functions - BLI_ghash.c (used a lot and not sure > its been profiled for speedups much already) > > On Wed, Oct 10, 2012 at 8:38 AM, Dwight Hutto <[email protected]> wrote: >> I had a question in an earlier thread about disassembled Blender. I'm >> not new to Assembly, but just a little naive. >> >> Has anyone ever tried to disassemble, and optimize instructions given >> by the compilers to the processors instruction set(cycle count)? >> >> If not, is this a good project to work on? >> >> -- >> Best Regards, >> David Hutto >> CEO: http://www.hitwebdevelopment.com >> _______________________________________________ >> Bf-committers mailing list >> [email protected] >> http://lists.blender.org/mailman/listinfo/bf-committers > > > > -- > - Campbell -- - Campbell _______________________________________________ Bf-committers mailing list [email protected] http://lists.blender.org/mailman/listinfo/bf-committers
