[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [MiNT] gcc-performace, was: Samba working



Am 13.12.2013, 23:20 Uhr, schrieb Jean-François Lemaire <jflemaire@skynet.be>:

fatfs.o

need on the firebee? I get with Vincent's cross-gcc:

     2.46s real     2.35s user     0.15s system

(+- 0.5s)

I get: 1m34.395s real,  1m25.315s user, 0m5.665s sys

Longer than I expected ..

I've found the config-parameter JITCacheSize for aranym with was set to 8192 (is that kB?), I changed it to 65536 which increases gcc-speed significantly:

time gcc -c -DHAVE_CONFIG_H -I.. -g -O3 -DKSH=1 -funsigned-char -Wall -DVDEBUG=0 -D_GNU_SOURCE -m68030 -o exec.o ../exec.c
    9.93s real     9.67s user     0.26s system

132|/usr/src/ksh/src/o-ksh}sh -v ./c
time gcc -c -DHAVE_CONFIG_H -I.. -g -O3 -DKSH=1 -funsigned-char -Wall -DVDEBUG=0 -D_GNU_SOURCE -m68030 -o exec.o ../exec.c
    8.64s real     8.46s user     0.18s system

Better than the former 23s, and quite close to 5-7s I'd expect.

With fatfs.c I get:

   17.88s real    16.45s user     1.43s sys

which proves that these times are not linear (which I also expected ;)). I don't think that hostfs (where this was run) has a great impact on the speed.

Anyway I don't think that the gcc-slowdown on atari comes from mintlib, I guess it spends most of the time in code-generation/optimization where mintlib is probably not very much involved. Maybe some of the other libs used have more potential for improvements.

-Helmut