On Fri 5 Mar 2010 21:19, vivi...@analog.com pondered: > Using built-in specs. > Target: bfin-uclinux > Configured with: /home/test/work/cruise/checkouts/toolchain/gcc-4.3/configure --build=i686-pc-linux-gnu --host=i686-pc-linux-gnu --target=bfin-uclinux --prefix=/home/test/work/cruise/test_scripts/toolchain/toolchain-build/toolchain-2010_Feb_25_18_41/bfin-uclinux --enable-threads=posix --with-sysroot=/home/test/work/cruise/test_scripts/toolchain/toolchain-build/toolchain-2010_Feb_25_18_41/bfin-uclinux/bfin-uclinux/runtime --disable-libstdcxx-pch --disable-symvers --disable-libssp --disable-libffi --enable-clocale=generic --enable-version-specific-runtime-libs --enable-__cxa_atexit --with-bugurl=URL:http://blackfin.uclinux.org/gf/project/toolchain/tracker --with-pkgversion=ADI-trunk/svn-3951 > Thread model: posix > gcc version 4.3.4 (ADI-trunk/svn-3951)
First - "Yeah!". second - we (eventually) need to get this going to a different place - where things can be graphed, to look at long term results/changes as we change things (to ensure there aren't any regressions). For now - this is as good a place as any - until the data format is stable, then I'll start parsing things. > Test result of whetstone > ############################## > Loops: 1000, Iterations: 1, Duration: 3 sec. > C Converted Double Precision Whetstones: 33.3 MIPS Can you do a: ---- make user/whetstone_only TONLY=results IP=10.64.204.74 | grep “|” | sort -rn -t “|” +4 ---- That should provide a little more meaningful information that just the default (and actually let us know how the compiler is working, and if the fast float is OK). > Test result of dhrystone > ############################## [snip] > Dhrystone Benchmark, Version 2.1 (Language: C) > Microseconds for one run through Dhrystone: 1.8 > Dhrystones per Second: 558971.5 Can we just get the output? I think there is also a results target too. That will make things easier to parse and graph. > Results of nbench > ############################## > BYTEmark* Native Mode Benchmark ver. 2 (10/95) > Index-split by Andrew D. Balsa (11/97) > Linux/Unix* port by Uwe F. Mayer (12/96,11/97) > > TEST : Iterations/sec. : Old Index : New Index > : : Pentium 90* : AMD K6/233* > --------------------:------------------:-------------:------------ > NUMERIC SORT : 81.927 : 2.10 : 0.69 > STRING SORT : 8.2547 : 3.69 : 0.57 > BITFIELD : 4.4834e+07 : 7.69 : 1.61 > FP EMULATION : 12.03 : 5.77 : 1.33 > FOURIER : 62.441 : 0.07 : 0.04 > ASSIGNMENT : 1.6405 : 6.24 : 1.62 > IDEA : 395.41 : 6.05 : 1.80 > HUFFMAN : 131.09 : 3.64 : 1.16 > NEURAL NET : 0.071602 : 0.12 : 0.05 > LU DECOMPOSITION : 2.284 : 0.12 : 0.09 > ==========================ORIGINAL BYTEMARK RESULTS========================== > INTEGER INDEX : 4.653 > FLOATING-POINT INDEX: 0.099 > Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0 > ==============================LINUX DATA BELOW=============================== > CPU : Analog Devices ADSP-BF537 500(MHz CCLK) 125(MHz SCLK) (mpu off) 500MHz > L2 Cache : 16 KB(L1 icache) 32 KB(L1 dcache) 0 KB(L2 cache) > OS : Linux 2.6.32.9-ADI-2010R1-pre-svn8357 > C compiler : bfin-uclinux-gcc > libc : static > MEMORY INDEX : 1.141 > INTEGER INDEX : 1.176 > FLOATING-POINT INDEX: 0.055 > Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38 > * Trademarks are property of their respective holder. Do we want more than just the default "-O2" for this one? > Results of edn > ############################## > Test cycles per loop > overhead 38.3 > vec_mpy1 547.8 > mac 653.8 > fir 7130.7 > fir_no_red_ld 12012.0 > latsynth 1291.8 > iir1 691.5 > codebook 332.3 > jpegdct 2151.7 This is good. I'll add a results (with different compiler options) soon. > Results of lmbench (run lat_ctx) > ############################## > > "size=0k ovr=9.88 > 2 8.55 > 8 19.98 > 16 24.18 > > > "size=16k ovr=49.84 > 2 96.60 > 8 100.33 > 16 99.26 > > > "size=64k ovr=422.23 > 2 30.13 > 8 37.27 > 16 38.89 Don't we do any more lmbench than just lat_ctx? _______________________________________________ Toolchain-devel mailing list Toolchain-devel@blackfin.uclinux.org https://blackfin.uclinux.org/mailman/listinfo/toolchain-devel
