Bug#618799: libvigraimpex: unacceptably slow on mips
Hi, On Sat, Mar 19, 2011 at 03:17:31PM +0100, Jakub Wilk wrote: For a reason on another the attached program[0] is disturbingly slow on gabrielli.debian.org. It takes 74 seconds to run it; even my qemu mips machine (600 bogomips) is ca. 10 times faster. Dear mips porters: - Could you test speed of the program on other mips hardware? Will do that later, currently my mips machine is busy building other packages, so speed results won't be representative. - Do you have any ideas what could be the cause of such a bad efficiency? gabrielli.d.o and some of the mips buildd don't have FPU, so the floating point instructions are emulated by the kernel. That explains why it is so slow. OTOH QEMU is emulating CPU with FPU by default (that can be changed), so that's why it is faster. The best is probably to only build this package on a buildd which has FPU support. I have tweaked the buildd configuration, so it should be the case now. Aurelien -- Aurelien Jarno GPG: 1024D/F1BCDB73 aurel...@aurel32.net http://www.aurel32.net -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#618799: libvigraimpex: unacceptably slow on mips
For a reason on another the attached program[0] is disturbingly slow on gabrielli.debian.org. It takes 74 seconds to run it; even my qemu mips machine (600 bogomips) is ca. 10 times faster. Dear mips porters: - Could you test speed of the program on other mips hardware? - Do you have any ideas what could be the cause of such a bad efficiency? [0] To compile it, you need to have libvigraimpex-dev installed, but no special compiler flags are needed. -- Jakub Wilk #include vigra/stdimage.hxx #include vigra/boundarytensor.hxx int main(int argc, char **argv) { vigra::DImage img(200, 300); vigra::DVector3Image bt(img.size()); boundaryTensor1(srcImageRange(img), destImage(bt), 2.0); }
Bug#618799: libvigraimpex: unacceptably slow on mips
Source: libvigraimpex Version: 1.7.1+dfsg-1 Severity: serious User: debian-m...@lists.debian.org Usertags: mips $ grep -m1 'all 100 trees' *{mips,arm}*log | sort -t _ -k3 libvigraimpex_1.7.1+dfsg-1_armel.log:all 100 trees have been learned in 54688.1 msec libvigraimpex_1.7.1+dfsg-3_armel.log:all 100 trees have been learned in 56391.8 msec libvigraimpex_1.7.1+dfsg-2_armel.log:all 100 trees have been learned in 57515.8 msec libvigraimpex_1.7.1+dfsg-3_mips.log:all 100 trees have been learned in 2.27253e+06 msec libvigraimpex_1.7.1+dfsg-2_mips.log:all 100 trees have been learned in 2.28524e+06 msec libvigraimpex_1.7.1+dfsg-1_mipsel.log:all 100 trees have been learned in 13943.6 msec libvigraimpex_1.7.1+dfsg-3_mipsel.log:all 100 trees have been learned in 16365.9 msec So it looks like for this particular test mips is ~40 times slower than armel and ~150 times slower than mipsel. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org