https://bugzilla.novell.com/show_bug.cgi?id=811546 https://bugzilla.novell.com/show_bug.cgi?id=811546#c2 Richard Biener changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |rguenther@suse.com AssignedTo|rguenther@suse.com |schwab@suse.com --- Comment #2 from Richard Biener 2013-03-26 08:55:51 UTC ---

with math functions sin(), cos(), exp(), rand(),

this is a glibc issue which now includes different versions for these functions (you can verify this by running the executable compiled on 12.3 on 12.2 and vice-versa). -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=811546 https://bugzilla.novell.com/show_bug.cgi?id=811546#c4 Stefan Ziegler changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEEDINFO |NEW InfoProvider|j9002j@yahoo.com | --- Comment #4 from Stefan Ziegler 2013-03-26 19:11:20 UTC --- (In reply to comment #3)

Please provide a test case.

Please see the C++ code example in opensuse forums, in https://forums.opensuse.org/english/other-forums/development/programming-scr... in post # 20. The running times are brought there as well. The slowdown in opensuse 12.3 is twice. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=811546 https://bugzilla.novell.com/show_bug.cgi?id=811546#c6 --- Comment #6 from Stefan Ziegler 2013-03-26 21:01:20 UTC --- Reportedly, replacing libm in /lib64/ by the one from opensuse 12.2 restores the speed; see post # 33 in https://forums.opensuse.org/english/other-forums/development/programming-scr... -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=811546 https://bugzilla.novell.com/show_bug.cgi?id=811546#c8 --- Comment #8 from Aaron Burgemeister 2013-04-09 13:28:09 UTC --- Created an attachment (id=534220) --> (http://bugzilla.novell.com/attachment.cgi?id=534220) Sample code from post #49 in the forums further demonstrating decreased performance This was from post #49 as visible here: https://forums.opensuse.org/english/other-forums/development/programming-scr... Also saw a large slowdown on comparable systems when running on 12.3 vs. 12.2. Using libm.so.6 from 12.2 x86_64 on the 12.3 x86_64 system increased performance by several times as demonstrated below (using this code): ab@mybox:/tmp> time ./a.out x = 875.574 real 0m36.207s user 0m36.145s sys 0m0.001s ab@mybox:/tmp> time LD_PRELOAD=/tmp/libm-from-12.2.so.6 ./a.out x = 875.574 real 0m13.249s user 0m13.226s sys 0m0.001s -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=811546 https://bugzilla.novell.com/show_bug.cgi?id=811546#c10 Andreas Jaeger changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |aj@suse.com --- Comment #10 from Andreas Jaeger 2013-04-09 18:14:38 UTC --- I just added the following to the forums threads: As one of the upstream glibc developers and openSUSE packagers, let me chime into this discussion: Upstream glibc contains many changes for the math library so that it is standard conform also in all corner cases and accurately calculates numbers - the goal is accuracy for all digits. This might lead in some cases to slower runtime execution. For the upstream sin/cos bugs that changed this behaviour for sin and cos, see http://sourceware.org/bugzilla/show_bug.cgi?id=13658. Glibc developers are aware that some functions are slower and have started now with addressing some of that slowness and reached already some improvements. Also, glibc now contains a benchmark test suite that can be run before/after making changes to check that we do not introduce slowdowns by accident. If you read comment 4 of http://sourceware.org/bugzilla/show_bug.cgi?id=13932, the worst case performance of pow has been improved already 4 times. Coming back to sin/cos: Previously the fsincos hardware instruction was used and it is highly inaccurate for larger values - and therefore we had to use a software implementation. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=811546 https://bugzilla.novell.com/show_bug.cgi?id=811546#c12 Franz Bernasek changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |fbernasek@netscape.net --- Comment #12 from Franz Bernasek 2013-05-09 09:37:46 UTC --- Hi, i have compile the sample on 12.3, with gcc and the same options #include int main() { for(int i=0; i<100000000; i++) { sin(0); } } Compiled with gcc -fno-builtin -O0 -std=c99 check_speed.c -Wl,-rpath=. -lm linux-cuda55:/hpc_programming/testprgs # time ./a.out real 0m1.054s user 0m1.054s sys 0m0.000s now i think its not slow :-) -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.