--0-2076963316-1123574270=:75895 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit
let me get it stright... you would say that you expect problems when mixing -ffast-math with -O9 and optimization? I have huge project (video encoder/decoder) that uses -ffast-math -O9 -mtune=prescott and some more switches all compiled with gcc 3.2 (Suse 9.1 with -march instead mtune) gcc 3.4 (FC3) and gcc 4.0 (FC4) running perfectly with no errors for quite a long while! ciao scipio Oleg Goldshmidt <[EMAIL PROTECTED]> wrote: Shachar Shemesh writes: > Actually, something extremely weird it going on here. The result > change, considerably, when I compile with or without "no-math-errno": > > [EMAIL PROTECTED]:/tmp$ gcc -O2 -fno-math-errno -o test -lm test.c && time > > ./test > > nan > > > > real 0m7.521s > > user 0m7.517s > > sys 0m0.003s > > Why did the answer turn into "nan"? > > Shachar Because when you mix no-math-errno with optimization you may break the proper IEEE spec for math functions. It's right there in gcc docs. The more common case is when you switch fast-math on (which is no-math-errno and a few other options bundled together). Don't ask me why the options exist: I will say that all errors should be checked. I suspect that there are architectures (Motorolas) that have lots of special math functions, including sqrt, sin, cos, etc implemented as single instructions, and the fast-math options exist for those. -- Oleg Goldshmidt | [EMAIL PROTECTED] | http://www.goldshmidt.org ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED] __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --0-2076963316-1123574270=:75895 Content-Type: text/html; charset=iso-8859-1 Content-Transfer-Encoding: 8bit <DIV>let me get it stright...</DIV> <DIV> </DIV> <DIV>you would say that you expect problems when mixing -ffast-math with -O9 and optimization?</DIV> <DIV> </DIV> <DIV>I have huge project (video encoder/decoder) that uses -ffast-math -O9 -mtune=prescott and some more switches all compiled with gcc 3.2 (Suse 9.1 with -march instead mtune)</DIV> <DIV>gcc 3.4 (FC3) and gcc 4.0 (FC4) running perfectly with no errors for quite a long while!</DIV> <DIV> </DIV> <DIV>ciao</DIV> <DIV>scipio <BR><BR><B><I>Oleg Goldshmidt <[EMAIL PROTECTED]></I></B> wrote:</DIV> <BLOCKQUOTE class=replbq style="PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #1010ff 2px solid">Shachar Shemesh <[EMAIL PROTECTED]>writes:<BR><BR>> Actually, something extremely weird it going on here. The result<BR>> change, considerably, when I compile with or without "no-math-errno":<BR><BR><SNIP><BR><BR>> > [EMAIL PROTECTED]:/tmp$ gcc -O2 -fno-math-errno -o test -lm test.c && time ./test<BR>> > nan<BR>> ><BR>> > real 0m7.521s<BR>> > user 0m7.517s<BR>> > sys 0m0.003s<BR>> <BR>> Why did the answer turn into "nan"?<BR>> <BR>> Shachar<BR><BR>Because when you mix no-math-errno with optimization you may break the<BR>proper IEEE spec for math functions. It's right there in gcc docs. The<BR>more common case is when you switch fast-math on (which is<BR>no-math-errno and a few other options bundled together).<BR><BR>Don't ask me why the options exist: I will say that all errors should<BR>be checked. I suspect that there are architectures (Motorolas) that<BR>have lots of special math functions, including sqrt, sin, cos, etc<BR>implemented as single instructions, and the fast-math options exist<BR>for those.<BR><BR>-- <BR>Oleg Goldshmidt | [EMAIL PROTECTED] | http://www.goldshmidt.org<BR><BR>=================================================================<BR>To unsubscribe, send mail to [EMAIL PROTECTED] with<BR>the word "unsubscribe" in the message body, e.g., run the command<BR>echo unsubscribe | mail [EMAIL PROTECTED]<BR><BR></BLOCKQUOTE><p>__________________________________________________<br>Do You Yahoo!?<br>Tired of spam? Yahoo! Mail has the best spam protection around <br>http://mail.yahoo.com --0-2076963316-1123574270=:75895-- ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]