Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Tom Lane writes: > Sean Chittenden <[EMAIL PROTECTED]> writes: > > Is there a place where -fno-fast-math > > could be used as a CC option if the CC is gcc? > > configure is what I had in mind ;-). I can't think of any part of the > code where we'd really want this sort of optimization enabled. Today I read that __FAST_MATH__ is defined if -ffast-math is used, so it should be easy to write a test in configure. -- Peter Eisentraut [EMAIL PROTECTED] ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
> > After looking through gcc, using -O and -ffast-math will create broken
> > code, but -O2 -ffast-math _should_ be okay.
>
> At least in the gcc shipped with Red Hat 7.2, it doesn't seem to matter:
> you get the wrong answer regardless of -O level. Here's the test case
> I used:
>
> [tgl@rh1 tgl]$ cat bug.c
> #include
>
> double d18000 = 18000.0;
>
> main() {
> int d = d18000 / 3600;
> printf("18000.0 / 3600 = %d\n", d);
> return 0;
> }
> [tgl@rh1 tgl]$ gcc bug.c
> [tgl@rh1 tgl]$ ./a.out
> 18000.0 / 3600 = 5-- right
> [tgl@rh1 tgl]$ gcc -O2 -ffast-math bug.c
> [tgl@rh1 tgl]$ ./a.out
> 18000.0 / 3600 = 4-- wrong
> -- I get 4 if -ffast-math, -O doesn't affect it
> [tgl@rh1 tgl]$ gcc -v
> Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/2.96/specs
> gcc version 2.96 2731 (Red Hat Linux 7.1 2.96-98)
Heh, chalk this one up as another Linux-ism then 'cause it's not
present in FreeBSD -stable or -current. This actually makes me feel
better about setting an option in the -devel port for turning on
compilation with -O3. -sc
stable$ gcc -v
Using builtin specs.
gcc version 2.95.4 20020320 [FreeBSD]
current$ gcc -v
Using built-in specs.
Configured with: FreeBSD/i386 system compiler
Thread model: posix
gcc version 3.2.1 [FreeBSD] 20020901 (prerelease)
--
Sean Chittenden
---(end of broadcast)---
TIP 2: you can get off all lists at once with the unregister command
(send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Sean Chittenden <[EMAIL PROTECTED]> writes:
> Is there a place where -fno-fast-math
> could be used as a CC option if the CC is gcc?
configure is what I had in mind ;-). I can't think of any part of the
code where we'd really want this sort of optimization enabled.
> After looking through gcc, using -O and -ffast-math will create broken
> code, but -O2 -ffast-math _should_ be okay.
At least in the gcc shipped with Red Hat 7.2, it doesn't seem to matter:
you get the wrong answer regardless of -O level. Here's the test case
I used:
[tgl@rh1 tgl]$ cat bug.c
#include
double d18000 = 18000.0;
main() {
int d = d18000 / 3600;
printf("18000.0 / 3600 = %d\n", d);
return 0;
}
[tgl@rh1 tgl]$ gcc bug.c
[tgl@rh1 tgl]$ ./a.out
18000.0 / 3600 = 5 -- right
[tgl@rh1 tgl]$ gcc -O2 -ffast-math bug.c
[tgl@rh1 tgl]$ ./a.out
18000.0 / 3600 = 4 -- wrong
-- I get 4 if -ffast-math, -O doesn't affect it
[tgl@rh1 tgl]$ gcc -v
Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/2.96/specs
gcc version 2.96 2731 (Red Hat Linux 7.1 2.96-98)
regards, tom lane
---(end of broadcast)---
TIP 3: if posting/reading through Usenet, please send an appropriate
subscribe-nomail command to [EMAIL PROTECTED] so that your
message can get through to the mailing list cleanly
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
> > Out of curiosity: why does -ffast-math break the datetime rounding code? What code bits is this for? Is there a place where -fno-fast-math could be used as a CC option if the CC is gcc? After looking through gcc, using -O and -ffast-math will create broken code, but -O2 -ffast-math _should_ be okay. If it's not, then -O2 -fno-fast-math is likely the correct work around for GCC. -sc -- Sean Chittenden ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Thomas Lockhart wrote: > ... > > > > Hard to imagine why anyone would want such an optimization. How much > > > > faster could it possibly be? > > > Back in ye bad olde days, there was probably an order-of-magnitude > > > difference between the speed of a float multiply and that of a float > > > divide; so this used to be a pretty standard sort of optimization. > > > I can remember doing the equivalent thing by hand in source code. > > > On modern hardware I doubt it makes much difference... > > The speed difference is still there; you can't do a divide with one pass > through a barrel shifter, which is how the multiply operations were sped > up. I vaguely recall something like a 40:1 ratio on a 604 PowerPC but > that may not be accurate, or may be for integers which have similar > issues. Wow, 40:1 is a major difference. > > And you would have to do a heck of a lot of them to see a difference. > > ?? Some applications *do* do a heck of a lot of them ;) Yep, I am sure. The question is whether there is ever a reason to use such a flag on the database code. My guess is no. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Tom Lane wrote: > Bruce Momjian <[EMAIL PROTECTED]> writes: > > Tom Lane wrote: > >> with -ffast-math gcc will "improve" it to > >> > >> int hour = time * 0.000278; > > > Hard to imagine why anyone would want such an optimization. How much > > faster could it possibly be? > > Back in ye bad olde days, there was probably an order-of-magnitude > difference between the speed of a float multiply and that of a float > divide; so this used to be a pretty standard sort of optimization. > I can remember doing the equivalent thing by hand in source code. > > On modern hardware I doubt it makes much difference... And you would have to do a heck of a lot of them to see a difference. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Bruce Momjian <[EMAIL PROTECTED]> writes: > Tom Lane wrote: >> with -ffast-math gcc will "improve" it to >> >> int hour = time * 0.000278; > Hard to imagine why anyone would want such an optimization. How much > faster could it possibly be? Back in ye bad olde days, there was probably an order-of-magnitude difference between the speed of a float multiply and that of a float divide; so this used to be a pretty standard sort of optimization. I can remember doing the equivalent thing by hand in source code. On modern hardware I doubt it makes much difference... regards, tom lane ---(end of broadcast)--- TIP 4: Don't 'kill -9' the postmaster
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Tom Lane wrote: > Ron Johnson <[EMAIL PROTECTED]> writes: > > Out of curiosity: why does -ffast-math break the datetime rounding code? > > We dug into this last night, and it turns out that the culprit is code > like > > int hour = time / 3600; > > where time is a double. This yields an exact result when done > correctly, but with -ffast-math gcc will "improve" it to > > int hour = time * 0.000278; > > the constant being the nearest double value to 1.0 / 3600.0. The > problem is that the constant is inexact and in fact is slightly too > large; so for example if time is exactly 18000.0, you get a resulting > hour value of 4, not 5, after truncation to integer. Repeated a couple > more times, what should have been 5:00:00 comes out as 4:59:60 ... Hard to imagine why anyone would want such an optimization. How much faster could it possibly be? I guess if you were doing only complex math approximations, it would be a win, but that isn't really proper for a database. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [NOVICE] [BUGS] Postgres storing time in strange manner
Ron Johnson <[EMAIL PROTECTED]> writes: > Out of curiosity: why does -ffast-math break the datetime rounding code? We dug into this last night, and it turns out that the culprit is code like int hour = time / 3600; where time is a double. This yields an exact result when done correctly, but with -ffast-math gcc will "improve" it to int hour = time * 0.000278; the constant being the nearest double value to 1.0 / 3600.0. The problem is that the constant is inexact and in fact is slightly too large; so for example if time is exactly 18000.0, you get a resulting hour value of 4, not 5, after truncation to integer. Repeated a couple more times, what should have been 5:00:00 comes out as 4:59:60 ... regards, tom lane ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
