[pgrouting-users] importance of -frounding-math compilation flag?

pgRouting pgrouting-users
1

I'm recompiling pgrouting on OS X Lion, using the clang compiler. clang does not have the -frounding-math option.

I found one reference that the clang equivalent is --disable-fpmath, but that just causes and error "unsupported option '--disable-fpmath'". I could find no documentation that clang has or had this option, or something similar.

Is the rounding behavior critical?

-----
William Kyngesburye <kyngchaos*at*kyngchaos*dot*com>
http://www.kyngchaos.com/

"This is a question about the past, is it? ... How can I tell that the past isn't a fiction designed to account for the discrepancy between my immediate physical sensations and my state of mind?"

- The Ruler of the Universe

2

On 3/16/2013 6:20 PM, William Kyngesburye wrote:

I'm recompiling pgrouting on OS X Lion, using the clang compiler.
clang does not have the -frounding-math option.

I found one reference that the clang equivalent is --disable-fpmath,
but that just causes and error "unsupported option
'--disable-fpmath'". I could find no documentation that clang has or
had this option, or something similar.

Is the rounding behavior critical?

I do not know for a fact one way or the other, But I'm guess that this options probably is not critical for MOST pgrouting functions.

Where it MIGHT be important is for the CGAL library and/or GAUL libraries. CGAL specifically does a lot of math. You might check those packages so see what they do with respect to fp rounding on OS X.

If anyone knows more about this please correct this AND we should document it somewhere for future reference.

Thanks,
   -Steve

3

On Mar 16, 2013, at 6:47 PM, Stephen Woodbridge wrote:

On 3/16/2013 6:20 PM, William Kyngesburye wrote:

I'm recompiling pgrouting on OS X Lion, using the clang compiler.
clang does not have the -frounding-math option.

I found one reference that the clang equivalent is --disable-fpmath,
but that just causes and error "unsupported option
'--disable-fpmath'". I could find no documentation that clang has or
had this option, or something similar.

Is the rounding behavior critical?

I do not know for a fact one way or the other, But I'm guess that this options probably is not critical for MOST pgrouting functions.

Where it MIGHT be important is for the CGAL library and/or GAUL libraries. CGAL specifically does a lot of math. You might check those packages so see what they do with respect to fp rounding on OS X.

If anyone knows more about this please correct this AND we should document it somewhere for future reference.

I didn't need to recompile CGAL, so I didn't see a problem there. But I see that it also uses -frounding-math for GCC, but nothing for clang. The changes log does say CGAL fully supports clang as of CGAL 4.1, so maybe rounding behavior is not an issue or the code works around it.

All I could find in GAUL is a macro define for ROUND which looks like proper rounding towards 0.

I found another reference, for CGAL and OSX/clang:

http://jamesgregson.blogspot.com/2012/06/cgal-build-issues-under-os-x-107.html

I guess we're screwed, if the rounding behavior is important to compiler optimization in CGAL and pgRouting. (the previous reference for the supposed --disable-fpmath option was from Fink. Never liked Fink.)

-----
William Kyngesburye <kyngchaos*at*kyngchaos*dot*com>
http://www.kyngchaos.com/

"We are at war with them. Neither in hatred nor revenge and with no particular pleasure I shall kill every ___ I can until the war is over. That is my duty."

"Don't you even hate 'em?"

"What good would it do if I did? If all the many millions of people of the allied nations devoted an entire year exclusively to hating the ____ it wouldn't kill one ___ nor shorten the war one day."

<Ha, ha> "And it might give 'em all stomach ulcers."

- Tarzan, on war

4

On 3/16/2013 8:36 PM, William Kyngesburye wrote:

On Mar 16, 2013, at 6:47 PM, Stephen Woodbridge wrote:

On 3/16/2013 6:20 PM, William Kyngesburye wrote:

I'm recompiling pgrouting on OS X Lion, using the clang
compiler. clang does not have the -frounding-math option.

I found one reference that the clang equivalent is
--disable-fpmath, but that just causes and error "unsupported
option '--disable-fpmath'". I could find no documentation that
clang has or had this option, or something similar.

Is the rounding behavior critical?

I do not know for a fact one way or the other, But I'm guess that
this options probably is not critical for MOST pgrouting
functions.

Where it MIGHT be important is for the CGAL library and/or GAUL
libraries. CGAL specifically does a lot of math. You might check
those packages so see what they do with respect to fp rounding on
OS X.

If anyone knows more about this please correct this AND we should
document it somewhere for future reference.

I didn't need to recompile CGAL, so I didn't see a problem there.
But I see that it also uses -frounding-math for GCC, but nothing for
clang. The changes log does say CGAL fully supports clang as of CGAL
4.1, so maybe rounding behavior is not an issue or the code works
around it.

All I could find in GAUL is a macro define for ROUND which looks like
proper rounding towards 0.

I found another reference, for CGAL and OSX/clang:

http://jamesgregson.blogspot.com/2012/06/cgal-build-issues-under-os-x-107.html

I guess we're screwed, if the rounding behavior is important to
compiler optimization in CGAL and pgRouting. (the previous reference
for the supposed --disable-fpmath option was from Fink. Never liked
Fink.)

OK, so CGAL is only used for building drive time polygons. The rest of pgRouting should work just fine. I am also considering removing CGAL in Rev 2.0 and using a much smaller and simpler code set that I have used in the past.

This rounding issue might be a more serious issue GEOS which is the underlying geometery library used in postGIS, so I wonder what they are doing about this.

-Steve

----- William Kyngesburye <kyngchaos*at*kyngchaos*dot*com>
http://www.kyngchaos.com/

"We are at war with them. Neither in hatred nor revenge and with no
particular pleasure I shall kill every ___ I can until the war is
over. That is my duty."

"Don't you even hate 'em?"

"What good would it do if I did? If all the many millions of people
of the allied nations devoted an entire year exclusively to hating
the ____ it wouldn't kill one ___ nor shorten the war one day."

<Ha, ha> "And it might give 'em all stomach ulcers."

- Tarzan, on war

_______________________________________________ Pgrouting-users
mailing list Pgrouting-users@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/pgrouting-users

5

On Mar 16, 2013, at 11:04 PM, Stephen Woodbridge wrote:

On 3/16/2013 8:36 PM, William Kyngesburye wrote:

On Mar 16, 2013, at 6:47 PM, Stephen Woodbridge wrote:

On 3/16/2013 6:20 PM, William Kyngesburye wrote:

I'm recompiling pgrouting on OS X Lion, using the clang
compiler. clang does not have the -frounding-math option.

I found one reference that the clang equivalent is
--disable-fpmath, but that just causes and error "unsupported
option '--disable-fpmath'". I could find no documentation that
clang has or had this option, or something similar.

Is the rounding behavior critical?

I do not know for a fact one way or the other, But I'm guess that
this options probably is not critical for MOST pgrouting
functions.

Where it MIGHT be important is for the CGAL library and/or GAUL
libraries. CGAL specifically does a lot of math. You might check
those packages so see what they do with respect to fp rounding on
OS X.

If anyone knows more about this please correct this AND we should
document it somewhere for future reference.

I didn't need to recompile CGAL, so I didn't see a problem there.
But I see that it also uses -frounding-math for GCC, but nothing for
clang. The changes log does say CGAL fully supports clang as of CGAL
4.1, so maybe rounding behavior is not an issue or the code works
around it.

All I could find in GAUL is a macro define for ROUND which looks like
proper rounding towards 0.

I found another reference, for CGAL and OSX/clang:

http://jamesgregson.blogspot.com/2012/06/cgal-build-issues-under-os-x-107.html

I guess we're screwed, if the rounding behavior is important to
compiler optimization in CGAL and pgRouting. (the previous reference
for the supposed --disable-fpmath option was from Fink. Never liked
Fink.)

OK, so CGAL is only used for building drive time polygons. The rest of pgRouting should work just fine. I am also considering removing CGAL in Rev 2.0 and using a much smaller and simpler code set that I have used in the past.

This rounding issue might be a more serious issue GEOS which is the underlying geometery library used in postGIS, so I wonder what they are doing about this.

I found a round function is GEOS, which is a complex (overly?) if-then. If all the rounding behavior affects is a round function, then GEOS is OK. Makes sense - it probably came out of converting the JTS Java to C. I read that the Java rounding behavior is down, not the standard to 0, so JTS probably had the function already.

-----
William Kyngesburye <kyngchaos*at*kyngchaos*dot*com>
http://www.kyngchaos.com/

"Oh, look, I seem to have fallen down a deep, dark hole. Now what does that remind me of? Ah, yes - life."

- Marvin