On Sun, Aug 05, 2007 at 11:20:18AM -0400, Tom Lane wrote:
> "Dirk Tilger" <[EMAIL PROTECTED]> writes:
> > Operating system: Linux with Intel compiler on ia64
>
> > I have been compiling postgresql 8.0, 8.1 and 8.2.4 with the Intel compiler
> > in the past successfully. This time something went wrong and although I
> > can't tell precisely how I triggered it, I have found a fix.
>
> BTW, what I found in googling suggested that (1) icc never has defined
> __ICC on 64-bit machines, and (2) although it does define __GNUC__ by
> default, there is a way to turn that off. So I surmise that your
> relevant change was adding a compiler flag that disabled the definition
> of __GNUC__. That doesn't seem to have stopped it from accepting
> GNU-style asm directives, though, so I kinda wonder what compiler
> behavior does change and what was the point of your flag change.
The compiler seems to have been called with:
| icc -mp -no-gcc -mcpu=itanium2 -mtune=itanium2
The manual page about '-no-gcc' says:
-no-gcc
Do not predefine the __GNUC__, __GNUC_MINOR__, and __GNUC_PATCHLEVEL__
macros.
FYI: we also apply the attached patch to postgre, so that the
'configure' script would not add the -mp1, when -mp was already
specified. From the manual:
| -mp Maintain floating-point precision (disables some opti-
| mizations). The -mp option restricts optimization to
| maintain declared precision and to ensure that float-
| ing-point arithmetic conforms more closely to the ANSI
| and IEEE standards. For most programs, specifying this
| option adversely affects performance. If you are not sure
| whether your application needs this option, try compiling
| and running your program both with and without it to
| evaluate the effects on both performance and precision.
|
|
| -mp1 Improve floating-point precision. -mp1 disables fewer
| optimizations and has less impact on performance than
| -mp.
Best regards,
Dirk Tilger.
--- postgresql-8.1.5/configure.in 2006-10-12 23:41:27.000000000 +0400
+++ postgresql-8.1.5-patched/configure.in 2007-04-01 18:35:40.000000000
+0400
@@ -266,11 +266,6 @@
PGAC_PROG_CC_CFLAGS_OPT([-Wdeclaration-after-statement])
PGAC_PROG_CC_CFLAGS_OPT([-Wendif-labels])
- else
- # Intel compiler has a bug/misoptimization in checking for
- # division by NAN (NaN == 0), -mp1 fixes it, so add it to the
- # CFLAGS.
- PGAC_PROG_CC_CFLAGS_OPT([-mp1])
fi
# Disable strict-aliasing rules; needed for gcc 3.3+
---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster
