> Actually, gethrtime is in section 3C:
>
> "(3C) These functions, together with those of Section 2,
> constitute the standard C library"
>
> And indeed, if we do some further digging:
>
> http://docs.hp.com/en/B2355-60130/gethrtime.3C.html (HP-UX 11iv3)
>
> So, this says more about Linux than about your coding.
Well, one should be able to just compile something simple and expect it to
work but often times the -Xc switch can stop you in your tracks.
Consider this itty bitty thing :
#include <stdio.h>
#include <sys/time.h>
hrtime_t start_hrt, end_hrt;
int
main(int argc, char *argv[])
{
start_hrt = gethrtime();
printf ( "Hello World!\n" );
end_hrt = gethrtime();
printf("time = %lld nsec\n", ( end_hrt - start_hrt ) );
return (0);
}
Pretty simple and you can compile it with Studio 8 ( or whatever ) thus :
first .. the old old compiler :
$ cc -V
cc: Sun C 5.5 Patch 112760-18 2005/06/14
usage: cc [ options] files. Use 'cc -flags' for details
the compiler switches :
$ cc -xstrconst -xildoff -xarch=generic -H -xtime -Xa -c -o hello.o hello.c
/usr/include/stdio.h
/usr/include/iso/stdio_iso.h
/usr/include/sys/feature_tests.h
/usr/include/sys/isa_defs.h
/usr/include/sys/va_list.h
/usr/include/stdio_tag.h
/usr/include/stdio_impl.h
/usr/include/sys/time.h
/usr/include/sys/types.h
/usr/include/sys/machtypes.h
/usr/include/sys/int_types.h
/usr/include/sys/select.h
/usr/include/sys/time.h
/usr/include/time.h
/usr/include/iso/time_iso.h
/usr/include/sys/time_impl.h
acomp real 0.8 user 0.3 sys 0.2
I like to see all the header files being traversed.
For something so simple, amazing isn't it?
$ cc -xarch=generic -o hello hello.o
$ file hello
hello: ELF 32-bit MSB executable SPARC Version 1, dynamically
linked, not stripped
$ ./hello
Hello World!
time = 3765000 nsec
$ /bin/ptime ./hello
Hello World!
time = 4000000 nsec
real 0.055
user 0.010
sys 0.051
$
So that works as expected until you go to -Xc for strict compliance with
some pedantic standards.
$ cc -xstrconst -xildoff -xarch=generic -H -xtime -Xc -\# -c -o hello.o hello.c
### Note: NLSPATH =
/opt/SUNWspro/prod/bin/../lib/locale/%L/LC_MESSAGES/%N.cat:/opt/SUNWspro/prod/bin/../../lib/locale/%L/LC_MESSAGES/%N.cat
### command line files and options (expanded):
### -c -H -Xc -xarch=generic -xildoff -xstrconst -xtime hello.c -o hello.o
/opt/SUNWspro/prod/bin/acomp -xldscope=global -i hello.c -y-fbe
-y/opt/SUNWspro/prod/bin/fbe -y-xarch=generic -y-o -yhello.o -y-verbose
-y-xmemalign=4s -y-xthreadvar=no%dynamic -y-comdat -xdbggen=no%dwarf2+stabs
-strconst -H -fsimple=0 -xarch=generic -fparam_ir -Qy -D__SunOS_5_8
-D__SUNPRO_C=0x550 -D__SVR4 -D__unix -D__sun -D__sparc
-D__BUILTIN_VA_ARG_INCR -xc99=%all,no%lib -Xc -D__PRAGMA_REDEFINE_EXTNAME
-I/opt/SUNWspro/prod/include/cc "-g/opt/SUNWspro/prod/bin/cc -xstrconst
-xildoff -xarch=generic -H -xtime -Xc -c -o hello.o " -D__SUN_PREFETCH
-destination_ir=yabe
/usr/include/stdio.h
/usr/include/iso/stdio_iso.h
/usr/include/sys/feature_tests.h
/usr/include/sys/isa_defs.h
/usr/include/sys/va_list.h
/usr/include/stdio_tag.h
/usr/include/stdio_impl.h
/usr/include/sys/time.h
/usr/include/sys/types.h
/usr/include/sys/machtypes.h
/usr/include/sys/int_types.h
/usr/include/sys/select.h
/usr/include/sys/time.h
/usr/include/time.h
/usr/include/iso/time_iso.h
"hello.c", line 12: operands have incompatible types:
union {double _d, array[2] of int _l} "-" union {double _d,
array[2] of int _l}
cc: acomp failed for hello.c
Note all the things defined with -D and then an underscore.
I think they are all covered off in the Sun Studio 8: C User's Guide but in
most cases I would use Studio 11. Today I am using Studio 8 for reasons to
be revealed eventually.
But I digress. Let's look at what went wrong there. It looks to me that we
have some funky datatype which is really a union. That would be this thing
:
/*
* Time expressed as a 64-bit nanosecond counter.
*/
typedef longlong_t hrtime_t;
I found that by tracking down the underlying definition for hrtime_t which
counts our nanoseconds ( supposedly ) :
$ grep hrtime_t /usr/include/sys/time.h
typedef longlong_t hrtime_t;
extern hrtime_t gethrtime(void);
extern hrtime_t gethrtime_unscaled(void);
extern hrtime_t gethrtime_max(void);
extern void scalehrtime(hrtime_t *);
extern void hrt2ts(hrtime_t, timestruc_t *);
extern hrtime_t ts2hrt(timestruc_t *);
extern void hrt2tv(hrtime_t, struct timeval *);
extern hrtime_t tv2hrt(struct timeval *);
extern void hrt2ts32(hrtime_t, timestruc32_t *);
hrtime_t gethrtime(void);
hrtime_t gethrvtime(void);
hrtime_t gethrtime();
hrtime_t gethrvtime();
A bit of poking around finds this in /usr/include/sys/time.h
/*
* Needed for longlong_t type. Placement of this due to <sys/types.h>
* including <sys/select.h> which relies on the presense of the itimerval
* structure.
*/
and eventually this :
typedef long long longlong_t;
is in /usr/include/sys/types.h so then why don't we just use that and be
done with it ? Why the union ? Does not matter much because the data being
returned by gethrtime(3C) is not a long long and does need to be some funky
union :
"hello.c", line 11: assignment type mismatch:
long long "=" union {double _d, array[2] of int _l}
"hello.c", line 13: assignment type mismatch:
long long "=" union {double _d, array[2] of int _l}
cc: acomp failed for hello.c
attempts to get around that ... fail.
Is this portable and part of some standard C library?
I don't see how really. But it compiles nicely on Solaris with -Xa and
maybe even a transitional switch but forget pedantic or strict compliance
with some ANSI standard unless you are willing to dig around a LOT to solve
the data type problem.
I'd love to hear suggestions *because* I do have code that I'd like to see
run on both Solaris and Linux which reports fine grained timing information
with some degree of accuracy.
Dennis
_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org
