WPS compilation error: PGC-F-0249 (byteswap.h)

Accelerated Computing HPC Compilers Legacy PGI Compilers
#1

I tried compiling the WPS package with WRF, and I get the following error:

pgcc -D_UNDERSCORE -DBYTESWAP -DLINUX -DIO_NETCDF -DIO_BINARY -DIO_GRIB1 -DBIT32 -D_GEOGRID -O -c cio.c

PGC-F-0249-#error – “Never use <bits/byteswap.h> directly; include <byteswap.h> instead.” (/opt/pgi/linux86/8.0-1/include/bits/byteswap.h: 21)
PGC/x86 Linux 8.0-1: compilation aborted
make[1]: [cio.o] Error 2 (ignored)

This is with WPS 2.2.1, and PGI Workstation 8.0-1 32-bit, on Fedora 10.


I took a look at the cio.c source file in the WPS/geogrid/src/ folder, and it has only the following includes:

#include <stdlib.h>
#include <stdio.h>
#include <string.h>

Did I miss yet more fine print in the installation guide? Or is this something wrong on the compiler end?

#2

Hi Daniel_S,

I don’t have a system with Fedora 10 installed nor am I able to reproduce this error on any other system (the source compiles fine for me). So unfortunately, I don’t know why you are seeing this error.

Can you please compile this file with “-P” and then look at the resulting “cio.i” file to determine which file is including “bits/byteswap.h”?

I’ll try and get our IT department to install Fedora 10 on a system, but it most likely wont be until after the holidays. Note that since Fedora 10 was released after our 8.0 release, it is not a currently supported OS.

  • Mat
#3

I tried a smaller test file with only:

// test.c
#include <stdlib.h>
//#include <stdio.h>
//#include <string.h>

and was able to reproduce the same error.

Using -P stripped out a little too much to easily tell what was coming from where, but I liked the idea, and noticed the -dI option. Here’s some abridged output using “pgcc -dI test.c”:

// test.c:1
#include /opt/pgi/linux86/8.0-1/include/stdlib.h

// /opt/pgi/linux86/8.0-1/include/stdlib.h:9
#include_next /usr/include/stdlib.h

// /usr/include/stdlib.h:320
#include /opt/pgi/linux86/8.0-1/include/sys/types.h

// /opt/pgi/linux86/8.0-1/include/sys/types.h:215
#include /usr/include/endian.h

// /usr/include/endian.h:61
#include /opt/pgi/linux86/8.0-1/include/bits/byteswap.h
PGC-F-0249-#error – “Never use <bits/byteswap.h> directly; include <byteswap.h> instead.” (/opt/pgi/linux86/8.0-1/include/bits/byteswap.h: 21)
PGC/x86 Linux 8.0-1: compilation aborted

I also noticed a

#ifdef __USE_BSD

line before a couple of blocks in that include chain.


I also got similar errors when recompiling NetCDF 4 using the instructions for an older version of the compiler (and an older version of NetCDF). Mind you, I encountered a few other errors first. On the first go, the configure script failed with:

checking size of short… 0
checking size of int… 0
checking size of long… 0
checking size of float… 0
checking size of double… 0
checking size of off_t… configure: error: cannot compute sizeof (off_t)
See `config.log’ for more details.

Removing the CC environment variable (and removing or adjusting CFLAGS too), and the configure script then runs fine, but then when I run make I get the same error above about byteswap.h from pgCC.

without any environment variables set, it builds fine, but it seemed to be using the system cc compiler, and pgf95.

#4

Hi Daniel_S,

We got Fedora 10 installed here and it appears they have changed a few header files which will require us to change our “bits/byteswap.h” file.

In particular, they have updated the system “endian.h” file to directly include “bits/byteswap.h”. This means in our version of "bits/byteswap.h, “endian.h” needs to be added to the list of header files that are allowed to call it directly. So by changing line 20 of “/opt/pgi/linux86/8.0-1/include/bits/byteswap.h” from:

#if !defined _BYTESWAP_H && !defined _NETINET_IN_H
# error "Never use <bits/byteswap.h> directly; include <byteswap.h> instead."
#endif

to

#if !defined _BYTESWAP_H && !defined _NETINET_IN_H && !defined  _ENDIAN_H
# error "Never use <bits/byteswap.h> directly; include <byteswap.h> instead."
#endif

you can work around the problem.

I’ve added a technical problem report (TPR#15472) and send it on to engineering to have the problem corrected in a future release.

  • Mat
#5

Ahh, thank you. This solved my problem quite nicely.

#6

Hi Daniel,

We’ve updated the 8.0-4 PGI header files to include this change.

  • Mat