pow__ii

Prof Brian Ripley Prof Brian Ripley <ripley@stats.ox.ac.uk>
Wed, 20 Jan 1999 18:03:15 +0000 (GMT)


> Date: Wed, 20 Jan 1999 12:04:04 -0500
> From: Paul Gilbert <pgilbert@bank-banque-canada.ca>
> To: Bodo Ahrens <Bodo.Ahrens@univie.ac.at>, R-devel 
<R-devel@stat.math.ethz.ch>
> Subject: pow__ii
> 
> > R code works but the shared objects built from fortran code do not find
> > the appropriate libs with functions like  __pow_ii or __epx at runtime.
> 
> I used to have endless trouble with  __pow_ii which as I recall could be be
> fixed by the correct incantation of the mix of compiler libraries and
> LD_LIBRARY_PATH. It seemed to be extremely sensitive and never easy to get
> right. The problem is that pow_ii is not used by R itself, and so when your

Wait a minute. With SC4.2 installed on Solaris, __pow_ii is in
/opt/SUNWspro/lib/libF77.so.3, and if you followed the instructions
installing the compilers /opt/SUNWspro/lib/ will be in the library
search path, ahead of the system libraries.  And as R does call libF77,
__pow_ii is in a library it has already dlopen-ed, so it should always
be accessible.  Try ldd R.binary:

        libSM.so.6 =>    /usr/openwin/lib/libSM.so.6
        libICE.so.6 =>   /usr/openwin/lib/libICE.so.6
        libX11.so.4 =>   /usr/openwin/lib/libX11.so.4
        libsocket.so.1 =>        /usr/lib/libsocket.so.1
        libnsl.so.1 =>   /usr/lib/libnsl.so.1
        libz.so =>       /usr/local/lib/libz.so
        libdl.so.1 =>    /usr/lib/libdl.so.1
        libm.so.1 =>     /opt/SUNWspro/lib/libm.so.1
        libF77.so.3 =>   /opt/SUNWspro/lib/libF77.so.3
        libsunmath.so.1 =>       /opt/SUNWspro/lib/libsunmath.so.1
        libc.so.1 =>     /usr/lib/libc.so.1
        libXext.so.0 =>  /usr/openwin/lib/libXext.so.0
        libmp.so.2 =>    /usr/lib/libmp.so.2

(Solaris 2.6 aka SunOS 5.6).

I have just rebuilt R using the SC4.2 compilers and sent Bodo Ahrens
lots of detailed check information.  It works for me on several different
Solaris/compiler combinations.

> fortran code does something like 4**2 (or using variables declared as integer)
> then a call to pow_ii is generated and you need to find the correct library. I

Not on my system: you need n1**n2.

> eventually fixed the problem by changing the fortran code to 4.0D0**2 (or
> 4.0D0**2.0D0 ? and declared variables as double). This generates  calls to a
> routine which is used by R itself, and I never had to worry about the 
libraries
> again.
> 
> This also worked better with Splus, which can also be problematic in this
> regard. I highly recommend this change to anyone distributing fortran code 
with
> a package.

Current Solaris versions of S-PLUS even check up LD_LIBRARY_PATH for you.
I have not seen problems with this since about 1995, and have just checked
this out:

toucan% cat test.f
      subroutine mypow(n1, n2, n3)
      n2 = n1**n3
      end
toucan% Splus COMPILE test.f
toucan% nm -g test.o
[Index]   Value      Size    Type  Bind  Other Shndx   Name

[8]     |         0|       0|NOTY |GLOB |0    |UNDEF  |__pow_ii
[7]     |         0|      28|FUNC |GLOB |0    |2      |mypow_

and this dyn.loads into S-PLUS 3.4 without error.

toucan% R SHLIB test.f
ld -G -o test.so test.o -L/usr/local/lib -lblas 

[It also works on a machine without libblas installed.]
toucan% nm -g test.so

test.so:

[Index]   Value      Size    Type  Bind  Other Shndx   Name

[21]    |     66252|       0|OBJT |GLOB |0    |6      |_DYNAMIC
[23]    |         0|       0|OBJT |GLOB |0    |ABS    |_PROCEDURE_LINKAGE_TABLE_
[20]    |         0|       0|NOTY |GLOB |0    |UNDEF  |__pow_ii
[22]    |     66332|       0|OBJT |GLOB |0    |7      |_edata
[18]    |     66332|       0|OBJT |GLOB |0    |8      |_end
[24]    |       716|       0|OBJT |GLOB |0    |5      |_etext
[19]    |       688|      28|FUNC |GLOB |0    |5      |mypow_

> dyn.load("./test.so")
> .Fortran("mypow", as.integer(4), integer(1), as.integer(3))
[[1]]
[1] 4

[[2]]
[1] 64

[[3]]
[1] 3

I suspect Paul like me remembers when this was a problem on SunOS4 with static
libraries, but I don't think it is a problem for S-PLUS.

Brian

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._