The value passed to the function overflows a char;
add -Msignextend as a work-around.
% cat rtest.c
#include
main () {
int i;
int val_diff (char,int);
i = val_diff (-96, 3232);
printf ("main: val_diff returns = %d; ", i);
if (i == 0) {
printf ("values are the same\n");
} else {
printf ("values are different\n");
}
}
int val_diff(char v, int val)
{
char c2;
int iret;
int foo (char);
#if defined(OK_WAY)
/*
* In order to get the cast of val to char to really take hold,
* we have to do something with it besides just compare.
* When compiling with -g -O0, just the assignment to c2 is enough;
* when compiling with default opt, we must also invoke foo().
*/
c2 = (char) val;
foo (c2);
printf ("val_diff: vals are %d and %d\n", v, c2);
iret = (v != c2);
return iret;
#elif defined(BAD_WAY)
printf ("val_diff: vals are %d and %d\n", v, (char) val);
return (v != (char) val);
#endif
}
int foo (char c) { }
A simple example fails to trigger an fp exception at -O2.
% cat y.c
main()
{
int y=5; int x=0;
printf("y/x = %dn",y/x);
}
% pgcc -o y y.c -V
pgcc 6.0-5a 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2005, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86 Linux/x86 6.0-5a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2005, STMicroelectronics, Inc. All Rights Reserved.
% ./y
Floating exception
% pgcc -O2 -o y y.c -V
pgcc 6.0-5a 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2005, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86 Linux/x86 6.0-5a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2005, STMicroelectronics, Inc. All Rights Reserved.
% ./y
y/x = 0
The following test case shows that compiling with '-O2' will
get different results than compiling with -O0, -O1, or -O3.
$ cat bits.c
#include
main() {
char val[2];
int i = 0;
val[0]='1';
val[1]='';
while(val[0]) {
val[0] <<= 1;
i++;
/* printf("After %d shifts, val is %u (0x%x)n", i,
(unsigned int) val[0], (unsigned int) val[0]); */
}
printf("Bits per byte is %dn", i);
}
$ pgcc -V
pgcc 6.0-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2005, STMicroelectronics, Inc. All Rights Reserved.
$ pgcc -o bits -O2 bits.c && ./bits
Bits per byte is 9 <-- Wrong!
$ pgcc -o bits -O0 bits.c && ./bits
Bits per byte is 8
$ pgcc -o bits -O1 bits.c && ./bits
Bits per byte is 8
$ pgcc -o bits -O3 bits.c && ./bits
Bits per byte is 8
A failing case at some optimization levels now works,
because -Msignextend is now default in the C compilers,
matching the behavior of gcc.
The functions erf and derf return the error function of x. erfc and derfc return 1.0-erf(x) and 1.0-
derf(x), respectively. They have been added for the Windows versions of the compilers.
Synopsis
real function erf(x)
real x
real function erfc(x)
real x
double precision function derf(x)
double precision x
double precision function derfc(x)
double precision x
RHEL 4.0 now supports NUMA on Opterons, and we need to integrate that
into our installation script.
Users program, when compiled -tp piii -O2 -Mlist -Mbyteswapio, causes compiler
to crash.
% pgf90 -O2 -Mlist -Mbyteswapio -tp k8-32 -c vacuum.f
% pgf90 -O2 -Mlist -Mbyteswapio -tp piii -c vacuum.f
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 33 (vacuum.f: 4774)
pgf90-Fatal-/usr/pgi/linux86/6.1/bin/p3/pgf902 TERMINATED by signal 11
Arguments to /usr/pgi/linux86/6.1/bin/p3/pgf902
/usr/pgi/linux86/6.1/bin/p3/pgf902 /tmp/pgf90aaaaallbar.ilm -fn vacuum.f
-opt 2 -terse 1 -inform warn -x 51 0x20 -x 119 0xa10000 -x 119 0x100000
-x 122 0x40 -x 123 0x1000 -x 127 4 -x 127 16 -x 119 0x40000000
-x 19 0x400000 -x 28 0x40000 -x 59 4 -x 124 0x1400 -y 15 2 -x 57 0x3b0000
-x 58 0x48000000 -astype 0 -x 124 1 -x 80 0x300 -y 80 0x1000
-cmdline '+pgf90 vacuum.f -O2 -Mlist -Mbyteswapio -tp piii -c'
-asm /tmp/pgf90baaaallbat.s
% pgf90 -O2 -Mlist -Mbyteswapio -tp p6 -c vacuum.f
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 33 (vacuum.f: 4774)
pgf90-Fatal-/usr/pgi/linux86/6.1/bin/p3/pgf902 TERMINATED by signal 11
Arguments to /usr/pgi/linux86/6.1/bin/p3/pgf902
/usr/pgi/linux86/6.1/bin/p3/pgf902 /tmp/pgf90aaaaaolbah.ilm
-fn vacuum.f -opt 2 -terse 1 -inform warn -x 51 0x20 -x 119 0xa10000
-x 119 0x100000 -x 122 0x40 -x 123 0x1000 -x 127 4 -x 127 16 -x 119 0x40000000
-x 19 0x400000 -x 28 0x40000 -x 59 4 -x 124 0x1400 -y 15 2 -x 57 0x3b0000
-x 58 0x48000000 -astype 0 -x 124 1 -x 80 0x300 -y 80 0x1000
-cmdline '+pgf90 vacuum.f -O2 -Mlist -Mbyteswapio -tp p6 -c'
-asm /tmp/pgf90baaaaolbai.s
User noted that a line with an implied DO loop was very slow to compile
! Compiling with the next line takes 1000 times longer than
integer :: ncount(MAX_SOUND)= (/(0,i=1,MAX_SOUND)/)
! compiling with this line !
integer :: ncount(MAX_SOUND)=0
Dummy Arguments have support as allocatables.
% more ad.f
subroutine ad(a)
real,dimension(:),allocatable :: a
if (allocated(a)) deallocate(a)
allocate(a(100))
end
% pgf90 -V -c ad.f
pgf90 6.2-5 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0198-Illegal use of a in ALLOCATE/DEALLOCATE (ad.f: 4)
PGF90-S-0084-Illegal use of symbol a - must be an allocatable array (ad.f: 5)
0 inform, 0 warnings, 2 severes, 0 fatal for ad
% settmp704
% pgf90 -V -c ad.f
pgf90 7.0-4 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-4
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-4
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
The following code ran much better with manual interchange of the loops.
Now the compiler matches the performance by manual interchange.
program tbnd
parameter (N=1000)
parameter (M=1000)
parameter (kbndwd= 9)
parameter (lbndwd=1+2*kbndwd)
real A(lbndwd,N)
real B(N)
c
do k = 1, 10000
do j=1,N
do i=1,lnbndwd
A(i,j) = 1.0
end do
A(kbndwd+1,j) = 100.0
B(j) = 2.0
end do
call bnd_LU_09(N,M,A,B)
end do
c
print *,"B(1) = ",B(1)
print *,"B(1000) = ",B(1000)
c
end
c
subroutine bnd_LU_09
& ( N , M ,
& A , b )
parameter (kbndwd= 9)
parameter (lbndwd=1+2*kbndwd)
real A(lbndwd,N)
real B(N)
do k=1,N-kbndwd
oAkk=1./A(kbndwd+1,k)
do i=1,kbndwd,1
A(i,k+kbndwd-i+1)=A(i,k+kbndwd-i+1)*oAkk
enddo
do j=1,kbndwd
do i=1,kbndwd
A(i+j,k+kbndwd-i+1)=A(i+j,k+kbndwd-i+1)
& -A(i,k+kbndwd-i+1)*A(kbndwd+1+j,k)
enddo
enddo
enddo
do k=N-kbndwd+1,N-1
oAkk=1./A(kbndwd+1,k)
do i=k+1,min(k+kbndwd,n)
A(k+kbndwd-i+1,i)=A(k+kbndwd-i+1,i)*oAkk
do j=1,kbndwd
A(k+kbndwd-i+1+j,i)=A(k+kbndwd-i+1+j,i)
& -A(k+kbndwd-i+1,i)*A(kbndwd+1+j,k)
enddo
enddo
enddo
do j=1,N-kbndwd
bj=B(j)
do mi=1,kbndwd
B(j+mi)=B(j+mi)-A(kbndwd-mi+1,j+mi)*bj
enddo
enddo
do j=N-kbndwd+1,N-1
do i=j+1,n
B(i)=B(i)-A(j+kbndwd-i+1,i)*B(j)
enddo
enddo
do j=N,kbndwd+1,-1
B(j)=B(j)/A(kbndwd+1,j)
bj=B(j)
do i=1,kbndwd
B(j-kbndwd+i-1)=B(j-kbndwd+i-1)
& -A(lbndwd-i+1,j-kbndwd+i-1)*Bj
enddo
enddo
do j=kbndwd,2,-1
B(j)=B(j)/A(kbndwd+1,j)
do i=1,j-1,1
B(i)=B(i)-A(j-i+kbndwd+1,i)*B(j)
enddo
enddo
B(1)=B(1)/A(kbndwd+1,1)
return
end
Users example caused ICE.
% pgf90 -fastsse opt.f -tp p6 -c
PGF90-S-0000-Internal compiler error. fr_free: rg already free or wrong owner 1 (opt.f: 4872)
PGF90-S-0000-Internal compiler error. stack87pos: reg not found 1 (opt.f: 4907)
..
..
and so on..
0 inform, 0 warnings, 22 severes, 0 fatal for gqtst
Users with parallel processes that generated scratch files found that the
naming scheme we used sometimes resulted in two processes having the same
scratch file name. The scheme has been modified to produce unique filenames
even in these instances.
We now support -dM to display predefined Macros that have been encountered
during preprocessing.
for example,
$ pgcc -dM -dI 1619.c
#define unix 1
#define __unix 1
#define __unix__ 1
#define linux 1
#define __linux 1
#define __linux__ 1
#define __NO_MATH_INLINES 1
#define __x86_64__ 1
#define __LONG_MAX__ 9223372036854775807L
#define __SIZE_TYPE__ unsigned long int
#define __PTRDIFF_TYPE__ long int
#define __amd64__ 1
#define __SSE__ 1
#define __MMX__ 1
#define __SSE2__ 1
#define __PGI 1
#define __LINE__ 0
#define __FILE__ 1
#define __DATE__ "Nov 8 2006"
#define __TIME__ "14:31:35"
#define __STDC__ 1
#define __STDC_VERSION__ 199901L
#define __PGIC__ 99
#define __PGIC_MINOR__ 99
#define __PGIC_PATCHLEVEL__ 99
The C99 standard defines a Macro __STDC_VERSION__. This has been implemented.
C9X.pdf Section 6.10.8 page 158.
"__STDC_VERSION__" is a decimal constant "199901L".
We have now implemented a switch that provides the predefines the compiler uses.
pgcc -dM or pgcc -dD or pgcc -dN or pgcc -dl
% pgcc -dM -V hello.c
pgcc 7.0-1 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86 Linux/x86 7.0-1
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
#define unix 1
#define __unix 1
#define __unix__ 1
#define linux 1
#define __linux 1
#define __linux__ 1
#define i386 1
#define __i386 1
#define __i386__ 1
#define __NO_MATH_INLINES 1
#define linux86 1
#define __SSE__ 1
#define __MMX__ 1
#define __SSE2__ 1
#define __SSE3__ 1
#define __PGI 1
#define __LINE__ 0
#define __FILE__ 1
#define __DATE__ "Dec 15 2006"
#define __TIME__ "16:18:29"
#define __STDC__ 1
#define __STDC_VERSION__ 199901L
#define __PGIC__ 7
#define __PGIC_MINOR__ 0
#define __PGIC_PATCHLEVEL__ 1
// /usr/include/stdio.h:27
#define _STDIO_H 1
// /usr/include/features.h:20
#define _FEATURES_H 1
// /usr/include/features.h:90
#undef __USE_ISOC99
#undef __USE_POSIX
#undef __USE_POSIX2
#undef __USE_POSIX199309
#undef __USE_POSIX199506
#undef __USE_XOPEN
#undef __USE_XOPEN_EXTENDED
#undef __USE_UNIX98
#undef __USE_XOPEN2K
#undef __USE_LARGEFILE
#undef __USE_LARGEFILE64
#undef __USE_FILE_OFFSET64
#undef __USE_BSD
#undef __USE_SVID
#undef __USE_MISC
#undef __USE_GNU
#undef __USE_REENTRANT
#undef __USE_FORTIFY_LEVEL
#undef __FAVOR_BSD
#undef __KERNEL_STRICT_NAMES
// /usr/include/features.h:118
#define __USE_ANSI 1
// /usr/include/features.h:131
#define __GNUC_PREREQ(maj,min) 0
// /usr/include/features.h:169
#define _BSD_SOURCE 1
#define _SVID_SOURCE 1
// /usr/include/features.h:179
#define __USE_ISOC99 1
// /usr/include/features.h:186
#define _POSIX_SOURCE 1
// /usr/include/features.h:190
#define _POSIX_C_SOURCE 199506L
// /usr/include/features.h:195
#define __USE_POSIX 1
// /usr/include/features.h:199
#define __USE_POSIX2 1
// /usr/include/features.h:203
#define __USE_POSIX199309 1
// /usr/include/features.h:207
#define __USE_POSIX199506 1
// /usr/include/features.h:246
#define __USE_MISC 1
// /usr/include/features.h:250
#define __USE_BSD 1
// /usr/include/features.h:254
#define __USE_SVID 1
// /usr/include/features.h:274
#define __STDC_IEC_559__ 1
#define __STDC_IEC_559_COMPLEX__ 1
// /usr/include/features.h:278
#define __STDC_ISO_10646__ 200009L
// /usr/include/features.h:286
#undef __GNU_LIBRARY__
#define __GNU_LIBRARY__ 6
// /usr/include/features.h:291
#define __GLIBC__ 2
#define __GLIBC_MINOR__ 3
...
..
.............and so on.
C++ code fails to print 'long double' data type using cout
% cat x.C
#include
int main()
{
using namespace std;
double biggy = 1.0e307;
cout << biggy << endl;
return 0;
}
% pgCC x.C
% ./a.out
Code:
1e+307
When I change
Code:
double
to
Code:
long double
then the output is
Code:
0
(or sometimes a "-" or nothing at all)
g++ seems to do it correct both times.
It was determined that the behavior of the exception switches -Ktrap=fp, -Kfptrap, -Knoieee and -Knofptrap
did not match the documentation. Also, the documentation should indicate that the main program only can be compiled with these switches, or the
registers being loaded will not be.
The following user's code fails because the prototype for the C99 functions fmax and fmin are not used since they guarded in the linux bits/mathcalls.h by GNU define flags.
Now the same result with -DNEED_C99_PROTO as without it.
cat test.c
#include
#include
#include
#ifdef NEED_C99_PROTO
double fmax(double, double);
double fmin(double, double);
#endif
int main(void) {
double k =0.0e0, l=100.0e0;
double mx = fmax(k,l);
double mn = fmin(k,l);
printf("%15.6E\n",fmin(k,l));
printf("%15.6f\n",mn);
printf("%15.6f\n",mx);
printf("%15.6E\n",fmax(k,l));
return 0;
}
User encountered an "out of bounds" error when running after compiling with "-C". But allocating an array with size 0 should not trigger such an error message. In addition, this seems to happen only in combination with a character array.
program alloc
implicit none
integer :: n
character(len=10), dimension(:),pointer :: val1
call machet(val1)
write(*,*) 'val1: size,len',size(val1),len(val1(1))
contains
subroutine machet(value)
integer :: status
integer :: n
character(len=*), dimension(:), pointer :: value
integer, dimension(:), pointer :: ival
n = 0
allocate(ival(n), stat=status)
if (status >0) then
write(*,*) ' error 1'
end if
allocate(value(n), stat=status)
if (status >0) then
write(*,*) ' error 2'
end if
end subroutine machet
end program alloc
Compiled with pgf90 -C alloc.f90 and run gives:
0: Subscript out of range for array value (alloc.f90: 23)
subscript=1, lower bound=1, upper bound=0, dimension=1
The thread limit on an OpenMP library, even when compiled with a SERVER license
will always take on the limit that the main() routine aquires when it
is compiled by the library user. The limit used to be 4 sockets; it is now
8 sockets.
Our install script, makelocalrc needs to be French-aware, along with other languages.
The front-end does not attempt to collapse an implied I/O do item when
-Mlarge_arrays/-mcmodel=medium is used. This causes a serious performance drop.
1) with mcmodel=small
> pgf77 test.f
> time a.out
check start!
check completed!
FORTRAN STOP
real 0m0.008s <----
user 0m0.004s
sys 0m0.000s
(2) with mcmodel=medium
> pgf77 -mcmodel=medium test.f
> time a.out
check start!
check completed!
FORTRAN STOP
real 0m0.150s <----
user 0m0.144s <---- too slow.
sys 0m0.008s
-Mchkptr causes an error during execution, probably because associated(x)
is treated like a dereference of x.
% cat test.f90
program nullptr
implicit none
integer, dimension(:), pointer :: itar, iptr
nullify (itar)
iptr => itar
if (associated(itar)) then
print *, "itar ASSOCIATED"
else
print *, "itar UNASSOCIATED"
end if
if (associated(iptr)) then
print *, "iptr ASSOCIATED"
else
print *, "iptr UNASSOCIATED"
end if
end program nullptr
% pgf90 -Mchkptr test.f90
% ./a.out
0: Null pointer for itar (test.f90: 5)
Documentation of the Routine setvbuf3f
Fortran I/O uses 3 types of buffering, unbuffered, line buffered, and fully buffered. Units 5 (stdin) and 6 (stdout) are line buffered. Unit 0 (stderr) is unbuffered. Disk files are fully buffered. These are generally the correct choices. You can use setvbuf3f to change a unit's buffering type and size of the buffer.
This function must be called after the unit is opened and before any I/O is done on the unit.
The typ parameter can have the following values, 0 specifies full buffering, 1 specifies line buffering, and 2 specifies unbuffered. The size parameter specifies the size of the buffer. Note: the underlying stdio implementation may silently restrict your choice of buffer size.
This function will return zero on success and non-zero on failure.
An example of where this function might be useful would be a long-running program that writes a small amount of data to a log file. If the log file is line buffered, you could check the log file for progress. If the log file were fully buffered, the data may not be written to disk until the program terminates.
INTERFACE
INTEGER FUNCTION setvbuf3f(lu,typ,size)
INTEGER lu,typ,size
END FUNCTION
END INTERFACE
A Fortran example missed an opportunity to hoist costly dsin from loop,
if ( .NOT.(ii.eq.jj-8) ) then
sum = sum + dsin(2.d0*pig)
else
sum = sum - dsin(5.d0*pig)
endif
A number of OpenMP routines are linked into programs by default, whether or
not the code is compiled -mp. This has been corrected.
An implied do loop causes the compiler to falsely report a severe error.
program test2
implicit none
integer :: i
integer, parameter :: nn=12
real, dimension(nn), parameter :: Alt1 = (/(3*i, i=1,nn)/)
real, dimension(nn), parameter :: Alt2 = (/(3.0*i, i=1,nn)/)
print*, "Hello World!", Alt1(11), Alt2(11)
end program test2
% pgf90 -o test test.f90 -V
pgf90 6.2-5 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
PGF90-S-0069-Illegal implied DO expression (test.f90: 8)
0 inform, 0 warnings, 12 severes, 0 fatal for test2
% pgf90 -o test test.f90 -V
pgf90 7.0-2a 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-2a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-2a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
Loop bounds that passed in a structure , and stored locally, aren't countable.
#include
typedef struct {
int is; // Start of i-loop
int ie; // End of i-loop
int isdot;
int iedot;
int js;
int je;
int jsdot;
int jedot;
int jstride; // j-stride size
} LoopLimits;
int foo1(double * v, LoopLimits l);
int foo2(double * v, LoopLimits l);
int main() {
LoopLimits a;
double v[500];
int i,j;
a.is = 0;
a.ie = 100;
a.jstride = 100;
for (i = 0; i < 500; i++) {
v[i] = 1.0;
}
j = foo1(v,a);
printf("foo = %dn",j);
j = foo2(v,a);
printf("foo = %dn",j);
}
int foo1(double * v, LoopLimits l) {
LoopLimits b;
double * restrict w = v + l.jstride;
b = l;
printf("Bounds are %d %dn",b.is,b.ie);
for (int i = b.is; i < b.ie; i++) {
v[i] = v[i] + w[i];
}
if (w[0] != v[0]) return 1;
return 0;
}
int foo2(double * v, LoopLimits l) {
LoopLimits b;
double * restrict w = v + l.jstride;
int i,is,ie;
b = l;
is = b.is;
ie = b.ie;
printf("Bounds are %d %dn",is,ie);
for (i = is; i < ie; i++) {
v[i] = v[i] + w[i];
}
if (w[0] != v[0]) return 1;
return 0;
}
> pgcc -fastsse -Minfo=all -Mneginfo=all t.c
foo1:
40, Loop not vectorized: not countable
foo2:
56, Generated 3 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
> ./a.out
Bounds are 0 100
foo = 1
Bounds are 0 100
foo = 1
The routine ICHAR in the initialization is producing wrong results.
$ cat test.f90
integer(8) :: actg(0:3) = (/ichar('a'),ichar('c'),ichar('t'),ichar('g')/)
write(*,'(1x,4i4)') actg
actg(0) = ichar('a')
actg(1) = ichar('c')
actg(2) = ichar('t')
actg(3) = ichar('g')
write(*,'(1x,4i4)') actg
end
% pgf90 -o test test.f90 -V
pgf90 6.2-5 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% ./test
97 97 97 97 <-----------------------wrong
97 99 116 103
% pgf90 -o test test.f90 -V
pgf90 7.0-2a 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-2a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-2a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./test
97 99 116 103
97 99 116 103
The test case shows a memory leak in the new allocatable
components feature of pgi 6.2-2.
In the assignment statement:
duplicate = Vector_(vector_length)
the function reference returns a user-defined type with an allocatable
component, and the variable duplicate is of the same type.
The function Vector_ contains an allocation of the component
ALLOCATE(this%component(num_elements))
which appears in the assembly listing as
call pgf90_alloc
However, in the main program there is no corresponding deallocation of the
current component of duplicate before the function Vector_ is called. As a
result the memory image for the program grows with each loop iteration.
We added large array coding considerations to the user guidea.
User's code failed with an internal compiler error (ICE).
% pgf90 -i4 -tp x64 -fastsse -Mnosecond_underscore -Mlarge_arrays -mp -fpic cadari.for -c
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-S-0000-Internal compiler error. flowgraph: node is zero 175 (cadari.for: 1221)
PGF90-F-0008-Error limit exceeded (cadari.for: 1221)
PGF90/x86 Linux/x86-64 6.2-5: compilation aborted
% pgf90 -i4 -tp x64 -fastsse -Mnosecond_underscore -Mlarge_arrays -mp -fpic cadari.for -c -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
This test case was extracted from the NIMROD application at ORNL.
This has been corrected.
% pgf90 -c modu.f test.f -V
pgf90 6.2-3 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
modu.f:
PGF90/any Linux/x86 6.2-3
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86 6.2-3
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
test.f:
PGF90/any Linux/x86 6.2-3
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
Lowering Error: pointer symbol seam_out$p is a member reference
PGF90-F-0000-Internal compiler error. Errors in Lowering 1 (test.f: 41)
PGF90/any Linux/x86 6.2-3: compilation aborted
pgcc cannot compile/preprocess files ending in .h.
This has been corrected.
User pointed out that we were missing an optimization opportunity.
Code like this
subroutine rtest (n, x, y)
real x(n)
do i = 1, n
y = y + x(i)
end do
end
optimizes better when written as
subroutine rtest(n,x,y)
real x(n),t
t = y
do i=1,n
t = t + x(i)
end do
y = t
end
We now take advantage of the opportunity.
When a user links the ACML library with Visual C++ an undefined reference to __builtin_stinit is created.
This has been corrected.
This information is missng from the Fortran Ref Manual.
- All fortran languages (pgf77, pgf90, pghpf) now support an OPEN
statement specifier which allows for byte-swapping I/O to be performed
on specific logical units. Previously, byte-swapping I/O was only enabled
by the command-line option, -byteswapio, and was applied to all unformatted
I/O operations which appeared in the files compiled -byteswapio.
The new
OPEN specifier is
CONVERT=char expr
where char expr is one of
- 'BIG_ENDIAN',
- 'LITTLE_ENDIAN', or
- 'NATIVE'.
The value 'BIG_ENDIAN' is used to convert big-endian format data files produced
by most RISC workstations and high-end servers to the little-endian format used
on Intel Architecture systems on-the-fly during file reads/writes. This value
assumes that the record layouts of unformatted sequential access and direct access
files are the same on the systems.
For the values 'LITTLE_ENDIAN' and 'NATIVE", byte-swapping is not performed during
file reads/writes since the little-endian format is used on Intel Architecture.
- All fortran (pgf77, pgf90, pghpf) now support a new value, 'BINARY',
to the FORM specifier in the OPEN statement which allows for binary
unformatted I/O to be performed on specific logical units. For an unformatted
file whose form is 'BINARY', the file is just viewed as a byte-stream file,
such as a file created by fwrite() calls in a C program; the data in the
file is not organized into records.
DWARF format debug information in C and C++
was defective. This has been corrected.
User's example causes the compiler to generate Assembler error messages.
pgf90 -c -r8 -fast test.f -tp k8-64
/tmp/pgf90i67hyCA5WwgY.s: Assembler messages:
/tmp/pgf90i67hyCA5WwgY.s:370: Error: suffix or operands invalid for `movsd'
/tmp/pgf90i67hyCA5WwgY.s:390: Error: suffix or operands invalid for `movsd'
/tmp/pgf90i67hyCA5WwgY.s:440: Error: suffix or operands invalid for `movsd'
/tmp/pgf90i67hyCA5WwgY.s:654: Error: suffix or operands invalid for `movsd'
/tmp/pgf90i67hyCA5WwgY.s:674: Error: suffix or operands invalid for `movsd'
/tmp/pgf90i67hyCA5WwgY.s:723: Error: suffix or operands invalid for `movsd'
The following example should report an error when -Mstandard is used.
% cat test.f90
real a(:)
allocate(a(100),stat=i)
print *, i
end
with 6.2-5, no errors
% pgf90 -Mstandard -c test.f90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
with 7.0-2
% pgf90 -Mstandard -c test.f90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0084-Illegal use of symbol a - must be an allocatable array (test.f90: 3)
0 inform, 0 warnings, 1 severes, 0 fatal for MAIN
User's program caused the compiler to be TERMINATED by signal 11
% pgf90 -V
pgf90 6.2-4 32-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
pgf90 -I. -c -o atomsettype.o atomsettype.f
pgf90 -I. -c -o forces.o forces.f
pgf90 -I. -c -o gear.o gear.f
pgf90 -I. -c -o disloc.o disloc.f
pgf90-Fatal-/tmp/pgi624/linux86/6.2/bin/newcg/pgf901 TERMINATED by signal 11
Arguments to /tmp/pgi624/linux86/6.2/bin/newcg/pgf901
/tmp/pgi624/linux86/6.2/bin/newcg/pgf901 disloc.f -opt 1 -terse 1 -inform warn -nohpf
-nostatic -x 119 0x100000 -x 19 0x400000 -x 59 4 -x 15 2 -x 49 0x400004 -x 51 0x20
-x 57 0x4c -x 58 0x10000 -x 124 0x1000 -x 57 0xfb0000 -x 58 0x78031040 -x 48 3328
-stdinc /tmp/pgi624/linux86/6.2/include:/usr/local/include:/usr/lib/gcc/x86_64-redhat-linux/4.1.0/include:
/usr/lib/gcc/x86_64-redhat-linux/4.1.0/32/include:/usr/include -def unix -def __unix
-def __unix__ -def linux -def __linux -def __linux__ -def i386 -def __i386 -def __i386__
-def __NO_MATH_INLINES -def linux86 -def __THROW= -def __extension__= -def __SSE__
-def __MMX__ -def __SSE2__ -idir . -nofreeform -vect 48 -output /tmp/pgf90Bmefr1blk-v1.ilm
make: *** [disloc.o] Error 127
Multiple source files in multiple directories are compiled, each with a different version of a modfile
that is named the same. This confusesd the 6.2 compiler. Now it will act to look first in the
directory where the compiled file is for the modfile needed.
In order to support user defined options, we added PREOPTIONS and POSTOPTIONS
and -nodefaultoptions
append PREOPTIONS=-DTESTPREOPTIONS;
append POSTOPTIONS=-DTESTPOSTOPTIONS;
pgcc -v -O2 -V7.0-1 -c /opt/QA/tests/art/src/scanner.c
/usr/pgi/linux86-64/7.0-1/bin/pgc /opt/QA/tests/art/src/scanner.c -opt 2 -x 119 0xa10000
-x 122 0x40 -x 123 0x1000 -x 127 4 -x 127 16 -x 19 0x400000 -x 28 0x40000 -quad
-x 120 0x80000000 -x 59 4 -x 59 4 -y 80 0x1000 -x 80 0x10800000 -astype 0 -stdinc
/usr/pgi/linux86-64/7.0-1/include:/usr/local/include:/usr/lib64/gcc/x86_64-suse-linux/4.0.2/include:
/usr/lib64/gcc/x86_64-suse-linux/4.0.2//include:/usr/include -def unix -def __unix
-def __unix__ -def linux -def __linux -def __linux__ -def __NO_MATH_INLINES -def __x86_64__
-def __LONG_MAX__=9223372036854775807L -def '__SIZE_TYPE__=unsigned long int'
-def '__PTRDIFF_TYPE__=long int' -def __THROW= -def __extension__= -def __amd64__
-def __SSE__ -def __MMX__ -def __SSE2__ -def __SSE3__ -predicate '#machine(x86_64)
#lint(off) #system(posix) #cpu(x86_64)' -def TESTPREOPTIONS -def TESTPOSTOPTIONS
-cmdline '+pgcc /opt/QA/tests/art/src/scanner.c -DTESTPREOPTIONS -v -O2 -V7.0-1 -c
-DTESTPOSTOPTIONS' -x 123 4 -x 123 0x80000000 -alwaysinline
/usr/pgi/linux86-64/7.0-1/lib/libintrinsics.il 4 -asm /tmp/pgccQOCg_V3qeHpY.s
User thinks the four subroutines below should all vectorize.
subroutine vmax(a,b,c,n,t1,t2)
real(8) a(n,n), b(n,n),c(n,n)
integer(8) n,t1(*),t2(*)
do j=1,n
do i=max(1,t1(j)),min(n,t2(j))
a(i,j)=max(b(i,j),c(i,j))
enddo
enddo
end
subroutine vmax1(a,b,c,n,t1,t2)
real(8) a(n,n), b(n,n),c(n,n)
integer(8) n,t1,t2
do j=1,n
do i=t1,t2
a(i,j)=max(b(i,j),c(i,j))
enddo
enddo
end
subroutine vmax2(a,b,c,n,m)
real(8) a(m,n), b(m,n),c(m,n)
integer(8) n,m
do j=1,n
do i=1,m
a(i,j)=max(b(i,j),c(i,j))
enddo
enddo
end
subroutine vmax3(a,b,c,n,m)
real(8) a(m,n), b(m,n),c(m,n)
integer(8) n,m
do j=1,n
do i=2,m
a(i,j)=max(b(i,j),c(i,j))
enddo
enddo
% pgf90 -c -Minfo -fastsse -Mfprelaxed test.f90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
vmax:
6, Loop unrolled 4 times
vmax1:
18, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
vmax2:
30, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
vmax3:
42, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
with the 7.0 compilers
----------------------
% pgf90 -c -Minfo -fastsse -Mfprelaxed test.f90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
vmax:
6, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
vmax1:
18, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
vmax2:
30, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
vmax3:
42, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
User code seg faults when using a passed procedure in an internal function. The code
compiled with 6.2-5, but failed during execution when encountering a call to a passed in function from an internal subroutine.
The SAVE attribute was not being implemented when -Msave was not
also set. This has been corrected.
DIM argument to minval and maxval is broken
% cat test.f90
program test_minval
implicit none
integer :: i
real(kind=8), dimension(3,4) :: a = reshape( (/ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 /) , (/ 3,4 /) )
real(kind=8) :: minmaxv(3,2,4)
print*,"a="
write(6,'(3f10.2)') (a(i,1),i=1,3)
write(6,'(3f10.2)') (a(i,2),i=1,3)
write(6,'(3f10.2)') (a(i,3),i=1,3)
write(6,'(3f10.2)') (a(i,4),i=1,3)
minmaxv(:,1,1) = minval(a,DIM=2)
minmaxv(:,2,1) = maxval(a,DIM=2)
print*,""
print*,"minval="
write(6,'(3f10.2)') minmaxv(:,1,1)
print*,""
print*,"maxval="
write(6,'(3f10.2)') minmaxv(:,2,1)
print*,""
end program test_minval
% pgf90 -V test.f90
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% ./a.out
a=
1.00 2.00 3.00
4.00 5.00 6.00
7.00 8.00 9.00
10.00 11.00 12.00
minval=
10.00 10.00 10.00 <---------------------------wrong
maxval=
10.00 10.00 10.00 <---------------------------wrong
% pgf90 -V test.f90
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./a.out
a=
1.00 2.00 3.00
4.00 5.00 6.00
7.00 8.00 9.00
10.00 11.00 12.00
minval=
1.00 2.00 3.00
maxval=
10.00 11.00 12.00
Program fails when compiled with -Mchkptr.
% cat test.f90
program test
type data_set
integer , pointer :: ld(:,:) => null()
character(72) , pointer :: cd(:,:) => null()
end type data_set
type(data_set), pointer :: p1
type(data_set), target :: td
character(72), target :: c1(5,6)
p1 => td
! p1%cd => c1
if (associated(p1%ld)) then
write(*,*) 'ld Yes'
else
write(*,*) 'ld No'
end if
if (associated(p1%cd)) then
write(*,*) 'cd Yes'
else
write(*,*) 'cd No'
end if
end
% pgf90 -V -o test test.f90 -Mchkptr -tp k8-64
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% ./test
ld No
0: Null pointer for p1%cd (test.f90: 20) <--------------wrong
% pgf90 -V -o test test.f90 -Mchkptr -tp k8-64
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./test
ld No
cd No <-------------------------------correct
Program gets wrong answers when a section is vectorized. A directive to turn off
the vectorizer causes the answers to differ.
% diff test.f test1.f
102d101
< cpgi$ novector
% pgf90 -fastsse -byteswapio -o test test.f -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -fastsse -byteswapio -o test1 test1.f -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% ./test
l, jd, kd, ld= 2 21 15 15
stuff diag(1,1,9)= 0.000000000000000
stuff h(1,1)= 3.0798156208940841E-020
stuff h(1,2)= 3.0798156208940841E-020
FORTRAN STOP
% ./test1
l, jd, kd, ld= 2 21 15 15
stuff diag(1,1,9)= 0.000000000000000
stuff h(1,1)= -2.584640104061622 <--------------------wrong
stuff h(1,2)= -2.584640104061622 <--------------------wrong
% pgf90 -fastsse -byteswapio -o test test.f -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -fastsse -byteswapio -o test1 test1.f -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./test
l, jd, kd, ld= 2 21 15 15
stuff diag(1,1,9)= 0.000000000000000
stuff h(1,1)= 3.0798156208940841E-020
stuff h(1,2)= 3.0798156208940841E-020
FORTRAN STOP
% ./test1
l, jd, kd, ld= 2 21 15 15
stuff diag(1,1,9)= 0.000000000000000
stuff h(1,1)= 3.0798156208940841E-020
stuff h(1,2)= 3.0798156208940841E-020
FORTRAN STOP
A program produces wrong answers at higher opt levels
% cat test1.f90
module z
real,dimension(:),allocatable :: b
end module z
subroutine subA
use z
implicit none
b=76.1
end subroutine subA
subroutine x(a,n)
use z
implicit none
integer,intent(in) :: n
real,dimension(n) :: a
allocate(b(n))
b(1:n) = a(1:n)
call subA
a(1:n) = b(1:n)
print *, "a=",a(1:n)
print *, "b=",b(1:n)
end subroutine x
program pgi_opt_bug
implicit none
real,dimension(:),allocatable :: array
integer :: array_size
array_size=5
allocate(array(array_size))
array=1.0
call x(array,array_size)
end program pgi_opt_bug
% pgf90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -O0 -o test1_0 test1.f90
% pgf90 -O1 -o test1_1 test1.f90
% pgf90 -fast -o test1_fast test1.f90
% ./test1_0
a= 76.10000 76.10000 76.10000 76.10000
76.10000
b= 76.10000 76.10000 76.10000 76.10000
76.10000
% ./test1_1
a= 76.10000 76.10000 76.10000 76.10000
76.10000
b= 76.10000 76.10000 76.10000 76.10000
76.10000
% ./test1_fast
a= 1.000000 1.000000 1.000000 1.000000
1.000000
b= 76.10000 76.10000 76.10000 76.10000
76.10000
% pgf90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -O0 -o test1_0 test1.f90
% pgf90 -O1 -o test1_1 test1.f90
% pgf90 -fast -o test1_fast test1.f90
% ./test1_0
a= 76.10000 76.10000 76.10000 76.10000
76.10000
b= 76.10000 76.10000 76.10000 76.10000
76.10000
% ./test1_1
a= 76.10000 76.10000 76.10000 76.10000
76.10000
b= 76.10000 76.10000 76.10000 76.10000
76.10000
% ./test1_fast
a= 76.10000 76.10000 76.10000 76.10000
76.10000
b= 76.10000 76.10000 76.10000 76.10000
76.10000
Program compiled -g did not link properly.
Corrected in 7.0.
% cat test.f90
program DoesNothing
use MyMod
end program DoesNothing
% cat mod.f90
module MyMod
contains
subroutine ModuleRoutine
integer, allocatable,save :: iAllocSave(:)
end subroutine ModuleRoutine
end module MyMod
% pgf90 -c mod.f90 -g
% pgf90 -o test test.f90 -g -V mod.o
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
test.f90:
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
mod.o: In function `..Dm_mymod':mod.f90:(.debug_info+0x8a): undefined reference to `.BSS2'
:mod.f90:(.debug_info+0xa2): undefined reference to `.BSS2'
:mod.f90:(.debug_info+0xae): undefined reference to `.BSS2'
:mod.f90:(.debug_info+0xba): undefined reference to `.BSS2
User who declares $TMPDIR as a directory that does not exist, causes the
compiler to hang.
This has been corrected.
The program is clearly defective, but the compiler should not abort,
but rather print useful messages.
% cat test.f90
real cspl,unk
logical derg
derg = 0.0
cspl=1.
unk=2.0
derg = derg + cspl
unk = unk + derg
end
% pgf90 -r8 test.f90
Lowering Error: unknown source type for conversion to double precision [ast=11,asttype=1,datatype=16]
Lowering Error: unknown source type for conversion to double precision [ast=11,asttype=1,datatype=16]
PGF90-F-0000-Internal compiler error. Errors in Lowering 2 (test.f90: 9)
PGF90/any Linux/x86-64 6.2-5: compilation aborted
User example seg faults with pgcc, but not other C compilers.
% cat bugs.c
#include
const int N = 2;
static void func(int i)
{
if(i)
return;
double arr[N];
}
int main()
{
printf("firstn");
func(0);
printf("secondn");
func(1);
printf("OKn");
return 0;
}
% pgcc -o bug2 bug2.c -V
pgcc 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86-64 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% ./bug2
Segmentation fault
% pgcc -o bug2 bug2.c -V
pgcc 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./bug2
firstnsecondnOKn
User gets a an error from legal code when compiled. It has been corrected.
The Installation instructions about where to put lmgrd so that it will
auto-start after reboot indicate directories like
/etc/rc.d/init.d
and on systems like SUSE 10 , /etc/rc.d and /etc/init.d
are on the same level. It looks like linux has been transitoning where these
directories are located, and on some systems, links have been
created to simulate the old structures. But not all of them.
The notes have been updated.
User's code fails with pgf77 -Minline , and Terminates.
This has been corrected.
% pgf77 -c -Minline test_pgi.f -V
pgf77 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGFTN/x86-64 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGFTN/x86-64 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
pgf77-Fatal-/tmp/pgi625/linux86-64/6.2/bin/pgftn TERMINATED by signal 11
....
% pgf77 -c -Minline test_pgi.f -V
pgf77 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGFTN/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGFTN/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
User code executes with a segfault.
% cat len.f90
program testlen
implicit none
character (len=10) :: b = ''
call sub1(b)
write(*,*) 'b is =',b, 'ENDOFSTR'
contains
subroutine sub1 (dset_scalar)
character(len=*), intent(out) :: dset_scalar
character(len=len(dset_scalar)), dimension(:), pointer :: dset
integer :: dim1 = 3
integer :: i
integer :: mem_status = 0
allocate(dset(dim1), stat=mem_status)
if (mem_status /= 0) then
stop 'allocate failed'
end if
do i = 1, dim1
write(*,*) '1 len = ',len(dset(i))
call sub2(dset,dim1)
end do
dset_scalar = dset(1)
end subroutine sub1
subroutine sub2(dset,dim1)
integer, intent(in) :: dim1
character(len=*), dimension(dim1) , intent(inout) :: dset
write(*,*) '2 len(dset) =',len(dset(1)),len(dset(2)),len(dset(3))
call date_and_time(dset(1),dset(2),dset(3))
end subroutine sub2
end program testlen
% pgf90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o len_625 len.f90
% pgf90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o len_702 len.f90
% ./len_625
1 len = 10
2 len(dset) = 0 0 0
1 len = 10
2 len(dset) = 0 0 0
1 len = 10
2 len(dset) = 0 0 0
Segmentation fault
% ./len_702
1 len = 10
2 len(dset) = 10 10 10
1 len = 10
2 len(dset) = 10 10 10
1 len = 10
2 len(dset) = 10 10 10
b is =20070227 ENDOFSTR
Typo in documentation. Was --use_pch, should be --use_dir
User code caused an Internal compiler error (ICE). This has been corrected.
% pgf90 -Mpreprocess -byteswapio -c FxcDenConvecAddF.f -Mvect -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-F-0000-Internal compiler error. wrong loop indices when computing distances 0 (FxcDenConvecAddF.f: 74)
2.0**66 causes a floating exception with -Ktrap=fp.
% more prob.f
program fpebug
integer ix
real x
ix = 66 ! fails for ix >= 64
x = 2.0**ix
write (6,*) x
end
% pgf77 -Ktrap=fp fpebug.f
% ./a.out
Floating point exception (core dumped)
% pgf77 fpebug.f
% ./a.out
7.3786976E+19
User's example causes compiler to abort.
% pgf90 -c -g params.F -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
*** glibc detected *** /tmp/pgi625/linux86-64/6.2/bin/pgf902: corrupted double-linked list: 0x0000000040435d60 ***
2aed7e200000-pgf90-Fatal-/tmp/pgi625/linux86-64/6.2/bin/pgf902 TERMINATED by signal 6
User's program causes ICE in the compiler.
% pgf90 -c p_full.f -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
Lowering Error: logical result for arithmetic operation [ast=170,asttype=4,datatype=16]
Lowering Error: logical result for arithmetic operation [ast=206,asttype=4,datatype=16]
Lowering Error: logical result for arithmetic operation [ast=208,asttype=4,datatype=16]
Lowering Error: logical result for arithmetic operation [ast=336,asttype=4,datatype=16]
PGF90-F-0000-Internal compiler error. Errors in Lowering 4 (p_full.f: 371)
PGF90/any Linux/x86-64 6.2-5: compilation aborted
% pgf90 -c p_full.f -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
Example fails with __attribute__((aligned (16))) in code.
> cat ss.c
#include
struct sigcontext {
long sc_fpstate[128] __attribute__((aligned (16)));
};
> pgCC -c ss.c
pgCC -c ss.c
PGCC-S-0000-Internal compiler error. unexpected edg alignment %d 16 ()
PGCC/x86 Linux/x86-64 6.2-5: compilation completed with severe errors.
WRF V2.2 Fails to compile. The problem has been corrected.
Setting the environment variable PGI_OBJSUFFIX to o
generates object files with a suffix of .o instead of .obj.
In bash, you can type:
PGI$ declare -x PGI_OBJSUFFIX=o
PGI$ pgf90 x.f -o xf
LINK : fatal error LNK1561: entry point must be defined
User requested renaming of certain libraries for their use.
Code seg faults with 6.2-5.
% cat test.f90
program assume
integer, dimension(-2:2) :: iarr
integer :: i
do i=-2,2
iarr(i) = i
enddo
call my_proc(iarr, lbound(iarr))
contains
subroutine my_proc(iarr, ilow)
integer, dimension(:) :: ilow
integer, dimension(ilow(1):) :: iarr
print *, ilow, lbound(iarr), ubound(iarr)
end subroutine my_proc
end program assume
% pgf90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o test_625 test.f90
% pgf90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o test_702 test.f90
% ./test_625
Segmentation fault
% ./test_702
-2 -2 2
An FAQ webpage, gives information about
how to run Windows executables developed with PGI compilers on a Windows system that
does not have the PGI compilers installed.
Example created an ICE when compiled with 6.2-5 or earlier pgf90.
% cat test.f90
Subroutine CalcVxc_SGGA( W1up, W1dn,SmoothExc)
Complex, Intent(INOUT) :: W1up(:), W1dn(:)
Real, INTENT(INOUT) :: SmoothExc
Complex, Pointer :: W2up(:), W2dn(:), Dave1(:),Dave2(:)
W2up = 0
W2dn = 0
W = 0
W2up(i) = CMPLX(dfxdgbgup,0)
W2dn(i) = CMPLX(dfxdgbgdn,0)
Dave1=W2up+W2dn
Dave2=W2up+W2dn
write(log_unit,*) 'mag of vxc**2-G',dot_product((W2up+W2dn),(W2up+W2dn))
! write(log_unit,*) 'mag of vxc**2-G',dot_product(Dave1,Dave2)
Return
End Subroutine
% pgf90 -r8 -c test.f90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0000-Internal compiler error. transform_call:Array Expression can't be here 99 (test.f90: 14)
PGF90-S-0000-Internal compiler error. transform_call:Array Expression can't be here 99 (test.f90: 14)
0 inform, 0 warnings, 2 severes, 0 fatal for calcvxc_sgga
% pgf90 -r8 -c test.f90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
Case demonstrates that OPEN and INQUIRE handle file names with a trailing NULL character differently.
% cat test.f
C Test of Inquire
LOGICAL LOP1, LOP2
CHARACTER*120 File1, File2
File1 = "foo.0"//CHAR(0)
File2 = "foo.0"
OPEN (23,FILE=File1)
INQUIRE(FILE=File1,OPENED=LOP1,NUMBER=NUM1)
write(*,*) "File1 ", File1, "LOP1 ", LOP1, "NUM1 ", NUM1
C Comment out the close below to get a failure on the second open
CLOSE(23)
OPEN (24,FILE=File2)
INQUIRE(FILE=File2,OPENED=LOP2,NUMBER=NUM2)
write(*,*) "File2 ", File2, "LOP2 ", LOP2, "NUM2 ", NUM2
end
% pgf90 -o test_625 test.f -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o test_702 test.f -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./test_625
File1
foo.0
LOP1 F NUM1 -1
File2
foo.0
LOP2 T NUM2 24
% ./test_702
File1
foo.0
LOP1 T NUM1 23
File2
foo.0
LOP2 T NUM2 24
A set of subroutines that use a common block containing a pointer
to a three element real array. The common block is made thread private
using the OMP directive !$OMP THREADPRIVATE. When the mp option is used
and an element in the array is modified all three elements in the
array are updated rather than the specified index.
% cat tp1.f
subroutine sub(iii)
integer, pointer :: alph(:)
common/zzzz/alph
!$omp threadprivate(/zzzz/)
!! common/yyyy/ibcd(11)
!!$omp threadprivate(/yyyy/)
! print *, iii
allocate(alph(11))
alph = 0
alph(iii+1) = iii+1
!! ibcd(iii+1) = iii+1
print *, alph(1),alph(2),alph(3),alph(4)
!! print *, ibcd(1),ibcd(2),ibcd(3),ibcd(4)
end
integer omp_get_thread_num
!$omp parallel
call sub(omp_get_thread_num())
!$omp endparallel
end
% cat tp2.f
subroutine sub(iii)
integer, pointer :: alph(:)
!$omp threadprivate(alph)
!! common/yyyy/ibcd(11)
!!$omp threadprivate(/yyyy/)
! print *, iii
allocate(alph(11))
alph = 0
alph(iii+1) = iii+1
!! ibcd(iii+1) = iii+1
print *, alph(1),alph(2),alph(3),alph(4)
!! print *, ibcd(1),ibcd(2),ibcd(3),ibcd(4)
end
integer omp_get_thread_num
!$omp parallel
call sub(omp_get_thread_num())
!$omp endparallel
end
% pgf90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -mp -o tp1_625 tp1.f
% pgf90 -mp -o tp2_625 tp2.f
% pgf90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -mp -o tp1_702 tp1.f
% pgf90 -mp -o tp2_702 tp2.f
% ./tp1_625
Warning: OMP_NUM_THREADS or NCPUS (4) greater than available cpus (2)
1 1 1 1
2 2 2 2
3 3 3 3
4 4 4 4
% ./tp2_625
Warning: OMP_NUM_THREADS or NCPUS (4) greater than available cpus (2)
1 1 1 1
2 2 2 2
3 3 3 3
4 4 4 4
% ./tp1_702
Warning: OMP_NUM_THREADS or NCPUS (4) greater than available cpus (2)
1 0 0 0
0 2 0 0
0 0 3 0
0 0 0 4
% ./tp2_702
Warning: OMP_NUM_THREADS or NCPUS (4) greater than available cpus (2)
1 0 0 0
0 2 0 0
0 0 3 0
0 0 0 4
IMPLIED DO error where there is none, and compiler crash where there is an error.
> cat test1.f
MODULE NOT_IMPLIED_DO
TYPE TEST
REAL R, R2(2,3)
END TYPE TEST
REAL, PARAMETER :: Q(2,3)=0
TYPE(TEST),PARAMETER :: BUGGED = TEST(0,Q)
END MODULE
> cat test2.f
MODULE NOT_IMPLIED_DO
TYPE TEST
REAL R, R2(2,3)
END TYPE TEST
REAL, PARAMETER :: Q(2,3)=0
TYPE(TEST),PARAMETER :: BUGGED = BLAH(0,Q)
END MODULE
> pgf90 -c test1.f
PGF90-S-0069-Illegal implied DO expression (test1.f: 8)
0 inform, 0 warnings, 1 severes, 0 fatal for not_implied_do
> pgf90 -c test2.f
PGF90-S-0000-Internal compiler error. unexpected ast type in
initialization expr 16 (test2.f: 8)
pgf90-Fatal-/opt/pgi/6.2.5/linux86-64/6.2/bin/pgf901 TERMINATED by
signal 11 Arguments to /opt/pgi/6.2.5/linux86-64/6.2/bin/pgf901
/opt/pgi/6.2.5/linux86-64/6.2/bin/pgf901 test2.f -opt 1 -terse 1
-inform warn -nohpf -nostatic -x 19 0x400000 -quad -x 59 4 -x 59 4 -x
15 2 -x 49 0x400004 -x 51 0x20 -x 57 0x4c -x 58 0x10000 -x 124 0x1000
-x 57 0xfb0000 -x 58 0x78031040 -x 48 4608 -x 49 0x100 -x 120 0x200
-stdinc
/opt/pgi/6.2.5/linux86-64/6.2/include:/usr/local/include:/usr/lib64/gcc-lib/x86_64-suse-linux/3.2.2/include:/usr/lib64/gcc-lib/x86_64-suse-linux/3.2.2//include:/usr/include
-def unix -def __unix -def __unix__ -def linux -def __linux -def
__linux__ -def __NO_MATH_INLINES -def __x86_64__ -def
__LONG_MAX__=9223372036854775807L -def '__SIZE_TYPE__=unsigned long
int' -def '__PTRDIFF_TYPE__=long int' -def __THROW= -def
__extension__= -def __amd64__ -def __SSE__ -def __MMX__ -def __SSE2__
-nofreeform -vect 48 -output /tmp/pgf90eLFgmPZs_Wtl.ilm
A simple program with an error causes a compiler ICE.
% cat test.f90
PROGRAM test
IMPLICIT NONE
G(2) = 0
END PROGRAM test
% pgf90 -V test.f90
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0072-Assignment operation illegal to external procedure g (test.f90: 3)
PGF90-S-0000-Internal compiler error. sym_of_ast: unexpected ast 12 (test.f90: 3)
PGF90-S-0038-Symbol, g, has not been explicitly declared (test.f90)
0 inform, 0 warnings, 3 severes, 0 fatal for test
% pgf90 -V test.f90
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0072-Assignment operation illegal to external procedure g (test.f90: 3)
PGF90-S-0038-Symbol, g, has not been explicitly declared (test.f90)
0 inform, 0 warnings, 2 severes, 0 fatal for test
Similar to 4028.
% cat test.f90
SUBROUTINE A
TYPE T
DOUBLE COMPLEX :: X
END TYPE T
TYPE(T) :: TT
DOUBLE PRECISION :: Z
TT%Y = TT%Y * Z
END SUBROUTINE
% pgf90 -V test.f90 -c
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0142-y is not a member of this RECORD (test.f90: 9)
PGF90-S-0142-y is not a member of this RECORD (test.f90: 9)
PGF90-S-0000-Internal compiler error. sym_of_ast: unexpected ast 0 (test.f90: 9)
0 inform, 0 warnings, 3 severes, 0 fatal for a
% pgf90 -V test.f90 -c
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0142-y is not a member of this RECORD (test.f90: 9)
PGF90-S-0142-y is not a member of this RECORD (test.f90: 9)
0 inform, 0 warnings, 2 severes, 0 fatal for a
A user's example causes the compiler to seg fault. Now it only reports a severe error.
User's example causes the an Internal Compiler Error (ICE).
% cat test.c
#include
struct bar_t {
bool a;
};
int foo() {
struct bar_t c, d;
c = d;
}
% pgcc -V -c test.c
pgcc 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86-64 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGC-S-0000-Internal compiler error. copy_elem, unexp. dt 11 (test.c: 9)
PGC-S-0000-Internal compiler error. copy_elem, unexp. dt 11 (test.c: 9)
PGC/x86-64 Linux/x86-64 6.2-5: compilation completed with severe errors
% pgcc -V -c test.c
pgcc 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
xxx
% cat test.f90
INTEGER,PARAMETER :: mat_ptr = int(floor(2d0))
END
% pgf90 -V -c test.f90
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90-S-0000-Internal compiler error. unexpected ast type in initialization expr 12 (test.f90: 1)
pgf90-Fatal-/tmp/pgi625/linux86-64/6.2/bin/pgf901 TERMINATED by signal 11
Arguments to ......
-def __SSE__ -def __MMX__ -def __SSE2__ -freeform -vect 48 -output /tmp/pgf90ygJciO_cmjYk.ilm
% pgf90 -V -c test.f90
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
Threadprivate global variable not used in contained subprograms.
% cat test.f90
implicit none
integer, save :: ix=0
!$OMP THREADPRIVATE(ix)
write(*,*) 'before ix =',ix
call process_options
write(*,*) 'ix after process_options ', ix
call initialize_gfsr
return
contains
subroutine process_options
ix=0
ix = 2*5
return
end subroutine process_options
subroutine initialize_gfsr
if (ix /= 0) then
write(*,*) 'hello'
endif
return
end subroutine initialize_gfsr
end
% pgf90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -mp -o test_625 test.f90
% pgf90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -mp -o test_702 test.f90
% ./test_625
before ix = 0
ix after process_options 0
% ./test_702
before ix = 0
ix after process_options 10
hello
The pgCC switch -MM should only give user include header file dependencies. Instead, it gives all headers
This has been corrected.
A user wanted to use a Module in a DLL. The subroutines exported properly and
you are able to export variables. When you tried to use the exported variables in a program,
it appears that there are two copies of the variables, one local and one in the module. There should only be one.
This has been corrected.
A major appication was finding errors in their standard tests. The problem was isolated and fixed.
The code below results in a segfault when compiled and run
on a linux x86_64 AMD platform. Removing the scalar
multiplication at the front of the outputted array (zfni) will make the
problem go away but it is an allowed operation, hence a bug.
program tst
implicit none
real(8) :: zfni
integer(8) :: iout(3),jout(3)
integer(8) :: ii,jj
real(8), dimension(205,205) :: phi
iout(1)=1
iout(2)=205
iout(3)=2
jout(1)=1
jout(2)=205
jout(3)=2
zfni=4.0
do jj=1,205
do ii=1,205
phi(ii,jj)=real(ii)*real(jj)
enddo
enddo
open (20,file='tmp',form='unformatted')
write (20)
zfni*phi(iout(1):iout(2):iout(3),jout(1):jout(2):jout(3))
close(20)
stop
end
Code illustrates a problem when an allocatable array in a dervived type is used on the LHS of an assignment in a forall construct.
Some Users commented that since PGI compilers run on Linux and Windows systems, we should
not refer to Netscape browsers in our documents. Moving to generic browser references.
The 64-bit libpgc.a library has a non-fpic object in it which can
ruin the build of another shared library that uses it.
This has been corrected.
call exit(code) fails to echo code.
% cat test.f90
program test
implicit none
integer :: code
print *,"Enter an integer: "
read *,code
call exit(code)
end program test
% pgf90 -o test_625 -V test.f90
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/any Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o test_702 -V test.f90
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./test_625
Enter an integer:
15
% echo $?
127
% ./test_702
Enter an integer:
15
% echo $?
15
An include file in the PGI include area contained the line #include "/usr/include/stdlib.h" should be #include_next "stdlib.h"
This has been corrected.
A program failed in release 6.2-5 because -Msignextend was not default, and it failed
without the switch. The reasons we did not make -Msignextend default have been corrected, and we
now agree with gcc in this regard.
A user pointed out that our documentation implied that we had three types of fast math routines,
but we only have two types. This has been corrected.
pgcc inline assembly fails with the user example.
% cat test.c
typedef struct { /* standard complex number declaration for single- */
float real; /* precision complex numbers */
float imag;
} complex;
typedef struct { complex e[3][3]; } su3_matrix;
/* These assembly inline versions work only with the gnu C compiler
on Opteron processors */
#define _inline_sse_mult_su3_an(aa,bb) \
{ \
__asm__ __volatile__ ("movlps %0, %%xmm7 \n\t" \
"movaps %%xmm7, %%xmm0 \n\t" \
"unpcklps %%xmm0, %%xmm0 \n\t" \
"movhlps %%xmm0, %%xmm3 \n\t" \
"movlhps %%xmm0, %%xmm0 \n\t" \
"movhps %1, %%xmm7 \n\t" \
"movaps %%xmm7, %%xmm1 \n\t" \
"unpckhps %%xmm1, %%xmm1 \n\t" \
"movhlps %%xmm1, %%xmm4 \n\t" \
"movlhps %%xmm1, %%xmm1 \n\t" \
"movlps %2, %%xmm10 \n\t" \
"movaps %%xmm10, %%xmm2 \n\t" \
"unpcklps %%xmm2, %%xmm2 \n\t" \
"movhlps %%xmm2, %%xmm5 \n\t" \
"movlhps %%xmm2, %%xmm2 \n\t" \
"movlps %3, %%xmm12 \n\t" \
"movhps %4, %%xmm12 \n\t" \
"mulps %%xmm12, %%xmm0 \n\t" \
"movlps %5, %%xmm13" \
: \
: \
"m" ((aa)->e[0][0]), \
"m" ((aa)->e[1][0]), \
"m" ((aa)->e[2][0]), \
"m" ((bb)->e[0][0]), \
"m" ((bb)->e[0][1]), \
"m" ((bb)->e[1][0])\
: \
"%xmm0", \
"%xmm1", \
"%xmm10", \
"%xmm12", \
"%xmm13", \
"%xmm2", \
"%xmm3", \
"%xmm4", \
"%xmm5", \
"%xmm7"); \
}
void mult(su3_matrix *m1, su3_matrix *m4) {
_inline_sse_mult_su3_an(m4,m1);
}
% pgcc -V -c test.c
pgcc 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86-64 Linux/x86-64 6.2-5
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
PGC-W-0352-Constraint/expression type mismatch in extended ASM operand 1 (test.c: 55)
PGC-W-0352-Constraint/expression type mismatch in extended ASM operand 2 (test.c: 55)
PGC-W-0352-Constraint/expression type mismatch in extended ASM operand 4 (test.c: 55)
PGC-W-0352-Constraint/expression type mismatch in extended ASM operand 5 (test.c: 55)
PGC/x86-64 Linux/x86-64 6.2-5: compilation completed with warnings
/tmp/pgccLG_bVPt2xoJ_.s: Assembler messages:
/tmp/pgccLG_bVPt2xoJ_.s:42: Error: suffix or operands invalid for `movhps'
/tmp/pgccLG_bVPt2xoJ_.s:47: Error: suffix or operands invalid for `movlps'
/tmp/pgccLG_bVPt2xoJ_.s:53: Error: suffix or operands invalid for `movhps'
/tmp/pgccLG_bVPt2xoJ_.s:55: Error: suffix or operands invalid for `movlps'
% settmp64702
% pgcc -V -c test.c
pgcc 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGC/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
Customer wanted specific I/O libraries created.
User code had erroneous Overlapping data initializations warnings. This has been corrected.
Program seg faults with character length >64, and TRANSFER function
% cat Fortran-TRANSFER-Bug.F90
PROGRAM testb
IMPLICIT none
CHARACTER(len=VECT_LENGHT) :: vect(1)
real :: mvect(3000)
CHARACTER(len=1), ALLOCATABLE :: cvect(:)
INTEGER :: length
vect(1)='a'!(/'ippaaa0','ippaaa1','ippaaa2'/)
WRITE(*,*) 'Say hi!'
mvect=TRANSFER(vect,mvect)
WRITE(*,*) 'Toto'
END PROGRAM testb
User program caused PGC-F-0000-Internal compiler error. filename too long 0 () failure.
This has been corrected.
Simple fortran OPEN test works with pgf77, fails with pgf90.
% more test.f
program test
open(1,file='test.out',access='append', FORM='FORMATTED')
write(1,*) 'append2 test'
close(1)
stop
end
% pgf77 -o test test.f
% test
FORTRAN STOP
% pgf90 -o test test.f
% ./test
PGFIO-F-201/OPEN/unit=1/illegal value for specifier.
File name = test.out
In source file test.f, at line number 2
Program runs corrected only when you include a module the progrma does not reference.
% cat testa.f90
MODULE G
IMPLICIT NONE
SAVE
TYPE TYPE_Y
INTEGER :: U=1
INTEGER :: V=2
INTEGER :: T=3
INTEGER :: Q=4
END TYPE TYPE_Y
TYPE(TYPE_Y) :: Y
END MODULE
PROGRAM TEST
! You need this with pgf90
! USE G
call mumble
END PROGRAM
SUBROUTINE MUMBLE
USE G
print *,"Y%V",Y%V
END SUBROUTINE
% cat testb.f90
MODULE G
IMPLICIT NONE
SAVE
TYPE TYPE_Y
INTEGER :: U=1
INTEGER :: V=2
INTEGER :: T=3
INTEGER :: Q=4
END TYPE TYPE_Y
TYPE(TYPE_Y) :: Y
END MODULE
PROGRAM TEST
! You need this with pgf90
USE G
call mumble
END PROGRAM
SUBROUTINE MUMBLE
USE G
print *,"Y%V",Y%V
END SUBROUTINE
% pgf90 -V
pgf90 6.2-5 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2006, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o testa_625 testa.f90
% pgf90 -o testb_625 testb.f90
% pgf90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgf90 -o testa_702 testa.f90
% pgf90 -o testb_702 testb.f90
% ./testa_625
Y%V 0
% ./testb_625
Y%V 2
% ./testa_702
Y%V 2
% ./testb_702
Y%V 2
CDK mpif90, mpif77, etc scripts need to act exactly like pgf90 -Mmpi or -Mmpi2. There
were differences in libs used and lib order, and that has been corrected.
User example caused pgCC to seg fault when compiling with -Mipa. This
has been corrected.
We did not include librt.so when -pgf77libs or -pgf90libs is used with pgcc.
This has been corrected.
pgcc and pgCC using VLA after a block that reads size causes a compiler ICE.
This has been corrected.
A user example would not link as expected with -mcmodel=medium, but would without the switch.
This has been corrected.
Our beta Windows SUA product exhibited errors when the executable is stripped.
This has been corrected.
liblapack and libblas were left out of the Windows compiler packages,
and this has been corrected.
32-bit pgCC does not read back 'long long' from file properly.
% more ioLongLong.cc
#include
#include
#include
#include
#include
#include
using namespace std;
int main (int argc, char ** argv)
{
const char fileName[] = "my8bytes.tmp";
#ifndef READ_ONLY
ofstream out (fileName);
assert (out != NULL);
const long long outLongLong = 0xdeadbeefdeadbeefLL;
out << outLongLong;
out.close();
cout << "wrote " << hex << outLongLong << " to ";
#endif
cout << "file " << fileName;
ifstream in (fileName);
assert (in != NULL);
long long inLongLong = 17;
in >> inLongLong;
in.close();
cout << ", read " << hex << inLongLong << endl;
exit (0);
return (0);
}
% g++ -o ioLongLong_gcc ioLongLong.cc
% ./ioLongLong_gcc
wrote deadbeefdeadbeef to file my8bytes.tmp, read deadbeefdeadbeef
With the file my8bytes.tmp created by the g++ version, let's read it with
the pgCC compiled versions.
% pgCC -DREAD_ONLY -tp px -o ioLongLong_px ioLongLong.cc -V
pgCC 7.0-2a 32-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
Edison Design Group C/C++ Front End, version 3.6 (Jan 26 2007 22:44:19)
Copyright 1988-2006 Edison Design Group, Inc.
PGCC/x86 Linux 7.0-2a
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% pgCC -DREAD_ONLY -tp k8-32 -o ioLongLong_k832 ioLongLong.cc
% pgCC -DREAD_ONLY -tp k8-64 -o ioLongLong_k864 ioLongLong.cc
% pgCC -DREAD_ONLY -tp x64 -o ioLongLong_x64 ioLongLong.cc
% pgCC -DREAD_ONLY -tp piv -o ioLongLong_piv ioLongLong.cc
Notice the 32-bits fail, while the 64-bit versions do not.
% ./ioLongLong_k864
file my8bytes.tmp, read deadbeefdeadbeef
% ./ioLongLong_x64
file my8bytes.tmp, read deadbeefdeadbeef
% ./ioLongLong_piv
file my8bytes.tmp, read 8000000000000000
% ./ioLongLong_k832
file my8bytes.tmp, read 8000000000000000
% ./ioLongLong_px
file my8bytes.tmp, read 8000000000000000
This test case points out a bug when pointers are in common.
We do allow them, but the storage order of the common is wrong.
Specifically, we placed the nonpointer variables after the pointer's
compiler-created variables.
The stack command in the win32 pgdbg is providing information about just the current routine even if the current routine is not main.
The stack command is working correctly in the win64 pgdbg. This problem affects the contents of the Stack Window in PVF.
SUA64: MP Correctness test report stack overflow when compiled with -Mchkstk.
Example:
pgcc -c -I/dev/fs/Q/tests/mp -fast -Mchkstk -V7.0-2 -mp /dev/fs/Q/tests/mp/check.c
pgf90 -c -I/dev/fs/Q/tests/mp -fast -Mchkstk -V7.0-2 -mp /dev/fs/Q/tests/mp/do10.f
pgf90 -fast -Mchkstk -V7.0-2 -mp do10.o check.o -o a.out
------------ executing test do10
./RUNPAR 4 ./a.out
Error: in routine sp2 there is a
Error: in routine sp2 there is a
Error: in routine sp2 there is a
stack overflow: thread 0, max 10561KB, used 2KB, request 64B
stack overflow: thread 0, max 10561KB, used 2KB, request 64B
stack overflow: thread 0, max 10561KB, used 2KB, request 64B
User thinks warnings about variables that do not become too long
until the compiler decorates them, are not worth reporting.
pgf90 -c geometry_class.f90 -Mstandard -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v$sd, truncated to 31 chars (geometry_class.f90: 549)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v117, truncated to 31 chars (geometry_class.f90: 549)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v11$sd, truncated to 31 chars (geometry_class.f90: 549)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v11$o, truncated to 31 chars (geometry_class.f90: 549)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v11$p, truncated to 31 chars (geometry_class.f90: 549)
0 inform, 5 warnings, 0 severes, 0 fatal for get_instance_geometry_2_geo
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v$sd, truncated to 31 chars (geometry_class.f90: 620)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v169, truncated to 31 chars (geometry_class.f90: 620)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v16$sd, truncated to 31 chars (geometry_class.f90: 620)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v16$o, truncated to 31 chars (geometry_class.f90: 620)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v16$p, truncated to 31 chars (geometry_class.f90: 620)
0 inform, 5 warnings, 0 severes, 0 fatal for process_input_geometry_geo
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v$sd, truncated to 31 chars (geometry_class.f90: 984)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v361, truncated to 31 chars (geometry_class.f90: 984)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v36$sd, truncated to 31 chars (geometry_class.f90: 984)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v36$o, truncated to 31 chars (geometry_class.f90: 984)
PGF90-W-0016-Identifier, get_instance_geometry_1_geo$v36$p, truncated to 31 chars (geometry_class.f90: 984)
0 inform, 5 warnings, 0 severes, 0 fatal for check_consistency_geom_2_geo
Release7.0-2 pgf90did not support !DEC$directives the way6.2-5did.
endparallel is not treated the same as end parallel in OpenMP
% cat test.f90
! extracted from OpenMP test omp1/F2_1_1_2_1a.f90
use omp_lib
implicit none
integer, parameter :: N=100000
double precision :: a(N),b(N),c(N)
integer i
call random_number(a)
call random_number(b)
!$omp paralleldo default(none) shared(a,b,c) schedule(static)
do i = 1, N
c(i) = b(i) * a(i)
enddo
!$omp endparallel do
print *, c(1),c(N)
end
Unknown options are now treated as errors instead of warnings. This
change makes it a compiler error to pass switches that are not known to the
compiler. But users want their build scripts to work as before.
Added switch -noswitcherror, to turn switch errors into warnings instead
also allows a user to 'set NOSWITCHERROR=1;' in a siterc file for the same
behavior globally.
OpenMP Directive NUM_THREADS extends outside scope of current section.
This test case was extracted from OpenMP test omp1/F2_2_1a.f90, which tests
if a NUM_THREADS clause on an omp parallel directive affects the next
omp parallel directive which does not have a NUM_THREADS clause appearing on it.
% cat test.f90
! extracted from OpenMP test omp1/F2_2_1a.f90
use omp_lib
implicit none
integer, parameter :: NT = 4
integer :: i = 0
call omp_set_dynamic(.false.)
call omp_set_num_threads(NT+4)
!$omp parallel default(none) shared(i)
!$omp atomic
i = i+1
!$omp end parallel
if(i /= NT+4) then
print*, '1 FAIL - wrong number of threads, i=',i
endif
i=0
!$omp parallel default(none) shared(i) NUM_THREADS(NT)
!$omp atomic
i = i+1
!$omp end parallel
if(i /= NT) then
print*, '2 FAIL - wrong number of threads, i=',i
endif
!
! this loop does not revert back to 8 threads
!
i=0
!$omp parallel default(none) shared(i)
!$omp atomic
i = i+1
!$omp end parallel
if(i /= NT+4) then
print*, '3 FAIL - wrong number of threads, i=',i
endif
END
% pgf90 -mp -o test test.f90 -V
pgf90 7.0-2 32-bit target on x86 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
% ./test
Warning: omp_set_num_threads (8) greater than available cpus (1)
3 FAIL - wrong number of threads, i= 4
The optimization of the code for the two test cases is significantly
different, with test1.f90 generating slower code. On some benchmark
codes the version without the TARGET attribute is more than twice as
fast. This is a significant performance issue.
% cat test1.f90
subroutine tt(b,c,n)
integer n
real b(n),c(n)
real , allocatable, TARGET :: a(:)
ALLOCATE(A(n))
do i=1,n
a(i)=b(i)*c(i)
enddo
call sub(a)
end
harrier% cat test2.f90
subroutine tt(b,c,n)
integer n
real b(n),c(n)
real , allocatable :: a(:)
ALLOCATE(A(n))
do i=1,n
a(i)=b(i)*c(i)
enddo
call sub(a)
end
% diff test1.f90 test2.f90
4c4
< real , allocatable, TARGET :: a(:)
---
> real , allocatable :: a(:)
% pgf90 -c -r8 -Minfo=all -fastsse test1.f90 -V
pgf90 7.0-2 64-bit target on x86-64 Linux
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
PGF90/x86-64 Linux 7.0-2
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2007, STMicroelectronics, Inc. All Rights Reserved.
tt:
6, Unrolled inner loop 4 times
Generated 2 prefetch instructions for this loop
% pgf90 -c -r8 -Minfo=all -fastsse test2.f90
tt:
6, Generated 4 alternate loops for the inner loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
Generated 2 prefetch instructions for this loop
Generated vector sse code for inner loop
