From eschnetz at harris.com  Wed Sep  5 12:48:53 2007
From: eschnetz at harris.com (Schnetzer, Everett)
Date: Wed, 5 Sep 2007 08:48:53 -0400
Subject: Compile problems with reference implementation and gnu 3.4.5
Message-ID: <E7191752EF4FAD498FD0E968F7A7DAF70AA340@mlbe2k5.cs.myharris.net>

Hi,

I am trying to port the reference implementation to Windows using MinGW
and gnu 3.4.5, and the doesn't like certain static methods within
template structs. The first error encountered while building the
tests/ref-impl programs is:

g++ -DUNIT_TEST_FAILURE_PRINT \
-I../../src -I/include \
-c math-matvec.cpp -o math-matvec.o
../../src/vsip/core/matvec.hpp: In function `typename
vsip::Promotion<Left, Right>::type
vsip::impl::impl_dot(vsip::const_Vector<T0, Block0>,
vsip::const_Vector<T1, Block1>) [with T0 = vsip::cscalar_f, T1 =
vsip::cscalar_f, Block0 = vsip::Dense<1u, vsip::cscalar_f,
vsip::tuple<0u, 1u, 2u>, vsip::Local_map>, Block1 = const
vsip::impl::Unary_expr_block<1u, vsip::impl::conj_functor,
vsip::Dense<1u, vsip::cscalar_f, vsip::tuple<0u, 1u, 2u>,
vsip::Local_map>, vsip::cscalar_f>]':
../../src/vsip/core/matvec.hpp:609:   instantiated from `typename
vsip::Promotion<std::complex<_Tp>, std::complex<Right> >::type
vsip::cvjdot(vsip::const_Vector<std::complex<_Tp>, Block0>,
vsip::const_Vector<std::complex<Right>, Block1>) [with T0 = float, T1 =
float, Block0 = vsip::Dense<1u, vsip::cscalar_f, vsip::tuple<0u, 1u,
2u>, vsip::Local_map>, Block1 = vsip::Dense<1u, vsip::cscalar_f,
vsip::tuple<0u, 1u, 2u>, vsip::Local_map>]'
math-matvec.cpp:83:   instantiated from here
../../src/vsip/core/matvec.hpp:164: error: `exec' is not a member of
`vsip::impl::Evaluator<vsip::impl::Op_prod_vv_dot,
vsip::impl::Return_scalar<std::complex<float> >,
vsip::impl::Op_list_2<vsip::Dense<1u, vsip::cscalar_f, vsip::tuple<0u,
1u, 2u>, vsip::Local_map>, const vsip::impl::Unary_expr_block<1u,
vsip::impl::conj_functor, vsip::Dense<1u, vsip::cscalar_f,
vsip::tuple<0u, 1u, 2u>, vsip::Local_map>, vsip::cscalar_f> >,
vsip::impl::Cvsip_tag>'

I cannot spot any errors in the source that would lead to this error.
The compiler doesn't have this issue with the GPL version of the code,
although there seems to be similar implementation patterns in that code
as well. The error does show up in the GPL version of the tests
(matvec-prod.cpp) if I force it to use the reference version via the
VSIP_IMPL_REF_IMPL flag. So I'm stuck trying to figure out why gnu 3.4.5
doesn't like the implementations of Evaluator in core/matvec.hpp. Any
clues would be appreciated.

Sincerely,

Everett Schnetzer
Software Engineer, Harris GCSD
eschnetz at harris.com
321-727-6607

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sourcerytools.com/pipermail/vsipl++/attachments/20070905/c2060fef/attachment.html>

From stefan at codesourcery.com  Fri Sep  7 22:31:49 2007
From: stefan at codesourcery.com (Stefan Seefeld)
Date: Fri, 07 Sep 2007 18:31:49 -0400
Subject: [vsipl++] Compile problems with reference implementation and
 gnu 3.4.5
In-Reply-To: <E7191752EF4FAD498FD0E968F7A7DAF70AA340@mlbe2k5.cs.myharris.net>
References: <E7191752EF4FAD498FD0E968F7A7DAF70AA340@mlbe2k5.cs.myharris.net>
Message-ID: <46E1D155.8040209@codesourcery.com>

Everett,

thanks for reporting this issue. I have tried to reproduce it on Linux
with g++ 3.4.5 (assuming that mingw 3.4.5 would use the same C++ frontend,
and thus expose the same problems as the stock g++ 3.4.5, but couldn't
see any problem. The compilation command you show works fine.

We will try to reproduce this on a Windows machine.

Meanwhile, can you provide us with some details as to how you configured
and built ? What options and commands did you use exactly ?

Thanks,
		Stefan

-- 
Stefan Seefeld
CodeSourcery
stefan at codesourcery.com
(650) 331-3385 x718


From eschnetz at harris.com  Mon Sep 10 13:01:48 2007
From: eschnetz at harris.com (Schnetzer, Everett)
Date: Mon, 10 Sep 2007 09:01:48 -0400
Subject: [vsipl++] Compile problems with reference implementation and gnu 3.4.5
In-Reply-To: <46E1D155.8040209@codesourcery.com>
References: <E7191752EF4FAD498FD0E968F7A7DAF70AA340@mlbe2k5.cs.myharris.net> <46E1D155.8040209@codesourcery.com>
Message-ID: <E7191752EF4FAD498FD0E968F7A7DAF70AA356@mlbe2k5.cs.myharris.net>

Stefan,

Here's the configure command and status:

.\configure --enable-ref-impl --disable-mpi --without-lapack
--enable-cvsip
 
--with-cvsip-prefix=/m/aip_eschnetz_ai0127_view/AIP_IRAD/COTS/i686-pc-mi
ngw32

...

configure: Summary
Build in maintainer-mode:                
Using config suffix:                     
Exceptions enabled:                      yes
With mpi enabled:                        no
With PAS enabled:                        no
With LAPACK:                             no
With SAL:                                no
With IPP:                                no
With C-VSIPL:                            yes
Using FFT backends:                      cvsip
  Provides float FFTs
  Provides double FFTs
Complex storage format:                  interleaved
Timer:                                   none
With Python bindings:                    no
configure: creating ./config.status
config.status: creating benchmarks/GNUmakefile.inc
config.status: creating benchmarks/hpec_kernel/GNUmakefile.inc
config.status: creating doc/Doxyfile
config.status: creating doc/GNUmakefile.inc
config.status: creating examples/GNUmakefile.inc
config.status: creating examples/makefile.standalone
config.status: creating GNUmakefile
config.status: creating scripting/GNUmakefile.inc
config.status: creating src/vsip/core/acconfig.hpp
config.status: creating src/vsip/GNUmakefile.inc
config.status: creating src/vsip/opt/cbe/alf/src/ppu/GNUmakefile.inc
config.status: creating src/vsip/opt/cbe/spu/GNUmakefile.inc
config.status: creating src/vsip_csl/GNUmakefile.inc
config.status: creating synopsis.py
config.status: creating tests/context-installed.pre
config.status: creating tests/context
config.status: creating tests/GNUmakefile.inc
config.status: creating tests/QMTest/configuration
config.status: creating vendor/clapack/blas/SRC/GNUmakefile
config.status: creating vendor/clapack/F2CLIBS/libF77/GNUmakefile
config.status: creating vendor/clapack/SRC/GNUmakefile
config.status: creating vendor/clapack/SRC/make.inc
config.status: creating vendor/GNUmakefile.inc
config.status: creating vendor/lapack/make.inc
config.status: creating vendor/lapack/SRC/GNUmakefile
config.status: creating vsipl++.pc
config.status: creating src/vsip/core/acconfig.hpp 

The build of the VSIPL++ library is done with "make -fGNUmakefile all",
and it succeeds. Then I change directories to tests/ref-impl and try
"make -fMakefile check". For the tests, the Makefile and Makefile.inc
files were updated for our directory structure. The error appears when
it gets to math-matvec.cpp file.

I've played some with the flags used to compile the tests (removed -O2
and -pedantic), but nothing has changed the result. I hope your attempt
on Windows reproduces the problem, but since it works for Linux, I'll
start looking for possible problems in our environment.

Thanks,

Everett


From stefan at codesourcery.com  Mon Sep 10 18:56:04 2007
From: stefan at codesourcery.com (Stefan Seefeld)
Date: Mon, 10 Sep 2007 14:56:04 -0400
Subject: patch: minor fixes
Message-ID: <46E59344.8040206@codesourcery.com>

The attached patch fixes three issues:

* The default-option logic in configure.ac was wrong such that --enable-ref-impl
  and --without-lapack couldn't coexist.

* The src/core/vsip/fft.hpp file had a typo (exposed when built in ref-impl mode).

* The tests/ref-impl/GNUmakefile is obsolete and outdated, and thus is removed now.

OK to check in ?

Thanks,
		Stefan


-- 
Stefan Seefeld
CodeSourcery
stefan at codesourcery.com
(650) 331-3385 x718
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch
URL: <http://sourcerytools.com/pipermail/vsipl++/attachments/20070910/f8fb5798/attachment.ksh>

From stefan at codesourcery.com  Mon Sep 10 18:59:11 2007
From: stefan at codesourcery.com (Stefan Seefeld)
Date: Mon, 10 Sep 2007 14:59:11 -0400
Subject: [vsipl++] patch: minor fixes
In-Reply-To: <46E59344.8040206@codesourcery.com>
References: <46E59344.8040206@codesourcery.com>
Message-ID: <46E593FF.7090301@codesourcery.com>

Sorry, the previous mail contained the unfiltered output of 'svn diff'.
here is the cleaned-up patch for the three involved files.

Thanks,
		Stefan

-- 
Stefan Seefeld
CodeSourcery
stefan at codesourcery.com
(650) 331-3385 x718


From stefan at codesourcery.com  Mon Sep 10 19:00:50 2007
From: stefan at codesourcery.com (Stefan Seefeld)
Date: Mon, 10 Sep 2007 15:00:50 -0400
Subject: [vsipl++] patch: minor fixes
In-Reply-To: <46E593FF.7090301@codesourcery.com>
References: <46E59344.8040206@codesourcery.com> <46E593FF.7090301@codesourcery.com>
Message-ID: <46E59462.3050208@codesourcery.com>

Stefan Seefeld wrote:
> Sorry, the previous mail contained the unfiltered output of 'svn diff'.
> here is the cleaned-up patch for the three involved files.

There. Sorry for the noise.

		Stefan

-- 
Stefan Seefeld
CodeSourcery
stefan at codesourcery.com
(650) 331-3385 x718
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch
URL: <http://sourcerytools.com/pipermail/vsipl++/attachments/20070910/f8e8b82f/attachment.ksh>

From stefan at codesourcery.com  Mon Sep 10 19:52:19 2007
From: stefan at codesourcery.com (Stefan Seefeld)
Date: Mon, 10 Sep 2007 15:52:19 -0400
Subject: [vsipl++] Compile problems with reference implementation and
 gnu 3.4.5
In-Reply-To: <E7191752EF4FAD498FD0E968F7A7DAF70AA356@mlbe2k5.cs.myharris.net>
References: <E7191752EF4FAD498FD0E968F7A7DAF70AA340@mlbe2k5.cs.myharris.net> <46E1D155.8040209@codesourcery.com> <E7191752EF4FAD498FD0E968F7A7DAF70AA356@mlbe2k5.cs.myharris.net>
Message-ID: <46E5A073.8040305@codesourcery.com>

Everett,

Schnetzer, Everett wrote:

> The build of the VSIPL++ library is done with "make -fGNUmakefile all",

(The -f option shouldn't be necessary. The reason we use GNUmakefile
 instead of Makefile is to make clear (to 'make') that we really require
 GNU make -- which detects GNUmakefiles automatically. Other versions of
 make won't be able to handle our build system anyway.)

> and it succeeds. Then I change directories to tests/ref-impl and try
> "make -fMakefile check". For the tests, the Makefile and Makefile.inc
> files were updated for our directory structure. The error appears when
> it gets to math-matvec.cpp file.

That may be the root of the problem. tests/ref-impl/ doesn't come with
a 'Makefile', only with a 'GNUmakefile'. That, however, is obsolete, and
not used by our testing harness.

(Our testing harness uses QMTest, and to invoke it you run 'make check'
 from the toplevel directory. We also support compiling individual tests
 in a way as not to require QMTest, though the ref-impl test suite doesn't
 seem to be supported that way. I would suggest to modify / add to the
 tests_cxx_sources variable in tests/GNUmakefile.inc to include tests/ref-impl/*.cpp.

 Once that is done, you can run 'make tests/ref-impl/math-matvec' from the toplevel
 directory.)

> I've played some with the flags used to compile the tests (removed -O2
> and -pedantic), but nothing has changed the result. I hope your attempt
> on Windows reproduces the problem, but since it works for Linux, I'll
> start looking for possible problems in our environment.

As I can't reproduce this with my (Linux-based) version of GCC 3.4.5, and
since this looks like a C++ frontend issue, rather than a platform-specific
problem, I'm tempted to believe that the cause is related to your writing
your own Makefile for tests/ref-impl/.

Regards,
		Stefan

-- 
Stefan Seefeld
CodeSourcery
stefan at codesourcery.com
(650) 331-3385 x718


From jules at codesourcery.com  Mon Sep 10 20:18:27 2007
From: jules at codesourcery.com (Jules Bergmann)
Date: Mon, 10 Sep 2007 16:18:27 -0400
Subject: [vsipl++] patch: minor fixes
In-Reply-To: <46E59462.3050208@codesourcery.com>
References: <46E59344.8040206@codesourcery.com> <46E593FF.7090301@codesourcery.com> <46E59462.3050208@codesourcery.com>
Message-ID: <46E5A693.70602@codesourcery.com>

Stefan Seefeld wrote:
> Stefan Seefeld wrote:
>> Sorry, the previous mail contained the unfiltered output of 'svn diff'.
>> here is the cleaned-up patch for the three involved files.
> 
> There. Sorry for the noise.
> 
> 		Stefan
> 
> 

This looks good.  Please check it in, thanks!  -- Jules

-- 
Jules Bergmann
CodeSourcery
jules at codesourcery.com
(650) 331-3385 x705


From cwilso11 at harris.com  Wed Sep 26 21:24:45 2007
From: cwilso11 at harris.com (Wilson, Charles (cwilso11))
Date: Wed, 26 Sep 2007 17:24:45 -0400
Subject: VSIPL++ and Visual C++
Message-ID: <1B90921D44561C4A977E2604A5247075068508@mlbe2k4.cs.myharris.net>

From
http://www.codesourcery.com/public/vsiplplusplus/sourceryvsipl++-1.3/qui
ckstart/ch02s03.html, it is apparent that /building/ CodeSourcery
VSIPL++ requires the Intel C++ compiler. Once built, though, can the
library be used from Microsoft Visual C++ 2005?  If VSIPL++ were a C
library, the answer would obviously be yes because the ABI for C is
standardized and the header files contain little actual code.  However,
CodeSourcery VSIPL++ is C++, which makes for ABI issues between
compilers.  Also, it makes extensive use of templates and template
meta-programming techniques so quite a bit of the functionality is in
the public header files -- which the "client compiler" needs to process
to generate instantiated code.

So

(1) can a pre-compiled VSIPL++ library (such as the one here:
http://www.codesourcery.com/vsiplplusplus/1.3/download.html) be used
from Microsoft Visual C++ 2005 -- is MSVC 2005 "ISO C++ compliant"
enough?

(2) If so, then why can't VSIPL++ be /built/ using MSVC2005 -- assuming
that either individual licenses for IPP and MKL are obtained, or those
(presumably optional) elements are disabled when running configure.

(3) If not, the what is the point of using the /Qvc8 switch when
building VSIPL++ with the Intel C++ compiler? (Why worry about the MS
ABI if all clients of the library are required to icc anyway?)

Thanks,
Charles Wilson


From stefan at codesourcery.com  Fri Sep 28 18:05:25 2007
From: stefan at codesourcery.com (Stefan Seefeld)
Date: Fri, 28 Sep 2007 14:05:25 -0400
Subject: [vsipl++] VSIPL++ and Visual C++
In-Reply-To: <1B90921D44561C4A977E2604A5247075068508@mlbe2k4.cs.myharris.net>
References: <1B90921D44561C4A977E2604A5247075068508@mlbe2k4.cs.myharris.net>
Message-ID: <46FD4265.9000001@codesourcery.com>

Charles,

thanks for your interest into Sourcery VSIPL++.

Wilson, Charles (cwilso11) wrote:

> (1) can a pre-compiled VSIPL++ library (such as the one here:
> http://www.codesourcery.com/vsiplplusplus/1.3/download.html) be used
> from Microsoft Visual C++ 2005 -- is MSVC 2005 "ISO C++ compliant"
> enough?

You are right, building Sourcery VSIPL++ requires a fairly standard-compliant
C++ compiler, and -- as we had to find out -- MSVC2005 unfortunately isn't
good enough.
Since most of the complex code is part of (public) headers, it will be
exposed to the end-user's compiler, too. Thus, MSVC2005 most likely won't
be able to compile applications using Sourcery VSIPL++.

> (2) If so, then why can't VSIPL++ be /built/ using MSVC2005 -- assuming
> that either individual licenses for IPP and MKL are obtained, or those
> (presumably optional) elements are disabled when running configure.
> 
> (3) If not, the what is the point of using the /Qvc8 switch when
> building VSIPL++ with the Intel C++ compiler? (Why worry about the MS
> ABI if all clients of the library are required to icc anyway?)

This flag may indeed not be very meaningful in this context.

Thanks,
		Stefan

-- 
Stefan Seefeld
CodeSourcery
stefan at codesourcery.com
(650) 331-3385 x718