Re: [eigen] Fwd: BenchTimer: small patch to avoid compiler warnings and

On Tue, Mar 2, 2010 at 3:25 PM, Eamon Nerbonne <emn13@xxxxxxxxxxxx> wrote:

I get the impression the previous mail never arrived. There seems to be something slightly odd going on with listmaster; it claims I'm subscribed with my normally primary email address eamon[at]..., but won't let me unsubscribe nor post. On the other hand, sending from this email address works.

In any case; the patch...

--eamon@xxxxxxxxxxxx - Tel#:+31-6-15142163

---------- Forwarded message ----------
From: Eamon Nerbonne <eamon@xxxxxxxxxxxxx>
Date: Tue, Mar 2, 2010 at 09:21
Subject: BenchTimer: small patch to avoid compiler warnings and add missing include
To: eigen@xxxxxxxxxxxxxxxxxxx

Hi everyone,

In the process of looking for a few perf. regressions, I came across two minor issues in BenchTimer.h:
- in windows, it defines the NOMINMAX and WIN32_LEAN_AND_MEAN. Those are common preprocessor symbols and may already be defined. To avoid warnings in gcc on windows, I wrapped them in an #ifndef.
- BenchTimer uses Vector2d but doesn't include Eigen/Core; so if you include BenchTimer before Eigen/Core (or without it), it won't compile. I added the appropriate include.

patch applied.

It also turns out gcc on windows throws a fit if you try to include Eigen/Core before BenchTimer.h; I'm not sure that's worth fixing (or even how to). What happens is that it complains about the various sse intrinsics; these are obviously being included by eigen, but they're also included by windows.h via BenchTimer. However, windows.h includes windef, which in turn has an extern "C" section that includes the intrinsics. GCC then complains that this new definition uses "C" linkage rather than the previous C++ linkage, eg:

c:\gcc\equation443\bin\../lib/gcc/x86_64-pc-mingw32/4.4.3/include/emmintrin.h:234: error: previous declaration of 'double __vector__ _mm_add_sd(double __vector__, double __vector__)' with 'C++' linkage
c:\gcc\equation443\bin\../lib/gcc/x86_64-pc-mingw32/4.4.3/../../../../x86_64-pc-mingw32/include/intrin.h:599: error: conflicts with new declaration with 'C' linkage

The problem is that the intrinsics are included twice from two different header files. That's really strange. We should check the file c:\gcc\equation443\bin\../lib/

gcc/x86_64-pc-mingw32/4.4.3/../../../../x86_64-pc-mingw32/include/intrin.h to see if this can be can be avoided via some defines...

gael.

--eamon@xxxxxxxxxxxx - Tel#:+31-6-15142163