Re: [lmi] Compiling takes longer with gcc-4.9.2

[Vadim Zeitlin]

On Sun, 17 Jan 2016 17:51:45 +0000 Greg Chicares <address@hidden> wrote:

GC> On 2015-12-22 00:28, Greg Chicares wrote:
GC> > My local tree contains makefile changes to use MinGW-w64 gcc-4.9.2,
GC> > with '-std=c++11', some new warning flags, and various other
GC> > adjustments.
GC> 
GC> I'm almost ready to commit those changes. First, however, I tried them
GC> out on different hardware, and found a PCH surprise. Timings quoted
GC> from the 2015-12-22 original used old hardware; new timings reported
GC> today use new hardware unless explicitly indicated otherwise.
GC> 
GC>   old: 2 x E5520,      3 Gbps sata ii,  wd black WD2003FZEX, debian-7.7
GC>   new: 2 x #5-2530 v3, 6 Gpbs sata iii, samsung 850 pro 1TB, debian-7.9
               ^^^^^^^
 This is probably E5-2630.

GC> I expect the new hardware to perform uniformly better than the old.

 I would obviously expect it as well... Something seems to be really wrong
here, especially with

GC> > make --jobs=8 -f install_wx.make > ../log 2>&1
GC> > 2662.23s user 1594.59s system 475% cpu 14:55.38 total
GC> 
GC> make --jobs=8 -f install_wx.make > ../log 2>&1
GC> 2727.15s user 6098.59s system 214% cpu 1:08:39.33 total

this, of course. It takes twice less of the CPU time yet 4 times the wall
time? I just can't fathom this.

GC> 
http://svn.savannah.nongnu.org/viewvc/lmi/trunk/install_wx.make?root=lmi&r1=6443&r2=6463&diff_format=u
GC> +  --disable-precomp-headers \
GC> 
GC> Removing that one line, I measure the same build, twice:
GC> 
GC> make --jobs=8 -f install_wx.make > ../log 2>&1
GC> 1848.43s user 1444.34s system 391% cpu 14:01.53 total
GC> 1843.52s user 1446.63s system 389% cpu 14:05.59 total

 Even these numbers are very disappointing. After upgrading to a much
faster CPU (http://ark.intel.com/compare/83356,40200 shows that, in
addition to using a newer and improved architecture, it also has 2.5 times
as much of L3 cache and 1.5 times the memory bandwidth without speaking of
the obviously twice greater number of cores and 10% higher frequency) I'd
expect much more than a measly ~5% speedup.

GC> The bottleneck might even be the preprocessor.

 So the IO layer then? It looks plausible but this would just mean that KVM
doesn't do its job well. I know you don't want to use it but I'm really
curious to know how faster would the compilation inside the same VM be
inside VMware.

GC> New hardware makes a small difference for gcc-4.9.2, though it seemed
GC> to make no difference for gcc-3.4.5 .

 I'd like to see your numbers for cross-compiling. Normally they should be
significantly better than mine (and so an order of magnitude better than
what you have now), if this is not the case neither I'd seriously start
suspecting some hardware problem in the new machine.

 Regards,
VZ