difficulties building PPL 0.11.2 without "--enable-shared"
Anthony Foiani
anthony.foiani@gmail.com
Wed May 18 21:29:00 GMT 2011
Yann, all --
On Wed, May 18, 2011 at 10:34 AM, Anthony Foiani
<anthony.foiani@gmail.com> wrote:
> Although, I suspect that a better answer (for the purposes of ct-ng,
> at least) is to simply limit the interface languages to C and C++. It
> might be worth experimenting to see if limiting to just C saves a few
> more seconds; I suspect that the C++ interface is basically "free",
> since that's the native language of the library, but who knows...
A quick test from earlier today (5 reps for each case, varying which
language interfaces were selected, and the "-j" make argument):
default / j1: 719s total, 143s avg
default / j2: 636s total, 127s avg
default / j6: 627s total, 125s avg
default / j8: 619s total, 123s avg
default / j12: 624s total, 124s avg
just_c / j1: 409s total, 81s avg
just_c / j2: 340s total, 68s avg
just_c / j6: 332s total, 66s avg
just_c / j8: 327s total, 65s avg
just_c / j12: 335s total, 67s avg
just_c_cpp / j1: 416s total, 83s avg
just_c_cpp / j2: 343s total, 68s avg
just_c_cpp / j6: 324s total, 64s avg
just_c_cpp / j8: 323s total, 64s avg
just_c_cpp / j12: 319s total, 63s avg
From this, I get:
1. It looks like it's almost twice as fast / half as long to build
only C and C++ instead of all the other interfaces (python, etc)
2. The parallel argument doesn't seem to make nearly as much of a
difference as I would have thought. (This system might be slightly
IO-bound for this, though.)
System spec:
$ sudo hdparm -t -T /dev/sda6
/dev/sda6:
Timing cached reads: 17922 MB in 2.00 seconds = 8978.38 MB/sec
Timing buffered disk reads: 244 MB in 3.01 seconds = 81.03 MB/sec
$ grep 'model name' /proc/cpuinfo | uniq -c
8 model name : Intel(R) Core(TM) i7 CPU 930 @ 2.80GHz
Will finish putting together that patchset later this evening.
Best regards,
Tony
--
For unsubscribe information see http://sourceware.org/lists.html#faq
More information about the crossgcc
mailing list