[lfs-dev] Oddities while testing.

Ken Moffat zarniwhoop at ntlworld.com
Sat Jun 3 18:38:00 PDT 2017


On Sat, Jun 03, 2017 at 07:56:39PM -0400, Alain Toussaint wrote:
> > First, it is really good to know of people who build svn regularly.
> 
> My futur plans is to configure jenkins or another kind of buildbot to
> do it unattended on a spare computer.
> 
> >  And glibc built without
> > problems, but perhaps that is because I pass CFLAGS and CXXFLAGS of
> > "-O2".
> 
> I do use CFLAGS and CXXFLAGS for the usb that are generic enough for
> x86_64. This one build is using -Os and -mtune=generic because I am
> building for any x86_64 computers (at work and at home) with one usb
> key targeting UEFI bioses (using the refind
> http://www.rodsbooks.com/refind/ bootloader), the other one targeting
> older bioses but essentially, I'm using ramdrives for the root image,
> thus the -Os build.
> 

I think that -Os might be iffy in gcc-7.1.  I'm not sure what
firefox's default is, but in the past parts have been -Os and for
gcc-7.1 we have to force -O2 to get all of it to work.

> Other flags used are the minimum sse flags featured by the oldest
> x86_64 chips ever built (amd athlon and intel pentium 4 64 bits) which
> is sse and sse2. From the documentation given by gcc (source:
> https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gcc/x86-Options.html#x86-Options):
> 
> "For the x86-32 compiler, you must use -march=cpu-type, -msse or
> -msse2 switches to enable SSE extensions and make this option
> effective. For the x86-64 compiler, these extensions are enabled by
> default."
> 
> "The resulting code should be considerably faster in the majority of
> cases and avoid the numerical instability problems of 387 code, but
> may break some existing code that expects temporaries to be 80 bits"
> 
> Of the many years that I used gentoo (since year 2004) before and
> since last year, LFS, this has never been the case and gcc always
> default to the 387 for floating point instructions and data.
> 
> To figure out the options used by default for the compiler:
> 
> gcc -c -Q --march=native --target=help
> 
> or in my case: gcc -c -Q --march=generic --target=help
> 

Hmm, on this build -march= (one -) with --target-help only gives me
the same output as --target-help.

> the set of CFLAGS:
> 
> "-Os -pipe -m128bit-long-double -m80387 -m64 -malign-data=cacheline
> -malign-double -mtune=generic -mfancy-math-387 -mfentry
> -mfp-ret-in-387 -mfpmath=sse -mglibc -mhard-float -mieee-fp -msse
> -msse2 -mincoming-stack-boundary=4 -mpreferred-stack-boundary=4
> -mno-sse4 -mred-zone -msseregparm"
> 
> Some of these are indeed redundant and I need to learn how to read and
> dissect gcc source code for my desired effect (pad to 16 bytes
> alignments, use sse units for math and zerofill 80 bit float to 128
> bit and finally, use all registers available).
> 
> > I don't build systemd, but I _had_ been using gold for most of my
> > builds in the last year (newer, shinier, claimed to be faster).
> 
> I never intended to use gold either but the configure script for
> systemd-233 look for it and build with it. In binutils 2.25, that
> worked fine but in 2.28, there was a regression in the code and a
> number of projects had build failures which was fixed in binutils svn.
> I'll be attempting a build later on to document it but at the moment,
> I stopped at the glibc build in chapter 6 which has more failures
> compared to my usual build targeting this computer (usually, I have
> only 2 tests that fail with a ton more CFLAGS but for this build, I
> have 352 FAIL in the testlogs):
> 
> A short sample (log file available on request):
> 
> FAIL: conform/ISO/assert.h/linknamespace
> FAIL: conform/ISO/locale.h/linknamespace
> FAIL: conform/ISO/math.h/linknamespace
> FAIL: conform/ISO/signal.h/linknamespace
> FAIL: conform/ISO/stdio.h/linknamespace
> FAIL: conform/ISO/stdlib.h/linknamespace
> FAIL: conform/ISO/string.h/linknamespace
> 
> I haven't installed the resulting glibc build but will rebuild again
> without any CFLAGS.
> 
> All the rest of the software build and test usually fine with my set of CFLAGS.
> 
> Alain

For the first of those failures,
https://sourceware.org/bugzilla/show_bug.cgi?id=19463
(from March last year, but apparently with a commit in March).

ĸen
-- 
I live in a city. I know sparrows from starlings.  After that
everything is a duck as far as I'm concerned.  -- Monstrous Regiment


More information about the lfs-dev mailing list