Bug 568320 – dev-libs/gmp[pgo] fails to build on arm with error: unknown type name 'clock_gettime'

• Status: RESOLVED FIXED
• Product: Gentoo Linux

[Description tt_1 2015-12-15 10:02:01 UTC]

Created attachment 419264 [details]
compressed build.log

see attached build.log for further details.

[Comment 1 tt_1 2015-12-15 10:03:48 UTC]

Created attachment 419266 [details]
output of emerge --info

[Comment 2 Conrad Kostecki 2016-01-03 22:37:37 UTC]

+1 same problem here

[Comment 3 Conrad Kostecki 2016-01-05 18:58:46 UTC]

That error is caused by LDFLAGS="-Wl,--as-needed".

When "-Wl,--as-needed" is not set, this error is gone. But after that, I am still unable to compile dev-libs/gmp[pgo], because I am hitting a new error:

libtool: compile:  armv7a-hardfloat-linux-gnueabi-gcc -DHAVE_CONFIG_H -I. -I/var/tmp/portage/dev-libs/gmp-6.1.0/work/gmp-6.1.0/mpz -I.. -D__GMP_WITHIN_GMP -I/var/tmp/portage/dev-libs/gmp-6.1.0/work/gmp-6.1.0 -march=armv7ve -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mtune=cortex-a7 -O3 -mfloat-abi=hard -mthumb -pipe -fomit-frame-pointer -c /var/tmp/portage/dev-libs/gmp-6.1.0/work/gmp-6.1.0/mpz/bin_uiui.c  -fPIC -DPIC -o .libs/bin_uiui.o
In file included from /var/tmp/portage/dev-libs/gmp-6.1.0/work/gmp-6.1.0/mpz/array_init.c:32:0:
/var/tmp/portage/dev-libs/gmp-6.1.0/work/gmp-6.1.0/gmp-impl.h:1384:34: error: unknown type name ‘mp_srcptv’
 __GMP_DECLSPEC void      mpn_toom44_mul (mp_ptr, mp_srcptr, mp_size_t, mp_srcptr, mp_size_t, mp_ptr);
                                  ^

[Comment 4 . 2016-07-09 12:16:13 UTC]

fails for armv6 too

[Comment 5 Conrad Kostecki 2016-11-17 23:07:25 UTC]

Created attachment 453690 [details, diff]
tuneup-clock-gettime.patch

Placing that patch fixed this for me.

The cause is simple.
When USE="pgo" is set, the eBuild runs:
./tune/tuneup | tee gmp.mparam.h.new

Apparently, something does wrong and the string "clock_gettime is 1.000ns accurate" is beeing written to gmp-mparam.h.new (which is later moved to gmp-mparam.h)

[Comment 6 tt_1 2016-11-18 19:09:31 UTC]

Problem solved, thanks a lot! 

This is a bit off the topic, but are there any speedtests to see wether the use of pgo speeds up something in the end?

[Comment 7 . 2016-11-19 15:07:52 UTC]

(In reply to tt_1 from comment #6)
> This is a bit off the topic, but are there any speedtests to see wether the
> use of pgo speeds up something in the end?

It definitely does not. Which is also the reason why every major compiler includes this technology.

[Comment 8 SpanKY 2016-11-26 21:54:22 UTC]

thanks guys.  i've sent a patch upstream for the issue.  let's see what they have to say about it.

https://gmplib.org/list-archives/gmp-bugs/2016-November/004032.html

[Comment 9 SpanKY 2016-11-27 01:07:34 UTC]

upstream has merged the fix, so pushed on our side:
https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=02c897444d280edc7330204b9fb43afb0a52d0ba