[PATCH] gnu: cross-base: Relax check for powerpc64le

The canonical triplet for powerpc64le-linux-gnu is actually

powerpc64le-unknown-linux-gnu, we should relax our matching here.

"powerpc64le-" prefix for "--with-long-double-128" instead of matching

full target.

---

gnu/packages/cross-base.scm | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)

LGTM!

But also wait for marusich's input.

Hi people,

Carl Dong <contact@carldong.me> writes:

This patch looks good to me. I confirmed it doesn't rebuild the world.

We can apply it to master. I will do this in the next few days unless

somebody else has further comments on this patch specifically.

While reviewing this patch, I noticed that surprisingly, we do not use

--with-long-double in the following places:

- The gcc package (a GCC v7.5.0 native compiler) on powerpc64le-linux

systems.

- The gcc-final package (a GCC v7.5.0 native compiler) on

powerpc64le-linux systems.

- The %bootstrap-gcc package (a GCC v5.5.0 native compiler) on

powerpc64le-linux systems.

- The %gcc-static package (a GCC v5.5.0 native compiler), when

cross-compiled for the powerpc64le-linux-gnu target on an x86_64-linux

system. (Note that this is actually the same GCC that on

powerpc64le-linux systems is known as %bootstrap-gcc.)

Obviously, %bootstrap-gcc is good enough to build gcc-final, and

gcc-final is good enough to build various things, including GNU Hello

and GNU Guix itself. We have verified this. However, Leo, do you think

it would be a good idea to add --with-long-double to the gcc and

gcc-final package definitions on powerpc64le-linux, anyway? Perhaps it

actually is needed, but we just haven't yet encountered the problems

that it would solve. What do you think, Leo?

You can inspect the packages to see this yourself, but you can also ask

Guix to tell you about the configure options. Here is what Guix says

about the configure options for the gcc and gcc-final packages on a

powerpc64le-linux system using commit

b0eb525ddba77420ff8c0930b9173b10f9ae342d:

And here is what Guix says about the configure options for %gcc-static,

when cross-compiled for powerpc64le-linux-gnu on an x86_64-linux system

using commit 662e7e28d576ada91fc9dec7d27c100666114f03, which is the one

we used to build the powerpc64le-linux bootstrap binaries:

Hello Chris,

I have no idea about cross compilers, but it would certainly be nice if

the final gcc supported long doubles.

As a check, you can have a look at the log file for building gmp,

obtained with "guix build --log-file gmp". If you grep for "long double",

it should contain a line

checking for long double... yes

Hm, just "having" long double is maybe not all we want.

From the gcc documentation at https://gcc.gnu.org/install/configure.html:

--with-long-double-128

Specify if long double type should be 128-bit by default on selected GNU/Linux architectures. If using --without-long-double-128, long double will be by default 64-bit, the same as double type. When neither of these configure options are used, the default will be 128-bit long double when built against GNU C Library 2.4 and later, 64-bit long double otherwise.

Of course we would like long double to have 128 bits, since otherwise there

is no difference with double. And at least with the current gcc and a glibc

from 2006 or later, the flag should not be neeed.

On my x86_64 machine, sizeof(long double) is 16 bytes with gcc@10, 5 or 7.

Could you try to compile and run the trivial program

#include <stdio.h>

void main () {

printf ("long double %i\n", sizeof (long double));

}

and see what it outputs on powerpc?

Well, while --with-long-double-128 should not be needed since it should be

the default, you may as well add it to the flags, since it is the desired

behaviour.

Andreas

On Thu, 2021-03-25 at 23:27 -0700, Chris Marusich wrote:

close 47349