#12830: Work around GCC 4.7.0 bug on ia64 and improve the GMP-ECM spkg
-------------------------------------------------------------------------------------------------------------------------------------------------+
Reporter: leif
|
Owner: leif
Type: enhancement
|
Status: needs_review
Priority: blocker
|
Milestone: sage-5.0
Component: packages
|
Resolution:
Keywords: spkg -march=native assembler error Darwin MacOS __GMP_CFLAGS
__MPIR_CFLAGS gmp.h GCC 4.7.0 ia64 Itanium bug impossible reload | Work
issues:
Report Upstream: N/A
|
Reviewers: Jeroen Demeyer
Authors: Leif Leonhardy
| Merged
in:
Dependencies:
|
Stopgaps:
-------------------------------------------------------------------------------------------------------------------------------------------------+
Changes (by leif):
* status: needs_work => needs_review
Old description:
> GCC 4.7.0 is broken on ia64 (Itanium), which also affects the GMP-ECM
> spkg (cf. #12751).
>
> ----
>
> Adding `-march=native` to `CFLAGS` may lead to assembler errors, e.g. on
> MacOS X with newer GCCs on newer CPUs (e.g. such supporting AVX, which
> Apple's assembler currently doesn't).
>
> Newer versions of MPIR don't define `__GMP_CFLAGS` (in `gmp.h`) to a
> ''string literal'', but instead to a ''preprocessor macro''
> (`__MPIR_CFLAGS`), which in turn is defined to the string we want.
>
> ----
>
> '''New spkg:'''
> [http://boxen.math.washington.edu/home/leif/Sage/spkgs/ecm-6.3.p6.spkg]
>
> '''md5sum:''' `ef29a2f1a4165bc3efc37ae7981f2b90 ecm-6.3.p6.spkg`
>
> === ecm-6.3.p6 (Leif Leonhardy, April 16th 2012) ===
> * #12830: Add a work-around for GCC 4.7.x on ia64 (Itanium), since GMP-
> ECM
> currently won't build with that and anything but `-O0` on that
> platform.
> * Use `\{1,\}` instead of `\+` in `sed` patterns, which is more
> portable.
> * Also support newer system-wide MPIR installations for printing their
> settings.
> * Use `patch` to apply patches. Since the pre-patched `configure` in
> `patches/` was created with a newer version of autotools (or, rather,
> the
> original `configure` was created with an outdated version), the patch
> would
> have been almost as large as the patched `configure` file itself.
> Hence
> I `autoreconf`ed the source tree with a patched `configure.in` (and
> almost
> the latest versions of autotools), then created a patch to `configure`
> from the resulting file(s). Note that therefore `src/` isn't really
> vanilla
> any more, although just the auto-generated files differ (which are
> still
> made from vanilla upstream sources, including `configure.in`).
> Add a "Patches" subsection and update "Special Update/Build
> Instructions".
> Remove files in `patches/` from `.hgignore` (and also remove pre-
> patched
> files from that directory); the patch to `configure` and the diff of
> `configure.in` are [now] under revision control, which IMHO makes
> sense.
> * Beautify (and simplify) the output with respect to options passed to
> `configure`; print the settings of a few more environment variables
> that
> GMP-ECM uses (in case they're set); add some messages, also mention
> `--enable-assert` etc. if `SAGE_DEBUG=yes`.
> * Remove unused test for GCC.
>
> === ecm-6.3.p5 (Leif Leonhardy, April 11th 2012) ===
> * #12830: Don't add `-march=native` if the assembler doesn't understand
> the
> instructions the compiler emits with that. (E.g. the Apple/XCode
> assembler
> on Darwin doesn't yet support AVX.)
> * Fix extraction of `__GMP_CC` and `__GMP_CFLAGS` (from `gmp.h`) for
> newer
> versions of MPIR, which define these to preprocessor macros rather
> than
> strings.
> * Redirect warnings and error messages to `stderr`; add some messages.
> * Correct `SPKG.txt` w.r.t. applied patches.
New description:
GCC 4.7.0 is broken on ia64 (Itanium), which also affects the GMP-ECM spkg
(cf. #12751).
----
Adding `-march=native` to `CFLAGS` may lead to assembler errors, e.g. on
MacOS X with newer GCCs on newer CPUs (e.g. such supporting AVX, which
Apple's assembler currently doesn't).
Newer versions of MPIR don't define `__GMP_CFLAGS` (in `gmp.h`) to a
''string literal'', but instead to a ''preprocessor macro''
(`__MPIR_CFLAGS`), which in turn is defined to the string we want.
----
'''New spkg:'''
[http://boxen.math.washington.edu/home/leif/Sage/spkgs/ecm-6.3.p7.spkg]
'''md5sum:''' `f171b4e4ca52dccf8da4f24aabfce394 ecm-6.3.p7.spkg`
=== ecm-6.3.p7 (Leif Leonhardy, April 19th 2012) ===
* #12830: Add `-m[no-]power*` to the processor-specific options
recognized;
these are used to enable/disable specific instruction set extensions of
the
POWER and PowerPC (PPC) CPUs.
=== ecm-6.3.p6 (Leif Leonhardy, April 16th 2012) ===
* #12830: Add a work-around for GCC 4.7.x on ia64 (Itanium), since GMP-
ECM
currently won't build with that and anything but `-O0` on that
platform.
* Use `\{1,\}` instead of `\+` in `sed` patterns, which is more portable.
* Also support newer system-wide MPIR installations for printing their
settings.
* Use `patch` to apply patches. Since the pre-patched `configure` in
`patches/` was created with a newer version of autotools (or, rather,
the
original `configure` was created with an outdated version), the patch
would
have been almost as large as the patched `configure` file itself.
Hence
I `autoreconf`ed the source tree with a patched `configure.in` (and
almost
the latest versions of autotools), then created a patch to `configure`
from the resulting file(s). Note that therefore `src/` isn't really
vanilla
any more, although just the auto-generated files differ (which are
still
made from vanilla upstream sources, including `configure.in`).
Add a "Patches" subsection and update "Special Update/Build
Instructions".
Remove files in `patches/` from `.hgignore` (and also remove pre-
patched
files from that directory); the patch to `configure` and the diff of
`configure.in` are [now] under revision control, which IMHO makes
sense.
* Beautify (and simplify) the output with respect to options passed to
`configure`; print the settings of a few more environment variables
that
GMP-ECM uses (in case they're set); add some messages, also mention
`--enable-assert` etc. if `SAGE_DEBUG=yes`.
* Remove unused test for GCC.
=== ecm-6.3.p5 (Leif Leonhardy, April 11th 2012) ===
* #12830: Don't add `-march=native` if the assembler doesn't understand
the
instructions the compiler emits with that. (E.g. the Apple/XCode
assembler
on Darwin doesn't yet support AVX.)
* Fix extraction of `__GMP_CC` and `__GMP_CFLAGS` (from `gmp.h`) for
newer
versions of MPIR, which define these to preprocessor macros rather than
strings.
* Redirect warnings and error messages to `stderr`; add some messages.
* Correct `SPKG.txt` w.r.t. applied patches.
--
Comment:
Replying to [comment:16 jdemeyer]:
> In your `case` statement:
> {{{
> -march=*|-mcpu=*|-mtune*)
> }}}
> add `-mpower*` which MPIR also uses.
Fixed in a p7. (I've added `-mpower*` and `-mno-power*`.)
--
Ticket URL: <http://trac.sagemath.org/sage_trac/ticket/12830#comment:18>
Sage <http://www.sagemath.org>
Sage: Creating a Viable Open Source Alternative to Magma, Maple, Mathematica,
and MATLAB
--
You received this message because you are subscribed to the Google Groups
"sage-trac" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to
[email protected].
For more options, visit this group at
http://groups.google.com/group/sage-trac?hl=en.
