[Buildroot] [PATCH 1/1] package/openblas: fix some x86 32bit builds

[Buildroot] [PATCH 1/1] package/openblas: fix some x86 32bit builds

[Julien Olivain]

OpenBLAS Makefile normally needs to explicitly define the CPU
architecture size (32 or 64bit) in the BINARY macro. See [1].

When an architecture supports both 64 and 32bit, the 32bit support
is sometimes implemented in OpenBLAS by overriding a fallback to an
anterior architecture. For example, if the build target architecture
is x86 Haswell 32bit, OpenBLAS build will override the arch to
Nehalem. See [2].

If the BINARY macro is undefined, the 32bit fallback will not happen,
sometimes leading to a link failure, with output:

    i686-buildroot-linux-gnu/bin/ar: strmm_kernel_LN.o: No such file or directory

This commit fixes those issues by explicitly defining the BINARY
macro. This issue has also been discussed upstream in [3] and [4].

Note: this issue was not introduced recently ([3] dates back from 2015),
and was also see in previous package version, for example in [5].

Fixes:
http://autobuild.buildroot.net/results/e1e/e1e2034a78799abe1bd28b036fa6f7d13322e42f

[1] https://github.com/xianyi/OpenBLAS/blob/v0.3.24/Makefile.rule#L50
[2] https://github.com/xianyi/OpenBLAS/blob/v0.3.24/Makefile.system#L113
[3] https://github.com/xianyi/OpenBLAS/issues/657
[4] https://github.com/xianyi/OpenBLAS/issues/1106
[5] http://autobuild.buildroot.net/results/5cd/5cdccd106b1de275ac75c39783e536107a31651f

Signed-off-by: Julien Olivain <ju.o@free.fr>
---
Patch tested on branch master at commit 3557a7b with commands:

    make check-package
    ...
    0 warnings generated

    utils/test-pkg -a -p openblas
    ...
    45 builds, 25 skipped, 0 build failed, 0 legal-info failed, 0 show-info failed

    support/testing/run-tests \
        -d dl -o output_folder \
        tests.package.test_openblas
    ...
    OK
---
 package/openblas/openblas.mk | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/package/openblas/openblas.mk b/package/openblas/openblas.mk
index 0d0e8e56dd..aacfc74b53 100644
--- a/package/openblas/openblas.mk
+++ b/package/openblas/openblas.mk
@@ -49,6 +49,12 @@ ifeq ($(BR2_STATIC_LIBS),y)
 OPENBLAS_MAKE_OPTS += NO_SHARED=1
 endif
 
+ifeq ($(BR2_ARCH_IS_64),y)
+OPENBLAS_MAKE_OPTS += BINARY=64
+else
+OPENBLAS_MAKE_OPTS += BINARY=32
+endif
+
 # binutils version <= 2.23.2 has a bug
 # (https://sourceware.org/bugzilla/show_bug.cgi?id=14887) where
 # whitespaces in ARM register specifications such as [ r1, #12 ] or [
-- 
2.41.0

_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot

Re: [Buildroot] [PATCH 1/1] package/openblas: fix some x86 32bit builds

[Yann E. MORIN]

Julien, All,

On 2023-09-15 21:33 +0200, Julien Olivain spake thusly:
> OpenBLAS Makefile normally needs to explicitly define the CPU
> architecture size (32 or 64bit) in the BINARY macro. See [1].
> 
> When an architecture supports both 64 and 32bit, the 32bit support
> is sometimes implemented in OpenBLAS by overriding a fallback to an
> anterior architecture. For example, if the build target architecture
> is x86 Haswell 32bit, OpenBLAS build will override the arch to
> Nehalem. See [2].
> 
> If the BINARY macro is undefined, the 32bit fallback will not happen,
> sometimes leading to a link failure, with output:
> 
>     i686-buildroot-linux-gnu/bin/ar: strmm_kernel_LN.o: No such file or directory
> 
> This commit fixes those issues by explicitly defining the BINARY
> macro. This issue has also been discussed upstream in [3] and [4].
> 
> Note: this issue was not introduced recently ([3] dates back from 2015),
> and was also see in previous package version, for example in [5].
> 
> Fixes:
> http://autobuild.buildroot.net/results/e1e/e1e2034a78799abe1bd28b036fa6f7d13322e42f
> 
> [1] https://github.com/xianyi/OpenBLAS/blob/v0.3.24/Makefile.rule#L50
> [2] https://github.com/xianyi/OpenBLAS/blob/v0.3.24/Makefile.system#L113
> [3] https://github.com/xianyi/OpenBLAS/issues/657
> [4] https://github.com/xianyi/OpenBLAS/issues/1106
> [5] http://autobuild.buildroot.net/results/5cd/5cdccd106b1de275ac75c39783e536107a31651f
> 
> Signed-off-by: Julien Olivain <ju.o@free.fr>

Thanks *a lot* for this very detailed and convincing explanation, with
appropriate references. That's exactly the kind of information that
increases confidence in a patch! 👍

Applied to master, thanks.

Regards,
Yann E. MORIN.

> ---
> Patch tested on branch master at commit 3557a7b with commands:
> 
>     make check-package
>     ...
>     0 warnings generated
> 
>     utils/test-pkg -a -p openblas
>     ...
>     45 builds, 25 skipped, 0 build failed, 0 legal-info failed, 0 show-info failed
> 
>     support/testing/run-tests \
>         -d dl -o output_folder \
>         tests.package.test_openblas
>     ...
>     OK
> ---
>  package/openblas/openblas.mk | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/package/openblas/openblas.mk b/package/openblas/openblas.mk
> index 0d0e8e56dd..aacfc74b53 100644
> --- a/package/openblas/openblas.mk
> +++ b/package/openblas/openblas.mk
> @@ -49,6 +49,12 @@ ifeq ($(BR2_STATIC_LIBS),y)
>  OPENBLAS_MAKE_OPTS += NO_SHARED=1
>  endif
>  
> +ifeq ($(BR2_ARCH_IS_64),y)
> +OPENBLAS_MAKE_OPTS += BINARY=64
> +else
> +OPENBLAS_MAKE_OPTS += BINARY=32
> +endif
> +
>  # binutils version <= 2.23.2 has a bug
>  # (https://sourceware.org/bugzilla/show_bug.cgi?id=14887) where
>  # whitespaces in ARM register specifications such as [ r1, #12 ] or [
> -- 
> 2.41.0
> 
> _______________________________________________
> buildroot mailing list
> buildroot@buildroot.org
> https://lists.buildroot.org/mailman/listinfo/buildroot

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 561 099 427 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot

Re: [Buildroot] [PATCH 1/1] package/openblas: fix some x86 32bit builds

[Peter Korsgaard]

>>>>> "Julien" == Julien Olivain <ju.o@free.fr> writes:

 > OpenBLAS Makefile normally needs to explicitly define the CPU
 > architecture size (32 or 64bit) in the BINARY macro. See [1].

 > When an architecture supports both 64 and 32bit, the 32bit support
 > is sometimes implemented in OpenBLAS by overriding a fallback to an
 > anterior architecture. For example, if the build target architecture
 > is x86 Haswell 32bit, OpenBLAS build will override the arch to
 > Nehalem. See [2].

 > If the BINARY macro is undefined, the 32bit fallback will not happen,
 > sometimes leading to a link failure, with output:

 >     i686-buildroot-linux-gnu/bin/ar: strmm_kernel_LN.o: No such file or directory

 > This commit fixes those issues by explicitly defining the BINARY
 > macro. This issue has also been discussed upstream in [3] and [4].

 > Note: this issue was not introduced recently ([3] dates back from 2015),
 > and was also see in previous package version, for example in [5].

 > Fixes:
 > http://autobuild.buildroot.net/results/e1e/e1e2034a78799abe1bd28b036fa6f7d13322e42f

 > [1] https://github.com/xianyi/OpenBLAS/blob/v0.3.24/Makefile.rule#L50
 > [2] https://github.com/xianyi/OpenBLAS/blob/v0.3.24/Makefile.system#L113
 > [3] https://github.com/xianyi/OpenBLAS/issues/657
 > [4] https://github.com/xianyi/OpenBLAS/issues/1106
 > [5] http://autobuild.buildroot.net/results/5cd/5cdccd106b1de275ac75c39783e536107a31651f

 > Signed-off-by: Julien Olivain <ju.o@free.fr>

Committed to 2023.02.x, 2023.05.x and 2023.08.x, thanks.

-- 
Bye, Peter Korsgaard
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot