Skip optimizations for big-endian PowerPC.
authorMauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com>
Wed, 20 May 2020 14:42:23 +0000 (15:42 +0100)
committerSébastien Villemot <sebastien@debian.org>
Wed, 20 May 2020 14:42:23 +0000 (15:42 +0100)
commitf028b3d26be5376431ee5897fc8052a9ffa40be0
treef71bb09701d2844b7b4137e3e2b6c5990f0a64a0
parent8747ef8af80ba7e136c95817607503f94d9a2f5f
Skip optimizations for big-endian PowerPC.

Bug-Debian: https://bugs.debian.org/766695
Forwarded: http://sourceforge.net/p/math-atlas/patches/65/
Reviewed-by: Sébastien Villemot <sebastien@debian.org>
Last-Update: 2014-10-28

Some of the existing optimized files/cases for PowerPC
contain assembly instructions with implicit big-endian
behavior - thus incorrect for the little-endian mode -
incurring tests failures during the build (i.e., FTBFS).
This is being worked on; this is the workaround for now.

Note: this is an alternative for Michael Normand's patch
'atlas.3.10.2-ppc64le_do_not_use_files_with_lvx.patch',
in order not to touch file lists/indexes, but only C
source code, so the C preprocessor can avoid code per-arch.

For more details, see:
https://bugzilla.redhat.com/show_bug.cgi?id=1080073#c40

Gbp-Pq: Name ppc64el-ifdef-files-with-lvx.patch
tune/blas/gemm/CASES/ATL_cmm4x4x128_av.c
tune/blas/gemm/CASES/ATL_dmm4x4x80_ppc.c
tune/blas/gemm/CASES/ATL_dmm8x4x2_vsx.c
tune/blas/gemm/CASES/ATL_smm4x4x128_av.c
tune/blas/gemm/CASES/ATL_smm8x2x4_av.c