Source: openblas Severity: normal X-Debbugs-Cc: henr...@debian.org Dear Maintainer,
I've found some people complain about Ubuntu/Debian's openblas package via twitter and they rebuild it from source, since it seem that not work as expected with SMP (see http://verifiedby.me/adiary/060 (In Japanese)). I'm not familiar with this issue, but it may be worth to check, IMHO. -- Hideki Yamane <henr...@iijmio-mail.jp>
>From f0dab10eaf513857293b9db9ed4dc46bc4fae5ee Mon Sep 17 00:00:00 2001 From: Hideki Yamane <henr...@debian.org> Date: Fri, 1 May 2020 02:24:49 +0900 Subject: [PATCH] add 0007-synchronize-FP-CSR-between-threads.patch --- ...7-synchronize-FP-CSR-between-threads.patch | 21 +++++++++++++++++++ debian/patches/series | 1 + 2 files changed, 22 insertions(+) create mode 100644 debian/patches/0007-synchronize-FP-CSR-between-threads.patch diff --git a/debian/patches/0007-synchronize-FP-CSR-between-threads.patch b/debian/patches/0007-synchronize-FP-CSR-between-threads.patch new file mode 100644 index 0000000..5878acc --- /dev/null +++ b/debian/patches/0007-synchronize-FP-CSR-between-threads.patch @@ -0,0 +1,21 @@ +From: Hideki Yamane <henr...@debian.org> +Date: Fri, 1 May 2020 02:12:56 +0900 +Subject: synchronize FP CSR between threads + +--- + Makefile.rule | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/Makefile.rule b/Makefile.rule +index a4465e4..2543daa 100644 +--- a/Makefile.rule ++++ b/Makefile.rule +@@ -208,7 +208,7 @@ NO_AFFINITY = 1 + # DEVICEDRIVER_ALLOCATION = 1 + + # If you need to synchronize FP CSR between threads (for x86/x86_64 only). +-# CONSISTENT_FPCSR = 1 ++CONSISTENT_FPCSR = 1 + + # If any gemm argument m, n or k is less or equal this threshold, gemm will be execute + # with single thread. (Actually in recent versions this is a factor proportional to the diff --git a/debian/patches/series b/debian/patches/series index 69b41f9..dd1d3e7 100644 --- a/debian/patches/series +++ b/debian/patches/series @@ -4,3 +4,4 @@ remove-openmp-warning.patch no-embedded-lapack.patch shared-blas-lapack.patch matgen-symbols-not-included.patch +0007-synchronize-FP-CSR-between-threads.patch -- 2.26.2