Bug#1021115: unifrac-tools: Baseline violation on x86 and FTBFS everywhere else

2022-10-02 Thread Étienne Mollier
Adrian Bunk, on 2022-10-02:
> I was talking about a different change that would have been wrong:
> 
> -ifeq ($(PERFORMING_CONDA_BUILD),True)
>  CPPFLAGS += -mtune=generic
> -else
> -CPPFLAGS += -mfma -march=native
> -endif
> 
> 
> This would have fixed the baseline violation,
> but broken the build on non-x86 in a different way.

Right, I see your point.  True the -mtune= options are not valid
on several architectures.  Thanks for the clarification!

Have a nice day,  :)
-- 
Étienne Mollier 
Fingerprint:  8f91 b227 c7d6 f2b1 948c  8236 793c f67e 8f0d 11da
Sent from /dev/pts/2, please excuse my verbosity.


signature.asc
Description: PGP signature


Bug#1021115: unifrac-tools: Baseline violation on x86 and FTBFS everywhere else

2022-10-02 Thread Adrian Bunk
On Sun, Oct 02, 2022 at 05:13:33PM +0200, Étienne Mollier wrote:
> Hi Adrian,
> 
> Adrian Bunk, on 2022-10-02:
> > src/Makefile:
> > ...
> > ifeq ($(PERFORMING_CONDA_BUILD),True)
> > CPPFLAGS += -mtune=generic
> > else
> > CPPFLAGS += -mfma -march=native
> > endif
> > ...
> > 
> > 
> > 
> > Please remove this block, it is a baseline violation on x86
> > and causes FTBFS everywhere else.
> > 
> > Please don't use the PERFORMING_CONDA_BUILD case instead:
> > -mtune=generic is already the Debian default on x86,
> > but it is not supported on some/all(?) other architectures.
> 
> Done, the PERFORMING_CONDA_BUILD variable is specific to conda
> distribution and normally not related to enabling SIMD support.
> Touching it on the debian package end would also enable a lot of
> other bits which we probably don't need or want.

I was talking about a different change that would have been wrong:

-ifeq ($(PERFORMING_CONDA_BUILD),True)
 CPPFLAGS += -mtune=generic
-else
-CPPFLAGS += -mfma -march=native
-endif


This would have fixed the baseline violation,
but broken the build on non-x86 in a different way.


> Thanks for your report!
> 
> Have a nice day,  :)

cu
Adrian



Bug#1021115: unifrac-tools: Baseline violation on x86 and FTBFS everywhere else

2022-10-02 Thread Étienne Mollier
Hi Adrian,

Adrian Bunk, on 2022-10-02:
> src/Makefile:
> ...
> ifeq ($(PERFORMING_CONDA_BUILD),True)
> CPPFLAGS += -mtune=generic
> else
> CPPFLAGS += -mfma -march=native
> endif
> ...
> 
> 
> 
> Please remove this block, it is a baseline violation on x86
> and causes FTBFS everywhere else.
> 
> Please don't use the PERFORMING_CONDA_BUILD case instead:
> -mtune=generic is already the Debian default on x86,
> but it is not supported on some/all(?) other architectures.

Done, the PERFORMING_CONDA_BUILD variable is specific to conda
distribution and normally not related to enabling SIMD support.
Touching it on the debian package end would also enable a lot of
other bits which we probably don't need or want.

Thanks for your report!

Have a nice day,  :)
-- 
Étienne Mollier 
Fingerprint:  8f91 b227 c7d6 f2b1 948c  8236 793c f67e 8f0d 11da
Sent from /dev/pts/2, please excuse my verbosity.
On air: Life Line Project - The King - Doom


signature.asc
Description: PGP signature


Bug#1021115: unifrac-tools: Baseline violation on x86 and FTBFS everywhere else

2022-10-02 Thread Adrian Bunk
Source: unifrac-tools
Version: 1.1.1-1
Severity: serious
Tags: ftbfs
X-Debbugs-Cc: Debian Med Packaging Team 
, Étienne Mollier 


https://buildd.debian.org/status/fetch.php?pkg=unifrac-tools=armel=1.1.1-1=1664713094=0

...
h5c++ -Wdate-time -D_FORTIFY_SOURCE=2 -fopenmp -mfma -march=native -Wextra 
-Wno-unused-parameter -Wall  -std=c++11 -pedantic -I. -O4 -fPIC 
-L/<>/debian/tmp/usr/lib -c tree.cpp -o tree.o
arm-linux-gnueabi-g++: error: unrecognized command-line option ‘-mfma’
...


https://buildd.debian.org/status/fetch.php?pkg=unifrac-tools=amd64=1.1.1-2=1664716667=0

...
h5c++ -Wdate-time -D_FORTIFY_SOURCE=2 -fopenmp -mfma -march=native -Wextra 
-Wno-unused-parameter -Wall  -std=c++11 -pedantic -I. -O4 -fPIC 
-L/<>/debian/tmp/usr/lib -c tree.cpp -o tree.o
...


src/Makefile:
...
ifeq ($(PERFORMING_CONDA_BUILD),True)
CPPFLAGS += -mtune=generic
else
CPPFLAGS += -mfma -march=native
endif
...



Please remove this block, it is a baseline violation on x86
and causes FTBFS everywhere else.

Please don't use the PERFORMING_CONDA_BUILD case instead:
-mtune=generic is already the Debian default on x86,
but it is not supported on some/all(?) other architectures.