lp:fftw3

Created by dobey on 2016-01-08 and last modified on 2017-10-10
Get this branch:
bzr branch lp:fftw3

Branch merges

Related bugs

Related blueprints

Branch information

Owner:
VCS imports
Project:
FFTW
Status:
Development

Import details

Import Status: Reviewed

This branch is an import of the HEAD branch of the Git repository at https://github.com/FFTW/fftw3.git.

The next import is scheduled to run in 1 hour.

Last successful import was 4 hours ago.

Import started 4 hours ago on pear and finished 4 hours ago taking 20 seconds — see the log
Import started 10 hours ago on pear and finished 10 hours ago taking 20 seconds — see the log
Import started 16 hours ago on pear and finished 16 hours ago taking 25 seconds — see the log
Import started 22 hours ago on pear and finished 22 hours ago taking 25 seconds — see the log
Import started on 2017-10-21 on russkaya and finished on 2017-10-21 taking 40 seconds — see the log
Import started on 2017-10-21 on russkaya and finished on 2017-10-21 taking 40 seconds — see the log
Import started on 2017-10-21 on pear and finished on 2017-10-21 taking 20 seconds — see the log
Import started on 2017-10-20 on russkaya and finished on 2017-10-20 taking 40 seconds — see the log
Import started on 2017-10-20 on pear and finished on 2017-10-20 taking 25 seconds — see the log
Import started on 2017-10-20 on pear and finished on 2017-10-20 taking 20 seconds — see the log

Recent revisions

3022. By Matteo Frigo <email address hidden> on 2017-10-10

Ditch --enable-debug-malloc and --enable-debug-alignment

We wrote DEBUG_MALLOC in 1997 to debug memory leaks. Nowadays
DEBUG_MALLOC is just confusing. Better tools are available, and
DEBUG_MALLOC is not thread-safe and it does not respect SIMD
alignment. It confused at least one user.

In the gcc-2.SOMETHING days, gcc would allocate doubles on the stack
at 4-byte boundary (vs. 8) reducing performance by a factor of 3.
That's when we introduced --enable-debug-alignment, which is totally
obsolete by now.

3021. By Matteo Frigo <email address hidden> on 2017-09-29

Use armv7a cycle counter unconditionally if HAVE_ARMV7A_CNTVCT

It looks like __ARM_ARCH_7A__ is not always defined. If the
user says HAVE_ARMV7A_CNTVCT, trust the user.

3020. By Matteo Frigo <email address hidden> on 2017-09-22

Merge pull request #110 from junghans/cmake

Minor cmake fixes

3019. By Matteo Frigo <email address hidden> on 2017-09-21

simd-vsx.h: don't use vpermxor

It seems like gcc-6 generates incorrect code when using vpermxor
(tested with qemu emulator, so there is a chance that gcc is right and
qemu is wrong). Disable the use of vpermxor and do the simple thing
(one multiplication + one permutation).

3018. By Matteo Frigo <email address hidden> on 2017-09-21

Remove AC_FUNC_{MALLOC,REALLOC,MMAP}

They don't do what I thought. E.g., AC_FUNC_MALLOC checks that
malloc(0) returns NULL, and defines malloc to be rpl_malloc otherwise.
We don't support rpl_malloc() and we don't care about malloc(0).

3017. By Matteo Frigo <email address hidden> on 2017-09-21

Dead-Code Police

3016. By Matteo Frigo <email address hidden> on 2017-09-21

Fixup fftw3-mpi.h

fftw3-mpi.h must include "fftw3.h", not "api/fftw3.h", because both
fftw3-mpi.h and fftw3.h will ultimately be installed in /usr/include.

Thus, as a special exception, mpi/Makefile.am must specify the include
path -I $(top_srcdir)/api.

3015. By Matteo Frigo <email address hidden> on 2017-09-20

Generate codlist.c only when MAINTAINER_MODE

The user is not supposed to regenerate .c files. In addition, the
generation rule is subtly nonportable (it depends on whether or not
'#' can be escaped in Makefiles, an issue that does not appear
settled.)

3014. By Matteo Frigo <email address hidden> on 2017-09-20

Generate {dft,rdft}/simd/{sse,sse2,avx,...}/*.c only when MAINTAINER_MODE

Users are not supposed to generate them. Apart from that, the
generation rule uses '$*' in an explicit make rule, which is
technically a GNU extension. (Works with {open,free}bsd, but breaks
Solaris.)

3013. By Matteo Frigo <email address hidden> on 2017-09-20

Restore the ability to build out of tree.

Before 1f3704b9, we had "-I $(top_srcdir)/foo -I $(top_srcdir)/bar".
After 1f3704b9, we had no -I specification at all, but automake wants
an explicit -I $(top_srcdir) in order to build out of tree.

Branch metadata

Branch format:
Branch format 7
Repository format:
Bazaar repository format 2a (needs bzr 1.16 or later)
This branch contains Public information 
Everyone can see this information.

Subscribers