| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This is exactly the same memcpy()-in-hot-path xxHash problem, see
community/xxhash commit log for details.
The bug makes "lz4 -1" or "lz4 -t" twice slower on x86_64 however
"lz4 -9" is barely affected (do not use "lz4 -b" for benchmarking
this change since it does not calculate the checksum).
Other aport changes:
- use short and fast "make check" (unlike extensive "make test")
for pre-v7 ARM architectures.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is the same memcpy()-in-hot-path xxHash problem as described
in community/xxhash commit message.
The bug makes "zstd -1" or "zstd -t" ~15% slower on x86_64 though
higher levels (>= 9) are almost unaffected.
Other aport changes:
- make sure zlib/lzma/lz4 support is not compiled in automatically
(status quo, just make it explicit);
- re-enable armv7 since it is certainly well-supported by upstream,
was disabled by some script in the first place and should not be
affected by "Bus error" (i.e. some alignment problem) mentioned in
commit 2bae66907...
- use faster and shorter "make check" (unlike extensive "make test")
on arm* builders.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Yes, it is 20 times slower on x86* than it should be because xxhash.c always
uses "safe" memcpy()-based methods for unaligned memory access (XXH_readXX)
irregardless of input alignment due to x86-default XXH_FORCE_ALIGN_CHECK=0.
This ends up with real memcpy() calls in hot path (with -O2 too).
The bug affects Alpine x86* (not just edge, but at least 3.8 too -- i.e. this
is not something introduced in 0.6.5) for aligned and unaligned inputs. Other
architectures are severely affected for unaligned inputs only.
The fix lifts the XXH_FORCE_MEMORY_ACCESS=1 condition to enable XXH_readXX
methods based on __attribute__((__packed__)) usage everywhere except ARMv6
(which is covered by its own case earlier).
This is safe and fast because the compiler will either:
- use direct storage access instructions on capable architectures such as
aarch64, armv7, ppc64le, s390x, x86* irregardless of input alignment;
- or use relatively fast LWL/LWR instructions on mips* with unaligned input;
- or use byte load/stores and shifts/ors on armel with unaligned input which
is still faster then memcpy() call.
All aports that use xxhash.c are likely affected. For example, community/zstd
suffers too though not so grave (~15% difference for "zstd -t" on big archive)
and main/lz4 is twice slower on basic compression levels.
Other aport changes:
- modernize;
- enable check(); it is short and fast so suitable for slow builders too.
The python part is left intact though newer version exists.
|
| |
|
|
|
|
|
|
|
|
| |
Provide some basic mips* support just to allow the build succeed.
Disabling on mips* is not an option cause the arch= list becomes
too long and ugly.
Do not bump pkgrel since no previously built mips* apks exist and
change does nothing for other architectures.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
While at it, drop depend that was not needed after all
|
| |
|
|
| |
needed by diffoscope
|
| | |
|
| |
|
|
|
| |
1) Disable use of CFC1 (FPU instruction) on all soft-float mips*.
2) Enable MIPS32 ISA (default is MIPS III) on 32-bit mips{el}.
|
| |
|
|
|
| |
1) Disable use of CFC1 (FPU instruction) on all soft-float mips*.
2) Enable MIPS32 ISA (default is MIPS III) on 32-bit mips{el}.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
./configure tries to update CFLAGS/LIBS with LIBNL3_CFLAGS/LIBS using '+=':
checking for LIBNL3... yes
./configure: line 14545: CFLAGS+= -I/usr/include/libnl3 : not found
./configure: line 14546: LIBS+= -lnl-genl-3 -lnl-3 : not found
so later build fails without proper libnl3 flags:
nbd-client.c:51:10: fatal error: netlink/netlink.h: No such file or directory
#include <netlink/netlink.h>
nbd-3.18 suffers from the same problem.
|
| | |
|
| |
|
|
|
|
|
| |
* disable opencv (raw and tiff options depends on opencv)
After ffmpeg update to 4+, opencv is broken and unless updated it can't
be built right now.
|
| |
|
|
|
| |
https://github.com/tklauser/llmnrd
Link-Local Multicast Resolution (LLMNR) Daemon
|
| | |
|
| | |
|
| |
|
|
|
| |
Re-organize CARCH check for brevity (so match armel as well)
and use mipsel* not mipsel to cover out-of-tree mipseln8hf.
|
| |
|
|
|
|
|
|
| |
Build fails at switch_mips_unix.h with "$fp cannot be used in asm here"
because of forced -O0 (thus without -fomit-frame-pointer) in APKBUILD.
As a quick fix, use the "normal" C*FLAGS on mips* only. Test still
passes on mipsel at least.
|
| |
|
|
|
|
| |
Use sgidefs.h from linux-headers and replace _ABI64 references with non-GCC
counterpart (_ABI64 is only defined with -mabi=64 thus -Wundef errors arise
in some dependent projects with -mabi=32).
|
| |
|
|
|
|
|
| |
1) Disable parallel build for "make check" to prevent sporadic test failures
(at least on relatively old x86_64 and dual-core 1200 MHz mipsel).
2) Pass DBM version explicitly (--with-dbm=db53) to avoid slow DB version search.
The package already depends on main/db which persists at 5.3.x since 2012.
|
| | |
|
| |
|
|
| |
Insert package version with sed instead of relying on 'git describe'.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Use the generic gcc/clang/suncc __BYTE_ORDER__ macro to detect platform
byte order instead of hard-coded architecture detection which is wrong
for ppc64le (always assumes POWER is big-endian) and lacks any knowledge
of mips*.
This change covers s390x too so remove s390x-support.patch.
|
| |
|
|
|
|
|
| |
Let build continue when git is not found. No changes in behaviour
because all attempts to use git (to get project revision etc) fail
due to missing .git folder in the source tarball and project falls
back to appversion.default in any case.
|
