According to [1] sections 7.4, 8.1, and 8.2, desktop and server
processors support fast unaligned access, but embedded systems likely
don't.
It's important that TUKLIB_FAST_UNALIGNED_ACCESS isn't defined when
-mstrict-align is in use because it will result in slower binaries
even if running on a processor that supports fast unaligned access.
It's because compilers will translate multibyte memcpy() to multiple
byte-by-byte instructions instead of wider loads and stores. The
compression times from [2] show this well:
Unaligned access CFLAGS Compression time
enabled -O2 -mno-strict-align 66.1 s
disabled -O2 -mno-strict-align 79.5 s
disabled -O2 -mstrict-align 79.9 s
enabled -O2 -mstrict-align 129.1 s
There currently (GCC 15.2) is no preprocessor macro on LoongArch
to detect if -mstrict-align or -mno-strict-align is in effect (the
default is -mno-strict-align). Use heuristics to detect which of the
flags is in effect.
[1] https://github.com/loongson/la-softdev-convention/blob/v0.2/la-softdev-convention.adoc
[2] https://github.com/tukaani-project/xz/pull/186#issuecomment-3494570304
Thanks-to: Li Chenggang <lichenggang@deepin.org>
Thanks-to: Xi Ruoyao
See: https://github.com/tukaani-project/xz/pull/186
PowerPC64LE wasn't tested but it seems like a safe change.
POWER8 supports unaligned access in little endian mode. Testing
on godbolt.org shows that GCC uses unaligned access by default.
The RISC-V macro __riscv_misaligned_fast is very new and not
in any stable compiler release yet.
Documentation in INSTALL was updated to match.
Documentation about an autodetection bug when using ARM64 GCC
with -mstrict-align was added to INSTALL.
CMake files weren't updated yet.
Add a configure option --enable-unsafe-type-punning to get the
old non-conforming memory access methods. It can be useful with
old compilers or in some other less typical situations but
shouldn't normally be used.
Omit the packed struct trick for unaligned access. While it's
best in some cases, this is simpler. If the memcpy trick doesn't
work, one can request unsafe type punning from configure.
Because CRC32/CRC64 code needs fast aligned reads, if no very
safe way to do it is found, type punning is used as a fallback.
This sucks but since it currently works in practice, it seems to
be the least bad option. It's never needed with GCC >= 4.7 or
Clang >= 3.6 since these support __builtin_assume_aligned and
thus fast aligned access can be done with the memcpy trick.
Other things:
- Support GCC/Clang __builtin_bswapXX
- Cleaner bswap fallback macros
- Minor cleanups
This replaces bswap.h and integer.h.
The tuklib module uses <byteswap.h> on GNU,
<sys/endian.h> on *BSDs and <sys/byteorder.h>
on Solaris, which may contain optimized code
like inline assembly.
