The crc.w.{b/h/w/d}.w instructions in LoongArch can calculate the CRC32
result for 1/2/4/8 bytes in a single operation. Using these is much
faster compared to the generic method.
Optimized CRC32 is enabled unconditionally on 64-bit LoongArch because
the LoongArch specification says that CRC32 instructions shall be
implemented for 64-bit processors. Optimized CRC32 isn't enabled for
32-bit LoongArch processors because not enough information is available
about them.
Co-authored-by: Lasse Collin <lasse.collin@tukaani.org>
Closes: https://github.com/tukaani-project/xz/pull/86
This addresses the issue I mentioned in
6c095a98fb and speeds up the Valgrind
job a bit, because non-xz tools aren't run unnecessarily with
Valgrind by the script tests.
It shouldn't make any difference because LIBS should be empty
at that point in configure. But prepending is the correct way
because in general the libraries being added might require other
libraries that come later on the command line.
It's more robust in case the compiler allows pre-C99 implicit function
declarations. If an x86 intrinsic is missing and gets treated as
implicit function, the linking step will very probably fail. This
isn't the only way to workaround implicit function declarations but
it might be the simplest and cleanest.
The problem hasn't been observed in the wild.
There are a couple more AC_COMPILE_IFELSE uses in configure.ac.
Of these, Landlock check calls prctl() and in theory could have
the same problem. In practice it doesn't as the check program
looks for several other things too. However, it was changed to
AC_LINK_IFELSE still to look more correct.
Similarly, m4/tuklib_cpucores.m4 and m4/tuklib_physmem.m4 were
updated although they haven't given any trouble either. They
have worked all these years because those check programs rely
on specific headers and types: if headers or types are missing,
compilation will fail. Using the linker makes these checks more
similar to the ones in cmake/tuklib_*.cmake which always link.
AC_COMPILE_IFELSE needed -Werror because Clang <= 14 would merely
warn about the unsupported attribute and implicit function declaration.
Changing to AC_LINK_IFELSE handles the implicit declaration because
the symbol __crc32d is unlikely to exist in libc.
Note that the other part of the check is that #include <arm_acle.h>
must work. If the header is missing, most compilers give an error
and the linking step won't be attempted.
Avoiding -Werror makes the check more robust in case CFLAGS contains
warning flags that break -Werror anyway (but this isn't the only check
in configure.ac that has this problem). Using AC_LINK_IFELSE also makes
the check more similar to how it is done in CMakeLists.txt.
While the CMake support has gotten a lot less testing than
the Autotools-based build, the supported features should now
be equal. The output may differ slightly, for example,
liblzma.pc may have
Libs.private: -pthread -lpthread
with Autotools on GNU/Linux. CMake doesn't put any options
in Libs.private because on modern glibc the pthread functions
are in libc. The options options aren't required to link static
liblzma into an application.
Autotools-based build doesn't generate or install
lib/cmake/liblzma-*.cmake files. This means that on most
platforms one cannot rely on
find_package(liblzma 5.2.5 REQUIRED CONFIG)
or such finding those files.
It was weird to add CMAKE_THREAD_LIBS_INIT in CMAKE_REQUIRED_LIBRARIES
only if CLOCK_MONOTONIC is available. Alternative would be to remove
the thread libs from CMAKE_REQUIRED_LIBRARIES after the check for
pthread_condattr_setclock() but keeping the libs should be fine too.
Then it's ready in case more pthread functions were wanted some day.
In CMake, check_c_source_compiles() always links too. With
link-time optimization, unused functions may get omitted if
main() doesn't depend on them. Consider the following which
tries to check if somefunction() is available when <someheader.h>
has been included:
#include <someheader.h>
int foo(void) { return somefunction(); }
int main(void) { return 0; }
LTO may omit foo() completely because the program as a whole doesn't
need it and then the program will link even if the symbol somefunction
isn't available in libc or other library being linked in, and then
the test may pass when it shouldn't.
What happens if <someheader.h> doesn't declare somefunction()?
Shouldn't the test fail in the compilation phase already? It should
but many compilers don't follow the C99 and later standards that
prohibit implicit function declarations. Instead such compilers
assume that somefunction() exists, compilation succeeds (with a
warning), and then linker with LTO omits the call to somefunction().
Change the tests so that they are part of main(). If compiler accepts
implicitly declared functions, LTO cannot omit them because it has to
assume that they might have side effects and thus linking will fail.
On the other hand, if the functions/intrinsics being used are supported,
they might get optimized away but in that case it's fine because they
really are supported.
It is fine to use __attribute__((target(...))) for main(). At least
it works with GCC 4.9 to 14.1 on x86-64.
Reported-by: Sam James <sam@gentoo.org>
CC from environment is used to initialize CMAKE_C_COMPILER so
setting CMAKE_C_COMPILER explicitly isn't needed.
The syntax in ci_build.bash was broken in case one wished to put
spaces in CC.
Prefix ARM64_RUNTIME_DETECTION with CRC_ and reorder it to be with
the other ARM64-specific lines. That macro isn't used outside this
file.
ARM64 CLMUL implementation doesn't exist yet and thus CRC64_ARM64_CLMUL
isn't used anywhere yet.
It's not ideal that the single-letter CRC utility macros are here
as they pollute the namespace of the LZ encoder files. Those could
be moved their own crc_macros.h like they were in 5.2.x but in practice
this is fine enough already.
LZ encoder needs lzma_crc32_table[0] but otherwise those tables
are private to the CRC code. In contrast, the other things in
check.h are needed in several places.
Now runtime detection of CLMUL support can pick between the CLMUL and
the generic assembly implementations. Whatever overhead this has for
builds that omit CLMUL completely isn't important because builds for
any non-ancient system is likely to include the CLMUL code too.
Handle the CRC tables in crcXX_fast.c files because now these files
are built even when assembly code is used.
If 32-bit x86 assembly is enabled then it will always be built even
if compiler flags were such that CLMUL would be allowed unconditionally.
That is, runtime detection will be used anyway. This keeps the build
rules simpler.
In LZ encoder, build and use lzma_lz_hash_table[256] if CLMUL CRC
is used without runtime detection. Previously this wasn't needed
because crc32_table.c included the lzma_crc32_table[][] in the build
unless encoder support had been disabled. Including an 8 KiB table
was silly when only 1 KiB is actually used. So now liblzma is 7 KiB
smaller if CLMUL is enabled without runtime detection.
It feels clearer this way, and when support for external SHA-256
is added, this will keep the order of the library detection the
same as in configure.ac (check for pthreads before libmd) although
it shouldn't matter in practice.
This simplifies things a little. Building liblzma with VS2013 probably
still worked but building the command line tools was not supported.
Microsoft ended support for VS2013 on 2024-04.
Make the available options and their behavior match
--enable-symbol-versions in configure.ac.
Don't enable symbol versions on Linux if not using glibc. Previously
the generic variant was selected on Microblaze or if using NVHPC
without checking that libc is glibc.
Leave the cache variable to "auto" or "yes" if that was specified
instead of setting it to the autodetected value by default. A downside
is that one cannot easily see which variant the autodetection code
has selected. The same applies to XZ_SANDBOX and XZ_THREADS though.
Also clarify that "yes" will fail if no threading support is found.
If no threading is wanted, it has to be disabled manually.
configure.ac doesn't behave this way at the moment. Instead it
assumes pthreads to be present if not targeting Windows. If pthreads
actually are missing, the build fails later.
The list was copied from configure.ac and should be kept in sync.
(Pretend that the deleted comment in CMakeLists.txt didn't exist.)
There is no need to add equivalent of --enable-werror as CMake >= 3.24
supports -DCMAKE_COMPILE_WARNING_AS_ERROR=ON.
This is for consistency with 4c81c9611f
where \040 has to be used because \0x20F gets interpret at three hex
digits. Octals escapes are never longer than three digits.
Xi Ruoyao
Lasse Collin
Sam James