Previously, when fetching elements from Sleef_rempitabsp
or Sleef_rempitabdp for infinity/nan inputs, we would
hit outside the table. This would trigger a global buffer
overflow error when running any of the trig functions with
an address sanitizer.
Fixes issue #464, on multiple arch. Furthermore tester3advsimd and
tester3advsimdnofma start passing when library is build with
ASAN enabled.

Prefer use of __riscv_vcreate_*() intrinsics, where appropriate.  This
does away with piecemeal assembly of a tuple register types and ensures
that every component is defined.  It also gets rid of uninitialised
variable warnings.

__riscv_vcreate intrinsics aren't available in older toolchains but will
be available in version 1.0 of the intrinsics interface.  In the interim
include substitute macros which try to fill the gap as best as possible.

Use a regular pattern of cast-then-insert, and extract-then-cast when
dealing with tuple registers with mixed component types.

Simplify two vcreate macros in helperrvv.h, remove a few unused macros, etc..

Make the vmask types for single-precision and double-precision
interfaces compatible, so that we have fewer divergent functions.

Keep the vopmask types incompatible, because the underlying vbool*_t
types still encode some underlying assumptions about the number of lanes
involved.

Make the still-incompatible function names distinct between SP and DP,
and then reconcile these with the preprocessor at the end of the file so
that the caller still sees the expected operations.  This will help when
it comes to generating inline headers in a future revision.

Updated CHANGELOG.
Version number already bumped in CMakeLists.txt.

Copyright year and some build instructions.

Macs do not implement SVE yet and Darwin does not support it.

Add missing headers when needed.

Add SLEEF_ prefix to SLEEF-specific and user facing cmake options.
Do not overwrite BUILD_SHARED_LIBS anymore.
Udpate docs.

...on all targets

This test suite covers the usage of scalar symbols
with auto-vectorisation, to test the GNU ABI
symbols generated by SLEEF.
However, some failures are recorded when emulating
these tests using user mode QEMU, for a
cross-compile build for AArch64 and s390x.
This is because this test suite does not get called
with the EMULATOR command necessary.
This fix adds the emulator to the test command
which makes the tests pass when cross-compiling
and using user mode QEMU emulation.

AArch64 tests build with llvm cross-compiler on linux, run and pass.

PR #477 pointed out some test failures with qemu.
Re-enabling these tests in GHA and testing locally
on x86 does not show errors.

* Move sources to deployed directory docs/.
* Sync sources with shibatch/shibatch.github.io
* Update copyright line in README and html.

And minor fixes on README and website.

These can already be visualised in source or on github.

No need to maintain 2 changelogs.
Besides, the actual changelog is behind the one on the website.
Just keep CHANGELOG.md updated and remove the other one.

Add a table for supported compilers and OS-es.
Add badges linked to CI, License and SLEEF paper, Spack package, ...
Update install and build instructions.

Keep toolchain files and rename them.

fix in libm and quad

This is a follow up from #454.

Individual COMMANDS in define_custom_command are not guaranteed to
execute sequentially, meaning that the inline headers may be broken if
doing a parallel build with make. Each command is extracted into its
own custom_command (some are merged, such as subsequent sed calls) so
that CMake can properly make sense of the dependencies and generate
consistent headers in parallel.

* Atomise creation of inline header, alias, dispscalar, simd disp, sleefquad header and build
* Reuse existing atomic inline libm header generation for CUDA

Now that custom_commands are atomic, parallel builds should be
supported with make.

* Use CMake cat instead of shell cat

This requires minimum CMake 3.18, which is not particularly
recent. This provides a better cross-platform way of concatenating
files.

Removes warnings about deprecation of old API for versions of OpenSSL above 3.0.
Needed to amend openssl static link line to avoid a crash in tester3 with old openssl.
Fixes #486.

This is sometimes defined in math.h.

These two functions were added to the MPFR API.
Fixes #458.

* Add support for the RISC-V Vector ISA
- RISC-V V 1.0 with (latest) intrinsics >=0.12.
- Enable bit manip extensions on RISC-V.
- Add RVVM*NOFMA configurations for determinism/bit-reproducibility across all architectures.

* Not supported yet
- DFT, QUAD, and GNUABI.
- Only upcoming versions of gcc (14+) have support for latest intrinsics. Will be supported when available in release version of gcc.

* CI Updates
- Add riscv64 CI on GitHub Actions.
- Add gcc (11 - #483) and llvm (17) builds on all architectures.
- Disable a few failing test variants, tracked in #484 and #485.
- Use same compiler versions across all builds.
- Add sysroot for dependencies.
- Nit: Reorder build-cross targets to group gcc/llvm per-arch together.

---------

Co-authored-by: GlassOfWhiskey <iacopo.c92@gmail.com>
Co-authored-by: Eric Love <eric.love@sifive.com>

* Add CI on Github Actions for x86_64, aarch64, armhf, ppc64 and s390x
* Disable -march=native for testervecabi - fails when compiling on a host CPU with AVX512f (#478)
* Enable qemu features for s390x, ppc64, and arm
* Add some architecture-specific cmake flags
* Use job dependencies to avoid duplicating building native
* Fix permissions in downloaded artifacts
* Disable arm inline headers - don't work (#480)
* Document why DISABLE_VXE2 on s390x

Co-authored-by: friendlyanon <friendlyanon@users.noreply.github.com>

* Use variables from GNUInstallDirs everywhere

* Componentize the install rules

* Add headers to the interface include path

* Add install rules for CMake package config

* Associate inline headers to a target

Previously inline_headers was a UTILITY target, which cannot be
installed, so it was renamed to inline_headers_util and a new
inline_headers INTERFACE library target is introduced instead.

Co-authored-by: friendlyanon <friendlyanon@users.noreply.github.com>

Co-authored-by: JT <tidswell@maths.usyd.edu.au>

This is yet another patch for cleaning up the code.
It mainly removes warning messages generated when compiling gnuabi_compatibility.c.
It unifies the code for detecting availability of vector extensions.
It removes unused variables.
It adds mveclibtest.c that tests if substitution of mveclib with sleefgnuabi works.

This patch adds OpenMP simd pragmas to more scalar functions.
It also adds more testing to check if the functions can be auto-vectorized.

With this patch, libm functions and libquadmath functions can be replaced with sleef functions by defining SLEEF_REPLACE_LIBM_FUNCS and SLEEF_REPLACE_LIBQUADMATH_FUNCS macros before including sleef.h and sleefquad.h, respectively. The replaced libm functions can be automatically vectorized with gcc on x86_64.