This patch adds a dispatcher for 128-bit wide vector functions in the quad library.

This patch adds a dispatcher for the scalar functions in the quad library.

With this patch, the inline headers can be generated with MinGW.
This patch also enables CI testing on MinGW build.

This is a combined patch including the following items.

* Define quad-precision constants in the header files
* Add macros for libquadmath compatibility
* Remove unions from helperpurec_scalar.h. Unions are removed from sleefquadinline_cuda.h, as a result.

With this patch, the pure C inline header can be included from CUDA programs along with the CUDA inline header.

This is a combined patch for eliminating most of the warning messages.

This is a combined patch for removing potential problems with the strict aliasing rule.
It also drops long double support for DFT.

With this patch, `-fno-tree-vrp` compiler option is added if the SVE code is compiled with GCC.

This patch adds support and build-only testing for iOS.
ios.toolchain.cmake is required for building, which can be downloaded from at https://github.com/leetal/ios-cmake

.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds preliminary support and build-only testing for android OS.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds quad-precision ldexp, ilogb, fma and hypot.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds quad-precision frexp and modf.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds quad-precision cbrt function.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

Another cleanup of the quad library

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds quad-precision fmod and remainder.

This patch adds quad-precision trunc, floor, ceil, round and rint.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds quad-precision asinh, acosh and atanh.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds build-only testing to the Jenkins configuration.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

* no message

* no message

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

* Simplify x86 arch check

* Fix sleef compilation when CMAKE_OSX_ARCHITECTURES is passed

Test plan:
  Run `cmake .. -DCMAKE_OSX_ARCHITECTURES=x86_64 -G Ninja; ninja` on M1 Mac
  Run `cmake .. -DCMAKE_OSX_ARCHITECTURES=arm64 -G Ninja; ninja` on x86 Mac

* Compile host executable as universable binaries on OS X
If multiarch option is passed

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

With this patch, double2 and float2 data types can be used instead of Sleef_double2 and Sleef_float2 for CUDA.
It also eliminates a need for including float.h when using the CUDA header file.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This pull request is made following issue https://github.com/shibatch/sleef/issues/371

.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch revises the algorithm for computing the error function.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch moves AArch32 CI testing from travis to jenkins.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch adds POWER9 support as mentioned in issue https://github.com/shibatch/sleef/issues/313

.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

With this patch, the quad functions can be used with header files in which all the functions are included
It also adds support for CUDA quad functions.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch revives the broken i386 support.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

With this patch, cost for each test is defined to speed-up testing.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

This patch fixes the problem pointed out in issue #354.
It also changes the CI setting for mac.

Co-authored-by: shibatch <shibatch.sf.net@gmail.com>

C standard does not define how sign bit should be handled during left shift, which triggers UBSAN runtime error:
```
src/libm/sleefsimdsp.c:3031:101: runtime error: left shift of 1 by 31 places cannot be represented in type 'int'
```
Discovered while working on https://github.com/pytorch/pytorch/pull/48275

C standard do not define have left shift should affect the sign bit, which results in the following runtime error if `vilogbk_vi_vd` is compiled by clang with sanitizer checks enabled:
```
sleef/src/libm/sleefsimddp.c:329:49: runtime error: left shift of 4095 by 20 places cannot be represented in type 'int'
```
Can be fixed by explicitly specifying type of shift 1st operand as unsigned: i.e. replacing `1` with `1U`

CMAKE_SYSTEM_PROCESSOR is set to "arm64" on Apple M1 machines

Discovered while working on https://github.com/pytorch/pytorch/issues/48145