This patch replaces Horner method which was used to evaluate polynomials with Estrin's method( https://en.wikipedia.org/wiki/Estrin%27s_scheme ) that allows more parallel computations with out-of-order execution.
This patch also introducing a new reduction method to tan.
With this patch, mainly computation for double-precision functions becomes faster, and the effect is like a few percent to 20 percent. For example, the ratio between execution time of the following functions before and after applying this patch is shown below.

Sleef_atan2d4_u35 : 1.21
Sleef_powd4_u10 : 1.17
Sleef_sind4_u35 : 1.10
Sleef_tand4_u10 : 1.04
Sleef_tand4_u35 : 1.17

This patch adds Sleef_strtoq and Sleef_qtostr which can be used to convert between a quad value and a string. These functions are not vectorized. The corresponding testers are also added.

This patch also adds functions for subtraction.

Intel compiler testing is temporarily disabled because of license expiration( https://github.com/shibatch/sleef/issues/238 ).

This is a part of implementation of issue #233 ( https://github.com/shibatch/sleef/issues/233 ).
At this point, add, mul, div and sqrt with testers are implemented. Remaining functions will be committed in the succeeding PRs.
As for vector extensions, SSE2, AVX, FMA4, AVX2, AV2_128, AVX512F, AdvSIMD and SVE are supported.

This quad-precision math library is built only if -DBUILD_QUAD option is given to cmake. For some time(1 year?), this sub-project is positioned at alpha development stage.

1. `-march=armv8-a+simd` is removed as it is not necessary (#232)
2. Delete output that is never generated (#231)

It also includes changes of CI setting for removing GCC/OSX testing on travis. This is because updating gcc with brew takes too much time now. Instead of this, build with gcc is now tested on Jenkins.

Merging 3.3.1 with aavpcs

The x86 dispatcher was build for SP when targeting DP.

This commit enables building `libsleef` and `libsleefgnuabi` with the
`aarch64_vector_pcs` attribute defined in the _Vector Function ABI
specification for AArch64_ [1].

The build must be configured with `-DFORCE_AAVPCS=On`. By default this
configure variable is set to `Off`.

[1] https://developer.arm.com/products/software-development-tools/hpc/arm-compiler-for-hpc/vector-function-abi

This patch reduces testing time to 50%.

I found a bug of tester in denormal/nonnumber handling of functions with two arguments.
This patch fixes that bug.
There is no change in the library itself.

This patch adds implementations of deterministic functions.

The SIMD source files(sleefsimd?p.c) are compiled twice for each vector extension, with DETERMINISTIC macro turned on and off.
Renaming by rename*.h is switched according to DETERMINISTIC macro.

When DETERMINISTIC macro is undefined, the function name xsin will be renamed to Sleef_sind2_u35sse2 with renamesse2.h, for example.

If DETERMINISTIC macro is defined, the function name xsin will be renamed to Sleef_cinz_sind2_u35sse2, for example.

iuty* and tester2y* are added in order to test the newly added deterministic functions. As a consequence, time for testing is increased to almost two times.

In the HTML doc, following items are added
* Changes in 3.3.1
* Update input domain of trigonometric functions

Changes in 3.3.1 is added to CHANGELOG.md

This patch adds support for FreeBSD.
Testing on Jenkins is added.
The bug pointed out on issue https://github.com/shibatch/sleef/issues/217 is also fixed.

This patch fixes build problem on i386(32-bit) linux systems.
Testing on Jenkins is added.

This patch improves the Payne-Hanek reduction algorithm.
The restriction on the input domain of trig functions is all removed.

These two functions are not documented.

This patch adds pure C scalar helpers with which scalar functions are implemented using the SIMD source code.
It also adds helper files for functions which do not use FMA on platforms with FMA support.

This patch adds NEON32+VFPV4 helper which has FMA support.

VFPV4 is supported on most of new 32-bit ARM CPUs, and the computation of some functions is much faster.
This patch does not include a dispatcher.