RFR: 8189105: AARCH64: create intrinsic for sin and cos

[Dmitrij Pochepko]

Hi,

please review patch for JDK-8189105: AARCH64: create intrinsic for sin 
and cos

This patch adds intrinsics for Math::sin(double) and Math::cos(double). 
It implements algorithm for sin and cos from sharedRuntimeTrig.cpp, 
which is a variation of classical libm algorithm implementation. Main 
difference is in heavy SIMD instructions usage, which shortens 
execution. Also few minor optimizations like pipelining instructions, 
range-check-based optimizations, optimized constants loading were applied.


Performance:

I created simple benchmark for sin and cos: 
http://cr.openjdk.java.net/~dpochepk/8189105/MathBench.java It checks 
few values, because of different algorithm branches are taken.

Results summary:

ThunderX2: up to x1.7 improvement

ThunderX: up to x1.4

Cortex A53: up to x1.4

Cortex A73: up to 1.65

Detailed results can be seen here: 
http://cr.openjdk.java.net/~dpochepk/8189105/math-sin.xls


Testing: I launched jdk jtreg test for sin/cos functions: 
(./test/jdk/java/lang/Math/SinCosCornerCasesTests.java) in both Xmixed 
and Xcomp modes

Also I launched respective JCK tests (api/java_lang/Math/sin* and 
api/java_lang/Math/cos*) in both Xmixed and Xcomp modes

All passed, no failures found.


webrev: http://cr.openjdk.java.net/~dpochepk/8189105/webrev.01/

CR: https://bugs.openjdk.java.net/browse/JDK-8189105


NOTE: This patch depends on my another patch for Math::log 
(JDK-8196402), which adds new instructions into assembler_aarch64.hpp: 
frecpe, faddpd, fmlavs, fmlsvs, fmulxvs


Thanks,

Dmitrij