Vector API and Bulk Evaluator - Expression Evaluation will never be the same in Java again

Full Changelog: v2.0.0...v2.0.2

Guess who the kid on the block is? Vector API bulk evaluator(SIMDVectorTurboEvaluator)and its compatibility partner, VectorTurboEvaluator. Both run bulk evaluations at roughly same speed(faster than Janino), and come with workers out of the box! VectorTurboEvaluator works because its code is mechanically sympathetic to the hardware running it,so auto-vectorization occurs.

Math evaluation will never be the same again on the JVM!

Desktop/C2 JIT - Math Expression Benchmark

Expression: 12*x1 + 3*x2 - 4*x3 + 5*x1 - x2 - 4*x3 + 2*x1 + x2
Mode: avgt, 30 iterations

Engine	dataSize	time (ns/op)
Janino	1,000,000	9.81ns
parser-ng-simd	1,000,000	7.49ns
Janino	67,108,864	9.65ns
parser-ng-simd	67,108,864	7.78ns

On Redmi 12(Android), VectorTurboEvaluator evaluated:
sin(z-x)+3*sin(5*x^2 + 4*y^2)

Mobile/ART - Redmi 12 Helio G88

Expression: sin(z-x)+3*sin(5*x^2 + 4*y^2)

dataSize	time (ns/op)
30	8235ns
3000000	903.4ns

Notes:

1. Small batches <1k hit interpreter overhead + JIT warmup
2. 3M elements = steady state after ART JIT compiles hot loop
3. No jdk.incubator.vector on Android. Performance limited by 128-bit NEON + A75 cores

Desktop/C2 + JDK24 Vector API - Intel i5-1135G7

AVX2, 2 cores, cache-hot

dataSize	Kernel	time (ns/op)
40000	Gelu	4.82ns
40000	Swiglu	10.61ns

Notes:

1. 40k elements = 200x200 matrix, steady state after C2 JIT
2. JDK24 --add-modules jdk.incubator.vector required for peak numbers
3. JDK21 ~15% slower. No AVX2 = ~2x slower

Platform	CPU	Expression	Engine	dataSize	time (ns/op)
Desktop	i7-6500U AVX2	2x sin + mul + pow	parser-ng-simd	1M	40ns
Desktop	i5-1135G7 AVX2	Gelu	parser-ng-simd Vector API	40k	4.82ns
Mobile	Helio G88 NEON	2x sin + mul + pow	parser-ng-simd ART	3M	903.4ns

Bug fixes in Standard mode's Matrix Algebra

Full Changelog: v1.2.0...v1.2.1
Version 1.2.1 fixes bugs in Matrix Algebra in the standard mode.

Matrix Algebra

Version 1.2.0 introduces pure Matrix algebra into ParserNG standard. Its great performance is not at par with what ParserNG Turbo can do( with MatrixTurboEvaluator), in terms of memory allocation optimization and matrix evaluation speed, but it ensures that Matrix Algebra is fully available alongside matrix functionality on ParserNG Standard also.

Full Changelog: v1.1.6...v1.2.0

v1.1.6

Full Changelog: v1.1.5...v1.1.6
added matrix_minor function to parser both in standard and turbo

Milking more speed out of turbo with v1.1.5

Full Changelog: v1.1.4...v1.1.5

Completely refactored the runtime variable mapping layer inside ScalarTurboEvaluator1 and ScalarTurboEvaluator2.

• Runtime Remapping Eliminated: Variable array positions are now baked directly into the MethodHandle topology at compile-time.
  The runtime engine now evaluates expressions by reading straight from the user's input arrays.
• 30%+ Evaluation Speed Burst: Microbenchmarks show arithmetic evaluation speeds dropping from ~18ns down to ~12.2ns, pulling within arm's reach of raw native Java performance (~6.4ns).
• Flat Memory Profile: GC allocation churn on hot evaluation paths remains at absolute zero.
  This guarantees stutter-free performance during heavy graph plotting or the soon-coming multi-million step differential equation loops.

Overzealous parser check that disabled nested stats fixed

Sorry, v1.1.3 tag was not pushed.
This release is targeted at an overzealous check in the parser that disabled the nested stats functionality e.g.
sort(1,2,3,listsum(4,5,9),2,8) should start working again now

Full Changelog: v1.1.2...v1.1.4

MatrixTurboEvaluator support for rot function

What's Changed

• Returned original readme, and linked from main readme by @judovana in #44
• Added jdk21+25 to gh workgflows by @judovana in #45
• Version 1.1.2 fixes bugs and makes MatrixTurboEvaluator natively support turbo execution of the rot function. Note that the ScalarTurboEvaluators already support it.
• FastCompositeExpression is now aware of its compiler as it now sports a getCompiler default method(which can be overriden to specify the turbo class that compiled it)

Full Changelog: v1.1.1...v1.1.2

Dynamic Version Implemented

Full Changelog: v1.1.0...v1.1.1

Dynamic Version Computation At runtime implemented

Rotor Function and ErrorLog upgrades

Bug fixes in Rotor and ErrorLog. Matrix of Points upgrade for Rotor

1. The rotor function(rot) has been upgraded to take an argument of a Matrix of Nx3 points to be rotated. It then outputs a Matrix of Nx3 rotated points.
2. There were bug fixes also in the handling of the rotation of functions. When a function is rotated now, the user is handed the output to solve or make the y(vertical coordinates) the subject of the formula, because the output is generally an implicit function.
3. Two point function rotation still supported for backwards compatibility.
4. ErrorLog upgraded

Full Changelog: v1.0.8...v1.1.0

Turbo-Android-Compatibility-Fixes

Full Changelog: v1.0.4...v1.0.8
1.0.8
Bug fixes and more Android compatibility issues resolved.

1.0.6
Bug fixes and Android compatibility issues resolved.

1.0.5
Features bug fixes and optimizations in the scanning/semantic analysis stages.