JOML Comparison

JOML Comparison

Feature Comparison

Feature	Jamma	JOML 1.10.8
Vector mutability	Immutable (records)	Mutable
Vector types	2d/f/i, 3d/f/i, 4d/f/i	2d/f, 3d/f, 4d/f
Integer vectors	✅ Vector2i, Vector3i, Vector4i	❌
Matrix types	2x2, 3x3, 4x4, 4x3 (d/f)	2x2, 3x3, 4x4, 4x3 (d/f)
Quaternion	Record (immutable)	Mutable class
Dual numbers	✅ DualNumberd/f	❌
ScalarMath	80+ methods	❌ (only basic Math)
Easing functions	34 functions	❌
Java records	✅ All vectors & quaternions	❌
JPMS module-info	✅ All modules	❌
MemorySegment I/O	✅ All types	Partial (NIO only)
SIMD (Vector API)	✅ Incubator module	✅ (later versions)
Parallel batch ops	✅ ParallelOps	❌
Off-heap memory	✅ MemoryOps	❌
Math.fma usage	✅ All dot products, lerps, matrix mul	Partial
Thread safety	✅ Trivially safe (records)	❌ Not safe
equals/hashCode	✅ Correct (records)	❌ Incomplete
LWJGL integration	✅ lwjgl-jamma module	✅ Native

Why Not JOML?

1. Immutability

JOML uses mutable objects everywhere. v.add(w) mutates v. Jamma's records fix this — v.add(w) returns a new vector, v is untouched. No defensive copies, no thread safety issues, correct equals/hashCode.

2. No Scalar Math

JOML provides vector and matrix types and nothing else. Jamma's ScalarMath covers 80+ functions.

3. No Integer Vectors

JOML has no Vector2i, Vector3i, or Vector4i. Jamma has all three.

4. No Dual Numbers

Automatic differentiation is essential for gradient-based optimization and numerical methods.

5. No JPMS

JOML ships without module-info.java. Jamma is fully modular.

6. No MemorySegment Support

JOML relies on NIO buffers only. Jamma supports both NIO buffers and MemorySegment.

7. No Parallel / SIMD Incubator

Jamma's incubator module provides SIMD vector ops and StructuredTaskScope-based parallelism.

8. Immutable Vectors Are Fast Enough

Modern JVMs eliminate short-lived allocations via escape analysis. Generational ZGC collects them at sub-millisecond cost.