Standalone SIMD library

Portable 128-bit SIMD for Ada.

Flyology SIMD provides explicit integer and floating-point vector operations for ordinary Ada programs. The scalar backend defines each result. NEON and x86 backends implement the same semantics behind private representations.

Read the setup guideRead the API

Experimental v0.1 API. GCC-based GNAT is the verified compiler family.

U8x16 registerlane 0 is the first loaded element
2F2E2D2C2B2A29282726252423222120
compare “ ”
0000 0000 0000 0001
portable meaning→private backend→machine code
10 vector familiesAll 128-bit signed, unsigned, and IEEE floating families.
Zero runtime dependenciesNo Flyology runtime, heap allocation, tasking, I/O, locks, or waits.
Coarse dispatchFeature selection occurs once for a complete buffer algorithm.
Portable means three things

Semantics first. Mechanism private.

The same operation contract applies on every target. Scalar source keeps programs buildable where no SIMD backend exists. Target-specific backends provide efficient code generation where the compiler and instruction set have been verified.

  1. 01Identical public semantics

    Wrapping, saturation, shifts, masks, floating-point edge cases, and memory extent have defined results.

  2. 02Scalar source portability

    The authoritative backend runs on every supported GCC-based GNAT target.

  3. 03Efficient target code

    AArch64 NEON and x86-64 SSE2 implement the complete 128-bit family. AVX2 accelerates selected buffer loops.

Fixed-width family

Types carry the lane shape.

Signed, unsigned, floating-point, and mask values do not convert implicitly. Operations such as Add_Wrap, Add_Saturate, Min_Number, and Load_Partial state their portable behavior.

Inspect the type family
U8x1616 × 8
I8x1616 × 8
U16x88 × 16
I16x88 × 16
U32x44 × 32
I32x44 × 32
U64x22 × 64
I64x22 × 64
F32x44 × 32
F64x22 × 64