Skip to content

Performance.md

Latest commit

 

History

History
106 lines (80 loc) · 3.88 KB

Performance.md

File metadata and controls

106 lines (80 loc) · 3.88 KB

Performance

Benchmark results, methodology, and memory layout analysis.

Overview

All benchmarks run using package-benchmark in release mode with .mega scaling factor (1,000,000 iterations per sample). Measured on Apple M4 Max, macOS 15, Swift 6.2.

Wall Clock (per operation)

Operation FixedPointDecimal Foundation.Decimal Speedup
Addition 0.67 ns 240 ns 359x
Subtraction 0.67 ns 283 ns 424x
Multiplication 8 ns 607 ns 79x
Division 8 ns 1,285 ns 168x
Comparison (<) 0.33 ns 300 ns 901x
Equality (==) 0.34 ns 317 ns 943x
Hash 5 ns 261 ns 48x
To Double 0.49 ns 271 ns 551x
String description 44 ns 1,045 ns 24x
JSON encode 320 ns 1,215 ns 3.8x
JSON decode 457 ns 831 ns 1.8x
init(significand:exponent:) 0.41 ns — —
init(Double) 1.4 ns 2,319 ns 1,622x
rounded(scale:) 2 ns 705 ns 349x

Instructions (per operation)

Operation FixedPointDecimal Foundation.Decimal Ratio
Addition 13 6,923 533x
Subtraction 14 8,442 603x
Multiplication 86 18,000 209x
Division 88 35,000 398x
Comparison (<) 10 8,829 883x
Equality (==) 10 9,056 906x
Hash 150 7,258 48x
To Double 12 7,892 658x
String description 1,393 31,000 22x
JSON encode 8,805 35,000 4x
JSON decode 9,577 16,000 1.7x
init(significand:exponent:) 7 — —
init(Double) 36 50,000 1,389x
rounded(scale:) 44 16,000 364x

Wrapping Operators (&+, &-, &*)

Wrapping operators perform no validation — no NaN checks, no overflow traps, no sentinel adjustment — matching Swift Int wrapping operator semantics.

Wall Clock (per operation)

Operation Checked Wrapping Speedup
Addition 0.67 ns 0.67 ns 1.00x
Subtraction 0.67 ns 0.67 ns 1.00x
Multiplication 8 ns 9 ns 0.83x (slower)

Instructions (per operation)

Operation Checked Wrapping Reduction
Addition 13 9 31% fewer
Subtraction 14 8 43% fewer
Multiplication 86 101 17% more

&+ and &- execute fewer instructions but show identical wall clock time — the NaN/overflow branches in checked +/- are always-not-taken and perfectly predicted by the CPU.

&* is actually slower than checked * because the precondition in * gives the optimizer proof that the result fits in Int64, enabling tighter code generation. Prefer * unless non-trapping overflow behavior is specifically needed.

Zero Heap Allocations

FixedPointDecimal performs zero heap allocations across all arithmetic, comparison, conversion, rounding, and string formatting operations.

For latency-sensitive applications processing thousands of values per tick, eliminating these allocations removes a significant source of jitter.

Memory Layout

Type Size Stride Values per cache line (64B)
FixedPointDecimal 8 bytes 8 bytes 8
Foundation.Decimal 20 bytes 24 bytes 2

For ContiguousArray with 5K elements:

  • FixedPointDecimal: 40 KB (fits in L1 cache)
  • Foundation.Decimal: 120 KB (exceeds L1, spills to L2)

Wire Size

Type Per-field size
FixedPointDecimal (as Int64 scalar) 8 bytes
Foundation.Decimal (as struct) 20 bytes

This compounds across messages containing multiple price/quantity fields.

Reproducing Benchmarks

cd Benchmarks
swift package benchmark