I am going to consider it axiomatic that no software method will compete with hardware instruction for square roots. The only difficulty is that .NET does not give us direct control over the hardware, as in the days of inline assembler for C code.
First, let's discuss the overall hardware perspective of x86.
The x86 floating-point FSQRT command has three prefixes: single, double, and extended (built-in precision of 80-bit FP registers), and 25-40% less time for single or double precision. See here for 32-bit x86 instructions.
, . , . ++ ( ), , #.
. FSQRT , FDIV, Intel , , .
#, , SSE SIMD, . JIT- , .
Intel (15 2010 ), , .NET Framework 4 SIMD:
[ Intel ... SIMD- #]
Mono JIT SIMD Mono 2.2:
[Mono: Release Note Mono 2.2]
Mono SIMD MS #:
fooobar.com/questions/728748/...
( !), Mono SIMD:
fooobar.com/questions/1799423/...