Compare RISC-V/Swift to Custom ISA/Go

[j1010001]

No description provided.

[jsproz]

The attached archive has the C++, RISC-V assembly, and hexdump of the imperative and recursive fibonacci functions. The assembly and hex files were created with the cadence-tool. The hex files can be executed using the cadence-tool. The imperative version is 10x faster than the recursive version. Currently the recursive version does not work for values greater than 9.

fibonacci.zip

[SupunS]

These are the numbers (execution time) for the custom VM/ISA:

Implementation	fib(9)	fib(14)	fib(46)
Imperative	0.745µ	1.239µ	5.260µ
Recursive	7.220µ	80.67µ	392.31s

(Runtime checks were disabled)

Benchmarked on my laptop: Apple M1 Pro, 8-Cores, 16GB RAM

Source code: vm_bench_test.go

[SupunS]

Fib(46) comparison:

                │  fib_riscv  │             fib_vm                   │
                │   sec/op    │    sec/op     vs base                │
FibImperative-8   718.6n ± 2%   3232.0n ± 1%  +349.73% (p=0.002 n=6)

                │ fib_riscv │         fib_vm              │
                │   B/op    │    B/op     vs base         │
FibImperative-8    0.0 ± 0%   264.0 ± 0%  ? (p=0.002 n=6)

                │ fib_riscv │         fib_vm              │
                │ allocs/op │ allocs/op   vs base         │
FibImperative-8   0.00 ± 0%   33.00 ± 0%  ? (p=0.002 n=6)