Are there any JIT mode optimizations flags? #1120
Description
Hi, I'm trying to compare daScript performance with the performance of my programming language (Spawn) in the Fibonacci example.
daScript's performance info states the following:
daScript Just-in-Time compilation is based on LLVM backend, and often outperforms C++, due to daScript providing more information to LLVM than AOT.
and provides the following table
daScript code I'm benchmarking is the following:
def fibI(n)
var last = 1
var cur = 0
for i in range(n)
let tmp = cur
cur += last
last = tmp
return cur
[export]
def main
print("{fibI(6511134)}")
I'm running it with just das fibI.das
Spawn code is the following:
fn main() {
print('${fib_i(6511134)}')
}
fn fib_i(n u64) -> u64 {
mut last := 1
mut cur := 0
for i in 0 .. n {
tmp := cur
cur += last
last = tmp
}
return cur
}
I'm building executable with ./spawnlang --cc clang --opt-fast --show-timings=false -o fib_sp ../fibI.sp
I'm comparing both daScript and Spawn using hyperfine hyperfine -N --warmup 5 "das fibI.das" ../../spawn/fib_sp
The results are the following
Benchmark 1: das fibI.das
Time (mean ± σ): 292.5 ms ± 11.0 ms [User: 276.2 ms, System: 13.6 ms]
Range (min … max): 275.4 ms … 319.6 ms 10 runs
Benchmark 2: ../../spawn/fib_sp
Time (mean ± σ): 4.5 ms ± 0.6 ms [User: 3.1 ms, System: 1.1 ms]
Range (min … max): 2.2 ms … 5.9 ms 574 runs
Summary
../../spawn/fib_sp ran
64.72 ± 8.79 times faster than das fibI.das
Am I doing something wrong? Should I add any optimization flags to daScript build command or something? I can't try AOT for now (see #1119), but seems JIT mode of daScript should perform sometimes better than AOT mode, and I have drastically difference in comparison with my compiled programming language that should have similar performance as AOT
I'm on Arch Linux, with lastest GCC and Clang and dev tools