Compiling with Intel icx

StarByte

An optimized compiler (icx) is available on Intel.com (oneAPI). On Intel hardware, I've found it runs about 6x faster than with PellesC or gcc (-O2). Using the clang compiler, it runs about 3x faster with icx.

Is it possible to get faster results with the rtc compiler?

Samir

Hi Starbyte,

Good question. I'm afraid it won't brings better results as even when compiled, the LuaRT program still uses the Lua interpreter.

StarByte

A much faster Lua is important.
Is it possible to rebuild the LUA interpreter with the Intel compiler icx or icpx (C/C++)?

Non-Intel computers also benefit from this...

Samir

It might be possible, but it will need to update the Makefile to support this compiler.
And sadly I don't have enough time to do it if someone can try I would appreciate it !

StarByte

I agree. It would be fantastic.

StarByte

I wrote a small program and included 'minilua'.
I benchmarked with six compilers and made various optimizations (-O3, fast-math, etc.)

VisualC (1.15 s, 600 MB)
LLVM (1.30 s, 660 MB)
icx oneAPI (1.40 s, 886 MB)
MinGW (2.10 s, 600 MB)
PellesC (2.25 s, 295 MB)

Unfortunately, oneAPI ended up in the middle of the pack.