[Scilab-users] Matlab vs Scilab perf

[Clément David]

Hi Pierre,

Le jeudi 02 mars 2017 à 22:12 +0100, Pierre Vuillemin a écrit :
> Concerning speed, while I agree that this is often not that important, the Julia Language has
> demonstrated that it is not necessarily unreachable for a "high level language". I guess it mainly
> relies on the performance of the JIT compiler at some point. 
> I remember reading that Scilab 6 should have a JIT compiler, but I might be mistaken?

Our approach differs from the Julia one as basically Julia does not have an interpreter at all, they
emit LLVM bytecode (ASM like language) on function definition (emitting one function per handled
datatype or per signature basically).

Our implementation is currently just a proof of concept (read : not shipped with scilab) but some
"language analysis" is shipped within Scilab 6.0.0 without accessible user-interface. We currently
emit one code-generation for the executed path using an "analysis" to infer types, sizes and some
simple behaviors (ones, zeros, find, etc...) on the "double" datatype only.

Thanks,

--
Clément