Re: Lua versus Ravi versus Luajit - a simple benchmark

[Ahmed Charles <acharles@...>]

> On Mar 5, 2015, at 12:17 PM, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> 
>> On 5 March 2015 at 11:55, Ahmed Charles <acharles@outlook.com> wrote:
>> Why does the Ravi version exist in a Debug directory? Are optimizations on?
> 
> Hi,
> The LLVM generated code is not affected by this - only affects the
> interpreter code. But that is not being measured here so it isn't a
> factor in the timings.
> 
> I admit though that I am new to LLVM and therefore may not be
> generating the best code. I do run the LLVM optimizer passes but the
> problem is that contrary to appearances LLVM is poorly documented, so
> it is not at all clear how the optimizer passes should be hooked up,
> and how well they are coping with the LLVM IR I am generating.
> Hopefully I will work out better ways as I learn more.
> 
> Regards
> Dibyendu

It's still a good idea to use release builds for perf testing. The jit is probably a large portion of the time and llvm is known for slow compile times in the jit since most people use it for offline compilation or scenarios where jit time is less of a factor.