[fpc-devel] just for information

[Jonas Maebe]

On 2020-07-12 09:37, Samuel Herzog via fpc-devel wrote:

> The results you can either verify yourself (by downloading sources or 
> executables)
> or look at the attached screenshot.
> 
> "Composite score" ist the average of the 5 different "benchmarks".
> 
> I am a little irritated about the scores of "sparse matmult" and "LU".

While the results are certainly interesting, keep in mind that SciMark 
(and LU in particular; I don't know about sparse matmult) is generally 
not a very good benchmark in terms of being representative for real 
world programs, especially as far as non-JIT compilers are concerned. 
See e.g. 
https://www.phoronix.com/forums/forum/software/programming-compilers/870805-gcc-6-1-vs-llvm-clang-3-9-compiler-performance?p=871540#post871540 
(last post on the page) by Honza Hubička, a gcc compiler developer (*).

Of course, compiler developers still tend to focus on it because it is 
used a lot to compare compiler performance and nobody likes to look bad. 
And there are of course still optimisations that will improve both those 
benchmarks and real world programs.


Jonas

(*) who has a very interesting blog on compiler optimisations, btw: 
http://hubicka.blogspot.com