Re: HLL syntax & structure suited to rapid compilation?

Tzvetan Mikov wrote:
> "Marco van de Voort" <marcov@toad.stack.nl> wrote in message
>> Tzvetan Mikov wrote:
>>> [... that Borland C/Delphi optimizer is not very good]
>>
>> But how did the overal performance differ?
>>[...]
>
> I have made some CPU-intensive comparisons between Windows compilers
> and in each case Borland C invariably came out at least 20% slower
> than Visual C or GCC. This is a very substancial difference in the
> context of comp.compilers, I think.


IMHO I'd say that that depends if that CPU-intensive is some numerical
benchmark or not. Most of the "benchmarks" are non relevant for real
world performance.


> In the context of real-world Windows programming and the typical
> applications that are developed with Delphi (GUI and database
> frontends?) such performance difference couldn't matter less... :-)


There still are gradations in performance between being the best in
numerical benchmarks and performance being irrelevant due to external
  :-)


If a compiler is somewhat optimizing for normal algo's, other things
get more important;


Things like


- language helpers (strings, dynamic arrays, exceptions)
- Is the basic architecture of the types that need
special treatment optimized
- Will simple or medium operations get inlined?
- Do the calls to helpers only use the minimal amount of registers,
and does the optimizer keep data in regs then?
- Does the compiler avoid generating helper calls if they are
strictly not necessary?


- inlining of small procs and/or switch to reg params for small procs.
      (and optimizing over it)
- Improving of operations on the native stringtype


are IMHO much more important than the degree of tail recursion etc.

Return to the comp.compilers page.
Search the comp.compilers archives again.