Re: HLL syntax & structure suited to rapid compilation?

Joachim Durchholz wrote:
> Marco van de Voort wrote:
>> [...] the most important reasons why e.g.
>> TP/BP is so fast, are
>> - [...]
>
> - Almost nonexistent optimization.


Certainly true, but not the dominant factor. Also non optimizing C
compilers are relatively slow, while somewhat optimizing Pascals (like
Delphi and FPC) are still very fast.


> I'm under the impression that compilers spend most of their time with
> optimization.


When they are actually compiling, maybe, mainly because optimization
adds several passes.


> TP did constant expression evaluation and dead code elimination, but
> that's about all: no liveness analysis, no register allocation. The
> resultant machine code was horrendous (but nobody noticed or cared, even
> in the 16 MHz age - so much for the relevance of efficiency).


IIRC it also did some CSE. But indeed it wasn't that much to look at.
I liked TopSpeed a lot at that time.


The main reason was that TP was mostly created in assembler.


Does somebody have an example of a heavily optimizing x86 16-bit compiler
that supports huge memory models?


The problem was that the constant segmentregister reloading in huge
mem models cost a lot of registers, and e.g. TopSpeeds simple keep
variables in reg often stumbled on it.


> That's Turbo Pascal. I never got around to looking at Borland Pascal
> output - I sincerely hope things have improved.


A bit maybe, but when BP came out, it still was the 386sx era.


Delphi is much better though.

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