[fpc-other] Interpreter speed

Bernd Oppolzer bernd.oppolzer at t-online.de
Sat Apr 29 18:15:09 CEST 2023


This may be slightly off-topic, but I can tell you some facts about my 
Stanford Pascal compiler,
which runs

- native on z Mainframe machines (which may count as a RISC machine, 
given the instruction set used)
- emulated by Hercules, which is a emulator of z Mainframes
- by emulating P-Code, which is a (sort of) byte code for Pascal ... the 
P-Code (which is pure text, portable)
is translated to a byte code representation before execution; static 
linking is also done in this stage

The times for compiling the compiler (first pass, 26.000 lines) are as 
follows:

- native on z machine: 0.1 seconds
- emulating z by Hercules using a very old z operating system: 10 to 12 
seconds
- similar (10 to 15 seconds), when running on Windows and emulating the 
P-Code;
this includes the time to translate the P-Code char representation ... 
depending on the
power of the used laptop etc., of course.

This makes a factor of 100 for the two emulation strategies, compared to 
the native execution.

The times are CPU times as reported by the builtin function CLOCK.

Windows example:

c:\work\pascal\work\src>pp pascal1

PCINT (Build 1.0 Jun 15 2022 08:21:21)

    **** STANFORD PASCAL COMPILER, OPPOLZER VERSION OF 2023.03 ****

    **** Compiler Summary ****
    ****      No Errors, no Warnings.
    ****   26058 LINE(S) READ,  243 PROCEDURE(S) COMPILED,
    ****   75130 P_INSTRUCTIONS GENERATED,  13.55 SECONDS IN COMPILATION.

*** EXIT Aufruf mit Parameter = 0 ***

HTH, kind regards

Bernd

http://bernd-oppolzer.de/job9.htm


Am 28.04.2023 um 09:20 schrieb Adriaan van Os via fpc-other:
> Out of curiosity — has anybody compared the speed of
>
> 1. interpreting a parsed syntax tree, versus
> 2. interpreting byte code, versus
> 3. interpreting a RISC CPU ?
>
> Regards,
>
> Adriaan van Os
>
> _______________________________________________
> fpc-other maillist  -  fpc-other at lists.freepascal.org
> https://lists.freepascal.org/cgi-bin/mailman/listinfo/fpc-other


More information about the fpc-other mailing list