[sdcc-devel] [ sdcc-Patches-2764475 ] Z80: divisions size/speed tradeoff

[SourceForge.net <no...@so...>]

Patches item #2764475, was opened at 2009-04-15 08:54
Message generated for change (Comment added) made by spth
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=300599&aid=2764475&group_id=599

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: z80 port
Group: None
>Status: Closed
>Resolution: Accepted
Priority: 5
Private: No
Submitted By: Marco Bodrato (parerga)
Assigned to: Nobody/Anonymous (nobody)
Summary: Z80: divisions size/speed tradeoff

Initial Comment:
This patch splits the division/modulus computing function in two cases: divisor up to 7 bits long, divisor with more than 7 bits. It may sound strange that the threshold is 7 and not 8... but look at the code and you will see! Both cases can be computed faster than with current code.

Drawback of the patch: code size increases! The _overall_ test-ucz80 result is 7,000 more bytes used, 150,000 ticks saved.

The "atmost7bits" version can be easily adapted to compute both quotient and reminder of a 32-bit long by a positive signed char (7 bits :-), storing dividend (and quotient) in HLDE and divisor in C (A holds the reminder). This code should then be used to speed-up ltoa (radix must be < 37 with the current implementation).

----------------------------------------------------------------------

>Comment By: Philipp Krause (spth)
Date: 2009-05-03 17:15

Message:
IMO the speedup is worth the increase in code size. I suppose we can make
up for the code size increase by further modularization of the asm
routines.

Applied in revision #5447.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=300599&aid=2764475&group_id=599