[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: The no goto religion
From:       James Richard Tyrer <tyrerj () acm ! org>
Date:       2007-08-05 1:46:19
Message-ID: 46B52BEB.1010200 () acm ! org
[Download RAW message or body]

Sylvain Joyeux wrote:
>>> You have no data to back up this assertion,
>> No, I have a published paper written by Donald Knuth.$
> Dated 1974. Microprocessor architecture evolved a lot since then, so any 
> performance-related data in that paper is crap nowadays.

I presume that you didn't read the paper.  His numbers are based on how 
the code would compile.  Not on actual execution in a specific 
architecture.  I don't think that there is any doubt that, in most 
cases, more instructions will result in longer execution time.

The main change in microprocessor architecture is the use of longer 
pipelines.  If you are familiar with MPU architecture you are probably 
aware that the main issue with long pipelines is the high (time) cost of 
each JUMP.  Aside from that, although processors run much faster and use 
various methods to speed up execution, the fact is that each machine 
language instruction still gets executed in an x86 type MPU.  Proposed 
architectures exist that don't execute every instruction, but these are 
not anywhere near reaching practical CPUs.

So, his paper is still valid with the possible exception of the JUMP 
issue.  Due to the high (time) cost of a JUMP instruction, there is a 
possibility where using extra code to avoid a JUMP would improve 
execution time.  This is not the case here since the redundant code in 
the code used as an example causes an extra jump.

Since your main line of argument appears to be to say 'crap' rather than 
engage in a useful instruction, I am going to classify such remarks as 
trolling and not respond further.

-- 
JRT
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<
[prev in list] [next in list] [prev in thread] [next in thread]