More of an outsiders thoughts of a summary of the supposed experts, for 
comment, that's why I noted up front my qualifications. :)
1: Should a compiler, do a reasonable job of generating code on a line 
by line basis, or is this not expected if the targets are a mixed bag of 
devices in a family?
2: When you specify a target device, should the compiler generate 
different code, then for a different device? This is excluding hardware 
specifics like ADC, EUARTS ,UARTS, etc. which I would assume be handled 
correctly for the device.
3: Should the optimizer then take that code generated on a line by line 
basis and improve it by eliminating duplicates, and unnecessary steps?

It appears numerous examples have been cited where the basic code 
generated was less than optimal quality, and as pointed out code size 
can be an issue, though I don't recall seeing actual numbers as to, how 
much larger the code got.


Wouter van Ooijen wrote:
>> Not an area where I have much expertise, but sounds like the original 
>> compiler was poorly done, and a lots of patches call optimizers are 
>> added. I would think a good compiler, would do a good job, based on 
>> individual lines or small groups of lines like a loop of code, and the 
>> optimizer would look at more of a macro view, to eliminate duplicate 
>> instructions, etc., say where a value is already in a register.
>>     
>
> It is not your area of expertise, but you seem to expect that your 
> conclusions are valid?
>
>   
-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist