Average number of instructions per line of kernel code

[Gordan Bobic]

On Thu, 30 Aug 2007, Mohamed Bamakhrama wrote:

>>> I have a question regarding the average number of assembly
>>> instructions per line of kernel code. I know that this is a difficult
>>> question since it depends on many factors such as the instruction set
>>> architecture, compiler used, optimizations used, type of code, coding
>>> style, etc...
>>> I would like to know a rough estimate for such a quantity for the
>>> kernel 2.4/2.6 code running on MIPS32 architecture compiled using GCC.
>>> My estimate is between 5-10 instructions. I googled for such a thing
>>> but couldn't find any useful papers/resources.
>>>
>>
>> I don't understand what use this metric is.
>>
>> For starters, not all lines are created equal.  Not all lines contain
>> code, not all lines with code produce output (hint: dead code).  Not all
>> lines contain a single statement, etc....
>
> I understand that it doesn't sound interesting at all for programmers.
> Nevertheless, this metric is interesting for hardware architects since
> it gives you a hint about the instruction cache behavior of your code.
> If somehow you know how many instructions can result from one
> statement *on average*, then it can help you somehow in calculating
> optimal parameters for the cache (assuming that you can find the
> length of the basic blocks within your C code).

Wow... We have reached the point where we're designing hardware for 
software rather than the other way around. Am I the only one who finds 
that really backwards and wrong?

Gordan