Re: Loop jamming!?

Stanley Chow wrote:
>> Loop jamming will result in a loop that is, in general, slower than
>> each of the original loops. The time penalty will depend on many
>> factors (including cache size, serial dependency, multiple ALU) and
>> in the extreme case, the time penalty can be zero wrt the slower loop.


  <cliff.click@Eng.Sun.COM> wrote:
>Actually, it's pretty easy for the jammed loop to run faster than the
>slowest loop, if the cache locality is right.


Then the compiler was not doing a good job for the slow loop - it should
have managed the cache better. In the extreme :-), it could invent
the "extra" loop and use loop jamming to speed up the slow loop.


--
Stanley Chow; schow@nortel.ca, (613) 763-2831
Nortel/BNR, PO Box 3511 Station C, Ottawa, Ontario
--

Return to the comp.compilers page.
Search the comp.compilers archives again.