This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: arm constant pool problems


> >I'm not saying the above is wrong, but I don't think it should be needed: 
> >the neg_pool_range attribute value should already take this into account 
> >(note that they are all slightly less than the maximum value).  I'll have 
> >a look at this, but it won't be soon since I'm going to be out of the 
> >office for most of the next two weeks.
> 
> Oh.  I assumed the difference between -1024 and -1012 was there just to take 
> care of pipelining.
> 

It needs to cover four things.  

1) The size of the pool entry
2) The location of the instruction in the output sequence of 
   the insn (only important if the insn creates more than one 
   instruction).
3) The pipeline offset
4) The pc-relative range of the instruction doing the load.

For forward fetches only 2) and 4) important to get right.  For backward 
fetches, they are all important.

R.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]