This is the mail archive of the gcc@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: PPro and egcs



  In message <m0xPxjy-0004edC@ocean.lucon.org>you write:
  > > 
  > > 
  > >   In message <m0xPxLV-0004edC@ocean.lucon.org>you write:
  > >   > I tried it. But it failed to compile _bb.o in libgcc2.a.
  > >   > It aborted in find_reloads (). It seems that there weren't
  > >   > enough registers for reload.
  > > Well, I'd be tempted to simplify it such that it only accepts regs
  > > and requires an inout operand.  That handles the most common cases
  > > and should require more registers than your typical add pattern.
  > > 
  > 
  > >From what I saw, gcc wound up trying to put those operands into
  > registers to satisfy the constraints, which caused the register
  > shortage.
No, there was a const_int in the original example, which caused gcc to
try and use the last constraint, which didn't work because the clobber
was teh same as one of the input registers.

If you fix the pattern to only use two registers (remember there's a
matching in/out) then it shouldn't be any more register pressure
sensitive than any other two operand instruction.


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