Ilya Leoshkevich wrote:
> + /* Return unannotated constant pool references, so that the corresponding
> + entries are added to the back-end-managed pool. Not doing so would
> result
> + in those entries being placed in the middle-end-managed pool, which
> would
> + in turn prevent merging them with identical ones in the back-end-managed
> + pool. */
I'm still not convinced this is necessary; how frequently does the case occur
that a constant is duplicated, and is it worth the extra code complexity?
(Also, note that the middle-end pool is shared across the whole translation
unit, while the back-end pool is duplicated in every function ... so if the
same constant is used by many functions, forcing it into the back-end pool
may *increase* overall size.)
In any case, does this even happen with your current code? You find the
unannotated reference here, and therefore copy the constant into the local
pool, but then replace_constant_pool_ref will still ignore the insn, and
thus the insn will actually continue to refer to the middle-end pool
constant, *not* the copy created in the local pool. Right?
> if (DISP_IN_RANGE (INTVAL (frame_off)))
> {
> - insn = gen_rtx_SET (frame_pointer,
> - gen_rtx_PLUS (Pmode, frame_pointer, frame_off));
> - insn = emit_insn (insn);
> + insn = emit_insn (copy_rtx (cfa));
> }
This seems unrelated?
Bye,
Ulrich
--
Dr. Ulrich Weigand
GNU/Linux compilers and toolchain
ulrich.weig...@de.ibm.com
