Re: Repost: RFA [4.1]: improvement to if-conversion and cross-jumping (PR20070)

[Joern RENNECKE <joern dot rennecke at st dot com>]

Ulrich Weigand wrote:

>  
>
> Those are the top-most insns of the match.  As I understand it, the [xy]_local_live
> bitmaps now correspond to the live information at the points immediately preceding
> those two insns.
>
>  
>
> > When the last insn of the match has been processed, bits 1 and 2 should 
> > be set for both blocks, and hence the bitmaps intersect.
> >    
>
> How come?  Before those two insns, register 1 in block X is *not* live,
> and same for register 2 in block Y.
>  
Oops, I got confused when you brougt up resolve_input_conflict.  Since the
registers are not live at the start, they are not inputs; they are used 
only locally.
Hence there is no input conflict, either.

Earlier you wrote:

> in try_crossjump_to_edge:
>
> 1299      for (i = 0; i < info.cur.local_count; i++)
> 1300        if (info.local_rvalue[i])
> 1301          emit_insn_before (gen_move_insn (info.x_local[i], 
info.y_local[i]),
> 1302                            y_active);
>
> where
>
> (gdb) print info.cur.local_count
> $2 = 2
> (gdb) call debug_rtx (info.x_local[0])
> (reg/f:SI 1 %r1 [83])
> (gdb) call debug_rtx (info.y_local[0])
> (reg/f:SI 2 %r2 [70])
> (gdb) call debug_rtx (info.x_local[1])
> (reg:SI 2 %r2 [86])
> (gdb) call debug_rtx (info.y_local[1])
> (reg:SI 1 %r1 [73])

You have to look at info.local_rvalue for what registers are actually 
inputs.