> -----Original Message-----
> From: Jeff Law [mailto:l...@redhat.com]
> Sent: Tuesday, November 06, 2012 4:51 AM
> To: Bin Cheng
> Cc: gcc-patches@gcc.gnu.org
> Subject: Re: [PATCH Version 2][RFA]Improving register pressure directed
hoist
>
> On 11/02/2012 02:34 AM, Bin Cheng wrote:
> >>
> > Also I don't understand why the bogus patch can catch more hoist
> > opportunities and improve code size, so please help if you have any
> > idea about this.
> Well, perturbing the code, particularly in a way which is supposed to
change
> the amount of register pressure is certainly going to affect the
allocators
> and reload.
>
> It shouldn't be that hard to look at results which differ between the two
> patches and analyze why they're different.
I will try to investigate this issue.
>
> >
> > 2012-11-02 Bin Cheng<bin.ch...@arm.com>
> >
> > * gcse.c: (struct bb_data): Add new fields, old_pressure, live_in
> > and backup.
> > (calculate_bb_reg_pressure): Initialize live_in and backup.
> > (update_bb_reg_pressure): New.
> > (should_hoist_expr_to_dom): Add new parameter from.
> > Monitor the change of reg pressure and use it to drive hoisting.
> > (hoist_code): Update LIVE and reg pressure information.
> >
> > gcc/testsuite/ChangeLog
> > 2012-11-02 Bin Cheng<bin.ch...@arm.com>
> >
> > * gcc.dg/hoist-register-pressure-3.c: New test.
> >
> >
> > +/* Update register pressure for BB when hoisting an expression from
> > + instruction FROM, if live ranges of inputs are shrunk. Also
> > + maintain live_in information if live range of register referred
> > + in FROM is shrunk.
> > +
> > + Return 0 if register pressure doesn't change, otherwise return
> > + the number by which register pressure is decreased.
> > +
> > + NOTE: Register pressure won't be increased in this function. */
> > +
> > +static int
> > +update_bb_reg_pressure (basic_block bb, rtx from,
> > + enum reg_class pressure_class, int nregs) {
> > + rtx dreg, insn;
> > + basic_block succ_bb;
> > + df_ref *op, op_ref;
> > + edge succ;
> > + edge_iterator ei;
> > + int decreased_pressure = 0;
> > +
> > + for (op = DF_INSN_USES (from); *op; op++)
> > + {
> > + dreg = DF_REF_REAL_REG (*op);
> > + /* The live range of register is shrunk only if it isn't:
> > + 1. referred on any path from the end of this block to EXIT, or
> > + 2. referred by insns other than FROM in this block. */
> > + FOR_EACH_EDGE (succ, ei, bb->succs)
> > + {
> > + succ_bb = succ->dest;
> > + if (succ_bb == EXIT_BLOCK_PTR)
> > + continue;
> > +
> > + if (bitmap_bit_p (BB_DATA (succ_bb)->live_in, REGNO (dreg)))
> > + break;
> > + }
> > + if (succ != NULL)
> > + continue;
> > +
> > + op_ref = DF_REG_USE_CHAIN (REGNO (dreg));
> > + for (; op_ref; op_ref = DF_REF_NEXT_REG (op_ref))
> > + {
> > + if (!DF_REF_INSN_INFO (op_ref))
> > + continue;
> > +
> > + insn = DF_REF_INSN (op_ref);
> > + if (BLOCK_FOR_INSN (insn) == bb
> > + && NONDEBUG_INSN_P (insn) && insn != from)
> > + break;
> > + }
> > +
> > + /* Decrease register pressure and update live_in information for
> > + this block. */
> > + if (!op_ref)
> > + {
> > + decreased_pressure += nregs;
> > + BB_DATA (bb)->max_reg_pressure[pressure_class] -= nregs;
> > + bitmap_clear_bit (BB_DATA (bb)->live_in, REGNO (dreg));
> > + }
> > + }
> > + return decreased_pressure;
> So we're looking to see if any of the registers used in FROM are used
after
> from. If none are used, then we decrease the register pressure by nregs
which
> appears to be a property of the the registers *set* in FROM.
> Is seems like there's some inconsistency here. Or have I misunderstood
> something?
>
> I'm not sure how much it matters in practice, except perhaps for
conversions
> and the like where the source and destination operands are different
modes.
Agreed, I missed the inconsistence of register_class/number between input
and output. I will fix this issue and measure the effect once I get back to
office.
Thanks for your comments.
