This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: state of 3.2.1-pre: how far from release?
- From: Michael Matz <matz at suse dot de>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: Geoff Keating <geoffk at geoffk dot org>,Richard Henderson <rth at redhat dot com>, Daniel Berlin <dan at dberlin dot org>,<gcc at gcc dot gnu dot org>
- Date: Wed, 6 Nov 2002 16:56:44 +0100 (CET)
- Subject: Re: state of 3.2.1-pre: how far from release?
Hi,
On Wed, 6 Nov 2002, David Edelsohn wrote:
> A larger fragment of the code generating the parallel is:
>
> operands[3] = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (count));
> op1 = replace_equiv_address (operands[1],
> force_reg (SImode, XEXP (operands[1], 0)));
>
> for (i = 0; i < count; i++)
> XVECEXP (operands[3], 0, i)
> = gen_rtx_SET (VOIDmode, gen_rtx_REG (SImode, regno + i),
> adjust_address (op1, SImode, i * 4));
>
> op1 contains the memref forced into a reg. This produces RTL like:
>
> (insn 1842 11909 11910 (set (reg/f:SI 929)
> (plus:SI (reg/f:SI 31 r31)
> (const_int 520 [0x208]))) 36 {*addsi3_internal1} (nil)
> (expr_list:REG_EQUIV (plus:SI (reg/f:SI 31 r31)
> (const_int 520 [0x208]))
> (nil)))
>
> (insn 1856 1853 16458 (parallel[
> (set (reg:SI 4 r4)
> (mem/s:SI (reg/f:SI 929) [0 S4 A32]))
> (set (reg:SI 5 r5)
> (mem/s:SI (plus:SI (reg/f:SI 929)
> (const_int 4 [0x4])) [0 S4 A32]))
> (set (reg:SI 6 r6)
> (mem/s:SI (plus:SI (reg/f:SI 929)
> (const_int 8 [0x8])) [0 S4 A32]))
> (set (reg:SI 7 r7)
> (mem/s:SI (plus:SI (reg/f:SI 929)
> (const_int 12 [0xc])) [0 S4 A32]))
> ] ) 322 {*rs6000.md:8746} (insn_list:REG_DEP_OUTPUT 1843 (insn_list:REG_DEP_OUTPUT 1846 (insn_list 1842 (insn_list:REG_DEP_ANTI 1849 (nil)))))
> (nil))
>
> eliminate_regs_in_insn() modification of the pseudo in the first
> expression does not affect the pseudo in the other expressions.
I must be dense or something. But eliminate_regs_in_insn() is only for
converting eliminable hardregs into their target plus constant, like for
instance frame to stack pointer elimination. The changing from pseudo
regs into their reg_renumber[] hardregs is done by alter_reg(). I.e. in
reload() after the
for (i = LAST_VIRTUAL_REGISTER + 1...) alter_reg (i, -1);
loop, all pseudos which have a hardreg should be replaced by it (i.e. if
reg 929 got a hardreg it should be consistently replaced by
reg_renumber[929]). I don't know how eliminate_regs_in_insn() should come
into play here (except maybe if r11 is somehow eliminable?).
Ciao,
Michael.