Re: [PATCH] PR optimization/12324

[Bernardo Innocenti <bernie at develer dot com>]

Jan Hubicka wrote:
> > On Sat, 27 Sep 2003, Jan Hubicka wrote:
> >
> > > the problem is that we deffer compound literals until after finalizing
> > > compilation unit (like we currently do for declarations without initializer).
> > > I don't see any reason for doing so.  Would be this OK?
> >
> > The point is that when the extension (gnu89 mode only, used by Linux
> > kernel) where compound literals can be used as static initializers (as if
> > they were the brace-enclosed list rather than an anonymous variable) is
> > used, the compound literal shouldn't itself be emitted - only the
> > initialized variable.  Does this patch preserve that property?
>
> I am not quite sure I follow this.  Can you show me some testcase?
> In non-unit-at-a-time the compound literal will be output always I guess
> after my patch, while in unit-at-a-time only when it is actually used.
> Perhaps we can add if (!flag_unit_at_a_time) around that line instead.

I've submitted this PR. The attached test case is a simplified case of
a real case from linux/drivers/net/fec.c:


static phy_info_t phy_info_am79c874 = {
       0x00022561,
       "AM79C874",
       (const phy_cmd_t []) {  /* config */
               /* limit to 10MBit because my protorype board
                * doesn't work with 100. */
               { mk_mii_read(MII_REG_CR), mii_parse_cr },
               { mk_mii_read(MII_REG_ANAR), mii_parse_anar },
               { mk_mii_read(MII_AM79C874_DR), mii_parse_am79c874_dr },
               { mk_mii_end, }
       },
       (const phy_cmd_t []) {  /* startup - enable interrupts */
               { mk_mii_write(MII_AM79C874_ICSR, 0xff00), NULL },
               { mk_mii_write(MII_REG_CR, 0x1200), NULL }, /* autonegotiate */
               { mk_mii_read(MII_REG_SR), mii_parse_sr },
               { mk_mii_end, }
       },

	[...]

};

All those mk_parse_foobar things are pointers to static functions.
The regression with 3.4 causes them to be omitted in the output
because the compiler doesn't notice they're being referenced.

--
 // Bernardo Innocenti - Develer S.r.l., R&D dept.
\X/  http://www.develer.com/

Please don't send Word attachments - http://www.gnu.org/philosophy/no-word-attachments.html