Re: [google] Dump inline decisions more wisely
Here is the new patch.
Thanks,
Dehao
Index: gcc/ipa-inline.c
===
--- gcc/ipa-inline.c(revision 182864)
+++ gcc/ipa-inline.c(working copy)
@@ -308,12 +308,12 @@
bfd_name = "unknown";
buf_size = strlen (bfd_name) + 1;
- if (flag_opt_info >= OPT_INFO_MED && profile_info)
+ if (profile_info)
buf_size += (2 * MAX_INT_LENGTH + 5);
buf = (char *) xmalloc (buf_size);
strcpy (buf, bfd_name);
- if (flag_opt_info >= OPT_INFO_MED && profile_info)
+ if (profile_info)
sprintf (buf,
"%s ("HOST_WIDEST_INT_PRINT_DEC", "HOST_WIDEST_INT_PRINT_DEC")",
buf, node->count, node->max_bb_count);
@@ -354,6 +354,9 @@
return buf;
}
+/* File static variable to denote if it is in ipa-inline pass. */
+static bool is_in_ipa_inline = false;
+
/* Dump the inline decision of EDGE to stderr. */
static void
@@ -364,6 +367,8 @@
const char *call_count_text;
struct cgraph_node *final_caller = edge->caller;
+ if (flag_opt_info < OPT_INFO_MED && !is_in_ipa_inline)
+return;
if (final_caller->global.inlined_to != NULL)
inline_chain_text = cgraph_node_call_chain (final_caller, &final_caller);
else
@@ -1193,6 +1198,8 @@
int min_size, max_size;
VEC (cgraph_edge_p, heap) *new_indirect_edges = NULL;
+ is_in_ipa_inline = true;
+
if (flag_indirect_inlining)
new_indirect_edges = VEC_alloc (cgraph_edge_p, heap, 8);
On Fri, Jan 6, 2012 at 1:39 PM, Xinliang David Li wrote:
> Not ideal but better. Ok with this change.
>
> David
>
> On Thu, Jan 5, 2012 at 5:47 PM, Dehao Chen wrote:
>> Or use a new global variable to denote whether it's in early-inline or
>> ipa-inline?
>>
>> Dehao
>>
>> On Fri, Jan 6, 2012 at 1:46 AM, Xinliang David Li wrote:
>>>
>>> Is there a better way to detect early inline phase and ipa_inline
>>> pass? Use always_inline_functions_inlined flag seems hacky.
>>>
>>> David
>>>
>>> On Wed, Jan 4, 2012 at 1:12 AM, Dehao Chen wrote:
>>> > Hi,
>>> >
>>> > This patch:
>>> >
>>> > * dump inline decisions with profile info whenever available.
>>> > * disable dump of einline decisions at OPT_INFO_MIN.
>>> >
>>> > Is it ok for google branches?
>>> >
>>> > thanks,
>>> > Dehao
>>> >
>>> > 2012-01-04 Dehao Chen
>>> >
>>> > * ipa-inline.c (cgraph_node_opt_info): Print profile info if
>>> > available
>>> > (dump_inline_decision): Disable einline dump at OPT_INFO_MIN
>>> >
>>> > Index: gcc/ipa-inline.c
>>> > ===
>>> > --- gcc/ipa-inline.c (revision 182864)
>>> > +++ gcc/ipa-inline.c (working copy)
>>> > @@ -308,12 +308,12 @@
>>> > bfd_name = "unknown";
>>> >
>>> > buf_size = strlen (bfd_name) + 1;
>>> > - if (flag_opt_info >= OPT_INFO_MED && profile_info)
>>> > + if (profile_info)
>>> > buf_size += (2 * MAX_INT_LENGTH + 5);
>>> > buf = (char *) xmalloc (buf_size);
>>> >
>>> > strcpy (buf, bfd_name);
>>> > - if (flag_opt_info >= OPT_INFO_MED && profile_info)
>>> > + if (profile_info)
>>> > sprintf (buf,
>>> > "%s ("HOST_WIDEST_INT_PRINT_DEC",
>>> > "HOST_WIDEST_INT_PRINT_DEC")",
>>> > buf, node->count, node->max_bb_count);
>>> > @@ -364,6 +364,16 @@
>>> > const char *call_count_text;
>>> > struct cgraph_node *final_caller = edge->caller;
>>> >
>>> > + if (flag_opt_info < OPT_INFO_MED)
>>> > + {
>>> > + tree decl = edge->caller->decl;
>>> > + if (decl)
>>> > + {
>>> > + struct function *fn = DECL_STRUCT_FUNCTION (decl);
>>> > + if (!fn || !fn->always_inline_functions_inlined)
>>> > + return;
>>> > + }
>>> > + }
>>> > if (final_caller->global.inlined_to != NULL)
>>> > inline_chain_text = cgraph_node_call_chain (final_caller,
>>> > &final_caller);
>>> > else
Re: [C++ Patch] deprecation of access declarations
On Thu, 29 Dec 2011, Fabien Chêne wrote: > As previously announced, here is a patch that deprecate access > declarations . I did a little tour in the GCC museum old-deja --and in > g++.dg as well -- to disinter and adjust those dusty tests. To avoid > false positive on invalid code, I have decided to only emit the > warning if the access declaration is actually valid. Mind suggesting a snippt for our release notes at http://gcc.gnu.org/gcc-4.7/changes.html ? (http://gcc.gnu.org/cvs.html has background on how to make changes to our web pages, though if you want to send just a patch against the online ones or a text snippet, I'll take care of the rest.) Gerald
Re: [PATCH] [MIPS] Add Octeon2 cpu support to GCC
On Wed, 4 Jan 2012, Andrew Pinski wrote: > Woops, I had missed that. Here is the patch which adds both octeon+ > and octeon2 to that list. > > I committed as obvious after a build. You may want to update the release notes as well (htdocs/gcc-4.7/changes.html). Gerald
[committed] Skip gcc.misc-tests/gcov-13.c and gcov-14.c on 32-bit hppa*-*-hpux*
Tests need to be skipped for reasons noted in PRs. Tested on
hppa2.0w-hp-hpux11.11 and hppa64-hp-hpux11.11. Committed to
trunk.
Dave
--
J. David Anglin dave.ang...@nrc-cnrc.gc.ca
National Research Council of Canada (613) 990-0752 (FAX: 952-6602)
2012-01-07 John David Anglin
PR gcov-profile/51715
PR gcov-profile/51717
* gcc.misc-tests/gcov-13.c: Skip on 32-bit hppa*-*-hpux*.
* gcc.misc-tests/gcov-14.c: Likewise.
Index: gcc.misc-tests/gcov-13.c
===
--- gcc.misc-tests/gcov-13.c(revision 182980)
+++ gcc.misc-tests/gcov-13.c(working copy)
@@ -4,6 +4,7 @@
/* { dg-require-weak "" } */
/* { dg-do run { target native } } */
/* { dg-additional-sources "gcovpart-13b.c" } */
+/* { dg-skip-if "weak ellision not supported" { { hppa*-*-hpux* } && { !
hppa*64*-*-* } } { "*" } { "" } } */
int __attribute__ ((weak)) weak ()
{
Index: gcc.misc-tests/gcov-14.c
===
--- gcc.misc-tests/gcov-14.c(revision 182980)
+++ gcc.misc-tests/gcov-14.c(working copy)
@@ -5,6 +5,7 @@
/* { dg-additional-options "-flat_namespace -undefined suppress" { target
*-*-darwin* } } */
/* { dg-require-weak "" } */
/* { dg-do run { target native } } */
+/* { dg-skip-if "undefined weak not supported" { { hppa*-*-hpux* } && { !
hppa*64*-*-* } } { "*" } { "" } } */
extern int __attribute__ ((weak)) Foo ();
Re: [google] Dump inline decisions more wisely
ok for google branches.
David
On Sat, Jan 7, 2012 at 1:46 AM, Dehao Chen wrote:
> Here is the new patch.
>
> Thanks,
> Dehao
>
> Index: gcc/ipa-inline.c
> ===
> --- gcc/ipa-inline.c (revision 182864)
> +++ gcc/ipa-inline.c (working copy)
> @@ -308,12 +308,12 @@
> bfd_name = "unknown";
>
> buf_size = strlen (bfd_name) + 1;
> - if (flag_opt_info >= OPT_INFO_MED && profile_info)
> + if (profile_info)
> buf_size += (2 * MAX_INT_LENGTH + 5);
> buf = (char *) xmalloc (buf_size);
>
> strcpy (buf, bfd_name);
> - if (flag_opt_info >= OPT_INFO_MED && profile_info)
> + if (profile_info)
> sprintf (buf,
> "%s ("HOST_WIDEST_INT_PRINT_DEC", "HOST_WIDEST_INT_PRINT_DEC")",
> buf, node->count, node->max_bb_count);
> @@ -354,6 +354,9 @@
> return buf;
> }
>
> +/* File static variable to denote if it is in ipa-inline pass. */
> +static bool is_in_ipa_inline = false;
> +
> /* Dump the inline decision of EDGE to stderr. */
>
> static void
> @@ -364,6 +367,8 @@
> const char *call_count_text;
> struct cgraph_node *final_caller = edge->caller;
>
> + if (flag_opt_info < OPT_INFO_MED && !is_in_ipa_inline)
> + return;
> if (final_caller->global.inlined_to != NULL)
> inline_chain_text = cgraph_node_call_chain (final_caller, &final_caller);
> else
> @@ -1193,6 +1198,8 @@
> int min_size, max_size;
> VEC (cgraph_edge_p, heap) *new_indirect_edges = NULL;
>
> + is_in_ipa_inline = true;
> +
> if (flag_indirect_inlining)
> new_indirect_edges = VEC_alloc (cgraph_edge_p, heap, 8);
>
> On Fri, Jan 6, 2012 at 1:39 PM, Xinliang David Li wrote:
>> Not ideal but better. Ok with this change.
>>
>> David
>>
>> On Thu, Jan 5, 2012 at 5:47 PM, Dehao Chen wrote:
>>> Or use a new global variable to denote whether it's in early-inline or
>>> ipa-inline?
>>>
>>> Dehao
>>>
>>> On Fri, Jan 6, 2012 at 1:46 AM, Xinliang David Li
>>> wrote:
Is there a better way to detect early inline phase and ipa_inline
pass? Use always_inline_functions_inlined flag seems hacky.
David
On Wed, Jan 4, 2012 at 1:12 AM, Dehao Chen wrote:
> Hi,
>
> This patch:
>
> * dump inline decisions with profile info whenever available.
> * disable dump of einline decisions at OPT_INFO_MIN.
>
> Is it ok for google branches?
>
> thanks,
> Dehao
>
> 2012-01-04 Dehao Chen
>
> * ipa-inline.c (cgraph_node_opt_info): Print profile info if
> available
> (dump_inline_decision): Disable einline dump at OPT_INFO_MIN
>
> Index: gcc/ipa-inline.c
> ===
> --- gcc/ipa-inline.c (revision 182864)
> +++ gcc/ipa-inline.c (working copy)
> @@ -308,12 +308,12 @@
> bfd_name = "unknown";
>
> buf_size = strlen (bfd_name) + 1;
> - if (flag_opt_info >= OPT_INFO_MED && profile_info)
> + if (profile_info)
> buf_size += (2 * MAX_INT_LENGTH + 5);
> buf = (char *) xmalloc (buf_size);
>
> strcpy (buf, bfd_name);
> - if (flag_opt_info >= OPT_INFO_MED && profile_info)
> + if (profile_info)
> sprintf (buf,
> "%s ("HOST_WIDEST_INT_PRINT_DEC",
> "HOST_WIDEST_INT_PRINT_DEC")",
> buf, node->count, node->max_bb_count);
> @@ -364,6 +364,16 @@
> const char *call_count_text;
> struct cgraph_node *final_caller = edge->caller;
>
> + if (flag_opt_info < OPT_INFO_MED)
> + {
> + tree decl = edge->caller->decl;
> + if (decl)
> + {
> + struct function *fn = DECL_STRUCT_FUNCTION (decl);
> + if (!fn || !fn->always_inline_functions_inlined)
> + return;
> + }
> + }
> if (final_caller->global.inlined_to != NULL)
> inline_chain_text = cgraph_node_call_chain (final_caller,
> &final_caller);
> else
PR tree-optimization/51680 (not inlining comdats)
Hi,
this patch fixes first half of the PR. The PR is about a function that is
supposed
to be inlined at -O2. The function contains two calls, one indirect and
inlining
the indirect call makes things a lot better. GCC however things that replacing
call
by two calls at -O2 is a bad idea.
This patch solves first part of it. The function in question is COMDAT. We
already
know that C++ coding style makes COMDATs quite rarely to be shared and at -Os we
already handle them sort of similarly to static ufnctions. We however don't do
that
at -O2. This patch makes it happen in both cases.
Bootstrapped/regtested x86_64-linux, comitted.
Will commit the testcase with the followup fix.
PR tree-optimization/51680
* ipa-inline.c (want_inline_small_function_p): Be more lax on functions
whose inlining reduce unit size.
Index: ipa-inline.c
===
--- ipa-inline.c(revision 182949)
+++ ipa-inline.c(working copy)
@@ -482,21 +482,13 @@ want_inline_small_function_p (struct cgr
e->inline_failed = CIF_MAX_INLINE_INSNS_SINGLE_LIMIT;
want_inline = false;
}
- else if (!DECL_DECLARED_INLINE_P (callee->decl)
- && !flag_inline_functions)
- {
- e->inline_failed = CIF_NOT_DECLARED_INLINED;
- want_inline = false;
- }
- else if (!DECL_DECLARED_INLINE_P (callee->decl)
- && growth >= MAX_INLINE_INSNS_AUTO)
- {
- e->inline_failed = CIF_MAX_INLINE_INSNS_AUTO_LIMIT;
- want_inline = false;
- }
- /* If call is cold, do not inline when function body would grow.
-Still inline when the overall unit size will shrink because the offline
-copy of function being eliminated.
+ /* Before giving up based on fact that caller size will grow, allow
+ functions that are called few times and eliminating the offline
+copy will lead to overall code size reduction.
+Not all of these will be handled by subsequent inlining of functions
+called once: in particular weak functions are not handled or funcitons
+that inline to multiple calls but a lot of bodies is optimized out.
+Finally we want to inline earlier to allow inlining of callbacks.
This is slightly wrong on aggressive side: it is entirely possible
that function is called many times with a context where inlining
@@ -509,24 +501,39 @@ want_inline_small_function_p (struct cgr
first, this situation is not a problem at all: after inlining all
"good" calls, we will realize that keeping the function around is
better. */
- else if (!cgraph_maybe_hot_edge_p (e)
- && (DECL_EXTERNAL (callee->decl)
+ else if (growth <= MAX_INLINE_INSNS_SINGLE
+ /* Unlike for functions called once, we play unsafe with
+ COMDATs. We can allow that since we know functions
+ in consideration are small (and thus risk is small) and
+ moreover grow estimates already accounts that COMDAT
+ functions may or may not disappear when eliminated from
+ current unit. With good probability making aggressive
+ choice in all units is going to make overall program
+ smaller.
+
+ Consequently we ask cgraph_can_remove_if_no_direct_calls_p
+ instead of
+ cgraph_will_be_removed_from_program_if_no_direct_calls */
+ && !DECL_EXTERNAL (callee->decl)
+ && cgraph_can_remove_if_no_direct_calls_p (callee)
+ && estimate_growth (callee) <= 0)
+ ;
+ else if (!DECL_DECLARED_INLINE_P (callee->decl)
+ && !flag_inline_functions)
+ {
+ e->inline_failed = CIF_NOT_DECLARED_INLINED;
+ want_inline = false;
+ }
+ else if (!DECL_DECLARED_INLINE_P (callee->decl)
+ && growth >= MAX_INLINE_INSNS_AUTO)
+ {
+ e->inline_failed = CIF_MAX_INLINE_INSNS_AUTO_LIMIT;
+ want_inline = false;
+ }
+ /* If call is cold, do not inline when function body would grow. */
+ else if (!cgraph_maybe_hot_edge_p (e))
- /* Unlike for functions called once, we play unsafe with
- COMDATs. We can allow that since we know functions
- in consideration are small (and thus risk is small) and
- moreover grow estimates already accounts that COMDAT
- functions may or may not disappear when eliminated from
- current unit. With good probability making aggressive
- choice in all units is going to make overall program
- smaller.
-
- Consequently we ask cgraph_can_remove_if_no_direct_calls_p
- instead of
- cgrap
Re: increase call_saved_regs[] in caller-save.c
On Mon, 2011-11-07 at 09:27 +0100, Michael Matz wrote: > One source of same valued pseudos are copies, and copy coalescing we (of > course) do implement. While digging through ira-color.c tracking down an IRA shuffle copy issue, I noticed we only seem to do real copy coalescing for spilled pseudos. It seems we rely on coloring to try and assign the same hard reg to pseudos connected by a copy so the copy can be removed as a nop. Looking at all the code used to do the cost preferencing to achieve that, I'm guessing just coalescing them would be a lot easier. Peter
