M
> Hi,
>
> On 08/29/2013 10:11 AM, Jan Hubicka wrote:
> >Paolo,
> >there seems to be one extra issue about this patch. It causes quite a twist
> >in libstdc++ exported symbols.
> >It is purpose of the patch to remove those that are going to be generated i
Richard, Jason,
I would apprechiate your opinion on this patch. It blocks all the code that
makes
use of ipa-devirt post LTO streaming.
The main part that I would like to know your opinion on is the ODR rule
implementation
by vtable comparsion (it looks obvious but I got it wrong once already)
Jakub,
I am adding you to CC since I put my current toughts on LTO and debug info
in here.
> > Fork-fire-forget is really a much simpler choice here IMO; no worries
> > about shared resources, less debug hassle.
>
> It might be not as cheap as it is on Linux hosts on other hosts of
> course. Als
> Hi,
>
> On 08/29/2013 02:19 PM, Jan Hubicka wrote:
> >So my belief is that it is safe to drop those symbols from
> >libstdc++. Every program/DSO using them have to define its own
> >copy of those symbols, so I believe removing them from libstdc++
> >won
> On Thu, Aug 29, 2013 at 02:47:46PM +0200, Paolo Carlini wrote:
> > On 08/29/2013 02:19 PM, Jan Hubicka wrote:
> > >So my belief is that it is safe to drop those symbols from
> > >libstdc++. Every program/DSO using them have to define its own
> > >copy of those
> > Said that, I now have the fork() patch in all my trees and enjoy 50% faster
> > WPA times. I changed my mind about claim that stremaing should be disk
> > bound -
> > it is hard to hope for disk boundness for something that should fit in
> > cache.
>
> It should at least limit its fork rate
Hi,
this patch makes COMDAT profiles right with LTO. (made jointly with Martin
Liska)
To recap the issue: COMDAT functions are produced many times. Each copy gets
its own set of counters and depending on inlining and linker decision, one
or more of copies of a given COMDAT function will get exe
> Great! Is this the LTO merging you were talking about in an earlier
> message, or the gcov runtime fix (that would presumably not be
> lto-specific)?
It is the LTO path - we need to merge profiles there anyway for his code
unification
work.
> > I have patch to track this. Moreover vforks seem
> > The esitmated profile is already there before reading the profile in, so we
> > only do not want to overwrite it. Does the following work for you?
>
> It does get the estimated frequencies on the bbs.
Good.
> > We wil also need to solve problem that in this case cgraph edges will have
> >
>
> Please instead remove this assert and put the checking into
> tree-cfg.c:verify_gimple_label where it should need no special-casing
> on cgraph_state.
>
> Otherwise the non-profile bits look ok.
OK, will do.
Thank you!
Honza
> >
> > The frequency condition needs to be done only when you walk predecestors -
> > when
> > you walk down the edge probabilities are just fine.
>
> True. For simplicity I think it should be fine to leave as-is so there
> isn't more special casing as the current approach works in both
> direct
> Jan Hubicka writes:
> >
> > Now for longer term, we want to make function CFGs independent of gimple
> > body
> > and we want to decide on instrumentation at linktime, so we won't require
> > user
> > to rebuild with -fprofile-use, just relink.
Hi,
this patch fixes gcc.dg/tree-ssa/attr-alias.c on x86-64 target.
The problem here is that we attempt to create local alias since we consider
weakref
of test() to be overwritable and at the same time available.
The change in cgraph_function_body_availability makes weakref to properly
inherit
av
> > Bootstrapped/regtesed x86_64-linux. Martin, please can you review the
> > change?
> >
> > * ipa-prop.c (ipa_set_jf_known_type): Check that component type
> > is a record type with BINFO.
> > (detect_type_change_ssa): Add comp_type argument.
> > (compute_complex_assign_jump_fu
.
Bootstrapped/regtested x86_64-linux, comitted.
Honza
Index: ChangeLog
===
--- ChangeLog (revision 202127)
+++ ChangeLog (working copy)
@@ -1,3 +1,7 @@
+2013-08-31 Jan Hubicka
+
+ * basic-block.h (apply_scale): Make scale
===
--- ChangeLog (revision 202128)
+++ ChangeLog (working copy)
@@ -1,5 +1,9 @@
2013-08-31 Jan Hubicka
+ * cgraph.c (cgraph_speculative_call_info): Fix ref lookup
+
+2013-08-31 Jan Hubicka
+
* basic-block.h (apply_scale): Make
Hi,
as noticed by Martin Liska, -O0 -fprofile-generate code won't land into
__gcov_fork. This is due to early exit from expand_builtin.
Fixed and will be comitted as obvious.
Honza
* builtins.c (expand_builtin): Do not exit early for gcov
instrumented functions.
* gcc.dg/f
Hi,
With Martin we made script for testing the profiling failures.
First do
ld --verbose >~/script
then apply
--- /home/jh/script22013-08-31 17:59:11.0 +0200
+++ /home/jh/script 2013-08-31 17:39:40.0 +0200
@@ -1,12 +1,3 @@
-GNU ld (GNU Binutils for Debian) 2.20.1-system.2
Hi,
remember_with_vars walks trees that are read from file (now unique)
and looks for fields that can contain pointers to vars or functions and
if so, it records them to global hashtable for later fixup.
This is quite wasteful, because the hash is querried many times.
We can simply walk all fields
Hi,
I run my script on execute testsuite and looked into few testcases. The problem
I found
was roundoff errors - i.e. when expanding switch we set 50% chage that out of
bound
value is above or bellow. Both gets rounded to 0, because switch is executed
once
and the value is bellow.
Partly this
Hi,
this patch implement speculative devirtualization. It is a trivial pass that
asks for targets of every polymorphic call in a program and if the list
contains one likely target, it produces an speculative call. No context
sensitive analysis is done at the moment. This call may or may not survi
> Missing test cases?
Good point. The testcases I have needs rest of the patches from the series to
hit the mainline. For now I have added the following to test the basic scenario
* g++.dg/ipa/devirt-15.C: New testcase.
Index: g++.dg/ipa/devirt-15.C
==
Hi,
once simple pass has grown up into set of not-so-trivial profile based
transformations.
This patch merely moves it to separate file. I intend to add toplevel comment
and fix
interesting problem Martin Liska noticed in followup patch.
I did some basic housekeeping of include files and depend
> On 08/19/2013 10:01 AM, Jan Hubicka wrote:
> >+ /* All equivalent types, if more than one. */
> >+ vec *types;
> >+ /* Set of all equivalent types, if NON-NULL. */
> >+ pointer_set_t * GTY((skip)) types_set;
>
> Why do you need both a vector and a
> 2013-08-27 Martin Jambor
>
> PR ipa/58106
> * ipa-prop.c (ipa_edge_duplication_hook): Always put new rdesc to the
> linked list. When finding the correct duplicate, also consider also
> the caller in additon to its inlined_to node.
>
> testsuite/
> * gcc.dg/ipa
> On Aug 21, 2013, at 11:47 PM, Jan Hubicka wrote:
> >> The problem is that DECL_ARGUMENTS of the thunk (aka _ZThn528_N1D3fooEv)
> >> is used during thunk code-generation, and thunk code-generation happens
> >> during the output of D::foo.
>
> > I see, I
> > + tree a;
> > +
> > + if (in_lto_p)
> > +cgraph_get_body (node);
>
> That looks gross. It cannot possibly be the correct fix for this.
DECL_ARGUMENTS/DECL_RESULT are now part of function body. cgraph_get_body is
there to load it for you when you need it. We are going to expand the func
>
> Well, there is still fallout from this change so I'm not convinced
> this will stay
> this way. Also we stream the function-decl that refers to these fields in
As far as I know there are two problems
1) problem with the thunk expansion not getting DECL_ARGUMENTS/DECL_RESULT
addressed by
> >
> > Well, there is still fallout from this change so I'm not convinced
> > this will stay
> > this way. Also we stream the function-decl that refers to these fields in
>
> As far as I know there are two problems
> 1) problem with the thunk expansion not getting DECL_ARGUMENTS/DECL_RESULT
>
fallout I caused
this week.
Honza
Index: ChangeLog
===
--- ChangeLog (revision 202173)
+++ ChangeLog (working copy)
@@ -1,5 +1,10 @@
2013-08-31 Jan Hubicka
+ * lto.c (mentions_vars_p_field_decl
>
> But we still refer to the local entity from TREE_TYPE of the function decl,
> no?
depending on definition of local entity. I tought we was discussing if
PARM_DECL
is local or not.
I spent some time thining about the whole streaming scheme and I think we do
have
important side cases handle
Hi,
the following testcase illustrate problem with the offset. Sadly it ICEs even
w/o
LTO:
evans:/abuild/jh/trunk-3/build-inst12/gcc/:[1]# ./xgcc -B ./ -O2 ~/tt.c
/root/tt.c: In function 'main':
/root/tt.c:24:11: warning: overflow in implicit constant conversion [-Woverflow]
b.b=
Hi,
also to avoid the ICE in the original testcase, we do not really need the
DECL_ARGUMENTS/RESULT_DECL lists. All we need is RESULT_DECL in the global
stream. The following one liner fixes the testcase and all variants of my
ulitimate death testcase that did not ICE in tree from beggining of Au
Hi,
gcc.dg/lto/20090218-1.c contains cross module call into always_inline function.
At -O0 -flto we used to report nothing since optimize_inline_calls was not
called. With
my change we report it as error.
I am not sure what is desired behaviour, but for now I am restoring the previous
situation
Hi,
this patch fixes gcc.dg/lto/pr56297 failure. Here we have two modules defining
global variable assigned to hard registers. Those variables do not go into
assembler name hash because they have no real assembler name and consequentely
they are not merged.
We however may end up with two declarat
Hi,
reading profile/instrumenting breaks basic blocks and introduces fake edges.
The broken basic blocks are not re-merged until after LTO streaming that is
wasteful. Fixed thus.
Profiledbotostrapped/regtsted ppc64-linux, comitted.
Index: tree-profile.c
==
> > The decision what should go to local or global stream is basically
> > motivated by
> > 1) everything needed for interprocedural optimization has to be global
> > 2) everything related to function bodies should be local.
>
> I'd rather formulate it as "everything not needed at WPA time sh
> What is the footprint impact of speculative devirtualization?
It is less than 2% of text section and once we solve problems with ipa-prop
tracking,
I hope it will be less.
I hope we now understand better how to devirtualize and I think we can improve
non-speculative
devirt noticeably.
>
> F
> On Mon, 2013-09-02 at 14:35 +0200, Martin Jambor wrote:
> > Hi,
> >
> > On Fri, Aug 30, 2013 at 03:21:22PM -0400, David Malcolm wrote:
> > > Apart from the GTY aspect, how do people feel about the patch series?
> > > FWIW I have vague thoughts about doing something similar for tree -
> > > doing
Hi,
this is third fallout of my change to remove DECL_ARGUMENTS/DECL_RESULT for
functions w/o
bodies I did not really anticipate.
Here removal of the arguments changes mangling algorithm if
set_decl_assembler_name is invoked late. This is something I wanted to get rid
of for a long time: we alr
Hi,
as Michael pointed out, I introduced long lines primarily because of large
indentation in cgraph_analyze_function. It makes the code to look better
if the polymorphic call logic in broken out.
Bootstrapping/regtesting x86_64-linux, will commit it once testing conlcude.
Honza
* cgraphu
> On 09/04/2013 06:04 PM, Jan Hubicka wrote:
> > this is third fallout of my change to remove DECL_ARGUMENTS/DECL_RESULT for
> > functions w/o
> > bodies I did not really anticipate.
> [...]
> > I would like to basically ask if it seems to make sense to go this r
> On 09/04/2013 10:49 AM, Bernd Schmidt wrote:
> >On 09/04/2013 06:04 PM, Jan Hubicka wrote:
> >>this is third fallout of my change to remove DECL_ARGUMENTS/DECL_RESULT for
> >>functions w/o
> >>bodies I did not really anticipate.
> >[...]
> >&g
inlining
until we have IPA value range propagation.
Comitted as obvoius.
Honza
Index: ChangeLog
===
--- ChangeLog (revision 202296)
+++ ChangeLog (working copy)
@@ -1,3 +1,6 @@
+2013-09-05 Jan Hubicka
+
+ * gcc.dg/autopar
Hi,
ipa-devirt uses mangled names to establish equivalence in between symbols.
I added assert for cloned names, since I convinced myself that otherwise
we can miss matches in between local static symbols. I also convinced
myself we should never rename.
I was wrong on both: declaration merging mak
Hi,
this is the patch I commited after testing on x86_64-linux.
Honza
Index: ChangeLog
===
*** ChangeLog (revision 202271)
--- ChangeLog (working copy)
***
*** 1,3
--- 1,9
+ 2013-09-04 Jan Hubicka
> > +2013-09-05 Jan Hubicka
> > +
> > + * gcc.dg/autopar/pr49960.c: Disable partial inlining
> > 2013-09-05 Richard Biener
>
> Please add a blank line between entries?
The actual commmit has the extra line. Not sure why the patch doesn't.
Sorry for that.
Honza
> Hi,
>
> see http://gcc.gnu.org/ml/gcc/2013-09/msg00028.html for the context.
> The patch sets DECL_NO_INLINE_WARNING_P on the non-inlinable part after
> splitting (an alternative would be to clear DECL_DECLARED_INLINE_P).
Sorry, I missed your mail and it seems that my original mail did not
hit
Hi,
ipa-devirt-11.C (invented by me) checks for series of events that lead
to an ICE. With local alias changes these events are not happening for multiple
reasons. This patch solves first problem: we now redirect call to a function
to call to its alias and that breaks detection of self recursion
Hi,
this patch makes tree-sra to do its job in the case where function has an alias.
There were two problems; first recursion is not detected correctly and second
we did not see the callers and thus skipped the function. Rest of tree-sra
seems
to work as expected.
Bootstrapped/regtsted x86_64-li
Hi,
this patch fixes ICE with -O0 -fdevirtualize where we try to access type
inheritance
graph that does not exist.
Bootstrapped/regtested x86_64-linux, comitted.
2013-09-06 Jan Hubicka
PR tree-optimization/58311
* ipa-devirt.c (gate_ipa_devirt): Only execute when optimizing
Hi,
this patch makes inlining of functions called once to work even if they are
called
by alias.
Bootstrapped/regtested x86_64-linux, comitted.
PR middle-end/58094
* ipa-inline.c (has_caller_p): New function.
(want_inline_function_to_all_callers_p): Use it.
(sum_c
> >+ 2013-09-04 Jan Hubicka
> >+
> >+PR middle-end/58201
> >+* cgraphunit.c (analyze_functions): Clear AUX fields
> >+after processing; initialize assembler name has.
> >+
> I checked and double checked and with this commit a C++ test regres
Hi,
this is a variant of patch I tested and comitted after discussion on
DECL_ARGUMENTS change.
Basically ARGUMENTS are now part of a functio nbody and we need to stream them
for thunks
in order to be able to expand them. The patch also fixes misplaced pop_cfun in
lto-streamer-in.c.
Bootstrapp
> > >+ 2013-09-04 Jan Hubicka
> > >+
> > >+ PR middle-end/58201
> > >+ * cgraphunit.c (analyze_functions): Clear AUX fields
> > >+ after processing; initialize assembler name has.
> > >+
> > I checked and double checked and
> Hi Honza,
>
> and thanks for the analysis, now I understand the issue a little more.
>
> On 09/07/2013 10:28 AM, Jan Hubicka wrote:
> >So it is just an accident that the line info is output sanely (if
> >line 9 is sane, I don't exactly know)
> I would say t
> Hi all, Jakub,
>
> On 09/07/2013 01:16 PM, Jakub Jelinek wrote:
> >As I wrote in the PR, IMHO mangle_decl should
> > location_t save_location = input_location;
> > input_location = DECL_SOURCE_LOCATION (decl);
> >...
> > input_location = save_location;
> >around the call,
> I had a look an
> > Hi all, Jakub,
> >
> > On 09/07/2013 01:16 PM, Jakub Jelinek wrote:
> > >As I wrote in the PR, IMHO mangle_decl should
> > > location_t save_location = input_location;
> > > input_location = DECL_SOURCE_LOCATION (decl);
> > >...
> > > input_location = save_location;
> > >around the call,
Hi,
this patch adds API for varpool node removal/insertion hooks that are fully
symmetric
to cgraph code.
Bootstrapped/regtested x86_64-linux after working around PR58340, will commit
it shortly.
Honza
* cgraph.h (varpool_node_hook, varpool_node_hook_list,
varpool_add_node_rem
> On Sun, Sep 8, 2013 at 9:23 AM, Jan Hubicka wrote:
> > Hi,
> > this patch adds API for varpool node removal/insertion hooks that are fully
> > symmetric
> > to cgraph code.
> >
> > Bootstrapped/regtested x86_64-linux after working around PR58340, wil
Hi,
this patch make symtab_remove_unreachable_nodes to use polymorphic call analysis
same was as the cgraph construction code. The use is basically simple - instead
of marking all virtual functions as potentially reachable until inlining,
we mark only those that appear in the list of possible targ
> Andi Kleen writes:
>
> Ping!
>
> This problem is still open, and afaik no better solution has been
> proposed. It is also a regression.
>
> Is it ok to commit if I rerun the tests and they pass?
OK.
Honza
Hi,
this hack I put on place since ipa-prop was breaking without it. Martin fixed
the problem
and I tested it can now be removed for Firefx build.
Bootstrapped/regtested x86_64-linux, comitted.
* ipa-prop.c (try_make_edge_direct_simple_call): Do not special case
speculative edge
Hi,
this patch fixes ICE when edge redirection happens just before summary
generation.
At that time we still don't want edges to be redirected when underlying
statements
are not.
Bootstrapped/regtested x86_64-linux, comitted.
* ipa.c (walk_polymorphic_call_targets): Fix redirection befo
> Hi HJ,
> You were right, and the change from epilogue_size_needed to size_needed was
> the
> rootcause of the bug.
> Here is the obvious fix for that, regtested and bootstrapped on i386/x86_64.
>
> Is it OK for trunk?
>
> Changelog:
> 2013-09-09 Michael Zolotukhin
>
> * config/i386/i
Hi,
ipa-profile does propagation across the CFG in attempt to prove that function is
cold. This is counter-productive when profile is read and we can easilly work
this out from count itself. This patch makes it less agressive in this setting
and it also put ipa-profile into busyness to make coun
> Hi,
>
> until recently we did not have thunks with incoming edges in the call
> graph. However, speculative devirtualization can introduce them which
We did, just not very often. Normal devirtualization code can produce them,
too.
> may lead to creation of new SCCs in the call graph. What i
> Here's version 2 of this patch series.
>
> This is now 6 patches.
>
> The first two patches correspond to the two patches from the original
> patch series (and must both be applied together to make sense). They
> convert cgraph_node and varpool_node to inherit from symtab_node_base,
> eliminat
> This patch is the handwritten part of the conversion of these types
> to C++; it requires the followup patch, which is autogenerated.
>
> It converts:
> struct GTY(()) symtab_node_base
> to:
> class GTY((user)) symtab_node_base
>
> and converts:
> struct GTY(()) cgraph_node
> to:
> stru
> Amongst other things, the rename_symtab.py script converts
> "symtab_node" to "symtab_node *".
>
> This will lead to broken code on declarations that declare
> more than one variable (only the first would get a "*"), so split
> up such declarations.
>
> gcc/
> * cgraphunit.c (analyze_func
> This patch is 439KB in size, so to avoid mailing-list limitations I've
> uploaded it to:
> http://dmalcolm.fedorapeople.org/gcc/large-patches/fd607e77883b8d3f8fb7c8091d0626dbbd31b9a5-0002-Automated-conversion-of-symtab-to-class-hierarchy.patch
OK,
thanks!
Honza
> gcc/
>
> * ipa-ref.h (symtab_node): Remove typedef to pointer type, as it
> clashes with the preferred name for the base class.
> (const_symtab_node): Remove redundant typedef.
This is OK, too.
Actually it is not only clash - it was really inconsistent to write
symtab_node nod
> gcc/
>
> Patch autogenerated by rename_symtab.py from
> https://github.com/davidmalcolm/gcc-refactoring-scripts
> revision 714e57b66b08fb29f47826fbb642707bdfa19520
OK,
thanks!
Honza
> * cgraph.c (gt_ggc_mx): Update for renaming of symtab_node_base
> to symtab_node
> (gt_pch_nx): Likewise.
> ---
> gcc/cgraph.c | 24
> 1 file changed, 12 insertions(+), 12 deletions(-)
>
> diff --git a/gcc/cgraph.c b/gcc/cgraph.c
> index bb2626d..abbff
Hi,
fold_ctor_reference expects VAR or FUNCTION_DECL as FROM_DECL parameter.
gimple_get_virt_method_for_binfo instead passes expression
(pointer_plus_expr (addr_expr vtable_decl) (integer_cst XXX))
This bug makes can_refere_from_current_unit_p to mistakely think that
the references are safe because
Hi,
this patch adds polymorphic call targets into the boundary of an unit. This
enables
more late devirtualization. I briefly measured size on effect of WPA->ltrans
files
on firefox and they still seem to be around 2GB.
Bootstrapped/regtested x86_64-linux, comitted.
* lto-cgraph.c: Inc
Hi,
All the latent issues wthat ipa-devirt has triggered has been hopefully
fixed. So I am enabling ipa-devirt with LTO.
To answer original David's comment on non-LTO performance, I did not really
noticed
any significant improvements/dergradations at -O2 with our benchmarks (SPEC2k,
2k6
and C++
> Hi,
>
> PR 58389 is another case where the reference removal code, which is
> now also called from the edge removal hook, ICEs when inlined edges
> are not removed in topological order, which ins not the case when we
> are removing unreachable nodes.
>
> I'd like to fix it by the patch below.
Hi,
this is a streghtening of current ipa-devirt type walking I was working on in
past weeks.
My initial implementation of inheritance tree analysis simply take OTR_TYPE
that is the type of class of the polymorphic call and OTR_TOKEN that is an
index of virtual method in the vtable. It always wal
)
@@ -1,3 +1,8 @@
+2013-09-13 Jan Hubicka
+
+ * ipa-inline-analysis.c (struct growth_data): Add node.
+ (do_estimate_growth_1): Fix detection of recursion.
+
2013-09-13 Jakub Jelinek
PR tree-optimization/58392
Index: ipa-inline-analysis.c
Hi,
with improvements on devirtualization side, the thunks are getting more
problematic
for optimizers. Inliner won't inline them, profiling won't instrument them and
there are other places where they bock optimization.
This patch solves part of the problem by making gimple thunks to go same way
> 2013-09-13 Jakub Jelinek
>
> * ipa-prop.c (ipa_compute_jump_functions_for_edge): Return early
> for internal calls.
That seems resonable. I wonder if we want to consider internal calls to form
callgarph edges at all: perhaps we can just modify cgraph builder+verifier
to skip the
Hi,
this patch solves problem with can_refer_decl_in_current_unit_p always
returning true for static functions. The devirtualization code can look
up a method in anonymous namespace that has been optimized out due to
lack of references to any of vtables mentioning it. In this case we need
to make
> On Sun, Sep 15, 2013 at 09:08:00PM +0200, Jan Hubicka wrote:
> > > 2013-09-13 Jakub Jelinek
> > >
> > > * ipa-prop.c (ipa_compute_jump_functions_for_edge): Return early
> > > for internal calls.
> >
> > That seems resonable. I wonder
> The attached patch fixes the regression in gcc.dg/attr-ifunc-4.c (PR 58398).
>
> The problem is that the resolver function just looks like an alias, but it
> actually is
> something completely different. So inlining the resolver function has to be
> avoided.
>
> The patch was bootstrapped and
Hi,
this patch should fix HP-PA bootstrap issue where we create local aliases but
the target
has no support for them.
Bootstrapped/regtested x86_64-linux (with aliases disabled) and commited.
PR middle-end/58329
* ipa-devirt.c (ipa_devirt): Be ready for symtab_nonoverwritable_ali
Hi,
this patch makes inliner to not inline functions with -O0 optimization attribute
and also to not inline into functions.
Bootstrapped/regtested x86_64-linux, comitted.
PR middle-end/58332
* gcc.c-torture/compile/pr58332.c: New testcase.
* cif-code.def (FUNCTION_NOT_OPTIM
> > > +cgraph_node *cgn = static_cast (x);
> > > +gt_ggc_m_11cgraph_edge (cgn->callers);
> > > +gt_ggc_m_11cgraph_node (cgn->origin);
> > > +gt_ggc_m_11cgraph_node (cgn->nested);
> > > +gt_ggc_m_11cgraph_node (cgn->next_nested);
> > > +
> On Wed, Sep 18, 2013 at 10:57:57AM +0200, Richard Biener wrote:
> > On Wed, 18 Sep 2013, Richard Earnshaw wrote:
> >
> > > On 16/09/13 15:13, Richard Biener wrote:
> > > > +void
> > > > +get_range_info (tree name, double_int &min, double_int &max,
> > > > +enum value_range_type &
+1,8 @@
+2013-09-19 Jan Hubicka
+
+ * cgraph.c (cgraph_create_edge_1): Avoid uninitialized read
+ of speculative flag.
+
2013-09-19 Jakub Jelinek
* omp-low.c (expand_omp_sections): Always pass len - 1 to
Index: cgraph.c
>
>
> decide_alg is being called from ix86_expand_movmem, from
> expand_builtin_memcpy, for the call at line 61 of go-append.c.
> __builtin_memcpy (n, a.__values, a.__count * element_size);
>
> I'm continuing to look.
Indeed it is problem of this patch - the issue is that generic64 had
du
>
> I did some experiment with code alignment. I found
> -fno-align-loops -fno-align-functions -fno-align-jumps
> had no negative performance impacts on current
> Intel processors while reducing code sizes by 1-2%.
> Should we use
>
> {&generic_cost, 0, 0, 0, 0, 0},
>
> instead?
Good, revisitin
> This patch is the handwritten part of the conversion of these types
> to C++; it requires the followup patch, which is autogenerated.
>
> It converts:
> struct symtab_node_base
> to:
> class symtab_node_base
>
> and converts:
> struct cgraph_node
> to:
> struct cgraph_node : public symt
> On Fri, Sep 20, 2013 at 6:00 AM, Zamyatin, Igor
> wrote:
> >
> > For x86 32-bit is still important due to mobile market so we would like to
> > make some measurements first to see how this new generic model works on 32
> > bits
I assume by mobile market you mean atom CPU these days (not Pent
Hi,
this is upated version of patch discussed at
http://gcc.gnu.org/ml/gcc-patches/2012-12/msg00841.html
It makes CORE tuning to more follow the optimization guidelines.
In particular it removes some tuning flags for features I implemented years
back specifically for K7/K8 chips that ended up in
> On Sat, Sep 21, 2013 at 3:51 PM, Xinliang David Li wrote:
> > On Sat, Sep 21, 2013 at 12:54 PM, Jan Hubicka wrote:
> >> Hi,
> >> this is upated version of patch discussed at
> >> http://gcc.gnu.org/ml/gcc-patches/2012-12/msg00841.html
> >>
> 2013-09-16 Wei Mi
>
> * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Separate
> PROCESSOR_COREI7_AVX out from PROCESSOR_COREI7.
> * gcc/config/i386/i386.c (ix86_option_override_internal): Ditto.
> (ix86_issue_rate): Ditto.
> (ia32_multipass_d
>
> I am not sure if AMD hardware has any limitations here. It fuses only
> cmp/test
This is what Agner Fog says:
A CMP or TEST instruction immediately followed by a conditional jump can be
fused into a single macro-op. This applies to all versions of the CMP and TEST
instructions and all cond
> Hello Richard,
>
> Not directly related to your patch but...
>
> On Sun, Sep 22, 2013 at 12:54 PM, Richard Sandiford wrote:
> > @@ -588,14 +589,17 @@ cond_exec_process_if_block (ce_if_block_
> > goto fail;
> > #endif
> >
> > - true_prob_val = find_reg_note (BB_END (test_bb), REG_BR_PROB,
Hi,
this is the last patch of the series.
This patch adds sanity check that all devirtualizations are among ones
predicted by possible_polymorphic_call_targets. This sanity check was
very useful to chase out numberous problems in this area.
The patch check the type based devirtualization to go i
601 - 700 of 5417 matches
Mail list logo
