[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 Martin Jambor jamborm at gcc dot gnu.org changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #17 from Martin Jambor jamborm at gcc dot gnu.org --- Author: jamborm Date: Mon May 13 10:16:09 2013 New Revision: 198821 URL: http://gcc.gnu.org/viewcvs?rev=198821root=gccview=rev Log: 2013-05-13 Martin Jambor mjam...@suse.cz PR middle-end/42371 * ipa-prop.h (IPA_UNDESCRIBED_USE): New macro. (ipa_constant_data): New type. (ipa_jump_func): Use ipa_constant_data to hold information about constant jump functions. (ipa_get_jf_constant): Adjust to jump function type changes. (ipa_get_jf_constant_rdesc): New function. (ipa_param_descriptor): New field controlled_uses. (ipa_get_controlled_uses): New function. (ipa_set_controlled_uses): Likewise. * ipa-ref.h (ipa_find_reference): Declare. * ipa-prop.c (ipa_cst_ref_desc): New type. (ipa_print_node_jump_functions_for_edge): Adjust for jump function type changes. (ipa_set_jf_constant): Likewise. Also create reference descriptions. New parameter cs. Adjust all callers. (ipa_analyze_params_uses): Detect uncontrolled and controlled uses. (remove_described_reference): New function. (jfunc_rdesc_usable): Likewise. (try_make_edge_direct_simple_call): Decrement controlled use count, attempt to remove reference if it hits zero. (combine_controlled_uses_counters): New function. (propagate_controlled_uses): Likewise. (ipa_propagate_indirect_call_infos): Call propagate_controlled_uses. (ipa_edge_duplication_hook): Duplicate reference descriptions. (ipa_print_node_params): Print described use counter. (ipa_write_jump_function): Adjust to jump function type changes. (ipa_read_jump_function): New parameter CS, pass it to ipa_set_jf_constant. Adjust caller. (ipa_write_node_info): Stream controlled use count (ipa_read_node_info): Likewise. * cgraph.c (cgraph_mark_address_taken_node): Bail out instead of asserting. * ipa-cp.c (ipcp_discover_new_direct_edges): Decrement controlled use count. Remove cloning-added reference if it reaches zero. * ipa-ref.c (ipa_find_reference): New function. testsuite/ * gcc.dg/ipa/remref-0.c: New test. * gcc.dg/ipa/remref-1a.c: Likewise. * gcc.dg/ipa/remref-1b.c: Likewise. * gcc.dg/ipa/remref-2a.c: Likewise. * gcc.dg/ipa/remref-2b.c: Likewise. Added: trunk/gcc/testsuite/gcc.dg/ipa/remref-0.c trunk/gcc/testsuite/gcc.dg/ipa/remref-1a.c trunk/gcc/testsuite/gcc.dg/ipa/remref-1b.c trunk/gcc/testsuite/gcc.dg/ipa/remref-2a.c trunk/gcc/testsuite/gcc.dg/ipa/remref-2b.c Modified: trunk/gcc/ChangeLog trunk/gcc/cgraph.c trunk/gcc/ipa-cp.c trunk/gcc/ipa-prop.c trunk/gcc/ipa-prop.h trunk/gcc/ipa-ref.c trunk/gcc/ipa-ref.h trunk/gcc/testsuite/ChangeLog
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 Martin Jambor jamborm at gcc dot gnu.org changed: What|Removed |Added URL||http://gcc.gnu.org/ml/gcc-p ||atches/2013-04/msg01032.htm ||l CC||jamborm at gcc dot gnu.org --- Comment #16 from Martin Jambor jamborm at gcc dot gnu.org 2013-04-17 15:58:17 UTC --- I have submitted a patch to address this issue: http://gcc.gnu.org/ml/gcc-patches/2013-04/msg01032.html
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #15 from Jan Hubicka hubicka at ucw dot cz 2013-01-19 21:48:12 UTC --- Clearing of address-taken does not work: two/1 (two) @0x7fafc0e29818 Type: function Visibility: prevailing_def_ironly Address is taken. References: Referring: main/3 (addr) Availability: available Function flags: analyzed body finalized Called by: Calls: that is, we do not re-compute address-taken after IPA inline transform stage and thus do not reclaim indirectly inlined functions that we no longer need. We can't recompute here because inlining transform happens during the final compilation stage, so the functions are already beging output to the assembly. IPA-prop needs to realize that indirect inlining is going to optimize out the address operation and remove the reference. I assume this is not too hard to implement - we only need to record when the function parameter is used _only_ of the stuff represented by jump functions and indirect edges. I was trying to convince Martin to implement this for a while :) Virtual functions are ven harder than indirect calls since we need to figure out the vtable will become unreachable at the IPA stage. Honza
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 Richard Biener rguenth at gcc dot gnu.org changed: What|Removed |Added Status|REOPENED|NEW --- Comment #14 from Richard Biener rguenth at gcc dot gnu.org 2013-01-18 10:25:04 UTC --- Clearing of address-taken does not work: two/1 (two) @0x7fafc0e29818 Type: function Visibility: prevailing_def_ironly Address is taken. References: Referring: main/3 (addr) Availability: available Function flags: analyzed body finalized Called by: Calls: that is, we do not re-compute address-taken after IPA inline transform stage and thus do not reclaim indirectly inlined functions that we no longer need.
Re: [Bug middle-end/42371] dead code not eliminated during folding with whole-program
Still a problem in latest 4.7, google/4.7, and trunk (4.8). So 4.6 was working but 4.7 isn't?
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #12 from Jan Hubicka hubicka at ucw dot cz 2013-01-17 10:27:10 UTC --- Still a problem in latest 4.7, google/4.7, and trunk (4.8). So 4.6 was working but 4.7 isn't?
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #13 from Matt Hargett matt at use dot net 2013-01-17 18:28:18 UTC --- No. 4.6 doesn't devirt (at -O2 or -O3) and therefore the DCE isn't relevant. At both -O2 and -O3, with and without -fwhole-program, both with and without adjustin declarations one()/two() to one(void)/two(void): 4.7 and google/4.7 both devirt correctly but the DCE on the function bodies doesn't happen. 4.8 also devirts correctly, but the DCE on the function bodies doesn't happen.
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 Matt Hargett matt at use dot net changed: What|Removed |Added Status|RESOLVED|REOPENED Resolution|FIXED | --- Comment #11 from Matt Hargett matt at use dot net 2013-01-17 00:09:32 UTC --- Still a problem in latest 4.7, google/4.7, and trunk (4.8).
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 Matt Hargett matt at use dot net changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED --- Comment #10 from Matt Hargett matt at use dot net 2011-04-12 18:29:28 UTC --- I just tested and it looks like this is actually fixed in 4.6 for this specific example and another one I had. I'm still seeing remnants with C++, but I'll file a separate bug for that.
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #9 from Matt Hargett matt at use dot net 2011-03-16 21:43:56 UTC --- Now that 4.7 stage 1 is open, let me know if there's anything else I can reasonably provide to provide examples, testing, etc. Thanks!
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 Jan Hubicka hubicka at gcc dot gnu.org changed: What|Removed |Added AssignedTo|hubicka at gcc dot gnu.org |mjambor at suse dot cz --- Comment #8 from Jan Hubicka hubicka at gcc dot gnu.org 2011-01-12 10:43:13 UTC --- Hi, the last problem was solved, sadly it is not last. We still keep the functions around because we don't see that indirect inlining effectively removes the reference. Martin probably should look into this for 4.7.
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #7 from Matt Hargett matt at use dot net 2011-01-11 23:44:22 UTC --- Is the last remaining issue with this test case fixed by the patch for PR46076?
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #5 from Matt Hargett matt at use dot net 2010-11-17 21:43:13 UTC --- GCC 4.6.0.20101116 still shows this problem, but also introduces new problems on top of it. Let me know if I should file a new bug for the regressions. Either way, is there any update on this enhancement getting into 4.6? Thanks!
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371 --- Comment #6 from Jan Hubicka hubicka at ucw dot cz 2010-11-17 23:01:58 UTC --- The problem with not recomputing address taken flag is solved. The testcase still failed to iling because of type mismatch (i.e. one() instead of one(void)). You probably could fix your sources there to work around this. Whether this will be fixed is Richard's call (we need to make constant propagation to take away the cast I guess). Technically this is regression to 4.3. Yes, please go ahead and report the new problems so there is enough time to fix them.
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
--- Comment #4 from matt at use dot net 2010-06-07 19:46 --- Let me know when this is implemented on trunk (preferrably by marking this report as resolved) and I'll test my proprietary test cases here. Thanks! -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
--- Comment #2 from hubicka at gcc dot gnu dot org 2010-01-22 19:09 --- Note that testcase mises static in front of one/two. This is something I plan to implement for next release. Last release has some preparation work for it, but at the moment we don't really build may edges nor other datastructures related with indirect calls. Honza -- hubicka at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |hubicka at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2009-12-15 13:38:42 |2010-01-22 19:09:51 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
--- Comment #3 from hubicka at gcc dot gnu dot org 2010-01-22 19:11 --- really an enhancement rather than bug. -- hubicka at gcc dot gnu dot org changed: What|Removed |Added Severity|normal |enhancement Last reconfirmed|2010-01-22 19:09:51 |2010-01-22 19:11:10 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371
[Bug middle-end/42371] dead code not eliminated during folding with whole-program
--- Comment #1 from rguenth at gcc dot gnu dot org 2009-12-15 13:38 --- This is because we never re-compute the address-taken flag of functions. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||hubicka at gcc dot gnu dot ||org Status|UNCONFIRMED |NEW Ever Confirmed|0 |1 Keywords||missed-optimization Last reconfirmed|-00-00 00:00:00 |2009-12-15 13:38:42 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42371
