Re: [PATCH] Make devirtualization use BINFO_VTABLE instead of BINFO_VIRTUALS

[Richard Guenther <richard.guenther@gmail.com>]

On Thu, Sep 1, 2011 at 8:52 PM, Jan Hubicka <hubicka@ucw.cz> wrote:
>> - Nevertheless, this method of devirtualization cannot automatically
>>   de-thunkize this-adjusting thunks and newly direct calls to them
>>   cannot be inlined because the inliner does not have this capability
>>   now.  This is in fact a regression from 4.6, and testcases
>>   ivinline-7.C and ivinline-9.C had to be XFAILed exactly for this
>>   reason.  The size of the memory savings and the fact that we do not
>>   devirtualize that much now make this an acceptable tradeoff, though.
>
> OK, the "dethunkization" was ugly anyway.  I guess we will need to add support
> for handling thunks in inliner but that can go incrementally.
>
>> 2011-08-31  Martin Jambor  <mjambor@suse.cz>
>>
>>       * cgraph.h (cgraph_indirect_call_info): Removed field thunk_delta.
>>       * gimple-fold.c (gimple_get_virt_method_for_binfo): Rewritten to use
>>       BINFO_VTABLE.  Parameter delta removed, all callers updated.
>>       * tree.c (free_lang_data_in_binfo): Clear BINFO_VIRTUALs instead
>>       BINFO_VTABLE.
>>       * cgraph.c (cgraph_make_edge_direct): Removed parameter delta, updated
>>       all calls.
>>       * cgraphunit.c (cgraph_redirect_edge_call_stmt_to_callee): Removed
>>       handling of thunk_delta.
>>       * ipa-cp.c (get_indirect_edge_target): Removed parameter delta.
>>       (devirtualization_time_bonus): Do not handle thunk deltas.
>>       (ipcp_discover_new_direct_edges): Likewise.
>>       * ipa-prop.c (ipa_make_edge_direct_to_target): Likewise.
>>       (try_make_edge_direct_simple_call): Likewise.
>>       (try_make_edge_direct_virtual_call): Likewise.
>>       * lto-cgraph.c (output_cgraph_opt_summary_p): Likewise.  Mark
>>       parameter set as unused.
>>       (output_edge_opt_summary): Likewise.  Mark both parameters as unused.
>>       * lto-cgraph.c (output_cgraph_opt_summary_p): Likewise.  Mark
>>       parameter set as unused.
>>       (output_edge_opt_summary): Likewise.  Mark both parameters as unused.
>>       (input_edge_opt_summary): Likewise.
>>       * lto-streamer-out.c (lto_output_ts_binfo_tree_pointers): Do not stream
>>       BINFO_VIRTUALS at all.
>>       * lto-streamer-in.c (lto_input_ts_binfo_tree_pointers): Likewise.
>>
>>       * testsuite/g++.dg/ipa/devirt-3.C: Added a distraction method.
>>       * testsuite/g++.dg/ipa/ivinline-7.C: Added a test for direct call
>>       discovery, xfailed test for inlining.
>>       * testsuite/g++.dg/ipa/ivinline-9.C: Likewise.
>
> The cgraph bits of the patch are OK, but I will leave the gimple-fold bits for Richi's approval.

Ok.

Thanks,
Richard.

> Can't we drop computation of BINFO_VIRTUALS from C++ FE completely now?
>
> Honza
>