On Wed, Oct 5, 2011 at 12:29 AM, Richard Henderson <r...@redhat.com> wrote:
> On 10/04/2011 03:10 PM, Bernd Schmidt wrote:
>>       * doc/invoke.texi (-fshrink-wrap): Document.
>>       * opts.c (default_options_table): Add it.
>>       * common.opt (fshrink-wrap): Add.
>>       * function.c (emit_return_into_block): Remove useless declaration.
>>       (record_hard_reg_uses_1, record_hard_reg_uses, frame_required_for_rtx,
>>       requires_stack_frame_p, gen_return_pattern): New static functions.
>>       (emit_return_into_block): New arg simple_p.  All callers changed.
>>       Use gen_return_pattern.
>>       (thread_prologue_and_epilogue_insns): Implement shrink-wrapping.
>>       * config/i386/i386.md (return): Expand into a simple_return.
>>       (simple_return): New expander):
>>       (simple_return_internal, simple_return_internal_long,
>>       simple_return_pop_internal_long, simple_return_indirect_internal):
>>       Renamed from return_internal, return_internal_long,
>>       return_pop_internal_long and return_indirect_internal; changed to use
>>       simple_return.
>>       * config/i386/i386.c (ix86_expand_epilogue): Adjust to expand
>>       simple returns.
>>       (ix86_pad_returns): Likewise.
>>       * function.h (struct rtl_data): Add member shrink_wrapped.
>>       * cfgcleanup.c (outgoing_edges_match): If shrink-wrapped, edges that
>>       are not jumps or sibcalls can't be compared.
>>
>>       * gcc.target/i386/sw-1.c: New test.
>
> Ok.
>
> As a followup, I think this option needs to be disabled for profiling
> and profile_after_prologue.  Should be a mere matter of frobbing the
> options at startup.

This breaks bootstrap on x86_64-linux.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50621

Richard.

>
> r~
>

Reply via email to