[Patch,committed] Fix PR/target 32313, MIPS bootstrap failure running gengtype in stage 2.
David Daney
ddaney@avtrex.com
Mon Jun 18 20:03:00 GMT 2007
Richard Sandiford wrote:
> David Daney <ddaney@avtrex.com> writes:
>
>> Committed as approved by Eric Christopher in the PR:
>>
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32313#c7
>>
>> 2007-06-18 David Daney <ddaney@avtrex.com
>>
>> PR target/32313
>> * config/mips/mips.c (mips_expand_call): Mark $gp as used by
>> local function call.
>>
>> Index: config/mips/mips.c
>> ===================================================================
>> --- config/mips/mips.c (revision 125765)
>> +++ config/mips/mips.c (working copy)
>> @@ -3562,8 +3562,13 @@ mips_expand_call (rtx result, rtx addr,
>>
>> insn = emit_call_insn (pattern);
>>
>> - /* Lazy-binding stubs require $gp to be valid on entry. */
>> - if (mips_ok_for_lazy_binding_p (orig_addr))
>> + /* Lazy-binding stubs require $gp to be valid on entry. So does the
>> + case of calling a local function with TARGET_ABICALLS and
>> + -mno-shared (a.k.a. TARGET_ABSOLUTE_ABICALLS). */
>> + if (mips_ok_for_lazy_binding_p (orig_addr)
>> + || (TARGET_ABSOLUTE_ABICALLS
>> + && (GET_CODE (orig_addr) == SYMBOL_REF)
>> + && targetm.binds_local_p (SYMBOL_REF_DECL (orig_addr))))
>> use_reg (&CALL_INSN_FUNCTION_USAGE (insn), pic_offset_table_rtx);
>> }
>>
>
> I don't think this is right. Locally-binding functions for -mno-shared
> should never use lazy-binding stubs, so should never need $gp to be
> initialised on entry. (Functions called directly should intialise $gp
> if they need it.) In the PR, why does get_output_file_with_visibility
> need $gp to be initialised on entry?
>
>
Sorry for the lack of clear thought in this case. I reverted the patch.
2007-06-18 David Daney <ddaney@avtrex.com
Revert:
2007-06-18 David Daney <ddaney@avtrex.com
PR target/32313
* config/mips/mips.c (mips_expand_call): Mark $gp as used by
local function call.
More information about the Gcc-patches
mailing list