[PATCH] IPA: fix reproducibility in IPA MOD REF

[Martin Liška]

On 11/18/21 19:22, Jan Hubicka wrote:
>> Supported LTO compression algorithms: zlib zstd
>> gcc version 12.0.0 20211118 (experimental) (GCC)
>> /usr/bin/ld: ./xxx.ltrans0.ltrans.o: warning: relocation against `lm_read_ctl_dict_size_n_lmclass_used' in read-only section `.text'
>> /usr/bin/ld: ./xxx.ltrans0.ltrans.o: relocation R_X86_64_PC32 against symbol `__ckd_calloc___elem_size' can not be used when making a shared object; recompile with -fPIC
>> /usr/bin/ld: final link failed: bad value
>> collect2: error: ld returned 1 exit status
>> /usr/bin/ld: ./yyy.ltrans0.ltrans.o: warning: relocation against `__ckd_calloc___n_elem' in read-only section `.text'
>> /usr/bin/ld: ./yyy.ltrans0.ltrans.o: relocation R_X86_64_PC32 against symbol `__ckd_calloc___elem_size' can not be used when making a shared object; recompile with -fPIC
>> /usr/bin/ld: final link failed: bad value
>> collect2: error: ld returned 1 exit status
>> diff: yyy.ltrans0.ltrans*optimized: No such file or directory
>> 54,55d53
>> < 	movslq	lm_read_ctl_dict_size_n_lmclass_used(%rip), %rax
>> < 	movl	$0, 0(%rbp,%rax,4)
>>
>> I tracked that it differs in tree DSE dump.
>>
>> May I install the patch?
> 
> No, I think it is bug of symbol_summary that get is causing a
> difference.

Isn't problem that the following code

               past_flags.reserve_exact (summary->arg_flags.length ());
               past_flags.splice (summary->arg_flags);
               past_retslot_flags = summary->retslot_flags;

is guarded with if (dump_file && ... )?



   It should be pure function. I think it is:
>    /* Getter for summary callgraph node pointer.  */
>    T* get (cgraph_node *node) ATTRIBUTE_PURE
>      {
>         return exists (node) ? (*m_vector)[node->get_summary_id ()] : NULL;
>      }
> It should not be using get_summary_id (which allocates it for no good
> reaosn) and simply check that summary_id is non-negative.

   /* Get summary id of the node.  */
   inline int get_summary_id ()
   {
     return m_summary_id;
   }

Where do you see the allocation?

Martin

> 
> Still I wonder how this can make code different - that looks like
> another bug somewhere.
> 
> Honza
>