This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Do not gather mem stats in run_exit_handles (PR middle-end/69919)


On Thu, 2016-02-25 at 12:07 +0100, Martin LiÅka wrote:
> Hello.
> 
> As discussed in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69919#c3
> , following patch
> guards usage of memory statistics infrastructure after a
> mem_alloc_description is already
> destructed.
> 
> Patch can lto-bootstrap on x86_64-linux-gnu with --enable-gather
> -detailed-mem-stats and
> survives bootstrap&regtests with --enable-gather-detailed-mem-stats
> on x86_64-linux-gnu with.
> 
> Ready to be installed?
> Martin
> 
> gcc/ChangeLog:
> 
> 2016-02-25  Martin Liska  <mliska@suse.cz>
> 
> 	PR middle-end/69919
> 	* alloc-pool.c (after_memory_report): New variable.
> 	* alloc-pool.h (base_pool_allocator ::release): Do not use
> 	the infrastructure if after_memory_report.
> 	* toplev.c (toplev::main): Mark after memory report.
> ---
>  gcc/alloc-pool.c | 1 +
>  gcc/alloc-pool.h | 5 ++++-
>  gcc/toplev.c     | 3 +++
>  3 files changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/gcc/alloc-pool.c b/gcc/alloc-pool.c
> index ae5e232..43d06f6 100644
> --- a/gcc/alloc-pool.c
> +++ b/gcc/alloc-pool.c
> @@ -25,6 +25,7 @@ along with GCC; see the file COPYING3.  If not see
>  
>  ALLOC_POOL_ID_TYPE last_id;
>  mem_alloc_description<pool_usage> pool_allocator_usage;
> +bool after_memory_report = false;
>  
>  /* Output per-alloc_pool memory usage statistics.  */
>  void
> diff --git a/gcc/alloc-pool.h b/gcc/alloc-pool.h
> index 8ccf089..3ead101 100644
> --- a/gcc/alloc-pool.h
> +++ b/gcc/alloc-pool.h
> @@ -25,6 +25,9 @@ along with GCC; see the file COPYING3.  If not see
>  
>  extern void dump_alloc_pool_statistics (void);
>  
> +/* Flag indicates whether memory statistics are gathered any longer.
>   */
> +extern bool after_memory_report;
> +
>  typedef unsigned long ALLOC_POOL_ID_TYPE;
>  
>  /* Last used ID.  */
> @@ -306,7 +309,7 @@ base_pool_allocator <TBlockAllocator>::release ()
>        TBlockAllocator::release (block);
>      }
>  
> -  if (GATHER_STATISTICS)
> +  if (GATHER_STATISTICS && !after_memory_report)
>      {
>        pool_allocator_usage.release_instance_overhead
>  	(this, (m_elts_allocated - m_elts_free) * m_elt_size);
> diff --git a/gcc/toplev.c b/gcc/toplev.c
> index 28c115d..c480bfc 100644
> --- a/gcc/toplev.c
> +++ b/gcc/toplev.c
> @@ -2107,6 +2107,9 @@ toplev::main (int argc, char **argv)
>  
>    finalize_plugins ();
>    location_adhoc_data_fini (line_table);
> +
> +  after_memory_report = true;
> +
>    if (seen_error () || werrorcount)
>      return (FATAL_EXIT_CODE);

Was this tested with "jit" in --enable-languages?   I ask because
toplev::main can be run repeatedly by libgccjit, and it looks like
nothing can reset after_memory_report.  (Though I don't typically build
with --enable-gather-detailed-mem-stats, and the jit docs don't mention
it).


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]