Re: [PATCH] Report failed block copies in the backward threader [PR126906]

On Mon, 17 Aug 2026, Aldy Hernandez wrote:

> When back_jt_path_registry::duplicate_thread_path fails, the
> registered path was dropped silently: the dump shows "Registering jump
> thread" with no matching outcome.  Every other way a thread dies is
> reported -- the profitability FAILs, and the registry's other drops
> via cancel_thread.  Report why the copy failed through cancel_thread
> like everywhere else.
> 
> Tested on ppc64le Linux.
> 
> OK?

It conflicts with my pushed patch, but otherwise OK.  Note I really
want some TLC here, this "late" failing in random places makes
hunting down the various checks difficult ...

> gcc/ChangeLog:
> 	PR tree-optimization/126906
> 
> 	* tree-ssa-threadupdate.h
> 	(back_jt_path_registry::duplicate_thread_path): Add failure_reason
> 	argument.
> 	* tree-ssa-threadupdate.cc
> 	(back_jt_path_registry::duplicate_thread_path): Set
> 	*FAILURE_REASON on the two failing exits.
> 	(back_jt_path_registry::update_cfg): Cancel the path with the
> 	reason when duplicate_thread_path fails.
> ---
>  gcc/tree-ssa-threadupdate.cc | 24 ++++++++++++++++++------
>  gcc/tree-ssa-threadupdate.h  |  3 ++-
>  2 files changed, 20 insertions(+), 7 deletions(-)
> 
> diff --git a/gcc/tree-ssa-threadupdate.cc b/gcc/tree-ssa-threadupdate.cc
> index db3520b42fc..73a995815c0 100644
> --- a/gcc/tree-ssa-threadupdate.cc
> +++ b/gcc/tree-ssa-threadupdate.cc
> @@ -2382,14 +2382,16 @@ back_jt_path_registry::adjust_paths_after_duplication (unsigned curr_path_num)
>     CURRENT_PATH_NO is an index into the global paths[] table
>     specifying the jump-thread path.
>  
> -   Returns false if it is unable to copy the region, true otherwise.  */
> +   Returns false if it is unable to copy the region, true otherwise.
> +   On failure *FAILURE_REASON says why.  */
>  
>  bool
>  back_jt_path_registry::duplicate_thread_path (edge entry,
>  					      edge exit,
>  					      basic_block *region,
>  					      unsigned n_region,
> -					      unsigned current_path_no)
> +					      unsigned current_path_no,
> +					      const char **failure_reason)
>  {
>    unsigned i;
>    class loop *loop = entry->dest->loop_father;
> @@ -2398,7 +2400,10 @@ back_jt_path_registry::duplicate_thread_path (edge entry,
>    profile_count curr_count;
>  
>    if (!can_copy_bbs_p (region, n_region))
> -    return false;
> +    {
> +      *failure_reason = "Cannot copy the blocks in the path";
> +      return false;
> +    }
>  
>    /* Some sanity checking.  Note that we do not check for all possible
>       missuses of the functions.  I.e. if you ask to copy something weird,
> @@ -2409,7 +2414,10 @@ back_jt_path_registry::duplicate_thread_path (edge entry,
>        /* We do not handle subloops, i.e. all the blocks must belong to the
>  	 same loop.  */
>        if (region[i]->loop_father != loop)
> -	return false;
> +	{
> +	  *failure_reason = "Path crosses loops";
> +	  return false;
> +	}
>      }
>  
>    initialize_original_copy_tables ();
> @@ -2638,16 +2646,20 @@ back_jt_path_registry::update_cfg (bool /*peel_loop_headers*/)
>        for (unsigned int j = 0; j < len - 1; j++)
>  	region[j] = (*path)[j]->e->dest;
>  
> -      if (duplicate_thread_path (entry, exit, region, len - 1, 0))
> +      const char *failure_reason = NULL;
> +      if (duplicate_thread_path (entry, exit, region, len - 1, 0,
> +				 &failure_reason))
>  	{
>  	  /* We do not update dominance info.  */
>  	  free_dominance_info (CDI_DOMINATORS);
>  	  visited_starting_edges.add (entry);
>  	  retval = true;
>  	  m_num_threaded_edges++;
> +	  path->release ();
>  	}
> +      else
> +	cancel_thread (path, failure_reason);
>  
> -      path->release ();
>        m_paths.unordered_remove (0);
>        free (region);
>      }
> diff --git a/gcc/tree-ssa-threadupdate.h b/gcc/tree-ssa-threadupdate.h
> index 7faac34fb8f..624bc10a64d 100644
> --- a/gcc/tree-ssa-threadupdate.h
> +++ b/gcc/tree-ssa-threadupdate.h
> @@ -116,7 +116,8 @@ private:
>    bool update_cfg (bool peel_loop_headers) override;
>    void adjust_paths_after_duplication (unsigned curr_path_num);
>    bool duplicate_thread_path (edge entry, edge exit, basic_block *region,
> -			      unsigned n_region, unsigned current_path_no);
> +			      unsigned n_region, unsigned current_path_no,
> +			      const char **failure_reason);
>    bool rewire_first_differing_edge (unsigned path_num, unsigned edge_num);
>  };
>  
> 

-- 
Richard Biener <[email protected]>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Jochen Jaser, Andrew McDonald, Abhinav Puri; (HRB 36809, AG Nuernberg)