This is the mail archive of the 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: [GOOGLE] Workaround a bug in AutoFDO which may leads to infinite loop for inlined assembly

On Sun, Apr 21, 2013 at 4:51 PM, Dehao Chen <> wrote:
> This patch fixed a bug in getting inline stacks: if there is no
> location info attached to a block, we should *not* try to get its
> function name because it could result in infinite loop.

Is the infinite loop within get_function_decl_from_block itself? If
so, for extra safety perhaps that routine should check if the location
is unknown and return early if it is. One other minor comment below,
otherwise it looks ok to me.


> Bootstrapped and passed all regression tests.
> Ok for gcc-4_7 branch?
> Thanks,
> Dehao
> Index: gcc/auto-profile.c
> ===================================================================
> --- gcc/auto-profile.c (revision 198117)
> +++ gcc/auto-profile.c (working copy)
> @@ -662,9 +662,10 @@ get_inline_stack_by_stmt (gimple stmt, tree decl,
>         block && (TREE_CODE (block) == BLOCK);
>         block = BLOCK_SUPERCONTEXT (block))
>      {
> -      tree decl = get_function_decl_from_block (block);
> +      tree decl;
>   continue;
> +      decl = get_function_decl_from_block (block);
>        loc = BLOCK_SOURCE_LOCATION (block);

Do you want to move up the assignment to loc and use loc in the new if

>        pos_stack[idx].file = expand_location (loc).file;
>        pos_stack[idx].line = expand_location (loc).line;

Teresa Johnson | Software Engineer | | 408-460-2413

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