Ping
On Mon, May 19, 2025 at 10:06 AM liuhongt <hongtao....@intel.com> wrote:
>
> From: "hongtao.liu" <hongtao....@intel.com>
>
> AutoFDO profile is a scaled profile, as a result, 0 sample does not
> mean never executed. especially there's profile from function
> body. Prevent combine_with_ipa_count·(ipa_count) from zeroing all
> bb->count.
>
> Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}
> OK for trunk?
>
> gcc/ChangeLog:
>
> PR gcov-profile/118551
> * predict.cc (estimate_bb_frequencies): Prevent
> combine_with_ipa_count·(ipa_count) from zeroing all bb->count
> when function body is known to be hot.
> ---
> gcc/predict.cc | 73 +++++++++++++++++++++++++++++---------------------
> 1 file changed, 42 insertions(+), 31 deletions(-)
>
> diff --git a/gcc/predict.cc b/gcc/predict.cc
> index ef31c48bfe2..7d4bf5261ad 100644
> --- a/gcc/predict.cc
> +++ b/gcc/predict.cc
> @@ -4105,40 +4105,51 @@ estimate_bb_frequencies ()
> if (freq_max < 16)
> freq_max = 16;
> profile_count ipa_count = ENTRY_BLOCK_PTR_FOR_FN (cfun)->count.ipa ();
> - cfun->cfg->count_max = profile_count::uninitialized ();
> - FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR_FOR_FN (cfun), NULL, next_bb)
> +
> + /* AutoFDO profile is a scaled profile, as a result, 0 sample does not
> + mean never executed. especially there's profile from function
> + body. Prevent combine_with_ipa_count·(ipa_count) from zeroing all
> + bb->count. */
> + if (!(ipa_count.quality () == AFDO
> + && cfun->cfg->count_max.quality () == AFDO
> + && !ipa_count.nonzero_p ()
> + && cfun->cfg->count_max.nonzero_p ()))
> {
> - sreal tmp = BLOCK_INFO (bb)->frequency;
> - if (tmp >= 1)
> + cfun->cfg->count_max = profile_count::uninitialized ();
> + FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR_FOR_FN (cfun), NULL, next_bb)
> {
> - gimple_stmt_iterator gsi;
> - tree decl;
> -
> - /* Self recursive calls can not have frequency greater than 1
> - or program will never terminate. This will result in an
> - inconsistent bb profile but it is better than greatly confusing
> - IPA cost metrics. */
> - for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
> - if (is_gimple_call (gsi_stmt (gsi))
> - && (decl = gimple_call_fndecl (gsi_stmt (gsi))) != NULL
> - && recursive_call_p (current_function_decl, decl))
> - {
> - if (dump_file)
> - fprintf (dump_file, "Dropping frequency of recursive call"
> - " in bb %i from %f\n", bb->index,
> - tmp.to_double ());
> - tmp = (sreal)9 / (sreal)10;
> - break;
> - }
> + sreal tmp = BLOCK_INFO (bb)->frequency;
> + if (tmp >= 1)
> + {
> + gimple_stmt_iterator gsi;
> + tree decl;
> +
> + /* Self recursive calls can not have frequency greater than 1
> + or program will never terminate. This will result in an
> + inconsistent bb profile but it is better than greatly
> confusing
> + IPA cost metrics. */
> + for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
> + if (is_gimple_call (gsi_stmt (gsi))
> + && (decl = gimple_call_fndecl (gsi_stmt (gsi))) != NULL
> + && recursive_call_p (current_function_decl, decl))
> + {
> + if (dump_file)
> + fprintf (dump_file, "Dropping frequency of recursive
> call"
> + " in bb %i from %f\n", bb->index,
> + tmp.to_double ());
> + tmp = (sreal)9 / (sreal)10;
> + break;
> + }
> + }
> + tmp = tmp * freq_max;
> + profile_count count = profile_count::from_gcov_type
> (tmp.to_nearest_int ());
> +
> + /* If we have profile feedback in which this function was never
> + executed, then preserve this info. */
> + if (!(bb->count == profile_count::zero ()))
> + bb->count = count.guessed_local ().combine_with_ipa_count
> (ipa_count);
> + cfun->cfg->count_max = cfun->cfg->count_max.max (bb->count);
> }
> - tmp = tmp * freq_max;
> - profile_count count = profile_count::from_gcov_type
> (tmp.to_nearest_int ());
> -
> - /* If we have profile feedback in which this function was never
> - executed, then preserve this info. */
> - if (!(bb->count == profile_count::zero ()))
> - bb->count = count.guessed_local ().combine_with_ipa_count (ipa_count);
> - cfun->cfg->count_max = cfun->cfg->count_max.max (bb->count);
> }
>
> free_aux_for_blocks ();
> --
> 2.34.1
>
--
BR,
Hongtao
