* [PATCH AutoFDO/3]Set profile_count accordingly for AutoFDO.
@ 2018-10-31 8:37 bin.cheng
2018-10-31 9:49 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: bin.cheng @ 2018-10-31 8:37 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 484 bytes --]
Hi,
This is a simple patch setting profile_count with according precision category in
AutoFDO, as well as setting profile_count for exit basic block.
Bootstrap and test on x86_64 in patch set. Is it OK?
Thanks,
bin
2018-10-31 Bin Cheng <bin.cheng@linux.alibaba.com>
* auto-profile.c (afdo_calculate_branch_prob): Convert profile_count
to afdo precision quality.
(afdo_callsite_hot_enough_for_early_inline): Ditto.
(afdo_annotate_cfg): Set profile_count for exit basic block.
[-- Attachment #2: 0003-Set-profile-count-accordingly-for-AutoFDO.patch --]
[-- Type: application/octet-stream, Size: 1896 bytes --]
From c88da703e3b514969027951946dd2a1fb78ec31a Mon Sep 17 00:00:00 2001
From: "chengbin.cb" <bin.cheng@linux.alibaba.com>
Date: Wed, 10 Oct 2018 16:10:20 +0800
Subject: [PATCH 3/4] Set profile count accordingly for AutoFDO.
---
gcc/auto-profile.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/gcc/auto-profile.c b/gcc/auto-profile.c
index 9316f4c477b..cde4f41c1d9 100644
--- a/gcc/auto-profile.c
+++ b/gcc/auto-profile.c
@@ -1400,7 +1400,7 @@ afdo_calculate_branch_prob (bb_set *annotated_bb, edge_set *annotated_edge)
edge e;
edge_iterator ei;
int num_unknown_succ = 0;
- profile_count total_count = profile_count::zero ();
+ profile_count total_count = profile_count::zero ().afdo ();
FOR_EACH_EDGE (e, ei, bb->succs)
{
@@ -1507,6 +1507,7 @@ afdo_annotate_cfg (const stmt_set &promoted_stmts)
= profile_count::from_gcov_type (s->head_count ()).afdo ();
ENTRY_BLOCK_PTR_FOR_FN (cfun)->count
= profile_count::from_gcov_type (s->head_count ()).afdo ();
+ EXIT_BLOCK_PTR_FOR_FN (cfun)->count = profile_count::zero ().afdo ();
profile_count max_count = ENTRY_BLOCK_PTR_FOR_FN (cfun)->count;
FOR_EACH_BB_FN (bb, cfun)
@@ -1686,11 +1687,12 @@ afdo_callsite_hot_enough_for_early_inline (struct cgraph_edge *edge)
if (count > 0)
{
bool is_hot;
+ profile_count pcount = profile_count::from_gcov_type (count).afdo ();
gcov_summary *saved_profile_info = profile_info;
/* At early inline stage, profile_info is not set yet. We need to
temporarily set it to afdo_profile_info to calculate hotness. */
profile_info = autofdo::afdo_profile_info;
- is_hot = maybe_hot_count_p (NULL, profile_count::from_gcov_type (count));
+ is_hot = maybe_hot_count_p (NULL, pcount);
profile_info = saved_profile_info;
return is_hot;
}
--
2.14.4.44.g2045bb6
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH AutoFDO/3]Set profile_count accordingly for AutoFDO.
2018-10-31 8:37 [PATCH AutoFDO/3]Set profile_count accordingly for AutoFDO bin.cheng
@ 2018-10-31 9:49 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2018-10-31 9:49 UTC (permalink / raw)
To: bin.cheng; +Cc: GCC Patches
On Wed, Oct 31, 2018 at 7:31 AM bin.cheng <bin.cheng@linux.alibaba.com> wrote:
>
> Hi,
> This is a simple patch setting profile_count with according precision category in
> AutoFDO, as well as setting profile_count for exit basic block.
>
> Bootstrap and test on x86_64 in patch set. Is it OK?
OK. It might look cleaner (and cheaper) to add a profile-quality argument
to from_gcov_type instead of calling .afdo() on the result. Your call.
Richard.
> Thanks,
> bin
> 2018-10-31 Bin Cheng <bin.cheng@linux.alibaba.com>
>
> * auto-profile.c (afdo_calculate_branch_prob): Convert profile_count
> to afdo precision quality.
> (afdo_callsite_hot_enough_for_early_inline): Ditto.
> (afdo_annotate_cfg): Set profile_count for exit basic block.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2018-10-31 9:14 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-10-31 8:37 [PATCH AutoFDO/3]Set profile_count accordingly for AutoFDO bin.cheng
2018-10-31 9:49 ` Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).