linux-sparse.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "Jan Pokorný" <pokorny_jan@seznam.cz>
To: sparse@chrisli.org
Cc: linux-sparse@vger.kernel.org
Subject: Re: [PATCH 2/2] better dealing with OP_PHISOURCE insn
Date: Sat, 16 Apr 2011 12:56:29 +0200	[thread overview]
Message-ID: <4DA975DD.2050908@seznam.cz> (raw)
In-Reply-To: <4DA8CC33.1050307@seznam.cz>

Evolution of patch I proposed on 2011-04-15 (to be applied instead
of my first 2/2 patch posted on 2011-04-09).

In fact, no need to use `trackable_pseudo' as we are only interested in
PSEUDO_REG and we do not expect this pseudo to be NULL (round of testing
with additional assert added hadn't proven such test is relevant).

Signed-off-by: Jan Pokorny <pokorny_jan@seznam.cz>
---
 liveness.c |   29 ++++++++++++++++++++---------
 1 files changed, 20 insertions(+), 9 deletions(-)

diff --git a/liveness.c b/liveness.c
index eeff0f7..060d599 100644
--- a/liveness.c
+++ b/liveness.c
@@ -13,21 +13,32 @@
 #include "flow.h"
 
 static void phi_defines(struct instruction * phi_node, pseudo_t target,
-	void (*defines)(struct basic_block *, struct instruction *, pseudo_t))
+	void (*defines)(struct basic_block *, struct instruction *, pseudo_t),
+	unsigned long generation)
 {
 	pseudo_t phi;
 	FOR_EACH_PTR(phi_node->phi_list, phi) {
-		struct instruction *def;
 		if (phi == VOID)
 			continue;
-		def = phi->def;
-		if (!def || !def->bb)
-			continue;
-		if (def->opcode == OP_PHI) {
-			phi_defines(def, target, defines);
+		if (!phi->def || !phi->def->bb)
 			continue;
+
+		/*
+		 * In case of "(PHISOURCE->PHI->)+ PHISOURCE->PHI" chain, move
+		 * "defines" information upstream to the very first PHISOURCE;
+		 * recursion guarded by generation marking of PHI instructions BBs.
+		 */
+		pseudo_t src_pseudo = phi->def->phi_src;
+		if (src_pseudo->type == PSEUDO_REG
+			&& src_pseudo->def->opcode == OP_PHI) {
+			phi_node->bb->generation = generation;
+			if (src_pseudo->def->bb->generation != generation) {
+				phi_defines(src_pseudo->def, target, defines, generation);
+				continue;
+			}
 		}
-		defines(def->bb, phi->def, target);
+
+		defines(phi->def->bb, phi->def, target);
 	} END_FOR_EACH_PTR(phi);
 }
 
@@ -103,7 +114,7 @@ static void track_instruction_usage(struct basic_block *bb, struct instruction *
 	/* Other */
 	case OP_PHI:
 		/* Phi-nodes are "backwards" nodes. Their def doesn't matter */
-		phi_defines(insn, insn->target, def);
+		phi_defines(insn, insn->target, def, ++bb_generation);
 		break;
 
 	case OP_PHISOURCE:
-- 
1.7.1

  reply	other threads:[~2011-04-16 10:56 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-04-09 11:53 [PATCH 1/2] better dealing with OP_PHISOURCE insn Jan Pokorný
2011-04-09 12:07 ` [PATCH 2/2] " Jan Pokorný
2011-04-09 12:12   ` Jan Pokorný
2011-04-14 10:10     ` Christopher Li
2011-04-15 22:52   ` Jan Pokorný
2011-04-16 10:56     ` Jan Pokorný [this message]
2011-04-16 11:16       ` Jan Pokorný
2011-04-16 14:49         ` Jan Pokorný
2011-04-16 16:21         ` Jan Pokorný

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4DA975DD.2050908@seznam.cz \
    --to=pokorny_jan@seznam.cz \
    --cc=linux-sparse@vger.kernel.org \
    --cc=sparse@chrisli.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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 NNTP newsgroup(s).