Skip to contentSkip to conversation

latencytop: optimize LT_BACKTRACEDEPTH loops a bit

DISCUSSION Dmitry Adamushko 2 participants 2 messages Feb 3, 2008 – Feb 4, 2008

lore
Share & tools
lore
#0 lore raw
Quoted 17 lines · Arjan van de Ven
Quoted 3 lines · Dmitry Adamushko

Subject: latencytop: optimize LT_BACKTRACEDEPTH loops a bit.

It looks like there is no need to loop any longer when 'same == 0'.

thanks for the contribution!
while I like your patch, I wonder if we should go even a little further in cleaning this up

Quoted 8 lines
(quoted diff) +1 -2
@@ -73,12 +73,12 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
                      continue;
              }
              for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) {
-                     if (latency_record[i].backtrace[q] !=
-                             lat->backtrace[q])
+                     unsigned long record = lat->backtrace[q];
 
Quoted 11 lines

Review account_global_scheduler_latency(struct task_struct *tsk, struct latency_record · line 83

Quoted 9 lines
(quoted diff) +3 -3
+                     if (latency_record[i].backtrace[q] != record)
                              same = 0;
-                     if (same && lat->backtrace[q] == 0)
-                             break;
-                     if (same && lat->backtrace[q] == ULONG_MAX)
+
+                     if (!same || record == 0 || record == ULONG_MAX)
                              break;
              }

I mean, we could make it look like this:

Yeah, I had some doubts regarding the '!same' case. We'd probably be better off taking a decision (i.e. break) from inside the first branch so to avoid the second one (I guess, it can be a bit more efficient, wrt the CPU's branch-prediction logic).

what about this one instead?

(I'd prefer to have a single 'if' for 'record == 0 || record == ULONG_MAX', it's just a matter of taste though)

---

Subject: latencytop: optimize LT_BACKTRACEDEPTH loops a bit.

It looks like there is no need to loop any longer when 'same == 0'.

---

kernel/latencytop.c +17 -10
index b4e3c85..5e4743d 100644
@@ -64,8 +64,8 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
 		return;
 
 	for (i = 0; i < MAXLR; i++) {
-		int q;
-		int same = 1;
+		int q, same = 1;
+
 		/* Nothing stored: */
 		if (!latency_record[i].backtrace[0]) {
 			if (firstnonnull > i)
@@ -73,12 +73,15 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
 			continue;
 		}
 		for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) {
-			if (latency_record[i].backtrace[q] !=
-				lat->backtrace[q])
+			unsigned long record = lat->backtrace[q];
+
+			if (latency_record[i].backtrace[q] != record) {
 				same = 0;
-			if (same && lat->backtrace[q] == 0)
 				break;
-			if (same && lat->backtrace[q] == ULONG_MAX)
+			}
+
+			/* 0 and ULONG_MAX entries denote the end of backtrace: */
+			if (record == 0 || record == ULONG_MAX)
 				break;
 		}
 		if (same) {
@@ -143,14 +146,18 @@ account_scheduler_latency(struct task_struct *tsk, int usecs, int inter)
 	for (i = 0; i < LT_SAVECOUNT ; i++) {
 		struct latency_record *mylat;
 		int same = 1;
+
 		mylat = &tsk->latency_record[i];
 		for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) {
-			if (mylat->backtrace[q] !=
-				lat.backtrace[q])
+			unsigned long record = lat.backtrace[q];
+
+			if (mylat->backtrace[q] != record) {
 				same = 0;
-			if (same && lat.backtrace[q] == 0)
 				break;
-			if (same && lat.backtrace[q] == ULONG_MAX)
+			}
+
+			/* 0 and ULONG_MAX entries denote the end of backtrace: */
+			if (record == 0 || record == ULONG_MAX)
 				break;
 		}
 		if (same) {
Signature · 2 lines
Best regards,
Dmitry Adamushko
#1 lore raw
Quoted 8 lines · Dmitry Adamushko

---

Subject: latencytop: optimize LT_BACKTRACEDEPTH loops a bit.

It looks like there is no need to loop any longer when 'same == 0'.

looks good to me; Ingo can you queue this one up ?