Comment #5 on issue 2275 by [email protected]: Disappointingly bad switch
performance on V8
http://code.google.com/p/v8/issues/detail?id=2275
Below change can improve about 150% switch kernel performace . But I'm not
sure how the limitation is decided. There should be lots of tradeoff about
memory consumption, compilation overhead and optimization benefit.
diff --git a/src/hydrogen.cc b/src/hydrogen.cc
index 7d0e228..b7aa144 100644
--- a/src/hydrogen.cc
+++ b/src/hydrogen.cc
@@ -3898,7 +3898,7 @@ void
HGraphBuilder::VisitSwitchStatement(SwitchStatement* stmt) {
ASSERT(current_block()->HasPredecessor());
// We only optimize switch statements with smi-literal smi comparisons,
// with a bounded number of clauses.
- const int kCaseClauseLimit = 128;
+ const int kCaseClauseLimit = 512;
ZoneList<CaseClause*>* clauses = stmt->cases();
int clause_count = clauses->length();
if (clause_count > kCaseClauseLimit) {
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
