Reviewers: Mads Ager, Description: Don't strength reduce divisions by 1 or 2 as they can often be handled by an optimistic inline idiv.
Please review this at http://codereview.chromium.org/125258 SVN Base: http://v8.googlecode.com/svn/branches/bleeding_edge/ Affected files: M src/parser.cc Index: src/parser.cc =================================================================== --- src/parser.cc (revision 2211) +++ src/parser.cc (working copy) @@ -2653,10 +2653,15 @@ double y_val = y->AsLiteral()->handle()->Number(); int64_t y_int = static_cast<int64_t>(y_val); // There are rounding issues with this optimization, but they don't - // apply if the number to be divided with has a reciprocal that can - // be precisely represented as a floating point number. This is - // the case if the number is an integer power of 2. - if (static_cast<double>(y_int) == y_val && IsPowerOf2(y_int)) { + // apply if the number to be divided with has a reciprocal that can be + // precisely represented as a floating point number. This is the case + // if the number is an integer power of 2. Negative integer powers of + // 2 work too, but for -2, -1, 1 and 2 we don't do the strength + // reduction because the inlined optimistic idiv has a reasonable + // chance of succeeding by producing a Smi answer with no remainder. + if (static_cast<double>(y_int) == y_val && + (IsPowerOf2(y_int) || IsPowerOf2(-y_int)) && + (y_int > 2 || y_int < -2)) { y = NewNumberLiteral(1 / y_val); op = Token::MUL; } --~--~---------~--~----~------------~-------~--~----~ v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev -~----------~----~----~----~------~----~------~--~---
