On Fri, Nov 25, 2011 at 10:11 AM, <[email protected]> wrote: > Uploaded a new patch set that renames "heavy" to "non_primitive" and adds > comments. > > I tried moving the counters to IsInlineable (see > http://codereview.chromium.**org/8700001/<http://codereview.chromium.org/8700001/>) > but don't really like it because there > is no good place to call ComputeAstNodeCountAndInlineab**leFlags() from. > This > function needs to be called after analyzing the scopes and before code > generation, because Jacob's ARM performance improvement CL is using the > "is_primitive" flag in the full-codegen. > > Those seem like different concerns, and I have a feeling that conflating them is a mistake.
I still think it's warty that we compute one predicate over the AST (IsInlineable) in one way, and the other related one (IsPrimitive) in a different way. Ideally, we would do them uniformly one way or the other. And long-term I'm doubtful that the conditions for inlining and the conditions for early optimization will always be the same and we'll have to split them up anyway. -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
