Mikhail, are you planning to implement the write-barrier inlining? That is necessary for GCv5 to be reasonably performing. The fast path for write barrier is very simple: just checking parameter p_target is smaller than nursery boudary and p_slot is not smaller. If the checking returns true, a slow path native function is called; otherwise it simply returns. The nursery space boundary can be a compilation time constant (at the moment, it's a variable with a constant value that is initialized by after gc initialization).
Thanks, xiaofeng On 11/27/06, Mikhail Fursov <[EMAIL PROTECTED]> wrote:
On 11/27/06, Salikh Zakirov <[EMAIL PROTECTED]> wrote: > > Mikhail Fursov wrote: > > > 2) Harmony-2008 status > > The patch was committed a month ago. Today it can't be applied cleanly. > ... > > Did you mean "... was submitted a month ago" ? > > AFAIU, "commited" = "committed to the SVN" > "submitted" = "uploaded to JIRA for review", but not yet committed. > > Yes this is a typo :) BTW I updated the patch and now it is synchronized with the trunk. Weldon, could you check it and if the patch is commited I can work on synchronization helpers this week? -- Mikhail Fursov
