On Mon, Mar 29, 2010 at 9:46 PM, Vyacheslav Egorov <[email protected]> wrote: > Do we have enough evidence for that, Vitaly?
I have no idea. I said "*if* this new version is faster". I just saw Anton changed one memcpy call that is used in array implementation. -- Vitaly > Judging from my recent experiments with memmove such replacement [even if it > looks beneficial] might result in unexpected performance degradation. > The other thing which bothers me: have anyone researched what MS cl does > here? They might have highly optimized memcpy/memmove version which uses > 128-bit mmx registers and other fancy tricks [it can even be implemented as > compiler intrinsic]. > So can we be sure that our improvements on Linux do not bring degradation of > Windows? > -- > Vyacheslav Egorov > > > On Mon, Mar 29, 2010 at 7:33 PM, Vitaly Repeshko <[email protected]> > wrote: >> >> On Mon, Mar 29, 2010 at 9:23 PM, <[email protected]> wrote: >> > Thanks a lot for comments, Vitaly >> > >> > >> > http://codereview.chromium.org/1530005/diff/1/4 >> > File src/utils.h (right): >> > >> > http://codereview.chromium.org/1530005/diff/1/4#newcode600 >> > src/utils.h:600: inline void CopyWords(Object** dst, Object** src, int >> > words_size) { >> > On 2010/03/29 17:02:04, Vitaly wrote: >> >> >> >> Maybe also implement CopyBytes and call it here? This way all clients >> > >> > of memcpy >> >> >> >> get a faster alternative. >> > >> > >> > I didn't see a lot of byte proper copying logic as we usually operate on >> > words, not bytes itself (maybe in string related code?), so if you don't >> > mind, I won't implement it right now. >> >> There is quite a few memcpy calls and some of them are in >> performance-sensitive places. If this new version is faster, doesn't >> it makes sense to use it there as well? >> >> >> -- Vitaly -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev To unsubscribe from this group, send email to v8-dev+unsubscribegooglegroups.com or reply to this email with the words "REMOVE ME" as the subject.
