Request for reviews (XS): 7123926: Some CTW test crash: !_control.contains(ctrl)
Vladimir Kozlov
vladimir.kozlov at oracle.com
Fri Jul 6 08:34:52 PDT 2012
I looked again on my changes and, I think, the lines with push_int();continue; should be also under (arg->outcnt() == 1)
condition. If we don't remove call to Integer.toString() - use its result. Otherwise we have to generate conversion from
int to String by hand for each use (see int_getChars()). As result Integer.toString() code will not be folded by later
optimizations for the case I showed in previous mail. I need to do more testing for this and will resend webrev later.
> I'm a little bit worried about this part after seeing your fix for 7181658. Do you think it's safe to use uncast()
It was another Vladimir (Ivanov) who fixed 7181658 :-)
> this way, or should we use eqv_uncast() for comparison at the use sites of StringConcat::argument_uncast()?
I think using uncast() in argument_uncast() is safe, it is used only in related code. The first time argument_uncast()
is called to find a Proj node from SB.toString() call. Then that Proj node passed to merge() method as 'arg'. And in the
merge() method argument_uncast() is called second time to compare with that passed 'arg'.
Thanks,
Vladimir
On 7/5/12 8:23 PM, Krystal Mok wrote:
> Comments inline:
>
> On Fri, Jul 6, 2012 at 10:38 AM, Vladimir Kozlov <vladimir.kozlov at oracle.com <mailto:vladimir.kozlov at oracle.com>> wrote:
>
> Thank you, Kris
>
>
> Krystal Mok wrote:
>
> Hi Vladimir,
>
> foo(Integer intobj) {
> String intstr = intobj.toString();
> String str = intstr + ", " + intstr;
>
> What should be the desired behavior for the case above?
> Is it all right to skip all Integer.toString() calls, and let later optimizations handle it?
>
>
> You have only one toString() call. And it will be kept because it is referenced by debug info which is copied from
> SB constructor to new char[] constructor (AllocateArray node). It is needed for deoptimization since the bytecode
> will be executed after that call.
>
> I see. So that's what you meant in the original "Keep case". That makes sense.
>
> The next code is the case when Integer.toString() will be eliminated by later optimizations:
>
> StringBuilder sb = new StringBuilder();
> String intstr = intobj.toString();
> String str = sb.append(intstr).append(__intstr).toString();
>
> I also applied your uncast() suggestion in argument_uncast() method.
>
> I'm a little bit worried about this part after seeing your fix for 7181658. Do you think it's safe to use uncast() this
> way, or should we use eqv_uncast() for comparison at the use sites of StringConcat::argument_uncast()?
>
> Thanks,
> Kris
>
> And I fixed typos.
>
> Thanks,
> Vladimir
>
>
> And, two typos:
>
> 541 // chain. It's node could be eliminated only if it's result
>
> It's -> Its. Other occurrences of "it's" should also be replaced with "its".
>
> 543 // An other limitation: it should be used only once because
>
> An other -> Another.
>
> - Kris
>
> On Fri, Jul 6, 2012 at 4:37 AM, Vladimir Kozlov <vladimir.kozlov at oracle.com <mailto:vladimir.kozlov at oracle.com>
> <mailto:vladimir.kozlov at __oracle.com <mailto:vladimir.kozlov at oracle.com>>> wrote:
>
> Vladimir Kozlov wrote:
>
> http://cr.openjdk.java.net/~____kvn/7123926/webrev <http://cr.openjdk.java.net/~__kvn/7123926/webrev>
>
> <http://cr.openjdk.java.net/~__kvn/7123926/webrev <http://cr.openjdk.java.net/~kvn/7123926/webrev>>
>
> 7123926: Some CTW test crash: !_control.contains(ctrl)
>
> Don't try to eliminate Integer::toString() call node during
> String concatenation optimization if it's result has several uses.
>
> Eliminate case:
> foo (Integer intob) {
> String str = "int: " + intobj;
>
> Keep case:
> foo (Integer intob) {
> String intstr = intobj.toString();
> String str = "int: " + intobj; // second use is in SB
> allocation
>
>
> Sorry, it should be: "int: " + intstr;
>
> Vladimir
>
>
> debug info
>
> Tested with failed test.
>
> Thanks,
> Vladimir
>
>
>
More information about the hotspot-compiler-dev
mailing list