Request for reviews (XS): Fix of 6658428, emitting popcount on platforms without support
vladimir.kozlov at oracle.com
Mon Mar 12 09:10:36 PDT 2012
Closed it as duplicate of an other.
On 3/12/12 8:45 AM, Nils Eliasson wrote:
> I opened 7153034 before I saw Vladimirs mail, and made this one:
> http://cr.openjdk.java.net/~neliasso/7153034/webrev.02/ <http://cr.openjdk.java.net/%7Eneliasso/7153034/webrev.02/>
> How do I close a bug as Will Not Fix or similar?
> Vladimir Kozlov skrev 2012-03-12 16:42:
>> On 3/12/12 8:39 AM, Nils Eliasson wrote:
>>> The count leading zeros intrinsic uses bsr on x86 when hw instruction not available, so It doesn't need a check.
>> You are right. Changes looks good.
> Vladimir Kozlov skrev 2012-03-12 16:28:
>> The bug id 7152957.
>> On X86 Count*Zeros instructions are predicated by flag UseCountLeadingZerosInstruction so you need to add check for
>> them also into match_rule_supported().
>> On 3/12/12 8:12 AM, Nils Eliasson wrote:
>>> http://cr.openjdk.java.net/~neliasso/0000000/webrev.01/ <http://cr.openjdk.java.net/%7Eneliasso/0000000/webrev.01/>
>>> Fix of 6658428, emitting popcount on platforms without support
>>> I introduced a bug in the intrinsic fix 6658428. The platform depedent code in Matcher::match_rule_supported didn't take
>>> actual hardware support into account, only if the matcher on this platforms support it.
>>> Have added checks in match_rule_supported on all platforms inline with what was available on sparc and arm.
>>> I haven't opened a new bug for this, should I do that, or reuse the one that broke things? Here is the fix anyway.
> Oracle <http://www.oracle.com>
> Nils Eliasson | Senior Member of Technical Staff
> Oracle Java Platform Group, JVM Engineering
> ORACLE Sweden
More information about the hotspot-compiler-dev