RFR(M) 8024924: Intrinsify java.lang.Math.addExact
stefan.karlsson at oracle.com
Thu Sep 19 12:01:26 PDT 2013
On 9/19/13 8:46 PM, Rickard Bäckman wrote:
> thanks for the review. I'll update the code accordingly.
> On Sep 19, 2013, at 8:15 PM, Vladimir Kozlov wrote:
>> You need to provide more description for public and include the link to the bug report which now is accessible from outside:
> Sorry missed that one.
>> When result of MathExact is not used we should replace it with top to remove uncommon trap.
> You are saying that if the ProjNode (result) doesn't have an out we should replace the MathExact with top() right?
>> I think it would be better to create new mathexactnode.?pp files for new mathexact nodes. Our *node.?pp files become too fat.
> Good idea.
>> In MultiNode::proj_out(proj) check is_FlagsProj() first since is_Proj() is also true for it (since it is subclass).
> No it doesn't. is_Proj() is controlled by the macros in nodes.hpp. FlagsProjNode is a child of CmpNode there and can't be a member of both. I hit the assert in proj_out()
>> Next test file misses Copyright header:
>> We are moving to creation of subcomponent subdirs for new tests. Could you move that into test/compiler/intrinsic/8024924 ?
For those who haven't seen this:
>> On 9/18/13 11:55 AM, Rickard Bäckman wrote:
>>> Hi all,
>>> can I please have reviews for this change.
>>> Jump on overflow is implemented on x86 32&64-bit.
More information about the hotspot-compiler-dev