-
Bug
-
Resolution: Fixed
-
P2
-
1.4.0
-
beta2
-
generic
-
solaris_2.5.1
The JVM must discover whether the system running it is capable
of handling v9 instructions. The present mechanism is esoteric,
depending on speical magic numbers resulting from a trap. It is not
evident that this code will still be correct for future processors or
releases of Solaris.
In order to add prefetching in Ladybird, a sysconf() call was used
to query the isalist of the system; this is the preferred mechanism
for determining the presence of an ultra3 ("vis2"). It would be
straightforward to use that same test for "sparcv9", replacing all
the esoteric code. This wasn't done in Ladybird only because of
stability concerns of being late in the development cycle.
It would make sense to do this at the same time as addressing 4453409
(adding prefetching) and 4453399 (overhauling atomics) for Merlin and
4484953 (don't use FXTOD on cheetah).
of handling v9 instructions. The present mechanism is esoteric,
depending on speical magic numbers resulting from a trap. It is not
evident that this code will still be correct for future processors or
releases of Solaris.
In order to add prefetching in Ladybird, a sysconf() call was used
to query the isalist of the system; this is the preferred mechanism
for determining the presence of an ultra3 ("vis2"). It would be
straightforward to use that same test for "sparcv9", replacing all
the esoteric code. This wasn't done in Ladybird only because of
stability concerns of being late in the development cycle.
It would make sense to do this at the same time as addressing 4453409
(adding prefetching) and 4453399 (overhauling atomics) for Merlin and
4484953 (don't use FXTOD on cheetah).
- relates to
-
JDK-4453409 Prefetching needs to be added to Merlin
-
- Closed
-
-
JDK-4453399 Atomic instruction code should be overhauled
-
- Closed
-