[aarch64-port-dev ] hg: aarch64-port/jdk8/hotspot: 2 new changesets

[Andrew Dinn]

On 08/11/13 14:05, adinn at redhat.com wrote:
> Changeset: 2e1e9bc263f5
> Author:    adinn
> Date:      2013-11-08 14:02 +0000
> URL:       http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/2e1e9bc263f5
> 
> made fp comparisons work correctly
> 
> ! src/cpu/aarch64/vm/aarch64.ad
> 
> Changeset: 1ceb49e72701
> Author:    adinn
> Date:      2013-11-08 14:04 +0000
> URL:       http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/1ceb49e72701
> 
> added special case rules for fp compare against zero
> 
> n.b. this change revealed an error in the small aarch64 simulator
> implementation of fcmpz. if you pull this change then you will also
> need to update to the latest simulator tree in order for it to run
> correctly.
> 
> ! src/cpu/aarch64/vm/aarch64.ad
> 
> 
> 

I am not 100% convinced that the FP comparison changes are working
correctly as I am now seeing an exit with code 15 when running eclipse.
This is happening after merging the FP compare code with the compressed
OOP optimizations Andrew Haley made in parallel. Since the latter were
tested using eclipse, my concern is that the FP code is incorrect in
some cases -- probably those involving NaNs.

When I made the original change (i.e. before the merge) I ran tests to
exercise the various available comparisons for NaNs, (ensuring they
actually exercised C2 compiled code rather than code reverted to
interpreted execution) and they all appeared to provide the correct results.

I will recheck these tests and also review the way the ideal code
operates to ensure that the comparisons it expects to occur actually
match those currently implemented.

regards,


Andrew Dinn
-----------