[aarch64-port-dev ] hg: aarch64-port/jdk8/hotspot: 4 new changesets

[Andrew Dinn]

Just wanted to note that with these changes (which are *on top* of Ed's
upgrade to jdk8-b110) javac now runs to completion again (array index
error was due to the arraycopy issue) and I manage to run eclipse with
-Xcomp for about 30 minutes on my machine (i.e. probably about 10
minutes on Andrew's hot box) before it breaks with a register allocator
error. javac has a problem if you don't provide -XX:-TieredCompilation
which I still need to investigate but overall this is great progress.

Once again, many thanks to Ed for making the merge so clean!

regards,

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

On 15/10/13 17:11, adinn at redhat.com wrote:
> Changeset: 658ead2379ef
> Author:    adinn
> Date:      2013-10-15 14:08 +0100
> URL:       http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/658ead2379ef
> 
> modify C1 to notify method names with . not / as package separator
> 
> this means that when you simbreak a method you can use the same format
> as when printed out by TraceCompilation  and hsdis. This also makes it
> uniform with C2 so a given breakpoint works with both compilers.
> 
> ! src/share/vm/c1/c1_Compilation.cpp
> 
> Changeset: 5075a0ca6a07
> Author:    adinn
> Date:      2013-10-15 14:12 +0100
> URL:       http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/5075a0ca6a07
> 
> ensure iRegP_R0 outputs can be matched with general iRegP inputs
> 
> ! src/cpu/aarch64/vm/aarch64.ad
> 
> Changeset: 75997cf311bb
> Author:    adinn
> Date:      2013-10-15 14:16 +0100
> URL:       http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/75997cf311bb
> 
> implement instructons matching CmpF3 and CmpD3 ideal nodes
> 
> i.e. cmp two floats/doubles and generate -1 / 0 / 1 for LTU / EQ / GT
> 
> ! src/cpu/aarch64/vm/aarch64.ad
> 
> Changeset: 965b63104307
> Author:    adinn
> Date:      2013-10-15 17:10 +0100
> URL:       http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/965b63104307
> 
> reverted checkcast_arraycopy stub to return -1 ^ num_left_to_copy if copy fails
> 
> the generic code in C2 expects this result and its better to avoid the
> maintenance/integration issues which arise from changing generic code
> if we don't have a compelling performance reason
> 
> ! src/cpu/aarch64/vm/c1_LIRAssembler_aarch64.cpp
> ! src/cpu/aarch64/vm/stubGenerator_aarch64.cpp
> 
>