[aarch64-port-dev ] RFR: 8135231: add support for vectorizing double precision sqrt

[Edward Nevill]

On Mon, Sep 28, 2015 at 5:12 PM, Andrew Haley <aph at redhat.com> wrote:

> I'm not seeing vectorization of the test case.  Please let me know
> exactly how to see the effect.
>
> I'm doing this:
>
> /home/aph/hs-comp/build/linux-aarch64-normal-server-slowdebug/images/jdk/bin/java
> \
> -XX:+SuperWordReductions -XX:LoopUnrollLimit=250 -XX:LoopMaxUnroll=16 \
> -XX:CompileThresholdScaling=0.1 -classpath \
> JTwork/classes/compiler/loopopts/superword -XX:+PrintCompilation \
> SumRedSqrt_Double
>
> I see a ton of faddd instructions, but no vector ops.
>
> Thanks,
> Andrew.
>
>
Nothing special.

-bash-4.2$ /home/ed/images/jdk/bin/java -XX:+UnlockDiagnosticVMOptions
-XX:+PrintAssembly SumRedSqrt_Double > asm
OpenJDK 64-Bit Server VM warning: PrintAssembly is enabled; turning on
DebugNonSafepoints to gain additional output
-bash-4.2$
-bash-4.2$
-bash-4.2$ fgrep fsqrt asm
  0x000003ff70c7fa60: fsqrt    d4, d4          ;*invokestatic sqrt
  0x000003ff70c7fa68: fsqrt    d1, d1          ;*invokestatic sqrt
  0x000003ff70c7fa74: fsqrt    d2, d2          ;*invokestatic sqrt
  0x000003ff70c80254: fsqrt    d4, d4          ;*invokestatic sqrt
  0x000003ff70c8025c: fsqrt    d1, d1          ;*invokestatic sqrt
  0x000003ff70c80268: fsqrt    d2, d2          ;*invokestatic sqrt
  0x000003ff70c83bcc: fsqrt    d18, d18        ;*invokestatic sqrt
  0x000003ff70c83be8: fsqrt    d16, d16
  0x000003ff70c83bec: fsqrt    d17, d17
  0x000003ff70c83ca8: fsqrt    v17.2d, v17.2d
  0x000003ff70c83cac: fsqrt    v19.2d, v20.2d
  0x000003ff70c83cb0: fsqrt    v18.2d, v18.2d
  0x000003ff70c83ccc: fsqrt    v18.2d, v18.2d
  0x000003ff70c83cd0: fsqrt    v19.2d, v19.2d
  0x000003ff70c83cdc: fsqrt    v19.2d, v20.2d
.........


Regards,
Ed.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20150929/09392ad2/attachment.html>