[10] RFR(S): 8186297: AARCH64: Intrinsify Unsafe.compareAndSetByte and compareAndSetShort

[Dmitrij Pochepko]

Hi,

please review small diff for  JDK-8186297: AARCH64: Intrinsify 
Unsafe.compareAndSetByte and compareAndSetShort

This small patch adds intrinsic for Unsafe::compareAndSetByte and 
Unsafe::compareAndSetShort in a same manner as existing 
compareAndSetInt/Long intrinsics are implemented.

Intrinsic has shorter code with less barriers and overall instruction count.

Generated assembly examples for short method:

private static void testMethod() {
   unsafe.compareAndSetByte(null, BYTE_ARRAY_ADDR, (byte)0, (byte)1);
}

are here:
  - baseline version: 
http://cr.openjdk.java.net/~dpochepk/8186297/baseline_byte_assembly.txt
  - intrinsified version: 
http://cr.openjdk.java.net/~dpochepk/8186297/intrinsic_byte_assembly.txt


I also created small benchmark which demonstrates performance 
difference(10-15%). Benchmark code is here: 
http://cr.openjdk.java.net/~dpochepk/8186297/UnsafeBench.java (should be 
launched with --add-opens java.base/jdk.internal.misc=ALL-UNNAMED to 
have access to unsafe)

Performance results for ThunderX T88 are here: 
http://cr.openjdk.java.net/~dpochepk/8186297/ThunderX_T88_results.txt

Performance results to Cortex A53 (R-Pi 3) are here: 
http://cr.openjdk.java.net/~dpochepk/8186297/CortexA53_results.txt

webrev: http://cr.openjdk.java.net/~dpochepk/8186297/webrev.01/

CR: https://bugs.openjdk.java.net/browse/JDK-8186297


I also run hotspot jtreg tests in compiler/unsafe/* and found no 
regressions.


Thanks,

Dmitrij

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20170816/987a7144/attachment.html>