August 2014 - aarch64-port-dev

JDK8: Get BUILTIN_SIM working again
by Edward Nevill 04 Sep '14

04 Sep '14

Oops, The encode_iso_array patch broke the builtin sim image because it tries to execute uqxtn instructions. The following patch gets it going again. Maybe we need a 'UseSIMD' which can be set false for BUILTIN_SIM build and the intrinsics in aarch64.ad can use predicate(UseSimd)? Regards, Ed. --- CUT HERE --- diff -r 1a507fdf6de6 src/cpu/aarch64/vm/macroAssembler_aarch64.cpp --- a/src/cpu/aarch64/vm/macroAssembler_aarch64.cpp Mon Aug 04 18:03:53 2014 +0100 +++ b/src/cpu/aarch64/vm/macroAssembler_aarch64.cpp Tue Aug 05 15:46:21 2014 +0100 @@ -3625,6 +3625,7 @@ mov(result, len); // Save initial len +#ifndef BUILTIN_SIM subs(len, len, 32); br(LT, LOOP_8); @@ -3663,6 +3664,9 @@ BIND(LOOP_1); adds(len, len, 8); br(LE, DONE); +#else + cbz(len, DONE); +#endif BIND(NEXT_1); ldrh(tmp1, Address(post(src, 2))); tst(tmp1, 0xff00); --- CUT HERE ---

2 1

JTreg failures in aarch64 jdk7u
by Edward Nevill 04 Sep '14

04 Sep '14

Hi, I have been looking at the jtreg hotspot failures in aarch64 jdk7. There are a large number of failures because many of the jtreg hotspot tests contain code which is specific to jdk8, even though it is a jdk7 build. This is because the hotspot sub-repo came originally from jdk8 (hs 25.0-b70) instead of the standard jdk7u hotspot (hs 24.80-b07) and there are many tests which depend on jdk8, either because the tests use jdk8 source features, or because they rely on jdk8 APIs. For example, all (38) of the mathexact tests are invalid and therefore fail because mathexact does not exist on jdk7. The following is a summary of the results i get for jdk7 hotspot Test results: passed: 343; failed: 102; error: 1 For comparison, i tried running the same tests on x86 jdk7 with the following results. Test results: passed: 311; failed: 134; error: 1 I then took the jtreg test suite from the hs24 build and ran it on both aarch64 and x86 vsns of jdk7 with the following results. aarch64: Test results: passed: 279; failed: 11; error: 2 x86: Test results: passed: 286; failed: 4; error: 2 Here aarch64 still gets more failures than x86 because some of the tests make assumptions about testing hs24 vs hs25 (eg. the behaviour of various -XX: options). However the number of test failures is much smaller. So, what is the best way to resolve this? 1) Prune the existing aarch64 jtreg hotspot tree to remove any tests which are specific to jdk8 or 2) Generate an exclude list to exclude the above tests but leave them in place in the tree or 3) Replace the existing aarch64 jtreg hotspot tree (from hs25) with the one from hs24 (which gets 11 failures) and then either prune or exclude the much smaller set of tests which depend on some feature of hs24. I am happy to implement whichever solution as long as the solution works and is satisfactory to everyone. All the best, Ed.

2 1

RFR: Optimise store of 0 byte into card table
by Edward Nevill 29 Aug '14

29 Aug '14

Hi, The following patch optimises the storing of 0 bytes into the card table. Firstly the existing code was generating mov wS, zr stlrb wS, [xN] because it didn't have a rule for storing 0. This has been optimised to stlrb zr, [xN] Note: I have only done this optimisation for bytes, should I also do it for 16, 32 & 64 bit values? How often do these actually occur? Secondly, if the byte in memory is already 0 it skips the store. Since in the vast majority of cases the byte is in fact 0 because the card is already dirty this avoids doing unnecessary STRLB instructions. So it generates ldr rScratch, [xN] cbz rScratch, skip stlrb zr, [xN] skip: This, in combination with the previous patch generates significant performance improvements on programs that do extensive stores of non volatile oops. OK to push? Ed. Patch also available at http://people.linaro.org/~edward.nevill/patches/memorder.patch in case there is any problem with the formatting below. --- CUT HERE --- # HG changeset patch # User Edward Nevill edward.nevill(a)linaro.org # Date 1409322430 -3600 # Fri Aug 29 15:27:10 2014 +0100 # Node ID 953a1b5e5b1726470045bfa0dbe1b2bff799b906 # Parent 4aa306297dafb02943645761f2477d0d95c4a157 Optimise store of 0 byte into card table diff -r 4aa306297daf -r 953a1b5e5b17 src/cpu/aarch64/vm/aarch64.ad --- a/src/cpu/aarch64/vm/aarch64.ad Fri Aug 29 11:12:45 2014 +0100 +++ b/src/cpu/aarch64/vm/aarch64.ad Fri Aug 29 15:27:10 2014 +0100 @@ -2160,6 +2160,18 @@ rscratch1, stlrb); %} + // Special case of storing 0 to volatile for storing into card table + enc_class aarch64_enc_stlrb0(memory mem) %{ + Label skip; + { + MacroAssembler _masm(&cbuf); + __ ldrb(rscratch1, as_Register($mem$$base)); + __ cbz(rscratch1, skip); + } + MOV_VOLATILE(zr, $mem$$base, $mem$$index, $mem$$scale, $mem$$disp, rscratch1, stlrb); + __ bind(skip); + %} + enc_class aarch64_enc_stlrh(iRegI src, memory mem) %{ MOV_VOLATILE(as_Register($src$$reg), $mem$$base, $mem$$index, $mem$$scale, $mem$$disp, rscratch1, stlrh); @@ -5909,6 +5921,19 @@ ins_pipe(pipe_class_memory); %} +// Special rule for store of 0 byte to volatile for card table +instruct storeB_volatile_imm0(immI0 zero, /* sync_memory*/indirect mem) +%{ + match(Set mem (StoreB mem zero)); + + ins_cost(VOLATILE_REF_COST); + format %{ "stlrb zr, $mem\t# byte" %} + + ins_encode(aarch64_enc_stlrb0(mem)); + + ins_pipe(pipe_class_memory); +%} + // Store Char/Short instruct storeC_volatile(iRegI src, /* sync_memory*/indirect mem) %{ diff -r 4aa306297daf -r 953a1b5e5b17 src/cpu/aarch64/vm/assembler_aarch64.hpp --- a/src/cpu/aarch64/vm/assembler_aarch64.hpp Fri Aug 29 11:12:45 2014 +0100 +++ b/src/cpu/aarch64/vm/assembler_aarch64.hpp Fri Aug 29 15:27:10 2014 +0100 @@ -1081,7 +1081,7 @@ Register Rn, enum operand_size sz, int op, int o0) { starti; f(sz, 31, 30), f(0b001000, 29, 24), f(op, 23, 21); - rf(Rs, 16), f(o0, 15), rf(Rt2, 10), rf(Rn, 5), rf(Rt1, 0); + rf(Rs, 16), f(o0, 15), rf(Rt2, 10), rf(Rn, 5), zrf(Rt1, 0); } #define INSN4(NAME, sz, op, o0) /* Four registers */ \ --- CUT HERE ---

1 0

hg: aarch64-port/jdk8/hotspot: Dont use a release store when storing an OOP in a non-volatile field.
by ed＠camswl.com 29 Aug '14

29 Aug '14

Changeset: 4aa306297daf Author: Edward Nevill edward.nevill(a)linaro.org Date: 2014-08-29 11:12 +0100 URL: http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/4aa306297daf Dont use a release store when storing an OOP in a non-volatile field. ! src/share/vm/opto/memnode.hpp ! src/share/vm/opto/parse2.cpp

1 0

Don't use a release store when storing an OOP in a non-volatile field.
by Andrew Haley 23 Aug '14

23 Aug '14

HotSpot always uses a release tore when writing OOP fields, because: // Non-volatile fields also need releasing stores if they hold an // object reference, because the object reference might point to // a freshly created object. This isn't needed on AArch64 because we generate a memory barrier when the object is created and there is an address dependency from a read instruction to a program-order-later read or write instruction when the value read by the first is used to compute the address used for the second. See 4.1, Address Dependencies, in http://www.cl.cam.ac.uk/~pes20/ppc-supplemental/test7.pdf for all the gory details. I suspect that every CPU on which HotSpot runs respects address dependencies in the same way. Andrew. # HG changeset patch # User aph # Date 1408722585 14400 # Fri Aug 22 11:49:45 2014 -0400 # Node ID 6e5128f86ad8bf307f0d7cea6e3bfd203cab8262 # Parent 6a5c5cb9dde89ea50fefe4bf2e41c02d62a5beec Don't use a release store when storing an OOP in a non-volatile field. diff -r 6a5c5cb9dde8 -r 6e5128f86ad8 src/share/vm/opto/parse2.cpp --- a/src/share/vm/opto/parse2.cpp Fri Aug 22 10:52:59 2014 -0400 +++ b/src/share/vm/opto/parse2.cpp Fri Aug 22 11:49:45 2014 -0400 @@ -1689,7 +1689,9 @@ a = pop(); // the array itself const TypeOopPtr* elemtype = _gvn.type(a)->is_aryptr()->elem()->make_oopptr(); const TypeAryPtr* adr_type = TypeAryPtr::OOPS; - Node* store = store_oop_to_array(control(), a, d, adr_type, c, elemtype, T_OBJECT, MemNode::release); + Node* store = store_oop_to_array(control(), a, d, adr_type, c, elemtype, T_OBJECT, + NOT_AARCH64(MemNode::release) + AARCH64_ONLY(MemNode::unordered)); break; } case Bytecodes::_lastore: { diff -r 6a5c5cb9dde8 -r 6e5128f86ad8 src/share/vm/opto/parse3.cpp --- a/src/share/vm/opto/parse3.cpp Fri Aug 22 10:52:59 2014 -0400 +++ b/src/share/vm/opto/parse3.cpp Fri Aug 22 11:49:45 2014 -0400 @@ -303,8 +303,12 @@ // Non-volatile fields also need releasing stores if they hold an // object reference, because the object reference might point to // a freshly created object. - StoreNode::release_if_reference(bt); - + NOT_AARCH64(StoreNode::release_if_reference(bt)) + // AArch64 doesn't need a release store because if there is an + // address dependency between a read and a write, then those + // memory accesses are observed in program order by all observers + // within the shareability domain. + AARCH64_ONLY(MemNode::unordered); // Store the value. Node* store; if (bt == T_OBJECT) {

2 2

hg: aarch64-port/jdk9/hotspot: 2 new changesets
by aph＠redhat.com 22 Aug '14

22 Aug '14

Changeset: 104401510ba6 Author: aph Date: 2014-08-22 10:51 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk9/hotspot/rev/104401510ba6 Backed out changeset 5e06422ea48b - agent/src/share/classes/sun/jvm/hotspot/debugger/MachineDescriptionAArch64.java ! src/share/vm/opto/parse2.cpp ! src/share/vm/opto/parse3.cpp Changeset: 6a5c5cb9dde8 Author: aph Date: 2014-08-22 10:52 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk9/hotspot/rev/6a5c5cb9dde8 Add MachineDescriptionAArch64 + agent/src/share/classes/sun/jvm/hotspot/debugger/MachineDescriptionAArch64.java

1 0

RFR: jdk9: Add missing file MachineDescriptionAArch64.java
by Edward Nevill 22 Aug '14

22 Aug '14

Hi, Add missing hotspot file agent/src/share/classes/sun/jvm/hotspot/debugger/MachineDescriptionAArch64.java Regards, Ed. --- CUT HERE --- # HG changeset patch # User Edward Nevill edward.nevill(a)linaro.org # Date 1408713905 -3600 # Fri Aug 22 14:25:05 2014 +0100 # Node ID 6d5e156d7136addd3dab0ff08863f4b960921b43 # Parent ed1a3a21d7cfb49d9a595cfab49d7f780d604e6d Add missing file MachineDescriptionAArch64.java diff -r ed1a3a21d7cf -r 6d5e156d7136 agent/src/share/classes/sun/jvm/hotspot/debugger/MachineDescriptionAArch64.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/MachineDescriptionAArch64.java Fri Aug 22 14:25:05 2014 +0100 @@ -0,0 +1,39 @@ +/* + * Copyright (c) 2003, 2008, Oracle and/or its affiliates. All rights reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA + * or visit www.oracle.com if you need additional information or have any + * questions. + * + */ + +package sun.jvm.hotspot.debugger; + +public class MachineDescriptionAArch64 extends MachineDescriptionTwosComplement implements MachineDescription { + public long getAddressSize() { + return 8; + } + + public boolean isLP64() { + return true; + } + + public boolean isBigEndian() { + return false; + } +} --- CUT HERE ---

2 1

hg: aarch64-port/jdk9/hotspot: Add MachineDescriptionAArch64
by aph＠redhat.com 22 Aug '14

22 Aug '14

Changeset: 5e06422ea48b Author: aph Date: 2014-08-22 10:35 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk9/hotspot/rev/5e06422ea48b Add MachineDescriptionAArch64 + agent/src/share/classes/sun/jvm/hotspot/debugger/MachineDescriptionAArch64.java ! src/share/vm/opto/parse2.cpp ! src/share/vm/opto/parse3.cpp

1 0

JTREG, SPECjbb2013 and Hadoop/Terasort results for OpenJDK 8 on AArch64
by Edward Nevill 22 Aug '14

22 Aug '14

This is a summary of the JTREG test results =========================================== The build and test results are cycled every 10 days. For detailed information on the test output please refer to: http://openjdk.linaro.org/openjdk8-jtreg-nightly-tests/summary/2014/234/sum… ------------------------------------------------------------------------------- client-release/hotspot ------------------------------------------------------------------------------- Build 0: aarch64/2014/jul/09 pass: 545; fail: 19 Build 1: aarch64/2014/jul/10 pass: 545; fail: 19 Build 2: aarch64/2014/jul/23 pass: 545; fail: 19 Build 3: aarch64/2014/jul/29 pass: 545; fail: 19 Build 4: aarch64/2014/jul/30 pass: 545; fail: 19 Build 5: aarch64/2014/aug/01 pass: 545; fail: 19 Build 6: aarch64/2014/aug/05 pass: 545; fail: 19 Build 7: aarch64/2014/aug/06 pass: 545; fail: 19 Build 8: aarch64/2014/aug/20 pass: 545; fail: 19 Build 9: aarch64/2014/aug/22 pass: 545; fail: 19 ------------------------------------------------------------------------------- client-release/jdk ------------------------------------------------------------------------------- Build 0: aarch64/2014/jul/09 pass: 5,238; fail: 217; error: 55 Build 1: aarch64/2014/jul/10 pass: 5,230; fail: 234; error: 46 Build 2: aarch64/2014/jul/23 pass: 5,281; fail: 182; error: 47 Build 3: aarch64/2014/jul/29 pass: 5,267; fail: 199; error: 44 Build 4: aarch64/2014/jul/30 pass: 5,279; fail: 185; error: 46 Build 5: aarch64/2014/aug/01 pass: 5,285; fail: 180; error: 45 Build 6: aarch64/2014/aug/05 pass: 5,280; fail: 183; error: 47 Build 7: aarch64/2014/aug/06 pass: 5,284; fail: 179; error: 47 Build 8: aarch64/2014/aug/20 pass: 5,283; fail: 182; error: 45 Build 9: aarch64/2014/aug/22 pass: 5,280; fail: 181; error: 49 ------------------------------------------------------------------------------- client-release/langtools ------------------------------------------------------------------------------- Build 0: aarch64/2014/jul/09 pass: 3,000; error: 15 Build 1: aarch64/2014/jul/10 pass: 2,999; error: 16 Build 2: aarch64/2014/jul/23 pass: 3,000; error: 15 Build 3: aarch64/2014/jul/29 pass: 2,999; error: 16 Build 4: aarch64/2014/jul/30 pass: 2,999; error: 16 Build 5: aarch64/2014/aug/01 pass: 2,999; error: 16 Build 6: aarch64/2014/aug/05 pass: 2,999; error: 16 Build 7: aarch64/2014/aug/06 pass: 2,999; error: 16 Build 8: aarch64/2014/aug/20 pass: 2,999; error: 16 Build 9: aarch64/2014/aug/22 pass: 2,999; error: 16 ------------------------------------------------------------------------------- server-release/hotspot ------------------------------------------------------------------------------- Build 0: aarch64/2014/jul/09 pass: 562; fail: 2 Build 1: aarch64/2014/jul/10 pass: 562; fail: 2 Build 2: aarch64/2014/jul/23 pass: 562; fail: 2 Build 3: aarch64/2014/jul/29 pass: 562; fail: 2 Build 4: aarch64/2014/jul/30 pass: 562; fail: 2 Build 5: aarch64/2014/aug/01 pass: 562; fail: 2 Build 6: aarch64/2014/aug/05 pass: 562; fail: 2 Build 7: aarch64/2014/aug/06 pass: 562; fail: 2 Build 8: aarch64/2014/aug/20 pass: 562; fail: 2 Build 9: aarch64/2014/aug/22 pass: 562; fail: 2 ------------------------------------------------------------------------------- server-release/jdk ------------------------------------------------------------------------------- Build 0: aarch64/2014/jul/09 pass: 5,231; fail: 231; error: 48 Build 1: aarch64/2014/jul/10 pass: 5,246; fail: 212; error: 52 Build 2: aarch64/2014/jul/23 pass: 5,291; fail: 174; error: 45 Build 3: aarch64/2014/jul/29 pass: 5,285; fail: 182; error: 43 Build 4: aarch64/2014/jul/30 pass: 5,292; fail: 172; error: 46 Build 5: aarch64/2014/aug/01 pass: 5,284; fail: 180; error: 46 Build 6: aarch64/2014/aug/05 pass: 5,289; fail: 176; error: 45 Build 7: aarch64/2014/aug/06 pass: 5,285; fail: 177; error: 48 Build 8: aarch64/2014/aug/20 pass: 5,290; fail: 172; error: 48 Build 9: aarch64/2014/aug/22 pass: 5,283; fail: 181; error: 46 ------------------------------------------------------------------------------- server-release/langtools ------------------------------------------------------------------------------- Build 0: aarch64/2014/jul/09 pass: 3,004; error: 11 Build 1: aarch64/2014/jul/10 pass: 3,004; error: 11 Build 2: aarch64/2014/jul/23 pass: 3,004; error: 11 Build 3: aarch64/2014/jul/29 pass: 3,004; error: 11 Build 4: aarch64/2014/jul/30 pass: 3,004; error: 11 Build 5: aarch64/2014/aug/01 pass: 3,004; error: 11 Build 6: aarch64/2014/aug/05 pass: 3,004; error: 11 Build 7: aarch64/2014/aug/06 pass: 3,004; error: 11 Build 8: aarch64/2014/aug/20 pass: 3,004; error: 11 Build 9: aarch64/2014/aug/22 pass: 3,004; error: 11 Previous results can be found here: http://openjdk.linaro.org/openjdk8-jtreg-nightly-tests/index.html SPECjbb2013 composite regression test completed =============================================== This test measures the relative performance of the server compiler running the SPECjbb2013 composite tests and compares the performance against the baseline performance of the server compiler taken on 2014-04-01. In accordance with [1], the SPECjbb2013 tests are run on a system which is not production ready and does not meet all the requirements for publishing compliant results. The numbers below shall be treated as non-compliant (nc) and are for experimental purposes only. Relative performance: Server max-jOPS (nc): 0.98x Relative performance: Server critical-jOPS (nc): 1.02x Details of the test setup and historical results may be found here: http://openjdk.linaro.org/SPECjbb2013-1.00-results/ [1] http://www.spec.org/fairuse.html#Academic Regression test Hadoop-Terasort completed ========================================= This test measures the performance of the server and client compilers running Hadoop sorting a 1GB file using Terasort and compares the performance against the baseline performance of the Zero interpreter and against the baseline performance of the client and server compilers on 2014-04-01. Relative performance: Zero: 1.0, Client: 50.77, Server: 84.52 Client 50.77 / Client 2014-04-01 (43.00): 1.18x Server 84.52 / Server 2014-04-01 (71.00): 1.19x Details of the test setup and historical results may be found here: http://openjdk.linaro.org/hadoop-terasort-benchmark-results/ This is a summary of the jcstress test results ============================================== The build and test results are cycled every 10 days. 2014-07-11 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/192/results/ 2014-07-23 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/204/results/ 2014-07-29 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/210/results/ 2014-07-30 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/211/results/ 2014-08-01 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/213/results/ 2014-08-05 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/217/results/ 2014-08-06 pass rate: 898/898, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/218/results/ 2014-08-20 pass rate: 872/872, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/232/results/ 2014-08-22 pass rate: 872/872, results: http://openjdk.linaro.org/jcstress-nightly-runs/2014/234/results/ For detailed information on the test output please refer to: http://openjdk.linaro.org/jcstress-nightly-runs/

1 0

hg: aarch64-port/jdk8/hotspot: 4 new changesets
by aph＠redhat.com 21 Aug '14

21 Aug '14

Changeset: 8ca3a150d97d Author: aph Date: 2014-08-21 11:56 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/8ca3a150d97d Unwind native AArch64 frames. ! src/os_cpu/linux_aarch64/vm/os_linux_aarch64.cpp Changeset: f5e4bc9d2b26 Author: aph Date: 2014-08-21 11:58 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/f5e4bc9d2b26 Add frame anchor fences. ! src/cpu/aarch64/vm/javaFrameAnchor_aarch64.hpp Changeset: 13b0e050a417 Author: aph Date: 2014-08-20 10:56 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/13b0e050a417 A more efficient sequence for C1_MacroAssembler::float_cmp. ! src/cpu/aarch64/vm/c1_MacroAssembler_aarch64.cpp Changeset: 72b78cf4cd32 Author: aph Date: 2014-08-21 12:35 -0400 URL: http://hg.openjdk.java.net/aarch64-port/jdk8/hotspot/rev/72b78cf4cd32 Add CNEG and CNEGW to macro assembler. ! src/cpu/aarch64/vm/aarch64.ad ! src/cpu/aarch64/vm/macroAssembler_aarch64.hpp

1 0