openj9
openj9 copied to clipboard
Published
20 hours ago •
eclipse-openj9
eclipse-openj9
OpenJ9 Outputs Incorrect Results with JIT Compiler
Java -version output
openjdk version "1.8.0_402-internal"
OpenJDK Runtime Environment (build 1.8.0_402-internal-user_2024_01_31_16_41-b00)
Eclipse OpenJ9 VM (build master-40a4d38b2, JRE 1.8.0 Linux amd64-64-Bit Compressed References 20240131_000000 (JIT enabled, AOT enabled)
OpenJ9 - 40a4d38b2
OMR - a74c0935b
JCL - 33602fcbe7 based on jdk8u402-b06)
javac 1.8.0_402-internal
Summary of problem
When I run the following code, openj9 outputs error results different from jdk. Since there is no error result when -Xint is turned on, I think it may be a JIT problem:
class Test {
int a = 400;
int[] m = new int[a];
int[] c = new int[a];
void n() {
for (int i = 46; i > 9; --i) {
m[i + 1] = 1;
c[i] = 2;
}
for (int i = 0; i < 20000; ++i) c = m;
}
void g() {
for (int i = 0; i < 20000; ++i)
n();
System.out.println(m[11]);
}
public static void main(String[] o) {
Test test = new Test();
test.g();
}
}
$ openj9-openjdk-jdk8/.../java Test
2
$ other-jdk/.../java Test
1
$ openj9-openjdk-jdk8/.../java -Xint Test
I investigated the causes of the problem and found the following. I hope it can be helpful to you in the process of fixing the bug.
$ openj9-openjdk-jdk8/.../java '-Xjit:{Test.n(*)*}(optLevel=hot,lastOptIndex=137)' Test
2
$ openj9-openjdk-jdk8/.../java '-Xjit:{Test.n(*)*}(optLevel=hot,lastOptIndex=136)' Test
1
$ openj9-openjdk-jdk8/.../java '-Xjit:{Test.n(*)*}(optLevel=warm)' Test
1
Thanks for reporting the problem. In my own runs, I needed to include the count=1,disableAsyncCompilation JIT compiler options, and I ended up with different lastOptIndex settings than yours:
$ openj9-openjdk-jdk8/bin/java '-Xjit:{Test.n(*)*}(optLevel=hot,lastOptIndex=122),count=1,disableAsyncCompilation' Test
1
root@ubuntu-1-hz1:/home/henry/docker-hostdir/defects/issue18956
$ openj9-openjdk-jdk8/bin/java '-Xjit:{Test.n(*)*}(optLevel=hot,lastOptIndex=123),count=1,disableAsyncCompilation' Test
2
Looking at log files, I found lastOptIndex=123 corresponded to Idiom Recognition:
<optimization id=123 name=idiomRecognition method=Test.n()V>
Performing 123: idiomRecognition
(Building use/def info)
PREPARTITION VN (Building value number info)
[ 260] 123.1 O^O NEWLOOPREDUCER: Reducing loop 4 to MemSet
May I ask you to run with the option '-Xjit:{Test.n(*)*}(optLevel=hot,lastOptIndex=137,optDetails,log=issue18955.log)' and check the log file to see whether Idiom Recognition corresponds to optIndex 137 in your tests?
When I run with the option '-Xjit:{Test.n(*)*}(optLevel=hot,lastOptIndex=137,optDetails,log=issue18955.log)', the log file indicates that the optIndex corresponds to localCSE.
<optimization id=137 name=localCSE method=Test.n()V>
Performing 137: localCSE
[ 387] 137.1 O^O LOCAL COMMON SUBEXPRESSION ELIMINATION: Local Common Subexpression Elimination propagating local #423 in node : 00007F7DB4F3EC90 PARENT : 00007F7DB4F3EC40 from node 00007F7DB4F3E970
O^O LOCAL COMMON SUBEXPRESSION ELIMINATION: Rhs of store def node : 00007F7DB4F3E9C0
[ 388] 137.2 O^O LOCAL COMMON SUBEXPRESSION ELIMINATION: Local Common Subexpression Elimination commoning node : 00007F7DB4F3F870 by available node : 00007F7DB4F3F8C0
[ 389] 137.3 O^O LOCAL COMMON SUBEXPRESSION ELIMINATION: Local Common Subexpression Elimination propagating local #423 in node : 00007F7DB4F3EF60 PARENT : 00007F7DB4F3EF10 from node 00007F7DB4F3EBF0
O^O LOCAL COMMON SUBEXPRESSION ELIMINATION: Rhs of store def node : 00007F7DB4F3EC40
[ 390] 137.4 O^O LOCAL COMMON SUBEXPRESSION ELIMINATION: Local Common Subexpression Elimination commoning node : 00007F7DB4F3F820 by available node : 00007F7DB4F3F8C0
<trees
title="Trees after localCSE"
method="Test.n()V"
hotness="hot">
The Optimization ID for Idiom Recognition remains 123. Below is the complete log file I obtained.
issue18955.log.1604928.1220.20240215.204700.1604928.log
We haven't investigated this problem enough to get a fix in for the 0.44.0 release. Moving to 0.45.
Hello, I found mis-compilation again in the localCSE optimization but using another testcase. To avoid cluttering the issue system, I'll post the relevant information here.
Could you please confirm if this is due to the same reason as the issue mentioned above? @hzongaro
class Test {
int a = 400;
long b;
volatile short c = 2;
double[] k = new double[a];
void d() {
int h = 1, f = 1;
do {
b = 2;
for (int g = 2; g < 66; g++) {
for (int j = 1; j < 2; ++j) k[f] = h -= b;
b <<= c;
}
} while (++f < 381);
System.out.println(h);
}
public static void main(String[] l) {
Test t = new Test();
for (int i = 0; i < 50; i++) t.d();
}
}
$ other-jdk/.../java Test
...
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
$ openj9-openjdk-jdk8/.../java '-Xjit:{Test.d(*)*}(optLevel=hot,lastOptIndex=99,optDetails,log=jit.log)' Test
...
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
-1431655511
# mis-compilation
$ openj9-openjdk-jdk8/.../java '-Xjit:{Test.d(*)*}(optLevel=hot,lastOptIndex=100,optDetails,log=jit.log)' Test
...
-1431655511
-1431655511
-48639
-48639
-48639
-48639
-48639
-48639
-48639
-48639
-48639
The relevant log files are located here: jit.log.1340959.68205.20240225.213645.1340959.log
I took a quick look at the log file for the original test case, and I have confirmed that it's a problem in Idiom Recognition. Before Idiom Recogition, the IL for the body of the loop looks like this:
n15n BBStart <block_4> (freq 10000) (in loop 4) [0x7f1c6f004ce0] bci=[-1,3,8] rc=0 vc=819 vn=- li=-1 udi=- nc=0
n35n istorei <array-shadow>[#229 Shadow] [flags 0x80000603 0x0 ] [0x7f1c6f005320] bci=[-1,17,9] rc=0 vc=819 vn=- li=-1 udi=- nc=2
n34n aladd (X>=0 internalPtr ) [0x7f1c6f0052d0] bci=[-1,17,9] rc=1 vc=819 vn=- li=- udi=- nc=2 flg=0x8100
n19n aload <temp slot 4>[#429 Auto] [flags 0x7 0x0 ] (X!=0 ) [0x7f1c6f004e20] bci=[-1,10,9] rc=1 vc=819 vn=- li=- udi=20 nc=0 flg=0x4
n33n lsub (highWordZero X>=0 cannotOverflow ) [0x7f1c6f005280] bci=[-1,17,9] rc=1 vc=819 vn=- li=- udi=- nc=2 flg=0x5100
n135n lmul (X>=0 cannotOverflow ) [0x7f1c6f0819f0] bci=[-1,17,9] rc=2 vc=819 vn=- li=- udi=- nc=2 flg=0x1100
n30n i2l (highWordZero X>=0 ) [0x7f1c6f005190] bci=[-1,17,9] rc=1 vc=819 vn=- li=- udi=- nc=1 flg=0x4100
n22n iload <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] (X>=0 cannotOverflow ) [0x7f1c6f004f10] bci=[-1,13,9] rc=2 vc=819 vn=- li=- udi=21 nc=0 flg=0x1100
n136n lconst 4 (highWordZero X!=0 X>=0 ) [0x7f1c6f081a40] bci=[-1,14,9] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x4104
n32n lconst -12 (X!=0 X<=0 ) [0x7f1c6f005230] bci=[-1,17,9] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x204
n25n iconst 1 (X!=0 X>=0 ) [0x7f1c6f005000] bci=[-1,16,9] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x104
n51n istorei <array-shadow>[#229 Shadow] [flags 0x80000603 0x0 ] [0x7f1c6f005820] bci=[-1,24,10] rc=0 vc=819 vn=- li=-1 udi=- nc=2
n50n aladd (X>=0 internalPtr ) [0x7f1c6f0057d0] bci=[-1,24,10] rc=1 vc=819 vn=- li=- udi=- nc=2 flg=0x8100
n37n aload <temp slot 3>[#428 Auto] [flags 0x7 0x0 ] (X!=0 ) [0x7f1c6f0053c0] bci=[-1,19,10] rc=1 vc=819 vn=- li=- udi=22 nc=0 flg=0x4
n49n lsub (highWordZero X>=0 cannotOverflow ) [0x7f1c6f005780] bci=[-1,24,10] rc=1 vc=819 vn=- li=- udi=- nc=2 flg=0x5100
n135n ==>lmul
n48n lconst -8 (X!=0 X<=0 ) [0x7f1c6f005730] bci=[-1,24,10] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x204
n41n iconst 2 (X!=0 X>=0 ) [0x7f1c6f005500] bci=[-1,23,10] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x104
n55n istore <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] [0x7f1c6f0800f0] bci=[-1,25,8] rc=0 vc=819 vn=- li=-1 udi=4 nc=1
n54n iadd (X>=0 cannotOverflow ) [0x7f1c6f0800a0] bci=[-1,25,8] rc=2 vc=819 vn=- li=- udi=- nc=2 flg=0x1100
n22n ==>iload
n53n iconst -1 (X!=0 X<=0 ) [0x7f1c6f080050] bci=[-1,25,8] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x204
n17n ificmpgt --> block_4 BBStart at n15n () [0x7f1c6f004d80] bci=[-1,6,8] rc=0 vc=819 vn=- li=-1 udi=- nc=2 flg=0x20
n54n ==>iadd
n14n iconst 9 (X!=0 X>=0 ) [0x7f1c6f004c90] bci=[-1,4,8] rc=1 vc=819 vn=- li=- udi=- nc=0 flg=0x104
n4n BBEnd </block_4> =====
Idiom Recognition replaces the loop with arrayset operations:
[ 260] 123.1 O^O NEWLOOPREDUCER: Reducing loop 4 to MemSet
...
n15n BBStart <block_4> (freq 10000) [0x7f1c6f004ce0] bci=[-1,3,8] rc=0 vc=831 vn=- li=- udi=- nc=0
n736n treetop [0x7f1c6f08d5c0] bci=[-1,17,9] rc=0 vc=0 vn=- li=- udi=- nc=1
n735n arrayset <arrayset>[#219 helper Method] [flags 0x400 0x0 ] () [0x7f1c6f08d570] bci=[-1,17,9] rc=1 vc=0 vn=- li=- udi=- nc=3 flg=0x20
n716n aladd (X>=0 internalPtr ) [0x7f1c6f08cf80] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=2 flg=0x8100
n717n aload <temp slot 4>[#429 Auto] [flags 0x7 0x0 ] (X!=0 ) [0x7f1c6f08cfd0] bci=[-1,10,9] rc=1 vc=831 vn=- li=- udi=13 nc=0 flg=0x4
n718n lsub (highWordZero X>=0 cannotOverflow ) [0x7f1c6f08d020] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=2 flg=0x5100
n719n lmul (X>=0 cannotOverflow ) [0x7f1c6f08d070] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=2 flg=0x1100
n720n i2l (highWordZero X>=0 ) [0x7f1c6f08d0c0] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=1 flg=0x4100
n727n iconst 10 (X!=0 X>=0 ) [0x7f1c6f08d2f0] bci=[-1,4,8] rc=1 vc=0 vn=- li=- udi=- nc=0 flg=0x104
n722n lconst 4 (highWordZero X!=0 X>=0 ) [0x7f1c6f08d160] bci=[-1,14,9] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x4104
n723n lconst -12 (X!=0 X<=0 ) [0x7f1c6f08d1b0] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x204
n724n iconst 1 (X!=0 X>=0 ) [0x7f1c6f08d200] bci=[-1,16,9] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x104
n734n lmul [0x7f1c6f08d520] bci=[-1,13,9] rc=1 vc=0 vn=- li=- udi=- nc=2
n732n i2l [0x7f1c6f08d480] bci=[-1,13,9] rc=1 vc=0 vn=- li=- udi=- nc=1
n728n isub [0x7f1c6f08d340] bci=[-1,13,9] rc=1 vc=0 vn=- li=- udi=- nc=2
n694n iload <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] (X>=0 cannotOverflow ) [0x7f1c6f08c8a0] bci=[-1,13,9] rc=2 vc=831 vn=- li=- udi=14 nc=0 flg=0x1100
n725n iconst 9 (X!=0 X>=0 ) [0x7f1c6f08d250] bci=[-1,4,8] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x104
n733n lconst 4 (highWordZero X!=0 X>=0 ) [0x7f1c6f08d4d0] bci=[-1,17,9] rc=1 vc=0 vn=- li=- udi=- nc=0 flg=0x4104
n715n treetop [0x7f1c6f08cf30] bci=[-1,24,10] rc=0 vc=0 vn=- li=- udi=- nc=1
n714n arrayset <arrayset>[#219 helper Method] [flags 0x400 0x0 ] () [0x7f1c6f08cee0] bci=[-1,24,10] rc=1 vc=0 vn=- li=- udi=- nc=3 flg=0x20
n695n aladd (X>=0 internalPtr ) [0x7f1c6f08c8f0] bci=[-1,24,10] rc=1 vc=831 vn=- li=- udi=- nc=2 flg=0x8100
n696n aload <temp slot 3>[#428 Auto] [flags 0x7 0x0 ] (X!=0 ) [0x7f1c6f08c940] bci=[-1,19,10] rc=1 vc=831 vn=- li=- udi=15 nc=0 flg=0x4
n697n lsub (highWordZero X>=0 cannotOverflow ) [0x7f1c6f08c990] bci=[-1,24,10] rc=1 vc=831 vn=- li=- udi=- nc=2 flg=0x5100
n698n lmul (X>=0 cannotOverflow ) [0x7f1c6f08c9e0] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=2 flg=0x1100
n699n i2l (highWordZero X>=0 ) [0x7f1c6f08ca30] bci=[-1,17,9] rc=1 vc=831 vn=- li=- udi=- nc=1 flg=0x4100
n706n iconst 10 (X!=0 X>=0 ) [0x7f1c6f08cc60] bci=[-1,4,8] rc=1 vc=0 vn=- li=- udi=- nc=0 flg=0x104
n701n lconst 4 (highWordZero X!=0 X>=0 ) [0x7f1c6f08cad0] bci=[-1,14,9] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x4104
n702n lconst -8 (X!=0 X<=0 ) [0x7f1c6f08cb20] bci=[-1,24,10] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x204
n703n iconst 2 (X!=0 X>=0 ) [0x7f1c6f08cb70] bci=[-1,23,10] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x104
n713n lmul [0x7f1c6f08ce90] bci=[-1,13,9] rc=1 vc=0 vn=- li=- udi=- nc=2
n711n i2l [0x7f1c6f08cdf0] bci=[-1,13,9] rc=1 vc=0 vn=- li=- udi=- nc=1
n707n isub [0x7f1c6f08ccb0] bci=[-1,13,9] rc=1 vc=0 vn=- li=- udi=- nc=2
n694n ==>iload
n704n iconst 9 (X!=0 X>=0 ) [0x7f1c6f08cbc0] bci=[-1,4,8] rc=1 vc=831 vn=- li=- udi=- nc=0 flg=0x104
n712n lconst 4 (highWordZero X!=0 X>=0 ) [0x7f1c6f08ce40] bci=[-1,24,10] rc=1 vc=0 vn=- li=- udi=- nc=0 flg=0x4104
n737n istore <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] [0x7f1c6f08d610] bci=[-1,4,8] rc=0 vc=0 vn=- li=- udi=- nc=1
n731n iconst 9 (X!=0 X>=0 ) [0x7f1c6f08d430] bci=[-1,4,8] rc=1 vc=0 vn=- li=- udi=- nc=0 flg=0x104
n744n goto --> block_5 BBStart at n1n [0x7f1c6f08d840] bci=[-1,4,8] rc=0 vc=0 vn=- li=- udi=- nc=0
n4n BBEnd </block_4> ===== [0x7f1c6f004970] bci=[-1,6,8] rc=0 vc=831 vn=- li=- udi=- nc=0
So it sets some elements of array m to 1 and then sets some elements of array c to 2. The problem is that after the first call to n, both m and c refer to the same array. So Idiom Recognition seems not to be taking into account the possibility that the same array could be being updated in the loop.
@Qeryu, I took a look at the second test that you mentioned in this issue. I believe that problem is a duplicate of either #18777 or #19218, both of which were fixed by @a7ehuo. Looking at the IL in the jit log that you posted, before PRE we see this IL accessing Test.b:
n9n BBStart <block_3> (freq 1432) (in loop 3) [0x7ff4ab004d00] bci=[-1,16,12] rc=0 vc=835 vn=- li=- udi=- nc=0
n19n lstorei Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] [0x7ff4ab005020] bci=[-1,8,11] rc=0 vc=835 vn=- li=- udi=- nc=2
n17n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab004f80] bci=[-1,4,11] rc=1 vc=835 vn=- li=- udi=31 nc=0 flg=0x104
n18n lconst 2 (highWordZero X!=0 X>=0 )
...
n43n BBStart <block_7> (freq 10000) (in loop 5) [0x7ff4ab0057a0] bci=[-1,22,13] rc=0 vc=835 vn=- li=- udi=- nc=0
n65n istore <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] [0x7ff4ab080410] bci=[-1,42,13] rc=0 vc=835 vn=- li=- udi=8 nc=1
n64n isub [0x7ff4ab0803c0] bci=[-1,40,13] rc=2 vc=835 vn=- li=- udi=- nc=2
n59n iload <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] (cannotOverflow ) [0x7ff4ab080230] bci=[-1,33,13] rc=1 vc=835 vn=- li=- udi=39 nc=0 flg=0x1000
n485n l2i [0x7ff4ab088750] bci=[-1,36,13] rc=1 vc=835 vn=- li=- udi=- nc=1
n62n lloadi Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] (cannotOverflow ) [0x7ff4ab080320] bci=[-1,36,13] rc=1 vc=835 vn=- li=- udi=- nc=1 flg=0x1000
n61n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab0802d0] bci=[-1,35,13] rc=1 vc=835 vn=- li=- udi=40 nc=0 flg=0x104
n76n dstorei <array-shadow>[#232 Shadow] [flags 0x80000606 0x0 ] [0x7ff4ab080780] bci=[-1,44,13] rc=0 vc=835 vn=- li=- udi=- nc=2
n75n aladd (X>=0 internalPtr ) [0x7ff4ab080730] bci=[-1,44,13] rc=1 vc=835 vn=- li=- udi=- nc=2 flg=0x8100
n55n aload <temp slot 9>[#479 Auto] [flags 0x7 0x0 ] (X!=0 ) [0x7ff4ab0800f0] bci=[-1,29,13] rc=1 vc=835 vn=- li=- udi=38 nc=0 flg=0x4
n74n lload <temp slot 8>[#478 Auto] [flags 0x4 0x0 ] (highWordZero X>=0 cannotOverflow ) [0x7ff4ab0806e0] bci=[-1,44,13] rc=1 vc=835 vn=- li=- udi=41 nc=0 flg=0x5100
n66n i2d [0x7ff4ab080460] bci=[-1,43,13] rc=1 vc=835 vn=- li=- udi=- nc=1
n64n ==>isub
...
n515n BBStart <block_53> (freq 1500) (in loop 5) [0x7ff4ab0890b0] bci=[-1,13,12] rc=0 vc=835 vn=- li=- udi=- nc=0
n519n lstorei Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] [0x7ff4ab0891f0] bci=[-1,61,14] rc=0 vc=835 vn=- li=- udi=- nc=2
n518n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab0891a0] bci=[-1,51,14] rc=2 vc=835 vn=- li=- udi=42 nc=0 flg=0x104
n520n lshl [0x7ff4ab089240] bci=[-1,60,14] rc=1 vc=835 vn=- li=- udi=- nc=2
n517n lloadi Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] (cannotOverflow ) [0x7ff4ab089150] bci=[-1,53,14] rc=1 vc=835 vn=- li=- udi=- nc=1 flg=0x1000
n518n ==>aload
n521n iloadi Test.c S[#435 volatile Shadow +24] [flags 0x2602 0x0 ] (cannotOverflow ) [0x7ff4ab089290] bci=[-1,57,14] rc=1 vc=835 vn=- li=- udi=- nc=1 flg=0x1000
n522n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab0892e0] bci=[-1,56,14] rc=1 vc=835 vn=- li=- udi=43 nc=0 flg=0x104
The first segment is assigning 2 to this.b in the outer do-while loop, the second is calculating h -= this.b and the third is calculating this.b <<= this.c.
After PRE, we have this:
Performing 99: partialRedundancyElimination
O^O PARTIAL REDUNDANCY ELIMINATION: 1Placing lloadi optimally : 00007FF4AB08F820 using symRef #480
O^O PARTIAL REDUNDANCY ELIMINATION: Eliminating redundant computation (store) : 00007FF4AB005020
[ 260] O^O PARTIAL REDUNDANCY ELIMINATION: Eliminating redundant computation (lloadi) : 00007FF4AB080320 in block_7, visit count 1077
...
n9n BBStart <block_3> (freq 1432) (in loop 3) [0x7ff4ab004d00] bci=[-1,16,12] rc=0 vc=1063 vn=- li=-1 udi=- nc=0
n19n lstorei Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] [0x7ff4ab005020] bci=[-1,8,11] rc=0 vc=1063 vn=- li=2 udi=- nc=2
n17n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab004f80] bci=[-1,4,11] rc=1 vc=1063 vn=- li=1 udi=31 nc=0 flg=0x104
n18n lconst 2 (highWordZero X!=0 X>=0 ) [0x7ff4ab004fd0] bci=[-1,5,11] rc=2 vc=1063 vn=- li=-1 udi=- nc=0 flg=0x4104
n849n lstore <temp slot 10>[#480 Auto] [flags 0x4 0x0 ] [0x7ff4ab08f910] bci=[-1,5,11] rc=0 vc=0 vn=- li=-1 udi=- nc=1
n18n ==>lconst 2
...
n43n BBStart <block_7> (freq 10000) (in loop 5) [0x7ff4ab0057a0] bci=[-1,22,13] rc=0 vc=1077 vn=- li=-1 udi=- nc=0
n65n istore <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] [0x7ff4ab080410] bci=[-1,42,13] rc=0 vc=1077 vn=- li=-1 udi=8 nc=1
n64n isub [0x7ff4ab0803c0] bci=[-1,40,13] rc=2 vc=1077 vn=- li=13 udi=- nc=2
n59n iload <auto slot 1>[#423 Auto] [flags 0x3 0x0 ] (cannotOverflow ) [0x7ff4ab080230] bci=[-1,33,13] rc=1 vc=1077 vn=- li=11 udi=39 nc=0 flg=0x1000
n485n l2i [0x7ff4ab088750] bci=[-1,36,13] rc=1 vc=1077 vn=- li=12 udi=- nc=1
n62n lload <temp slot 10>[#480 Auto] [flags 0x4 0x0 ] (cannotOverflow createdByPRE ) [0x7ff4ab080320] bci=[-1,36,13] rc=1 vc=1077 vn=- li=2 udi=- nc=0 flg=0x41000
n76n dstorei <array-shadow>[#232 Shadow] [flags 0x80000606 0x0 ] [0x7ff4ab080780] bci=[-1,44,13] rc=0 vc=1077 vn=- li=16 udi=- nc=2
n75n aladd (X>=0 internalPtr ) [0x7ff4ab080730] bci=[-1,44,13] rc=1 vc=1077 vn=- li=-1 udi=- nc=2 flg=0x8100
n55n aload <temp slot 9>[#479 Auto] [flags 0x7 0x0 ] (X!=0 ) [0x7ff4ab0800f0] bci=[-1,29,13] rc=1 vc=1077 vn=- li=9 udi=38 nc=0 flg=0x4
n74n lload <temp slot 8>[#478 Auto] [flags 0x4 0x0 ] (highWordZero X>=0 cannotOverflow ) [0x7ff4ab0806e0] bci=[-1,44,13] rc=1 vc=1077 vn=- li=14 udi=41 nc=0 flg=0x5100
n66n i2d [0x7ff4ab080460] bci=[-1,43,13] rc=1 vc=1077 vn=- li=15 udi=- nc=1
n64n ==>isub
...
n515n BBStart <block_53> (freq 1500) (in loop 5) [0x7ff4ab0890b0] bci=[-1,13,12] rc=0 vc=1079 vn=- li=-1 udi=- nc=0
n519n lstorei Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] () [0x7ff4ab0891f0] bci=[-1,61,14] rc=0 vc=1079 vn=- li=2 udi=- nc=2 flg=0x8
n518n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab0891a0] bci=[-1,51,14] rc=2 vc=1079 vn=- li=1 udi=42 nc=0 flg=0x104
n520n lshl () [0x7ff4ab089240] bci=[-1,60,14] rc=1 vc=1079 vn=- li=- udi=- nc=2 flg=0x8
n517n lloadi Test.b J[#425 Shadow +8] [flags 0x604 0x0 ] (cannotOverflow ) [0x7ff4ab089150] bci=[-1,53,14] rc=1 vc=1079 vn=- li=2 udi=- nc=1 flg=0x1000
n518n ==>aload
n521n iloadi Test.c S[#435 volatile Shadow +24] [flags 0x2602 0x0 ] (cannotOverflow ) [0x7ff4ab089290] bci=[-1,57,14] rc=1 vc=1079 vn=- li=- udi=- nc=1 flg=0x1008
n522n aload <'this' parm LTest;>[#422 Parm] [flags 0x40000107 0x0 ] (X!=0 X>=0 ) [0x7ff4ab0892e0] bci=[-1,56,14] rc=1 vc=1079 vn=- li=1 udi=43 nc=0 flg=0x104
So it sets both this.b and a temporary variable #480 with the value 2 in the outer do-while, uses #480 in the update of h, but at the end, it only calculates this.b <<= this.c - the temporary variable #480 is not similarly updated.
I wasn't ever able to reproduce this problem, so I can't verify that it's fixed by either of the fixes that applied for #18777 or #19218. May I ask you to test whether you are still able to reproduce the problem?
Regardless, we'll look at the Idiom Recognition problem in this issue.
@Spencer-Comin, may I ask you to take a look at this problem in Idiom Recognition?