In VarHandles, we have more through Unsafe tests that are currently present in Hotspot. One of these tests asserts in LinearScan like this:
# Internal Error (/opt/jprt/T/P1/125327.sandoz/s/hotspot/src/share/vm/c1/c1_LinearScan.cpp:5597), pid=17455, tid=17463
# assert(false) failed: cannot spill interval that is used in first instruction (possible reason: no register found)
The problem seems to triggered on i586 only, not on x86_64. I have self-built several JDK 9 builds, and every build seems to fail; the oldest one I tried is jdk9-b80 -- others fail to compile with recent toolchains. Note that public JDK 9 EA builds, including 9b92 work fine, because asserts are stripped from there. It is unknown if the assert signifies a real trouble gone unnoticed in product bits. In product bits, LinearScan bails out, and skips C1 compile of the method:
compilation bailout: LinearScan: no register found
494 205 1 UnsafeTwoCASLong::testAccess (27 bytes) COMPILE SKIPPED: LinearScan: no register found (retry at different tier)
The minimized offending test:
http://cr.openjdk.java.net/~shade/8143930/UnsafeTwoCASLong.java
# Internal Error (/opt/jprt/T/P1/125327.sandoz/s/hotspot/src/share/vm/c1/c1_LinearScan.cpp:5597), pid=17455, tid=17463
# assert(false) failed: cannot spill interval that is used in first instruction (possible reason: no register found)
The problem seems to triggered on i586 only, not on x86_64. I have self-built several JDK 9 builds, and every build seems to fail; the oldest one I tried is jdk9-b80 -- others fail to compile with recent toolchains. Note that public JDK 9 EA builds, including 9b92 work fine, because asserts are stripped from there. It is unknown if the assert signifies a real trouble gone unnoticed in product bits. In product bits, LinearScan bails out, and skips C1 compile of the method:
compilation bailout: LinearScan: no register found
494 205 1 UnsafeTwoCASLong::testAccess (27 bytes) COMPILE SKIPPED: LinearScan: no register found (retry at different tier)
The minimized offending test:
http://cr.openjdk.java.net/~shade/8143930/UnsafeTwoCASLong.java
- blocks
-
JDK-8143628 Fork sun.misc.Unsafe and jdk.internal.misc.Unsafe native method tables
-
- Resolved
-
- relates to
-
JDK-8142314 Bug in C1 ControlFlowOptimizer::delete_unnecessary_jumps with bytecode profiling
-
- Resolved
-