RFR: 8326615: C1/C2 don't handle allocation failure properly during initialization (RuntimeStub::new_runtime_stub fatal crash) [v5]

Damon Fenacci dfenacci at openjdk.org
Tue Jul 23 15:26:34 UTC 2024


On Tue, 9 Jul 2024 15:05:55 GMT, Vladimir Kozlov <kvn at openjdk.org> wrote:

>> Max used code cache size running a simple HelloWorld:
>> |                | Int   | C1 only | C2 only | C1/C2 tiered |
>> |----------------|-------|---------|---------|--------------|
>> | linux-x64 (16 core, avx512)      | 785Kb | 5400Kb  | 1521Kb  | 6219Kb       |
>> | macosx-aarch64 (10 core) | 673Kb | 3210Kb  | 1357Kb  | 4701Kb       |
>> | linux-aarch64  (16 core) | 671Kb | 6142Kb  | 1462Kb  | 6890Kb       |
>
> @dafedafe, can you show full output (including version lines) of `java -XX:+PrintCodeCache -XX:TieredStopAtLevel=1 -version` on linux-x64?
> I want to see what consumes CodeCache in your experiments.

@vnkozlov here it is (sorry for the delay, I updated the repo as well but the results are similar):

build/linux-x64/jdk/bin/java -XX:+PrintCodeCache -XX:+TieredCompilation -XX:TieredStopAtLevel=1 -version
java version "24-internal" 2025-03-18
Java(TM) SE Runtime Environment (build 24-internal-2024-07-23-1502438.dfenacci...)
Java HotSpot(TM) 64-Bit Server VM (build 24-internal-2024-07-23-1502438.dfenacci..., mixed mode, emulated-client)
CodeCache: size=49152Kb used=3486Kb max_used=3486Kb free=45665Kb
 bounds [0x000070510d000000, 0x000070510d370000, 0x0000705110000000]
 total_blobs=916, nmethods=556, adapters=284, full_count=0
Compilation: enabled, stopped_count=0, restarted_count=0


There seem to be many more methods that get compiled with C1 than with C2 when just running `-version`:

- `java -XX:+PrintCodeCache -XX:+TieredCompilation -XX:TieredStopAtLevel=1 -XX:+PrintCompilation -version`:[C1.log](https://github.com/user-attachments/files/16350954/C1.log)

- `java -XX:+PrintCodeCache -XX:-TieredCompilation -XX:+PrintCompilation -version`: [C2.log](https://github.com/user-attachments/files/16350951/C2.log)

-------------

PR Comment: https://git.openjdk.org/jdk/pull/19280#issuecomment-2245551951


More information about the hotspot-compiler-dev mailing list