Out-of-memory in dawn_spirv_cross_glsl_fast_fuzzer |
||||||
Issue descriptionDetailed report: https://clusterfuzz.com/testcase?key=6431235930587136 Fuzzer: libFuzzer_dawn_spirv_cross_glsl_fast_fuzzer Job Type: libfuzzer_chrome_ubsan Platform Id: linux Crash Type: Out-of-memory (exceeds 2048 MB) Crash Address: Crash State: dawn_spirv_cross_glsl_fast_fuzzer Sanitizer: undefined (UBSAN) Regressed: https://clusterfuzz.com/revisions?job=libfuzzer_chrome_ubsan&range=609741:609751 Reproducer Testcase: https://clusterfuzz.com/download?testcase_id=6431235930587136 Issue filed automatically. See https://chromium.googlesource.com/chromium/src/+/master/testing/libfuzzer/reference.md for more information.
,
Nov 23
Automatically adding ccs based on OWNERS file / target commit history. If this is incorrect, please add ClusterFuzz-Wrong label.
,
Nov 23
,
Nov 23
These all boil down to a very high bound id being set, i.e. the SPIRV binary says it is going to use a ton of ids, so the cross compiler is pre-allocating storage for them. If you crank up the memory limit the operation does succeed, so it isn't actually a real world crasher. I am looking into if these large bounds are valid, I suspect they are. And if they are is this a completely intended behaviour, or is there something else that should be being done wrt allocating id data structures.
,
Nov 24
,
Dec 1
ClusterFuzz testcase 6431235930587136 appears to be flaky, updating reproducibility label.
,
Dec 1
Please ignore the last comment about testcase being unreproducible. The testcase is still reproducible. This happened due to a code refactoring on ClusterFuzz side, and the underlying root cause is now fixed. Resetting the label back to Reproducible. Sorry about the inconvenience caused from these incorrect notifications.
,
Dec 4
This will either be resolved by performing some sort of optimization to pack ids in the new shaderc wrapper for this code or directly resolving https://github.com/KhronosGroup/SPIRV-Cross/issues/781. |
||||||
►
Sign in to add a comment |
||||||
Comment 1 by ClusterFuzz
, Nov 23