OpenCL / Nvidia causing Segmentation Fault at startup

[patriote1k]

Hello, thank you for this repo, very interesting :)

I have cuda 12.8 installed on my system but the software failed when starting with OpenCL configuration (using the default one from example).
I have a 3090 gpu and I run using ubuntu 22.

I have a segmentation fault happening :

########## BEGIN transformed configuration ##########
{
  "command": "Find",
  "finder": {
    "keyProducerJavaRandom": [
      {
        "keyProducerJavaRandomInstance": "SECURE_RANDOM",
        "customSeed": 0,
        "keyProducerId": "exampleKeyProducerSecureRandomId",
        "privateKeyMaxNumBits": 256
      }
    ],
    "consumerJava": {
      "lmdbConfigurationReadOnly": {
        "lmdbDirectory": "lmdb",
        "useProxyOptimal": true,
        "logStatsOnInit": true,
        "logStatsOnClose": false
      },
      "printStatisticsEveryNSeconds": 10,
      "threads": 4,      "delayEmptyConsumer": 50,      "queueSize": 4,
      "runtimePublicKeyCalculationCheck": false,
      "enableVanity": false,
      "vanityPattern": "1[Ee][Mm][Ii][Ll].*"
    },
    "producerJava": [],
    "producerJavaSecretsFiles": [],
    "producerOpenCL": [
      {
        "platformIndex": 0,
        "deviceType": -1,
        "deviceIndex": 0,
        "maxResultReaderThreads": 4,
        "delayBlockedReader": 50,
        "keyProducerId": "exampleKeyProducerSecureRandomId",
        "batchSizeInBits": 16,
        "batchUsePrivateKeyIncrement": true,
        "logSecretBase": false,
        "runOnce": false
      }
    ]
  }
}
########## END   transformed configuration ##########

2025-05-12 15:58:21,193 [main] INFO  net.ladenthin.bitcoinaddressfinder.cli.Main - Find
2025-05-12 15:58:21,220 [main] INFO  net.ladenthin.bitcoinaddressfinder.Finder - startKeyProducer
2025-05-12 15:58:21,221 [main] INFO  net.ladenthin.bitcoinaddressfinder.Finder - startConsumer
2025-05-12 15:58:21,354 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - ##### BEGIN: LMDB stats #####
2025-05-12 15:58:21,354 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - ... this may take a lot of time ...
2025-05-12 15:58:21,355 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - DatabaseSize: 5237.953125 MiB
2025-05-12 15:58:21,356 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - IncreasedCounter: 0
2025-05-12 15:58:21,356 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - IncreasedSum: 0.0 MiB
2025-05-12 15:58:21,357 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - Stat: Stat{branchPages=0, depth=1, entries=1, leafPages=1, overflowPages=0, pageSize=4096}
2025-05-12 15:58:27,654 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - LMDB contains 124362254 unique entries.
2025-05-12 15:58:27,654 [main] INFO  net.ladenthin.bitcoinaddressfinder.persistence.lmdb.LMDBPersistence - ##### END: LMDB stats #####
2025-05-12 15:58:27,655 [pool-3-thread-1] INFO  net.ladenthin.bitcoinaddressfinder.ConsumerJava - start consumeKeysRunner
2025-05-12 15:58:27,655 [pool-3-thread-2] INFO  net.ladenthin.bitcoinaddressfinder.ConsumerJava - start consumeKeysRunner
2025-05-12 15:58:27,655 [pool-3-thread-4] INFO  net.ladenthin.bitcoinaddressfinder.ConsumerJava - start consumeKeysRunner
2025-05-12 15:58:27,655 [pool-3-thread-3] INFO  net.ladenthin.bitcoinaddressfinder.ConsumerJava - start consumeKeysRunner
2025-05-12 15:58:27,655 [main] INFO  net.ladenthin.bitcoinaddressfinder.Finder - configureProducer
2025-05-12 15:58:27,657 [main] INFO  net.ladenthin.bitcoinaddressfinder.Finder - initProducer
2025-05-12 15:58:27,657 [main] INFO  net.ladenthin.bitcoinaddressfinder.ProducerOpenCL - Init producer.
2025-05-12 15:58:37,659 [pool-2-thread-1] INFO  net.ladenthin.bitcoinaddressfinder.ConsumerJava - Statistics: [Checked 0 M keys in 0 minutes] [0 k keys/second] [0 M keys/minute] [Times an empty consumer: 800] [Average contains time: 0 ms] [keys queue size: 0] [Hits: 0]
Segmentation fault      (core dumped)

I checked the stracktrace looks like that :

#1 0x0000798b9d26bead in ?? () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
(...etc)
#2030 0x0000798b9d42b42f in ?? () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
#2031 0x0000798b9d1079cc in ?? () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
#2032 0x0000798b9d0d7859 in ?? () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
#2033 0x0000798b9d0c9151 in ?? () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
#2034 0x0000798b9d0c0905 in ?? () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
#2035 0x0000798b9d0c43ff in NvCliCompileBitcode () from /lib/x86_64-linux-gnu/libnvidia-nvvm.so.4
#2036 0x0000798bc0176ed8 in ?? () from /lib/x86_64-linux-gnu/libnvidia-opencl.so.1
#2037 0x0000798bc017df2c in ?? () from /lib/x86_64-linux-gnu/libnvidia-opencl.so.1
#2038 0x0000798df00c1770 in Java_org_jocl_CL_clBuildProgramNative () from /tmp/libJOCL_2_0_5-linux-x86_64.so
#2039 0x0000798e0bc6b9c0 in ?? ()
#2040 0x0000000000000000 in ?? ()

I'm wondering if it's not a version compatiblity issue with cuda, opencl etc. it works ok on cpu

Any hints ?

@bernardladenthin may you share your minimal system used for benchmarks with nvidia gpus ?

[bernardladenthin]

Thanks for the report! I've already refactored a lot and will release a new version in the next few days. Alternatively, the current main branch can be built using Maven.

The issue is not fully clear to me either, it does seem to work under Linux generally. The segfault likely comes from an incompatibility between CUDA 12.8 and the OpenCL runtime (JOCL/NVVM). If possible, try with an earlier CUDA version to confirm.