Segmentation Fault Running OCCA in Metal Mode on Mac

[nonni-treble]

Hi,

Is anyone using OCCA in 'Metal' mode on macOS?

I've successfully compiled OCCA, and it recognizes the Metal device:

❯ ./occa info 
    ========+======================+==============
     CPU(s) | Processor Name       | Apple M4 Max
            | SIMD Instruction Set | N/A
            | SIMD Width           | 32 bits
            | L1d Cache Size       |   64 KB
            | L1i Cache Size       |  128 KB
            | L2 Cache Size        |    4 MB
            | L3 Cache Size        | 0 bytes
    ========+======================+==============
     Metal  | Device Name          | Apple M4 Max
            | Device ID            | 0
            | Memory               | 27 GB
    ========+======================+==============

However, when I try to run the 01_add_vectors example, I get a segmentation fault:

❯ ./main -d "{mode: 'Metal', device_id: 0}"  
{mode: 'Metal', device_id: 0}
[1]    8389 segmentation fault  ./main -d "{mode: 'Metal', device_id: 0}"

The stack trace:

* thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x5)
  * frame #0: 0x00000001002ffce4 launcher_binary`addVectors + 80
    frame #1: 0x0000000100dd7ad0 libocca.dylib`occa::sys::runFunction(f=(launcher_binary`addVectors), argc=5, args=0x000060000240c000) at runFunction.cpp_codegen:24:5
    frame #2: 0x0000000100db48c4 libocca.dylib`occa::serial::kernel::run(this=0x000000013b808a60) const at kernel.cpp:54:7
    frame #3: 0x0000000100c24ae8 libocca.dylib`occa::launchedModeKernel_t::launcherRun(this=0x000000013b8085b0) const at launchedKernel.cpp:60:21
    frame #4: 0x0000000100c248f0 libocca.dylib`occa::launchedModeKernel_t::run(this=0x000000013b8085b0) const at launchedKernel.cpp:33:7
    frame #5: 0x0000000100b593d8 libocca.dylib`occa::kernel::run(this=0x000000016fdfeda0) const at kernel.cpp:179:17
    frame #6: 0x0000000100b59a78 libocca.dylib`occa::kernel::operator()(this=0x000000016fdfeda0, arg1=0x000000016fdfebd0, arg2=0x000000016fdfebb0, arg3=0x000000016fdfeb90, arg4=0x000000016fdfeb70) const at kernelOperators.cpp_codegen:44:3
    frame #7: 0x0000000100004544 main`main(argc=3, argv=0x000000016fdff338) at main.cpp:89:3
    frame #8: 0x0000000194468274 dyld`start + 2840

System Information:

• macOS Version: 15.3.2
• Compiler: Clang 19.1.7 (installed via Homebrew)

I'm using OCCA 2.0, but I've also tried 1.6 with the same results.

Has anyone encountered this issue, or does anyone have suggestions for debugging Metal support in OCCA?

Thanks in advance!