runnnlayertests segmentation fault (zero pointer)

[ThomasZoschke]

./katago runnnlayertests

Only nonsense 0 values in output tabular. Then SEGFAULT.

In lldb you can easily see it is a NULL pointer.

0, 0,
0, 0,
0, 0,
0, 0,
0, 0,

---

Process 72129 stopped

• thread Additional networks? #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x0)
  frame #0: 0x0000000239099b0c MetalPerformanceShadersGraphllvm::FailureOr<float> mlir::getSingleFloatValue<float>(mlir::ElementsAttr) + 996 MetalPerformanceShadersGraphmlir::getSingleFloatValue:
  -> 0x239099b0c <+996>: ldr s0, [x19, x9, lsl compiling on Windows #2]
  0x239099b10 <+1000>: str s0, [sp, #0xc]
  0x239099b14 <+1004>: add x8, x8, #0x1
  0x239099b18 <+1008>: cmp x21, x8
  Target 0: (katago) stopped.
  (lldb)

I've been using latest homebrew katago on Apple MacBook Pro M5. That's Silicon built with Metal.

[ThomasZoschke]

WHAT THE F*** ADDED THE NONSENSE HYPERLINKS?

[ChinChangYang]

I can't reproduce the issue on M3 Max. Could you try building from source? See MacOS section in https://github.com/lightvector/KataGo/blob/master/Compiling.md

[ChinChangYang]

I can reproduce this issue now. This issue is related to #1192.

[lightvector]

Thanks! I attempted a fix, I think this is now addressed in c4a5947

[ChinChangYang]

Confirm. This issue is fixed at 11399d2.