larsgeb / m1-gpu-cpp Goto Github PK

Apologies if this is not the right place for such a question but I found the repo and associated blog posts very informative.

I also read your interesting post on CSR GEMV:
https://larsgeb.github.io/2022/03/31/mkl-scipy.html

I was thus wondering if you had experimented with sparse matrix operations in MSL, say SpMV (even if using a naive implementation initiall)?

Got this error when building 02-General Array Operations:

Executing task: xcrun -sdk macosx metal -c ~/m1-gpu-cpp/02-GeneralArrayOperations/ops.metal -o ~/m1-gpu-cpp/02-GeneralArrayOperations/ops.air

~/m1-gpu-cpp/02-GeneralArrayOperations/ops.metal:52:13: error: kernel function cannot have input declarations with 'thread_execution_width' and 'threads_per_simdgroup' attributes
kernel void inspector(
^
~/m1-gpu-cpp/02-GeneralArrayOperations/ops.metal:67:24: note: conflicting declaration with 'thread_execution_width' attribute here
uint thread_execution_width [[thread_execution_width]],
^
~/m1-gpu-cpp/02-GeneralArrayOperations/ops.metal:74:24: note: conflicting declaration with 'threads_per_simdgroup' attribute here
uint threads_per_simdgroup [[threads_per_simdgroup]],
^
1 error generated.

I was able to build the project, but I got the error below during the execution, do you know why?

2023-05-28 13:17:47.203280+0200 test_cpp[75480:1219121] Metal GPU Frame Capture Enabled
2023-05-28 13:17:47.203665+0200 test_cpp[75480:1219121] Metal API Validation Enabled
Failed to find the default library.
Assertion failed: (commandBuffer != nullptr), function sendComputeCommand, file MetalAdder.cpp, line 79.
(lldb)