Yes, deleting the app's folder in ~/Library/Containers/ fixes the problem. I don't have to restart the Mac, just deleting the folder resets the state of the app which makes it use the default window size.

That post on the Swift Forums is about using a generic type for a Matrix struct. The post here is about using an Array or a buffer for the underlying data storage in a Matrix struct.

What do you mean by "no copying to temporary buffers when passing to BLAS"? In my example code, where am I copying to temporary buffers?

Regarding the *= operator, I can implement it using the function shown below for both the Array based solution and the buffer based solution. So I don't understand how this is supposed to be different for the array or buffer approach. func *= (lhs: inout Matrix<Double>, rhs: Matrix<Double>) { lhs = lhs * rhs }

So the BLAS and LAPACK functions that come with Accelerate are optimized for Apple Silicon?

The BLAS page in the Accelerate documentation does say it is "Apple's implementation of BLAS". See here. As you suggest, I guess it's safe to say that the BLAS functions are optimized for Apple Silicon since it's Apple's implementation of the routines. I can't find any Accelerate documentation about the LAPACK functions. But I guess we can assume those functions are optimized too since the BLAS functions appear to be optimized.

I asked a question on the Swift Forums about redundant functions in Accelerate where I noticed some of the vDSP functions are similar to BLAS functions. This led to my post on here because Steve suggested I ask the question on the Apple Developer Forums. However, on here, I rephrased the question specifically about BLAS and LAPACK instead of other Accelerate features like vDSP.

All of these functions are for 32-bit integers (Int32) what about 64-bit integers (Int64 and Int)?

Yeah, I thought I would try my luck here and see if I get different answers. I'm always conflicted on where to ask questions: on here or on the swift forum.

I tried this but I have to create two arrays to get the results (see reply below). One array contains Float values and the result array contains Double values. It would be preferred to just create the Double array and fill it with random values from BNNS.