A few of the reference kernels in iree/hal/vmla/op_kernels_generic.h have particularly poor performance. While we expect the LLVM ahead-of-time backend to be more viable as a deployment target, having a faster reference backend is still generally useful.

Some of the slow kernels are already labeled:

https://github

We currently parse this as one location, where the filename is "Decoupled.scala 308:19 Decoupled.scala" and it is at file/col=309:14.

We should instead parse this as two different locations, and use an MLIR FusedLoc to represent them.

Since #384 we can make use of the IR verification features of MLIR.

Our BConv op has a few parameter combinations where it can throw during Init or Prepare. In the converter we try to make sure that none of those cases will ev