Paul Fultz II

For type-erased classes this will now check that the class implements the methods its asking for. Currently it only check the method names and correct parameters, it is still missing...

Better simplification of shape transformation operators such as reshape/transpose/broadcast

9

Currently, in migraphx, we only simplify shape transformation of the same operator such as repeated transpose or reshapes. However, this will simplify across reshape/transpose/broadcast. It produces a much simpler set...

Fuse reductions with MLIR with multi-outputs

5

This issue has two parts. The first part is to fuse reductions(including split reductions) with MLIR, including any pointwise. The second part is to use multiple outputs when fusing, to...

Reuse inner buffers when generating kernels for blockwise reductions

2

Due to the use of `__syncthreads` in the `reduce` methods registers are not reused. We can reuse them directly by assigning to them with `r.inner([](auto& y, auto x) { y...

Preload transposed inputs in LDS for pointwise kernels

1

To improve performance for transpose kernels we should load the transposed inputs into LDS directly, and then read from LDS instead. We have function like `preload_copy` which will do this...