[mlir][SparseTensor] Enable VLA ops in index value generation
authorJavier Setoain <javier.setoain@gmail.com>
Tue, 26 Apr 2022 13:39:29 +0000 (14:39 +0100)
committerJavier Setoain <javier.setoain@gmail.com>
Thu, 28 Apr 2022 08:39:07 +0000 (09:39 +0100)
commit6301574206b39f72edb957f3b069f3892c117d4b
tree4ebad7cf3e7e8e5ecff8dba1c8b4f6bc7c5773a4
parent515f890033ccaad3890333513731e4c5b6db2482
[mlir][SparseTensor] Enable VLA ops in index value generation

Current index value generation uses fixed-length vector ops, this patch
adds an alterantive codegen path compatible with scalable vectors by
using `LLVM::StepVectorOp`.

Differential Revision: https://reviews.llvm.org/D124454
mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp