[Mlir-commits] [mlir] [memref] Support non-scalar copies in `reinterpret_cast` elision (PR #203873)
ioana ghiban
llvmlistbot at llvm.org
Mon Jun 29 04:50:02 PDT 2026
================
----------------
ioghiban wrote:
if the `reinterpret_cast` result shape is all 1s, every result index is zero, so result strides don't contribute to the address calculation. I wouldn't mention offset here, as the reasoning comes from the result indices. WDYT?
https://github.com/llvm/llvm-project/pull/203873
More information about the Mlir-commits
mailing list