[PATCH] D80208: [CodeGen] Add support for extracting elements of scalable vectors

[Eli Friedman via Phabricator via llvm-commits]

efriedma added inline comments.


================
Comment at: llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:5431
+    // returns a vector type that is guaranteed to only have one element,
+    // which is the case for fixed width vectors.
     if (N1.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
----------------
This comment doesn't seem quite right.

For fixed-width vectors, we only do this for vectors with exactly one element, but that's not because it's semantically incorrect for vectors with more than one element. It's easy to convert indexing on a subvector to indexing on the original vector.  The issue is just that it would make legalization more complex: we don't want to do the transform for vectors with more than one element without a profitability check.


================
Comment at: llvm/test/CodeGen/AArch64/sve-extract-element.ll:171
+; CHECK-NEXT:    mov w8, #9
+; CHECK-NEXT:    whilels p0.d, #0, x8
+; CHECK-NEXT:    lastb d0, p0, z0.d
----------------
This doesn't assemble.  (I assume you meant to refer to xzr?)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D80208/new/

https://reviews.llvm.org/D80208