[PATCH] D69275: Add constrained int->FP intrinsics

[Kevin P. Neal via Phabricator via llvm-commits]

kpn marked an inline comment as done.
kpn added inline comments.


================
Comment at: llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:9145
 
 SDValue SelectionDAG::UnrollVectorOp(SDNode *N, unsigned ResNE) {
+  assert((N->getNumValues() == 1 ||
----------------
craig.topper wrote:
> When are these changes needed? The STRICT_UINT_TO_FP handling in LegalizeVectorOps always scalarizes and goes through ExpandStrictFPOp.
VectorLegalizer::ExpandUINT_TO_FLOAT() uses UnrollVectorOp() if the required instructions are not available. I have a change where it gets called from the top of ExpandStrictFPOp() and instead returns an empty value if it would have called UnrollVectorOp(). I think that may eliminate the need to modify UnrollVectorOp() since it will then fall through into the ExpandUINT_TO_FLOAT() logic instead. If it works it'll generate better code, and eliminating complexity would be nice.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69275/new/

https://reviews.llvm.org/D69275