[PATCH] D26518: [AArch64] Compute the Newton series for reciprocals natively

[James Molloy via llvm-commits]

jmolloy requested changes to this revision.
jmolloy added a reviewer: jmolloy.
jmolloy added a comment.
This revision now requires changes to proceed.

Hi Evandro,

Some comments (purely technical this time, thank goodness!) :)

James



================
Comment at: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:4658
                                        DAG, ExtraSteps)) {
-      UseOneConst = true;
+      SDLoc DL(Operand);
+      EVT VT = Operand.getValueType();
----------------
This seems a bit of a hack. Can we at least get a comment saying this is working around the model somehow?

Can Sanjay comment on how the model was meant to be used where targets have native step support?


================
Comment at: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:4681
+        // Correct the result if the operand is 0.0.
+        Estimate = DAG.getNode(VT.isVector() ? ISD::VSELECT : ISD::SELECT, DL,
+                               VT, Eq, Operand, Estimate);
----------------
Why do you need this? Is this to preserve sign-of-zero?

You don't need this if the original operation was marked nsz or fast.


Repository:
  rL LLVM

https://reviews.llvm.org/D26518