[PATCH] D48467: [X86] Recognize a splat of negate in isFNEG

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: lib/Target/X86/X86ISelLowering.cpp:36937
+  }
+  auto Opc = Op.getOpcode();
+  if (Opc == ISD::INSERT_VECTOR_ELT) {
----------------
unsigned - only use auto when the type is very obvious


================
Comment at: lib/Target/X86/X86ISelLowering.cpp:36965
+    // Extract constant bits and see if they are all sign bit masks. Ignore the
+    // undef elements.
+    if (getTargetConstantBitsFromNode(Op1, Op1.getScalarValueSizeInBits(),
----------------
Do we have test coverage for ignoring the undefs?


================
Comment at: lib/Target/X86/X86ISelLowering.cpp:36970
+                                      /* AllowPartialUndefs */ false)) {
+      for (unsigned I = 0; I < EltBits.size(); I++)
+        if (!UndefElts[I] && !EltBits[I].isSignMask())
----------------
```
for (unsigned I = 0, E = EltBits.size(); I < E; I++)
```


================
Comment at: lib/Target/X86/X86ISelLowering.cpp:36980
+    std::swap(Op0, Op1);
+  return Negate(Op0, Op1);
 }
----------------
Why did you create the lambda? Why not just inline Negate?


Repository:
  rL LLVM

https://reviews.llvm.org/D48467