[PATCH] D34515: [ARM] Materialise some boolean values to avoid a branch

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

The use of clz is very similar to what the PowerPC backend does; I wonder if we can pick up any additional tricks from there.



================
Comment at: lib/Target/ARM/ARMISelLowering.cpp:10015-10025
+  if (N->getOpcode() == ARMISD::SUBC) {
+    // (SUBC (ADDE 0, 0, C), 1) -> C
+    SDValue LHS = N->getOperand(0);
+    SDValue RHS = N->getOperand(1);
+    if (LHS->getOpcode() == ARMISD::ADDE &&
+        isNullConstant(LHS->getOperand(0)) &&
+        isNullConstant(LHS->getOperand(1)) && isOneConstant(RHS)) {
----------------
rogfer01 wrote:
> I think this may make the `ADDC` combiner above redundant as `ADDC x, -1` will usually become `SUBC x, 1`. I'll try to see if I can remove the `ADDC` one.
Did you end up checking this?


================
Comment at: lib/Target/ARM/ARMISelLowering.cpp:7443
+  if (isNullConstant(Carry))
+    return SDValue();
+
----------------
When do we hit this case?  I would expect that normally DAGCombiner::visitADDCARRY would combine this away.


================
Comment at: lib/Target/ARM/ARMISelLowering.cpp:12245
 
+  if (!VT.isInteger())
+      return SDValue();
----------------
Do we generate ARMISD::CMOV for values which aren't integers?


https://reviews.llvm.org/D34515