[PATCH] D102398: [RISCV] Optimize or/xor with immediate

[Luís Marques via Phabricator via llvm-commits]

luismarques added a comment.

We need to optimize integer materialization, both in general and to take advantage of bitmanip. I suppose that the baseline test would be less bad if we had done that, but that this optimization would still be worth it?



================
Comment at: llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp:1293
+  if (auto *ConstOp = dyn_cast<ConstantSDNode>(N)) {
+    // The immediate operand must have only use.
+    if (!(ConstOp->hasOneUse()))
----------------
The code already seems self-explanatory. Remove?


================
Comment at: llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp:1299
+      return false;
+    // The immediate should has only two set bits.
+    if (countPopulation(ConstOp->getZExtValue()) == 2) {
----------------
"The immediate must have exactly two bits set."


================
Comment at: llvm/lib/Target/RISCV/RISCVInstrInfoB.td:90
 
+// Check if this mask has only two set bits and is used only once.
+def BSETINVTwoBitsMask : ComplexPattern<XLenVT, 1, "selectTwoBitsMask">;
----------------
"only two set bits" -> "two set bits", otherwise it seems it's "up to two bits"


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D102398/new/

https://reviews.llvm.org/D102398