[PowerPC] better instruction selection for OR (XOR) with a 32-bit immediate

- recommitting after fixing a test failure on MacOS

On PPC64, OR (XOR) with a 32-bit immediate can be done with only two instructions, i.e. ori + oris.
But the current LLVM generates three or four instructions for this purpose (and also it clobbers one GPR).

This patch makes PPC backend generate ori + oris (xori + xoris) for OR (XOR) with a 32-bit immediate.

e.g. (x | 0xFFFFFFFF) should be

	ori 3, 3, 65535
	oris 3, 3, 65535

but LLVM generates without this patch

	li 4, 0
	oris 4, 4, 65535
	ori 4, 4, 65535
	or 3, 3, 4

Differential Revision: https://reviews.llvm.org/D34757

llvm-svn: 311538

1 files changed, 51 insertions, 0 deletions

diff --git a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 4867f77bdc5..2eba6c7fa99 100644
--- a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -133,6 +133,12 @@ namespace {
     void PreprocessISelDAG() override;
     void PostprocessISelDAG() override;
 
+    /// getI16Imm - Return a target constant with the specified value, of type
+    /// i16.
+    inline SDValue getI16Imm(unsigned Imm, const SDLoc &dl) {
+      return CurDAG->getTargetConstant(Imm, dl, MVT::i16);
+    }
+
     /// getI32Imm - Return a target constant with the specified value, of type
     /// i32.
     inline SDValue getI32Imm(unsigned Imm, const SDLoc &dl) {
@@ -450,6 +456,12 @@ static bool isInt32Immediate(SDValue N, unsigned &Imm) {
   return isInt32Immediate(N.getNode(), Imm);
 }
 
+/// isInt64Immediate - This method tests to see if the value is a 64-bit
+/// constant operand. If so Imm will receive the 64-bit value.
+static bool isInt64Immediate(SDValue N, uint64_t &Imm) {
+  return isInt64Immediate(N.getNode(), Imm);
+}
+
 static unsigned getBranchHint(unsigned PCC, FunctionLoweringInfo *FuncInfo,
                               const SDValue &DestMBB) {
   assert(isa<BasicBlockSDNode>(DestMBB));
@@ -3375,12 +3387,51 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
       }
     }
 
+    // OR with a 32-bit immediate can be handled by ori + oris
+    // without creating an immediate in a GPR.
+    uint64_t Imm64 = 0;
+    bool IsPPC64 = PPCSubTarget->isPPC64();
+    if (IsPPC64 && isInt64Immediate(N->getOperand(1), Imm64) &&
+        (Imm64 & ~0xFFFFFFFFuLL) == 0) {
+      // If ImmHi (ImmHi) is zero, only one ori (oris) is generated later.
+      uint64_t ImmHi = Imm64 >> 16;
+      uint64_t ImmLo = Imm64 & 0xFFFF;
+      if (ImmHi != 0 && ImmLo != 0) {
+        SDNode *Lo = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64,
+                                            N->getOperand(0),
+                                            getI16Imm(ImmLo, dl));
+        SDValue Ops1[] = { SDValue(Lo, 0), getI16Imm(ImmHi, dl)};
+        CurDAG->SelectNodeTo(N, PPC::ORIS8, MVT::i64, Ops1);
+        return;
+      }
+    }
+
     // Other cases are autogenerated.
     break;
   }
   case ISD::XOR: {
     if (tryLogicOpOfCompares(N))
       return;
+
+    // XOR with a 32-bit immediate can be handled by xori + xoris
+    // without creating an immediate in a GPR.
+    uint64_t Imm64 = 0;
+    bool IsPPC64 = PPCSubTarget->isPPC64();
+    if (IsPPC64 && isInt64Immediate(N->getOperand(1), Imm64) &&
+        (Imm64 & ~0xFFFFFFFFuLL) == 0) {
+      // If ImmHi (ImmHi) is zero, only one xori (xoris) is generated later.
+      uint64_t ImmHi = Imm64 >> 16;
+      uint64_t ImmLo = Imm64 & 0xFFFF;
+      if (ImmHi != 0 && ImmLo != 0) {
+        SDNode *Lo = CurDAG->getMachineNode(PPC::XORI8, dl, MVT::i64,
+                                            N->getOperand(0),
+                                            getI16Imm(ImmLo, dl));
+        SDValue Ops1[] = { SDValue(Lo, 0), getI16Imm(ImmHi, dl)};
+        CurDAG->SelectNodeTo(N, PPC::XORIS8, MVT::i64, Ops1);
+        return;
+      }
+    }
+
     break;
   }
   case ISD::ADD: {