[PATCH] D13239: [ARM] Modify codegen for memcpy intrinsic to prefer LDM/STM.

[Tim Northover via llvm-commits]

t.p.northover added a comment.

Looks good to me now, with 1+epsilon nits:


================
Comment at: lib/Target/ARM/ARMBaseInstrInfo.cpp:1252
@@ +1251,3 @@
+
+  LDM.addOperand(MI->getOperand(3)).addImm(ARMCC::AL).addReg(0);
+  STM.addOperand(MI->getOperand(2)).addImm(ARMCC::AL).addReg(0);
----------------
This is idiomatically:

    AddDefaultPred(LDM.addOperand(MI->getOperand(3)));

================
Comment at: lib/Target/ARM/ARMBaseInstrInfo.cpp:1259-1263
@@ +1258,7 @@
+  llvm::SmallVector<unsigned, 6> ScratchRegs(NumScratch);
+  std::transform(MI->operands_begin() + 5, MI->operands_end(),
+                 ScratchRegs.begin(),
+                 [] (const MachineOperand &Op) {
+                   return Op.getReg();
+                 });
+  std::sort(ScratchRegs.begin(), ScratchRegs.end(),
----------------
I'm not entirely convinced this is clearer than

    for(unsigned I = 5; I < MI->getNumOperands(); ++I)
      ScratchRegs.push_back(MI->getOperand(I).getReg());

but that's just bike-shedding and probably personal biases, feel free to leave it if you disagree.


http://reviews.llvm.org/D13239