[ARM] Constant Materialize: imms with specific value can be encoded into mov.w

[weimingz]

Summary

Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes.

[jmolloy]

LGTM, thanks!

[weimingz]

The original test case doesn't explicitly specify triple. On my build for arm-linux, the original test is OK.
But for apple triple, the behavior is different for O0 and default opt level.

lib/Target/ARM/ARMISelDAGToDAG.cpp

bool ARMDAGToDAGISel::isShifterOpProfitable(const SDValue &Shift,
   // R << 2 is free.
   return ShOpcVal == ARM_AM::lsl &&
          (ShAmt == 2 || (Subtarget->isSwift() && ShAmt == 1));
 }
 
 unsigned ARMDAGToDAGISel::ConstantMaterializationCost(unsigned Val) const {
   if (Subtarget->isThumb()) {
     if (Val <= 255) return 1;                               // MOV
-    if (Subtarget->hasV6T2Ops() && Val <= 0xffff) return 1; // MOVW
+    if (Subtarget->hasV6T2Ops() &&
+        (Val <= 0xffff || ARM_AM::getT2SOImmValSplatVal(Val) != -1))
+      return 1; // MOVW
     if (Val <= 510) return 2;                               // MOV + ADDi8
     if (~Val <= 255) return 2;                              // MOV + MVN
     if (ARM_AM::isThumbImmShiftedVal(Val)) return 2;        // MOV + LSL
   } else {
     if (ARM_AM::getSOImmVal(Val) != -1) return 1;           // MOV
     if (ARM_AM::getSOImmVal(~Val) != -1) return 1;          // MVN
     if (Subtarget->hasV6T2Ops() && Val <= 0xffff) return 1; // MOVW
     if (ARM_AM::isSOImmTwoPartVal(Val)) return 2;           // two instrs

test/CodeGen/ARM/subtarget-no-movt.ll

 ; NO-USE-MOVT: [[L0]]:
 ; NO-USE-MOVT: .long 2296237089
 
 define i32 @foo1(i32 %a) {
   %1 = xor i32 -1998730207, %a
   ret i32 %1
 }
 
+; NO-OPTION-LABEL: {{_?}}foo2
+; NO-OPTION: mov.w  r0, #-536813568
+; NO-OPTION-LABEL-NOT: .long
+
+; USE-MOVT-LABEL: {{_?}}foo2
+; USE-MOVT: mov.w r0, #-536813568
+; USE-MOVT-NOT: .long
+
+; NO-USE-MOVT-LABEL: {{_?}}foo2
+; NO-USE-MOVT: mov.w  r0, #-536813568
+; NO-USE-MOVT-NOT: .long
+define i32 @foo2() {
+  %1 = load i32, i32* inttoptr (i32 -536813568 to i32*) ; load from 0xe000e000
+  ret i32 %1
+}
+
+
 attributes #0 = { "target-features"="+no-movt" }