s390: optimize MathSqrt

R=joransiu@ca.ibm.com, bjaideep@ca.ibm.com
BUG=

Review-Url: https://codereview.chromium.org/2615683003
Cr-Commit-Position: refs/heads/master@{#42077}

src/crankshaft/s390/lithium-codegen-s390.cc

@@ -3521,9 +3521,13 @@ void LCodeGen::DoMathFround(LMathFround* instr) {
 }
 
 void LCodeGen::DoMathSqrt(LMathSqrt* instr) {
-  DoubleRegister input = ToDoubleRegister(instr->value());
   DoubleRegister result = ToDoubleRegister(instr->result());
-  __ sqdbr(result, input);
+  LOperand* input = instr->value();
+  if (input->IsDoubleRegister()) {
+    __ Sqrt(result, ToDoubleRegister(instr->value()));
+  } else {
+    __ Sqrt(result, ToMemOperand(input));
+  }
 }
 
 void LCodeGen::DoMathPowHalf(LMathPowHalf* instr) {

src/crankshaft/s390/lithium-s390.cc

@@ -1058,7 +1058,7 @@ LInstruction* LChunkBuilder::DoMathExp(HUnaryMathOperation* instr) {
 }
 
 LInstruction* LChunkBuilder::DoMathSqrt(HUnaryMathOperation* instr) {
-  LOperand* input = UseRegisterAtStart(instr->value());
+  LOperand* input = UseAtStart(instr->value());
   LMathSqrt* result = new (zone()) LMathSqrt(input);
   return DefineAsRegister(result);
 }

src/s390/macro-assembler-s390.cc

@@ -3342,6 +3342,17 @@ void MacroAssembler::MulP(Register dst, const MemOperand& opnd) {
 #endif
 }
 
+void MacroAssembler::Sqrt(DoubleRegister result, DoubleRegister input) {
+  sqdbr(result, input);
+}
+void MacroAssembler::Sqrt(DoubleRegister result, const MemOperand& input) {
+  if (is_uint12(input.offset())) {
+    sqdb(result, input);
+  } else {
+    ldy(result, input);
+    sqdbr(result, result);
+  }
+}
 //----------------------------------------------------------------------------
 //  Add Instructions
 //----------------------------------------------------------------------------

src/s390/macro-assembler-s390.h

@@ -323,6 +323,10 @@ class MacroAssembler : public Assembler {
   // Divide
   void DivP(Register dividend, Register divider);
 
+  // Square root
+  void Sqrt(DoubleRegister result, DoubleRegister input);
+  void Sqrt(DoubleRegister result, const MemOperand& input);
+
   // Compare
   void Cmp32(Register src1, Register src2);
   void CmpP(Register src1, Register src2);