s390: optimize for compares

1. use ltr/ltgr when possible
2. combine compares with possible load

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

Review-Url: https://codereview.chromium.org/2696343002
Cr-Commit-Position: refs/heads/master@{#43265}

src/compiler/s390/instruction-codes-s390.h

@@ -135,6 +135,10 @@ namespace compiler {
   V(S390_LoadWordU16)              \
   V(S390_LoadWordS32)              \
   V(S390_LoadWordU32)              \
+  V(S390_LoadAndTestWord32)        \
+  V(S390_LoadAndTestWord64)        \
+  V(S390_LoadAndTestFloat32)       \
+  V(S390_LoadAndTestFloat64)       \
   V(S390_LoadReverse16RR)          \
   V(S390_LoadReverse32RR)          \
   V(S390_LoadReverse64RR)          \

src/compiler/s390/instruction-scheduler-s390.cc

@@ -130,6 +130,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_LoadReverse16RR:
     case kS390_LoadReverse32RR:
     case kS390_LoadReverse64RR:
+    case kS390_LoadAndTestWord32:
+    case kS390_LoadAndTestWord64:
+    case kS390_LoadAndTestFloat32:
+    case kS390_LoadAndTestFloat64:
       return kNoOpcodeFlags;
 
     case kS390_LoadWordS8:

src/s390/assembler-s390.cc

@@ -1250,7 +1250,6 @@ void Assembler::rrfe_form(Opcode op, Condition m3, Condition m4, Register r1,
 // end of S390 Instruction generation
 
 // start of S390 instruction
-RXE_FORM_EMIT(ceb, CEB)
 SS1_FORM_EMIT(ed, ED)
 SS1_FORM_EMIT(mvn, MVN)
 SS1_FORM_EMIT(nc, NC)
@@ -1863,6 +1862,16 @@ void Assembler::sdb(DoubleRegister r1, const MemOperand& opnd) {
            opnd.offset());
 }
 
+void Assembler::ceb(DoubleRegister r1, const MemOperand& opnd) {
+  rxe_form(CEB, Register::from_code(r1.code()), opnd.rx(), opnd.rb(),
+           opnd.offset());
+}
+
+void Assembler::cdb(DoubleRegister r1, const MemOperand& opnd) {
+  rxe_form(CDB, Register::from_code(r1.code()), opnd.rx(), opnd.rb(),
+           opnd.offset());
+}
+
 // Square Root (LB)
 void Assembler::sqdb(DoubleRegister r1, const MemOperand& opnd) {
   rxe_form(SQDB, Register::from_code(r1.code()), opnd.rx(), opnd.rb(),

src/s390/assembler-s390.h

@@ -1006,8 +1006,6 @@ class Assembler : public AssemblerBase {
   }
 
   // S390 instruction sets
-  RXE_FORM(cdb);
-  RXE_FORM(ceb);
   RXE_FORM(ddb);
   SS1_FORM(ed);
   RRF2_FORM(fidbr);
@@ -1200,6 +1198,7 @@ class Assembler : public AssemblerBase {
 
   // Floating Point Compare Instructions
   void cdb(DoubleRegister r1, const MemOperand& opnd);
+  void ceb(DoubleRegister r1, const MemOperand& opnd);
 
   // Floating Point Arithmetic Instructions
   void adb(DoubleRegister r1, const MemOperand& opnd);

src/s390/disasm-s390.cc

@@ -1434,7 +1434,13 @@ bool Decoder::DecodeSixByte(Instruction* instr) {
       Format(instr, "stdy\t'f1,'d2('r2d,'r3)");
       break;
     case ADB:
-      Format(instr, "adb\t'r1,'d1('r2d, 'r3)");
+      Format(instr, "adb\t'f1,'d1('r2d, 'r3)");
+      break;
+    case CDB:
+      Format(instr, "cdb\t'f1,'d1('r2d, 'r3)");
+      break;
+    case CEB:
+      Format(instr, "ceb\t'f1,'d1('r2d, 'r3)");
       break;
     case SDB:
       Format(instr, "sdb\t'r1,'d1('r2d, 'r3)");

src/s390/macro-assembler-s390.cc

@@ -3307,7 +3307,10 @@ void MacroAssembler::MulHighU32(Register dst, Register src1,
 
 void MacroAssembler::Mul32WithOverflowIfCCUnequal(Register dst, Register src1,
                                                   const MemOperand& src2) {
+  Register result = dst;
+  if (src2.rx().is(dst) || src2.rb().is(dst)) dst = r0;
   Generate_Mul32WithOverflowIfCCUnequal(msgf);
+  if (!result.is(dst)) llgfr(result, dst);
 }
 
 void MacroAssembler::Mul32WithOverflowIfCCUnequal(Register dst, Register src1,

src/s390/simulator-s390.cc

@@ -1850,6 +1850,11 @@ double Simulator::ReadDouble(intptr_t addr) {
   return *ptr;
 }
 
+float Simulator::ReadFloat(intptr_t addr) {
+  float* ptr = reinterpret_cast<float*>(addr);
+  return *ptr;
+}
+
 // Returns the limit of the stack area to enable checking for stack overflows.
 uintptr_t Simulator::StackLimit(uintptr_t c_limit) const {
   // The simulator uses a separate JS stack. If we have exhausted the C stack,
@@ -12542,9 +12547,16 @@ EVALUATE(KEB) {
 }
 
 EVALUATE(CEB) {
-  UNIMPLEMENTED();
-  USE(instr);
-  return 0;
+  DCHECK_OPCODE(CEB);
+
+  DECODE_RXE_INSTRUCTION(r1, b2, x2, d2);
+  int64_t b2_val = (b2 == 0) ? 0 : get_register(b2);
+  int64_t x2_val = (x2 == 0) ? 0 : get_register(x2);
+  intptr_t d2_val = d2;
+  float r1_val = get_float32_from_d_register(r1);
+  float fval = ReadFloat(b2_val + x2_val + d2_val);
+  SetS390ConditionCode<float>(r1_val, fval);
+  return length;
 }
 
 EVALUATE(AEB) {

src/s390/simulator-s390.h

@@ -304,6 +304,7 @@ class Simulator {
 
   inline int64_t ReadDW(intptr_t addr);
   inline double ReadDouble(intptr_t addr);
+  inline float ReadFloat(intptr_t addr);
   inline void WriteDW(intptr_t addr, int64_t value);
 
   // S390