S390: Use FIEBRA/FIDBRA for FP Floor/Ceil

The existing FloatFloor/Ceiling functions had an issue with handling
Number.MIN_VALUE (0x1 ==> 5e-324), and would incorrectly return
MIN_VALUE instead of 0/1.  Simplify the entire sequence by using the
available hardware instruction with the appropriate rounding mode
(POS_INF for Ceiling, NEG_INF for Floor).

Add missing FIEBRA/FIDBRA cases to disassembler.

R=jyan@ca.ibm.com,michael_dawson@ca.ibm.com,mbrandy@us.ibm.com
BUG=

Review URL: https://codereview.chromium.org/1853973002

Cr-Commit-Position: refs/heads/master@{#35239}

src/compiler/s390/code-generator-s390.cc

@@ -1203,13 +1203,12 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       ASSEMBLE_FLOAT_UNOP(sqebr);
       break;
     case kS390_FloorFloat:
-      //      ASSEMBLE_FLOAT_UNOP_RC(frim);
-      __ FloatFloor32(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
-                      kScratchReg);
+      __ fiebra(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+                v8::internal::Assembler::FIDBRA_ROUND_TOWARD_NEG_INF);
       break;
     case kS390_CeilFloat:
-      __ FloatCeiling32(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
-                        kScratchReg, kScratchDoubleReg);
+      __ fiebra(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+                v8::internal::Assembler::FIDBRA_ROUND_TOWARD_POS_INF);
       break;
     case kS390_TruncateFloat:
       __ fiebra(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
@@ -1235,12 +1234,12 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       ASSEMBLE_FLOAT_UNOP(sqdbr);
       break;
     case kS390_FloorDouble:
-      __ FloatFloor64(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
-                      kScratchReg);
+      __ fidbra(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+                v8::internal::Assembler::FIDBRA_ROUND_TOWARD_NEG_INF);
       break;
     case kS390_CeilDouble:
-      __ FloatCeiling64(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
-                        kScratchReg, kScratchDoubleReg);
+      __ fidbra(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+                v8::internal::Assembler::FIDBRA_ROUND_TOWARD_POS_INF);
       break;
     case kS390_TruncateDouble:
       __ fidbra(i.OutputDoubleRegister(), i.InputDoubleRegister(0),

src/s390/disasm-s390.cc

@@ -308,14 +308,18 @@ int Decoder::FormatOption(Instruction* instr, const char* format) {
 int Decoder::FormatMask(Instruction* instr, const char* format) {
   DCHECK(format[0] == 'm');
   int32_t value = 0;
-  if ((format[1] == '1')) {  // prints the mask format in bit 8-12
+  if ((format[1] == '1')) {  // prints the mask format in bits 8-12
     value = reinterpret_cast<RRInstruction*>(instr)->R1Value();
     out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", value);
     return 2;
-  } else if (format[1] == '2') {  // mask format in bit 16 - 19
+  } else if (format[1] == '2') {  // mask format in bits 16-19
     value = reinterpret_cast<RXInstruction*>(instr)->B2Value();
     out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", value);
     return 2;
+  } else if (format[1] == '3') {  // mask format in bits 20-23
+    value = reinterpret_cast<RRFInstruction*>(instr)->M4Value();
+    out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", value);
+    return 2;
   }
 
   out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", value);
@@ -991,6 +995,12 @@ bool Decoder::DecodeFourByte(Instruction* instr) {
     case FLOGR:
       Format(instr, "flogr\t'r5,'r6");
       break;
+    case FIEBRA:
+      Format(instr, "fiebra\t'f5,'m2,'f6,'m3");
+      break;
+    case FIDBRA:
+      Format(instr, "fidbra\t'f5,'m2,'f6,'m3");
+      break;
     // TRAP4 is used in calling to native function. it will not be generated
     // in native code.
     case TRAP4: {

src/s390/macro-assembler-s390.cc

@@ -2407,160 +2407,6 @@ void MacroAssembler::TryInt32Floor(Register result, DoubleRegister double_input,
   bind(&exception);
 }
 
-void MacroAssembler::FloatCeiling32(DoubleRegister double_output,
-                                    DoubleRegister double_input,
-                                    Register scratch,
-                                    DoubleRegister double_scratch) {
-  Label not_zero, no_nan_inf, done, do_ceil;
-  Register scratch2 = r0;
-
-  // Move high word into scratch
-  MovFloatToInt(scratch, double_input);
-
-  // Test for NaN/Inf which results in NaN/Inf respectively
-  static const uint32_t float32ExponentMask = 0x7f800000u;
-  ExtractBitMask(scratch2, scratch, float32ExponentMask);
-  CmpLogical32(scratch2, Operand(0xff));
-  bne(&no_nan_inf, Label::kNear);
-  Move(double_output, double_input);
-  b(&done);
-  bind(&no_nan_inf);
-
-  // Test for double_input in (-1, -0) which results in -0
-  LoadFloat32Literal(double_scratch, -1.0, scratch2);
-  cebr(double_input, double_scratch);
-  ble(&do_ceil, Label::kNear);
-  Cmp32(scratch, Operand::Zero());
-  bgt(&do_ceil, Label::kNear);
-  bne(&not_zero, Label::kNear);
-
-  // double_input = +/- 0 which results in +/- 0 respectively
-  Move(double_output, double_input);
-  b(&done);
-  bind(&not_zero);
-
-  // double_output = -0
-  llihf(scratch2, Operand(0x80000000));
-  ldgr(double_output, scratch2);
-  b(&done);
-  bind(&do_ceil);
-
-  // Regular case
-  // cgdbr(Condition(6), scratch, double_input);
-  // cdfbr(double_output, scratch);
-  fiebra(double_output, double_input, FIDBRA_ROUND_TOWARD_POS_INF);
-  bind(&done);
-}
-
-void MacroAssembler::FloatFloor32(DoubleRegister double_output,
-                                  DoubleRegister double_input,
-                                  Register scratch) {
-  Label not_zero, no_nan_inf, done, do_floor;
-  Register scratch2 = r0;
-
-  // Move high word into scratch
-  MovFloatToInt(scratch, double_input);
-
-  // Test for NaN/Inf which results in NaN/Inf respectively
-  static const uint32_t float32ExponentMask = 0x7f800000u;
-  ExtractBitMask(scratch2, scratch, float32ExponentMask);
-  CmpLogical32(scratch2, Operand(0xff));
-  bne(&no_nan_inf, Label::kNear);
-  Move(double_output, double_input);
-  b(&done);
-  bind(&no_nan_inf);
-
-  // Test for double_input=+/- 0 which results in +/- 0 respectively
-  ltebr(double_input, double_input);
-  bne(&do_floor, Label::kNear);
-  Move(double_output, double_input);
-  b(&done);
-  bind(&do_floor);
-
-  // Regular case
-  // cgdbr(Condition(7), scratch, double_input);
-  // cdfbr(double_output, scratch);
-  fiebra(double_output, double_input, FIDBRA_ROUND_TOWARD_NEG_INF);
-  bind(&done);
-}
-
-void MacroAssembler::FloatCeiling64(DoubleRegister double_output,
-                                    DoubleRegister double_input,
-                                    Register scratch,
-                                    DoubleRegister double_scratch) {
-  Label not_zero, no_nan_inf, done, do_ceil;
-  Register scratch2 = r0;
-
-  // Move high word into scratch
-  StoreDouble(double_input, MemOperand(sp, -kDoubleSize));
-  LoadlW(scratch, MemOperand(sp, -kDoubleSize + Register::kExponentOffset));
-
-  // Test for NaN/Inf which results in NaN/Inf respectively
-  ExtractBitMask(scratch2, scratch, HeapNumber::kExponentMask);
-  CmpLogicalP(scratch2, Operand(0x7ff));
-  bne(&no_nan_inf, Label::kNear);
-  Move(double_output, double_input);
-  b(&done);
-  bind(&no_nan_inf);
-
-  // Test for double_input in (-1, -0) which results in -0
-  LoadDoubleLiteral(double_scratch, -1.0, scratch2);
-  cdbr(double_input, double_scratch);
-  ble(&do_ceil, Label::kNear);
-  Cmp32(scratch, Operand::Zero());
-  bgt(&do_ceil, Label::kNear);
-  bne(&not_zero, Label::kNear);
-
-  // double_input = +/- 0 which results in +/- 0 respectively
-  Move(double_output, double_input);
-  b(&done);
-  bind(&not_zero);
-
-  // double_output = -0
-  llihf(scratch2, Operand(0x80000000));
-  ldgr(double_output, scratch2);
-  b(&done);
-  bind(&do_ceil);
-
-  // Regular case
-  // cgdbr(Condition(6), scratch, double_input);
-  // cdfbr(double_output, scratch);
-  fidbra(double_output, double_input, FIDBRA_ROUND_TOWARD_POS_INF);
-  bind(&done);
-}
-
-void MacroAssembler::FloatFloor64(DoubleRegister double_output,
-                                  DoubleRegister double_input,
-                                  Register scratch) {
-  Label not_zero, no_nan_inf, done, do_floor;
-  Register scratch2 = r0;
-
-  // Move high word into scratch
-  StoreDouble(double_input, MemOperand(sp, -kDoubleSize));
-  LoadlW(scratch, MemOperand(sp, -kDoubleSize + Register::kExponentOffset));
-
-  // Test for NaN/Inf which results in NaN/Inf respectively
-  ExtractBitMask(scratch2, scratch, HeapNumber::kExponentMask);
-  CmpLogicalP(scratch2, Operand(0x7ff));
-  bne(&no_nan_inf, Label::kNear);
-  Move(double_output, double_input);
-  b(&done);
-  bind(&no_nan_inf);
-
-  // Test for double_input=+/- 0 which results in +/- 0 respectively
-  ltdbr(double_input, double_input);
-  bne(&do_floor, Label::kNear);
-  Move(double_output, double_input);
-  b(&done);
-  bind(&do_floor);
-
-  // Regular case
-  // cgdbr(Condition(7), scratch, double_input);
-  // cdfbr(double_output, scratch);
-  fidbra(double_output, double_input, FIDBRA_ROUND_TOWARD_NEG_INF);
-  bind(&done);
-}
-
 void MacroAssembler::TryInlineTruncateDoubleToI(Register result,
                                                 DoubleRegister double_input,
                                                 Label* done) {

src/s390/macro-assembler-s390.h

@@ -1180,22 +1180,6 @@ class MacroAssembler : public Assembler {
                      Register input_high, Register scratch,
                      DoubleRegister double_scratch, Label* done, Label* exact);
 
-  // Perform ceiling of float in input_register and store in double_output.
-  void FloatCeiling32(DoubleRegister double_output, DoubleRegister double_input,
-                      Register scratch, DoubleRegister double_scratch);
-
-  // Perform floor of float in input_register and store in double_output.
-  void FloatFloor32(DoubleRegister double_output, DoubleRegister double_input,
-                    Register scratch);
-
-  // Perform ceiling of double in input_register and store in double_output.
-  void FloatCeiling64(DoubleRegister double_output, DoubleRegister double_input,
-                      Register scratch, DoubleRegister double_scratch);
-
-  // Perform floor of double in input_register and store in double_output.
-  void FloatFloor64(DoubleRegister double_output, DoubleRegister double_input,
-                    Register scratch);
-
   // Performs a truncating conversion of a floating point number as used by
   // the JS bitwise operations. See ECMA-262 9.5: ToInt32. Goes to 'done' if it
   // succeeds, otherwise falls through if result is saturated. On return