MIPS: [turbofan] Implement Float32Round(TiesEven|RoundUp|RoundDown|Truncate).

src/compiler/mips64/code-generator-mips64.cc

 // Copyright 2014 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/compiler/code-generator.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
 #include "src/compiler/osr.h"
 #include "src/mips/macro-assembler-mips.h"
@@ skipping 174 matching lines @@
     __ dsrl(at, kScratchReg, 31);
     __ dsll(at, at, 31);
     __ mthc1(at, result_);
   }
 
  private:
   DoubleRegister const result_;
 };
 
 
+class OutOfLineRound32 : public OutOfLineCode {
+ public:
+  OutOfLineRound32(CodeGenerator* gen, DoubleRegister result)
+      : OutOfLineCode(gen), result_(result) {}
+
+  void Generate() final {
+    // Handle rounding to zero case where sign has to be preserved.
+    // High bits of float input already in kScratchReg.
+    __ srl(at, kScratchReg, 31);
+    __ sll(at, at, 31);
+    __ mtc1(at, result_);
+  }
+
+ private:
+  DoubleRegister const result_;
+};
+
+
 class OutOfLineTruncate final : public OutOfLineRound {
  public:
   OutOfLineTruncate(CodeGenerator* gen, DoubleRegister result)
       : OutOfLineRound(gen, result) {}
 };
 
 
 class OutOfLineFloor final : public OutOfLineRound {
  public:
   OutOfLineFloor(CodeGenerator* gen, DoubleRegister result)
       : OutOfLineRound(gen, result) {}
 };
 
 
 class OutOfLineCeil final : public OutOfLineRound {
  public:
   OutOfLineCeil(CodeGenerator* gen, DoubleRegister result)
       : OutOfLineRound(gen, result) {}
 };
 
 
 class OutOfLineTiesEven final : public OutOfLineRound {
  public:
   OutOfLineTiesEven(CodeGenerator* gen, DoubleRegister result)
       : OutOfLineRound(gen, result) {}
 };
 
+class OutOfLineTruncate32 final : public OutOfLineRound32 {

[titzer, 2015/11/26 08:43:03]

Looks like the out of line code is exactly the same for all four cases, so maybe
we can just drop these extra classes?

[dusan.milosavljevic, 2015/11/26 12:52:37]

Right, too much duplicated code.

[dusan.milosavljevic, 2015/11/26 12:52:37]

Done.

+ public:
+  OutOfLineTruncate32(CodeGenerator* gen, DoubleRegister result)
+      : OutOfLineRound32(gen, result) {}
+};
+
+
+class OutOfLineFloor32 final : public OutOfLineRound32 {
+ public:
+  OutOfLineFloor32(CodeGenerator* gen, DoubleRegister result)
+      : OutOfLineRound32(gen, result) {}
+};
+
+
+class OutOfLineCeil32 final : public OutOfLineRound32 {
+ public:
+  OutOfLineCeil32(CodeGenerator* gen, DoubleRegister result)
+      : OutOfLineRound32(gen, result) {}
+};
+
+
+class OutOfLineTiesEven32 final : public OutOfLineRound32 {
+ public:
+  OutOfLineTiesEven32(CodeGenerator* gen, DoubleRegister result)
+      : OutOfLineRound32(gen, result) {}
+};
+
 
 class OutOfLineRecordWrite final : public OutOfLineCode {
  public:
   OutOfLineRecordWrite(CodeGenerator* gen, Register object, Register index,
                        Register value, Register scratch0, Register scratch1,
                        RecordWriteMode mode)
       : OutOfLineCode(gen),
         object_(object),
         index_(index),
         value_(value),
@@ skipping 213 matching lines @@
               Operand(HeapNumber::kExponentBias + HeapNumber::kMantissaBits)); \
     __ mov_d(i.OutputDoubleRegister(), i.InputDoubleRegister(0));              \
     __ asm_instr(i.OutputDoubleRegister(), i.InputDoubleRegister(0));          \
     __ dmfc1(at, i.OutputDoubleRegister());                                    \
     __ Branch(USE_DELAY_SLOT, ool->entry(), eq, at, Operand(zero_reg));        \
     __ cvt_d_l(i.OutputDoubleRegister(), i.OutputDoubleRegister());            \
     __ bind(ool->exit());                                                      \
     __ bind(&done);                                                            \
   } while (0)
 
+#define ASSEMBLE_ROUND_FLOAT_TO_FLOAT(asm_instr, operation)                    \
+  do {                                                                         \
+    int32_t kFloat32ExponentBias = 127;                                        \
+    int32_t kFloat32MantissaBits = 23;                                         \
+    int32_t kFloat32ExponentBits = 8;                                          \
+    auto ool =                                                                 \
+        new (zone()) OutOfLine##operation##32(this, i.OutputDoubleRegister()); \
+    Label done;                                                                \
+    __ mfc1(kScratchReg, i.InputDoubleRegister(0));                            \
+    __ Ext(at, kScratchReg, kFloat32MantissaBits, kFloat32ExponentBits);       \
+    __ Branch(USE_DELAY_SLOT, &done, hs, at,                                   \
+              Operand(kFloat32ExponentBias + kFloat32MantissaBits));           \
+    __ mov_s(i.OutputDoubleRegister(), i.InputDoubleRegister(0));              \
+    __ asm_instr(i.OutputDoubleRegister(), i.InputDoubleRegister(0));          \
+    __ mfc1(at, i.OutputDoubleRegister());                                     \
+    __ Branch(USE_DELAY_SLOT, ool->entry(), eq, at, Operand(zero_reg));        \
+    __ cvt_s_w(i.OutputDoubleRegister(), i.OutputDoubleRegister());            \
+    __ bind(ool->exit());                                                      \
+    __ bind(&done);                                                            \
+  } while (0)
 
 void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
   int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
   if (sp_slot_delta > 0) {
     __ daddiu(sp, sp, sp_slot_delta * kPointerSize);
   }
   if (frame()->needs_frame()) {
     __ Pop(ra, fp);
   }
   frame_access_state()->SetFrameAccessToDefault();
@@ skipping 408 matching lines @@
                i.InputDoubleRegister(1));
       break;
     case kMips64MinD:
       __ min_d(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                i.InputDoubleRegister(1));
       break;
     case kMips64Float64RoundDown: {
       ASSEMBLE_ROUND_DOUBLE_TO_DOUBLE(floor_l_d, Floor);
       break;
     }
+    case kMips64Float32RoundDown: {
+      ASSEMBLE_ROUND_FLOAT_TO_FLOAT(floor_w_s, Floor);
+      break;
+    }
     case kMips64Float64RoundTruncate: {
       ASSEMBLE_ROUND_DOUBLE_TO_DOUBLE(trunc_l_d, Truncate);
       break;
     }
+    case kMips64Float32RoundTruncate: {
+      ASSEMBLE_ROUND_FLOAT_TO_FLOAT(trunc_w_s, Truncate);
+      break;
+    }
     case kMips64Float64RoundUp: {
       ASSEMBLE_ROUND_DOUBLE_TO_DOUBLE(ceil_l_d, Ceil);
       break;
     }
+    case kMips64Float32RoundUp: {
+      ASSEMBLE_ROUND_FLOAT_TO_FLOAT(ceil_w_s, Ceil);
+      break;
+    }
     case kMips64Float64RoundTiesEven: {
       ASSEMBLE_ROUND_DOUBLE_TO_DOUBLE(round_l_d, TiesEven);
       break;
     }
+    case kMips64Float32RoundTiesEven: {
+      ASSEMBLE_ROUND_FLOAT_TO_FLOAT(round_w_s, TiesEven);
+      break;
+    }
     case kMips64Float64Max: {
       // (b < a) ? a : b
       if (kArchVariant == kMips64r6) {
         __ cmp_d(OLT, i.OutputDoubleRegister(), i.InputDoubleRegister(1),
                  i.InputDoubleRegister(0));
         __ sel_d(i.OutputDoubleRegister(), i.InputDoubleRegister(1),
                  i.InputDoubleRegister(0));
       } else {
         __ c_d(OLT, i.InputDoubleRegister(0), i.InputDoubleRegister(1));
         // Left operand is result, passthrough if false.
@@ skipping 823 matching lines @@
       padding_size -= v8::internal::Assembler::kInstrSize;
     }
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8