MIPS:[turbofan] Improve unordered comparisons for boolean materialization.

src/compiler/mips/code-generator-mips.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/mips/macro-assembler-mips.h"
 #include "src/scopes.h"
@@ skipping 244 matching lines @@
       return lt;
     case kNotOverflow:
       return ge;
     default:
       break;
   }
   UNREACHABLE();
   return kNoCondition;
 }
 
+FPUCondition FlagsConditionToConditionCmpD(bool& predicate,
+                                           FlagsCondition condition) {
+  switch (condition) {
+    case kEqual:
+      predicate = true;
+      return EQ;
+    case kNotEqual:
+      predicate = false;
+      return EQ;
+    case kUnsignedLessThan:
+      predicate = true;
+      return OLT;
+    case kUnsignedLessThanOrEqual:
+      predicate = true;
+      return OLE;
+    case kUnorderedEqual:
+    case kUnorderedNotEqual:
+      predicate = true;
+      break;
+    default:
+      predicate = true;
+      break;
+  }
+  UNREACHABLE();
+  return kNoFPUCondition;
+}
+
 }  // namespace
 
 
 #define ASSEMBLE_CHECKED_LOAD_FLOAT(width, asm_instr)                         \
   do {                                                                        \
     auto result = i.Output##width##Register();                                \
     auto ool = new (zone()) OutOfLineLoad##width(this, result);               \
     if (instr->InputAt(0)->IsRegister()) {                                    \
       auto offset = i.InputRegister(0);                                       \
       __ Branch(USE_DELAY_SLOT, ool->entry(), hs, offset, i.InputOperand(1)); \
@@ skipping 508 matching lines @@
   Condition cc = kNoCondition;
 
   // MIPS does not have condition code flags, so compare and branch are
   // implemented differently than on the other arch's. The compare operations
   // emit mips psuedo-instructions, which are checked and handled here.
 
   // For materializations, we use delay slot to set the result true, and
   // in the false case, where we fall thru the branch, we reset the result
   // false.
 
-  // TODO(plind): Add CHECK() to ensure that test/cmp and this branch were
-  //    not separated by other instructions.
   if (instr->arch_opcode() == kMipsTst) {
     cc = FlagsConditionToConditionTst(condition);
     __ And(at, i.InputRegister(0), i.InputOperand(1));
     __ Branch(USE_DELAY_SLOT, &done, cc, at, Operand(zero_reg));
     __ li(result, Operand(1));  // In delay slot.
 
   } else if (instr->arch_opcode() == kMipsAddOvf ||
              instr->arch_opcode() == kMipsSubOvf) {
     // kMipsAddOvf, SubOvf emits negative result to 'kCompareReg' on overflow.
     cc = FlagsConditionToConditionOvf(condition);
     __ Branch(USE_DELAY_SLOT, &done, cc, kCompareReg, Operand(zero_reg));
     __ li(result, Operand(1));  // In delay slot.
 
-
   } else if (instr->arch_opcode() == kMipsCmp) {
     Register left = i.InputRegister(0);
     Operand right = i.InputOperand(1);
     cc = FlagsConditionToConditionCmp(condition);
     __ Branch(USE_DELAY_SLOT, &done, cc, left, right);
     __ li(result, Operand(1));  // In delay slot.
 
   } else if (instr->arch_opcode() == kMipsCmpD) {
     FPURegister left = i.InputDoubleRegister(0);
     FPURegister right = i.InputDoubleRegister(1);
-    // TODO(plind): Provide NaN-testing macro-asm function without need for
-    // BranchF.
-    FPURegister dummy1 = f0;
-    FPURegister dummy2 = f2;
-    switch (condition) {
-      case kEqual:
-        // TODO(plind):  improve the NaN testing throughout this function.
-        __ BranchF(NULL, &false_value, kNoCondition, dummy1, dummy2);
-        cc = eq;
-        break;
-      case kNotEqual:
-        __ BranchF(USE_DELAY_SLOT, NULL, &done, kNoCondition, dummy1, dummy2);
-        __ li(result, Operand(1));  // In delay slot - returns 1 on NaN.
-        cc = ne;
-        break;
-      case kUnsignedLessThan:
-        __ BranchF(NULL, &false_value, kNoCondition, dummy1, dummy2);
-        cc = lt;
-        break;
-      case kUnsignedGreaterThanOrEqual:
-        __ BranchF(USE_DELAY_SLOT, NULL, &done, kNoCondition, dummy1, dummy2);
-        __ li(result, Operand(1));  // In delay slot - returns 1 on NaN.
-        cc = ge;
-        break;
-      case kUnsignedLessThanOrEqual:
-        __ BranchF(NULL, &false_value, kNoCondition, dummy1, dummy2);
-        cc = le;
-        break;
-      case kUnsignedGreaterThan:
-        __ BranchF(USE_DELAY_SLOT, NULL, &done, kNoCondition, dummy1, dummy2);
-        __ li(result, Operand(1));  // In delay slot - returns 1 on NaN.
-        cc = gt;
-        break;
-      default:
-        UNSUPPORTED_COND(kMipsCmp, condition);
-        break;
+
+    bool predicate;
+    FPUCondition cc = FlagsConditionToConditionCmpD(predicate, condition);
+    if (!IsMipsArchVariant(kMips32r6)) {
+      __ li(result, Operand(1));
+      __ c(cc, D, left, right);
+      if (predicate) {
+        __ Movf(result, zero_reg);
+      } else {
+        __ Movt(result, zero_reg);
+      }
+    } else {
+      __ cmp(cc, L, kDoubleCompareReg, left, right);
+      __ mfc1(at, kDoubleCompareReg);
+      __ srl(result, at, 31);  // Cmp returns all 1s for true.
+      if (!predicate)          // Toggle result for not equal.
+        __ xori(result, result, 1);
     }
-    __ BranchF(USE_DELAY_SLOT, &done, NULL, cc, left, right);
-    __ li(result, Operand(1));  // In delay slot - branch taken returns 1.
-                                // Fall-thru (branch not taken) returns 0.
-
+    return;
   } else {
     PrintF("AssembleArchBranch Unimplemented arch_opcode is : %d\n",
            instr->arch_opcode());
     TRACE_UNIMPL();
     UNIMPLEMENTED();
   }
-  // Fallthru case is the false materialization.
+
+  // Fallthrough case is the false materialization.
   __ bind(&false_value);
   __ li(result, Operand(0));
   __ bind(&done);
 }
 
 
 void CodeGenerator::AssembleDeoptimizerCall(int deoptimization_id) {
   Address deopt_entry = Deoptimizer::GetDeoptimizationEntry(
       isolate(), deoptimization_id, Deoptimizer::LAZY);
   __ Call(deopt_entry, RelocInfo::RUNTIME_ENTRY);
@@ skipping 266 matching lines @@
     }
   }
   MarkLazyDeoptSite();
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8