Adds more SIMD instructions to arm64.

runtime/vm/intermediate_language_arm.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"  // Needed here to get TARGET_ARCH_ARM.
 #if defined(TARGET_ARCH_ARM)
 
 #include "vm/intermediate_language.h"
 
 #include "vm/cpu.h"
@@ skipping 3908 matching lines @@
   const QRegister result = locs()->out(0).fpu_reg();
   const DRegister dresult0 = EvenDRegisterOf(result);
   const DRegister dresult1 = OddDRegisterOf(result);
   const SRegister sresult0 = EvenSRegisterOf(dresult0);
   const SRegister sresult1 = OddSRegisterOf(dresult0);
   const SRegister sresult2 = EvenSRegisterOf(dresult1);
   const SRegister sresult3 = OddSRegisterOf(dresult1);
 
   const DRegister dvalue0 = EvenDRegisterOf(value);
   const DRegister dvalue1 = OddDRegisterOf(value);
+  const SRegister svalue0 = EvenSRegisterOf(dvalue0);
+  const SRegister svalue1 = OddSRegisterOf(dvalue0);
+  const SRegister svalue2 = EvenSRegisterOf(dvalue1);
+  const SRegister svalue3 = OddSRegisterOf(dvalue1);
 
   const DRegister dtemp0 = DTMP;
   const DRegister dtemp1 = OddDRegisterOf(QTMP);
 
   // For some cases the vdup instruction requires fewer
   // instructions. For arbitrary shuffles, use vtbl.
 
   switch (op_kind()) {
     case MethodRecognizer::kFloat32x4ShuffleX:
-      __ vdup(kWord, result, dvalue0, 0);
-      __ vcvtds(dresult0, sresult0);
+      __ vcvtds(dresult0, svalue0);
       break;
     case MethodRecognizer::kFloat32x4ShuffleY:
-      __ vdup(kWord, result, dvalue0, 1);
-      __ vcvtds(dresult0, sresult0);
+      __ vcvtds(dresult0, svalue1);
       break;
     case MethodRecognizer::kFloat32x4ShuffleZ:
-      __ vdup(kWord, result, dvalue1, 0);
-      __ vcvtds(dresult0, sresult0);
+      __ vcvtds(dresult0, svalue2);
       break;
     case MethodRecognizer::kFloat32x4ShuffleW:
-      __ vdup(kWord, result, dvalue1, 1);
-      __ vcvtds(dresult0, sresult0);
+      __ vcvtds(dresult0, svalue3);
       break;
     case MethodRecognizer::kInt32x4Shuffle:
     case MethodRecognizer::kFloat32x4Shuffle:
       if (mask_ == 0x00) {
         __ vdup(kWord, result, dvalue0, 0);
       } else if (mask_ == 0x55) {
         __ vdup(kWord, result, dvalue0, 1);
       } else if (mask_ == 0xAA) {
         __ vdup(kWord, result, dvalue1, 0);
       } else  if (mask_ == 0xFF) {
@@ skipping 976 matching lines @@
   summary->set_out(0, Location::RequiresFpuRegister());
   return summary;
 }
 
 
 void BinaryInt32x4OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   const QRegister left = locs()->in(0).fpu_reg();
   const QRegister right = locs()->in(1).fpu_reg();
   const QRegister result = locs()->out(0).fpu_reg();
   switch (op_kind()) {
-    case Token::kBIT_AND: {
-      __ vandq(result, left, right);
-      break;
-    }
-    case Token::kBIT_OR: {
-      __ vorrq(result, left, right);
-      break;
-    }
-    case Token::kBIT_XOR: {
-      __ veorq(result, left, right);
-      break;
-    }
-    case Token::kADD:
-      __ vaddqi(kWord, result, left, right);
-      break;
-    case Token::kSUB:
-      __ vsubqi(kWord, result, left, right);
-      break;
+    case Token::kBIT_AND: __ vandq(result, left, right); break;
+    case Token::kBIT_OR: __ vorrq(result, left, right); break;
+    case Token::kBIT_XOR: __ veorq(result, left, right); break;
+    case Token::kADD: __ vaddqi(kWord, result, left, right); break;
+    case Token::kSUB: __ vsubqi(kWord, result, left, right); break;
     default: UNREACHABLE();
   }
 }
 
 
 LocationSummary* MathUnaryInstr::MakeLocationSummary(Isolate* isolate,
                                                      bool opt) const {
   if ((kind() == MathUnaryInstr::kSin) || (kind() == MathUnaryInstr::kCos)) {
     const intptr_t kNumInputs = 1;
     const intptr_t kNumTemps = 0;
@@ skipping 1462 matching lines @@
   compiler->GenerateCall(token_pos(),
                          &label,
                          PcDescriptors::kOther,
                          locs());
   __ Drop(ArgumentCount());  // Discard arguments.
 }
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM