Merge pull request #3 from fenghaitao/master

Optimize Float32x4Array and Int32x4Array element loading and storing

src/arm/assembler-arm.h

@@ -366,6 +366,34 @@ struct QwNeonRegister {
     return r;
   }
 
+  static int ToAllocationIndex(QwNeonRegister reg) {
+    ASSERT(reg.code() < kMaxNumRegisters);
+    return reg.code();
+  }
+
+  static const char* AllocationIndexToString(int index) {
+    ASSERT(index >= 0 && index < kMaxNumRegisters);
+    const char* const names[] = {
+      "q0",
+      "q1",
+      "q2",
+      "q3",
+      "q4",
+      "q5",
+      "q6",
+      "q7",
+      "q8",
+      "q9",
+      "q10",
+      "q11",
+      "q12",
+      "q13",
+      "q14",
+      "q15",
+    };
+    return names[index];
+  }
+
   bool is_valid() const {
     return (0 <= code_) && (code_ < kMaxNumRegisters);
   }
@@ -385,6 +413,7 @@ struct QwNeonRegister {
 
 
 typedef QwNeonRegister QuadRegister;
+typedef QwNeonRegister SIMD128Register;
 
 
 // Support for the VFP registers s0 to s31 (d0 to d15).

src/arm/cpu-arm.cc

@@ -56,6 +56,12 @@ bool CPU::SupportsCrankshaft() {
 }
 
 
+bool CPU::SupportsSIMD128InCrankshaft() {
+  // Not Implemented.
+  return false;
+}
+
+
 void CPU::FlushICache(void* start, size_t size) {
   // Nothing to do flushing no instructions.
   if (size == 0) {

src/arm/deoptimizer-arm.cc

@@ -113,7 +113,7 @@ void Deoptimizer::SetPlatformCompiledStubRegisters(
 }
 
 
-void Deoptimizer::CopyDoubleRegisters(FrameDescription* output_frame) {
+void Deoptimizer::CopySIMD128Registers(FrameDescription* output_frame) {
   for (int i = 0; i < DwVfpRegister::kMaxNumRegisters; ++i) {
     double double_value = input_->GetDoubleRegister(i);
     output_frame->SetDoubleRegister(i, double_value);
@@ -210,7 +210,7 @@ void Deoptimizer::EntryGenerator::Generate() {
 
   // Copy VFP registers to
   // double_registers_[DoubleRegister::kMaxNumAllocatableRegisters]
-  int double_regs_offset = FrameDescription::double_registers_offset();
+  int double_regs_offset = FrameDescription::simd128_registers_offset();
   for (int i = 0; i < DwVfpRegister::kMaxNumAllocatableRegisters; ++i) {
     int dst_offset = i * kDoubleSize + double_regs_offset;
     int src_offset = i * kDoubleSize + kNumberOfRegisters * kPointerSize;
@@ -284,7 +284,7 @@ void Deoptimizer::EntryGenerator::Generate() {
   __ CheckFor32DRegs(ip);
 
   __ ldr(r1, MemOperand(r0, Deoptimizer::input_offset()));
-  int src_offset = FrameDescription::double_registers_offset();
+  int src_offset = FrameDescription::simd128_registers_offset();
   for (int i = 0; i < DwVfpRegister::kMaxNumRegisters; ++i) {
     if (i == kDoubleRegZero.code()) continue;
     if (i == kScratchDoubleReg.code()) continue;
@@ -350,6 +350,18 @@ void FrameDescription::SetCallerFp(unsigned offset, intptr_t value) {
 }
 
 
+double FrameDescription::GetDoubleRegister(unsigned n) const {
+  ASSERT(n < 2 * ARRAY_SIZE(simd128_registers_));
+  return simd128_registers_[n / 2].d[n % 2];
+}
+
+
+void FrameDescription::SetDoubleRegister(unsigned n, double value) {
+  ASSERT(n < 2 * ARRAY_SIZE(simd128_registers_));
+  simd128_registers_[n / 2].d[n % 2] = value;
+}
+
+
 #undef __
 
 } }  // namespace v8::internal

src/cpu.h

@@ -107,6 +107,8 @@ class CPU V8_FINAL BASE_EMBEDDED {
 
   static bool SupportsCrankshaft();
 
+  static bool SupportsSIMD128InCrankshaft();
+
   // Flush instruction cache.
   static void FlushICache(void* start, size_t size);