[x64] Implement vmovd and vmovq AVX instructions.

src/x64/lithium-codegen-x64.cc

 // Copyright 2013 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.
 
 #if V8_TARGET_ARCH_X64
 
 #include "src/base/bits.h"
 #include "src/code-factory.h"
 #include "src/code-stubs.h"
 #include "src/hydrogen-osr.h"
@@ skipping 3586 matching lines @@
 
 void LCodeGen::DoMathFloor(LMathFloor* instr) {
   XMMRegister xmm_scratch = double_scratch0();
   Register output_reg = ToRegister(instr->result());
   XMMRegister input_reg = ToDoubleRegister(instr->value());
 
   if (CpuFeatures::IsSupported(SSE4_1)) {
     CpuFeatureScope scope(masm(), SSE4_1);
     if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
       // Deoptimize if minus zero.
-      __ movq(output_reg, input_reg);
+      __ Movq(output_reg, input_reg);
       __ subq(output_reg, Immediate(1));
       DeoptimizeIf(overflow, instr, Deoptimizer::kMinusZero);
     }
     __ roundsd(xmm_scratch, input_reg, kRoundDown);
     __ cvttsd2si(output_reg, xmm_scratch);
     __ cmpl(output_reg, Immediate(0x1));
     DeoptimizeIf(overflow, instr, Deoptimizer::kOverflow);
   } else {
     Label negative_sign, done;
     // Deoptimize on unordered.
@@ skipping 40 matching lines @@
   const XMMRegister xmm_scratch = double_scratch0();
   Register output_reg = ToRegister(instr->result());
   XMMRegister input_reg = ToDoubleRegister(instr->value());
   XMMRegister input_temp = ToDoubleRegister(instr->temp());
   static int64_t one_half = V8_INT64_C(0x3FE0000000000000);  // 0.5
   static int64_t minus_one_half = V8_INT64_C(0xBFE0000000000000);  // -0.5
 
   Label done, round_to_zero, below_one_half;
   Label::Distance dist = DeoptEveryNTimes() ? Label::kFar : Label::kNear;
   __ movq(kScratchRegister, one_half);
-  __ movq(xmm_scratch, kScratchRegister);
+  __ Movq(xmm_scratch, kScratchRegister);
   __ ucomisd(xmm_scratch, input_reg);
   __ j(above, &below_one_half, Label::kNear);
 
   // CVTTSD2SI rounds towards zero, since 0.5 <= x, we use floor(0.5 + x).
   __ addsd(xmm_scratch, input_reg);
   __ cvttsd2si(output_reg, xmm_scratch);
   // Overflow is signalled with minint.
   __ cmpl(output_reg, Immediate(0x1));
   DeoptimizeIf(overflow, instr, Deoptimizer::kOverflow);
   __ jmp(&done, dist);
 
   __ bind(&below_one_half);
   __ movq(kScratchRegister, minus_one_half);
-  __ movq(xmm_scratch, kScratchRegister);
+  __ Movq(xmm_scratch, kScratchRegister);
   __ ucomisd(xmm_scratch, input_reg);
   __ j(below_equal, &round_to_zero, Label::kNear);
 
   // CVTTSD2SI rounds towards zero, we use ceil(x - (-0.5)) and then
   // compare and compensate.
-  __ movq(input_temp, input_reg);  // Do not alter input_reg.
+  __ Movapd(input_temp, input_reg);  // Do not alter input_reg.
   __ subsd(input_temp, xmm_scratch);
   __ cvttsd2si(output_reg, input_temp);
   // Catch minint due to overflow, and to prevent overflow when compensating.
   __ cmpl(output_reg, Immediate(0x1));
   DeoptimizeIf(overflow, instr, Deoptimizer::kOverflow);
 
   __ Cvtlsi2sd(xmm_scratch, output_reg);
   __ ucomisd(xmm_scratch, input_temp);
   __ j(equal, &done, dist);
   __ subl(output_reg, Immediate(1));
   // No overflow because we already ruled out minint.
   __ jmp(&done, dist);
 
   __ bind(&round_to_zero);
   // We return 0 for the input range [+0, 0.5[, or [-0.5, 0.5[ if
   // we can ignore the difference between a result of -0 and +0.
   if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
-    __ movq(output_reg, input_reg);
+    __ Movq(output_reg, input_reg);
     __ testq(output_reg, output_reg);
     DeoptimizeIf(negative, instr, Deoptimizer::kMinusZero);
   }
   __ Set(output_reg, 0);
   __ bind(&done);
 }
 
 
 void LCodeGen::DoMathFround(LMathFround* instr) {
   XMMRegister input_reg = ToDoubleRegister(instr->value());
@@ skipping 20 matching lines @@
   XMMRegister input_reg = ToDoubleRegister(instr->value());
   DCHECK(ToDoubleRegister(instr->result()).is(input_reg));
 
   // Note that according to ECMA-262 15.8.2.13:
   // Math.pow(-Infinity, 0.5) == Infinity
   // Math.sqrt(-Infinity) == NaN
   Label done, sqrt;
   // Check base for -Infinity.  According to IEEE-754, double-precision
   // -Infinity has the highest 12 bits set and the lowest 52 bits cleared.
   __ movq(kScratchRegister, V8_INT64_C(0xFFF0000000000000));
-  __ movq(xmm_scratch, kScratchRegister);
+  __ Movq(xmm_scratch, kScratchRegister);
   __ ucomisd(xmm_scratch, input_reg);
   // Comparing -Infinity with NaN results in "unordered", which sets the
   // zero flag as if both were equal.  However, it also sets the carry flag.
   __ j(not_equal, &sqrt, Label::kNear);
   __ j(carry, &sqrt, Label::kNear);
   // If input is -Infinity, return Infinity.
   __ xorps(input_reg, input_reg);
   __ subsd(input_reg, xmm_scratch);
   __ jmp(&done, Label::kNear);
 
@@ skipping 1557 matching lines @@
   __ ClampUint8(input_reg);
 
   __ bind(&done);
 }
 
 
 void LCodeGen::DoDoubleBits(LDoubleBits* instr) {
   XMMRegister value_reg = ToDoubleRegister(instr->value());
   Register result_reg = ToRegister(instr->result());
   if (instr->hydrogen()->bits() == HDoubleBits::HIGH) {
-    __ movq(result_reg, value_reg);
+    __ Movq(result_reg, value_reg);
     __ shrq(result_reg, Immediate(32));
   } else {
-    __ movd(result_reg, value_reg);
+    __ Movd(result_reg, value_reg);
   }
 }
 
 
 void LCodeGen::DoConstructDouble(LConstructDouble* instr) {
   Register hi_reg = ToRegister(instr->hi());
   Register lo_reg = ToRegister(instr->lo());
   XMMRegister result_reg = ToDoubleRegister(instr->result());
   XMMRegister xmm_scratch = double_scratch0();
-  __ movd(result_reg, hi_reg);
+  __ Movd(result_reg, hi_reg);
   __ psllq(result_reg, 32);
-  __ movd(xmm_scratch, lo_reg);
+  __ Movd(xmm_scratch, lo_reg);
   __ orps(result_reg, xmm_scratch);
 }
 
 
 void LCodeGen::DoAllocate(LAllocate* instr) {
   class DeferredAllocate final : public LDeferredCode {
    public:
     DeferredAllocate(LCodeGen* codegen, LAllocate* instr)
         : LDeferredCode(codegen), instr_(instr) { }
     void Generate() override { codegen()->DoDeferredAllocate(instr_); }
@@ skipping 554 matching lines @@
   RecordSafepoint(Safepoint::kNoLazyDeopt);
 }
 
 
 #undef __
 
 }  // namespace internal
 }  // namespace v8
 
 #endif  // V8_TARGET_ARCH_X64