Revert SSE assembler changes

runtime/vm/assembler_x64.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"
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/assembler.h"
 #include "vm/heap.h"
 #include "vm/memory_region.h"
@@ skipping 511 matching lines @@
   ASSERT(src <= XMM15);
   ASSERT(dst <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitREX_RB(dst, src);
   EmitUint8(0x0F);
   EmitUint8(0x28);
   EmitXmmRegisterOperand(dst & 7, src);
 }
 
 
-void Assembler::movups(XmmRegister dst, const Address& src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x10);
-  EmitOperand(dst & 7, src);
-}
-
-
-void Assembler::movups(const Address& dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(src, dst);
-  EmitUint8(0x0F);
-  EmitUint8(0x11);
-  EmitOperand(src & 7, dst);
-}
-
-
 void Assembler::addsd(XmmRegister dst, XmmRegister src) {
   ASSERT(src <= XMM15);
   ASSERT(dst <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0xF2);
   EmitREX_RB(dst, src);
   EmitUint8(0x0F);
   EmitUint8(0x58);
   EmitXmmRegisterOperand(dst & 7, src);
 }
@@ skipping 27 matching lines @@
   ASSERT(src <= XMM15);
   ASSERT(dst <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0xF2);
   EmitREX_RB(dst, src);
   EmitUint8(0x0F);
   EmitUint8(0x5E);
   EmitXmmRegisterOperand(dst & 7, src);
 }
 
-void Assembler::addps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x58);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::subps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x5C);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::divps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x5E);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::mulps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x59);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::minps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x5D);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::maxps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x5F);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::andps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x54);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
-void Assembler::andps(XmmRegister dst, const Address& src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x54);
-  EmitOperand(dst & 7, src);
-}
-
-
-void Assembler::orps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x56);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-void Assembler::notps(XmmRegister dst) {
-  static const struct ALIGN16 {
-    uint32_t a;
-    uint32_t b;
-    uint32_t c;
-    uint32_t d;
-  } float_not_constant =
-      { 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
-  movq(TMP, Immediate(reinterpret_cast<intptr_t>(&float_not_constant)));
-  xorps(dst, Address(TMP, 0));
-}
-
-
-void Assembler::negateps(XmmRegister dst) {
-  static const struct ALIGN16 {
-    uint32_t a;
-    uint32_t b;
-    uint32_t c;
-    uint32_t d;
-  } float_negate_constant =
-      { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
-  movq(TMP, Immediate(reinterpret_cast<intptr_t>(&float_negate_constant)));
-  xorps(dst, Address(TMP, 0));
-}
-
-
-void Assembler::absps(XmmRegister dst) {
-  static const struct ALIGN16 {
-    uint32_t a;
-    uint32_t b;
-    uint32_t c;
-    uint32_t d;
-  } float_absolute_constant =
-      { 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF };
-  movq(TMP, Immediate(reinterpret_cast<intptr_t>(&float_absolute_constant)));
-  andps(dst, Address(TMP, 0));
-}
-
-
-void Assembler::zerowps(XmmRegister dst) {
-  static const struct ALIGN16 {
-    uint32_t a;
-    uint32_t b;
-    uint32_t c;
-    uint32_t d;
-  } float_zerow_constant =
-      { 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000 };
-  movq(TMP, Immediate(reinterpret_cast<intptr_t>(&float_zerow_constant)));
-  andps(dst, Address(TMP, 0));
-}
-
-
-void Assembler::cmppseq(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC2);
-  EmitXmmRegisterOperand(dst & 7, src);
-  EmitUint8(0x0);
-}
-
-
-void Assembler::cmppsneq(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC2);
-  EmitXmmRegisterOperand(dst & 7, src);
-  EmitUint8(0x4);
-}
-
-
-void Assembler::cmppslt(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC2);
-  EmitXmmRegisterOperand(dst & 7, src);
-  EmitUint8(0x1);
-}
-
-
-void Assembler::cmppsle(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC2);
-  EmitXmmRegisterOperand(dst & 7, src);
-  EmitUint8(0x2);
-}
-
-
-void Assembler::cmppsnlt(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC2);
-  EmitXmmRegisterOperand(dst & 7, src);
-  EmitUint8(0x5);
-}
-
-
-void Assembler::cmppsnle(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC2);
-  EmitXmmRegisterOperand(dst & 7, src);
-  EmitUint8(0x6);
-}
-
-
-void Assembler::sqrtps(XmmRegister dst) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, dst);
-  EmitUint8(0x0F);
-  EmitUint8(0x51);
-  EmitXmmRegisterOperand(dst & 7, dst);
-}
-
-
-void Assembler::rsqrtps(XmmRegister dst) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, dst);
-  EmitUint8(0x0F);
-  EmitUint8(0x52);
-  EmitXmmRegisterOperand(dst & 7, dst);
-}
-
-
-void Assembler::reciprocalps(XmmRegister dst) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, dst);
-  EmitUint8(0x0F);
-  EmitUint8(0x53);
-  EmitXmmRegisterOperand(dst & 7, dst);
-}
-
-
-void Assembler::set1ps(XmmRegister dst, Register tmp1, const Immediate& imm) {
-  // Load 32-bit immediate value into tmp1.
-  movl(tmp1, imm);
-  // Move value from tmp1 into dst.
-  movd(dst, tmp1);
-  // Broadcast low lane into other three lanes.
-  shufps(dst, dst, Immediate(0x0));
-}
-
-
-void Assembler::shufps(XmmRegister dst, XmmRegister src, const Immediate& imm) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0xC6);
-  EmitXmmRegisterOperand(dst & 7, src);
-  ASSERT(imm.is_uint8());
-  EmitUint8(imm.value());
-}
-
 
 void Assembler::comisd(XmmRegister a, XmmRegister b) {
   ASSERT(a <= XMM15);
   ASSERT(b <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitREX_RB(a, b);
   EmitUint8(0x0F);
   EmitUint8(0x2F);
   EmitXmmRegisterOperand(a & 7, b);
 }
 
 
 void Assembler::movmskpd(Register dst, XmmRegister src) {
   ASSERT(src <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitREX_RB(dst, src);
   EmitUint8(0x0F);
   EmitUint8(0x50);
   EmitXmmRegisterOperand(dst & 7, src);
 }
 
 
-void Assembler::movmskps(Register dst, XmmRegister src) {
-  ASSERT(src <= XMM15);
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x50);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
-
 void Assembler::sqrtsd(XmmRegister dst, XmmRegister src) {
   ASSERT(dst <= XMM15);
   ASSERT(src <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0xF2);
   EmitREX_RB(dst, src);
   EmitUint8(0x0F);
   EmitUint8(0x51);
   EmitXmmRegisterOperand(dst & 7, src);
 }
@@ skipping 15 matching lines @@
   ASSERT(src <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitREX_RB(dst, src);
   EmitUint8(0x0F);
   EmitUint8(0x57);
   EmitXmmRegisterOperand(dst & 7, src);
 }
 
 
-void Assembler::xorps(XmmRegister dst, const Address& src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x57);
-  EmitOperand(dst & 7, src);
-}
-
-
-void Assembler::xorps(XmmRegister dst, XmmRegister src) {
-  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  EmitREX_RB(dst, src);
-  EmitUint8(0x0F);
-  EmitUint8(0x57);
-  EmitXmmRegisterOperand(dst & 7, src);
-}
-
 void Assembler::andpd(XmmRegister dst, const Address& src) {
   ASSERT(dst <= XMM15);
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitOperandREX(dst, src, REX_NONE);
   EmitUint8(0x0F);
   EmitUint8(0x54);
   EmitOperand(dst & 7, src);
 }
 
@@ skipping 1268 matching lines @@
 
 void Assembler::EnterCallRuntimeFrame(intptr_t frame_space) {
   EnterFrame(0);
 
   // Preserve volatile CPU registers.
   for (intptr_t i = 0; i < kNumberOfVolatileCpuRegisters; i++) {
     pushq(volatile_cpu_registers[i]);
   }
 
   // Preserve all XMM registers except XMM0
-  subq(RSP, Immediate((kNumberOfXmmRegisters - 1) * kFpuRegisterSize));
+  subq(RSP, Immediate((kNumberOfXmmRegisters - 1) * kDoubleSize));
   // Store XMM registers with the lowest register number at the lowest
   // address.
   intptr_t offset = 0;
   for (intptr_t reg_idx = 1; reg_idx < kNumberOfXmmRegisters; ++reg_idx) {
     XmmRegister xmm_reg = static_cast<XmmRegister>(reg_idx);
-    movups(Address(RSP, offset), xmm_reg);
-    offset += kFpuRegisterSize;
+    movsd(Address(RSP, offset), xmm_reg);
+    offset += kDoubleSize;
   }
 
   ReserveAlignedFrameSpace(frame_space);
 }
 
 
 void Assembler::LeaveCallRuntimeFrame() {
   // RSP might have been modified to reserve space for arguments
   // and ensure proper alignment of the stack frame.
   // We need to restore it before restoring registers.
   const intptr_t kPushedRegistersSize =
       kNumberOfVolatileCpuRegisters * kWordSize +
-      kNumberOfVolatileXmmRegisters * kFpuRegisterSize;
+      kNumberOfVolatileXmmRegisters * kDoubleSize;
   leaq(RSP, Address(RBP, -kPushedRegistersSize));
 
   // Restore all XMM registers except XMM0
   // XMM registers have the lowest register number at the lowest address.
   intptr_t offset = 0;
   for (intptr_t reg_idx = 1; reg_idx < kNumberOfXmmRegisters; ++reg_idx) {
     XmmRegister xmm_reg = static_cast<XmmRegister>(reg_idx);
-    movups(xmm_reg, Address(RSP, offset));
-    offset += kFpuRegisterSize;
+    movsd(xmm_reg, Address(RSP, offset));
+    offset += kDoubleSize;
   }
   addq(RSP, Immediate(offset));
 
   // Restore volatile CPU registers.
   for (intptr_t i = kNumberOfVolatileCpuRegisters - 1; i >= 0; i--) {
     popq(volatile_cpu_registers[i]);
   }
 
   leave();
 }
@@ skipping 191 matching lines @@
 
 const char* Assembler::FpuRegisterName(FpuRegister reg) {
   ASSERT((0 <= reg) && (reg < kNumberOfXmmRegisters));
   return xmm_reg_names[reg];
 }
 
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_X64