[Turbofan] Support 128 bit moves/swaps for x64.

src/compiler/x64/code-generator-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.
 
 #include "src/compiler/code-generator.h"
 
 #include "src/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
@@ skipping 2347 matching lines @@
       }
     }
   } else if (source->IsFPRegister()) {
     XMMRegister src = g.ToDoubleRegister(source);
     if (destination->IsFPRegister()) {
       XMMRegister dst = g.ToDoubleRegister(destination);
       __ Movapd(dst, src);
     } else {
       DCHECK(destination->IsFPStackSlot());
       Operand dst = g.ToOperand(destination);
-      __ Movsd(dst, src);
+      MachineRepresentation rep =
+          LocationOperand::cast(source)->representation();
+      if (rep != MachineRepresentation::kSimd128) {
+        __ Movsd(dst, src);
+      } else {
+        __ Movups(dst, src);
+      }
     }
   } else if (source->IsFPStackSlot()) {
     DCHECK(destination->IsFPRegister() || destination->IsFPStackSlot());
     Operand src = g.ToOperand(source);
+    MachineRepresentation rep = LocationOperand::cast(source)->representation();
     if (destination->IsFPRegister()) {
       XMMRegister dst = g.ToDoubleRegister(destination);
-      __ Movsd(dst, src);
+      if (rep != MachineRepresentation::kSimd128) {
+        __ Movsd(dst, src);
+      } else {
+        __ Movups(dst, src);
+      }
     } else {
       Operand dst = g.ToOperand(destination);
-      __ Movsd(kScratchDoubleReg, src);
-      __ Movsd(dst, kScratchDoubleReg);
+      if (rep != MachineRepresentation::kSimd128) {
+        __ Movsd(kScratchDoubleReg, src);
+        __ Movsd(dst, kScratchDoubleReg);
+      } else {
+        __ Movups(kScratchDoubleReg, src);
+        __ Movups(dst, kScratchDoubleReg);
+      }
     }
   } else {
     UNREACHABLE();
   }
 }
 
 
 void CodeGenerator::AssembleSwap(InstructionOperand* source,
                                  InstructionOperand* destination) {
   X64OperandConverter g(this, nullptr);
@@ skipping 11 matching lines @@
     __ pushq(src);
     frame_access_state()->IncreaseSPDelta(1);
     Operand dst = g.ToOperand(destination);
     __ movq(src, dst);
     frame_access_state()->IncreaseSPDelta(-1);
     dst = g.ToOperand(destination);
     __ popq(dst);
   } else if ((source->IsStackSlot() && destination->IsStackSlot()) ||
              (source->IsFPStackSlot() && destination->IsFPStackSlot())) {
     // Memory-memory.
-    Register tmp = kScratchRegister;
     Operand src = g.ToOperand(source);
     Operand dst = g.ToOperand(destination);
-    __ movq(tmp, dst);
-    __ pushq(src);
-    frame_access_state()->IncreaseSPDelta(1);
-    src = g.ToOperand(source);
-    __ movq(src, tmp);
-    frame_access_state()->IncreaseSPDelta(-1);
-    dst = g.ToOperand(destination);
-    __ popq(dst);
+    MachineRepresentation rep = LocationOperand::cast(source)->representation();
+    if (rep != MachineRepresentation::kSimd128) {
+      Register tmp = kScratchRegister;
+      __ movq(tmp, dst);
+      __ pushq(src);
+      frame_access_state()->IncreaseSPDelta(1);
+      src = g.ToOperand(source);
+      __ movq(src, tmp);
+      frame_access_state()->IncreaseSPDelta(-1);
+      dst = g.ToOperand(destination);
+      __ popq(dst);
+    } else {
+      // Use the XOR trick to swap without a temporary.
+      __ Movups(kScratchDoubleReg, src);
+      __ xorps(kScratchDoubleReg, dst);  // scratch contains src ^ dst.

[Benedikt Meurer, 2016/07/11 03:43:18]

Please do the same for xorps (sorry, missed this one).

[bbudge, 2016/07/11 17:52:44]

Done. I noticed that there are existing references to xorps and vxorps, and that
these are invoked by special kSSE* and kAVX* opcodes. Is this a pattern we would
like to eliminate in a similar way?

+      __ Movups(src, kScratchDoubleReg);
+      __ xorps(kScratchDoubleReg, dst);  // scratch contains src.
+      __ Movups(dst, kScratchDoubleReg);
+      __ xorps(kScratchDoubleReg, src);  // scratch contains dst.
+      __ Movups(src, kScratchDoubleReg);
+    }
   } else if (source->IsFPRegister() && destination->IsFPRegister()) {
     // XMM register-register swap.
     XMMRegister src = g.ToDoubleRegister(source);
     XMMRegister dst = g.ToDoubleRegister(destination);
-    __ Movapd(kScratchDoubleReg, src);
-    __ Movapd(src, dst);
-    __ Movapd(dst, kScratchDoubleReg);
+    MachineRepresentation rep = LocationOperand::cast(source)->representation();
+    if (rep != MachineRepresentation::kSimd128) {
+      __ Movapd(kScratchDoubleReg, src);
+      __ Movapd(src, dst);
+      __ Movapd(dst, kScratchDoubleReg);
+    } else {
+      __ Movups(kScratchDoubleReg, src);
+      __ Movups(src, dst);
+      __ Movups(dst, kScratchDoubleReg);
+    }
   } else if (source->IsFPRegister() && destination->IsFPStackSlot()) {
     // XMM register-memory swap.
     XMMRegister src = g.ToDoubleRegister(source);
     Operand dst = g.ToOperand(destination);
-    __ Movsd(kScratchDoubleReg, src);
-    __ Movsd(src, dst);
-    __ Movsd(dst, kScratchDoubleReg);
+    MachineRepresentation rep = LocationOperand::cast(source)->representation();
+    if (rep != MachineRepresentation::kSimd128) {
+      __ Movsd(kScratchDoubleReg, src);
+      __ Movsd(src, dst);
+      __ Movsd(dst, kScratchDoubleReg);
+    } else {
+      __ Movups(kScratchDoubleReg, src);
+      __ Movups(src, dst);
+      __ Movups(dst, kScratchDoubleReg);
+    }
   } else {
     // No other combinations are possible.
     UNREACHABLE();
   }
 }
 
 
 void CodeGenerator::AssembleJumpTable(Label** targets, size_t target_count) {
   for (size_t index = 0; index < target_count; ++index) {
     __ dq(targets[index]);
@@ skipping 14 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8