SIMD plumbing

runtime/vm/flow_graph_compiler_ia32.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_IA32.
 #if defined(TARGET_ARCH_IA32)
 
 #include "vm/flow_graph_compiler.h"
 
 #include "lib/error.h"
@@ skipping 1318 matching lines @@
   __ Drop(1);
   __ jmp(skip_call);
   __ Bind(&fall_through);
 }
 
 
 void FlowGraphCompiler::SaveLiveRegisters(LocationSummary* locs) {
   // TODO(vegorov): consider saving only caller save (volatile) registers.
   const intptr_t xmm_regs_count = locs->live_registers()->fpu_regs_count();
   if (xmm_regs_count > 0) {
-    __ subl(ESP, Immediate(xmm_regs_count * kDoubleSize));
+    __ subl(ESP, Immediate(xmm_regs_count * kFpuRegisterSize));
     // Store XMM registers with the lowest register number at the lowest
     // address.
     intptr_t offset = 0;
     for (intptr_t reg_idx = 0; reg_idx < kNumberOfXmmRegisters; ++reg_idx) {
       XmmRegister xmm_reg = static_cast<XmmRegister>(reg_idx);
       if (locs->live_registers()->ContainsFpuRegister(xmm_reg)) {
-        __ movsd(Address(ESP, offset), xmm_reg);
-        offset += kDoubleSize;
+        __ movups(Address(ESP, offset), xmm_reg);

[Vyacheslav Egorov (Google), 2013/03/17 18:17:10]

I wonder how much slower does this make runtime calls when there are live
doubles.

[Cutch, 2013/03/17 21:52:37]

Can we add a tag in RegisterSet which indicates if an FPU register contains
64-bit or 128-bit data and issue the optimal move?

+        offset += kFpuRegisterSize;
       }
     }
-    ASSERT(offset == (xmm_regs_count * kDoubleSize));
+    ASSERT(offset == (xmm_regs_count * kFpuRegisterSize));
   }
 
   // Store general purpose registers with the highest register number at the
   // lowest address.
   for (intptr_t reg_idx = 0; reg_idx < kNumberOfCpuRegisters; ++reg_idx) {
     Register reg = static_cast<Register>(reg_idx);
     if (locs->live_registers()->ContainsRegister(reg)) {
       __ pushl(reg);
     }
   }
@@ skipping 10 matching lines @@
     }
   }
 
   const intptr_t xmm_regs_count = locs->live_registers()->fpu_regs_count();
   if (xmm_regs_count > 0) {
     // XMM registers have the lowest register number at the lowest address.
     intptr_t offset = 0;
     for (intptr_t reg_idx = 0; reg_idx < kNumberOfXmmRegisters; ++reg_idx) {
       XmmRegister xmm_reg = static_cast<XmmRegister>(reg_idx);
       if (locs->live_registers()->ContainsFpuRegister(xmm_reg)) {
-        __ movsd(xmm_reg, Address(ESP, offset));
-        offset += kDoubleSize;
+        __ movups(xmm_reg, Address(ESP, offset));
+        offset += kFpuRegisterSize;
       }
     }
-    ASSERT(offset == (xmm_regs_count * kDoubleSize));
+    ASSERT(offset == (xmm_regs_count * kFpuRegisterSize));
     __ addl(ESP, Immediate(offset));
   }
 }
 
 
 void FlowGraphCompiler::EmitTestAndCall(const ICData& ic_data,
                                         Register class_id_reg,
                                         intptr_t arg_count,
                                         const Array& arg_names,
                                         Label* deopt,
@@ skipping 312 matching lines @@
   __ popl(ECX);
   __ popl(EAX);
 }
 
 
 #undef __
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_IA32