[turbofan] Add CalleeSavedFPRegisters to CallDescriptor.

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/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
 #include "src/scopes.h"
@@ skipping 1426 matching lines @@
   __ call(deopt_entry, RelocInfo::RUNTIME_ENTRY);
 }
 
 
 void CodeGenerator::AssemblePrologue() {
   CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
   int stack_slots = frame()->GetSpillSlotCount();
   if (descriptor->kind() == CallDescriptor::kCallAddress) {
     __ pushq(rbp);
     __ movq(rbp, rsp);
+    int register_save_area_size = 0;
     const RegList saves = descriptor->CalleeSavedRegisters();
     if (saves != 0) {  // Save callee-saved registers.
-      int register_save_area_size = 0;
       for (int i = Register::kNumRegisters - 1; i >= 0; i--) {
         if (!((1 << i) & saves)) continue;
         __ pushq(Register::from_code(i));
         register_save_area_size += kPointerSize;
       }
+    }
+    const RegList saves_fp = descriptor->CalleeSavedFPRegisters();
+    if (saves_fp != 0) {  // Save callee-saved XMM registers.
+      const uint32_t saves_fp_count = base::bits::CountPopulation32(saves_fp);
+      const int stack_size = saves_fp_count * 16;
+      // Adjust the stack pointer.
+      __ subp(rsp, Immediate(stack_size));
+      // Store the registers on the stack.
+      int slot_idx = 0;
+      for (int i = XMMRegister::kMaxNumRegisters - 1; i >= 0; i--) {
+        if (!((1 << i) & saves_fp)) continue;
+        __ movdqu(Operand(rsp, 16 * slot_idx), XMMRegister::from_code(i));
+        slot_idx++;
+      }
+      register_save_area_size += stack_size;
+    }
+    if (register_save_area_size > 0) {
       frame()->SetRegisterSaveAreaSize(register_save_area_size);
     }
   } else if (descriptor->IsJSFunctionCall()) {
     CompilationInfo* info = this->info();
     __ Prologue(info->IsCodePreAgingActive());
     frame()->SetRegisterSaveAreaSize(
         StandardFrameConstants::kFixedFrameSizeFromFp);
   } else if (needs_frame_) {
     __ StubPrologue();
     frame()->SetRegisterSaveAreaSize(
@@ skipping 24 matching lines @@
 
 void CodeGenerator::AssembleReturn() {
   CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
   int stack_slots = frame()->GetSpillSlotCount();
   if (descriptor->kind() == CallDescriptor::kCallAddress) {
     if (frame()->GetRegisterSaveAreaSize() > 0) {
       // Remove this frame's spill slots first.
       if (stack_slots > 0) {
         __ addq(rsp, Immediate(stack_slots * kPointerSize));
       }
+      // Restore registers.
+      const RegList saves_fp = descriptor->CalleeSavedFPRegisters();
+      if (saves_fp != 0) {
+        const uint32_t saves_fp_count = base::bits::CountPopulation32(saves_fp);
+        const int stack_size = saves_fp_count * 16;
+        // Load the registers from the stack.
+        int slot_idx = 0;
+        for (int i = 0; i < XMMRegister::kMaxNumRegisters; i++) {
+          if (!((1 << i) & saves_fp)) continue;
+          __ movdqu(Operand(rsp, 16 * slot_idx), XMMRegister::from_code(i));

[paul.l..., 2015/06/23 02:15:02]

Swap params here, per our slack discussion.

[akos.palfi.imgtec, 2015/06/23 19:48:57]

Done.

+          slot_idx++;
+        }
+        // Adjust the stack pointer.
+        __ addp(rsp, Immediate(stack_size));
+      }
       const RegList saves = descriptor->CalleeSavedRegisters();
-      // Restore registers.
       if (saves != 0) {
         for (int i = 0; i < Register::kNumRegisters; i++) {
           if (!((1 << i) & saves)) continue;
           __ popq(Register::from_code(i));
         }
       }
       __ popq(rbp);  // Pop caller's frame pointer.
       __ ret(0);
     } else {
       // No saved registers.
@@ skipping 202 matching lines @@
     }
   }
   MarkLazyDeoptSite();
 }
 
 #undef __
 
 }  // namespace internal
 }  // namespace compiler
 }  // namespace v8