Stop creating dummy call frames when calling out to C functions

runtime/vm/intermediate_language_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/intermediate_language.h"
 
 #include "vm/dart_entry.h"
@@ skipping 3963 matching lines @@
       break;
     default: UNREACHABLE();
   }
 }
 
 
 LocationSummary* MathUnaryInstr::MakeLocationSummary(bool opt) const {
   if ((kind() == MethodRecognizer::kMathSin) ||
       (kind() == MethodRecognizer::kMathCos)) {
     const intptr_t kNumInputs = 1;
-    const intptr_t kNumTemps = 0;
+    const intptr_t kNumTemps = 1;
     LocationSummary* summary =
         new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kCall);
     summary->set_in(0, Location::FpuRegisterLocation(XMM1));
+    summary->set_temp(0, Location::RegisterLocation(CALLEE_SAVED));

[regis, 2014/02/26 23:53:12]

CALLEE_SAVED -> EDI));  // Callee saved.

[Cutch, 2014/02/27 15:52:37]

Done.

     summary->set_out(Location::FpuRegisterLocation(XMM1));
     return summary;
   }
+  ASSERT(kind() == MethodRecognizer::kMathSqrt);
   const intptr_t kNumInputs = 1;
   const intptr_t kNumTemps = 0;
   LocationSummary* summary =
       new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
   summary->set_in(0, Location::RequiresFpuRegister());
   summary->set_out(Location::RequiresFpuRegister());
   return summary;
 }
 
 
 void MathUnaryInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   if (kind() == MethodRecognizer::kMathSqrt) {
     __ sqrtsd(locs()->out().fpu_reg(), locs()->in(0).fpu_reg());
   } else {
-    __ EnterFrame(0);
+    ASSERT((kind() == MethodRecognizer::kMathSin) ||
+           (kind() == MethodRecognizer::kMathCos));
+    // Save ESP.
+    __ movl(locs()->temp(0).reg(), ESP);
     __ ReserveAlignedFrameSpace(kDoubleSize * InputCount());
     __ movsd(Address(ESP, 0), locs()->in(0).fpu_reg());
     __ CallRuntime(TargetFunction(), InputCount());
     __ fstpl(Address(ESP, 0));
     __ movsd(locs()->out().fpu_reg(), Address(ESP, 0));
-    __ leave();
+    // Restore ESP.
+    __ movl(ESP, locs()->temp(0).reg());
   }
 }
 
 
 LocationSummary* MathMinMaxInstr::MakeLocationSummary(bool opt) const {
   if (result_cid() == kDoubleCid) {
     const intptr_t kNumInputs = 2;
     const intptr_t kNumTemps = 1;
     LocationSummary* summary =
         new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
@@ skipping 269 matching lines @@
 }
 
 
 void FloatToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   __ cvtss2sd(locs()->out().fpu_reg(), locs()->in(0).fpu_reg());
 }
 
 
 LocationSummary* InvokeMathCFunctionInstr::MakeLocationSummary(bool opt) const {
   ASSERT((InputCount() == 1) || (InputCount() == 2));
-  const intptr_t kNumTemps = 0;
+  const intptr_t kNumTemps = 1;
   LocationSummary* result =
       new LocationSummary(InputCount(), kNumTemps, LocationSummary::kCall);
+  result->set_temp(0, Location::RegisterLocation(CALLEE_SAVED));
   result->set_in(0, Location::FpuRegisterLocation(XMM1));
   if (InputCount() == 2) {
     result->set_in(1, Location::FpuRegisterLocation(XMM2));
   }
   if (recognized_kind() == MethodRecognizer::kMathDoublePow) {
     result->AddTemp(Location::RegisterLocation(EAX));
     result->AddTemp(Location::FpuRegisterLocation(XMM4));
   }
   result->set_out(Location::FpuRegisterLocation(XMM3));
   return result;
 }
 
 
 void InvokeMathCFunctionInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
-  __ EnterFrame(0);
+  // Save ESP.
+  __ movl(locs()->temp(0).reg(), ESP);
   __ ReserveAlignedFrameSpace(kDoubleSize * InputCount());
   for (intptr_t i = 0; i < InputCount(); i++) {
     __ movsd(Address(ESP, kDoubleSize * i), locs()->in(i).fpu_reg());
   }
   Label do_call, skip_call;
   if (recognized_kind() == MethodRecognizer::kMathDoublePow) {
     // Pseudo code:
     // if (exponent == 0.0) return 1.0;
     // if (base == 1.0) return 1.0;
     // if (base.isNaN || exponent.isNaN) {
     //    return double.NAN;
     // }
     XmmRegister base = locs()->in(0).fpu_reg();
     XmmRegister exp = locs()->in(1).fpu_reg();
     XmmRegister result = locs()->out().fpu_reg();
     Register temp = locs()->temp(0).reg();

[siva, 2014/02/27 00:00:26]

The index here is wrong, we are essentially using the same temp as the one used
for saving ESP.

[Cutch, 2014/02/27 15:52:37]

Done.

     XmmRegister zero_temp = locs()->temp(1).fpu_reg();
 
     Label check_base_is_one;
     // Check if exponent is 0.0 -> return 1.0;
     __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(0)));
     __ movsd(zero_temp, FieldAddress(temp, Double::value_offset()));
     __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(1)));
     __ movsd(result, FieldAddress(temp, Double::value_offset()));
     // 'result' contains 1.0.
     __ comisd(exp, zero_temp);
@@ skipping 11 matching lines @@
     // Returns NaN.
     __ movsd(result, base);
     __ jmp(&skip_call, Assembler::kNearJump);
     // exp is Nan case is handled correctly in the C-library.
   }
   __ Bind(&do_call);
   __ CallRuntime(TargetFunction(), InputCount());
   __ fstpl(Address(ESP, 0));
   __ movsd(locs()->out().fpu_reg(), Address(ESP, 0));
   __ Bind(&skip_call);
-  __ leave();
+  // Restore ESP.
+  __ movl(ESP, locs()->temp(0).reg());
 }
 
 
 LocationSummary* MergedMathInstr::MakeLocationSummary(bool opt) const {
   if (kind() == MergedMathInstr::kTruncDivMod) {
     const intptr_t kNumInputs = 2;
     const intptr_t kNumTemps = 1;
     LocationSummary* summary =
         new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
     // Both inputs must be writable because they will be untagged.
@@ skipping 1002 matching lines @@
                          PcDescriptors::kOther,
                          locs());
   __ Drop(2);  // Discard type arguments and receiver.
 }
 
 }  // namespace dart
 
 #undef __
 
 #endif  // defined TARGET_ARCH_IA32