Add optimization for pow(base, 0.5) to the other architectures as well.

runtime/vm/intermediate_language_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"  // Needed here to get TARGET_ARCH_X64.
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/intermediate_language.h"
 
 #include "vm/dart_entry.h"
@@ skipping 4650 matching lines @@
 
 
 void InvokeMathCFunctionInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   // Save RSP.
   __ movq(locs()->temp(kSavedSpTempIndex).reg(), RSP);
   __ ReserveAlignedFrameSpace(0);
   __ movaps(XMM0, locs()->in(0).fpu_reg());
   if (InputCount() == 2) {
     ASSERT(locs()->in(1).fpu_reg() == XMM1);
   }
-  // For pow-function return NaN if exponent is NaN.
-  Label do_call, skip_call;
+
+  Label skip_call;
   if (recognized_kind() == MethodRecognizer::kMathDoublePow) {
     // Pseudo code:
-    // if (exponent == 0.0) return 0.0;
+    // if (exponent == 0.0) return 1.0;
     // if (base == 1.0) return 1.0;
     // if (base.isNaN || exponent.isNaN) {
     //    return double.NAN;
     // }
+    // if (base != -Infinity && exponent == 0.5) {
+    //   if (base == 0.0) return 0.0;
+    //   return sqrt(value);
+    // }
     XmmRegister base = locs()->in(0).fpu_reg();
     XmmRegister exp = locs()->in(1).fpu_reg();
     XmmRegister result = locs()->out(0).fpu_reg();
     Register temp = locs()->temp(kObjectTempIndex).reg();
     XmmRegister zero_temp = locs()->temp(kDoubleTempIndex).fpu_reg();
 
-    // Check if exponent is 0.0 -> return 1.0;
+    Label try_sqrt, check_base, return_nan;
     __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(0)), PP);
     __ movsd(zero_temp, FieldAddress(temp, Double::value_offset()));
     __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(1)), PP);
     __ movsd(result, FieldAddress(temp, Double::value_offset()));
-    // 'result' contains 1.0.
-    Label exp_is_nan;
+
+    // exponent == 0.0 -> return 1.0;
     __ comisd(exp, zero_temp);
-    __ j(PARITY_EVEN, &exp_is_nan, Assembler::kNearJump);  // NaN.
-    __ j(EQUAL, &skip_call, Assembler::kNearJump);  // exp is 0, result is 1.0.
+    __ j(PARITY_EVEN, &check_base, Assembler::kNearJump);
+    __ j(EQUAL, &skip_call, Assembler::kNearJump);  // 'result' is 1.0.
 
-    Label base_is_nan;
-    // Checks if base == 1.0.
+    __ Bind(&check_base);
+    // Note: 'exp' could be NaN.
+
+    // base == 1.0 -> return 1.0;
     __ comisd(base, result);
-    __ j(PARITY_EVEN, &base_is_nan, Assembler::kNearJump);
-    __ j(EQUAL, &skip_call, Assembler::kNearJump);  // base and result are 1.0
-    __ jmp(&do_call, Assembler::kNearJump);
+    __ j(PARITY_EVEN, &return_nan, Assembler::kNearJump);
+    __ j(EQUAL, &skip_call, Assembler::kNearJump);
+    // Note: 'base' could be NaN.
+    __ comisd(exp, base);
+    // Neither 'exp' nor 'base' is NaN.
+    __ j(PARITY_ODD, &try_sqrt, Assembler::kNearJump);
+    // Return NaN.
+    __ Bind(&return_nan);
+    __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(NAN)), PP);
+    __ movsd(result, FieldAddress(temp, Double::value_offset()));
+    __ jmp(&skip_call);
 
-    __ Bind(&base_is_nan);
-    // Returns NaN.
-    __ movsd(result, base);
+    Label do_pow, return_zero;
+    __ Bind(&try_sqrt);
+    // Before calling pow, check if we could use sqrt instead of pow.
+    __ LoadObject(temp,
+        Double::ZoneHandle(Double::NewCanonical(-INFINITY)), PP);
+    __ movsd(result, FieldAddress(temp, Double::value_offset()));
+    // base == -Infinity -> call pow;
+    __ comisd(base, result);
+    __ j(EQUAL, &do_pow, Assembler::kNearJump);
+
+    // exponent == 0.5 ?
+    __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(0.5)), PP);
+    __ movsd(result, FieldAddress(temp, Double::value_offset()));
+    __ comisd(exp, result);
+    __ j(NOT_EQUAL, &do_pow, Assembler::kNearJump);
+
+    // base == 0 -> return 0;
+    __ comisd(base, zero_temp);
+    __ j(EQUAL, &return_zero, Assembler::kNearJump);
+
+    __ sqrtsd(result, base);
     __ jmp(&skip_call, Assembler::kNearJump);
 
-    __ Bind(&exp_is_nan);
-    // Checks if base == 1.0.
-    __ comisd(base, result);
-    __ j(PARITY_EVEN, &base_is_nan, Assembler::kNearJump);
-    __ j(EQUAL, &skip_call, Assembler::kNearJump);  // base and result are 1.0
-    __ movsd(result, exp);  // result is NaN
-    __ jmp(&skip_call, Assembler::kNearJump);
+    __ Bind(&return_zero);
+    __ movsd(result, zero_temp);
+    __ jmp(&skip_call);
+
+    __ Bind(&do_pow);
   }
-  __ Bind(&do_call);
   __ CallRuntime(TargetFunction(), InputCount());
   __ movaps(locs()->out(0).fpu_reg(), XMM0);
   __ Bind(&skip_call);
   // Restore RSP.
   __ movq(RSP, locs()->temp(kSavedSpTempIndex).reg());
 }
 
 
 LocationSummary* MergedMathInstr::MakeLocationSummary(bool opt) const {
   if (kind() == MergedMathInstr::kTruncDivMod) {
@@ skipping 672 matching lines @@
                          PcDescriptors::kOther,
                          locs());
   __ Drop(ArgumentCount());  // Discard arguments.
 }
 
 }  // namespace dart
 
 #undef __
 
 #endif  // defined TARGET_ARCH_X64