Changeset 93375 in webkit for trunk/Source/JavaScriptCore/dfg/DFGOperations.cpp

Timestamp:

Aug 18, 2011, 5:18:49 PM (14 years ago)

Author:

[email protected]

Message:

DFG non-speculative JIT does not inline the double case of ValueAdd
​https://bugs.webkit.org/show_bug.cgi?id=66025

Reviewed by Gavin Barraclough.

This is a 1.3% win on Kraken overall, with >=8% speed-ups on a few
benchmarks (imaging-darkroom, stanford-crypto-pbkdf2,
stanford-crypto-sha256-iterative). It looks like it might have
a speed-up in SunSpider (though not statistically significant or
particularly reproducible) and a slight slow-down in V8 (0.14%,
not statistically significant). It does slow down v8-crypto by
1.5%.

• dfg/DFGJITCodeGenerator.cpp:

(JSC::DFG::JITCodeGenerator::isKnownInteger):
(JSC::DFG::JITCodeGenerator::isKnownNumeric):

• dfg/DFGNonSpeculativeJIT.cpp:

(JSC::DFG::NonSpeculativeJIT::knownConstantArithOp):
(JSC::DFG::NonSpeculativeJIT::basicArithOp):

• dfg/DFGOperations.cpp:

File:

• trunk/Source/JavaScriptCore/dfg/DFGOperations.cpp (modified) (1 diff)

trunk/Source/JavaScriptCore/dfg/DFGOperations.cpp

@@ -124 +124 @@
     JSValue op1 = JSValue::decode(encodedOp1);
     JSValue op2 = JSValue::decode(encodedOp2);
-
-    if (op1.isInt32() && op2.isInt32()) {
-        int64_t result64 = static_cast<int64_t>(op1.asInt32()) + static_cast<int64_t>(op2.asInt32());
-        int32_t result32 = static_cast<int32_t>(result64);
-        if (LIKELY(result32 == result64))
-            return JSValue::encode(jsNumber(result32));
-        return JSValue::encode(jsNumber((double)result64));
-    }
-    
-    double number1;
-    double number2;
-    if (op1.getNumber(number1) && op2.getNumber(number2))
-        return JSValue::encode(jsNumber(number1 + number2));
+    
+    ASSERT(!op1.isNumber() || !op2.isNumber());
 
     return JSValue::encode(jsAddSlowCase(exec, op1, op2));