<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN"
"http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head><meta http-equiv="content-type" content="text/html; charset=utf-8" />
<title>[184220] trunk/Source/JavaScriptCore</title>
</head>
<body>
<style type="text/css"><!--
#msg dl.meta { border: 1px #006 solid; background: #369; padding: 6px; color: #fff; }
#msg dl.meta dt { float: left; width: 6em; font-weight: bold; }
#msg dt:after { content:':';}
#msg dl, #msg dt, #msg ul, #msg li, #header, #footer, #logmsg { font-family: verdana,arial,helvetica,sans-serif; font-size: 10pt; }
#msg dl a { font-weight: bold}
#msg dl a:link { color:#fc3; }
#msg dl a:active { color:#ff0; }
#msg dl a:visited { color:#cc6; }
h3 { font-family: verdana,arial,helvetica,sans-serif; font-size: 10pt; font-weight: bold; }
#msg pre { overflow: auto; background: #ffc; border: 1px #fa0 solid; padding: 6px; }
#logmsg { background: #ffc; border: 1px #fa0 solid; padding: 1em 1em 0 1em; }
#logmsg p, #logmsg pre, #logmsg blockquote { margin: 0 0 1em 0; }
#logmsg p, #logmsg li, #logmsg dt, #logmsg dd { line-height: 14pt; }
#logmsg h1, #logmsg h2, #logmsg h3, #logmsg h4, #logmsg h5, #logmsg h6 { margin: .5em 0; }
#logmsg h1:first-child, #logmsg h2:first-child, #logmsg h3:first-child, #logmsg h4:first-child, #logmsg h5:first-child, #logmsg h6:first-child { margin-top: 0; }
#logmsg ul, #logmsg ol { padding: 0; list-style-position: inside; margin: 0 0 0 1em; }
#logmsg ul { text-indent: -1em; padding-left: 1em; }#logmsg ol { text-indent: -1.5em; padding-left: 1.5em; }
#logmsg > ul, #logmsg > ol { margin: 0 0 1em 0; }
#logmsg pre { background: #eee; padding: 1em; }
#logmsg blockquote { border: 1px solid #fa0; border-left-width: 10px; padding: 1em 1em 0 1em; background: white;}
#logmsg dl { margin: 0; }
#logmsg dt { font-weight: bold; }
#logmsg dd { margin: 0; padding: 0 0 0.5em 0; }
#logmsg dd:before { content:'\00bb';}
#logmsg table { border-spacing: 0px; border-collapse: collapse; border-top: 4px solid #fa0; border-bottom: 1px solid #fa0; background: #fff; }
#logmsg table th { text-align: left; font-weight: normal; padding: 0.2em 0.5em; border-top: 1px dotted #fa0; }
#logmsg table td { text-align: right; border-top: 1px dotted #fa0; padding: 0.2em 0.5em; }
#logmsg table thead th { text-align: center; border-bottom: 1px solid #fa0; }
#logmsg table th.Corner { text-align: left; }
#logmsg hr { border: none 0; border-top: 2px dashed #fa0; height: 1px; }
#header, #footer { color: #fff; background: #636; border: 1px #300 solid; padding: 6px; }
#patch { width: 100%; }
#patch h4 {font-family: verdana,arial,helvetica,sans-serif;font-size:10pt;padding:8px;background:#369;color:#fff;margin:0;}
#patch .propset h4, #patch .binary h4 {margin:0;}
#patch pre {padding:0;line-height:1.2em;margin:0;}
#patch .diff {width:100%;background:#eee;padding: 0 0 10px 0;overflow:auto;}
#patch .propset .diff, #patch .binary .diff {padding:10px 0;}
#patch span {display:block;padding:0 10px;}
#patch .modfile, #patch .addfile, #patch .delfile, #patch .propset, #patch .binary, #patch .copfile {border:1px solid #ccc;margin:10px 0;}
#patch ins {background:#dfd;text-decoration:none;display:block;padding:0 10px;}
#patch del {background:#fdd;text-decoration:none;display:block;padding:0 10px;}
#patch .lines, .info {color:#888;background:#fff;}
--></style>
<div id="msg">
<dl class="meta">
<dt>Revision</dt> <dd><a href="http://trac.webkit.org/projects/webkit/changeset/184220">184220</a></dd>
<dt>Author</dt> <dd>benjamin@webkit.org</dd>
<dt>Date</dt> <dd>2015-05-12 14:28:43 -0700 (Tue, 12 May 2015)</dd>
</dl>
<h3>Log Message</h3>
<pre>[JSC] Make the NegZero backward propagated flags of ArithMod stricter
https://bugs.webkit.org/show_bug.cgi?id=144897
Patch by Benjamin Poulain <bpoulain@apple.com> on 2015-05-12
Reviewed by Geoffrey Garen.
The NegZero flags of ArithMod were the same as ArithDiv: both children were
marked as needing to handle NegativeZero.
Lucky for us, ArithMod is quite a bit different than ArithDiv.
First, the sign of the result is completely independent from
the sign of the divisor. A zero on the divisor always produces a NaN.
That's great, we can remove the NodeBytecodeNeedsNegZero
from the flags propagated to child2.
Second, the sign of the result is always the same as the sign of
the dividend. A dividend of zero produces a zero of same sign
unless the divisor is zero (in which case the result is NaN).
This is great too: we can just pass the flags we got into
ArithMod.
With those two out of the way, we can make a faster version of ArithRound
for Kraken's oscillator. Since we no longer care about negative zero,
rounding becomes cast<int32>(value + 0.5). This gives ~3% faster runtime
on the benchmark.
Unfortunatelly, most of the time is spent in FTL and the same optimization
does not apply well just yet: rdar://problem/20904149.
* dfg/DFGBackwardsPropagationPhase.cpp:
(JSC::DFG::BackwardsPropagationPhase::propagate):
Never add NodeBytecodeNeedsNegZero unless needed by the users of this node.
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileArithRound):
Faster Math.round() when negative zero is not important.
* tests/stress/arith-modulo-node-behaviors.js: Added.
(moduloWithNegativeZeroDividend):
(moduloWithUnusedNegativeZeroDividend):
(moduloWithNegativeZeroDivisor):</pre>
<h3>Modified Paths</h3>
<ul>
<li><a href="#trunkSourceJavaScriptCoreChangeLog">trunk/Source/JavaScriptCore/ChangeLog</a></li>
<li><a href="#trunkSourceJavaScriptCoredfgDFGBackwardsPropagationPhasecpp">trunk/Source/JavaScriptCore/dfg/DFGBackwardsPropagationPhase.cpp</a></li>
<li><a href="#trunkSourceJavaScriptCoredfgDFGSpeculativeJITcpp">trunk/Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp</a></li>
</ul>
<h3>Added Paths</h3>
<ul>
<li><a href="#trunkSourceJavaScriptCoretestsstressarithmodulonodebehaviorsjs">trunk/Source/JavaScriptCore/tests/stress/arith-modulo-node-behaviors.js</a></li>
</ul>
</div>
<div id="patch">
<h3>Diff</h3>
<a id="trunkSourceJavaScriptCoreChangeLog"></a>
<div class="modfile"><h4>Modified: trunk/Source/JavaScriptCore/ChangeLog (184219 => 184220)</h4>
<pre class="diff"><span>
<span class="info">--- trunk/Source/JavaScriptCore/ChangeLog 2015-05-12 21:24:37 UTC (rev 184219)
+++ trunk/Source/JavaScriptCore/ChangeLog 2015-05-12 21:28:43 UTC (rev 184220)
</span><span class="lines">@@ -1,3 +1,47 @@
</span><ins>+2015-05-12 Benjamin Poulain <bpoulain@apple.com>
+
+ [JSC] Make the NegZero backward propagated flags of ArithMod stricter
+ https://bugs.webkit.org/show_bug.cgi?id=144897
+
+ Reviewed by Geoffrey Garen.
+
+ The NegZero flags of ArithMod were the same as ArithDiv: both children were
+ marked as needing to handle NegativeZero.
+
+ Lucky for us, ArithMod is quite a bit different than ArithDiv.
+
+ First, the sign of the result is completely independent from
+ the sign of the divisor. A zero on the divisor always produces a NaN.
+ That's great, we can remove the NodeBytecodeNeedsNegZero
+ from the flags propagated to child2.
+
+ Second, the sign of the result is always the same as the sign of
+ the dividend. A dividend of zero produces a zero of same sign
+ unless the divisor is zero (in which case the result is NaN).
+ This is great too: we can just pass the flags we got into
+ ArithMod.
+
+ With those two out of the way, we can make a faster version of ArithRound
+ for Kraken's oscillator. Since we no longer care about negative zero,
+ rounding becomes cast<int32>(value + 0.5). This gives ~3% faster runtime
+ on the benchmark.
+
+ Unfortunatelly, most of the time is spent in FTL and the same optimization
+ does not apply well just yet: rdar://problem/20904149.
+
+ * dfg/DFGBackwardsPropagationPhase.cpp:
+ (JSC::DFG::BackwardsPropagationPhase::propagate):
+ Never add NodeBytecodeNeedsNegZero unless needed by the users of this node.
+
+ * dfg/DFGSpeculativeJIT.cpp:
+ (JSC::DFG::SpeculativeJIT::compileArithRound):
+ Faster Math.round() when negative zero is not important.
+
+ * tests/stress/arith-modulo-node-behaviors.js: Added.
+ (moduloWithNegativeZeroDividend):
+ (moduloWithUnusedNegativeZeroDividend):
+ (moduloWithNegativeZeroDivisor):
+
</ins><span class="cx"> 2015-05-12 Mark Lam <mark.lam@apple.com>
</span><span class="cx">
</span><span class="cx"> Refactor MachineStackMarker.cpp so that it's easier to reason about MachineThreads::Thread.
</span></span></pre></div>
<a id="trunkSourceJavaScriptCoredfgDFGBackwardsPropagationPhasecpp"></a>
<div class="modfile"><h4>Modified: trunk/Source/JavaScriptCore/dfg/DFGBackwardsPropagationPhase.cpp (184219 => 184220)</h4>
<pre class="diff"><span>
<span class="info">--- trunk/Source/JavaScriptCore/dfg/DFGBackwardsPropagationPhase.cpp 2015-05-12 21:24:37 UTC (rev 184219)
+++ trunk/Source/JavaScriptCore/dfg/DFGBackwardsPropagationPhase.cpp 2015-05-12 21:28:43 UTC (rev 184220)
</span><span class="lines">@@ -319,11 +319,11 @@
</span><span class="cx"> }
</span><span class="cx">
</span><span class="cx"> case ArithMod: {
</span><del>- flags |= NodeBytecodeUsesAsNumber | NodeBytecodeNeedsNegZero;
</del><ins>+ flags |= NodeBytecodeUsesAsNumber;
</ins><span class="cx"> flags &= ~NodeBytecodeUsesAsOther;
</span><span class="cx">
</span><span class="cx"> node->child1()->mergeFlags(flags);
</span><del>- node->child2()->mergeFlags(flags);
</del><ins>+ node->child2()->mergeFlags(flags & ~NodeBytecodeNeedsNegZero);
</ins><span class="cx"> break;
</span><span class="cx"> }
</span><span class="cx">
</span></span></pre></div>
<a id="trunkSourceJavaScriptCoredfgDFGSpeculativeJITcpp"></a>
<div class="modfile"><h4>Modified: trunk/Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp (184219 => 184220)</h4>
<pre class="diff"><span>
<span class="info">--- trunk/Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp 2015-05-12 21:24:37 UTC (rev 184219)
+++ trunk/Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp 2015-05-12 21:28:43 UTC (rev 184220)
</span><span class="lines">@@ -3550,18 +3550,34 @@
</span><span class="cx">
</span><span class="cx"> SpeculateDoubleOperand value(this, node->child1());
</span><span class="cx"> FPRReg valueFPR = value.fpr();
</span><ins>+
+ if (producesInteger(node->arithRoundingMode()) && !shouldCheckNegativeZero(node->arithRoundingMode())) {
+ FPRTemporary oneHalf(this);
+ GPRTemporary roundedResultAsInt32(this);
+ FPRReg oneHalfFPR = oneHalf.fpr();
+ GPRReg resultGPR = roundedResultAsInt32.gpr();
+
+ static const double halfConstant = 0.5;
+ m_jit.loadDouble(MacroAssembler::TrustedImmPtr(&halfConstant), oneHalfFPR);
+ m_jit.addDouble(valueFPR, oneHalfFPR);
+
+ JITCompiler::Jump truncationFailed = m_jit.branchTruncateDoubleToInt32(oneHalfFPR, resultGPR);
+ speculationCheck(Overflow, JSValueRegs(), node, truncationFailed);
+ int32Result(resultGPR, node);
+ return;
+ }
+
</ins><span class="cx"> flushRegisters();
</span><span class="cx"> FPRResult roundedResultAsDouble(this);
</span><span class="cx"> FPRReg resultFPR = roundedResultAsDouble.fpr();
</span><span class="cx"> callOperation(jsRound, resultFPR, valueFPR);
</span><del>-
</del><span class="cx"> if (producesInteger(node->arithRoundingMode())) {
</span><span class="cx"> GPRTemporary roundedResultAsInt32(this);
</span><span class="cx"> FPRTemporary scratch(this);
</span><span class="cx"> FPRReg scratchFPR = scratch.fpr();
</span><span class="cx"> GPRReg resultGPR = roundedResultAsInt32.gpr();
</span><span class="cx"> JITCompiler::JumpList failureCases;
</span><del>- m_jit.branchConvertDoubleToInt32(resultFPR, resultGPR, failureCases, scratchFPR, shouldCheckNegativeZero(node->arithRoundingMode()));
</del><ins>+ m_jit.branchConvertDoubleToInt32(resultFPR, resultGPR, failureCases, scratchFPR);
</ins><span class="cx"> speculationCheck(Overflow, JSValueRegs(), node, failureCases);
</span><span class="cx">
</span><span class="cx"> int32Result(resultGPR, node);
</span></span></pre></div>
<a id="trunkSourceJavaScriptCoretestsstressarithmodulonodebehaviorsjs"></a>
<div class="addfile"><h4>Added: trunk/Source/JavaScriptCore/tests/stress/arith-modulo-node-behaviors.js (0 => 184220)</h4>
<pre class="diff"><span>
<span class="info">--- trunk/Source/JavaScriptCore/tests/stress/arith-modulo-node-behaviors.js (rev 0)
+++ trunk/Source/JavaScriptCore/tests/stress/arith-modulo-node-behaviors.js 2015-05-12 21:28:43 UTC (rev 184220)
</span><span class="lines">@@ -0,0 +1,104 @@
</span><ins>+// Verify that the dividend propagate the NeedsNegZero if the dividend requires it.
+function moduloWithNegativeZeroDividend(a, b, c)
+{
+ var temp = a * b;
+ return temp % c;
+}
+noInline(moduloWithNegativeZeroDividend);
+
+// Warm up with integers. The test for NegZero should not be eliminated here.
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithNegativeZeroDividend(i, 5, 5);
+ if (result !== 0)
+ throw "moduloWithNegativeZeroDividend(i, 5, 5), returned: " + result;
+}
+
+for (var i = 1; i < 1e7; ++i) {
+ // Produce negative zero in the multiplication.
+ var result = moduloWithNegativeZeroDividend(-i, 0, 2);
+ if (!(result === 0 && (1/result) === -Infinity))
+ throw "moduloWithNegativeZeroDividend(-i, 0, 2) failed, returned: " + result;
+
+ // A negative dividend can produce negative zero results.
+ var result = moduloWithNegativeZeroDividend(-i, 5, 5);
+ if (!(result === 0 && (1/result) === -Infinity))
+ throw "moduloWithNegativeZeroDividend(-i, 5, 5) failed, returned: " + result;
+}
+
+// Edge cases.
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithNegativeZeroDividend(-i, 0, Infinity);
+ if (!(result === 0 && (1/result) === -Infinity))
+ throw "moduloWithNegativeZeroDividend(-i, 0, Infinity) failed, returned: " + result;
+
+ var result = moduloWithNegativeZeroDividend(-i, 0, -Infinity);
+ if (!(result === 0 && (1/result) === -Infinity))
+ throw "moduloWithNegativeZeroDividend(-i, 0, -Infinity) failed, returned: " + result;
+
+ var result = moduloWithNegativeZeroDividend(-i, 0, NaN);
+ if (result === result)
+ throw "moduloWithNegativeZeroDividend(-i, 0, NaN) failed, returned: " + result;
+}
+
+
+// In this case, the negative zero is irrelevant. The Neg Zero check can be eliminated.
+function moduloWithUnusedNegativeZeroDividend(a, b, c)
+{
+ var temp = a * b;
+ return (temp % c) | 0;
+}
+noInline(moduloWithUnusedNegativeZeroDividend);
+
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithUnusedNegativeZeroDividend(i, 5, 5);
+ if (result !== 0)
+ throw "moduloWithUnusedNegativeZeroDividend(i, 5, 5), returned: " + result;
+}
+
+// Produce negative zero in the multiplication.
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithUnusedNegativeZeroDividend(-i, 0, 2);
+ if (!(result === 0 && (1/result) === Infinity))
+ throw "moduloWithUnusedNegativeZeroDividend(-i, 0, 2) failed, returned: " + result;
+}
+
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithUnusedNegativeZeroDividend(-i, 0, Infinity);
+ if (!(result === 0 && (1/result) === Infinity))
+ throw "moduloWithUnusedNegativeZeroDividend(-i, 0, Infinity) failed, returned: " + result;
+
+ var result = moduloWithUnusedNegativeZeroDividend(-i, 0, -Infinity);
+ if (!(result === 0 && (1/result) === Infinity))
+ throw "moduloWithUnusedNegativeZeroDividend(-i, 0, -Infinity) failed, returned: " + result;
+
+ var result = moduloWithUnusedNegativeZeroDividend(-i, 0, NaN);
+ if (result !== 0)
+ throw "moduloWithUnusedNegativeZeroDividend(-i, 0, NaN) failed, returned: " + result;
+}
+
+
+// The sign of the divisor is completely irrelevant. This should never fail on negative zero divisors.
+function moduloWithNegativeZeroDivisor(a, b, c)
+{
+ var temp = a * b;
+ return c % temp;
+}
+noInline(moduloWithNegativeZeroDivisor);
+
+// Warm up with integers.
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithNegativeZeroDivisor(i, 2, i);
+ if (result !== i)
+ throw "moduloWithNegativeZeroDividend(i, 2, i), returned: " + result;
+
+ var result = moduloWithNegativeZeroDivisor(-i, 2, i);
+ if (result !== i)
+ throw "moduloWithNegativeZeroDividend(-i, 2, i), returned: " + result;
+}
+
+// Produce negative zero in the multiplication.
+for (var i = 1; i < 1e7; ++i) {
+ var result = moduloWithNegativeZeroDivisor(-i, 0, 2);
+ if (result === result)
+ throw "moduloWithNegativeZeroDivisor(-i, 0, 2) failed, returned: " + result;
+}
</ins><span class="cx">\ No newline at end of file
</span></span></pre>
</div>
</div>
</body>
</html>