Modified: trunk/Source/_javascript_Core/ChangeLog (195727 => 195728)
--- trunk/Source/_javascript_Core/ChangeLog 2016-01-28 04:42:36 UTC (rev 195727)
+++ trunk/Source/_javascript_Core/ChangeLog 2016-01-28 05:21:57 UTC (rev 195728)
@@ -1,3 +1,24 @@
+2016-01-27 Filip Pizlo <[email protected]>
+
+ B3 IntRange analysis should know more about shifting
+ https://bugs.webkit.org/show_bug.cgi?id=153568
+
+ Reviewed by Benjamin Poulain.
+
+ This teaches the IntRange analysis that the result of a right shift is usually better than
+ the worst-case mask based on the shift amount. In fact, you can reach useful conclusions
+ from looking at the IntRange of the input. This helps because Octane/crypto does something
+ like:
+
+ CheckMul((@x & $268435455) >> 14, @y >> 14, ...)
+
+ If you consider just the shifts, then this may overflow. But if you consider that @x is
+ first masked, then the IntRange coming out of the first shift is tight enough to prove that
+ the CheckMul cannot overflow.
+
+ * b3/B3ReduceStrength.cpp:
+ * b3/testb3.cpp:
+
2016-01-27 Benjamin Poulain <[email protected]>
[JSC] adjustFrameAndStackInOSRExitCompilerThunk() can trash values in FTL
Modified: trunk/Source/_javascript_Core/b3/B3ReduceStrength.cpp (195727 => 195728)
--- trunk/Source/_javascript_Core/b3/B3ReduceStrength.cpp 2016-01-28 04:42:36 UTC (rev 195727)
+++ trunk/Source/_javascript_Core/b3/B3ReduceStrength.cpp 2016-01-28 05:21:57 UTC (rev 195728)
@@ -88,6 +88,8 @@
bool verbose = false;
+// FIXME: This IntRange stuff should be refactored into a general constant propagator. It's weird
+// that it's just sitting here in this file.
class IntRange {
public:
IntRange()
@@ -162,25 +164,6 @@
}
}
- template<typename T>
- static IntRange rangeForSShr(int32_t shiftAmount)
- {
- return IntRange(top<T>().min() >> shiftAmount, top<T>().max() >> shiftAmount);
- }
-
- static IntRange rangeForSShr(int32_t shiftAmount, Type type)
- {
- switch (type) {
- case Int32:
- return rangeForSShr<int32_t>(shiftAmount);
- case Int64:
- return rangeForSShr<int64_t>(shiftAmount);
- default:
- RELEASE_ASSERT_NOT_REACHED();
- return IntRange();
- }
- }
-
int64_t min() const { return m_min; }
int64_t max() const { return m_max; }
@@ -255,8 +238,8 @@
template<typename T>
IntRange shl(int32_t shiftAmount)
{
- T newMin = static_cast<T>(m_min) << shiftAmount;
- T newMax = static_cast<T>(m_max) << shiftAmount;
+ T newMin = static_cast<T>(m_min) << static_cast<T>(shiftAmount);
+ T newMax = static_cast<T>(m_max) << static_cast<T>(shiftAmount);
if ((newMin >> shiftAmount) != static_cast<T>(m_min))
newMin = std::numeric_limits<T>::min();
@@ -280,6 +263,61 @@
}
template<typename T>
+ IntRange sShr(int32_t shiftAmount)
+ {
+ T newMin = static_cast<T>(m_min) >> static_cast<T>(shiftAmount);
+ T newMax = static_cast<T>(m_max) >> static_cast<T>(shiftAmount);
+
+ return IntRange(newMin, newMax);
+ }
+
+ IntRange sShr(int32_t shiftAmount, Type type)
+ {
+ switch (type) {
+ case Int32:
+ return sShr<int32_t>(shiftAmount);
+ case Int64:
+ return sShr<int64_t>(shiftAmount);
+ default:
+ RELEASE_ASSERT_NOT_REACHED();
+ return IntRange();
+ }
+ }
+
+ template<typename T>
+ IntRange zShr(int32_t shiftAmount)
+ {
+ // This is an awkward corner case for all of the other logic.
+ if (!shiftAmount)
+ return *this;
+
+ // If the input range may be negative, then all we can say about the output range is that it
+ // will be masked. That's because -1 right shifted just produces that mask.
+ if (m_min < 0)
+ return rangeForZShr<T>(shiftAmount);
+
+ // If the input range is non-negative, then this just brings the range closer to zero.
+ typedef typename std::make_unsigned<T>::type UnsignedT;
+ UnsignedT newMin = static_cast<UnsignedT>(m_min) >> static_cast<UnsignedT>(shiftAmount);
+ UnsignedT newMax = static_cast<UnsignedT>(m_max) >> static_cast<UnsignedT>(shiftAmount);
+
+ return IntRange(newMin, newMax);
+ }
+
+ IntRange zShr(int32_t shiftAmount, Type type)
+ {
+ switch (type) {
+ case Int32:
+ return zShr<int32_t>(shiftAmount);
+ case Int64:
+ return zShr<int64_t>(shiftAmount);
+ default:
+ RELEASE_ASSERT_NOT_REACHED();
+ return IntRange();
+ }
+ }
+
+ template<typename T>
IntRange add(const IntRange& other)
{
if (couldOverflowAdd<T>(other))
@@ -1955,19 +1993,24 @@
break;
case SShr:
- if (value->child(1)->hasInt32())
- return IntRange::rangeForSShr(value->child(1)->asInt32(), value->type());
+ if (value->child(1)->hasInt32()) {
+ return rangeFor(value->child(0), timeToLive - 1).sShr(
+ value->child(1)->asInt32(), value->type());
+ }
break;
case ZShr:
- if (value->child(1)->hasInt32())
- return IntRange::rangeForZShr(value->child(1)->asInt32(), value->type());
+ if (value->child(1)->hasInt32()) {
+ return rangeFor(value->child(0), timeToLive - 1).zShr(
+ value->child(1)->asInt32(), value->type());
+ }
break;
case Shl:
- if (value->child(1)->hasInt32())
+ if (value->child(1)->hasInt32()) {
return rangeFor(value->child(0), timeToLive - 1).shl(
value->child(1)->asInt32(), value->type());
+ }
break;
case Add:
Modified: trunk/Source/_javascript_Core/b3/testb3.cpp (195727 => 195728)
--- trunk/Source/_javascript_Core/b3/testb3.cpp 2016-01-28 04:42:36 UTC (rev 195727)
+++ trunk/Source/_javascript_Core/b3/testb3.cpp 2016-01-28 05:21:57 UTC (rev 195728)
@@ -7908,6 +7908,45 @@
CHECK(invoke<int>(*code) == 42);
}
+void testCheckMul64SShr()
+{
+ Procedure proc;
+ BasicBlock* root = proc.addBlock();
+ Value* arg1 = root->appendNew<Value>(
+ proc, SShr, Origin(),
+ root->appendNew<ArgumentRegValue>(proc, Origin(), GPRInfo::argumentGPR0),
+ root->appendNew<Const32Value>(proc, Origin(), 1));
+ Value* arg2 = root->appendNew<Value>(
+ proc, SShr, Origin(),
+ root->appendNew<ArgumentRegValue>(proc, Origin(), GPRInfo::argumentGPR1),
+ root->appendNew<Const32Value>(proc, Origin(), 1));
+ CheckValue* checkMul = root->appendNew<CheckValue>(proc, CheckMul, Origin(), arg1, arg2);
+ checkMul->append(arg1);
+ checkMul->append(arg2);
+ checkMul->setGenerator(
+ [&] (CCallHelpers& jit, const StackmapGenerationParams& params) {
+ AllowMacroScratchRegisterUsage allowScratch(jit);
+ CHECK(params.size() == 2);
+ CHECK(params[0].isGPR());
+ CHECK(params[1].isGPR());
+ jit.convertInt64ToDouble(params[0].gpr(), FPRInfo::fpRegT0);
+ jit.convertInt64ToDouble(params[1].gpr(), FPRInfo::fpRegT1);
+ jit.mulDouble(FPRInfo::fpRegT1, FPRInfo::fpRegT0);
+ jit.emitFunctionEpilogue();
+ jit.ret();
+ });
+ root->appendNew<ControlValue>(
+ proc, Return, Origin(),
+ root->appendNew<Value>(proc, IToD, Origin(), checkMul));
+
+ auto code = compile(proc);
+
+ CHECK(invoke<double>(*code, 0ll, 42ll) == 0.0);
+ CHECK(invoke<double>(*code, 1ll, 42ll) == 0.0);
+ CHECK(invoke<double>(*code, 42ll, 42ll) == (42.0 / 2.0) * (42.0 / 2.0));
+ CHECK(invoke<double>(*code, 10000000000ll, 10000000000ll) == 25000000000000000000.0);
+}
+
template<typename LeftFunctor, typename RightFunctor, typename InputType>
void genericTestCompare(
B3::Opcode opcode, const LeftFunctor& leftFunctor, const RightFunctor& rightFunctor,
@@ -11340,6 +11379,8 @@
RUN(testSShrShl64(42000000000, 8, 8));
RUN(testSShrShl64(-42000000000, 8, 8));
+ RUN(testCheckMul64SShr());
+
if (tasks.isEmpty())
usage();
