Changeset 222380 in webkit for trunk/Source/JavaScriptCore/ChangeLog

Timestamp:

Sep 22, 2017, 1:22:44 AM (8 years ago)

Author:

Yusuke Suzuki

Message:

[DFG][FTL] Profile array vector length for array allocation
​https://bugs.webkit.org/show_bug.cgi?id=177051

Reviewed by Saam Barati.

JSTests:

• microbenchmarks/new-array-buffer-vector-profile.js: Added.

(target):

Source/JavaScriptCore:

Currently, NewArrayBuffer allocation is penalized by JSC: While empty array gets 25 vector size (BASE_CONTIGUOUS_VECTOR_LEN),
new_array_buffer case gets 3 vector size (BASE_CONTIGUOUS_VECTOR_LEN). Surely, new_array_buffer can get larger vector size
if the number of its constant elements is larger than 3. But these created array may be grown by push() operation after
the allocation. In this case, new_array_buffer is penalized compared to empty array allocation.

empty array allocation,

var array = [];
array.push(0);
array.push(1);
array.push(2);
array.push(3);
array.push(4);

v.s. new_array_buffer case,

var array = [0];
array.push(1);
array.push(2);
array.push(3);
array.push(4);

In this case, the latter becomes slow. While we have a chance to reduce memory usage if new_array_buffer is not grown (and a bit likely),
we should allocate 3 to 25 vector size if it is likely grown. So we should get profile on the resulted array.

We select 25 to make it fit to one of size classes.

In this patch, we extend ArrayAllocationProfile to record vector length. And use this information when allocating array for new_array_buffer.
If the number of new_array_buffer constants is <= 25, array vector size would become 3 to 25 based on profiling. If the number of its constants
is larger than 25, we just use it for allocation as before.

Added microbenchmark and SixSpeed spread-literal.es5 shows improvement.

new-array-buffer-vector-profile 67.4706+-3.7625 ^{28.4249+-1.9025} definitely 2.3736x faster
spread-literal.es5 133.1443+-9.2253 ^{95.2667+-0.5740} definitely 1.3976x faster

• bytecode/ArrayAllocationProfile.cpp:

(JSC::ArrayAllocationProfile::updateProfile):
(JSC::ArrayAllocationProfile::updateIndexingType): Deleted.

• bytecode/ArrayAllocationProfile.h:

(JSC::ArrayAllocationProfile::selectIndexingType):
(JSC::ArrayAllocationProfile::vectorLengthHint):
(JSC::ArrayAllocationProfile::ArrayAllocationProfile): Deleted.

• bytecode/CodeBlock.cpp:

(JSC::CodeBlock::updateAllArrayPredictions):

• dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::parseBlock):

• dfg/DFGGraph.cpp:

(JSC::DFG::Graph::dump):

• dfg/DFGNode.h:

(JSC::DFG::Node::vectorLengthHint):

• dfg/DFGOperations.cpp:
• dfg/DFGOperations.h:
• dfg/DFGSpeculativeJIT64.cpp:

(JSC::DFG::SpeculativeJIT::compile):

• ftl/FTLLowerDFGToB3.cpp:

(JSC::FTL::DFG::LowerDFGToB3::compileArraySlice):
(JSC::FTL::DFG::LowerDFGToB3::compileNewArrayBuffer):
(JSC::FTL::DFG::LowerDFGToB3::compileNewArrayWithSize):
(JSC::FTL::DFG::LowerDFGToB3::allocateJSArray):
(JSC::FTL::DFG::LowerDFGToB3::allocateUninitializedContiguousJSArrayInternal):
(JSC::FTL::DFG::LowerDFGToB3::allocateUninitializedContiguousJSArray):

• runtime/ArrayConventions.h:
• runtime/JSArray.h:

(JSC::JSArray::tryCreate):

File:

• trunk/Source/JavaScriptCore/ChangeLog (modified) (1 diff)

trunk/Source/JavaScriptCore/ChangeLog

@@ +1 @@
+2017-09-21  Yusuke Suzuki  <[email protected]>
+
+        [DFG][FTL] Profile array vector length for array allocation
+        https://bugs.webkit.org/show_bug.cgi?id=177051
+
+        Reviewed by Saam Barati.
+
+        Currently, NewArrayBuffer allocation is penalized by JSC: While empty array gets 25 vector size (BASE_CONTIGUOUS_VECTOR_LEN),
+        new_array_buffer case gets 3 vector size (BASE_CONTIGUOUS_VECTOR_LEN). Surely, new_array_buffer can get larger vector size
+        if the number of its constant elements is larger than 3. But these created array may be grown by `push()` operation after
+        the allocation. In this case, new_array_buffer is penalized compared to empty array allocation.
+
+            empty array allocation,
+
+            var array = [];
+            array.push(0);
+            array.push(1);
+            array.push(2);
+            array.push(3);
+            array.push(4);
+
+            v.s. new_array_buffer case,
+
+            var array = [0];
+            array.push(1);
+            array.push(2);
+            array.push(3);
+            array.push(4);
+
+        In this case, the latter becomes slow. While we have a chance to reduce memory usage if new_array_buffer is not grown (and a bit likely),
+        we should allocate 3 to 25 vector size if it is likely grown. So we should get profile on the resulted array.
+
+        We select 25 to make it fit to one of size classes.
+
+        In this patch, we extend ArrayAllocationProfile to record vector length. And use this information when allocating array for new_array_buffer.
+        If the number of new_array_buffer constants is <= 25, array vector size would become 3 to 25 based on profiling. If the number of its constants
+        is larger than 25, we just use it for allocation as before.
+
+        Added microbenchmark and SixSpeed spread-literal.es5 shows improvement.
+
+            new-array-buffer-vector-profile       67.4706+-3.7625     ^     28.4249+-1.9025        ^ definitely 2.3736x faster
+            spread-literal.es5                   133.1443+-9.2253     ^     95.2667+-0.5740        ^ definitely 1.3976x faster
+
+        * bytecode/ArrayAllocationProfile.cpp:
+        (JSC::ArrayAllocationProfile::updateProfile):
+        (JSC::ArrayAllocationProfile::updateIndexingType): Deleted.
+        * bytecode/ArrayAllocationProfile.h:
+        (JSC::ArrayAllocationProfile::selectIndexingType):
+        (JSC::ArrayAllocationProfile::vectorLengthHint):
+        (JSC::ArrayAllocationProfile::ArrayAllocationProfile): Deleted.
+        * bytecode/CodeBlock.cpp:
+        (JSC::CodeBlock::updateAllArrayPredictions):
+        * dfg/DFGByteCodeParser.cpp:
+        (JSC::DFG::ByteCodeParser::parseBlock):
+        * dfg/DFGGraph.cpp:
+        (JSC::DFG::Graph::dump):
+        * dfg/DFGNode.h:
+        (JSC::DFG::Node::vectorLengthHint):
+        * dfg/DFGOperations.cpp:
+        * dfg/DFGOperations.h:
+        * dfg/DFGSpeculativeJIT64.cpp:
+        (JSC::DFG::SpeculativeJIT::compile):
+        * ftl/FTLLowerDFGToB3.cpp:
+        (JSC::FTL::DFG::LowerDFGToB3::compileArraySlice):
+        (JSC::FTL::DFG::LowerDFGToB3::compileNewArrayBuffer):
+        (JSC::FTL::DFG::LowerDFGToB3::compileNewArrayWithSize):
+        (JSC::FTL::DFG::LowerDFGToB3::allocateJSArray):
+        (JSC::FTL::DFG::LowerDFGToB3::allocateUninitializedContiguousJSArrayInternal):
+        (JSC::FTL::DFG::LowerDFGToB3::allocateUninitializedContiguousJSArray):
+        * runtime/ArrayConventions.h:
+        * runtime/JSArray.h:
+        (JSC::JSArray::tryCreate):
+
 2017-09-21  Joseph Pecoraro  <[email protected]>