Wed, 06 Feb 2013 08:42:19 -0400
8007545: jjs input evalinput need to be NOT_ENUMERABLE
Reviewed-by: sundar, lagergren
Contributed-by: james.laskey@oracle.com
1 This document describes system properties that are used for internal
2 debugging and instrumentation purposes, along with the system loggers,
3 which are used for the same thing.
5 This document is intended as a developer resource, and it is not
6 needed as Nashorn documentation for normal usage. Flags and system
7 properties described herein are subject to change without notice.
9 =====================================
10 1. System properties used internally
11 =====================================
13 This documentation of the system property flags assume that the
14 default value of the flag is false, unless otherwise specified.
16 SYSTEM PROPERTY: -Dnashorn.unstable.relink.threshold=x
18 This property controls how many call site misses are allowed before a
19 callsite is relinked with "apply" semantics to never change again.
20 In the case of megamorphic callsites, this is necessary, or the
21 program would spend all its time swapping out callsite targets. Dynalink
22 has a default value (currently 8 relinks) for this property if it
23 is not explicitly set.
26 SYSTEM PROPERTY: -Dnashorn.compiler.split.threshold=x
28 This will change the node weight that requires a subgraph of the IR to
29 be split into several classes in order not to run out of bytecode space.
30 The default value is 0x8000 (32768).
33 SYSTEM PROPERTY: -Dnashorn.callsiteaccess.debug
35 See the description of the access logger below. This flag is
36 equivalent to enabling the access logger with "info" level.
39 SYSTEM PROPERTY: -Dnashorn.compiler.intarithmetic
41 Arithmetic operations in Nashorn (except bitwise ones) typically
42 coerce the operands to doubles (as per the JavaScript spec). To switch
43 this off and remain in integer mode, for example for "var x = a&b; var
44 y = c&d; var z = x*y;", use this flag. This will force the
45 multiplication of variables that are ints to be done with the IMUL
46 bytecode and the result "z" to become an int.
48 WARNING: Note that is is experimental only to ensure that type support
49 exists for all primitive types. The generated code is unsound. This
50 will be the case until we do optimizations based on it. There is a CR
51 in Nashorn to do better range analysis, and ensure that this is only
52 done where the operation can't overflow into a wider type. Currently
53 no overflow checking is done, so at the moment, until range analysis
54 has been completed, this option is turned off.
56 We've experimented by using int arithmetic for everything and putting
57 overflow checks afterwards, which would recompute the operation with
58 the correct precision, but have yet to find a configuration where this
59 is faster than just using doubles directly, even if the int operation
60 does not overflow. Getting access to a JVM intrinsic that does branch
61 on overflow would probably alleviate this.
63 There is also a problem with this optimistic approach if the symbol
64 happens to reside in a local variable slot in the bytecode, as those
65 are strongly typed. Then we would need to split large sections of
66 control flow, so this is probably not the right way to go, while range
67 analysis is. There is a large difference between integer bytecode
68 without overflow checks and double bytecode. The former is
69 significantly faster.
72 SYSTEM PROPERTY: -Dnashorn.codegen.debug, -Dnashorn.codegen.debug.trace=<x>
74 See the description of the codegen logger below.
77 SYSTEM_PROPERTY: -Dnashorn.fields.debug
79 See the description on the fields logger below.
82 SYSTEM PROPERTY: -Dnashorn.fields.dual
84 When this property is true, Nashorn will attempt to use primitive
85 fields for AccessorProperties (currently just AccessorProperties, not
86 spill properties). Memory footprint for script objects will increase,
87 as we need to maintain both a primitive field (a long) as well as an
88 Object field for the property value. Ints are represented as the 32
89 low bits of the long fields. Doubles are represented as the
90 doubleToLongBits of their value. This way a single field can be used
91 for all primitive types. Packing and unpacking doubles to their bit
92 representation is intrinsified by the JVM and extremely fast.
94 While dual fields in theory runs significantly faster than Object
95 fields due to reduction of boxing and memory allocation overhead,
96 there is still work to be done to make this a general purpose
97 solution. Research is ongoing.
99 In the future, this might complement or be replaced by experimental
100 feature sun.misc.TaggedArray, which has been discussed on the mlvm
101 mailing list. TaggedArrays are basically a way to share data space
102 between primitives and references, and have the GC understand this.
104 As long as only primitive values are written to the fields and enough
105 type information exists to make sure that any reads don't have to be
106 uselessly boxed and unboxed, this is significantly faster than the
107 standard "Objects only" approach that currently is the default. See
108 test/examples/dual-fields-micro.js for an example that runs twice as
109 fast with dual fields as without them. Here, the compiler, can
110 determine that we are dealing with numbers only throughout the entire
111 property life span of the properties involved.
113 If a "real" object (not a boxed primitive) is written to a field that
114 has a primitive representation, its callsite is relinked and an Object
115 field is used forevermore for that particular field in that
116 PropertyMap and its children, even if primitives are later assigned to
117 it.
119 As the amount of compile time type information is very small in a
120 dynamic language like JavaScript, it is frequently the case that
121 something has to be treated as an object, because we don't know any
122 better. In reality though, it is often a boxed primitive is stored to
123 an AccessorProperty. The fastest way to handle this soundly is to use
124 a callsite typecheck and avoid blowing the field up to an Object. We
125 never revert object fields to primitives. Ping-pong:ing back and forth
126 between primitive representation and Object representation would cause
127 fatal performance overhead, so this is not an option.
129 For a general application the dual fields approach is still slower
130 than objects only fields in some places, about the same in most cases,
131 and significantly faster in very few. This is due the program using
132 primitives, but we still can't prove it. For example "local_var a =
133 call(); field = a;" may very well write a double to the field, but the
134 compiler dare not guess a double type if field is a local variable,
135 due to bytecode variables being strongly typed and later non
136 interchangeable. To get around this, the entire method would have to
137 be replaced and a continuation retained to restart from. We believe
138 that the next steps we should go through are instead:
140 1) Implement method specialization based on callsite, as it's quite
141 frequently the case that numbers are passed around, but currently our
142 function nodes just have object types visible to the compiler. For
143 example "var b = 17; func(a,b,17)" is an example where two parameters
144 can be specialized, but the main version of func might also be called
145 from another callsite with func(x,y,"string").
147 2) This requires lazy jitting as the functions have to be specialized
148 per callsite.
150 Even though "function square(x) { return x*x }" might look like a
151 trivial function that can always only take doubles, this is not
152 true. Someone might have overridden the valueOf for x so that the
153 toNumber coercion has side effects. To fulfil JavaScript semantics,
154 the coercion has to run twice for both terms of the multiplication
155 even if they are the same object. This means that call site
156 specialization is necessary, not parameter specialization on the form
157 "function square(x) { var xd = (double)x; return xd*xd; }", as one
158 might first think.
160 Generating a method specialization for any variant of a function that
161 we can determine by types at compile time is a combinatorial explosion
162 of byte code (try it e.g. on all the variants of am3 in the Octane
163 benchmark crypto.js). Thus, this needs to be lazy
165 3) Possibly optimistic callsite writes, something on the form
167 x = y; //x is a field known to be a primitive. y is only an object as
168 far as we can tell
170 turns into
172 try {
173 x = (int)y;
174 } catch (X is not an integer field right now | ClassCastException e) {
175 x = y;
176 }
178 Mini POC shows that this is the key to a lot of dual field performance
179 in seemingly trivial micros where one unknown object, in reality
180 actually a primitive, foils it for us. Very common pattern. Once we
181 are "all primitives", dual fields runs a lot faster than Object fields
182 only.
184 We still have to deal with objects vs primitives for local bytecode
185 slots, possibly through code copying and versioning.
188 SYSTEM PROPERTY: -Dnashorn.compiler.symbol.trace=[<x>[,*]],
189 -Dnashorn.compiler.symbol.stacktrace=[<x>[,*]]
191 When this property is set, creation and manipulation of any symbol
192 named "x" will show information about when the compiler changes its
193 type assumption, bytecode local variable slot assignment and other
194 data. This is useful if, for example, a symbol shows up as an Object,
195 when you believe it should be a primitive. Usually there is an
196 explanation for this, for example that it exists in the global scope
197 and type analysis has to be more conservative.
199 Several symbols names to watch can be specified by comma separation.
201 If no variable name is specified (and no equals sign), all symbols
202 will be watched
204 By using "stacktrace" instead of or together with "trace", stack
205 traces will be displayed upon symbol changes according to the same
206 semantics.
209 SYSTEM PROPERTY: nashorn.lexer.xmlliterals
211 If this property it set, it means that the Lexer should attempt to
212 parse XML literals, which would otherwise generate syntax
213 errors. Warning: there are currently no unit tests for this
214 functionality.
216 XML literals, when this is enabled, end up as standard LiteralNodes in
217 the IR.
220 SYSTEM_PROPERTY: nashorn.debug
222 If this property is set to true, Nashorn runs in Debug mode. Debug
223 mode is slightly slower, as for example statistics counters are enabled
224 during the run. Debug mode makes available a NativeDebug instance
225 called "Debug" in the global space that can be used to print property
226 maps and layout for script objects, as well as a "dumpCounters" method
227 that will print the current values of the previously mentioned stats
228 counters.
230 These functions currently exists for Debug:
232 "map" - print(Debug.map(x)) will dump the PropertyMap for object x to
233 stdout (currently there also exist functions called "embedX", where X
234 is a value from 0 to 3, that will dump the contents of the embed pool
235 for the first spill properties in any script object and "spill", that
236 will dump the contents of the growing spill pool of spill properties
237 in any script object. This is of course subject to change without
238 notice, should we change the script object layout.
240 "methodHandle" - this method returns the method handle that is used
241 for invoking a particular script function.
243 "identical" - this method compares two script objects for reference
244 equality. It is a == Java comparison
246 "dumpCounters" - will dump the debug counters' current values to
247 stdout.
249 Currently we count number of ScriptObjects in the system, number of
250 Scope objects in the system, number of ScriptObject listeners added,
251 removed and dead (without references).
253 We also count number of ScriptFunctions, ScriptFunction invocations
254 and ScriptFunction allocations.
256 Furthermore we count PropertyMap statistics: how many property maps
257 exist, how many times were property maps cloned, how many times did
258 the property map history cache hit, prevent new allocations, how many
259 prototype invalidations were done, how many time the property map
260 proto cache hit.
262 Finally we count callsite misses on a per callsite bases, which occur
263 when a callsite has to be relinked, due to a previous assumption of
264 object layout being invalidated.
267 SYSTEM PROPERTY: nashorn.methodhandles.debug,
268 nashorn.methodhandles.debug=create
270 If this property is enabled, each MethodHandle related call that uses
271 the java.lang.invoke package gets its MethodHandle intercepted and an
272 instrumentation printout of arguments and return value appended to
273 it. This shows exactly which method handles are executed and from
274 where. (Also MethodTypes and SwitchPoints). This can be augmented with
275 more information, for example, instance count, by subclassing or
276 further extending the TraceMethodHandleFactory implementation in
277 MethodHandleFactory.java.
279 If the property is specialized with "=create" as its option,
280 instrumentation will be shown for method handles upon creation time
281 rather than at runtime usage.
284 SYSTEM PROPERTY: nashorn.methodhandles.debug.stacktrace
286 This does the same as nashorn.methodhandles.debug, but when enabled
287 also dumps the stack trace for every instrumented method handle
288 operation. Warning: This is enormously verbose, but provides a pretty
289 decent "grep:able" picture of where the calls are coming from.
291 See the description of the codegen logger below for a more verbose
292 description of this option
295 SYSTEM PROPERTY: nashorn.scriptfunction.specialization.disable
297 There are several "fast path" implementations of constructors and
298 functions in the NativeObject classes that, in their original form,
299 take a variable amount of arguments. Said functions are also declared
300 to take Object parameters in their original form, as this is what the
301 JavaScript specification mandates.
303 However, we often know quite a lot more at a callsite of one of these
304 functions. For example, Math.min is called with a fixed number (2) of
305 integer arguments. The overhead of boxing these ints to Objects and
306 folding them into an Object array for the generic varargs Math.min
307 function is an order of magnitude slower than calling a specialized
308 implementation of Math.min that takes two integers. Specialized
309 functions and constructors are identified by the tag
310 @SpecializedFunction and @SpecializedConstructor in the Nashorn
311 code. The linker will link in the most appropriate (narrowest types,
312 right number of types and least number of arguments) specialization if
313 specializations are available.
315 Every ScriptFunction may carry specializations that the linker can
316 choose from. This framework will likely be extended for user defined
317 functions. The compiler can often infer enough parameter type info
318 from callsites for in order to generate simpler versions with less
319 generic Object types. This feature depends on future lazy jitting, as
320 there tend to be many calls to user defined functions, some where the
321 callsite can be specialized, some where we mostly see object
322 parameters even at the callsite.
324 If this system property is set to true, the linker will not attempt to
325 use any specialized function or constructor for native objects, but
326 just call the generic one.
329 SYSTEM PROPERTY: nashorn.tcs.miss.samplePercent=<x>
331 When running with the trace callsite option (-tcs), Nashorn will count
332 and instrument any callsite misses that require relinking. As the
333 number of relinks is large and usually produces a lot of output, this
334 system property can be used to constrain the percentage of misses that
335 should be logged. Typically this is set to 1 or 5 (percent). 1% is the
336 default value.
339 SYSTEM_PROPERTY: nashorn.profilefile=<filename>
341 When running with the profile callsite options (-pcs), Nashorn will
342 dump profiling data for all callsites to stderr as a shutdown hook. To
343 instead redirect this to a file, specify the path to the file using
344 this system property.
347 ===============
348 2. The loggers.
349 ===============
351 It is very simple to create your own logger. Use the DebugLogger class
352 and give the subsystem name as a constructor argument.
354 The Nashorn loggers can be used to print per-module or per-subsystem
355 debug information with different levels of verbosity. The loggers for
356 a given subsystem are available are enabled by using
358 --log=<systemname>[:<level>]
360 on the command line.
362 Here <systemname> identifies the name of the subsystem to be logged
363 and the optional colon and level argument is a standard
364 java.util.logging.Level name (severe, warning, info, config, fine,
365 finer, finest). If the level is left out for a particular subsystem,
366 it defaults to "info". Any log message logged as the level or a level
367 that is more important will be output to stderr by the logger.
369 Several loggers can be enabled by a single command line option, by
370 putting a comma after each subsystem/level tuple (or each subsystem if
371 level is unspecified). The --log option can also be given multiple
372 times on the same command line, with the same effect.
374 For example: --log=codegen,fields:finest is equivalent to
375 --log=codegen:info --log=fields:finest
377 The subsystems that currently support logging are:
380 * compiler
382 The compiler is in charge of turning source code and function nodes
383 into byte code, and installs the classes into a class loader
384 controlled from the Context. Log messages are, for example, about
385 things like new compile units being allocated. The compiler has global
386 settings that all the tiers of codegen (e.g. Lower and CodeGenerator)
387 use.s
390 * codegen
392 The code generator is the emitter stage of the code pipeline, and
393 turns the lowest tier of a FunctionNode into bytecode. Codegen logging
394 shows byte codes as they are being emitted, line number information
395 and jumps. It also shows the contents of the bytecode stack prior to
396 each instruction being emitted. This is a good debugging aid. For
397 example:
399 [codegen] #41 line:2 (f)_afc824e
400 [codegen] #42 load symbol x slot=2
401 [codegen] #43 {1:O} load int 0
402 [codegen] #44 {2:I O} dynamic_runtime_call GT:ZOI_I args=2 returnType=boolean
403 [codegen] #45 signature (Ljava/lang/Object;I)Z
404 [codegen] #46 {1:Z} ifeq ternary_false_5402fe28
405 [codegen] #47 load symbol x slot=2
406 [codegen] #48 {1:O} goto ternary_exit_107c1f2f
407 [codegen] #49 ternary_false_5402fe28
408 [codegen] #50 load symbol x slot=2
409 [codegen] #51 {1:O} convert object -> double
410 [codegen] #52 {1:D} neg
411 [codegen] #53 {1:D} convert double -> object
412 [codegen] #54 {1:O} ternary_exit_107c1f2f
413 [codegen] #55 {1:O} return object
415 shows a ternary node being generated for the sequence "return x > 0 ?
416 x : -x"
418 The first number on the log line is a unique monotonically increasing
419 emission id per bytecode. There is no guarantee this is the same id
420 between runs. depending on non deterministic code
421 execution/compilation, but for small applications it usually is. If
422 the system variable -Dnashorn.codegen.debug.trace=<x> is set, where x
423 is a bytecode emission id, a stack trace will be shown as the
424 particular bytecode is about to be emitted. This can be a quick way to
425 determine where it comes from without attaching the debugger. "Who
426 generated that neg?"
428 The --log=codegen option is equivalent to setting the system variable
429 "nashorn.codegen.debug" to true.
432 * lower
434 This is the first lowering pass.
436 Lower is a code generation pass that turns high level IR nodes into
437 lower level one, for example substituting comparisons to RuntimeNodes
438 and inlining finally blocks.
440 Lower is also responsible for determining control flow information
441 like end points.
444 * attr
446 The lowering annotates a FunctionNode with symbols for each identifier
447 and transforms high level constructs into lower level ones, that the
448 CodeGenerator consumes.
450 Lower logging typically outputs things like post pass actions,
451 insertions of casts because symbol types have been changed and type
452 specialization information. Currently very little info is generated by
453 this logger. This will probably change.
456 * finalize
458 This --log=finalize log option outputs information for type finalization,
459 the third tier of the compiler. This means things like placement of
460 specialized scope nodes or explicit conversions.
463 * fields
465 The --log=fields option (at info level) is equivalent to setting the
466 system variable "nashorn.fields.debug" to true. At the info level it
467 will only show info about type assumptions that were invalidated. If
468 the level is set to finest, it will also trace every AccessorProperty
469 getter and setter in the program, show arguments, return values
470 etc. It will also show the internal representation of respective field
471 (Object in the normal case, unless running with the dual field
472 representation)