Andreas Gal
|
ca4e6419ee
|
shaver's favorite operating system of choice doesn't like templates, so de-template tracker since we use it with LInsp only anyway.
|
2008-07-10 18:42:04 -07:00 |
|
shaver@mozilla.org
|
6a5deaf550
|
Decouple jscntxt.h from jstracer.h so that xpconnect doesn't try to include all of
nanojit.
|
2008-07-10 17:29:16 -04:00 |
|
shaver@mozilla.org
|
2a0f9b6245
|
can't have zero-sized arrays in ISO C++, says gcc
|
2008-07-10 09:29:42 -04:00 |
|
Brendan Eich
|
20a5a9ef31
|
1. Fix inc to address the right result stack slot; 2. Require via assertions that the interpreter memoize implicit gvars (patch to do that next; bitwise-and will assert until that lands).
|
2008-07-09 15:15:32 -07:00 |
|
Brendan Eich
|
d8cf234b33
|
Memoize global name gets and sets as if they refer to a declared gvar.
|
2008-07-09 11:42:31 -07:00 |
|
Andreas Gal
|
05554d05b3
|
Stricter typing of the tracker code path which now only accepts jsval as suggested by Brendan.
|
2008-07-09 00:29:23 -07:00 |
|
Andreas Gal
|
d6b834b98a
|
Move builtin[] table in jsbuiltin.cpp
|
2008-07-08 23:48:06 -07:00 |
|
David Anderson
|
fcf8410125
|
Builds against latest nanojit; merged VMSideExitInfo into SideExit
|
2008-07-08 17:16:51 -07:00 |
|
Andreas Gal
|
4550dd77f0
|
Introduce a generic stack frame walking macro. This eliminates the 6 (hg pull) redundant blocks of code that all walked the stack in 6 different ways, each with their individuals quirks.
|
2008-07-08 16:29:23 -07:00 |
|
Andreas Gal
|
56397d3c0e
|
Attempt at cleaning up the typemap/stack frame iteration code.
|
2008-07-07 18:35:38 -07:00 |
|
Andreas Gal
|
0fe5682c00
|
Make gvar ops use the global varobj values we unbox onto the native stack. The stack offset handling needs more work.
|
2008-07-07 17:21:54 -07:00 |
|
Andreas Gal
|
1d36b97516
|
Unbox global variables onto the native stack and re-box them. This has certain synchronization implications but will make top-level scripts quite a bit faster.
|
2008-07-07 15:15:56 -07:00 |
|
Brendan Eich
|
175ce14f5b
|
Indentation and comment wrapping (at 79, 99 looks too long and most comments still avoid going past 80+/-).
|
2008-07-07 02:55:03 -07:00 |
|
Brendan Eich
|
6d65efee60
|
Use jsopcode.tbl for tracer JSOP_* method decls; trim trailing whitespace.
|
2008-07-07 02:21:04 -07:00 |
|
Andreas Gal
|
a38e8a96c3
|
Add support to demote stores of constants that are currently represented as float but are really integers. bitwise.js is now compiled complete fp-casts free.
|
2008-07-07 01:05:53 -07:00 |
|
Andreas Gal
|
efa3309778
|
Demote floating point comparisons even if one side is constant (bug in nanojit, filed as #443884 against tamarin). Also demote add/sub/mul using the integer overflow detection side exit code that Ed adopted upstream. Tight loops (bitwise) are not emitted completely as integer code after the initial compilation using doubles triggered a speculative demotion of the context slots.
|
2008-07-07 00:43:40 -07:00 |
|
Andreas Gal
|
8ce0de1ad7
|
Sink type conversions into the side exit by updating the map and seeing though the cast in the store (in ExitFilter). Add a whole bunch of asserts on the varios speculative type states to make sure we get the state machine right. Import speculated integer values as integers into the trace (indicate in entry map).
|
2008-07-07 00:10:22 -07:00 |
|
Andreas Gal
|
34c2283213
|
Added loop-tail driven context slot type demotion. Without all the buzzwords, what this essentially means is that we detect if the last value that the trace leaves in a slot (which is the value that flows along the backedge back to the loop header) is known to originate from an integer value via i2f, we flag the slot as type integer and recompile the trace. We do this where type stability is certain (result of an and, i.e.) or where its very likely (++operator). If the speculation fails, the same analysis flags the slot as blocked, which means it will always be double. The hope is that this analysis converges quickly (1-2 recompilations tops).
|
2008-07-06 22:35:19 -07:00 |
|
Andreas Gal
|
efee2381b5
|
Make typemap uint8_t and fix order of arguments in call to BoxDouble.
|
2008-07-06 20:25:49 -07:00 |
|
Andreas Gal
|
aeeff83beb
|
Move the type level from int/double to number. All traces start out as double in all slots, and denote and promote to/from int as needed. The FuncFilter optimizes on-trace casting and elininates redundant f->i-> chains. More optimization needed on this of course, and this code is now a bit slower than the previous integer-register use. However, this does solve the q += 2.5 issues. The heap access code does not properly cast yet and is likely unstable.
|
2008-07-06 15:55:04 -07:00 |
|
shaver@mozilla.org
|
b28afaa80d
|
Trace GETGVAR and INCGVAR, though not yet correctly.
|
2008-07-06 16:59:59 -04:00 |
|
Andreas Gal
|
d30b7d1c8a
|
Add unary and binary helpers that automatically demote and promote when dealing with integer operations (not used yet, need loop typemap peeling in place first.)
|
2008-07-06 13:16:34 -07:00 |
|
Andreas Gal
|
5816dee707
|
Fix template, should use typename, not class.
|
2008-07-06 12:03:51 -07:00 |
|
Andreas Gal
|
1343901446
|
Add ExitFilter, which builds side exit typemaps for us. This had to be moved into a filter, because this has to happen after we eliminate redundant i2f-f2i chains as we can sink those into the side exit by simply storing the unpromoted/undemoted value and just flip the type in the exit map.
|
2008-07-06 11:48:41 -07:00 |
|
Andreas Gal
|
84ec537a0c
|
Merge.
|
2008-07-06 11:19:06 -07:00 |
|
Andreas Gal
|
9501bcd749
|
Added doubleToUint32 builtin and make tracker a template.
|
2008-07-06 10:59:55 -07:00 |
|
shaver@mozilla.org
|
d881a494aa
|
fix builtin_UnboxInt32 signature and name shape_ins for debugging
|
2008-07-06 13:58:59 -04:00 |
|
shaver@mozilla.org
|
45b3826103
|
Trace JSOP_NAME, and refactor out jsval unboxing.
|
2008-07-06 13:38:55 -04:00 |
|
shaver@mozilla.org
|
06a27ca367
|
begin on JSOP_NAME, refactor stobj_get_slot to chain better
|
2008-07-06 12:15:55 -04:00 |
|
Andreas Gal
|
100f4ea5e0
|
Added support for semi-stable loop variables. Compiling for(...) q += 2.5; is ridiculously difficult because it flip-flops between int and double. Add support to promote integer values to doubles at the loop tail if at loop entry we expect a double. Since this isn't possible the other way around, we have to get luck that we catch a path into the loop where q is already double. For this we add 3 trigger points (10, 13, 37). We will try three times to record a trace at those iteration counts of a loop. If none succeed the loop is blacklisted. This probably needs more tuning down the road.
|
2008-07-05 23:21:53 -07:00 |
|
Andreas Gal
|
41efb6c488
|
Merge and rename jsIf to ifop.
|
2008-07-05 22:23:34 -07:00 |
|
Andreas Gal
|
0b28897ad3
|
Add helpers for i->f and f->i conversion. The f->i path goes via builtin calls.
|
2008-07-05 22:04:58 -07:00 |
|
shaver@mozilla.org
|
a268883f3c
|
Implement JSOP_IFEQ, JSOP_IFNE, JSOP_DUP, JSOP_DUP2.
|
2008-07-05 23:47:24 -04:00 |
|
Andreas Gal
|
075e3f00a6
|
Add casting for objects and cleanup casting code.
|
2008-07-05 20:30:11 -07:00 |
|
Andreas Gal
|
0245a6ba7c
|
Believe it or not NEG can actually overflow the int32 range, so add an overflow bailout. This will only trap for -0x80000000.
|
2008-07-05 19:25:38 -07:00 |
|
Andreas Gal
|
48e459ba73
|
Added blacklisting of recording points where we failed to complete a trace (overly aggressive at this point, needs tuning, we want to try several times for each point.)
|
2008-07-05 19:15:00 -07:00 |
|
Andreas Gal
|
cf8e2a555b
|
Added boolean boxing/unboxing code.
|
2008-07-05 18:34:41 -07:00 |
|
Andreas Gal
|
888d2f9aee
|
Add support for some misc opcodes including binary and arithmetic and/or/not.
|
2008-07-05 17:26:00 -07:00 |
|
Andreas Gal
|
19152ffb1f
|
Signal error from the boxing/unboxing using magic cookies since gcc seems to very seriously object to the use of uint64 return values during a fastcall (horribly inefficient code).
|
2008-07-05 16:28:03 -07:00 |
|
Andreas Gal
|
d765c3ead2
|
Finish SETELEM for int and double values.
|
2008-07-05 14:00:32 -07:00 |
|
Andreas Gal
|
1e9c9451e0
|
Merge unboxing code into GETELEM code.
|
2008-07-05 13:44:48 -07:00 |
|
Andreas Gal
|
503bf13dac
|
Merge.
|
2008-07-05 11:46:08 -07:00 |
|
Andreas Gal
|
f9981c8870
|
Added box_int_jsval for the store path. Use only there.
|
2008-07-05 11:45:53 -07:00 |
|
Andreas Gal
|
ba7e66df92
|
Code generators to access object slots and native code callouts (builtins) for boxing doubles and ints. Ints have to be boxed through a native code helper on read and write (BoxInt32 and UnboxInt32), because we sometimes have to cast internally to double to store 32-bit values. We don't want a separate trace in this case, so we have to do this inline in a helper. Also a couple of modifications to shaver's code. Always make sure to check types (JSOP_NEG).
|
2008-07-05 11:35:55 -07:00 |
|
Brendan Eich
|
ab0e34e2ae
|
Try to keep 64-bit portability via size_t instead of unsigned, jsuword for uintptr instead of long, etc.
|
2008-07-05 11:18:26 -07:00 |
|
shaver@mozilla.org
|
de59f88d9c
|
Beginning of SETELEM/GETELEM tracing for dense arrays. Needs computed-offset
insLoad before they'll work.
|
2008-07-05 09:47:59 -04:00 |
|
Andreas Gal
|
2df2214a9e
|
Introduce asInt and asDouble to check for the type of values based on the actual value since some 32-bit integers hide out in doubles.
|
2008-07-04 23:53:29 -07:00 |
|
Andreas Gal
|
e35d351b28
|
Added support for SET_VAR again. bitwise.js working now.
|
2008-07-04 15:21:56 -07:00 |
|
Andreas Gal
|
ab1dc289ba
|
Fixed some stack handling and trace activation issues. We can run trace.js again.
|
2008-07-04 15:07:05 -07:00 |
|
Andreas Gal
|
5e9ca6ef0e
|
Cleanup of stack handling.
|
2008-07-04 13:23:42 -07:00 |
|
Andreas Gal
|
9b8f5f75a8
|
Added back support for the instructions required to compile trace.js.
|
2008-07-04 03:06:18 -07:00 |
|
Andreas Gal
|
1503932551
|
Properly switch tracer on and off depending on loopEdge and abort signaling from trace recorder recording functions.
|
2008-07-04 01:10:57 -07:00 |
|
Andreas Gal
|
efb72df02a
|
Re-integrate trace recording and trace activation infrastructure. Model more closely after Tamarin.
|
2008-07-04 00:51:30 -07:00 |
|
Andreas Gal
|
731614bfe9
|
Create a stub for each opcode in TraceRecorder and invoke them from the stubs that are pointed to by the recorder dispatch-table.
|
2008-07-03 23:57:57 -07:00 |
|
Andreas Gal
|
62d0c7bbfb
|
Switch back to mozilla-central jsinter.cpp. Lets try a different approach to attach the tracer to the interpreter.
|
2008-07-03 22:08:13 -07:00 |
|
Andreas Gal
|
9a1bbdcdc6
|
Fix unnecessary type prefix (reported by bc, doesn't pass gcc 4.2.3)
|
2008-07-03 17:02:19 -07:00 |
|
Andreas Gal
|
2cd7b6993b
|
VOIDs hang out in the BOOLEAN value space, so we have to treat them like a boolean on the trace. For ValuetoECMAInt32 all special boolean values except TRUE produce 0, and TRUE produces 1. Use a conditional move to implement this.
|
2008-07-02 19:11:07 -07:00 |
|
Andreas Gal
|
b137f2da3b
|
Introduce a VMSideExitInfo structure (holds the exit typemap only for now).
|
2008-07-02 18:06:24 -07:00 |
|
Andreas Gal
|
523ba92b3d
|
Use fragmento to track fragments and add proper trace activiation code.
|
2008-07-02 17:19:07 -07:00 |
|
Andreas Gal
|
6f3af33edf
|
Add type stability check for loop variables. Rename readstack to import.
|
2008-07-02 15:14:43 -07:00 |
|
Andreas Gal
|
e70272bfab
|
Removed attempt to imply ints into doubles on the fly. This can't work. Use proper float loads where needed and add some initial code for builtin functions.
|
2008-07-02 14:38:12 -07:00 |
|
Andreas Gal
|
f3fba20c9c
|
Create type maps during trace entry and in each side exit and store them in the LIR using LIR_skip. Use these type maps during trace entry and exit.
|
2008-07-01 19:43:10 -07:00 |
|
Andreas Gal
|
39891951f4
|
Merge.
|
2008-07-01 13:36:33 -07:00 |
|
shaver@mozilla.org
|
50d599cb99
|
label arg/var/stack/sp/state for easier trace-reading
|
2008-07-01 08:06:02 -04:00 |
|
Andreas Gal
|
d3e7a57a68
|
Fix deallocation bug in the recorder. We will have to lift more code Tamarin's Interpreter to stabilize the recording.
|
2008-07-01 02:37:07 -07:00 |
|
Andreas Gal
|
84274f1b33
|
Save the VM registers into markRegs at opcode entry and restore that state when ending or aborting recording. The same info is pushed into sideexit and used by guards. With this change we can eliminate the hack in MONITOR_BRANCH that had to adjust the stack pointer depending on the opcode type, and it also allows us to trace through the recently added boolean guard opcode fusing. This improves trace code quality, since only the pre-conditional check state is saved. The guard restores into that state, so the trace code doesn't have to emit the value of the conditional evaluation onto the stack, saving a handful ops.
|
2008-06-30 18:31:29 -07:00 |
|
Andreas Gal
|
87396e1870
|
Rework memory management, don't leak memory in the filter pipeline.
|
2008-06-30 18:08:32 -07:00 |
|
Andreas Gal
|
19b4bcdc69
|
Save the side exit state at entry in to the opcode (since we want to restart ops) and enable store filter.
|
2008-06-30 17:12:52 -07:00 |
|
Andreas Gal
|
5d1d997c9d
|
Housekeeping (add comments, removed some dead code.)
|
2008-06-30 09:36:10 -07:00 |
|
Andreas Gal
|
d8d25140f9
|
Mockup of trace execution. Speedup is 10x for a tight loop. Keep in mind that this is a hack and the trace code is not optimized yet.
|
2008-06-29 23:02:22 -07:00 |
|
Andreas Gal
|
b1e24b8c98
|
Execute trace code. Boxing back the side exit state is not handled yet.
|
2008-06-29 20:56:06 -07:00 |
|
Andreas Gal
|
e267cb120c
|
Fix native frame addressing (reported by vlad.)
|
2008-06-29 17:17:39 -07:00 |
|
Andreas Gal
|
253b2572b2
|
Initialize jump table when side-entering the interpreter. Keep track of cx and sp directly in the recorder instead of going through the tracker.
|
2008-06-29 13:53:40 -07:00 |
|
Andreas Gal
|
ef71223de2
|
Remember JSContext instead of trying to keep track of frame pointer.
|
2008-06-28 18:47:12 -07:00 |
|
Andreas Gal
|
628a1f6985
|
Cleanup trace abort/end code.
|
2008-06-28 18:19:21 -07:00 |
|
Andreas Gal
|
ba35b66f90
|
Make internal form of nativeFrameSize private.
|
2008-06-28 17:22:08 -07:00 |
|
Andreas Gal
|
6df975a5b5
|
Rewritten frame management. Use cx->fp->regs chain instead of direct passing of those structures.
|
2008-06-28 17:14:06 -07:00 |
|
Andreas Gal
|
552425ee07
|
Hook up pc and sp to the tracer in order to generate proper PC/SP adjustment code.
|
2008-06-28 11:00:18 -07:00 |
|
Andreas Gal
|
1f37ae3e62
|
Small cleanups and licensing blurb housekeeping.
|
2008-06-28 09:58:15 -07:00 |
|
shaver@mozilla.org
|
4fc71515e0
|
Add guards for XML objects, and some tracer bits to accommodate them.
|
2008-06-27 23:58:06 -04:00 |
|
Andreas Gal
|
e5d4f398ef
|
Merge.
|
2008-06-27 18:09:33 -07:00 |
|
Andreas Gal
|
40f1a90a78
|
Move recorder initialization into TraceRecorder and encapsulate its state.
|
2008-06-27 18:06:50 -07:00 |
|
Andreas Gal
|
2c370239e5
|
Add a way to calculate the current size of the native frame.
|
2008-06-27 16:24:17 -07:00 |
|
shaver@mozilla.org
|
06500a1a50
|
merge
|
2008-06-27 12:01:28 -04:00 |
|
shaver@mozilla.org
|
af19dd5627
|
rename ENABLE_TRACER to TRACING_ENABLED and explicitly parameterize on cx
|
2008-06-27 11:41:59 -04:00 |
|
Andreas Gal
|
afd5f3b1e3
|
Clarified argument names in guard code and explain better why we guard for overflow the way we do.
|
2008-06-27 07:28:10 -07:00 |
|
Andreas Gal
|
e59076860a
|
Instead of guarding on null -> boolean conversion just emit a null compare and use the boolean result.
|
2008-06-27 06:53:10 -07:00 |
|
Andreas Gal
|
0c5757a21f
|
Add a new option -j to enable the JIT. The default is off so we can easily benchmark the overhead without the tracer.
|
2008-06-27 00:22:53 -08:00 |
|
Andreas Gal
|
a21ca6a788
|
Prepare LIR_ov to be used as soon Ed adds it and add a few comments that explain how overflow is handled.
|
2008-06-26 23:34:50 -08:00 |
|
Andreas Gal
|
49fe7a95eb
|
More cleanup and code move into TraceRecorder and prepare for LIR_ov.
|
2008-06-26 23:22:16 -08:00 |
|
Andreas Gal
|
f357ff14f0
|
Move the recorder functionality into TraceRecorder to unclutter trace inlines.
|
2008-06-26 21:58:57 -08:00 |
|
Andreas Gal
|
0c98250a90
|
Move set/get into recorder and introduce init.
|
2008-06-26 21:41:10 -08:00 |
|
Andreas Gal
|
431b7b810a
|
Moved loading context slots into the recorder (more to follow).
|
2008-06-26 21:33:33 -08:00 |
|
Andreas Gal
|
693ee7f0cc
|
Turn macros into inline functions and start prepatations for inlining function calls.
|
2008-06-26 21:19:11 -08:00 |
|
Andreas Gal
|
c7bab6721b
|
Drop JS prefix from classes related to traceing.
|
2008-06-26 19:49:04 -08:00 |
|
Andreas Gal
|
ad5ceed9cd
|
Merge.
|
2008-06-26 19:46:51 -08:00 |
|
Andreas Gal
|
a1b42735cd
|
Backed out changeset fa82b7eda72a
|
2008-06-26 19:44:23 -08:00 |
|
Andreas Gal
|
5b4f0aaeaa
|
Allocate recorder dynamically to avoid having a vtable in the struct holding the reference. Emit writebacks for every update to the tracker, not just stack writes.
|
2008-06-26 19:37:28 -08:00 |
|
Andreas Gal
|
407b9cdf75
|
Parameterize tracker.
|
2008-06-26 17:13:54 -08:00 |
|
Andreas Gal
|
6d192a7054
|
Added end-of-trace detection and make sure trace loops back to the loop header.
|
2008-06-22 09:30:04 -07:00 |
|
Andreas Gal
|
78da0668af
|
Start attaching our tracer to nanojit. This is incomplete and meant for review by David only.
|
2008-06-21 14:55:32 -07:00 |
|
Andreas Gal
|
33ecda6e85
|
Added danderson's test cases for nanojit. Execute with nanojit() from JS shell.
|
2008-06-19 18:28:15 -07:00 |
|
Andreas Gal
|
179be35727
|
Notify the recorder of all primitives as they are recorded. If the recorder doesn't support a primitive we abort the trace. In debugging mode this also stops the VM.
|
2008-06-11 20:22:00 -07:00 |
|
Andreas Gal
|
448d1ba7e3
|
Merge with be's last push.
|
2008-06-11 18:13:43 -07:00 |
|
Andreas Gal
|
bb1b206714
|
Use an error property in the recorder instead of a return value to indicate errors and make sure DO_OP aborts the trace once we go into the error state in the tracer.
|
2008-06-11 17:59:28 -07:00 |
|
Brendan Eich
|
a64ed1a5ee
|
native_pointer_to_jsval should not lose low-order bits (plus random style nits).
|
2008-06-11 17:21:15 -07:00 |
|
Andreas Gal
|
54dd5103e9
|
Track data flow through variables in the recorder. Set an error flag if the recorder signals an error.
|
2008-06-11 17:15:49 -07:00 |
|
Brendan Eich
|
74c80d5597
|
Kill trailing spaces.
|
2008-06-11 13:07:24 -07:00 |
|
Andreas Gal
|
b484f790e4
|
Start hooking the recorder into the tracer.
|
2008-06-11 12:12:29 -07:00 |
|
Andreas Gal
|
fef28e9f56
|
Allow recorder to abort recording.
|
2008-06-11 11:29:35 -07:00 |
|
Andreas Gal
|
17d0d41459
|
Added branch monitoring using a single unified branch frequency counter.
|
2008-06-10 16:49:05 -07:00 |
|
Andreas Gal
|
a1bf0819b2
|
Removed JSOP_HEADER code. Its too costly. This patch restore essentially the same performance as unmodified SM.
|
2008-06-10 14:44:51 -07:00 |
|
Brendan Eich
|
04b1ac5592
|
Merge.
|
2008-06-04 00:09:57 -07:00 |
|
Brendan Eich
|
894dd03e2a
|
First stage of loop table work; bitmap free space management and GC hook-up still to come.
|
2008-06-03 23:52:28 -07:00 |
|
Andreas Gal
|
9684e2070f
|
Introduced a JavaScript recording script that is loaded dynamically.
|
2008-06-03 21:01:23 -07:00 |
|
Andreas Gal
|
be39054e45
|
Store the recorder script in JSTraceMonitor and make sure its traced by the GC. Also clean up the tracing of JSTraceMonitor.
|
2008-06-03 17:35:30 -07:00 |
|
Andreas Gal
|
8b6c4d02c6
|
Remove InitTacer. Pointless for JSRuntime.
|
2008-06-03 17:04:42 -07:00 |
|
Brendan Eich
|
9c2b039d6f
|
JSOP_HEADER takes a byte index of loop header counting from script->loopBase, and related changes.
|
2008-06-03 14:48:47 -07:00 |
|
Brendan Eich
|
946fa41d8c
|
Fiddle loop table slot interface and impl in hope of freeing slots over time.
|
2008-06-02 18:06:33 -07:00 |
|
Brendan Eich
|
3db9424289
|
Merge and undo js_AllocateLoopTableSlot reparam.
|
2008-06-02 17:28:39 -07:00 |
|
Brendan Eich
|
01514c4b16
|
No JS_TRACER ifdefs, keep line len < 80, reparameterize jstracer.cpp functions, js_OpLength helper.
|
2008-06-02 17:19:41 -07:00 |
|
Brendan Eich
|
5b116ae998
|
Merge, style nits, no locking for tracing.
|
2008-05-31 15:29:54 -07:00 |
|
Andreas Gal
|
e461a72736
|
The table is now per-thread in a multi-threaded environment, and per-runtime otherwise. During code generation we merely allocate a loop table slot to each loop. Each thread will enlarge the table as needed in JSOP_HEADER.
|
2008-05-30 18:58:43 -07:00 |
|