Changes between Version 2 and Version 3 of SquirrelFish

Timestamp:

Apr 20, 2008, 4:46:39 PM (17 years ago)

Author:

mjs@apple.com

Comment:

--

SquirrelFish

@@ -2 +2 @@
 == !SquirrelFish ==
 
+SquirrelFish is an incremental rewrite of JavaScriptCore to turn it into a bytecode interpreter. It is a direct-dispatch register VM. Currently it is showing promising performance results, even at a fairly early stage, and we are working to fix the remaining blockers to merging to trunk.
+
 '''Blockers to merge to trunk'''
 
-See SquirrelFishBlockers
-
-'''Major bits left over:'''
-  * Getters and Setters
-  * toString/valueOf
-     * It occurs to me that these could both (s|g)etters and toString could be "trivially" handled through the standard js -> native -> js call model -- at least in the short-medium term
-  * eval
-
-'''List of tasks in no order -- pick one, tell everyone what you're doing:
+See SquirrelFishBlockers.
+
+'''Lost optimizations'''
+
+Some optimizations have been temporarily lost as part of the squirrelfish work. Other speedups seem to make up for them, so they could be restored either before or after merging back to trunk.
+
+ * optimized multiscope access [https://bugs.webkit.org/show_bug.cgi?id=18645 18645]
+ * optimized access to global built-ins (originally done in [http://trac.webkit.org/projects/webkit/changeset/31226 r31226]) [https://bugs.webkit.org/show_bug.cgi?id=18646 18646]
+ * static type inference of things like "numeric less than" and "string add" [https://bugs.webkit.org/show_bug.cgi?id=18647 18647]
+
+
+
+'''What people are working on:
 '''
 
@@ -18 +24 @@
 '''
 
-Re-entry into Machine::privateExecute:
-  * More indirection in the register file
-  * Don't overwrite nested register frames when re-entering global execution
-  * Provide API for re-entering non-global execution (i.e., function callbacks)
-
-Arguments object
+Fixing scope chain handling, JS regression test failures.
 
 '''Cameron is working on:
 '''
 
+JS regression test failures.
+
 Better code generation. We have been pondering whether to have a separate peephole optimization pass or to incorporate peephole optimization into code generation. Either way, we should look at some code generation algorithms based on tile matching. We also want to choose an approach that will be compatible with planned extensions, e.g. superinstructions.
 
-Constant pool GC
-
 '''Oliver is working on:
 '''
 
-Making VM throw exceptions for invalid behaviour; Finally blocks.
+Getters & setters.
 
 '''Sam is working on (when he sees fit to do so):
@@ -43 +44 @@
 '''
 
-ScopeChain hackery
-
-'''You could take something from Geoff, or make something up yourself, or do one of these:'''
-
-Leftover opcodes:
-
- * ArgumentsNode
- * AssignErrorNode
- * BreakpointCheckStatement
- * ConstDeclNode
- * ConstStatementNode
- * ElementNode
- * EvalFunctionCallNode
- * ParameterNode
- * PostDecConstNode
- * PostfixErrorNode
- * PreDecConstNode
- * PrefixErrorNode
- * PropertyNode
- * ReadModifyConstNode
- * ThrowNode
- * TryNode
-
-Where SunSpider tests currently fail codegen:
-
- * 3d-cube: ungettableGetter
- * 3d-morph: ungettableGetter
- * 3d-raytrace: crash on failing toObject conversion for base of call (bad Args to constructor)
- * access-binary-trees: crash on failing toObject conversion for base of call (bad Args to constructor)
- * access-fannkuch: crash on failing toObject conversion for base of call (bad Args to constructor)
- * access-nbody: new expr (NBodySystem)
- * access-nsieve: ungettableGetter
- * bitops-3bit-bits-in-byte: SUCCESS
- * bitops-bits-in-byte: SUCCESS
- * bitops-bitwise-and: SUCCESS
- * bitops-nsieve-bits: ungettableGetter
- * controlflow-recursive: ungettableGetter
- * crypto-aes: crash on failing toObject conversion for base of call (bad Args to constructor)
- * crypto-md5: ungettableGetter
- * crypto-sha1: ungettableGetter
- * date-format-tofte: local eval - eval(ia[ij] + "()")
- * date-format-xparb: bracket call - this[func]()
- * math-cordic: ungettableGetter
- * math-partial-sums: crash on failing toObject conversion for base of call (bad Args to constructor)
- * math-spectral-norm: ungettableGetter
- * regexp-dna: regexp literal
- * string-base64: crash on failing toObject conversion for base of call (bad Args to constructor)
- * string-fasta: crash on failing toObject conversion for base of call (bad Args to constructor)
- * string-tagcloud: trying to get from a base register containing 0x0 (bad codegen?)
- * string-unpack-code: crash on failing toObject conversion for base of call (bad Args to constructor)
- * string-validate-input: regexp literal
-
-Harness failures:
-
- * sunspider-analyze-results: KJS::resolve SHOULD NEVER BE REACHED
- * sunspider-compare-results: SUCCESS (but not running yet)
- * sunspider-standalone-compare: KJS::resolve SHOULD NEVER BE REACHED
- * sunspider-standalone-driver: new expr (invoking Array constructor)
+Organizing remaining blocker issues.
+
+'''Assorted notes, some of these may be obsolete'''
 
 Optimize dynamic scopes that aren't closures not to save the environment on return
@@ -113 +59 @@
 
 {{{
-optimized multiscope access
-optimized access to global built-ins (http://trac.webkit.org/projects/webkit/changeset/31226)
-static type inference of things like "numeric less than" and "string add"
 
 // WARNING: If code generation wants to optimize resolves to parent scopes,