d55c07198d
Add TODO about more permissive label handling in registerizeHarder
2014-01-23 22:37:37 +11:00
d21f2ea3ad
Simplify handling of switch statements in registerizeHarder.
...
It now assumes they're of the very restricted for that's actually
produced by emscripten, which both simplified the code and uncovered
opportunities for further register reduction in a testcase.
2014-01-23 22:37:31 +11:00
a912e8271c
Reduce memory usage when assigning registers in a block.
...
Rather than pre-calculating liveness states and then processing nodes
in execution order, we process nodes in reverse execution order and
use the liveness info implicit in this traversal. Names are assigned
a register when they're used for the first time, and the register is
freed when we reach an assignment.
2014-01-23 22:32:40 +11:00
2266ef7cac
assign to null to clear a local variable, avoid delete
2014-01-21 17:50:46 -08:00
55cadd9dc5
fix typo
2014-01-21 17:50:46 -08:00
c744bcec81
avoid nonstandard use of splice
2014-01-21 17:50:46 -08:00
b604a07c61
rename a variable to our conventions
2014-01-21 17:50:46 -08:00
68d6af077f
Registerize based on full liveness analysis.
...
This does a much-more-expensive but much-more-thorough registerization
pass based a live-variable analysis for each function.
azakai: perform this on -O3 and above
2014-01-21 17:50:46 -08:00
7a8c26b65f
Minify loop labels while we're minifying local names.
2014-01-21 17:43:09 +11:00
dd4745db3b
Fix a param-handling bug in normalizeAsm
2014-01-20 21:50:09 +11:00
dd7771e94e
handle unsigned reads in safe heap
2014-01-17 16:12:09 -08:00
d0342ae398
use double for float arguments to safe heap
2014-01-17 10:19:47 -08:00
a310827131
enable a form of safe heap in asm, using js optimizer pass to ensure full coverage and support for fastcomp
2014-01-16 11:27:39 -08:00
e14e6ea75c
add todo
2014-01-15 21:35:30 -08:00
7600935358
Merge pull request #2016 from rfk/rfk/minify-names-separately
...
Split name-minification into a separate pass from registerization.
2014-01-15 21:33:17 -08:00
375761d4a5
Address review comments.
2014-01-16 16:19:22 +11:00
c7cb560f93
do not track eliminable variables from one switch statement to another, if they have any dependencies; fixes #2003
2014-01-15 21:06:50 -08:00
ba54ecbc93
Split name-minification into a separate pass from registerization.
2014-01-16 01:02:43 +11:00
60a86d0fcb
fix aggressiveVariableElimination bug and add testing
2014-01-14 20:54:14 -08:00
ed095a832d
do minified name generation in js-optimizer.js, so that we have no more fixed limit on the # of minified names
2014-01-08 14:07:46 -08:00
e4e8063f6d
handle variables that assign a value including themselves in aggressiveVariableElimination
2014-01-07 14:05:53 -08:00
582e3e62db
make aggressiveVariableElimination usable through a setting
2014-01-07 12:24:40 -08:00
9e1c9c1577
handle if-chains without an else in the outliner
2013-12-27 12:52:48 -08:00
59051ea514
remove outdated comment
2013-12-24 10:56:36 -08:00
508d4f0241
make eliminator not get confused by calls without a return but with a coercion
2013-12-23 18:42:28 -08:00
5f00b61350
don't get confused by temporary empty nodes when calculating assigns in the loop variable optimizer
2013-12-23 17:52:54 -08:00
b86abc2ecf
fix bug where close-together tempDoublePtr operations could cross each other
2013-12-23 15:43:01 -08:00
8825277bf3
comment on how tempDoublePtr is optimized
2013-12-22 21:29:26 -08:00
47e803a625
Optimize away masks on stores to HEAP8 and HEAP16.
...
Optimize HEAP8[i] = x & 255 to HEAP8[i] = x, and similar for HEAP16.
2013-12-19 15:28:27 -08:00
e238b7e11f
minify in asm arg coercions that denormalizeAsm generates, to handle Math.fround calls
2013-11-11 10:23:40 -08:00
eb8172ed78
minify in asm var definitions that denormalizeAsm generates, to handle Math.fround calls
2013-11-11 10:06:29 -08:00
d420405180
Math.* have no side effects
2013-11-09 09:55:18 -08:00
16a5d12dc1
consolidate side effect checking code
2013-11-09 09:29:34 -08:00
e203c5b86c
remove unneeded var
2013-11-09 09:22:24 -08:00
7f5c2dec95
optimize hasSideEffects
2013-11-09 09:21:25 -08:00
8d20a1424a
fix float/int bitcast optimization to recognize and preserve Math_fround
2013-11-08 18:54:13 -08:00
eb3ec20f8b
fix registerize pass for float32
2013-11-08 18:18:12 -08:00
d041b0fe9f
initial work for float32 support in js optimizer
2013-11-08 18:08:42 -08:00
c715cec68f
invalidate calls on conditionals in js optimizer, to avoid eliminating into a non-executing code path
2013-11-08 14:54:18 -08:00
028a3210fc
avoid X-(-Y) in js optimizer; fixes #1685
2013-10-09 13:22:24 -04:00
800ec1f87b
look for metadata in js optimizer starting at the end
2013-10-06 10:33:50 -07:00
8b76d21dae
update js optimizer tests
2013-10-03 10:53:57 -07:00
c182633f18
emit final missing returns in compiler itself
2013-10-03 10:53:57 -07:00
db47cbe4a2
assert on vars being in front of functions straight out of the js compiler
2013-10-02 17:31:37 -07:00
6e8ba9140b
optimize Math.x to Math_x
2013-10-02 17:31:37 -07:00
f2965d6c36
assert in normalizeAsm
2013-10-02 17:31:36 -07:00
e48b5037b6
abort registerize at the proper early time when we need to
2013-09-11 21:33:56 -07:00
b805225b2a
disable registerize when seeing inline js in non-asm mode, as we do not protect it from replacements there; fixes #1614
2013-09-11 21:03:35 -07:00
139faf0677
fix outlining log message
2013-09-07 13:49:37 -07:00
490d55b399
properly outline the while (1) switch pattern
2013-09-06 21:56:41 -07:00
Ryan Kelly
Alon Zakai
Dan Gohman