Doug Coleman
82992f6dd4
reverse-here -> reverse!
2009-10-28 14:40:15 -05:00
Slava Pestov
43af9b06a4
compiler.cfg.linear-scan.live-intervals: dead-value-error is never thrown anymore
2009-08-30 05:15:18 -05:00
Slava Pestov
f6a836d1e9
compiler.cfg.linear-scan now supports partial sync-points where all registers are spilled; taking advantage of this, there are new trigonometric intrinsics which yield a 2x performance boost on benchmark.struct-arrays and a 25% boost on benchmark.partial-sums
2009-08-30 04:52:01 -05:00
Slava Pestov
88d15dfbea
compiler.cfg: new system to track when results of analyses need to be recomputed (reverse post order, linear order, predecessors, dominance, loops). Passes can now call needs-predecessors, needs-dominance, needs-loops at the beginning, and cfg-changed, predecessors-changd at the end. Linearization order now takes loop nesting into account, and linear scan now uses linearization order instead of RPO.
2009-08-08 20:02:56 -05:00
Slava Pestov
8b887630a0
compiler.cfg: convert code into two-operand form before SSA destruction; SSA destruction now operates on a relaxed SSA form where multiple defs of the same vreg are allowed, but only within a single basic block. This makes linear scan's coalescing redundant, allowing it to be removed completely
2009-08-05 18:57:46 -05:00
Slava Pestov
8677d56466
compiler.cfg.linear-scan: fix case where a register can be made available for only a part of a live interval's lifetime, but there are no more usages after the split location. This case never came up until global stack analysis, at which point it started to be exercised on x86-32
2009-07-30 02:13:30 -05:00
Slava Pestov
bbd4e4af3e
compiler.cfg: Minor optimization. Instructions can now only ever produce a single value; this eliminates 1array constructions and some iterations
2009-07-28 12:29:07 -05:00
Slava Pestov
c355c21de6
Move liveness back to compiler.cfg.liveness, fix a corner-case where it didn't work on some forms of non-SSA code
2009-07-22 06:07:28 -05:00
Slava Pestov
5d9f7b0ed2
compiler.cfg: Major restructuring -- do not compute liveness before local optimization, and instead change local optimizations to be more permissive of undefined values. Now, liveness is only computed once, after phi elimination and before register allocation. This means liveness analysis does not need to take phi nodes into account and can now use the new compiler.cfg.dataflow-analysis framework
2009-07-22 03:08:28 -05:00
Slava Pestov
9d831baa9c
compiler.cfg.linear-scan: algorithmic optimizations
2009-07-18 00:14:39 -05:00
Slava Pestov
c5e00da9f8
compiler.cfg.linear-scan.live-intervals: remove bogus assertion
2009-07-16 02:42:01 -05:00
Slava Pestov
39e660d8f6
compiler.cfg.linear-scan: code cleanup
2009-07-07 03:45:27 -05:00
Slava Pestov
15134e61e0
compiler.cfg.linear-scan: Re-implement spilling, add some additional runtime assertions, simplify assignment pass since it doesn't have to track spill slots anymore; just assume a live value that's not in active or inactive is spilled
2009-07-07 03:28:55 -05:00
Slava Pestov
d6f45a5216
compiler.cfg.linear-scan: minor fixes
2009-07-05 22:51:53 -05:00
Slava Pestov
14bce3b400
compiler.cfg.linear-scan: debugging spilling and resolve pass
2009-07-01 17:41:07 -05:00
Slava Pestov
fff4a1dbbd
compiler.cfg.linear-scan: fix coalescing to take lifetime holes into account
2009-06-23 21:32:51 -05:00
Slava Pestov
941d7a1eeb
Various linear scan fixes
...
insert spill before reload to fix x86-32 regression
inactive splitting: if all inactive intervals' registers are in use, don't fail
fix stack analysis tests
2009-06-13 17:35:40 -05:00
Slava Pestov
d09ece8152
Split up compiler.cfg.linear-scan.allocation into a number of sub-vocabularies; start work on compiler.cfg.linear-scan.resolve; start work on inactive interval splitting
2009-06-11 17:55:14 -05:00
Slava Pestov
ee67b7c9a8
compiler.cfg.linear-scan: re-do interval splitting to operate on live ranges; add inactive set processing
2009-06-04 18:53:02 -05:00
Slava Pestov
9096acea6f
Linear scan: build live ranges
2009-06-03 03:22:27 -05:00
Slava Pestov
968a9bb666
Various codegen improvements:
...
- new-insn word to construct instructions
- cache RPO in the CFG
- re-organize low-level optimizer so that MR is built after register allocation
- register allocation now stores instruction numbers in the instructions themselves
- split defs-vregs into defs-vregs and temp-vregs
2009-05-29 13:11:34 -05:00
Slava Pestov
d11a779fdd
RicE
2008-11-02 03:58:32 -06:00
Slava Pestov
8df1aba71d
Coalescing
2008-11-02 01:49:57 -06:00
Slava Pestov
db4db19cd9
Start working on coalescing
2008-10-28 02:38:37 -07:00
Slava Pestov
ba0ed21e1d
Better error message
2008-10-22 21:58:37 -05:00
Slava Pestov
f092622fac
CFG IR is now pure SSA
2008-10-20 01:56:28 -05:00
Slava Pestov
c0d89b061e
Fixing register allocator prspilling
2008-10-19 01:10:21 -05:00
Slava Pestov
f436fd0c0f
Merging in new codegen
2008-10-07 16:16:50 -05:00