factor/basis/compiler/cfg/value-numbering/value-numbering.factor

! Copyright (C) 2008, 2009 Slava Pestov.
! See http://factorcode.org/license.txt for BSD license.
USING: namespaces assocs kernel accessors
sorting sets sequences arrays
cpu.architecture
sequences.deep
compiler.cfg
compiler.cfg.rpo
compiler.cfg.def-use
compiler.cfg.instructions
compiler.cfg.value-numbering.graph
compiler.cfg.value-numbering.expressions
compiler.cfg.value-numbering.simplify
compiler.cfg.value-numbering.rewrite ;
IN: compiler.cfg.value-numbering

! Local value numbering.

: >copy ( insn -- insn/##copy )
    dup defs-vreg dup vreg>vn vn>vreg
    2dup eq? [ 2drop ] [ any-rep \ ##copy new-insn nip ] if ;

GENERIC: process-instruction ( insn -- insn' )

M: insn process-instruction
    dup rewrite
    [ process-instruction ]
    [ dup defs-vreg [ dup number-values >copy ] when ] ?if ;

M: array process-instruction
    [ process-instruction ] map ;

: value-numbering-step ( insns -- insns' )
    init-value-graph
    [ process-instruction ] map flatten ;

: value-numbering ( cfg -- cfg' )
    [ value-numbering-step ] local-optimization

    cfg-changed predecessors-changed ;
compiler.cfg.value-numbering: factor out value renaming into a separate compiler.cfg.renaming vocabulary 2009-07-10 00:14:26 -04:00			`! Copyright (C) 2008, 2009 Slava Pestov.`
Fixing register allocator prspilling 2008-10-19 02:10:21 -04:00			`! See http://factorcode.org/license.txt for BSD license.`
compiler.cfg.value-numbering: insert ##copy instructions for instructions whose expressions simplify. While subsequent usages are replaced with the instruction computing the simplified vreg locally, global usages may exist of the original instruction. In this case, the ##copy is not dead 2009-07-24 06:30:30 -04:00			`USING: namespaces assocs kernel accessors`
compiler: new inline intrinsic for <displaced-alien> where the inputs have known types; value numbering now eliminates unnecessary allocation of displaced aliens if the result is immediately unboxed again 2009-08-27 01:06:19 -04:00			`sorting sets sequences arrays`
compiler.cfg: virtual registers are integers now, and representations are stored off to the side. Fix bug in representation selection that would manifest if a value was used as a float and a fixnum in different branches; cannot globally unbox float in this case 2009-08-08 05:02:18 -04:00			`cpu.architecture`
compiler: new inline intrinsic for <displaced-alien> where the inputs have known types; value numbering now eliminates unnecessary allocation of displaced aliens if the result is immediately unboxed again 2009-08-27 01:06:19 -04:00			`sequences.deep`
compiler.cfg: Fix regressions from recent changes 2009-07-14 21:43:06 -04:00			`compiler.cfg`
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 04:08:28 -04:00			`compiler.cfg.rpo`
compiler: eliminate boilerplate by centralizing info in declarative INSN: syntax 2009-09-02 07:22:37 -04:00			`compiler.cfg.def-use`
compiler.cfg.value-numbering: insert ##copy instructions for instructions whose expressions simplify. While subsequent usages are replaced with the instruction computing the simplified vreg locally, global usages may exist of the original instruction. In this case, the ##copy is not dead 2009-07-24 06:30:30 -04:00			`compiler.cfg.instructions`
Value numbering 2008-10-22 22:59:07 -04:00			`compiler.cfg.value-numbering.graph`
			`compiler.cfg.value-numbering.expressions`
Improved value numbering 2008-10-23 03:49:26 -04:00			`compiler.cfg.value-numbering.simplify`
			`compiler.cfg.value-numbering.rewrite ;`
Fixing register allocator prspilling 2008-10-19 02:10:21 -04:00			`IN: compiler.cfg.value-numbering`

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 21:02:56 -04:00			`! Local value numbering.`

compiler.cfg: Fixing test failures 2009-07-24 07:08:07 -04:00			`: >copy ( insn -- insn/##copy )`
compiler: eliminate boilerplate by centralizing info in declarative INSN: syntax 2009-09-02 07:22:37 -04:00			`dup defs-vreg dup vreg>vn vn>vreg`
compiler.cfg: virtual registers are integers now, and representations are stored off to the side. Fix bug in representation selection that would manifest if a value was used as a float and a fixnum in different branches; cannot globally unbox float in this case 2009-08-08 05:02:18 -04:00			`2dup eq? [ 2drop ] [ any-rep \ ##copy new-insn nip ] if ;`
compiler.cfg.value-numbering: factor out value renaming into a separate compiler.cfg.renaming vocabulary 2009-07-10 00:14:26 -04:00
compiler.cfg.value-numbering: insert ##copy instructions for instructions whose expressions simplify. While subsequent usages are replaced with the instruction computing the simplified vreg locally, global usages may exist of the original instruction. In this case, the ##copy is not dead 2009-07-24 06:30:30 -04:00			`GENERIC: process-instruction ( insn -- insn' )`

			`M: insn process-instruction`
			`dup rewrite`
compiler: eliminate boilerplate by centralizing info in declarative INSN: syntax 2009-09-02 07:22:37 -04:00			`[ process-instruction ]`
			`[ dup defs-vreg [ dup number-values >copy ] when ] ?if ;`
compiler.cfg.value-numbering: factor out value renaming into a separate compiler.cfg.renaming vocabulary 2009-07-10 00:14:26 -04:00
compiler: new inline intrinsic for <displaced-alien> where the inputs have known types; value numbering now eliminates unnecessary allocation of displaced aliens if the result is immediately unboxed again 2009-08-27 01:06:19 -04:00			`M: array process-instruction`
			`[ process-instruction ] map ;`

New local-optimization combinator removes some boilerplate 2009-05-26 20:56:56 -04:00			`: value-numbering-step ( insns -- insns' )`
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 04:08:28 -04:00			`init-value-graph`
compiler: new inline intrinsic for <displaced-alien> where the inputs have known types; value numbering now eliminates unnecessary allocation of displaced aliens if the result is immediately unboxed again 2009-08-27 01:06:19 -04:00			`[ process-instruction ] map flatten ;`
Refactoring low-level optimizer to support stack analysis pass 2009-05-26 20:31:19 -04:00
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 14:11:34 -04:00			`: value-numbering ( cfg -- cfg' )`
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 21:02:56 -04:00			`[ value-numbering-step ] local-optimization`

			`cfg-changed predecessors-changed ;`