steve
/
factor
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

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

db4
Slava Pestov 2009-08-05 18:57:46 -05:00
parent eb3bd1edea
commit 370f4c081d
13 changed files with 57 additions and 164 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
basis/compiler/cfg/linear-scan/allocation/allocation.factor
View File

@ -3,7 +3,6 @@
USING: accessors assocs heaps kernel namespaces sequences fry math
math.order combinators arrays sorting compiler.utilities
compiler.cfg.linear-scan.live-intervals
compiler.cfg.linear-scan.allocation.coalescing
compiler.cfg.linear-scan.allocation.spilling
compiler.cfg.linear-scan.allocation.splitting
compiler.cfg.linear-scan.allocation.state ;
@ -29,13 +28,11 @@ IN: compiler.cfg.linear-scan.allocation
second 0 = ; inline
: assign-register ( new -- )
dup coalesce? [ coalesce ] [
dup register-status {
{ [ dup no-free-registers? ] [ drop assign-blocked-register ] }
{ [ 2dup register-available? ] [ register-available ] }
[ drop assign-blocked-register ]
} cond
] if ;
dup register-status {
{ [ dup no-free-registers? ] [ drop assign-blocked-register ] }
{ [ 2dup register-available? ] [ register-available ] }
[ drop assign-blocked-register ]
} cond ;
: handle-interval ( live-interval -- )
[

35
basis/compiler/cfg/linear-scan/allocation/coalescing/coalescing.factor
View File

@ -1,35 +0,0 @@
! Copyright (C) 2009 Slava Pestov.
! See http://factorcode.org/license.txt for BSD license.
USING: accessors kernel sequences namespaces assocs fry
combinators.short-circuit
compiler.cfg.linear-scan.live-intervals
compiler.cfg.linear-scan.allocation.state ;
IN: compiler.cfg.linear-scan.allocation.coalescing
: active-interval ( vreg -- live-interval )
dup [ dup active-intervals-for [ vreg>> = ] with find nip ] when ;
: avoids-inactive-intervals? ( live-interval -- ? )
dup vreg>> inactive-intervals-for
[ intervals-intersect? not ] with all? ;
: coalesce? ( live-interval -- ? )
{
[ copy-from>> active-interval ]
[ [ start>> ] [ copy-from>> active-interval end>> ] bi = ]
[ avoids-inactive-intervals? ]
} 1&& ;
: reuse-spill-slot ( old new -- )
[ vreg>> spill-slots get at ] dip '[ _ vreg>> spill-slots get set-at ] when* ;
: reuse-register ( old new -- )
reg>> >>reg drop ;
: (coalesce) ( old new -- )
[ add-active ] [ [ delete-active ] [ add-handled ] bi ] bi* ;
: coalesce ( live-interval -- )
dup copy-from>> active-interval
[ reuse-spill-slot ] [ reuse-register ] [ (coalesce) ] 2tri ;

2
basis/compiler/cfg/linear-scan/allocation/splitting/splitting.factor
View File

@ -45,7 +45,7 @@ ERROR: splitting-atomic-interval ;
f >>spill-to ; inline
: split-after ( after -- after' )
f >>copy-from f >>reg f >>reload-from ; inline
f >>reg f >>reload-from ; inline
:: split-interval ( live-interval n -- before after )
live-interval n check-split

3
basis/compiler/cfg/linear-scan/debugger/debugger.factor
View File

@ -18,9 +18,8 @@ IN: compiler.cfg.linear-scan.debugger
: interval-picture ( interval -- str )
[ uses>> picture ]
[ copy-from>> unparse ]
[ vreg>> unparse ]
tri 3array ;
bi 2array ;
: live-intervals. ( seq -- )
[ interval-picture ] map simple-table. ;

20
basis/compiler/cfg/linear-scan/linear-scan-tests.factor
View File

@ -470,7 +470,6 @@ USING: math.private ;
clone dup [ start>> ] [ end>> ] bi <live-range> 1vector >>ranges
] map ;
! Coalescing interacted badly with splitting
[ ] [
{
T{ live-interval
@ -478,7 +477,6 @@ USING: math.private ;
{ start 14 }
{ end 17 }
{ uses V{ 14 15 16 17 } }
{ copy-from V int-regs 67 }
}
T{ live-interval
{ vreg V int-regs 67 }
@ -503,7 +501,6 @@ USING: math.private ;
{ start 10 }
{ end 18 }
{ uses V{ 10 11 12 18 } }
{ copy-from V int-regs 56 }
}
T{ live-interval
{ vreg V int-regs 60 }
@ -559,7 +556,6 @@ USING: math.private ;
{ start 44 }
{ end 56 }
{ uses V{ 44 45 45 46 56 } }
{ copy-from V int-regs 3686445 }
}
T{ live-interval
{ vreg V int-regs 3686198 }
@ -572,7 +568,6 @@ USING: math.private ;
{ start 46 }
{ end 49 }
{ uses V{ 46 47 47 49 } }
{ copy-from V int-regs 3686449 }
}
T{ live-interval
{ vreg V int-regs 3686196 }
@ -603,7 +598,6 @@ USING: math.private ;
{ start 49 }
{ end 52 }
{ uses V{ 49 50 50 52 } }
{ copy-from V int-regs 3686454 }
}
T{ live-interval
{ vreg V int-regs 3686461 }
@ -622,42 +616,36 @@ USING: math.private ;
{ start 54 }
{ end 76 }
{ uses V{ 54 55 55 76 } }
{ copy-from V int-regs 3686464 }
}
T{ live-interval
{ vreg V int-regs 3686470 }
{ start 58 }
{ end 60 }
{ uses V{ 58 59 59 60 } }
{ copy-from V int-regs 3686469 }
}
T{ live-interval
{ vreg V int-regs 3686469 }
{ start 56 }
{ end 58 }
{ uses V{ 56 57 57 58 } }
{ copy-from V int-regs 3686449 }
}
T{ live-interval
{ vreg V int-regs 3686473 }
{ start 60 }
{ end 62 }
{ uses V{ 60 61 61 62 } }
{ copy-from V int-regs 3686470 }
}
T{ live-interval
{ vreg V int-regs 3686479 }
{ start 62 }
{ end 64 }
{ uses V{ 62 63 63 64 } }
{ copy-from V int-regs 3686473 }
}
T{ live-interval
{ vreg V int-regs 3686735 }
{ start 78 }
{ end 96 }
{ uses V{ 78 79 79 96 } }
{ copy-from V int-regs 3686372 }
}
T{ live-interval
{ vreg V int-regs 3686482 }
@ -688,7 +676,6 @@ USING: math.private ;
{ start 66 }
{ end 75 }
{ uses V{ 66 67 67 75 } }
{ copy-from V int-regs 3686483 }
}
T{ live-interval
{ vreg V int-regs 3687509 }
@ -719,7 +706,6 @@ USING: math.private ;
{ start 69 }
{ end 74 }
{ uses V{ 69 70 70 74 } }
{ copy-from V int-regs 3686491 }
}
T{ live-interval
{ vreg V int-regs 3687778 }
@ -762,7 +748,6 @@ USING: math.private ;
{ start 72 }
{ end 74 }
{ uses V{ 72 73 73 74 } }
{ copy-from V int-regs 3686499 }
}
T{ live-interval
{ vreg V int-regs 3687780 }
@ -877,7 +862,6 @@ USING: math.private ;
{ start 27 }
{ end 30 }
{ uses V{ 27 28 28 30 } }
{ copy-from V int-regs 3686300 }
}
T{ live-interval
{ vreg V int-regs 3686306 }
@ -950,7 +934,6 @@ USING: math.private ;
{ start 243 }
{ end 245 }
{ uses V{ 243 244 244 245 } }
{ copy-from V int-regs 3687845 }
}
T{ live-interval
{ vreg V int-regs 3687850 }
@ -1119,7 +1102,6 @@ USING: math.private ;
{ start 141 }
{ end 143 }
{ uses V{ 141 142 142 143 } }
{ copy-from V int-regs 3687377 }
}
T{ live-interval
{ vreg V int-regs 3687381 }
@ -1174,7 +1156,6 @@ USING: math.private ;
{ start 293 }
{ end 295 }
{ uses V{ 293 294 294 295 } }
{ copy-from V int-regs 3687087 }
}
T{ live-interval
{ vreg V int-regs 3687403 }
@ -1345,7 +1326,6 @@ USING: math.private ;
{ start 78 }
{ end 96 }
{ uses V{ 78 79 96 } }
{ copy-from V int-regs 6372 }
}
T{ live-interval
{ vreg V int-regs 6483 }

12
basis/compiler/cfg/linear-scan/live-intervals/live-intervals.factor
View File

@ -13,8 +13,7 @@ C: <live-range> live-range
TUPLE: live-interval
vreg
reg spill-to reload-from
start end ranges uses
copy-from ;
start end ranges uses ;
GENERIC: covers? ( insn# obj -- ? )
@ -102,15 +101,6 @@ M: vreg-insn compute-live-intervals*
[ [ temp-vregs ] 2dip '[ [ _ ] dip _ handle-temp ] each ]
3tri ;
: record-copy ( insn -- )
[ dst>> live-intervals get at ] [ src>> ] bi >>copy-from drop ;
M: ##copy compute-live-intervals*
[ call-next-method ] [ record-copy ] bi ;
M: ##copy-float compute-live-intervals*
[ call-next-method ] [ record-copy ] bi ;
: handle-live-out ( bb -- )
live-out keys
basic-block get [ block-from ] [ block-to ] bi

7
basis/compiler/cfg/mr/mr.factor
View File

@ -1,12 +1,11 @@
! Copyright (C) 2009 Slava Pestov.
! See http://factorcode.org/license.txt for BSD license.
USING: compiler.cfg.linearization compiler.cfg.two-operand
compiler.cfg.gc-checks compiler.cfg.linear-scan
compiler.cfg.build-stack-frame compiler.cfg.rpo ;
USING: compiler.cfg.linearization compiler.cfg.gc-checks
compiler.cfg.linear-scan compiler.cfg.build-stack-frame
compiler.cfg.rpo ;
IN: compiler.cfg.mr
: build-mr ( cfg -- mr )
convert-two-operand
insert-gc-checks
linear-scan
flatten-cfg

2
basis/compiler/cfg/optimizer/optimizer.factor
View File

@ -11,6 +11,7 @@ compiler.cfg.value-numbering
compiler.cfg.copy-prop
compiler.cfg.dce
compiler.cfg.write-barrier
compiler.cfg.two-operand
compiler.cfg.ssa.destruction
compiler.cfg.empty-blocks
compiler.cfg.predecessors
@ -42,6 +43,7 @@ SYMBOL: check-optimizer?
copy-propagation
eliminate-dead-code
eliminate-write-barriers
convert-two-operand
destruct-ssa
delete-empty-blocks
?check

28
basis/compiler/cfg/ssa/destruction/destruction.factor
View File

@ -58,9 +58,13 @@ SYMBOL: copies
GENERIC: prepare-insn ( insn -- )
M: ##copy prepare-insn
: prepare-copy ( insn -- )
[ dst>> ] [ src>> ] bi 2array copies get push ;
M: ##copy prepare-insn prepare-copy ;
M: ##copy-float prepare-insn prepare-copy ;
M: ##phi prepare-insn
[ dst>> ] [ inputs>> values ] bi
[ eliminate-copy ] with each ;
@ -81,8 +85,10 @@ M: insn prepare-insn drop ;
[ 2drop ] [ eliminate-copy ] if
] assoc-each ;
UNION: copy-insn ##copy ##copy-float ;
: useless-copy? ( ##copy -- ? )
dup ##copy? [ [ dst>> ] [ src>> ] bi eq? ] [ drop f ] if ;
dup copy-insn? [ [ dst>> ] [ src>> ] bi eq? ] [ drop f ] if ;
: perform-renaming ( cfg -- )
leader-map get keys [ dup leader ] H{ } map>assoc renamings set
@ -95,13 +101,11 @@ M: insn prepare-insn drop ;
] each-basic-block ;
: destruct-ssa ( cfg -- cfg' )
dup cfg-has-phis? [
dup construct-cssa
dup compute-defs
dup compute-dominance
compute-ssa-live-sets
dup compute-live-ranges
dup prepare-coalescing
process-copies
dup perform-renaming
] when ;
dup construct-cssa
dup compute-defs
dup compute-dominance
compute-ssa-live-sets
dup compute-live-ranges
dup prepare-coalescing
process-copies
dup perform-renaming ;

8
basis/compiler/cfg/ssa/interference/interference.factor
View File

@ -6,6 +6,11 @@ compiler.cfg.def-use compiler.cfg.dominance
compiler.cfg.ssa.interference.live-ranges ;
IN: compiler.cfg.ssa.interference
! Interference testing using SSA properties. Actually the only SSA property
! used here is that definitions dominate uses; because of this, the input
! is allowed to have multiple definitions of each vreg as long as they're
! all in the same basic block. This is needed because two-operand conversion
! runs before coalescing, which uses SSA interference testing.
<PRIVATE
:: kill-after-def? ( vreg1 vreg2 bb -- ? )
@ -47,9 +52,10 @@ PRIVATE>
[ 2drop 2drop f ]
} cond ;
! Debug this stuff later
<PRIVATE
! Debug this stuff later
: quadratic-test? ( seq1 seq2 -- ? ) [ length ] bi@ + 10 < ;
: quadratic-test ( seq1 seq2 -- ? )

9
basis/compiler/cfg/ssa/interference/live-ranges/live-ranges.factor
View File

@ -11,8 +11,13 @@ IN: compiler.cfg.ssa.interference.live-ranges
SYMBOLS: local-def-indices local-kill-indices ;
: record-def ( n vregs -- )
dup [ local-def-indices get set-at ] [ 2drop ] if ;
: record-def ( n vreg -- )
! We allow multiple defs of a vreg as long as they're
! all in the same basic block
dup [
local-def-indices get 2dup key?
[ 3drop ] [ set-at ] if
] [ 2drop ] if ;
: record-uses ( n vregs -- )
local-kill-indices get '[ _ set-at ] with each ;

22
basis/compiler/cfg/two-operand/two-operand-tests.factor
View File

@ -14,25 +14,3 @@ compiler.cfg.registers cpu.architecture namespaces tools.test ;
T{ ##sub f V int-regs 1 V int-regs 2 V int-regs 3 }
} (convert-two-operand)
] unit-test
[
V{
T{ ##sub f V int-regs 1 V int-regs 1 V int-regs 2 }
}
] [
{
T{ ##sub f V int-regs 1 V int-regs 1 V int-regs 2 }
} (convert-two-operand)
] unit-test
[
V{
T{ ##copy f V int-regs 4 V int-regs 1 }
T{ ##copy f V int-regs 1 V int-regs 2 }
T{ ##sub f V int-regs 1 V int-regs 1 V int-regs 4 }
}
] [
{
T{ ##sub f V int-regs 1 V int-regs 2 V int-regs 1 }
} (convert-two-operand)
] unit-test

60
basis/compiler/cfg/two-operand/two-operand.factor
View File

@ -5,27 +5,17 @@ compiler.cfg.registers compiler.cfg.instructions
compiler.cfg.rpo cpu.architecture ;
IN: compiler.cfg.two-operand
! This pass runs after SSA coalescing and normalizes instructions
! to fit the x86 two-address scheme. Possibilities are:
! 1) x = x op y
! 2) x = y op x
! 3) x = y op z
! In case 1, there is nothing to do.
! In case 2, we convert to
! z = y
! z = z op x
! x = z
! In case 3, we convert to
! This pass runs before SSA coalescing and normalizes instructions
! to fit the x86 two-address scheme. Since the input is in SSA,
! it suffices to convert
!
! x = y op z
!
! to
!
! x = y
! x = x op z
! In case 2 and case 3, linear scan coalescing will eliminate a
! copy if the value y is never used again.
!
! We don't bother with ##add, ##add-imm, ##sub-imm or ##mul-imm
! since x86 has LEA and IMUL instructions which are effectively
! three-operand addition and multiplication, respectively.
@ -59,37 +49,15 @@ GENERIC: convert-two-operand* ( insn -- )
{ double-float-regs [ ##copy-float ] }
} case ; inline
: case-1? ( insn -- ? ) [ dst>> ] [ src1>> ] bi = ; inline
: case-1 ( insn -- ) , ; inline
: case-2? ( insn -- ? ) [ dst>> ] [ src2>> ] bi = ; inline
: case-2 ( insn -- )
dup dst>> reg-class>> next-vreg
[ swap src2>> emit-copy ]
[ drop [ src2>> ] [ src1>> ] bi emit-copy ]
[ >>src2 dup dst>> >>src1 , ]
2tri ; inline
: case-3 ( insn -- )
M: two-operand-insn convert-two-operand*
[ [ dst>> ] [ src1>> ] bi emit-copy ]
[ dup dst>> >>src1 , ]
bi ; inline
M: two-operand-insn convert-two-operand*
{
{ [ dup case-1? ] [ case-1 ] }
{ [ dup case-2? ] [ case-2 ] }
[ case-3 ]
} cond ; inline
bi ;
M: ##not convert-two-operand*
dup [ dst>> ] [ src>> ] bi = [
[ [ dst>> ] [ src>> ] bi ##copy ]
[ dup dst>> >>src ]
bi
] unless , ;
[ [ dst>> ] [ src>> ] bi emit-copy ]
[ dup dst>> >>src , ]
bi ;
M: insn convert-two-operand* , ;