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

sever lingering dependencies on simd from compiler

db4
Joe Groff 2009-11-11 16:08:40 -06:00
parent 324889e2ac
commit e323071c44
10 changed files with 135 additions and 118 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -7,7 +7,6 @@ compiler.cfg.intrinsics.alien
compiler.cfg.intrinsics.allot
compiler.cfg.intrinsics.fixnum
compiler.cfg.intrinsics.float
compiler.cfg.intrinsics.simd
compiler.cfg.intrinsics.slots
compiler.cfg.intrinsics.misc
compiler.cfg.comparisons ;
@ -23,7 +22,6 @@ QUALIFIED: classes.tuple.private
QUALIFIED: math.private
QUALIFIED: math.integers.private
QUALIFIED: math.floats.private
QUALIFIED: math.vectors.simd.intrinsics
QUALIFIED: math.libm
IN: compiler.cfg.intrinsics

93
basis/compiler/cfg/value-numbering/rewrite/rewrite.factor
View File

@ -2,7 +2,7 @@
! See http://factorcode.org/license.txt for BSD license.
USING: accessors combinators combinators.short-circuit arrays
fry kernel layouts math namespaces sequences cpu.architecture
math.bitwise math.order math.vectors.simd.intrinsics classes
math.bitwise math.order classes
vectors locals make alien.c-types io.binary grouping
compiler.cfg
compiler.cfg.registers
@ -463,100 +463,9 @@ M: ##alien-signed-2 rewrite rewrite-alien-addressing ;
M: ##alien-signed-4 rewrite rewrite-alien-addressing ;
M: ##alien-float rewrite rewrite-alien-addressing ;
M: ##alien-double rewrite rewrite-alien-addressing ;
M: ##alien-vector rewrite rewrite-alien-addressing ;
M: ##set-alien-integer-1 rewrite rewrite-alien-addressing ;
M: ##set-alien-integer-2 rewrite rewrite-alien-addressing ;
M: ##set-alien-integer-4 rewrite rewrite-alien-addressing ;
M: ##set-alien-float rewrite rewrite-alien-addressing ;
M: ##set-alien-double rewrite rewrite-alien-addressing ;
M: ##set-alien-vector rewrite rewrite-alien-addressing ;
! Some lame constant folding for SIMD intrinsics. Eventually this
! should be redone completely.
: rewrite-shuffle-vector-imm ( insn expr -- insn' )
2dup [ rep>> ] bi@ eq? [
[ [ dst>> ] [ src>> vn>vreg ] bi* ]
[ [ shuffle>> ] bi@ nths ]
[ drop rep>> ]
2tri \ ##shuffle-vector-imm new-insn
] [ 2drop f ] if ;
: (fold-shuffle-vector-imm) ( shuffle bytes -- bytes' )
2dup length swap length /i group nths concat ;
: fold-shuffle-vector-imm ( insn expr -- insn' )
[ [ dst>> ] [ shuffle>> ] bi ] dip value>>
(fold-shuffle-vector-imm) \ ##load-constant new-insn ;
M: ##shuffle-vector-imm rewrite
dup src>> vreg>expr {
{ [ dup shuffle-vector-imm-expr? ] [ rewrite-shuffle-vector-imm ] }
{ [ dup reference-expr? ] [ fold-shuffle-vector-imm ] }
{ [ dup constant-expr? ] [ fold-shuffle-vector-imm ] }
[ 2drop f ]
} cond ;
: (fold-scalar>vector) ( insn bytes -- insn' )
[ [ dst>> ] [ rep>> rep-components ] bi ] dip <repetition> concat
\ ##load-constant new-insn ;
: fold-scalar>vector ( insn expr -- insn' )
value>> over rep>> {
{ float-4-rep [ float>bits 4 >le (fold-scalar>vector) ] }
{ double-2-rep [ double>bits 8 >le (fold-scalar>vector) ] }
[ [ untag-fixnum ] dip rep-component-type heap-size >le (fold-scalar>vector) ]
} case ;
M: ##scalar>vector rewrite
dup src>> vreg>expr dup constant-expr?
[ fold-scalar>vector ] [ 2drop f ] if ;
M: ##xor-vector rewrite
dup [ src1>> vreg>vn ] [ src2>> vreg>vn ] bi eq?
[ [ dst>> ] [ rep>> ] bi \ ##zero-vector new-insn ] [ drop f ] if ;
: vector-not? ( expr -- ? )
{
[ not-vector-expr? ]
[ {
[ xor-vector-expr? ]
[ [ src1>> ] [ src2>> ] bi [ vn>expr fill-vector-expr? ] either? ]
} 1&& ]
} 1|| ;
GENERIC: vector-not-src ( expr -- vreg )
M: not-vector-expr vector-not-src src>> vn>vreg ;
M: xor-vector-expr vector-not-src
dup src1>> vn>expr fill-vector-expr? [ src2>> ] [ src1>> ] if vn>vreg ;
M: ##and-vector rewrite
{
{ [ dup src1>> vreg>expr vector-not? ] [
{
[ dst>> ]
[ src1>> vreg>expr vector-not-src ]
[ src2>> ]
[ rep>> ]
} cleave \ ##andn-vector new-insn
] }
{ [ dup src2>> vreg>expr vector-not? ] [
{
[ dst>> ]
[ src2>> vreg>expr vector-not-src ]
[ src1>> ]
[ rep>> ]
} cleave \ ##andn-vector new-insn
] }
[ drop f ]
} cond ;
M: ##andn-vector rewrite
dup src1>> vreg>expr vector-not? [
{
[ dst>> ]
[ src1>> vreg>expr vector-not-src ]
[ src2>> ]
[ rep>> ]
} cleave \ ##and-vector new-insn
] [ drop f ] if ;

120
basis/compiler/cfg/value-numbering/simd/simd.factor Normal file
View File

@ -0,0 +1,120 @@
! Copyright (C) 2008, 2009 Slava Pestov.
! See http://factorcode.org/license.txt for BSD license.
USING: accessors combinators combinators.short-circuit arrays
fry kernel layouts math namespaces sequences cpu.architecture
math.bitwise math.order classes
vectors locals make alien.c-types io.binary grouping
math.vectors.simd
compiler.cfg
compiler.cfg.registers
compiler.cfg.comparisons
compiler.cfg.instructions
compiler.cfg.value-numbering.expressions
compiler.cfg.value-numbering.graph
compiler.cfg.value-numbering.rewrite
compiler.cfg.value-numbering.simplify ;
IN: compiler.cfg.value-numbering.simd
M: ##alien-vector rewrite rewrite-alien-addressing ;
M: ##set-alien-vector rewrite rewrite-alien-addressing ;
! Some lame constant folding for SIMD intrinsics. Eventually this
! should be redone completely.
: rewrite-shuffle-vector-imm ( insn expr -- insn' )
2dup [ rep>> ] bi@ eq? [
[ [ dst>> ] [ src>> vn>vreg ] bi* ]
[ [ shuffle>> ] bi@ nths ]
[ drop rep>> ]
2tri \ ##shuffle-vector-imm new-insn
] [ 2drop f ] if ;
: (fold-shuffle-vector-imm) ( shuffle bytes -- bytes' )
2dup length swap length /i group nths concat ;
: fold-shuffle-vector-imm ( insn expr -- insn' )
[ [ dst>> ] [ shuffle>> ] bi ] dip value>>
(fold-shuffle-vector-imm) \ ##load-constant new-insn ;
M: ##shuffle-vector-imm rewrite
dup src>> vreg>expr {
{ [ dup shuffle-vector-imm-expr? ] [ rewrite-shuffle-vector-imm ] }
{ [ dup reference-expr? ] [ fold-shuffle-vector-imm ] }
{ [ dup constant-expr? ] [ fold-shuffle-vector-imm ] }
[ 2drop f ]
} cond ;
: (fold-scalar>vector) ( insn bytes -- insn' )
[ [ dst>> ] [ rep>> rep-length ] bi ] dip <repetition> concat
\ ##load-constant new-insn ;
: fold-scalar>vector ( insn expr -- insn' )
value>> over rep>> {
{ float-4-rep [ float>bits 4 >le (fold-scalar>vector) ] }
{ double-2-rep [ double>bits 8 >le (fold-scalar>vector) ] }
[ [ untag-fixnum ] dip rep-component-type heap-size >le (fold-scalar>vector) ]
} case ;
M: ##scalar>vector rewrite
dup src>> vreg>expr dup constant-expr?
[ fold-scalar>vector ] [ 2drop f ] if ;
M: ##xor-vector rewrite
dup [ src1>> vreg>vn ] [ src2>> vreg>vn ] bi eq?
[ [ dst>> ] [ rep>> ] bi \ ##zero-vector new-insn ] [ drop f ] if ;
: vector-not? ( expr -- ? )
{
[ not-vector-expr? ]
[ {
[ xor-vector-expr? ]
[ [ src1>> ] [ src2>> ] bi [ vn>expr fill-vector-expr? ] either? ]
} 1&& ]
} 1|| ;
GENERIC: vector-not-src ( expr -- vreg )
M: not-vector-expr vector-not-src src>> vn>vreg ;
M: xor-vector-expr vector-not-src
dup src1>> vn>expr fill-vector-expr? [ src2>> ] [ src1>> ] if vn>vreg ;
M: ##and-vector rewrite
{
{ [ dup src1>> vreg>expr vector-not? ] [
{
[ dst>> ]
[ src1>> vreg>expr vector-not-src ]
[ src2>> ]
[ rep>> ]
} cleave \ ##andn-vector new-insn
] }
{ [ dup src2>> vreg>expr vector-not? ] [
{
[ dst>> ]
[ src2>> vreg>expr vector-not-src ]
[ src1>> ]
[ rep>> ]
} cleave \ ##andn-vector new-insn
] }
[ drop f ]
} cond ;
M: ##andn-vector rewrite
dup src1>> vreg>expr vector-not? [
{
[ dst>> ]
[ src1>> vreg>expr vector-not-src ]
[ src2>> ]
[ rep>> ]
} cleave \ ##and-vector new-insn
] [ drop f ] if ;
M: scalar>vector-expr simplify*
src>> vn>expr {
{ [ dup vector>scalar-expr? ] [ src>> ] }
[ drop f ]
} cond ;
M: shuffle-vector-imm-expr simplify*
[ src>> ] [ shuffle>> ] [ rep>> rep-length iota ] tri
sequence= [ drop f ] unless ;

12
basis/compiler/cfg/value-numbering/simplify/simplify.factor
View File

@ -1,7 +1,7 @@
! Copyright (C) 2008, 2009 Slava Pestov.
! See http://factorcode.org/license.txt for BSD license.
USING: kernel accessors combinators classes math layouts
sequences math.vectors.simd.intrinsics
sequences
compiler.cfg.instructions
compiler.cfg.value-numbering.graph
compiler.cfg.value-numbering.expressions ;
@ -130,16 +130,6 @@ M: box-displaced-alien-expr simplify*
[ 2drop f ]
} cond ;
M: scalar>vector-expr simplify*
src>> vn>expr {
{ [ dup vector>scalar-expr? ] [ src>> ] }
[ drop f ]
} cond ;
M: shuffle-vector-imm-expr simplify*
[ src>> ] [ shuffle>> ] [ rep>> rep-components iota ] tri
sequence= [ drop f ] unless ;
M: expr simplify* drop f ;
: simplify ( expr -- vn )

2
basis/compiler/cfg/value-numbering/value-numbering-tests.factor
View File

@ -4,7 +4,7 @@ cpu.architecture tools.test kernel math combinators.short-circuit
accessors sequences compiler.cfg.predecessors locals compiler.cfg.dce
compiler.cfg.ssa.destruction compiler.cfg.loop-detection
compiler.cfg.representations compiler.cfg assocs vectors arrays
layouts literals namespaces alien ;
layouts literals namespaces alien compiler.cfg.value-numbering.simd ;
IN: compiler.cfg.value-numbering.tests
: trim-temps ( insns -- insns )

3
basis/math/vectors/simd/mirrors/mirrors.factor Normal file
View File

@ -0,0 +1,3 @@
USING: math.vectors.simd mirrors ;
IN: math.vectors.simd.mirrors
INSTANCE: simd-128 enumerated-sequence

7
basis/math/vectors/simd/simd.factor
View File

@ -320,7 +320,7 @@ M: simd-128 n+v [ simd-with ] keep v+ ; inline
M: simd-128 n-v [ simd-with ] keep v- ; inline
M: simd-128 n*v [ simd-with ] keep v* ; inline
M: simd-128 n/v [ simd-with ] keep v/ ; inline
M: simd-128 v+n over simd-with v+
M: simd-128 v+n over simd-with v+ ; inline
M: simd-128 v-n over simd-with v- ; inline
M: simd-128 v*n over simd-with v* ; inline
M: simd-128 v/n over simd-with v/ ; inline
@ -336,3 +336,8 @@ M: simd-128 vshuffle ( u perm -- v )
"compiler.tree.propagation.simd" require
"compiler.cfg.intrinsics.simd" require
"compiler.cfg.value-numbering.simd" require
"mirrors" vocab [
"math.vectors.simd.mirrors" require
] when

3
basis/math/vectors/vectors.factor
View File

@ -97,7 +97,7 @@ M: object vbitor [ bitor ] 2map ;
GENERIC: vbitxor ( u v -- w )
M: object vbitxor [ bitxor ] 2map ;
GENERIC: vbitnot ( u -- w )
M: object vbitnot [ bitnot ] 2map ;
M: object vbitnot [ bitnot ] map ;
GENERIC# vbroadcast 1 ( u n -- v )
M:: object vbroadcast ( u n -- v ) u length n u nth <repetition> u like ;
@ -133,6 +133,7 @@ GENERIC: (vmerge-tail) ( u v -- t )
M: object (vmerge-tail) over length 2 /i '[ _ tail-slice ] bi@ [ zip ] keep concat-as ;
GENERIC: (vmerge) ( u v -- h t )
M: object (vmerge)
[ (vmerge-head) ] [ (vmerge-tail) ] 2bi ; inline
GENERIC: vmerge ( u v -- w )

2
basis/specialized-arrays/mirrors/mirrors.factor
View File

@ -4,5 +4,3 @@ USING: mirrors specialized-arrays math.vectors ;
IN: specialized-arrays.mirrors
INSTANCE: specialized-array enumerated-sequence
INSTANCE: simd-128 enumerated-sequence
INSTANCE: simd-256 enumerated-sequence

9
basis/specialized-arrays/specialized-arrays.factor
View File

@ -2,8 +2,7 @@
! See http://factorcode.org/license.txt for BSD license.
USING: accessors alien alien.c-types alien.data alien.parser
assocs byte-arrays classes compiler.units functors kernel lexer
libc math math.vectors math.vectors.private
math.vectors.specialization namespaces
libc math math.vectors math.vectors.private namespaces
parser prettyprint.custom sequences sequences.private strings
summary vocabs vocabs.loader vocabs.parser vocabs.generated
words fry combinators make ;
@ -69,8 +68,6 @@ TUPLE: A
[ drop \ T bad-byte-array-length ] unless
<direct-A> ; inline
M: A new-underlying drop byte-array>A ;
M: A clone [ underlying>> clone ] [ length>> ] bi <direct-A> ; inline
M: A length length>> ; inline
@ -96,8 +93,6 @@ M: A resize
M: A byte-length length \ T heap-size * ; inline
M: A element-type drop \ T ; inline
M: A direct-array-syntax drop \ A@ ;
M: A pprint-delims drop \ A{ \ } ;
@ -109,8 +104,6 @@ SYNTAX: A@ scan-object scan-object <direct-A> suffix! ;
INSTANCE: A specialized-array
A T c-type-boxed-class f specialize-vector-words
;FUNCTOR
GENERIC: (underlying-type) ( c-type -- c-type' )