1 ! (c)2009 Slava Pestov, Joe Groff bsd license
2 USING: accessors alien alien.c-types alien.data combinators
3 sequences.cords cpu.architecture fry generalizations grouping
4 kernel libc locals math math.libm math.order math.ranges
5 math.vectors sequences sequences.private specialized-arrays
7 QUALIFIED-WITH: alien.c-types c
9 c:char c:short c:int c:longlong
10 c:uchar c:ushort c:uint c:ulonglong
12 IN: math.vectors.simd.intrinsics
14 : assert-positive ( x -- y ) ;
18 : >bitwise-vector-rep ( rep -- rep' )
20 { float-4-rep [ uint-4-rep ] }
21 { double-2-rep [ ulonglong-2-rep ] }
25 : >uint-vector-rep ( rep -- rep' )
27 { longlong-2-rep [ ulonglong-2-rep ] }
28 { int-4-rep [ uint-4-rep ] }
29 { short-8-rep [ ushort-8-rep ] }
30 { char-16-rep [ uchar-16-rep ] }
34 : >int-vector-rep ( rep -- rep' )
36 { float-4-rep [ int-4-rep ] }
37 { double-2-rep [ longlong-2-rep ] }
40 : >float-vector-rep ( rep -- rep' )
42 { int-4-rep [ float-4-rep ] }
43 { longlong-2-rep [ double-2-rep ] }
46 : [byte>rep-array] ( rep -- class )
48 { char-16-rep [ [ byte-array>char-array ] ] }
49 { uchar-16-rep [ [ byte-array>uchar-array ] ] }
50 { short-8-rep [ [ byte-array>short-array ] ] }
51 { ushort-8-rep [ [ byte-array>ushort-array ] ] }
52 { int-4-rep [ [ byte-array>int-array ] ] }
53 { uint-4-rep [ [ byte-array>uint-array ] ] }
54 { longlong-2-rep [ [ byte-array>longlong-array ] ] }
55 { ulonglong-2-rep [ [ byte-array>ulonglong-array ] ] }
56 { float-4-rep [ [ byte-array>float-array ] ] }
57 { double-2-rep [ [ byte-array>double-array ] ] }
60 : [>rep-array] ( rep -- class )
62 { char-16-rep [ [ >char-array ] ] }
63 { uchar-16-rep [ [ >uchar-array ] ] }
64 { short-8-rep [ [ >short-array ] ] }
65 { ushort-8-rep [ [ >ushort-array ] ] }
66 { int-4-rep [ [ >int-array ] ] }
67 { uint-4-rep [ [ >uint-array ] ] }
68 { longlong-2-rep [ [ >longlong-array ] ] }
69 { ulonglong-2-rep [ [ >ulonglong-array ] ] }
70 { float-4-rep [ [ >float-array ] ] }
71 { double-2-rep [ [ >double-array ] ] }
74 : [<rep-array>] ( rep -- class )
76 { char-16-rep [ [ 16 (char-array) ] ] }
77 { uchar-16-rep [ [ 16 (uchar-array) ] ] }
78 { short-8-rep [ [ 8 (short-array) ] ] }
79 { ushort-8-rep [ [ 8 (ushort-array) ] ] }
80 { int-4-rep [ [ 4 (int-array) ] ] }
81 { uint-4-rep [ [ 4 (uint-array) ] ] }
82 { longlong-2-rep [ [ 2 (longlong-array) ] ] }
83 { ulonglong-2-rep [ [ 2 (ulonglong-array) ] ] }
84 { float-4-rep [ [ 4 (float-array) ] ] }
85 { double-2-rep [ [ 2 (double-array) ] ] }
88 : rep-tf-values ( rep -- t f )
89 float-vector-rep? [ -1 bits>double 0.0 ] [ -1 0 ] if ;
91 : >rep-array ( a rep -- a' )
92 [byte>rep-array] call( a -- a' ) ; inline
93 : 2>rep-array ( a b rep -- a' b' )
94 [byte>rep-array] '[ _ call( a -- a' ) ] bi@ ; inline
95 : <rep-array> ( rep -- a' )
96 [<rep-array>] call( -- a' ) ; inline
98 : components-map ( a rep quot -- c )
99 [ >rep-array ] dip map underlying>> ; inline
100 : components-2map ( a b rep quot -- c )
101 [ 2>rep-array ] dip 2map underlying>> ; inline
102 : components-reduce ( a rep quot -- x )
103 [ >rep-array [ ] ] dip map-reduce ; inline
105 : bitwise-components-map ( a rep quot -- c )
106 [ >bitwise-vector-rep >rep-array ] dip map underlying>> ; inline
107 : bitwise-components-2map ( a b rep quot -- c )
108 [ >bitwise-vector-rep 2>rep-array ] dip 2map underlying>> ; inline
109 : bitwise-components-reduce ( a rep quot -- x )
110 [ >bitwise-vector-rep >rep-array [ ] ] dip map-reduce ; inline
112 :: (vshuffle) ( a elts rep -- c )
113 a rep >rep-array :> a'
114 rep <rep-array> :> c'
116 from rep rep-length 1 - bitand
120 c' underlying>> ; inline
124 : (simd-v+) ( a b rep -- c ) [ + ] components-2map ;
125 : (simd-v-) ( a b rep -- c ) [ - ] components-2map ;
126 : (simd-vneg) ( a rep -- c ) [ neg ] components-map ;
127 :: (simd-v+-) ( a b rep -- c )
128 a b rep 2>rep-array :> ( a' b' )
129 rep <rep-array> :> c'
130 0 rep rep-length 1 - 2 <range> [| n |
131 n a' nth-unsafe n b' nth-unsafe -
134 n 1 + a' nth-unsafe n 1 + b' nth-unsafe +
135 n 1 + c' set-nth-unsafe
138 : (simd-vs+) ( a b rep -- c )
139 dup rep-component-type '[ + _ c-type-clamp ] components-2map ;
140 : (simd-vs-) ( a b rep -- c )
141 dup rep-component-type '[ - _ c-type-clamp ] components-2map ;
142 : (simd-vs*) ( a b rep -- c )
143 dup rep-component-type '[ * _ c-type-clamp ] components-2map ;
144 : (simd-v*) ( a b rep -- c ) [ * ] components-2map ;
145 : (simd-v*high) ( a b rep -- c )
146 dup rep-component-type heap-size -8 * '[ * _ shift ] components-2map ;
147 :: (simd-v*hs+) ( a b rep -- c )
148 rep { char-16-rep uchar-16-rep } member-eq?
149 [ uchar-16-rep char-16-rep ]
150 [ rep rep ] if :> ( a-rep b-rep )
151 b-rep widen-vector-rep signed-rep :> wide-rep
152 wide-rep rep-component-type :> wide-type
153 a a-rep >rep-array 2 <groups> :> a'
154 b b-rep >rep-array 2 <groups> :> b'
157 [ [ second ] bi@ * ] 2bi +
158 wide-type c-type-clamp
159 ] wide-rep <rep-array> 2map-as underlying>> ;
160 : (simd-v/) ( a b rep -- c ) [ / ] components-2map ;
161 : (simd-vavg) ( a b rep -- c )
162 [ + dup integer? [ 1 + -1 shift ] [ 0.5 * ] if ] components-2map ;
163 : (simd-vmin) ( a b rep -- c ) [ min ] components-2map ;
164 : (simd-vmax) ( a b rep -- c ) [ max ] components-2map ;
165 : (simd-v.) ( a b rep -- n )
166 [ 2>rep-array [ [ first ] bi@ * ] 2keep ] keep
167 1 swap rep-length [a,b) [ '[ _ swap nth-unsafe ] bi@ * + ] with with each ;
168 : (simd-vsqrt) ( a rep -- c ) [ fsqrt ] components-map ;
169 : (simd-vsad) ( a b rep -- c ) 2>rep-array [ - abs ] [ + ] 2map-reduce ;
170 : (simd-sum) ( a rep -- n ) [ + ] components-reduce ;
171 : (simd-vabs) ( a rep -- c ) [ abs ] components-map ;
172 : (simd-vbitand) ( a b rep -- c ) [ bitand ] bitwise-components-2map ;
173 : (simd-vbitandn) ( a b rep -- c ) [ [ bitnot ] dip bitand ] bitwise-components-2map ;
174 : (simd-vbitor) ( a b rep -- c ) [ bitor ] bitwise-components-2map ;
175 : (simd-vbitxor) ( a b rep -- c ) [ bitxor ] bitwise-components-2map ;
176 : (simd-vbitnot) ( a rep -- c ) [ bitnot ] bitwise-components-map ;
177 : (simd-vand) ( a b rep -- c ) [ bitand ] bitwise-components-2map ;
178 : (simd-vandn) ( a b rep -- c ) [ [ bitnot ] dip bitand ] bitwise-components-2map ;
179 : (simd-vor) ( a b rep -- c ) [ bitor ] bitwise-components-2map ;
180 : (simd-vxor) ( a b rep -- c ) [ bitxor ] bitwise-components-2map ;
181 : (simd-vnot) ( a rep -- c ) [ bitnot ] bitwise-components-map ;
182 : (simd-vlshift) ( a n rep -- c ) swap '[ _ shift ] bitwise-components-map ;
183 : (simd-vrshift) ( a n rep -- c ) swap '[ _ neg shift ] bitwise-components-map ;
184 : (simd-hlshift) ( a n rep -- c )
185 drop head-slice* 16 0 pad-head ;
186 : (simd-hrshift) ( a n rep -- c )
187 drop tail-slice 16 0 pad-tail ;
188 : (simd-vshuffle-elements) ( a n rep -- c ) [ rep-length 0 pad-tail ] keep (vshuffle) ;
189 : (simd-vshuffle-bytes) ( a b rep -- c ) drop uchar-16-rep (vshuffle) ;
190 :: (simd-vmerge-head) ( a b rep -- c )
191 a b rep 2>rep-array :> ( a' b' )
192 rep <rep-array> :> c'
193 rep rep-length 2 /i iota [| n |
194 n a' nth-unsafe n 2 * c' set-nth-unsafe
195 n b' nth-unsafe n 2 * 1 + c' set-nth-unsafe
198 :: (simd-vmerge-tail) ( a b rep -- c )
199 a b rep 2>rep-array :> ( a' b' )
200 rep <rep-array> :> c'
201 rep rep-length 2 /i :> len
203 n len + a' nth-unsafe n 2 * c' set-nth-unsafe
204 n len + b' nth-unsafe n 2 * 1 + c' set-nth-unsafe
207 : (simd-v<=) ( a b rep -- c )
208 dup rep-tf-values '[ <= _ _ ? ] components-2map ;
209 : (simd-v<) ( a b rep -- c )
210 dup rep-tf-values '[ < _ _ ? ] components-2map ;
211 : (simd-v=) ( a b rep -- c )
212 dup rep-tf-values '[ = _ _ ? ] components-2map ;
213 : (simd-v>) ( a b rep -- c )
214 dup rep-tf-values '[ > _ _ ? ] components-2map ;
215 : (simd-v>=) ( a b rep -- c )
216 dup rep-tf-values '[ >= _ _ ? ] components-2map ;
217 : (simd-vunordered?) ( a b rep -- c )
218 dup rep-tf-values '[ unordered? _ _ ? ] components-2map ;
219 : (simd-vany?) ( a rep -- ? ) [ bitor ] bitwise-components-reduce zero? not ;
220 : (simd-vall?) ( a rep -- ? ) [ bitand ] bitwise-components-reduce zero? not ;
221 : (simd-vnone?) ( a rep -- ? ) [ bitor ] bitwise-components-reduce zero? ;
222 : (simd-v>float) ( a rep -- c )
223 [ >rep-array [ >float ] ] [ >float-vector-rep <rep-array> ] bi map-as underlying>> ;
224 : (simd-v>integer) ( a rep -- c )
225 [ >rep-array [ >integer ] ] [ >int-vector-rep <rep-array> ] bi map-as underlying>> ;
226 : (simd-vpack-signed) ( a b rep -- c )
227 [ 2>rep-array cord-append ]
228 [ narrow-vector-rep [ <rep-array> ] [ rep-component-type ] bi ] bi
229 '[ _ c-type-clamp ] swap map-as underlying>> ;
230 : (simd-vpack-unsigned) ( a b rep -- c )
231 [ 2>rep-array cord-append ]
232 [ narrow-vector-rep >uint-vector-rep [ <rep-array> ] [ rep-component-type ] bi ] bi
233 '[ _ c-type-clamp ] swap map-as underlying>> ;
234 : (simd-vunpack-head) ( a rep -- c )
235 [ >rep-array ] [ widen-vector-rep [ rep-length ] [ [>rep-array] ] bi ] bi
236 [ head-slice ] dip call( a' -- c' ) underlying>> ;
237 : (simd-vunpack-tail) ( a rep -- c )
238 [ >rep-array ] [ widen-vector-rep [ rep-length ] [ [>rep-array] ] bi ] bi
239 [ tail-slice ] dip call( a' -- c' ) underlying>> ;
240 : (simd-with) ( n rep -- v )
241 [ rep-length swap '[ _ ] ] [ <rep-array> ] bi replicate-as
243 : (simd-gather-2) ( m n rep -- v ) <rep-array> [ 2 set-firstn ] keep underlying>> ;
244 : (simd-gather-4) ( m n o p rep -- v ) <rep-array> [ 4 set-firstn ] keep underlying>> ;
245 : (simd-select) ( a n rep -- x ) [ swap ] dip >rep-array nth-unsafe ;
247 : alien-vector ( c-ptr n rep -- value )
248 [ swap <displaced-alien> ] dip rep-size memory>byte-array ;
249 : set-alien-vector ( value c-ptr n rep -- )
250 [ swap <displaced-alien> swap ] dip rep-size memcpy ;
252 "compiler.cfg.intrinsics.simd" require
253 "compiler.tree.propagation.simd" require
254 "compiler.cfg.value-numbering.simd" require