1 ! Copyright (C) 2009 Slava Pestov.
2 ! See http://factorcode.org/license.txt for BSD license.
3 USING: alien alien.c-types alien.data assocs combinators
4 cpu.architecture compiler.cfg.comparisons fry generalizations
5 kernel libc macros math
6 math.vectors.conversion.backend
7 sequences sets effects accessors namespaces
8 lexer parser vocabs.parser words arrays math.vectors ;
9 IN: math.vectors.simd.intrinsics
11 ERROR: bad-simd-call ;
15 : simd-effect ( word -- effect )
16 stack-effect [ in>> "rep" suffix ] [ out>> ] bi <effect> ;
17 : simd-conversion-effect ( word -- effect )
18 stack-effect [ in>> but-last "rep" suffix ] [ out>> ] bi <effect> ;
22 V{ } clone simd-ops set-global
24 : (SIMD-OP:) ( accum quot -- accum )
26 scan-word dup name>> "(simd-" ")" surround create-in
27 [ nip [ bad-simd-call ] define ]
29 '[ _ dip set-stack-effect ]
30 [ 2array simd-ops get push ]
34 [ simd-effect ] (SIMD-OP:) ;
36 SYNTAX: SIMD-CONVERSION-OP:
37 [ simd-conversion-effect ] (SIMD-OP:) ;
71 SIMD-OP: (vmerge-head)
72 SIMD-OP: (vmerge-tail)
83 SIMD-CONVERSION-OP: (v>float)
84 SIMD-CONVERSION-OP: (v>integer)
85 SIMD-CONVERSION-OP: (vpack-signed)
86 SIMD-CONVERSION-OP: (vpack-unsigned)
87 SIMD-CONVERSION-OP: (vunpack-head)
88 SIMD-CONVERSION-OP: (vunpack-tail)
90 : (simd-with) ( x rep -- v ) bad-simd-call ;
91 : (simd-gather-2) ( a b rep -- v ) bad-simd-call ;
92 : (simd-gather-4) ( a b c d rep -- v ) bad-simd-call ;
93 : (simd-select) ( v n rep -- x ) bad-simd-call ;
95 : assert-positive ( x -- y ) ;
97 : alien-vector ( c-ptr n rep -- value )
98 ! Inefficient version for when intrinsics are missing
99 [ swap <displaced-alien> ] dip rep-size memory>byte-array ;
101 : set-alien-vector ( value c-ptr n rep -- )
102 ! Inefficient version for when intrinsics are missing
103 [ swap <displaced-alien> swap ] dip rep-size memcpy ;
107 : rep-components ( rep -- n )
108 16 swap rep-component-type heap-size /i ; foldable
110 : rep-coercer ( rep -- quot )
112 { [ dup int-vector-rep? ] [ [ >fixnum ] ] }
113 { [ dup float-vector-rep? ] [ [ >float ] ] }
114 } cond nip ; foldable
116 : rep-coerce ( value rep -- value' )
117 rep-coercer call( value -- value' ) ; inline
119 CONSTANT: rep-gather-words
121 { 2 (simd-gather-2) }
122 { 4 (simd-gather-4) }
125 : rep-gather-word ( rep -- word )
126 rep-components rep-gather-words at ;
130 MACRO: (simd-boa) ( rep -- quot )
137 '[ _ _ napply _ _ execute ] ;
139 GENERIC# supported-simd-op? 1 ( rep intrinsic -- ? )
141 : (%unpack-reps) ( -- reps )
142 %merge-vector-reps [ int-vector-rep? ] filter
143 %unpack-vector-head-reps union ;
145 : (%abs-reps) ( -- reps )
146 cc> %compare-vector-reps [ int-vector-rep? ] filter
147 %xor-vector-reps [ float-vector-rep? ] filter
149 { uchar-16-rep ushort-8-rep uint-4-rep ulonglong-2-rep } union ;
151 M: vector-rep supported-simd-op?
153 { \ (simd-v+) [ %add-vector-reps ] }
154 { \ (simd-vs+) [ %saturated-add-vector-reps ] }
155 { \ (simd-v+-) [ %add-sub-vector-reps ] }
156 { \ (simd-v-) [ %sub-vector-reps ] }
157 { \ (simd-vs-) [ %saturated-sub-vector-reps ] }
158 { \ (simd-vneg) [ %sub-vector-reps ] }
159 { \ (simd-v*) [ %mul-vector-reps ] }
160 { \ (simd-vs*) [ %saturated-mul-vector-reps ] }
161 { \ (simd-v/) [ %div-vector-reps ] }
162 { \ (simd-vmin) [ %min-vector-reps ] }
163 { \ (simd-vmax) [ %max-vector-reps ] }
164 { \ (simd-v.) [ %dot-vector-reps ] }
165 { \ (simd-vsqrt) [ %sqrt-vector-reps ] }
166 { \ (simd-sum) [ %horizontal-add-vector-reps ] }
167 { \ (simd-vabs) [ (%abs-reps) ] }
168 { \ (simd-vbitand) [ %and-vector-reps ] }
169 { \ (simd-vbitandn) [ %andn-vector-reps ] }
170 { \ (simd-vbitor) [ %or-vector-reps ] }
171 { \ (simd-vbitxor) [ %xor-vector-reps ] }
172 { \ (simd-vbitnot) [ %xor-vector-reps ] }
173 { \ (simd-vand) [ %and-vector-reps ] }
174 { \ (simd-vandn) [ %andn-vector-reps ] }
175 { \ (simd-vor) [ %or-vector-reps ] }
176 { \ (simd-vxor) [ %xor-vector-reps ] }
177 { \ (simd-vnot) [ %xor-vector-reps ] }
178 { \ (simd-vlshift) [ %shl-vector-reps ] }
179 { \ (simd-vrshift) [ %shr-vector-reps ] }
180 { \ (simd-hlshift) [ %horizontal-shl-vector-reps ] }
181 { \ (simd-hrshift) [ %horizontal-shr-vector-reps ] }
182 { \ (simd-vshuffle) [ %shuffle-vector-reps ] }
183 { \ (simd-(vmerge-head)) [ %merge-vector-reps ] }
184 { \ (simd-(vmerge-tail)) [ %merge-vector-reps ] }
185 { \ (simd-(v>float)) [ %integer>float-vector-reps ] }
186 { \ (simd-(v>integer)) [ %float>integer-vector-reps ] }
187 { \ (simd-(vpack-signed)) [ %signed-pack-vector-reps ] }
188 { \ (simd-(vpack-unsigned)) [ %unsigned-pack-vector-reps ] }
189 { \ (simd-(vunpack-head)) [ (%unpack-reps) ] }
190 { \ (simd-(vunpack-tail)) [ (%unpack-reps) ] }
191 { \ (simd-v<=) [ cc<= %compare-vector-reps ] }
192 { \ (simd-v<) [ cc< %compare-vector-reps ] }
193 { \ (simd-v=) [ cc= %compare-vector-reps ] }
194 { \ (simd-v>) [ cc> %compare-vector-reps ] }
195 { \ (simd-v>=) [ cc>= %compare-vector-reps ] }
196 { \ (simd-vunordered?) [ cc/<>= %compare-vector-reps ] }
197 { \ (simd-gather-2) [ %gather-vector-2-reps ] }
198 { \ (simd-gather-4) [ %gather-vector-4-reps ] }
199 { \ (simd-vany?) [ %test-vector-reps ] }
200 { \ (simd-vall?) [ %test-vector-reps ] }
201 { \ (simd-vnone?) [ %test-vector-reps ] }