1 ! Copyright (C) 2005, 2010 Slava Pestov.
2 ! See http://factorcode.org/license.txt for BSD license.
3 USING: accessors assocs alien alien.c-types arrays strings
4 cpu.x86.assembler cpu.x86.assembler.private cpu.x86.assembler.operands
5 cpu.x86.features cpu.x86.features.private cpu.architecture kernel
6 kernel.private math memory namespaces make sequences words system
7 layouts combinators math.order math.vectors fry locals compiler.constants
8 byte-arrays io macros quotations classes.algebra compiler
10 compiler.cfg.registers
11 compiler.cfg.instructions
12 compiler.cfg.intrinsics
13 compiler.cfg.comparisons
14 compiler.cfg.stack-frame
15 compiler.codegen.fixup ;
16 QUALIFIED-WITH: alien.c-types c
17 FROM: layouts => cell ;
21 ! Add some methods to the assembler to be more useful to the backend
22 M: label JMP 0 JMP rc-relative label-fixup ;
23 M: label JUMPcc [ 0 ] dip JUMPcc rc-relative label-fixup ;
25 M: x86 vector-regs float-regs ;
27 HOOK: stack-reg cpu ( -- reg )
29 HOOK: frame-reg cpu ( -- reg )
31 HOOK: reserved-stack-space cpu ( -- n )
33 HOOK: extra-stack-space cpu ( stack-frame -- n )
35 : stack@ ( n -- op ) stack-reg swap [+] ;
37 : special-offset ( m -- n )
38 stack-frame get extra-stack-space +
39 reserved-stack-space + ;
41 : special@ ( n -- op ) special-offset stack@ ;
43 : spill@ ( n -- op ) spill-offset special@ ;
45 : param@ ( n -- op ) reserved-stack-space + stack@ ;
47 : gc-root-offsets ( seq -- seq' )
48 [ n>> spill-offset special-offset cell + ] map f like ;
50 : decr-stack-reg ( n -- )
51 dup 0 = [ drop ] [ stack-reg swap SUB ] if ;
53 : incr-stack-reg ( n -- )
54 dup 0 = [ drop ] [ stack-reg swap ADD ] if ;
56 : align-stack ( n -- n' ) 16 align ;
58 M: x86 stack-frame-size ( stack-frame -- i )
59 [ (stack-frame-size) ]
60 [ extra-stack-space ] bi +
61 reserved-stack-space +
65 ! Must be a volatile register not used for parameter passing or
67 HOOK: temp-reg cpu ( -- reg )
69 HOOK: pic-tail-reg cpu ( -- reg )
71 M: x86 complex-addressing? t ;
73 M: x86 fused-unboxing? t ;
75 M: x86 test-instruction? t ;
77 M: x86 immediate-store? immediate-comparand? ;
79 M: x86 %load-immediate dup 0 = [ drop dup XOR ] [ MOV ] if ;
81 M: x86 %load-reference
82 [ swap 0 MOV rc-absolute-cell rel-literal ]
83 [ \ f type-number MOV ]
86 HOOK: ds-reg cpu ( -- reg )
87 HOOK: rs-reg cpu ( -- reg )
89 : reg-stack ( n reg -- op ) swap cells neg [+] ;
91 GENERIC: loc>operand ( loc -- operand )
93 M: ds-loc loc>operand n>> ds-reg reg-stack ;
94 M: rs-loc loc>operand n>> rs-reg reg-stack ;
96 M: x86 %peek loc>operand MOV ;
98 M: x86 %replace loc>operand swap MOV ;
103 { [ dup not ] [ drop \ f type-number MOV ] }
104 { [ dup fixnum? ] [ tag-fixnum MOV ] }
105 [ [ HEX: ffffffff MOV ] dip rc-absolute rel-literal ]
108 : (%inc) ( n reg -- ) swap cells dup 0 > [ ADD ] [ neg SUB ] if ; inline
109 M: x86 %inc-d ( n -- ) ds-reg (%inc) ;
110 M: x86 %inc-r ( n -- ) rs-reg (%inc) ;
112 M: x86 %call ( word -- ) 0 CALL rc-relative rel-word-pic ;
114 : xt-tail-pic-offset ( -- n )
115 #! See the comment in vm/cpu-x86.hpp
118 HOOK: %prepare-jump cpu ( -- )
120 M: x86 %jump ( word -- )
122 0 JMP rc-relative rel-word-pic-tail ;
124 M: x86 %jump-label ( label -- ) 0 JMP rc-relative label-fixup ;
126 M: x86 %return ( -- ) 0 RET ;
128 : (%slot) ( obj slot scale tag -- op ) neg <indirect> ; inline
129 : (%slot-imm) ( obj slot tag -- op ) slot-offset [+] ; inline
131 M: x86 %slot ( dst obj slot scale tag -- ) (%slot) MOV ;
132 M: x86 %slot-imm ( dst obj slot tag -- ) (%slot-imm) MOV ;
133 M: x86 %set-slot ( src obj slot scale tag -- ) (%slot) swap MOV ;
134 M: x86 %set-slot-imm ( src obj slot tag -- ) (%slot-imm) swap MOV ;
136 :: two-operand ( dst src1 src2 rep -- dst src )
137 dst src2 eq? dst src1 eq? not and [ "Cannot handle this case" throw ] when
141 :: one-operand ( dst src rep -- dst )
145 M: x86 %add 2over eq? [ nip ADD ] [ [+] LEA ] if ;
146 M: x86 %add-imm 2over eq? [ nip ADD ] [ [+] LEA ] if ;
147 M: x86 %sub int-rep two-operand SUB ;
148 M: x86 %sub-imm 2over eq? [ nip SUB ] [ neg [+] LEA ] if ;
149 M: x86 %mul int-rep two-operand IMUL2 ;
150 M: x86 %mul-imm IMUL3 ;
151 M: x86 %and int-rep two-operand AND ;
152 M: x86 %and-imm int-rep two-operand AND ;
153 M: x86 %or int-rep two-operand OR ;
154 M: x86 %or-imm int-rep two-operand OR ;
155 M: x86 %xor int-rep two-operand XOR ;
156 M: x86 %xor-imm int-rep two-operand XOR ;
157 M: x86 %shl-imm int-rep two-operand SHL ;
158 M: x86 %shr-imm int-rep two-operand SHR ;
159 M: x86 %sar-imm int-rep two-operand SAR ;
161 M: x86 %min int-rep two-operand [ CMP ] [ CMOVG ] 2bi ;
162 M: x86 %max int-rep two-operand [ CMP ] [ CMOVL ] 2bi ;
164 M: x86 %not int-rep one-operand NOT ;
165 M: x86 %neg int-rep one-operand NEG ;
167 M: x86 %bit-count POPCNT ;
169 ! A bit of logic to avoid using MOVSS/MOVSD for reg-reg moves
170 ! since this induces partial register stalls
171 GENERIC: copy-register* ( dst src rep -- )
172 GENERIC: copy-memory* ( dst src rep -- )
174 M: int-rep copy-register* drop MOV ;
175 M: tagged-rep copy-register* drop MOV ;
176 M: float-rep copy-register* drop MOVAPS ;
177 M: double-rep copy-register* drop MOVAPS ;
178 M: float-4-rep copy-register* drop MOVAPS ;
179 M: double-2-rep copy-register* drop MOVAPS ;
180 M: vector-rep copy-register* drop MOVDQA ;
182 M: object copy-memory* copy-register* ;
183 M: float-rep copy-memory* drop MOVSS ;
184 M: double-rep copy-memory* drop MOVSD ;
186 : ?spill-slot ( obj -- obj ) dup spill-slot? [ n>> spill@ ] when ;
188 M: x86 %copy ( dst src rep -- )
189 2over eq? [ 3drop ] [
190 [ [ ?spill-slot ] bi@ ] dip
191 2over [ register? ] both? [ copy-register* ] [ copy-memory* ] if
194 : fixnum-overflow ( label dst src1 src2 cc quot -- )
195 swap [ [ int-rep two-operand ] dip call ] dip
201 M: x86 %fixnum-add ( label dst src1 src2 cc -- )
202 [ ADD ] fixnum-overflow ;
204 M: x86 %fixnum-sub ( label dst src1 src2 cc -- )
205 [ SUB ] fixnum-overflow ;
207 M: x86 %fixnum-mul ( label dst src1 src2 cc -- )
208 [ IMUL2 ] fixnum-overflow ;
210 M: x86 %unbox-alien ( dst src -- )
211 alien-offset [+] MOV ;
213 M:: x86 %unbox-any-c-ptr ( dst src -- )
218 src \ f type-number CMP
220 ! Compute tag in dst register
223 ! Is the object an alien?
224 dst alien type-number CMP
225 ! Add an offset to start of byte array's data
226 dst src byte-array-offset [+] LEA
228 ! If so, load the offset and add it to the address
229 dst src alien-offset [+] MOV
233 : alien@ ( reg n -- op ) cells alien type-number - [+] ;
235 M:: x86 %box-alien ( dst src temp -- )
238 dst \ f type-number MOV
241 dst 5 cells alien temp %allot
242 dst 1 alien@ \ f type-number MOV ! base
243 dst 2 alien@ \ f type-number MOV ! expired
244 dst 3 alien@ src MOV ! displacement
245 dst 4 alien@ src MOV ! address
249 :: %box-displaced-alien/f ( dst displacement -- )
250 dst 1 alien@ \ f type-number MOV
251 dst 3 alien@ displacement MOV
252 dst 4 alien@ displacement MOV ;
254 :: %box-displaced-alien/alien ( dst displacement base temp -- )
255 ! Set new alien's base to base.base
256 temp base 1 alien@ MOV
257 dst 1 alien@ temp MOV
259 ! Compute displacement
260 temp base 3 alien@ MOV
261 temp displacement ADD
262 dst 3 alien@ temp MOV
265 temp base 4 alien@ MOV
266 temp displacement ADD
267 dst 4 alien@ temp MOV ;
269 :: %box-displaced-alien/byte-array ( dst displacement base temp -- )
270 dst 1 alien@ base MOV
271 dst 3 alien@ displacement MOV
272 temp base displacement byte-array-offset [++] LEA
273 dst 4 alien@ temp MOV ;
275 :: %box-displaced-alien/dynamic ( dst displacement base temp -- )
277 "not-alien" define-label
281 temp tag-mask get AND
284 temp \ f type-number CMP
287 ! Yes, it is f. Fill in new object
288 dst displacement %box-displaced-alien/f
292 "not-f" resolve-label
295 temp alien type-number CMP
298 dst displacement base temp %box-displaced-alien/alien
303 ! Is base a byte array? It has to be, by now...
304 "not-alien" resolve-label
306 dst displacement base temp %box-displaced-alien/byte-array ;
308 M:: x86 %box-displaced-alien ( dst displacement base temp base-class -- )
312 ! If displacement is zero, return the base
314 displacement displacement TEST
317 ! Displacement is non-zero, we're going to be allocating a new
319 dst 5 cells alien temp %allot
322 dst 2 alien@ \ f type-number MOV
324 dst displacement base temp
326 { [ base-class \ f class<= ] [ 2drop %box-displaced-alien/f ] }
327 { [ base-class \ alien class<= ] [ %box-displaced-alien/alien ] }
328 { [ base-class \ byte-array class<= ] [ %box-displaced-alien/byte-array ] }
329 [ %box-displaced-alien/dynamic ]
335 ! The 'small-reg' mess is pretty crappy, but its only used on x86-32.
336 ! On x86-64, all registers have 8-bit versions. However, a similar
337 ! problem arises for shifts, where the shift count must be in CL, and
338 ! so one day I will fix this properly by adding precoloring to the
339 ! register allocator.
341 HOOK: has-small-reg? cpu ( reg size -- ? )
343 CONSTANT: have-byte-regs { EAX ECX EDX EBX }
345 M: x86.32 has-small-reg?
347 { 8 [ have-byte-regs member-eq? ] }
352 M: x86.64 has-small-reg? 2drop t ;
354 : small-reg-that-isn't ( exclude -- reg' )
355 [ have-byte-regs ] dip
356 [ native-version-of ] map
357 '[ _ member-eq? not ] find nip ;
359 : with-save/restore ( reg quot -- )
360 [ drop PUSH ] [ call ] [ drop POP ] 2tri ; inline
362 :: with-small-register ( dst exclude size quot: ( new-dst -- ) -- )
363 ! If the destination register overlaps a small register with
364 ! 'size' bits, we call the quot with that. Otherwise, we find a
365 ! small register that is not in exclude, and call quot, saving and
366 ! restoring the small register.
367 dst size has-small-reg? [ dst quot call ] [
368 exclude small-reg-that-isn't
369 [ quot call ] with-save/restore
372 :: %alien-integer-getter ( dst exclude address bits quot -- )
373 dst exclude bits [| new-dst |
374 new-dst dup bits n-bit-version-of dup address MOV
376 dst new-dst int-rep %copy
377 ] with-small-register ; inline
379 : %alien-unsigned-getter ( dst exclude address bits -- )
380 [ MOVZX ] %alien-integer-getter ; inline
382 : %alien-signed-getter ( dst exclude address bits -- )
383 [ MOVSX ] %alien-integer-getter ; inline
385 :: %alien-integer-setter ( value exclude address bits -- )
386 value exclude bits [| new-value |
387 new-value value int-rep %copy
388 address new-value bits n-bit-version-of MOV
389 ] with-small-register ; inline
391 : (%memory) ( base displacement scale offset rep c-type -- exclude address rep c-type )
392 [ [ [ 2array ] 2keep ] 2dip <indirect> ] 2dip ;
394 : (%memory-imm) ( base offset rep c-type -- exclude address rep c-type )
395 [ [ drop 1array ] [ [+] ] 2bi ] 2dip ;
397 : (%load-memory) ( dst exclude address rep c-type -- )
400 { c:char [ 8 %alien-signed-getter ] }
401 { c:uchar [ 8 %alien-unsigned-getter ] }
402 { c:short [ 16 %alien-signed-getter ] }
403 { c:ushort [ 16 %alien-unsigned-getter ] }
404 { c:int [ 32 %alien-signed-getter ] }
405 { c:uint [ 32 [ 2drop ] %alien-integer-getter ] }
407 ] [ [ drop ] 2dip %copy ] ?if ;
409 M: x86 %load-memory ( dst base displacement scale offset rep c-type -- )
410 (%memory) (%load-memory) ;
412 M: x86 %load-memory-imm ( dst base offset rep c-type -- )
413 (%memory-imm) (%load-memory) ;
415 : (%store-memory) ( src exclude address rep c-type -- )
418 { c:char [ 8 %alien-integer-setter ] }
419 { c:uchar [ 8 %alien-integer-setter ] }
420 { c:short [ 16 %alien-integer-setter ] }
421 { c:ushort [ 16 %alien-integer-setter ] }
422 { c:int [ 32 %alien-integer-setter ] }
423 { c:uint [ 32 %alien-integer-setter ] }
425 ] [ [ nip swap ] dip %copy ] ?if ;
427 M: x86 %store-memory ( src base displacement scale offset rep c-type -- )
428 (%memory) (%store-memory) ;
430 M: x86 %store-memory-imm ( src base offset rep c-type -- )
431 (%memory-imm) (%store-memory) ;
433 : shift-count? ( reg -- ? ) { ECX RCX } member-eq? ;
435 :: emit-shift ( dst src quot -- )
444 ECX native-version-of [
446 drop dst CL quot call
451 M: x86 %shl int-rep two-operand [ SHL ] emit-shift ;
452 M: x86 %shr int-rep two-operand [ SHR ] emit-shift ;
453 M: x86 %sar int-rep two-operand [ SAR ] emit-shift ;
455 HOOK: %mov-vm-ptr cpu ( reg -- )
457 HOOK: %vm-field-ptr cpu ( reg offset -- )
459 : load-zone-offset ( nursery-ptr -- )
460 "nursery" vm-field-offset %vm-field-ptr ;
462 : load-allot-ptr ( nursery-ptr allot-ptr -- )
463 [ drop load-zone-offset ] [ swap [] MOV ] 2bi ;
465 : inc-allot-ptr ( nursery-ptr n -- )
466 [ [] ] dip data-alignment get align ADD ;
468 : store-header ( temp class -- )
469 [ [] ] [ type-number tag-header ] bi* MOV ;
471 : store-tagged ( dst tag -- )
474 M:: x86 %allot ( dst size class nursery-ptr -- )
475 nursery-ptr dst load-allot-ptr
476 dst class store-header
477 dst class store-tagged
478 nursery-ptr size inc-allot-ptr ;
480 HOOK: %mark-card cpu ( card temp -- )
481 HOOK: %mark-deck cpu ( card temp -- )
483 :: (%write-barrier) ( temp1 temp2 -- )
485 temp1 temp2 %mark-card
486 temp1 deck-bits card-bits - SHR
487 temp1 temp2 %mark-deck ;
489 M:: x86 %write-barrier ( src slot scale tag temp1 temp2 -- )
490 temp1 src slot scale tag (%slot) LEA
491 temp1 temp2 (%write-barrier) ;
493 M:: x86 %write-barrier-imm ( src slot tag temp1 temp2 -- )
494 temp1 src slot tag (%slot-imm) LEA
495 temp1 temp2 (%write-barrier) ;
497 M:: x86 %check-nursery-branch ( label size cc temp1 temp2 -- )
498 temp1 load-zone-offset
501 temp2 temp1 2 cells [+] CMP
503 { cc<= [ label JLE ] }
504 { cc/<= [ label JG ] }
507 M: x86 %alien-global ( dst symbol library -- )
508 [ 0 MOV ] 2dip rc-absolute-cell rel-dlsym ;
510 M: x86 %epilogue ( n -- ) cell - incr-stack-reg ;
512 :: (%boolean) ( dst temp insn -- )
513 dst \ f type-number MOV
514 temp 0 MOV \ t rc-absolute-cell rel-literal
515 dst temp insn execute ; inline
517 : %boolean ( dst cc temp -- )
519 { cc< [ \ CMOVL (%boolean) ] }
520 { cc<= [ \ CMOVLE (%boolean) ] }
521 { cc> [ \ CMOVG (%boolean) ] }
522 { cc>= [ \ CMOVGE (%boolean) ] }
523 { cc= [ \ CMOVE (%boolean) ] }
524 { cc/= [ \ CMOVNE (%boolean) ] }
527 M:: x86 %compare ( dst src1 src2 cc temp -- )
529 dst cc temp %boolean ;
531 M:: x86 %test ( dst src1 src2 cc temp -- )
533 dst cc temp %boolean ;
535 : (%compare-tagged) ( src1 src2 -- )
536 [ HEX: ffffffff CMP ] dip rc-absolute rel-literal ;
538 M:: x86 %compare-integer-imm ( dst src1 src2 cc temp -- )
540 dst cc temp %boolean ;
542 M:: x86 %test-imm ( dst src1 src2 cc temp -- )
544 dst cc temp %boolean ;
546 : (%compare-imm) ( src1 src2 -- )
548 { [ dup fixnum? ] [ tag-fixnum CMP ] }
549 { [ dup not ] [ drop \ f type-number CMP ] }
550 [ (%compare-tagged) ]
553 M:: x86 %compare-imm ( dst src1 src2 cc temp -- )
554 src1 src2 (%compare-imm)
555 dst cc temp %boolean ;
557 : %branch ( label cc -- )
567 M:: x86 %compare-branch ( label src1 src2 cc -- )
571 M:: x86 %compare-integer-imm-branch ( label src1 src2 cc -- )
575 M:: x86 %test-branch ( label src1 src2 cc -- )
579 M:: x86 %test-imm-branch ( label src1 src2 cc -- )
583 M:: x86 %compare-imm-branch ( label src1 src2 cc -- )
584 src1 src2 (%compare-imm)
587 M: x86 %add-float double-rep two-operand ADDSD ;
588 M: x86 %sub-float double-rep two-operand SUBSD ;
589 M: x86 %mul-float double-rep two-operand MULSD ;
590 M: x86 %div-float double-rep two-operand DIVSD ;
591 M: x86 %min-float double-rep two-operand MINSD ;
592 M: x86 %max-float double-rep two-operand MAXSD ;
593 M: x86 %sqrt SQRTSD ;
595 : %clear-unless-in-place ( dst src -- )
596 over = [ drop ] [ dup XORPS ] if ;
598 M: x86 %single>double-float [ %clear-unless-in-place ] [ CVTSS2SD ] 2bi ;
599 M: x86 %double>single-float [ %clear-unless-in-place ] [ CVTSD2SS ] 2bi ;
601 M: x86 %integer>float [ drop dup XORPS ] [ CVTSI2SD ] 2bi ;
602 M: x86 %float>integer CVTTSD2SI ;
604 : %cmov-float= ( dst src -- )
606 "no-move" define-label
608 "no-move" get [ JNE ] [ JP ] bi
610 "no-move" resolve-label
613 : %cmov-float/= ( dst src -- )
615 "no-move" define-label
622 "no-move" resolve-label
625 :: (%compare-float) ( dst src1 src2 cc temp compare -- )
627 { cc< [ src2 src1 \ compare execute( a b -- ) dst temp \ CMOVA (%boolean) ] }
628 { cc<= [ src2 src1 \ compare execute( a b -- ) dst temp \ CMOVAE (%boolean) ] }
629 { cc> [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVA (%boolean) ] }
630 { cc>= [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVAE (%boolean) ] }
631 { cc= [ src1 src2 \ compare execute( a b -- ) dst temp \ %cmov-float= (%boolean) ] }
632 { cc<> [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVNE (%boolean) ] }
633 { cc<>= [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVNP (%boolean) ] }
634 { cc/< [ src2 src1 \ compare execute( a b -- ) dst temp \ CMOVBE (%boolean) ] }
635 { cc/<= [ src2 src1 \ compare execute( a b -- ) dst temp \ CMOVB (%boolean) ] }
636 { cc/> [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVBE (%boolean) ] }
637 { cc/>= [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVB (%boolean) ] }
638 { cc/= [ src1 src2 \ compare execute( a b -- ) dst temp \ %cmov-float/= (%boolean) ] }
639 { cc/<> [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVE (%boolean) ] }
640 { cc/<>= [ src1 src2 \ compare execute( a b -- ) dst temp \ CMOVP (%boolean) ] }
643 M: x86 %compare-float-ordered ( dst src1 src2 cc temp -- )
644 \ COMISD (%compare-float) ;
646 M: x86 %compare-float-unordered ( dst src1 src2 cc temp -- )
647 \ UCOMISD (%compare-float) ;
649 : %jump-float= ( label -- )
651 "no-jump" define-label
654 "no-jump" resolve-label
657 : %jump-float/= ( label -- )
660 :: (%compare-float-branch) ( label src1 src2 cc compare -- )
662 { cc< [ src2 src1 \ compare execute( a b -- ) label JA ] }
663 { cc<= [ src2 src1 \ compare execute( a b -- ) label JAE ] }
664 { cc> [ src1 src2 \ compare execute( a b -- ) label JA ] }
665 { cc>= [ src1 src2 \ compare execute( a b -- ) label JAE ] }
666 { cc= [ src1 src2 \ compare execute( a b -- ) label %jump-float= ] }
667 { cc<> [ src1 src2 \ compare execute( a b -- ) label JNE ] }
668 { cc<>= [ src1 src2 \ compare execute( a b -- ) label JNP ] }
669 { cc/< [ src2 src1 \ compare execute( a b -- ) label JBE ] }
670 { cc/<= [ src2 src1 \ compare execute( a b -- ) label JB ] }
671 { cc/> [ src1 src2 \ compare execute( a b -- ) label JBE ] }
672 { cc/>= [ src1 src2 \ compare execute( a b -- ) label JB ] }
673 { cc/= [ src1 src2 \ compare execute( a b -- ) label %jump-float/= ] }
674 { cc/<> [ src1 src2 \ compare execute( a b -- ) label JE ] }
675 { cc/<>= [ src1 src2 \ compare execute( a b -- ) label JP ] }
678 M: x86 %compare-float-ordered-branch ( label src1 src2 cc -- )
679 \ COMISD (%compare-float-branch) ;
681 M: x86 %compare-float-unordered-branch ( label src1 src2 cc -- )
682 \ UCOMISD (%compare-float-branch) ;
684 MACRO: available-reps ( alist -- )
685 ! Each SSE version adds new representations and supports
687 unzip { } [ append ] accumulate rest swap suffix
688 [ [ 1quotation ] map ] bi@ zip
689 reverse [ { } ] suffix
692 M: x86 %alien-vector-reps
694 { sse? { float-4-rep } }
695 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
700 { double-2-rep [ dup XORPS ] }
701 { float-4-rep [ dup XORPS ] }
705 M: x86 %zero-vector-reps
707 { sse? { float-4-rep } }
708 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
713 { double-2-rep [ dup [ XORPS ] [ CMPEQPS ] 2bi ] }
714 { float-4-rep [ dup [ XORPS ] [ CMPEQPS ] 2bi ] }
718 M: x86 %fill-vector-reps
720 { sse? { float-4-rep } }
721 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
724 M:: x86 %gather-vector-4 ( dst src1 src2 src3 src4 rep -- )
727 dst src1 float-4-rep %copy
733 dst src1 int-4-rep %copy
740 M: x86 %gather-vector-4-reps
742 ! Can't do this with sse1 since it will want to unbox
743 ! double-precision floats and convert to single precision
744 { sse2? { float-4-rep int-4-rep uint-4-rep } }
747 M:: x86 %gather-int-vector-4 ( dst src1 src2 src3 src4 rep -- )
749 dst src1 32-bit-version-of 0 PINSRD
750 dst src2 32-bit-version-of 1 PINSRD
751 dst src3 32-bit-version-of 2 PINSRD
752 dst src4 32-bit-version-of 3 PINSRD ;
754 M: x86 %gather-int-vector-4-reps
756 { sse4.1? { int-4-rep uint-4-rep } }
759 M:: x86 %gather-vector-2 ( dst src1 src2 rep -- )
762 dst src1 double-2-rep %copy
766 dst src1 longlong-2-rep %copy
771 M: x86 %gather-vector-2-reps
773 { sse2? { double-2-rep longlong-2-rep ulonglong-2-rep } }
776 M:: x86.64 %gather-int-vector-2 ( dst src1 src2 rep -- )
781 M: x86.64 %gather-int-vector-2-reps
783 { sse4.1? { longlong-2-rep ulonglong-2-rep } }
786 : sse1-float-4-shuffle ( dst shuffle -- )
788 { { 0 1 2 3 } [ drop ] }
789 { { 0 1 0 1 } [ dup MOVLHPS ] }
790 { { 2 3 2 3 } [ dup MOVHLPS ] }
791 { { 0 0 1 1 } [ dup UNPCKLPS ] }
792 { { 2 2 3 3 } [ dup UNPCKHPS ] }
796 : float-4-shuffle ( dst shuffle -- )
799 { { 0 0 2 2 } [ dup MOVSLDUP ] }
800 { { 1 1 3 3 } [ dup MOVSHDUP ] }
801 [ sse1-float-4-shuffle ]
803 ] [ sse1-float-4-shuffle ] if ;
805 : int-4-shuffle ( dst shuffle -- )
807 { { 0 1 2 3 } [ drop ] }
808 { { 0 0 1 1 } [ dup PUNPCKLDQ ] }
809 { { 2 2 3 3 } [ dup PUNPCKHDQ ] }
810 { { 0 1 0 1 } [ dup PUNPCKLQDQ ] }
811 { { 2 3 2 3 } [ dup PUNPCKHQDQ ] }
815 : longlong-2-shuffle ( dst shuffle -- )
816 first2 [ 2 * dup 1 + ] bi@ 4array int-4-shuffle ;
818 : >float-4-shuffle ( double-2-shuffle -- float-4-shuffle )
819 [ 2 * { 0 1 } n+v ] map concat ;
821 M:: x86 %shuffle-vector-imm ( dst src shuffle rep -- )
823 dst shuffle rep signed-rep {
824 { double-2-rep [ >float-4-shuffle float-4-shuffle ] }
825 { float-4-rep [ float-4-shuffle ] }
826 { int-4-rep [ int-4-shuffle ] }
827 { longlong-2-rep [ longlong-2-shuffle ] }
830 M: x86 %shuffle-vector-imm-reps
832 { sse? { float-4-rep } }
833 { sse2? { double-2-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
836 M:: x86 %shuffle-vector-halves-imm ( dst src1 src2 shuffle rep -- )
837 dst src1 src2 rep two-operand
839 { double-2-rep [ >float-4-shuffle SHUFPS ] }
840 { float-4-rep [ SHUFPS ] }
843 M: x86 %shuffle-vector-halves-imm-reps
845 { sse? { float-4-rep } }
846 { sse2? { double-2-rep } }
849 M: x86 %shuffle-vector ( dst src shuffle rep -- )
852 M: x86 %shuffle-vector-reps
854 { ssse3? { float-4-rep double-2-rep longlong-2-rep ulonglong-2-rep int-4-rep uint-4-rep short-8-rep ushort-8-rep char-16-rep uchar-16-rep } }
857 M: x86 %merge-vector-head
860 { double-2-rep [ MOVLHPS ] }
861 { float-4-rep [ UNPCKLPS ] }
862 { longlong-2-rep [ PUNPCKLQDQ ] }
863 { int-4-rep [ PUNPCKLDQ ] }
864 { short-8-rep [ PUNPCKLWD ] }
865 { char-16-rep [ PUNPCKLBW ] }
868 M: x86 %merge-vector-tail
871 { double-2-rep [ UNPCKHPD ] }
872 { float-4-rep [ UNPCKHPS ] }
873 { longlong-2-rep [ PUNPCKHQDQ ] }
874 { int-4-rep [ PUNPCKHDQ ] }
875 { short-8-rep [ PUNPCKHWD ] }
876 { char-16-rep [ PUNPCKHBW ] }
879 M: x86 %merge-vector-reps
881 { sse? { float-4-rep } }
882 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
885 M: x86 %signed-pack-vector
888 { int-4-rep [ PACKSSDW ] }
889 { short-8-rep [ PACKSSWB ] }
892 M: x86 %signed-pack-vector-reps
894 { sse2? { short-8-rep int-4-rep } }
897 M: x86 %unsigned-pack-vector
900 { int-4-rep [ PACKUSDW ] }
901 { short-8-rep [ PACKUSWB ] }
904 M: x86 %unsigned-pack-vector-reps
906 { sse2? { short-8-rep } }
907 { sse4.1? { int-4-rep } }
910 M: x86 %tail>head-vector ( dst src rep -- )
912 { float-4-rep [ drop UNPCKHPD ] }
913 { double-2-rep [ drop UNPCKHPD ] }
914 [ drop [ %copy ] [ drop PUNPCKHQDQ ] 3bi ]
917 M: x86 %unpack-vector-head ( dst src rep -- )
919 { char-16-rep [ PMOVSXBW ] }
920 { uchar-16-rep [ PMOVZXBW ] }
921 { short-8-rep [ PMOVSXWD ] }
922 { ushort-8-rep [ PMOVZXWD ] }
923 { int-4-rep [ PMOVSXDQ ] }
924 { uint-4-rep [ PMOVZXDQ ] }
925 { float-4-rep [ CVTPS2PD ] }
928 M: x86 %unpack-vector-head-reps ( -- reps )
930 { sse2? { float-4-rep } }
931 { sse4.1? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep } }
934 M: x86 %integer>float-vector ( dst src rep -- )
936 { int-4-rep [ CVTDQ2PS ] }
939 M: x86 %integer>float-vector-reps
941 { sse2? { int-4-rep } }
944 M: x86 %float>integer-vector ( dst src rep -- )
946 { float-4-rep [ CVTTPS2DQ ] }
949 M: x86 %float>integer-vector-reps
951 { sse2? { float-4-rep } }
954 : (%compare-float-vector) ( dst src rep double single -- )
955 [ double-2-rep eq? ] 2dip if ; inline
957 : %compare-float-vector ( dst src rep cc -- )
959 { cc< [ [ CMPLTPD ] [ CMPLTPS ] (%compare-float-vector) ] }
960 { cc<= [ [ CMPLEPD ] [ CMPLEPS ] (%compare-float-vector) ] }
961 { cc= [ [ CMPEQPD ] [ CMPEQPS ] (%compare-float-vector) ] }
962 { cc<>= [ [ CMPORDPD ] [ CMPORDPS ] (%compare-float-vector) ] }
963 { cc/< [ [ CMPNLTPD ] [ CMPNLTPS ] (%compare-float-vector) ] }
964 { cc/<= [ [ CMPNLEPD ] [ CMPNLEPS ] (%compare-float-vector) ] }
965 { cc/= [ [ CMPNEQPD ] [ CMPNEQPS ] (%compare-float-vector) ] }
966 { cc/<>= [ [ CMPUNORDPD ] [ CMPUNORDPS ] (%compare-float-vector) ] }
969 :: (%compare-int-vector) ( dst src rep int64 int32 int16 int8 -- )
970 rep signed-rep :> rep'
972 { longlong-2-rep [ int64 call ] }
973 { int-4-rep [ int32 call ] }
974 { short-8-rep [ int16 call ] }
975 { char-16-rep [ int8 call ] }
978 : %compare-int-vector ( dst src rep cc -- )
980 { cc= [ [ PCMPEQQ ] [ PCMPEQD ] [ PCMPEQW ] [ PCMPEQB ] (%compare-int-vector) ] }
981 { cc> [ [ PCMPGTQ ] [ PCMPGTD ] [ PCMPGTW ] [ PCMPGTB ] (%compare-int-vector) ] }
984 M: x86 %compare-vector ( dst src1 src2 rep cc -- )
985 [ [ two-operand ] keep ] dip
986 over float-vector-rep?
987 [ %compare-float-vector ]
988 [ %compare-int-vector ] if ;
990 : %compare-vector-eq-reps ( -- reps )
992 { sse? { float-4-rep } }
993 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep } }
994 { sse4.1? { longlong-2-rep ulonglong-2-rep } }
997 : %compare-vector-ord-reps ( -- reps )
999 { sse? { float-4-rep } }
1000 { sse2? { double-2-rep char-16-rep short-8-rep int-4-rep } }
1001 { sse4.2? { longlong-2-rep } }
1004 M: x86 %compare-vector-reps
1006 { [ dup { cc= cc/= cc/<>= cc<>= } member-eq? ] [ drop %compare-vector-eq-reps ] }
1007 [ drop %compare-vector-ord-reps ]
1010 : %compare-float-vector-ccs ( cc -- ccs not? )
1012 { cc< [ { { cc< f } } f ] }
1013 { cc<= [ { { cc<= f } } f ] }
1014 { cc> [ { { cc< t } } f ] }
1015 { cc>= [ { { cc<= t } } f ] }
1016 { cc= [ { { cc= f } } f ] }
1017 { cc<> [ { { cc< f } { cc< t } } f ] }
1018 { cc<>= [ { { cc<>= f } } f ] }
1019 { cc/< [ { { cc/< f } } f ] }
1020 { cc/<= [ { { cc/<= f } } f ] }
1021 { cc/> [ { { cc/< t } } f ] }
1022 { cc/>= [ { { cc/<= t } } f ] }
1023 { cc/= [ { { cc/= f } } f ] }
1024 { cc/<> [ { { cc/= f } { cc/<>= f } } f ] }
1025 { cc/<>= [ { { cc/<>= f } } f ] }
1028 : %compare-int-vector-ccs ( cc -- ccs not? )
1030 { cc< [ { { cc> t } } f ] }
1031 { cc<= [ { { cc> f } } t ] }
1032 { cc> [ { { cc> f } } f ] }
1033 { cc>= [ { { cc> t } } t ] }
1034 { cc= [ { { cc= f } } f ] }
1035 { cc/= [ { { cc= f } } t ] }
1040 M: x86 %compare-vector-ccs
1041 swap float-vector-rep?
1042 [ %compare-float-vector-ccs ]
1043 [ %compare-int-vector-ccs ] if ;
1045 :: %test-vector-mask ( dst temp mask vcc -- )
1047 { vcc-any [ dst dst TEST dst temp \ CMOVNE (%boolean) ] }
1048 { vcc-none [ dst dst TEST dst temp \ CMOVE (%boolean) ] }
1049 { vcc-all [ dst mask CMP dst temp \ CMOVE (%boolean) ] }
1050 { vcc-notall [ dst mask CMP dst temp \ CMOVNE (%boolean) ] }
1053 : %move-vector-mask ( dst src rep -- mask )
1055 { double-2-rep [ MOVMSKPS HEX: f ] }
1056 { float-4-rep [ MOVMSKPS HEX: f ] }
1057 [ drop PMOVMSKB HEX: ffff ]
1060 M:: x86 %test-vector ( dst src temp rep vcc -- )
1061 dst src rep %move-vector-mask :> mask
1062 dst temp mask vcc %test-vector-mask ;
1064 :: %test-vector-mask-branch ( label temp mask vcc -- )
1066 { vcc-any [ temp temp TEST label JNE ] }
1067 { vcc-none [ temp temp TEST label JE ] }
1068 { vcc-all [ temp mask CMP label JE ] }
1069 { vcc-notall [ temp mask CMP label JNE ] }
1072 M:: x86 %test-vector-branch ( label src temp rep vcc -- )
1073 temp src rep %move-vector-mask :> mask
1074 label temp mask vcc %test-vector-mask-branch ;
1076 M: x86 %test-vector-reps
1078 { sse? { float-4-rep } }
1079 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1082 M: x86 %add-vector ( dst src1 src2 rep -- )
1083 [ two-operand ] keep
1085 { float-4-rep [ ADDPS ] }
1086 { double-2-rep [ ADDPD ] }
1087 { char-16-rep [ PADDB ] }
1088 { uchar-16-rep [ PADDB ] }
1089 { short-8-rep [ PADDW ] }
1090 { ushort-8-rep [ PADDW ] }
1091 { int-4-rep [ PADDD ] }
1092 { uint-4-rep [ PADDD ] }
1093 { longlong-2-rep [ PADDQ ] }
1094 { ulonglong-2-rep [ PADDQ ] }
1097 M: x86 %add-vector-reps
1099 { sse? { float-4-rep } }
1100 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1103 M: x86 %saturated-add-vector ( dst src1 src2 rep -- )
1104 [ two-operand ] keep
1106 { char-16-rep [ PADDSB ] }
1107 { uchar-16-rep [ PADDUSB ] }
1108 { short-8-rep [ PADDSW ] }
1109 { ushort-8-rep [ PADDUSW ] }
1112 M: x86 %saturated-add-vector-reps
1114 { sse2? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep } }
1117 M: x86 %add-sub-vector ( dst src1 src2 rep -- )
1118 [ two-operand ] keep
1120 { float-4-rep [ ADDSUBPS ] }
1121 { double-2-rep [ ADDSUBPD ] }
1124 M: x86 %add-sub-vector-reps
1126 { sse3? { float-4-rep double-2-rep } }
1129 M: x86 %sub-vector ( dst src1 src2 rep -- )
1130 [ two-operand ] keep
1132 { float-4-rep [ SUBPS ] }
1133 { double-2-rep [ SUBPD ] }
1134 { char-16-rep [ PSUBB ] }
1135 { uchar-16-rep [ PSUBB ] }
1136 { short-8-rep [ PSUBW ] }
1137 { ushort-8-rep [ PSUBW ] }
1138 { int-4-rep [ PSUBD ] }
1139 { uint-4-rep [ PSUBD ] }
1140 { longlong-2-rep [ PSUBQ ] }
1141 { ulonglong-2-rep [ PSUBQ ] }
1144 M: x86 %sub-vector-reps
1146 { sse? { float-4-rep } }
1147 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1150 M: x86 %saturated-sub-vector ( dst src1 src2 rep -- )
1151 [ two-operand ] keep
1153 { char-16-rep [ PSUBSB ] }
1154 { uchar-16-rep [ PSUBUSB ] }
1155 { short-8-rep [ PSUBSW ] }
1156 { ushort-8-rep [ PSUBUSW ] }
1159 M: x86 %saturated-sub-vector-reps
1161 { sse2? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep } }
1164 M: x86 %mul-vector ( dst src1 src2 rep -- )
1165 [ two-operand ] keep
1167 { float-4-rep [ MULPS ] }
1168 { double-2-rep [ MULPD ] }
1169 { short-8-rep [ PMULLW ] }
1170 { ushort-8-rep [ PMULLW ] }
1171 { int-4-rep [ PMULLD ] }
1172 { uint-4-rep [ PMULLD ] }
1175 M: x86 %mul-vector-reps
1177 { sse? { float-4-rep } }
1178 { sse2? { double-2-rep short-8-rep ushort-8-rep } }
1179 { sse4.1? { int-4-rep uint-4-rep } }
1182 M: x86 %mul-high-vector ( dst src1 src2 rep -- )
1183 [ two-operand ] keep
1185 { short-8-rep [ PMULHW ] }
1186 { ushort-8-rep [ PMULHUW ] }
1189 M: x86 %mul-high-vector-reps
1191 { sse2? { short-8-rep ushort-8-rep } }
1194 M: x86 %mul-horizontal-add-vector ( dst src1 src2 rep -- )
1195 [ two-operand ] keep
1197 { char-16-rep [ PMADDUBSW ] }
1198 { uchar-16-rep [ PMADDUBSW ] }
1199 { short-8-rep [ PMADDWD ] }
1202 M: x86 %mul-horizontal-add-vector-reps
1204 { sse2? { short-8-rep } }
1205 { ssse3? { char-16-rep uchar-16-rep } }
1208 M: x86 %div-vector ( dst src1 src2 rep -- )
1209 [ two-operand ] keep
1211 { float-4-rep [ DIVPS ] }
1212 { double-2-rep [ DIVPD ] }
1215 M: x86 %div-vector-reps
1217 { sse? { float-4-rep } }
1218 { sse2? { double-2-rep } }
1221 M: x86 %min-vector ( dst src1 src2 rep -- )
1222 [ two-operand ] keep
1224 { char-16-rep [ PMINSB ] }
1225 { uchar-16-rep [ PMINUB ] }
1226 { short-8-rep [ PMINSW ] }
1227 { ushort-8-rep [ PMINUW ] }
1228 { int-4-rep [ PMINSD ] }
1229 { uint-4-rep [ PMINUD ] }
1230 { float-4-rep [ MINPS ] }
1231 { double-2-rep [ MINPD ] }
1234 M: x86 %min-vector-reps
1236 { sse? { float-4-rep } }
1237 { sse2? { uchar-16-rep short-8-rep double-2-rep } }
1238 { sse4.1? { char-16-rep ushort-8-rep int-4-rep uint-4-rep } }
1241 M: x86 %max-vector ( dst src1 src2 rep -- )
1242 [ two-operand ] keep
1244 { char-16-rep [ PMAXSB ] }
1245 { uchar-16-rep [ PMAXUB ] }
1246 { short-8-rep [ PMAXSW ] }
1247 { ushort-8-rep [ PMAXUW ] }
1248 { int-4-rep [ PMAXSD ] }
1249 { uint-4-rep [ PMAXUD ] }
1250 { float-4-rep [ MAXPS ] }
1251 { double-2-rep [ MAXPD ] }
1254 M: x86 %max-vector-reps
1256 { sse? { float-4-rep } }
1257 { sse2? { uchar-16-rep short-8-rep double-2-rep } }
1258 { sse4.1? { char-16-rep ushort-8-rep int-4-rep uint-4-rep } }
1261 M: x86 %avg-vector ( dst src1 src2 rep -- )
1262 [ two-operand ] keep
1264 { uchar-16-rep [ PAVGB ] }
1265 { ushort-8-rep [ PAVGW ] }
1268 M: x86 %avg-vector-reps
1270 { sse2? { uchar-16-rep ushort-8-rep } }
1274 [ two-operand ] keep
1276 { float-4-rep [ HEX: ff DPPS ] }
1277 { double-2-rep [ HEX: ff DPPD ] }
1280 M: x86 %dot-vector-reps
1282 { sse4.1? { float-4-rep double-2-rep } }
1286 [ two-operand ] keep
1288 { uchar-16-rep [ PSADBW ] }
1291 M: x86 %sad-vector-reps
1293 { sse2? { uchar-16-rep } }
1296 M: x86 %horizontal-add-vector ( dst src1 src2 rep -- )
1297 [ two-operand ] keep
1299 { float-4-rep [ HADDPS ] }
1300 { double-2-rep [ HADDPD ] }
1301 { int-4-rep [ PHADDD ] }
1302 { short-8-rep [ PHADDW ] }
1305 M: x86 %horizontal-add-vector-reps
1307 { sse3? { float-4-rep double-2-rep } }
1308 { ssse3? { int-4-rep uint-4-rep short-8-rep ushort-8-rep } }
1311 M: x86 %horizontal-shl-vector-imm ( dst src1 src2 rep -- )
1312 two-operand PSLLDQ ;
1314 M: x86 %horizontal-shl-vector-imm-reps
1316 { sse2? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep float-4-rep double-2-rep } }
1319 M: x86 %horizontal-shr-vector-imm ( dst src1 src2 rep -- )
1320 two-operand PSRLDQ ;
1322 M: x86 %horizontal-shr-vector-imm-reps
1324 { sse2? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep float-4-rep double-2-rep } }
1327 M: x86 %abs-vector ( dst src rep -- )
1329 { char-16-rep [ PABSB ] }
1330 { short-8-rep [ PABSW ] }
1331 { int-4-rep [ PABSD ] }
1334 M: x86 %abs-vector-reps
1336 { ssse3? { char-16-rep short-8-rep int-4-rep } }
1339 M: x86 %sqrt-vector ( dst src rep -- )
1341 { float-4-rep [ SQRTPS ] }
1342 { double-2-rep [ SQRTPD ] }
1345 M: x86 %sqrt-vector-reps
1347 { sse? { float-4-rep } }
1348 { sse2? { double-2-rep } }
1351 M: x86 %and-vector ( dst src1 src2 rep -- )
1352 [ two-operand ] keep
1354 { float-4-rep [ ANDPS ] }
1355 { double-2-rep [ ANDPS ] }
1359 M: x86 %and-vector-reps
1361 { sse? { float-4-rep } }
1362 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1365 M: x86 %andn-vector ( dst src1 src2 rep -- )
1366 [ two-operand ] keep
1368 { float-4-rep [ ANDNPS ] }
1369 { double-2-rep [ ANDNPS ] }
1373 M: x86 %andn-vector-reps
1375 { sse? { float-4-rep } }
1376 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1379 M: x86 %or-vector ( dst src1 src2 rep -- )
1380 [ two-operand ] keep
1382 { float-4-rep [ ORPS ] }
1383 { double-2-rep [ ORPS ] }
1387 M: x86 %or-vector-reps
1389 { sse? { float-4-rep } }
1390 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1393 M: x86 %xor-vector ( dst src1 src2 rep -- )
1394 [ two-operand ] keep
1396 { float-4-rep [ XORPS ] }
1397 { double-2-rep [ XORPS ] }
1401 M: x86 %xor-vector-reps
1403 { sse? { float-4-rep } }
1404 { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1407 M: x86 %shl-vector ( dst src1 src2 rep -- )
1408 [ two-operand ] keep
1410 { short-8-rep [ PSLLW ] }
1411 { ushort-8-rep [ PSLLW ] }
1412 { int-4-rep [ PSLLD ] }
1413 { uint-4-rep [ PSLLD ] }
1414 { longlong-2-rep [ PSLLQ ] }
1415 { ulonglong-2-rep [ PSLLQ ] }
1418 M: x86 %shl-vector-reps
1420 { sse2? { short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
1423 M: x86 %shr-vector ( dst src1 src2 rep -- )
1424 [ two-operand ] keep
1426 { short-8-rep [ PSRAW ] }
1427 { ushort-8-rep [ PSRLW ] }
1428 { int-4-rep [ PSRAD ] }
1429 { uint-4-rep [ PSRLD ] }
1430 { ulonglong-2-rep [ PSRLQ ] }
1433 M: x86 %shr-vector-reps
1435 { sse2? { short-8-rep ushort-8-rep int-4-rep uint-4-rep ulonglong-2-rep } }
1438 M: x86 %shl-vector-imm %shl-vector ;
1439 M: x86 %shl-vector-imm-reps %shl-vector-reps ;
1440 M: x86 %shr-vector-imm %shr-vector ;
1441 M: x86 %shr-vector-imm-reps %shr-vector-reps ;
1443 : scalar-sized-reg ( reg rep -- reg' )
1444 rep-size 8 * n-bit-version-of ;
1446 M: x86 %integer>scalar drop MOVD ;
1448 :: %scalar>integer-32 ( dst src rep -- )
1451 dst 32-bit-version-of src MOVD
1452 dst dst 32-bit-version-of
1453 2dup eq? [ 2drop ] [ MOVSX ] if
1456 dst 32-bit-version-of src MOVD
1458 { short-scalar-rep [
1459 dst 32-bit-version-of src MOVD
1460 dst dst 16-bit-version-of MOVSX
1462 { ushort-scalar-rep [
1463 dst 32-bit-version-of src MOVD
1464 dst dst 16-bit-version-of MOVZX
1467 dst 32-bit-version-of src MOVD
1468 dst { } 8 [| tmp-dst |
1469 tmp-dst dst int-rep %copy
1470 tmp-dst tmp-dst 8-bit-version-of MOVSX
1471 dst tmp-dst int-rep %copy
1472 ] with-small-register
1474 { uchar-scalar-rep [
1475 dst 32-bit-version-of src MOVD
1476 dst { } 8 [| tmp-dst |
1477 tmp-dst dst int-rep %copy
1478 tmp-dst tmp-dst 8-bit-version-of MOVZX
1479 dst tmp-dst int-rep %copy
1480 ] with-small-register
1484 M: x86.32 %scalar>integer ( dst src rep -- ) %scalar>integer-32 ;
1486 M: x86.64 %scalar>integer ( dst src rep -- )
1488 { longlong-scalar-rep [ MOVD ] }
1489 { ulonglong-scalar-rep [ MOVD ] }
1490 [ %scalar>integer-32 ]
1493 M: x86 %vector>scalar %copy ;
1495 M: x86 %scalar>vector %copy ;
1497 M:: x86 %spill ( src rep dst -- )
1500 M:: x86 %reload ( dst rep src -- )
1503 M:: x86 %store-reg-param ( src reg rep -- )
1506 M:: x86 %store-stack-param ( src n rep -- )
1507 n param@ src rep %copy ;
1509 HOOK: struct-return@ cpu ( n -- operand )
1511 M: x86 %prepare-struct-area ( dst -- )
1512 f struct-return@ LEA ;
1514 M: x86 %alien-indirect ( src -- )
1517 M: x86 %loop-entry 16 alignment [ NOP ] times ;
1519 M:: x86 %restore-context ( temp1 temp2 -- )
1520 #! Load Factor stack pointers on entry from C to Factor.
1522 ds-reg temp1 "datastack" context-field-offset [+] MOV
1523 rs-reg temp1 "retainstack" context-field-offset [+] MOV ;
1525 M:: x86 %save-context ( temp1 temp2 -- )
1526 #! Save Factor stack pointers in case the C code calls a
1527 #! callback which does a GC, which must reliably trace
1530 temp2 stack-reg cell neg [+] LEA
1531 temp1 "callstack-top" context-field-offset [+] temp2 MOV
1532 temp1 "datastack" context-field-offset [+] ds-reg MOV
1533 temp1 "retainstack" context-field-offset [+] rs-reg MOV ;
1535 M: x86 value-struct? drop t ;
1537 M: x86 immediate-arithmetic? ( n -- ? )
1538 HEX: -80000000 HEX: 7fffffff between? ;
1540 M: x86 immediate-bitwise? ( n -- ? )
1541 HEX: -80000000 HEX: 7fffffff between? ;
1543 : next-stack@ ( n -- operand )
1544 #! nth parameter from the next stack frame. Used to box
1545 #! input values to callbacks; the callback has its own
1546 #! stack frame set up, and we want to read the frame
1547 #! set up by the caller.
1548 frame-reg swap 2 cells + [+] ;
1553 enable-float-intrinsics
1554 enable-float-functions
1555 enable-float-min/max
1559 [ { (sse-version) popcnt? } compile ] with-optimizer
1561 "Factor requires SSE2, which your CPU does not support." print