factor/basis/cpu/x86/x86.factor

790 lines
23 KiB
Factor

! Copyright (C) 2005, 2010 Slava Pestov.
! See http://factorcode.org/license.txt for BSD license.
USING: accessors alien arrays assocs byte-arrays classes.algebra
classes.struct combinators compiler compiler.cfg
compiler.cfg.comparisons compiler.cfg.instructions
compiler.cfg.intrinsics compiler.cfg.registers
compiler.cfg.stack-frame compiler.codegen.gc-maps
compiler.codegen.labels compiler.codegen.relocation compiler.constants
compiler.units cpu.architecture cpu.x86.assembler
cpu.x86.assembler.operands cpu.x86.assembler.private cpu.x86.features
cpu.x86.features.private fry io kernel layouts locals make math
math.order memory namespaces sequences system vm vocabs ;
QUALIFIED-WITH: alien.c-types c
FROM: math => float ;
IN: cpu.x86
! Add some methods to the assembler to be more useful to the backend
M: label JMP 0 JMP rc-relative label-fixup ;
M: label JUMPcc [ 0 ] dip JUMPcc rc-relative label-fixup ;
M: x86 vector-regs float-regs ;
HOOK: stack-reg cpu ( -- reg )
HOOK: reserved-stack-space cpu ( -- n )
HOOK: pic-tail-reg cpu ( -- reg )
: stack@ ( n -- op ) stack-reg swap [+] ;
: special-offset ( m -- n )
reserved-stack-space + ;
: spill@ ( n -- op ) spill-offset special-offset stack@ ;
: (%inc) ( n reg -- ) swap cells dup 0 > [ ADD ] [ neg SUB ] if ; inline
: decr-stack-reg ( n -- )
[
dup cell = [ drop pic-tail-reg PUSH ] [ stack-reg swap SUB ] if
] unless-zero ;
: incr-stack-reg ( n -- )
[
dup cell = [ drop pic-tail-reg POP ] [ stack-reg swap ADD ] if
] unless-zero ;
: align-stack ( n -- n' ) 16 align ;
M: x86 stack-frame-size ( stack-frame -- i )
(stack-frame-size)
reserved-stack-space +
cell +
align-stack ;
M: x86 complex-addressing? t ;
M: x86 fused-unboxing? t ;
M: x86 test-instruction? t ;
M: x86 immediate-store? immediate-comparand? ;
M: x86 %load-immediate [ dup XOR ] [ MOV ] if-zero ;
M: x86 %load-reference
[ swap 0 MOV rc-absolute-cell rel-literal ]
[ \ f type-number MOV ]
if* ;
HOOK: ds-reg cpu ( -- reg )
HOOK: rs-reg cpu ( -- reg )
: reg-stack ( n reg -- op ) swap cells neg [+] ;
GENERIC: loc>operand ( loc -- operand )
M: ds-loc loc>operand n>> ds-reg reg-stack ;
M: rs-loc loc>operand n>> rs-reg reg-stack ;
M: x86 %peek loc>operand MOV ;
M: x86 %replace loc>operand swap MOV ;
M: x86 %replace-imm
loc>operand swap
{
{ [ dup not ] [ drop \ f type-number MOV ] }
{ [ dup fixnum? ] [ tag-fixnum MOV ] }
[ [ 0xffffffff MOV ] dip rc-absolute rel-literal ]
} cond ;
M: x86 %clear ( loc -- )
297 swap %replace-imm ;
M: x86 %inc ( loc -- )
[ n>> ] [ ds-loc? ds-reg rs-reg ? ] bi (%inc) ;
M: x86 %call ( word -- ) 0 CALL rc-relative rel-word-pic ;
: xt-tail-pic-offset ( -- n )
! See the comment in vm/cpu-x86.hpp
4 1 + ; inline
HOOK: %prepare-jump cpu ( -- )
M: x86 %jump ( word -- )
%prepare-jump
0 JMP rc-relative rel-word-pic-tail ;
M: x86 %jump-label ( label -- ) 0 JMP rc-relative label-fixup ;
M: x86 %return ( -- ) 0 RET ;
: (%slot) ( obj slot scale tag -- op ) neg <indirect> ; inline
: (%slot-imm) ( obj slot tag -- op ) slot-offset [+] ; inline
M: x86 %slot ( dst obj slot scale tag -- ) (%slot) MOV ;
M: x86 %slot-imm ( dst obj slot tag -- ) (%slot-imm) MOV ;
M: x86 %set-slot ( src obj slot scale tag -- ) (%slot) swap MOV ;
M: x86 %set-slot-imm ( src obj slot tag -- ) (%slot-imm) swap MOV ;
:: two-operand ( dst src1 src2 rep -- dst src )
dst src2 eq? dst src1 eq? not and [ "Cannot handle this case" throw ] when
dst src1 rep %copy
dst src2 ; inline
:: one-operand ( dst src rep -- dst )
dst src rep %copy
dst ; inline
M: x86 %add 2over eq? [ nip ADD ] [ [+] LEA ] if ;
M: x86 %add-imm 2over eq? [ nip ADD ] [ [+] LEA ] if ;
M: x86 %sub int-rep two-operand SUB ;
M: x86 %sub-imm 2over eq? [ nip SUB ] [ neg [+] LEA ] if ;
M: x86 %mul int-rep two-operand IMUL2 ;
M: x86 %mul-imm IMUL3 ;
M: x86 %and int-rep two-operand AND ;
M: x86 %and-imm int-rep two-operand AND ;
M: x86 %or int-rep two-operand OR ;
M: x86 %or-imm int-rep two-operand OR ;
M: x86 %xor int-rep two-operand XOR ;
M: x86 %xor-imm int-rep two-operand XOR ;
M: x86 %shl-imm int-rep two-operand SHL ;
M: x86 %shr-imm int-rep two-operand SHR ;
M: x86 %sar-imm int-rep two-operand SAR ;
M: x86 %min int-rep two-operand [ CMP ] [ CMOVG ] 2bi ;
M: x86 %max int-rep two-operand [ CMP ] [ CMOVL ] 2bi ;
M: x86 %not int-rep one-operand NOT ;
M: x86 %neg int-rep one-operand NEG ;
M: x86 %log2 BSR ;
M: x86 %bit-count POPCNT ;
! A bit of logic to avoid using MOVSS/MOVSD for reg-reg moves
! since this induces partial register stalls
GENERIC: copy-register* ( dst src rep -- )
GENERIC: copy-memory* ( dst src rep -- )
M: int-rep copy-register* drop MOV ;
M: tagged-rep copy-register* drop MOV ;
M: object copy-memory* copy-register* ;
: ?spill-slot ( obj -- obj ) dup spill-slot? [ n>> spill@ ] when ;
M: x86 %copy ( dst src rep -- )
2over eq? [ 3drop ] [
[ [ ?spill-slot ] bi@ ] dip
2over [ register? ] both? [ copy-register* ] [ copy-memory* ] if
] if ;
: fixnum-overflow ( label dst src1 src2 cc quot -- )
swap [ [ int-rep two-operand ] dip call ] dip
{
{ cc-o [ JO ] }
{ cc/o [ JNO ] }
} case ; inline
M: x86 %fixnum-add ( label dst src1 src2 cc -- )
[ ADD ] fixnum-overflow ;
M: x86 %fixnum-sub ( label dst src1 src2 cc -- )
[ SUB ] fixnum-overflow ;
M: x86 %fixnum-mul ( label dst src1 src2 cc -- )
[ IMUL2 ] fixnum-overflow ;
M: x86 %unbox-alien ( dst src -- )
alien-offset [+] MOV ;
M:: x86 %unbox-any-c-ptr ( dst src -- )
<label> :> end
dst dst XOR
! Is the object f?
src \ f type-number CMP
end JE
! Compute tag in dst register
dst src MOV
dst tag-mask get AND
! Is the object an alien?
dst alien type-number CMP
! Add an offset to start of byte array's data
dst src byte-array-offset [+] LEA
end JNE
! If so, load the offset and add it to the address
dst src alien-offset [+] MOV
end resolve-label ;
: alien@ ( reg n -- op ) cells alien type-number - [+] ;
M:: x86 %box-alien ( dst src temp -- )
<label> :> end
dst \ f type-number MOV
src src TEST
end JE
dst 5 cells alien temp %allot
dst 1 alien@ \ f type-number MOV ! base
dst 2 alien@ \ f type-number MOV ! expired
dst 3 alien@ src MOV ! displacement
dst 4 alien@ src MOV ! address
end resolve-label ;
:: %box-displaced-alien/f ( dst displacement -- )
dst 1 alien@ \ f type-number MOV
dst 3 alien@ displacement MOV
dst 4 alien@ displacement MOV ;
:: %box-displaced-alien/alien ( dst displacement base temp -- )
! Set new alien's base to base.base
temp base 1 alien@ MOV
dst 1 alien@ temp MOV
! Compute displacement
temp base 3 alien@ MOV
temp displacement ADD
dst 3 alien@ temp MOV
! Compute address
temp base 4 alien@ MOV
temp displacement ADD
dst 4 alien@ temp MOV ;
:: %box-displaced-alien/byte-array ( dst displacement base temp -- )
dst 1 alien@ base MOV
dst 3 alien@ displacement MOV
temp base displacement byte-array-offset [++] LEA
dst 4 alien@ temp MOV ;
:: %box-displaced-alien/dynamic ( dst displacement base temp end -- )
<label> :> not-f
<label> :> not-alien
! Check base type
temp base MOV
temp tag-mask get AND
! Is base f?
temp \ f type-number CMP
not-f JNE
! Yes, it is f. Fill in new object
dst displacement %box-displaced-alien/f
end JMP
not-f resolve-label
! Is base an alien?
temp alien type-number CMP
not-alien JNE
dst displacement base temp %box-displaced-alien/alien
! We are done
end JMP
! Is base a byte array? It has to be, by now...
not-alien resolve-label
dst displacement base temp %box-displaced-alien/byte-array ;
M:: x86 %box-displaced-alien ( dst displacement base temp base-class -- )
<label> :> end
! If displacement is zero, return the base
dst base MOV
displacement displacement TEST
end JE
! Displacement is non-zero, we're going to be allocating a new
! object
dst 5 cells alien temp %allot
! Set expired to f
dst 2 alien@ \ f type-number MOV
dst displacement base temp
{
{ [ base-class \ f class<= ] [ 2drop %box-displaced-alien/f ] }
{ [ base-class \ alien class<= ] [ %box-displaced-alien/alien ] }
{ [ base-class \ byte-array class<= ] [ %box-displaced-alien/byte-array ] }
[ end %box-displaced-alien/dynamic ]
} cond
end resolve-label ;
! The 'small-reg' mess is pretty crappy, but its only used on x86-32.
! On x86-64, all registers have 8-bit versions. However, a similar
! problem arises for shifts, where the shift count must be in CL, and
! so one day I will fix this properly by adding precoloring to the
! register allocator.
HOOK: has-small-reg? cpu ( reg size -- ? )
CONSTANT: have-byte-regs { EAX ECX EDX EBX }
M: x86.32 has-small-reg?
{
{ 8 [ have-byte-regs member-eq? ] }
{ 16 [ drop t ] }
{ 32 [ drop t ] }
} case ;
M: x86.64 has-small-reg? 2drop t ;
: small-reg-that-isn't ( exclude -- reg' )
[ have-byte-regs ] dip
[ native-version-of ] map
'[ _ member-eq? not ] find nip ;
: with-save/restore ( reg quot -- )
[ drop PUSH ] [ call ] [ drop POP ] 2tri ; inline
:: with-small-register ( dst exclude size quot: ( new-dst -- ) -- )
! If the destination register overlaps a small register with
! 'size' bits, we call the quot with that. Otherwise, we find a
! small register that is not in exclude, and call quot, saving and
! restoring the small register.
dst size has-small-reg? [ dst quot call ] [
exclude small-reg-that-isn't
[ quot call ] with-save/restore
] if ; inline
:: (%convert-integer) ( dst src bits quot -- )
dst { src } bits [| new-dst |
new-dst src int-rep %copy
new-dst dup bits n-bit-version-of quot call
dst new-dst int-rep %copy
] with-small-register ; inline
: %zero-extend ( dst src bits -- )
[ MOVZX ] (%convert-integer) ; inline
: %sign-extend ( dst src bits -- )
[ MOVSX ] (%convert-integer) ; inline
M: x86 %convert-integer ( dst src c-type -- )
{
{ c:char [ 8 %sign-extend ] }
{ c:uchar [ 8 %zero-extend ] }
{ c:short [ 16 %sign-extend ] }
{ c:ushort [ 16 %zero-extend ] }
{ c:int [ 32 %sign-extend ] }
{ c:uint [ 32 [ 2drop ] (%convert-integer) ] }
} case ;
:: %alien-integer-getter ( dst exclude address bits quot -- )
dst exclude bits [| new-dst |
new-dst dup bits n-bit-version-of dup address MOV
quot call
dst new-dst int-rep %copy
] with-small-register ; inline
: %alien-unsigned-getter ( dst exclude address bits -- )
[ MOVZX ] %alien-integer-getter ; inline
: %alien-signed-getter ( dst exclude address bits -- )
[ MOVSX ] %alien-integer-getter ; inline
:: %alien-integer-setter ( value exclude address bits -- )
value exclude bits [| new-value |
new-value value int-rep %copy
address new-value bits n-bit-version-of MOV
] with-small-register ; inline
: (%memory) ( base displacement scale offset rep c-type -- exclude address rep c-type )
[ [ [ 2array ] 2keep ] 2dip <indirect> ] 2dip ;
: (%memory-imm) ( base offset rep c-type -- exclude address rep c-type )
[ [ drop 1array ] [ [+] ] 2bi ] 2dip ;
: (%load-memory) ( dst exclude address rep c-type -- )
[
{
{ c:char [ 8 %alien-signed-getter ] }
{ c:uchar [ 8 %alien-unsigned-getter ] }
{ c:short [ 16 %alien-signed-getter ] }
{ c:ushort [ 16 %alien-unsigned-getter ] }
{ c:int [ 32 %alien-signed-getter ] }
{ c:uint [ 32 [ 2drop ] %alien-integer-getter ] }
} case
] [ [ drop ] 2dip %copy ] ?if ;
M: x86 %load-memory ( dst base displacement scale offset rep c-type -- )
(%memory) (%load-memory) ;
M: x86 %load-memory-imm ( dst base offset rep c-type -- )
(%memory-imm) (%load-memory) ;
: (%store-memory) ( src exclude address rep c-type -- )
[
{
{ c:char [ 8 %alien-integer-setter ] }
{ c:uchar [ 8 %alien-integer-setter ] }
{ c:short [ 16 %alien-integer-setter ] }
{ c:ushort [ 16 %alien-integer-setter ] }
{ c:int [ 32 %alien-integer-setter ] }
{ c:uint [ 32 %alien-integer-setter ] }
} case
] [ [ nip swap ] dip %copy ] ?if ;
M: x86 %store-memory ( src base displacement scale offset rep c-type -- )
(%memory) (%store-memory) ;
M: x86 %store-memory-imm ( src base offset rep c-type -- )
(%memory-imm) (%store-memory) ;
: shift-count? ( reg -- ? ) { ECX RCX } member-eq? ;
:: emit-shift ( dst src quot -- )
src shift-count? [
dst CL quot call
] [
dst shift-count? [
dst src XCHG
src CL quot call
dst src XCHG
] [
ECX native-version-of [
CL src MOV
drop dst CL quot call
] with-save/restore
] if
] if ; inline
M: x86 %shl int-rep two-operand [ SHL ] emit-shift ;
M: x86 %shr int-rep two-operand [ SHR ] emit-shift ;
M: x86 %sar int-rep two-operand [ SAR ] emit-shift ;
HOOK: %vm-field-ptr cpu ( reg offset -- )
: load-zone-offset ( nursery-ptr -- )
"nursery" vm offset-of %vm-field-ptr ;
: load-allot-ptr ( nursery-ptr allot-ptr -- )
[ drop load-zone-offset ] [ swap [] MOV ] 2bi ;
: inc-allot-ptr ( nursery-ptr n -- )
[ [] ] dip data-alignment get align ADD ;
: store-header ( temp class -- )
[ [] ] [ type-number tag-header ] bi* MOV ;
: store-tagged ( dst tag -- )
type-number OR ;
M:: x86 %allot ( dst size class nursery-ptr -- )
nursery-ptr dst load-allot-ptr
dst class store-header
dst class store-tagged
nursery-ptr size inc-allot-ptr ;
HOOK: %mark-card cpu ( card temp -- )
HOOK: %mark-deck cpu ( card temp -- )
:: (%write-barrier) ( temp1 temp2 -- )
temp1 card-bits SHR
temp1 temp2 %mark-card
temp1 deck-bits card-bits - SHR
temp1 temp2 %mark-deck ;
M:: x86 %write-barrier ( src slot scale tag temp1 temp2 -- )
temp1 src slot scale tag (%slot) LEA
temp1 temp2 (%write-barrier) ;
M:: x86 %write-barrier-imm ( src slot tag temp1 temp2 -- )
temp1 src slot tag (%slot-imm) LEA
temp1 temp2 (%write-barrier) ;
M:: x86 %check-nursery-branch ( label size cc temp1 temp2 -- )
temp1 load-zone-offset
temp2 temp1 [] MOV
temp2 size ADD
temp2 temp1 2 cells [+] CMP
cc {
{ cc<= [ label JLE ] }
{ cc/<= [ label JG ] }
} case ;
M: x86 gc-root-offset
n>> spill-offset special-offset cell + cell /i ;
M: x86 %call-gc ( gc-map -- )
\ minor-gc %call
gc-map-here ;
M: x86 %alien-global ( dst symbol library -- )
[ 0 MOV ] 2dip rc-absolute-cell rel-dlsym ;
M: x86 %prologue ( n -- ) cell - decr-stack-reg ;
M: x86 %epilogue ( n -- ) cell - incr-stack-reg ;
:: (%boolean) ( dst temp insn -- )
dst \ f type-number MOV
temp 0 MOV \ t rc-absolute-cell rel-literal
dst temp insn execute ; inline
: %boolean ( dst cc temp -- )
swap order-cc {
{ cc< [ \ CMOVL (%boolean) ] }
{ cc<= [ \ CMOVLE (%boolean) ] }
{ cc> [ \ CMOVG (%boolean) ] }
{ cc>= [ \ CMOVGE (%boolean) ] }
{ cc= [ \ CMOVE (%boolean) ] }
{ cc/= [ \ CMOVNE (%boolean) ] }
} case ;
M:: x86 %compare ( dst src1 src2 cc temp -- )
src1 src2 CMP
dst cc temp %boolean ;
M:: x86 %test ( dst src1 src2 cc temp -- )
src1 src2 TEST
dst cc temp %boolean ;
: (%compare-tagged) ( src1 src2 -- )
[ 0xffffffff CMP ] dip rc-absolute rel-literal ;
M:: x86 %compare-integer-imm ( dst src1 src2 cc temp -- )
src1 src2 CMP
dst cc temp %boolean ;
M:: x86 %test-imm ( dst src1 src2 cc temp -- )
src1 src2 TEST
dst cc temp %boolean ;
: (%compare-imm) ( src1 src2 -- )
{
{ [ dup fixnum? ] [ tag-fixnum CMP ] }
{ [ dup not ] [ drop \ f type-number CMP ] }
[ (%compare-tagged) ]
} cond ;
M:: x86 %compare-imm ( dst src1 src2 cc temp -- )
src1 src2 (%compare-imm)
dst cc temp %boolean ;
: %branch ( label cc -- )
order-cc {
{ cc< [ JL ] }
{ cc<= [ JLE ] }
{ cc> [ JG ] }
{ cc>= [ JGE ] }
{ cc= [ JE ] }
{ cc/= [ JNE ] }
} case ;
M:: x86 %compare-branch ( label src1 src2 cc -- )
src1 src2 CMP
label cc %branch ;
M:: x86 %compare-integer-imm-branch ( label src1 src2 cc -- )
src1 src2 CMP
label cc %branch ;
M:: x86 %test-branch ( label src1 src2 cc -- )
src1 src2 TEST
label cc %branch ;
M:: x86 %test-imm-branch ( label src1 src2 cc -- )
src1 src2 TEST
label cc %branch ;
M:: x86 %compare-imm-branch ( label src1 src2 cc -- )
src1 src2 (%compare-imm)
label cc %branch ;
M:: x86 %dispatch ( src temp -- )
! Load jump table base.
temp 0xffffffff MOV
building get length :> start
0 rc-absolute-cell rel-here
! Add jump table base
temp src 0x7f [++] JMP
building get length :> end
! Fix up the displacement above
cell alignment
[ end start - + building get set-last ]
[ (align-code) ]
bi ;
M:: x86 %spill ( src rep dst -- )
dst src rep %copy ;
M:: x86 %reload ( dst rep src -- )
dst src rep %copy ;
M:: x86 %local-allot ( dst size align offset -- )
dst offset local-allot-offset special-offset stack@ LEA ;
: next-stack@ ( n -- operand )
! nth parameter from the next stack frame. Used to box
! input values to callbacks; the callback has its own
! stack frame set up, and we want to read the frame
! set up by the caller.
[ frame-reg ] dip 2 cells + reserved-stack-space + [+] ;
: return-reg ( rep -- reg )
reg-class-of return-regs at first ;
HOOK: %load-stack-param cpu ( vreg rep n -- )
HOOK: %store-stack-param cpu ( vreg rep n -- )
HOOK: %load-reg-param cpu ( vreg rep reg -- )
HOOK: %store-reg-param cpu ( vreg rep reg -- )
HOOK: %discard-reg-param cpu ( rep reg -- )
: %load-return ( dst rep -- )
dup return-reg %load-reg-param ;
: %store-return ( dst rep -- )
dup return-reg %store-reg-param ;
HOOK: %prepare-var-args cpu ( -- )
HOOK: %cleanup cpu ( n -- )
M:: x86 %alien-assembly ( reg-inputs
stack-inputs
reg-outputs
dead-outputs
cleanup
stack-size
quot -- )
stack-inputs [ first3 %store-stack-param ] each
reg-inputs [ first3 %store-reg-param ] each
%prepare-var-args
quot call( -- )
cleanup %cleanup
reg-outputs [ first3 %load-reg-param ] each
dead-outputs [ first2 %discard-reg-param ] each ;
M: x86 %alien-invoke ( reg-inputs stack-inputs reg-outputs dead-outputs cleanup stack-size symbols dll gc-map -- )
'[ _ _ _ %c-invoke ] %alien-assembly ;
M:: x86 %alien-indirect ( src reg-inputs stack-inputs reg-outputs dead-outputs cleanup stack-size gc-map -- )
reg-inputs stack-inputs reg-outputs dead-outputs cleanup stack-size [
src ?spill-slot CALL
gc-map gc-map-here
] %alien-assembly ;
HOOK: %begin-callback cpu ( -- )
M: x86 %callback-inputs ( reg-outputs stack-outputs -- )
[ [ first3 %load-reg-param ] each ]
[ [ first3 %load-stack-param ] each ] bi*
%begin-callback ;
HOOK: %end-callback cpu ( -- )
M: x86 %callback-outputs ( reg-inputs -- )
%end-callback
[ first3 %store-reg-param ] each ;
M: x86 %loop-entry 16 alignment [ NOP ] times ;
M:: x86 %save-context ( temp1 temp2 -- )
! Save Factor stack pointers in case the C code calls a
! callback which does a GC, which must reliably trace
! all roots.
temp1 %context
temp2 stack-reg cell neg [+] LEA
temp1 "callstack-top" context offset-of [+] temp2 MOV
temp1 "datastack" context offset-of [+] ds-reg MOV
temp1 "retainstack" context offset-of [+] rs-reg MOV ;
M: x86 value-struct? drop t ;
M: x86 long-long-odd-register? f ;
M: x86 float-right-align-on-stack? f ;
M: x86 immediate-arithmetic? ( n -- ? )
-0x80000000 0x7fffffff between? ;
M: x86 immediate-bitwise? ( n -- ? )
-0x80000000 0x7fffffff between? ;
:: %cmov-float= ( dst src -- )
<label> :> no-move
no-move [ JNE ] [ JP ] bi
dst src MOV
no-move resolve-label ;
:: %cmov-float/= ( dst src -- )
<label> :> no-move
<label> :> move
move JP
no-move JE
move resolve-label
dst src MOV
no-move resolve-label ;
:: (%compare-float) ( dst src1 src2 cc temp compare -- )
cc {
{ cc< [ src2 src1 compare call( a b -- ) dst temp \ CMOVA (%boolean) ] }
{ cc<= [ src2 src1 compare call( a b -- ) dst temp \ CMOVAE (%boolean) ] }
{ cc> [ src1 src2 compare call( a b -- ) dst temp \ CMOVA (%boolean) ] }
{ cc>= [ src1 src2 compare call( a b -- ) dst temp \ CMOVAE (%boolean) ] }
{ cc= [ src1 src2 compare call( a b -- ) dst temp \ %cmov-float= (%boolean) ] }
{ cc<> [ src1 src2 compare call( a b -- ) dst temp \ CMOVNE (%boolean) ] }
{ cc<>= [ src1 src2 compare call( a b -- ) dst temp \ CMOVNP (%boolean) ] }
{ cc/< [ src2 src1 compare call( a b -- ) dst temp \ CMOVBE (%boolean) ] }
{ cc/<= [ src2 src1 compare call( a b -- ) dst temp \ CMOVB (%boolean) ] }
{ cc/> [ src1 src2 compare call( a b -- ) dst temp \ CMOVBE (%boolean) ] }
{ cc/>= [ src1 src2 compare call( a b -- ) dst temp \ CMOVB (%boolean) ] }
{ cc/= [ src1 src2 compare call( a b -- ) dst temp \ %cmov-float/= (%boolean) ] }
{ cc/<> [ src1 src2 compare call( a b -- ) dst temp \ CMOVE (%boolean) ] }
{ cc/<>= [ src1 src2 compare call( a b -- ) dst temp \ CMOVP (%boolean) ] }
} case ; inline
:: %jump-float= ( label -- )
<label> :> no-jump
no-jump JP
label JE
no-jump resolve-label ;
: %jump-float/= ( label -- )
[ JNE ] [ JP ] bi ;
:: (%compare-float-branch) ( label src1 src2 cc compare -- )
cc {
{ cc< [ src2 src1 compare call( a b -- ) label JA ] }
{ cc<= [ src2 src1 compare call( a b -- ) label JAE ] }
{ cc> [ src1 src2 compare call( a b -- ) label JA ] }
{ cc>= [ src1 src2 compare call( a b -- ) label JAE ] }
{ cc= [ src1 src2 compare call( a b -- ) label %jump-float= ] }
{ cc<> [ src1 src2 compare call( a b -- ) label JNE ] }
{ cc<>= [ src1 src2 compare call( a b -- ) label JNP ] }
{ cc/< [ src2 src1 compare call( a b -- ) label JBE ] }
{ cc/<= [ src2 src1 compare call( a b -- ) label JB ] }
{ cc/> [ src1 src2 compare call( a b -- ) label JBE ] }
{ cc/>= [ src1 src2 compare call( a b -- ) label JB ] }
{ cc/= [ src1 src2 compare call( a b -- ) label %jump-float/= ] }
{ cc/<> [ src1 src2 compare call( a b -- ) label JE ] }
{ cc/<>= [ src1 src2 compare call( a b -- ) label JP ] }
} case ;
enable-min/max
enable-log2
M:: x86 %bit-test ( dst src1 src2 temp -- )
src1 src2 BT
dst temp \ CMOVB (%boolean) ;
enable-bit-test
: check-sse ( -- )
"Checking for multimedia extensions... " write flush
sse-version
[ sse-string " detected" append print ]
[ 20 < "cpu.x86.x87" "cpu.x86.sse" ? require ] bi ;
: check-popcnt ( -- )
enable-popcnt? [
"Building with POPCNT support" print
enable-bit-count
] when ;
: check-cpu-features ( -- )
[ { (sse-version) popcnt? } compile ] with-optimizer
check-sse
check-popcnt ;