func gcmp(as int, lhs *gc.Node, rhs *gc.Node) *obj.Prog {
if lhs.Op != gc.OREGISTER {
gc.Fatalf("bad operands to gcmp: %v %v", gc.Oconv(int(lhs.Op), 0), gc.Oconv(int(rhs.Op), 0))
}
p := rawgins(as, rhs, nil)
raddr(lhs, p)
return p
}
/*
* insert n into reg slot of p
*/
func raddr(n *gc.Node, p *obj.Prog) {
var a obj.Addr
gc.Naddr(&a, n)
if a.Type != obj.TYPE_REG {
if n != nil {
gc.Fatalf("bad in raddr: %v", gc.Oconv(int(n.Op), 0))
} else {
gc.Fatalf("bad in raddr: <null>")
}
p.Reg = 0
} else {
p.Reg = a.Reg
}
}
// jmptoset returns ASETxx for AJxx.
func jmptoset(jmp int) int {
switch jmp {
case x86.AJEQ:
return x86.ASETEQ
case x86.AJNE:
return x86.ASETNE
case x86.AJLT:
return x86.ASETLT
case x86.AJCS:
return x86.ASETCS
case x86.AJLE:
return x86.ASETLE
case x86.AJLS:
return x86.ASETLS
case x86.AJGT:
return x86.ASETGT
case x86.AJHI:
return x86.ASETHI
case x86.AJGE:
return x86.ASETGE
case x86.AJCC:
return x86.ASETCC
case x86.AJMI:
return x86.ASETMI
case x86.AJOC:
return x86.ASETOC
case x86.AJOS:
return x86.ASETOS
case x86.AJPC:
return x86.ASETPC
case x86.AJPL:
return x86.ASETPL
case x86.AJPS:
return x86.ASETPS
}
gc.Fatalf("jmptoset: no entry for %v", gc.Oconv(jmp, 0))
panic("unreachable")
}
/*
* return Axxx for Oxxx on type t.
*/
func optoas(op gc.Op, t *gc.Type) int {
if t == nil {
gc.Fatalf("optoas: t is nil")
}
// avoid constant conversions in switches below
const (
OMINUS_ = uint32(gc.OMINUS) << 16
OLSH_ = uint32(gc.OLSH) << 16
ORSH_ = uint32(gc.ORSH) << 16
OADD_ = uint32(gc.OADD) << 16
OSUB_ = uint32(gc.OSUB) << 16
OMUL_ = uint32(gc.OMUL) << 16
ODIV_ = uint32(gc.ODIV) << 16
OOR_ = uint32(gc.OOR) << 16
OAND_ = uint32(gc.OAND) << 16
OXOR_ = uint32(gc.OXOR) << 16
OEQ_ = uint32(gc.OEQ) << 16
ONE_ = uint32(gc.ONE) << 16
OLT_ = uint32(gc.OLT) << 16
OLE_ = uint32(gc.OLE) << 16
OGE_ = uint32(gc.OGE) << 16
OGT_ = uint32(gc.OGT) << 16
OCMP_ = uint32(gc.OCMP) << 16
OAS_ = uint32(gc.OAS) << 16
OHMUL_ = uint32(gc.OHMUL) << 16
)
a := int(obj.AXXX)
switch uint32(op)<<16 | uint32(gc.Simtype[t.Etype]) {
default:
gc.Fatalf("optoas: no entry for op=%v type=%v", gc.Oconv(int(op), 0), t)
case OEQ_ | gc.TBOOL,
OEQ_ | gc.TINT8,
OEQ_ | gc.TUINT8,
OEQ_ | gc.TINT16,
OEQ_ | gc.TUINT16,
OEQ_ | gc.TINT32,
OEQ_ | gc.TUINT32,
OEQ_ | gc.TINT64,
OEQ_ | gc.TUINT64,
OEQ_ | gc.TPTR32,
OEQ_ | gc.TPTR64,
OEQ_ | gc.TFLOAT32,
OEQ_ | gc.TFLOAT64:
a = ppc64.ABEQ
case ONE_ | gc.TBOOL,
ONE_ | gc.TINT8,
ONE_ | gc.TUINT8,
ONE_ | gc.TINT16,
ONE_ | gc.TUINT16,
ONE_ | gc.TINT32,
ONE_ | gc.TUINT32,
ONE_ | gc.TINT64,
ONE_ | gc.TUINT64,
ONE_ | gc.TPTR32,
ONE_ | gc.TPTR64,
ONE_ | gc.TFLOAT32,
ONE_ | gc.TFLOAT64:
a = ppc64.ABNE
case OLT_ | gc.TINT8, // ACMP
OLT_ | gc.TINT16,
OLT_ | gc.TINT32,
OLT_ | gc.TINT64,
OLT_ | gc.TUINT8,
// ACMPU
OLT_ | gc.TUINT16,
OLT_ | gc.TUINT32,
OLT_ | gc.TUINT64,
OLT_ | gc.TFLOAT32,
// AFCMPU
OLT_ | gc.TFLOAT64:
a = ppc64.ABLT
case OLE_ | gc.TINT8, // ACMP
OLE_ | gc.TINT16,
OLE_ | gc.TINT32,
OLE_ | gc.TINT64,
OLE_ | gc.TUINT8,
// ACMPU
OLE_ | gc.TUINT16,
OLE_ | gc.TUINT32,
OLE_ | gc.TUINT64:
// No OLE for floats, because it mishandles NaN.
// Front end must reverse comparison or use OLT and OEQ together.
a = ppc64.ABLE
case OGT_ | gc.TINT8,
OGT_ | gc.TINT16,
OGT_ | gc.TINT32,
OGT_ | gc.TINT64,
OGT_ | gc.TUINT8,
OGT_ | gc.TUINT16,
OGT_ | gc.TUINT32,
OGT_ | gc.TUINT64,
OGT_ | gc.TFLOAT32,
OGT_ | gc.TFLOAT64:
a = ppc64.ABGT
case OGE_ | gc.TINT8,
OGE_ | gc.TINT16,
OGE_ | gc.TINT32,
OGE_ | gc.TINT64,
OGE_ | gc.TUINT8,
OGE_ | gc.TUINT16,
OGE_ | gc.TUINT32,
OGE_ | gc.TUINT64:
// No OGE for floats, because it mishandles NaN.
// Front end must reverse comparison or use OLT and OEQ together.
a = ppc64.ABGE
case OCMP_ | gc.TBOOL,
OCMP_ | gc.TINT8,
OCMP_ | gc.TINT16,
OCMP_ | gc.TINT32,
OCMP_ | gc.TPTR32,
OCMP_ | gc.TINT64:
a = ppc64.ACMP
case OCMP_ | gc.TUINT8,
OCMP_ | gc.TUINT16,
OCMP_ | gc.TUINT32,
OCMP_ | gc.TUINT64,
OCMP_ | gc.TPTR64:
a = ppc64.ACMPU
case OCMP_ | gc.TFLOAT32,
OCMP_ | gc.TFLOAT64:
a = ppc64.AFCMPU
case OAS_ | gc.TBOOL,
OAS_ | gc.TINT8:
a = ppc64.AMOVB
case OAS_ | gc.TUINT8:
a = ppc64.AMOVBZ
case OAS_ | gc.TINT16:
a = ppc64.AMOVH
case OAS_ | gc.TUINT16:
a = ppc64.AMOVHZ
case OAS_ | gc.TINT32:
a = ppc64.AMOVW
case OAS_ | gc.TUINT32,
OAS_ | gc.TPTR32:
a = ppc64.AMOVWZ
case OAS_ | gc.TINT64,
OAS_ | gc.TUINT64,
OAS_ | gc.TPTR64:
a = ppc64.AMOVD
case OAS_ | gc.TFLOAT32:
a = ppc64.AFMOVS
case OAS_ | gc.TFLOAT64:
a = ppc64.AFMOVD
case OADD_ | gc.TINT8,
OADD_ | gc.TUINT8,
OADD_ | gc.TINT16,
OADD_ | gc.TUINT16,
OADD_ | gc.TINT32,
OADD_ | gc.TUINT32,
OADD_ | gc.TPTR32,
OADD_ | gc.TINT64,
OADD_ | gc.TUINT64,
OADD_ | gc.TPTR64:
a = ppc64.AADD
case OADD_ | gc.TFLOAT32:
a = ppc64.AFADDS
case OADD_ | gc.TFLOAT64:
a = ppc64.AFADD
case OSUB_ | gc.TINT8,
OSUB_ | gc.TUINT8,
OSUB_ | gc.TINT16,
OSUB_ | gc.TUINT16,
OSUB_ | gc.TINT32,
OSUB_ | gc.TUINT32,
OSUB_ | gc.TPTR32,
OSUB_ | gc.TINT64,
OSUB_ | gc.TUINT64,
OSUB_ | gc.TPTR64:
a = ppc64.ASUB
case OSUB_ | gc.TFLOAT32:
a = ppc64.AFSUBS
case OSUB_ | gc.TFLOAT64:
a = ppc64.AFSUB
case OMINUS_ | gc.TINT8,
OMINUS_ | gc.TUINT8,
OMINUS_ | gc.TINT16,
OMINUS_ | gc.TUINT16,
OMINUS_ | gc.TINT32,
OMINUS_ | gc.TUINT32,
OMINUS_ | gc.TPTR32,
OMINUS_ | gc.TINT64,
OMINUS_ | gc.TUINT64,
OMINUS_ | gc.TPTR64:
a = ppc64.ANEG
case OAND_ | gc.TINT8,
OAND_ | gc.TUINT8,
OAND_ | gc.TINT16,
OAND_ | gc.TUINT16,
OAND_ | gc.TINT32,
OAND_ | gc.TUINT32,
OAND_ | gc.TPTR32,
OAND_ | gc.TINT64,
OAND_ | gc.TUINT64,
OAND_ | gc.TPTR64:
a = ppc64.AAND
case OOR_ | gc.TINT8,
OOR_ | gc.TUINT8,
OOR_ | gc.TINT16,
OOR_ | gc.TUINT16,
OOR_ | gc.TINT32,
OOR_ | gc.TUINT32,
OOR_ | gc.TPTR32,
OOR_ | gc.TINT64,
OOR_ | gc.TUINT64,
OOR_ | gc.TPTR64:
a = ppc64.AOR
case OXOR_ | gc.TINT8,
OXOR_ | gc.TUINT8,
OXOR_ | gc.TINT16,
OXOR_ | gc.TUINT16,
OXOR_ | gc.TINT32,
OXOR_ | gc.TUINT32,
OXOR_ | gc.TPTR32,
OXOR_ | gc.TINT64,
OXOR_ | gc.TUINT64,
OXOR_ | gc.TPTR64:
a = ppc64.AXOR
// TODO(minux): handle rotates
//case CASE(OLROT, TINT8):
//case CASE(OLROT, TUINT8):
//case CASE(OLROT, TINT16):
//case CASE(OLROT, TUINT16):
//case CASE(OLROT, TINT32):
//case CASE(OLROT, TUINT32):
//case CASE(OLROT, TPTR32):
//case CASE(OLROT, TINT64):
//case CASE(OLROT, TUINT64):
//case CASE(OLROT, TPTR64):
// a = 0//???; RLDC?
// break;
case OLSH_ | gc.TINT8,
OLSH_ | gc.TUINT8,
OLSH_ | gc.TINT16,
OLSH_ | gc.TUINT16,
OLSH_ | gc.TINT32,
OLSH_ | gc.TUINT32,
OLSH_ | gc.TPTR32,
OLSH_ | gc.TINT64,
OLSH_ | gc.TUINT64,
OLSH_ | gc.TPTR64:
a = ppc64.ASLD
case ORSH_ | gc.TUINT8,
ORSH_ | gc.TUINT16,
ORSH_ | gc.TUINT32,
ORSH_ | gc.TPTR32,
ORSH_ | gc.TUINT64,
ORSH_ | gc.TPTR64:
a = ppc64.ASRD
case ORSH_ | gc.TINT8,
ORSH_ | gc.TINT16,
ORSH_ | gc.TINT32,
ORSH_ | gc.TINT64:
a = ppc64.ASRAD
// TODO(minux): handle rotates
//case CASE(ORROTC, TINT8):
//case CASE(ORROTC, TUINT8):
//case CASE(ORROTC, TINT16):
//case CASE(ORROTC, TUINT16):
//case CASE(ORROTC, TINT32):
//case CASE(ORROTC, TUINT32):
//case CASE(ORROTC, TINT64):
//case CASE(ORROTC, TUINT64):
// a = 0//??? RLDC??
// break;
case OHMUL_ | gc.TINT64:
a = ppc64.AMULHD
case OHMUL_ | gc.TUINT64,
OHMUL_ | gc.TPTR64:
a = ppc64.AMULHDU
case OMUL_ | gc.TINT8,
OMUL_ | gc.TINT16,
OMUL_ | gc.TINT32,
OMUL_ | gc.TINT64:
a = ppc64.AMULLD
case OMUL_ | gc.TUINT8,
OMUL_ | gc.TUINT16,
OMUL_ | gc.TUINT32,
OMUL_ | gc.TPTR32,
// don't use word multiply, the high 32-bit are undefined.
OMUL_ | gc.TUINT64,
OMUL_ | gc.TPTR64:
// for 64-bit multiplies, signedness doesn't matter.
a = ppc64.AMULLD
case OMUL_ | gc.TFLOAT32:
a = ppc64.AFMULS
case OMUL_ | gc.TFLOAT64:
a = ppc64.AFMUL
case ODIV_ | gc.TINT8,
ODIV_ | gc.TINT16,
ODIV_ | gc.TINT32,
ODIV_ | gc.TINT64:
a = ppc64.ADIVD
case ODIV_ | gc.TUINT8,
ODIV_ | gc.TUINT16,
ODIV_ | gc.TUINT32,
ODIV_ | gc.TPTR32,
ODIV_ | gc.TUINT64,
ODIV_ | gc.TPTR64:
a = ppc64.ADIVDU
case ODIV_ | gc.TFLOAT32:
a = ppc64.AFDIVS
case ODIV_ | gc.TFLOAT64:
a = ppc64.AFDIV
}
return a
}
/*
* generate comparison of nl, nr, both 64-bit.
* nl is memory; nr is constant or memory.
*/
func cmp64(nl *gc.Node, nr *gc.Node, op gc.Op, likely int, to *obj.Prog) {
var lo1 gc.Node
var hi1 gc.Node
var lo2 gc.Node
var hi2 gc.Node
var r1 gc.Node
var r2 gc.Node
split64(nl, &lo1, &hi1)
split64(nr, &lo2, &hi2)
// compare most significant word;
// if they differ, we're done.
t := hi1.Type
gc.Regalloc(&r1, gc.Types[gc.TINT32], nil)
gc.Regalloc(&r2, gc.Types[gc.TINT32], nil)
gins(arm.AMOVW, &hi1, &r1)
gins(arm.AMOVW, &hi2, &r2)
gins(arm.ACMP, &r1, &r2)
gc.Regfree(&r1)
gc.Regfree(&r2)
var br *obj.Prog
switch op {
default:
gc.Fatalf("cmp64 %v %v", gc.Oconv(int(op), 0), t)
// cmp hi
// bne L
// cmp lo
// beq to
// L:
case gc.OEQ:
br = gc.Gbranch(arm.ABNE, nil, -likely)
// cmp hi
// bne to
// cmp lo
// bne to
case gc.ONE:
gc.Patch(gc.Gbranch(arm.ABNE, nil, likely), to)
// cmp hi
// bgt to
// blt L
// cmp lo
// bge to (or bgt to)
// L:
case gc.OGE,
gc.OGT:
gc.Patch(gc.Gbranch(optoas(gc.OGT, t), nil, likely), to)
br = gc.Gbranch(optoas(gc.OLT, t), nil, -likely)
// cmp hi
// blt to
// bgt L
// cmp lo
// ble to (or jlt to)
// L:
case gc.OLE,
gc.OLT:
gc.Patch(gc.Gbranch(optoas(gc.OLT, t), nil, likely), to)
br = gc.Gbranch(optoas(gc.OGT, t), nil, -likely)
}
// compare least significant word
t = lo1.Type
gc.Regalloc(&r1, gc.Types[gc.TINT32], nil)
gc.Regalloc(&r2, gc.Types[gc.TINT32], nil)
gins(arm.AMOVW, &lo1, &r1)
gins(arm.AMOVW, &lo2, &r2)
gins(arm.ACMP, &r1, &r2)
gc.Regfree(&r1)
gc.Regfree(&r2)
// jump again
gc.Patch(gc.Gbranch(optoas(op, t), nil, likely), to)
// point first branch down here if appropriate
if br != nil {
gc.Patch(br, gc.Pc)
}
splitclean()
splitclean()
}
/*
* attempt to generate 64-bit
* res = n
* return 1 on success, 0 if op not handled.
*/
func cgen64(n *gc.Node, res *gc.Node) {
if res.Op != gc.OINDREG && res.Op != gc.ONAME {
gc.Dump("n", n)
gc.Dump("res", res)
gc.Fatalf("cgen64 %v of %v", gc.Oconv(int(n.Op), 0), gc.Oconv(int(res.Op), 0))
}
l := n.Left
var t1 gc.Node
if !l.Addable {
gc.Tempname(&t1, l.Type)
gc.Cgen(l, &t1)
l = &t1
}
var hi1 gc.Node
var lo1 gc.Node
split64(l, &lo1, &hi1)
switch n.Op {
default:
gc.Fatalf("cgen64 %v", gc.Oconv(int(n.Op), 0))
case gc.OMINUS:
var lo2 gc.Node
var hi2 gc.Node
split64(res, &lo2, &hi2)
gc.Regalloc(&t1, lo1.Type, nil)
var al gc.Node
gc.Regalloc(&al, lo1.Type, nil)
var ah gc.Node
gc.Regalloc(&ah, hi1.Type, nil)
gins(arm.AMOVW, &lo1, &al)
gins(arm.AMOVW, &hi1, &ah)
gmove(ncon(0), &t1)
p1 := gins(arm.ASUB, &al, &t1)
p1.Scond |= arm.C_SBIT
gins(arm.AMOVW, &t1, &lo2)
gmove(ncon(0), &t1)
gins(arm.ASBC, &ah, &t1)
gins(arm.AMOVW, &t1, &hi2)
gc.Regfree(&t1)
gc.Regfree(&al)
gc.Regfree(&ah)
splitclean()
splitclean()
return
case gc.OCOM:
gc.Regalloc(&t1, lo1.Type, nil)
gmove(ncon(^uint32(0)), &t1)
var lo2 gc.Node
var hi2 gc.Node
split64(res, &lo2, &hi2)
var n1 gc.Node
gc.Regalloc(&n1, lo1.Type, nil)
gins(arm.AMOVW, &lo1, &n1)
gins(arm.AEOR, &t1, &n1)
gins(arm.AMOVW, &n1, &lo2)
gins(arm.AMOVW, &hi1, &n1)
gins(arm.AEOR, &t1, &n1)
gins(arm.AMOVW, &n1, &hi2)
gc.Regfree(&t1)
gc.Regfree(&n1)
splitclean()
splitclean()
return
// binary operators.
// common setup below.
case gc.OADD,
gc.OSUB,
gc.OMUL,
gc.OLSH,
gc.ORSH,
gc.OAND,
gc.OOR,
gc.OXOR,
gc.OLROT:
break
}
// setup for binary operators
r := n.Right
if r != nil && !r.Addable {
var t2 gc.Node
gc.Tempname(&t2, r.Type)
gc.Cgen(r, &t2)
r = &t2
}
var hi2 gc.Node
var lo2 gc.Node
if gc.Is64(r.Type) {
split64(r, &lo2, &hi2)
}
var al gc.Node
gc.Regalloc(&al, lo1.Type, nil)
var ah gc.Node
gc.Regalloc(&ah, hi1.Type, nil)
// Do op. Leave result in ah:al.
switch n.Op {
default:
gc.Fatalf("cgen64: not implemented: %v\n", n)
// TODO: Constants
case gc.OADD:
var bl gc.Node
gc.Regalloc(&bl, gc.Types[gc.TPTR32], nil)
var bh gc.Node
gc.Regalloc(&bh, gc.Types[gc.TPTR32], nil)
gins(arm.AMOVW, &hi1, &ah)
gins(arm.AMOVW, &lo1, &al)
gins(arm.AMOVW, &hi2, &bh)
gins(arm.AMOVW, &lo2, &bl)
p1 := gins(arm.AADD, &bl, &al)
p1.Scond |= arm.C_SBIT
gins(arm.AADC, &bh, &ah)
gc.Regfree(&bl)
gc.Regfree(&bh)
// TODO: Constants.
case gc.OSUB:
var bl gc.Node
gc.Regalloc(&bl, gc.Types[gc.TPTR32], nil)
var bh gc.Node
gc.Regalloc(&bh, gc.Types[gc.TPTR32], nil)
gins(arm.AMOVW, &lo1, &al)
gins(arm.AMOVW, &hi1, &ah)
gins(arm.AMOVW, &lo2, &bl)
gins(arm.AMOVW, &hi2, &bh)
p1 := gins(arm.ASUB, &bl, &al)
p1.Scond |= arm.C_SBIT
gins(arm.ASBC, &bh, &ah)
gc.Regfree(&bl)
gc.Regfree(&bh)
// TODO(kaib): this can be done with 4 regs and does not need 6
case gc.OMUL:
var bl gc.Node
gc.Regalloc(&bl, gc.Types[gc.TPTR32], nil)
var bh gc.Node
gc.Regalloc(&bh, gc.Types[gc.TPTR32], nil)
var cl gc.Node
gc.Regalloc(&cl, gc.Types[gc.TPTR32], nil)
var ch gc.Node
gc.Regalloc(&ch, gc.Types[gc.TPTR32], nil)
// load args into bh:bl and bh:bl.
gins(arm.AMOVW, &hi1, &bh)
gins(arm.AMOVW, &lo1, &bl)
gins(arm.AMOVW, &hi2, &ch)
gins(arm.AMOVW, &lo2, &cl)
// bl * cl -> ah al
p1 := gins(arm.AMULLU, nil, nil)
p1.From.Type = obj.TYPE_REG
p1.From.Reg = bl.Reg
p1.Reg = cl.Reg
p1.To.Type = obj.TYPE_REGREG
p1.To.Reg = ah.Reg
p1.To.Offset = int64(al.Reg)
//print("%v\n", p1);
// bl * ch + ah -> ah
p1 = gins(arm.AMULA, nil, nil)
p1.From.Type = obj.TYPE_REG
p1.From.Reg = bl.Reg
p1.Reg = ch.Reg
p1.To.Type = obj.TYPE_REGREG2
p1.To.Reg = ah.Reg
p1.To.Offset = int64(ah.Reg)
//print("%v\n", p1);
// bh * cl + ah -> ah
p1 = gins(arm.AMULA, nil, nil)
p1.From.Type = obj.TYPE_REG
p1.From.Reg = bh.Reg
p1.Reg = cl.Reg
p1.To.Type = obj.TYPE_REGREG2
p1.To.Reg = ah.Reg
p1.To.Offset = int64(ah.Reg)
//print("%v\n", p1);
gc.Regfree(&bh)
gc.Regfree(&bl)
gc.Regfree(&ch)
gc.Regfree(&cl)
// We only rotate by a constant c in [0,64).
// if c >= 32:
// lo, hi = hi, lo
// c -= 32
// if c == 0:
// no-op
// else:
// t = hi
// shld hi:lo, c
// shld lo:t, c
case gc.OLROT:
v := uint64(r.Int())
var bl gc.Node
gc.Regalloc(&bl, lo1.Type, nil)
var bh gc.Node
gc.Regalloc(&bh, hi1.Type, nil)
if v >= 32 {
// reverse during load to do the first 32 bits of rotate
v -= 32
gins(arm.AMOVW, &hi1, &bl)
gins(arm.AMOVW, &lo1, &bh)
} else {
gins(arm.AMOVW, &hi1, &bh)
gins(arm.AMOVW, &lo1, &bl)
}
if v == 0 {
gins(arm.AMOVW, &bh, &ah)
gins(arm.AMOVW, &bl, &al)
} else {
// rotate by 1 <= v <= 31
// MOVW bl<<v, al
// MOVW bh<<v, ah
// OR bl>>(32-v), ah
// OR bh>>(32-v), al
gshift(arm.AMOVW, &bl, arm.SHIFT_LL, int32(v), &al)
gshift(arm.AMOVW, &bh, arm.SHIFT_LL, int32(v), &ah)
gshift(arm.AORR, &bl, arm.SHIFT_LR, int32(32-v), &ah)
gshift(arm.AORR, &bh, arm.SHIFT_LR, int32(32-v), &al)
}
gc.Regfree(&bl)
gc.Regfree(&bh)
case gc.OLSH:
var bl gc.Node
gc.Regalloc(&bl, lo1.Type, nil)
var bh gc.Node
gc.Regalloc(&bh, hi1.Type, nil)
gins(arm.AMOVW, &hi1, &bh)
gins(arm.AMOVW, &lo1, &bl)
var p6 *obj.Prog
var s gc.Node
var n1 gc.Node
var creg gc.Node
var p1 *obj.Prog
var p2 *obj.Prog
var p3 *obj.Prog
var p4 *obj.Prog
var p5 *obj.Prog
if r.Op == gc.OLITERAL {
v := uint64(r.Int())
if v >= 64 {
// TODO(kaib): replace with gins(AMOVW, nodintconst(0), &al)
// here and below (verify it optimizes to EOR)
gins(arm.AEOR, &al, &al)
gins(arm.AEOR, &ah, &ah)
} else if v > 32 {
gins(arm.AEOR, &al, &al)
// MOVW bl<<(v-32), ah
gshift(arm.AMOVW, &bl, arm.SHIFT_LL, int32(v-32), &ah)
} else if v == 32 {
gins(arm.AEOR, &al, &al)
gins(arm.AMOVW, &bl, &ah)
} else if v > 0 {
// MOVW bl<<v, al
gshift(arm.AMOVW, &bl, arm.SHIFT_LL, int32(v), &al)
// MOVW bh<<v, ah
gshift(arm.AMOVW, &bh, arm.SHIFT_LL, int32(v), &ah)
// OR bl>>(32-v), ah
gshift(arm.AORR, &bl, arm.SHIFT_LR, int32(32-v), &ah)
} else {
gins(arm.AMOVW, &bl, &al)
gins(arm.AMOVW, &bh, &ah)
}
goto olsh_break
}
gc.Regalloc(&s, gc.Types[gc.TUINT32], nil)
gc.Regalloc(&creg, gc.Types[gc.TUINT32], nil)
if gc.Is64(r.Type) {
// shift is >= 1<<32
var cl gc.Node
var ch gc.Node
split64(r, &cl, &ch)
gmove(&ch, &s)
gins(arm.ATST, &s, nil)
p6 = gc.Gbranch(arm.ABNE, nil, 0)
gmove(&cl, &s)
splitclean()
} else {
gmove(r, &s)
p6 = nil
}
gins(arm.ATST, &s, nil)
// shift == 0
p1 = gins(arm.AMOVW, &bl, &al)
p1.Scond = arm.C_SCOND_EQ
p1 = gins(arm.AMOVW, &bh, &ah)
p1.Scond = arm.C_SCOND_EQ
p2 = gc.Gbranch(arm.ABEQ, nil, 0)
// shift is < 32
gc.Nodconst(&n1, gc.Types[gc.TUINT32], 32)
gmove(&n1, &creg)
gins(arm.ACMP, &s, &creg)
// MOVW.LO bl<<s, al
p1 = gregshift(arm.AMOVW, &bl, arm.SHIFT_LL, &s, &al)
p1.Scond = arm.C_SCOND_LO
// MOVW.LO bh<<s, ah
p1 = gregshift(arm.AMOVW, &bh, arm.SHIFT_LL, &s, &ah)
p1.Scond = arm.C_SCOND_LO
// SUB.LO s, creg
p1 = gins(arm.ASUB, &s, &creg)
p1.Scond = arm.C_SCOND_LO
// OR.LO bl>>creg, ah
p1 = gregshift(arm.AORR, &bl, arm.SHIFT_LR, &creg, &ah)
p1.Scond = arm.C_SCOND_LO
// BLO end
p3 = gc.Gbranch(arm.ABLO, nil, 0)
// shift == 32
p1 = gins(arm.AEOR, &al, &al)
p1.Scond = arm.C_SCOND_EQ
p1 = gins(arm.AMOVW, &bl, &ah)
p1.Scond = arm.C_SCOND_EQ
p4 = gc.Gbranch(arm.ABEQ, nil, 0)
// shift is < 64
gc.Nodconst(&n1, gc.Types[gc.TUINT32], 64)
gmove(&n1, &creg)
gins(arm.ACMP, &s, &creg)
// EOR.LO al, al
p1 = gins(arm.AEOR, &al, &al)
p1.Scond = arm.C_SCOND_LO
// MOVW.LO creg>>1, creg
p1 = gshift(arm.AMOVW, &creg, arm.SHIFT_LR, 1, &creg)
p1.Scond = arm.C_SCOND_LO
// SUB.LO creg, s
p1 = gins(arm.ASUB, &creg, &s)
p1.Scond = arm.C_SCOND_LO
// MOVW bl<<s, ah
p1 = gregshift(arm.AMOVW, &bl, arm.SHIFT_LL, &s, &ah)
p1.Scond = arm.C_SCOND_LO
p5 = gc.Gbranch(arm.ABLO, nil, 0)
// shift >= 64
if p6 != nil {
gc.Patch(p6, gc.Pc)
}
gins(arm.AEOR, &al, &al)
gins(arm.AEOR, &ah, &ah)
gc.Patch(p2, gc.Pc)
gc.Patch(p3, gc.Pc)
gc.Patch(p4, gc.Pc)
gc.Patch(p5, gc.Pc)
gc.Regfree(&s)
gc.Regfree(&creg)
olsh_break:
gc.Regfree(&bl)
gc.Regfree(&bh)
case gc.ORSH:
var bl gc.Node
gc.Regalloc(&bl, lo1.Type, nil)
var bh gc.Node
gc.Regalloc(&bh, hi1.Type, nil)
gins(arm.AMOVW, &hi1, &bh)
gins(arm.AMOVW, &lo1, &bl)
var p4 *obj.Prog
var p5 *obj.Prog
var n1 gc.Node
var p6 *obj.Prog
var s gc.Node
var p1 *obj.Prog
var p2 *obj.Prog
var creg gc.Node
var p3 *obj.Prog
if r.Op == gc.OLITERAL {
v := uint64(r.Int())
if v >= 64 {
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->31, al
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &al)
// MOVW bh->31, ah
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &ah)
} else {
gins(arm.AEOR, &al, &al)
gins(arm.AEOR, &ah, &ah)
}
} else if v > 32 {
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->(v-32), al
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, int32(v-32), &al)
// MOVW bh->31, ah
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &ah)
} else {
// MOVW bh>>(v-32), al
gshift(arm.AMOVW, &bh, arm.SHIFT_LR, int32(v-32), &al)
gins(arm.AEOR, &ah, &ah)
}
} else if v == 32 {
gins(arm.AMOVW, &bh, &al)
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->31, ah
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &ah)
} else {
gins(arm.AEOR, &ah, &ah)
}
} else if v > 0 {
// MOVW bl>>v, al
gshift(arm.AMOVW, &bl, arm.SHIFT_LR, int32(v), &al)
// OR bh<<(32-v), al
gshift(arm.AORR, &bh, arm.SHIFT_LL, int32(32-v), &al)
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->v, ah
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, int32(v), &ah)
} else {
// MOVW bh>>v, ah
gshift(arm.AMOVW, &bh, arm.SHIFT_LR, int32(v), &ah)
}
} else {
gins(arm.AMOVW, &bl, &al)
gins(arm.AMOVW, &bh, &ah)
}
goto orsh_break
}
gc.Regalloc(&s, gc.Types[gc.TUINT32], nil)
gc.Regalloc(&creg, gc.Types[gc.TUINT32], nil)
if gc.Is64(r.Type) {
// shift is >= 1<<32
var ch gc.Node
var cl gc.Node
split64(r, &cl, &ch)
gmove(&ch, &s)
gins(arm.ATST, &s, nil)
var p1 *obj.Prog
if bh.Type.Etype == gc.TINT32 {
p1 = gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &ah)
} else {
p1 = gins(arm.AEOR, &ah, &ah)
}
p1.Scond = arm.C_SCOND_NE
p6 = gc.Gbranch(arm.ABNE, nil, 0)
gmove(&cl, &s)
splitclean()
} else {
gmove(r, &s)
p6 = nil
}
gins(arm.ATST, &s, nil)
// shift == 0
p1 = gins(arm.AMOVW, &bl, &al)
p1.Scond = arm.C_SCOND_EQ
p1 = gins(arm.AMOVW, &bh, &ah)
p1.Scond = arm.C_SCOND_EQ
p2 = gc.Gbranch(arm.ABEQ, nil, 0)
// check if shift is < 32
gc.Nodconst(&n1, gc.Types[gc.TUINT32], 32)
gmove(&n1, &creg)
gins(arm.ACMP, &s, &creg)
// MOVW.LO bl>>s, al
p1 = gregshift(arm.AMOVW, &bl, arm.SHIFT_LR, &s, &al)
p1.Scond = arm.C_SCOND_LO
// SUB.LO s,creg
p1 = gins(arm.ASUB, &s, &creg)
p1.Scond = arm.C_SCOND_LO
// OR.LO bh<<(32-s), al
p1 = gregshift(arm.AORR, &bh, arm.SHIFT_LL, &creg, &al)
p1.Scond = arm.C_SCOND_LO
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->s, ah
p1 = gregshift(arm.AMOVW, &bh, arm.SHIFT_AR, &s, &ah)
} else {
// MOVW bh>>s, ah
p1 = gregshift(arm.AMOVW, &bh, arm.SHIFT_LR, &s, &ah)
}
p1.Scond = arm.C_SCOND_LO
// BLO end
p3 = gc.Gbranch(arm.ABLO, nil, 0)
// shift == 32
p1 = gins(arm.AMOVW, &bh, &al)
p1.Scond = arm.C_SCOND_EQ
if bh.Type.Etype == gc.TINT32 {
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &ah)
} else {
gins(arm.AEOR, &ah, &ah)
}
p4 = gc.Gbranch(arm.ABEQ, nil, 0)
// check if shift is < 64
gc.Nodconst(&n1, gc.Types[gc.TUINT32], 64)
gmove(&n1, &creg)
gins(arm.ACMP, &s, &creg)
// MOVW.LO creg>>1, creg
p1 = gshift(arm.AMOVW, &creg, arm.SHIFT_LR, 1, &creg)
p1.Scond = arm.C_SCOND_LO
// SUB.LO creg, s
p1 = gins(arm.ASUB, &creg, &s)
p1.Scond = arm.C_SCOND_LO
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->(s-32), al
p1 := gregshift(arm.AMOVW, &bh, arm.SHIFT_AR, &s, &al)
p1.Scond = arm.C_SCOND_LO
} else {
// MOVW bh>>(v-32), al
p1 := gregshift(arm.AMOVW, &bh, arm.SHIFT_LR, &s, &al)
p1.Scond = arm.C_SCOND_LO
}
// BLO end
p5 = gc.Gbranch(arm.ABLO, nil, 0)
// s >= 64
if p6 != nil {
gc.Patch(p6, gc.Pc)
}
if bh.Type.Etype == gc.TINT32 {
// MOVW bh->31, al
gshift(arm.AMOVW, &bh, arm.SHIFT_AR, 31, &al)
} else {
gins(arm.AEOR, &al, &al)
}
gc.Patch(p2, gc.Pc)
gc.Patch(p3, gc.Pc)
gc.Patch(p4, gc.Pc)
gc.Patch(p5, gc.Pc)
gc.Regfree(&s)
gc.Regfree(&creg)
orsh_break:
gc.Regfree(&bl)
gc.Regfree(&bh)
// TODO(kaib): literal optimizations
// make constant the right side (it usually is anyway).
// if(lo1.op == OLITERAL) {
// nswap(&lo1, &lo2);
// nswap(&hi1, &hi2);
// }
// if(lo2.op == OLITERAL) {
// // special cases for constants.
// lv = mpgetfix(lo2.val.u.xval);
// hv = mpgetfix(hi2.val.u.xval);
// splitclean(); // right side
// split64(res, &lo2, &hi2);
// switch(n->op) {
// case OXOR:
// gmove(&lo1, &lo2);
// gmove(&hi1, &hi2);
// switch(lv) {
// case 0:
// break;
// case 0xffffffffu:
// gins(ANOTL, N, &lo2);
// break;
// default:
// gins(AXORL, ncon(lv), &lo2);
// break;
// }
// switch(hv) {
// case 0:
// break;
// case 0xffffffffu:
// gins(ANOTL, N, &hi2);
// break;
// default:
// gins(AXORL, ncon(hv), &hi2);
// break;
// }
// break;
// case OAND:
// switch(lv) {
// case 0:
// gins(AMOVL, ncon(0), &lo2);
// break;
// default:
// gmove(&lo1, &lo2);
// if(lv != 0xffffffffu)
// gins(AANDL, ncon(lv), &lo2);
// break;
// }
// switch(hv) {
// case 0:
// gins(AMOVL, ncon(0), &hi2);
// break;
// default:
// gmove(&hi1, &hi2);
// if(hv != 0xffffffffu)
// gins(AANDL, ncon(hv), &hi2);
// break;
// }
// break;
// case OOR:
// switch(lv) {
// case 0:
// gmove(&lo1, &lo2);
// break;
// case 0xffffffffu:
// gins(AMOVL, ncon(0xffffffffu), &lo2);
// break;
// default:
// gmove(&lo1, &lo2);
// gins(AORL, ncon(lv), &lo2);
// break;
// }
// switch(hv) {
// case 0:
// gmove(&hi1, &hi2);
// break;
// case 0xffffffffu:
// gins(AMOVL, ncon(0xffffffffu), &hi2);
// break;
// default:
// gmove(&hi1, &hi2);
// gins(AORL, ncon(hv), &hi2);
// break;
// }
// break;
// }
// splitclean();
// splitclean();
// goto out;
// }
case gc.OXOR,
gc.OAND,
gc.OOR:
var n1 gc.Node
gc.Regalloc(&n1, lo1.Type, nil)
gins(arm.AMOVW, &lo1, &al)
gins(arm.AMOVW, &hi1, &ah)
gins(arm.AMOVW, &lo2, &n1)
gins(optoas(n.Op, lo1.Type), &n1, &al)
gins(arm.AMOVW, &hi2, &n1)
gins(optoas(n.Op, lo1.Type), &n1, &ah)
gc.Regfree(&n1)
}
if gc.Is64(r.Type) {
splitclean()
}
splitclean()
split64(res, &lo1, &hi1)
gins(arm.AMOVW, &al, &lo1)
gins(arm.AMOVW, &ah, &hi1)
splitclean()
//out:
gc.Regfree(&al)
gc.Regfree(&ah)
}
/*
* return Axxx for Oxxx on type t.
*/
func optoas(op gc.Op, t *gc.Type) int {
if t == nil {
gc.Fatalf("optoas: t is nil")
}
// avoid constant conversions in switches below
const (
OMINUS_ = uint32(gc.OMINUS) << 16
OLSH_ = uint32(gc.OLSH) << 16
ORSH_ = uint32(gc.ORSH) << 16
OADD_ = uint32(gc.OADD) << 16
OSUB_ = uint32(gc.OSUB) << 16
OMUL_ = uint32(gc.OMUL) << 16
ODIV_ = uint32(gc.ODIV) << 16
OMOD_ = uint32(gc.OMOD) << 16
OOR_ = uint32(gc.OOR) << 16
OAND_ = uint32(gc.OAND) << 16
OXOR_ = uint32(gc.OXOR) << 16
OEQ_ = uint32(gc.OEQ) << 16
ONE_ = uint32(gc.ONE) << 16
OLT_ = uint32(gc.OLT) << 16
OLE_ = uint32(gc.OLE) << 16
OGE_ = uint32(gc.OGE) << 16
OGT_ = uint32(gc.OGT) << 16
OCMP_ = uint32(gc.OCMP) << 16
OPS_ = uint32(gc.OPS) << 16
OPC_ = uint32(gc.OPC) << 16
OAS_ = uint32(gc.OAS) << 16
OHMUL_ = uint32(gc.OHMUL) << 16
OSQRT_ = uint32(gc.OSQRT) << 16
OADDR_ = uint32(gc.OADDR) << 16
OINC_ = uint32(gc.OINC) << 16
ODEC_ = uint32(gc.ODEC) << 16
OLROT_ = uint32(gc.OLROT) << 16
ORROTC_ = uint32(gc.ORROTC) << 16
OEXTEND_ = uint32(gc.OEXTEND) << 16
)
a := obj.AXXX
switch uint32(op)<<16 | uint32(gc.Simtype[t.Etype]) {
default:
gc.Fatalf("optoas: no entry %v-%v", gc.Oconv(int(op), 0), t)
case OADDR_ | gc.TPTR32:
a = x86.ALEAL
case OADDR_ | gc.TPTR64:
a = x86.ALEAQ
case OEQ_ | gc.TBOOL,
OEQ_ | gc.TINT8,
OEQ_ | gc.TUINT8,
OEQ_ | gc.TINT16,
OEQ_ | gc.TUINT16,
OEQ_ | gc.TINT32,
OEQ_ | gc.TUINT32,
OEQ_ | gc.TINT64,
OEQ_ | gc.TUINT64,
OEQ_ | gc.TPTR32,
OEQ_ | gc.TPTR64,
OEQ_ | gc.TFLOAT32,
OEQ_ | gc.TFLOAT64:
a = x86.AJEQ
case ONE_ | gc.TBOOL,
ONE_ | gc.TINT8,
ONE_ | gc.TUINT8,
ONE_ | gc.TINT16,
ONE_ | gc.TUINT16,
ONE_ | gc.TINT32,
ONE_ | gc.TUINT32,
ONE_ | gc.TINT64,
ONE_ | gc.TUINT64,
ONE_ | gc.TPTR32,
ONE_ | gc.TPTR64,
ONE_ | gc.TFLOAT32,
ONE_ | gc.TFLOAT64:
a = x86.AJNE
case OPS_ | gc.TBOOL,
OPS_ | gc.TINT8,
OPS_ | gc.TUINT8,
OPS_ | gc.TINT16,
OPS_ | gc.TUINT16,
OPS_ | gc.TINT32,
OPS_ | gc.TUINT32,
OPS_ | gc.TINT64,
OPS_ | gc.TUINT64,
OPS_ | gc.TPTR32,
OPS_ | gc.TPTR64,
OPS_ | gc.TFLOAT32,
OPS_ | gc.TFLOAT64:
a = x86.AJPS
case OPC_ | gc.TBOOL,
OPC_ | gc.TINT8,
OPC_ | gc.TUINT8,
OPC_ | gc.TINT16,
OPC_ | gc.TUINT16,
OPC_ | gc.TINT32,
OPC_ | gc.TUINT32,
OPC_ | gc.TINT64,
OPC_ | gc.TUINT64,
OPC_ | gc.TPTR32,
OPC_ | gc.TPTR64,
OPC_ | gc.TFLOAT32,
OPC_ | gc.TFLOAT64:
a = x86.AJPC
case OLT_ | gc.TINT8,
OLT_ | gc.TINT16,
OLT_ | gc.TINT32,
OLT_ | gc.TINT64:
a = x86.AJLT
case OLT_ | gc.TUINT8,
OLT_ | gc.TUINT16,
OLT_ | gc.TUINT32,
OLT_ | gc.TUINT64:
a = x86.AJCS
case OLE_ | gc.TINT8,
OLE_ | gc.TINT16,
OLE_ | gc.TINT32,
OLE_ | gc.TINT64:
a = x86.AJLE
case OLE_ | gc.TUINT8,
OLE_ | gc.TUINT16,
OLE_ | gc.TUINT32,
OLE_ | gc.TUINT64:
a = x86.AJLS
case OGT_ | gc.TINT8,
OGT_ | gc.TINT16,
OGT_ | gc.TINT32,
OGT_ | gc.TINT64:
a = x86.AJGT
case OGT_ | gc.TUINT8,
OGT_ | gc.TUINT16,
OGT_ | gc.TUINT32,
OGT_ | gc.TUINT64,
OLT_ | gc.TFLOAT32,
OLT_ | gc.TFLOAT64:
a = x86.AJHI
case OGE_ | gc.TINT8,
OGE_ | gc.TINT16,
OGE_ | gc.TINT32,
OGE_ | gc.TINT64:
a = x86.AJGE
case OGE_ | gc.TUINT8,
OGE_ | gc.TUINT16,
OGE_ | gc.TUINT32,
OGE_ | gc.TUINT64,
OLE_ | gc.TFLOAT32,
OLE_ | gc.TFLOAT64:
a = x86.AJCC
case OCMP_ | gc.TBOOL,
OCMP_ | gc.TINT8,
OCMP_ | gc.TUINT8:
a = x86.ACMPB
case OCMP_ | gc.TINT16,
OCMP_ | gc.TUINT16:
a = x86.ACMPW
case OCMP_ | gc.TINT32,
OCMP_ | gc.TUINT32,
OCMP_ | gc.TPTR32:
a = x86.ACMPL
case OCMP_ | gc.TINT64,
OCMP_ | gc.TUINT64,
OCMP_ | gc.TPTR64:
a = x86.ACMPQ
case OCMP_ | gc.TFLOAT32:
a = x86.AUCOMISS
case OCMP_ | gc.TFLOAT64:
a = x86.AUCOMISD
case OAS_ | gc.TBOOL,
OAS_ | gc.TINT8,
OAS_ | gc.TUINT8:
a = x86.AMOVB
case OAS_ | gc.TINT16,
OAS_ | gc.TUINT16:
a = x86.AMOVW
case OAS_ | gc.TINT32,
OAS_ | gc.TUINT32,
OAS_ | gc.TPTR32:
a = x86.AMOVL
case OAS_ | gc.TINT64,
OAS_ | gc.TUINT64,
OAS_ | gc.TPTR64:
a = x86.AMOVQ
case OAS_ | gc.TFLOAT32:
a = x86.AMOVSS
case OAS_ | gc.TFLOAT64:
a = x86.AMOVSD
case OADD_ | gc.TINT8,
OADD_ | gc.TUINT8:
a = x86.AADDB
case OADD_ | gc.TINT16,
OADD_ | gc.TUINT16:
a = x86.AADDW
case OADD_ | gc.TINT32,
OADD_ | gc.TUINT32,
OADD_ | gc.TPTR32:
a = x86.AADDL
case OADD_ | gc.TINT64,
OADD_ | gc.TUINT64,
OADD_ | gc.TPTR64:
a = x86.AADDQ
case OADD_ | gc.TFLOAT32:
a = x86.AADDSS
case OADD_ | gc.TFLOAT64:
a = x86.AADDSD
case OSUB_ | gc.TINT8,
OSUB_ | gc.TUINT8:
a = x86.ASUBB
case OSUB_ | gc.TINT16,
OSUB_ | gc.TUINT16:
a = x86.ASUBW
case OSUB_ | gc.TINT32,
OSUB_ | gc.TUINT32,
OSUB_ | gc.TPTR32:
a = x86.ASUBL
case OSUB_ | gc.TINT64,
OSUB_ | gc.TUINT64,
OSUB_ | gc.TPTR64:
a = x86.ASUBQ
case OSUB_ | gc.TFLOAT32:
a = x86.ASUBSS
case OSUB_ | gc.TFLOAT64:
a = x86.ASUBSD
case OINC_ | gc.TINT8,
OINC_ | gc.TUINT8:
a = x86.AINCB
case OINC_ | gc.TINT16,
OINC_ | gc.TUINT16:
a = x86.AINCW
case OINC_ | gc.TINT32,
OINC_ | gc.TUINT32,
OINC_ | gc.TPTR32:
a = x86.AINCL
case OINC_ | gc.TINT64,
OINC_ | gc.TUINT64,
OINC_ | gc.TPTR64:
a = x86.AINCQ
case ODEC_ | gc.TINT8,
ODEC_ | gc.TUINT8:
a = x86.ADECB
case ODEC_ | gc.TINT16,
ODEC_ | gc.TUINT16:
a = x86.ADECW
case ODEC_ | gc.TINT32,
ODEC_ | gc.TUINT32,
ODEC_ | gc.TPTR32:
a = x86.ADECL
case ODEC_ | gc.TINT64,
ODEC_ | gc.TUINT64,
ODEC_ | gc.TPTR64:
a = x86.ADECQ
case OMINUS_ | gc.TINT8,
OMINUS_ | gc.TUINT8:
a = x86.ANEGB
case OMINUS_ | gc.TINT16,
OMINUS_ | gc.TUINT16:
a = x86.ANEGW
case OMINUS_ | gc.TINT32,
OMINUS_ | gc.TUINT32,
OMINUS_ | gc.TPTR32:
a = x86.ANEGL
case OMINUS_ | gc.TINT64,
OMINUS_ | gc.TUINT64,
OMINUS_ | gc.TPTR64:
a = x86.ANEGQ
case OAND_ | gc.TBOOL,
OAND_ | gc.TINT8,
OAND_ | gc.TUINT8:
a = x86.AANDB
case OAND_ | gc.TINT16,
OAND_ | gc.TUINT16:
a = x86.AANDW
case OAND_ | gc.TINT32,
OAND_ | gc.TUINT32,
OAND_ | gc.TPTR32:
a = x86.AANDL
case OAND_ | gc.TINT64,
OAND_ | gc.TUINT64,
OAND_ | gc.TPTR64:
a = x86.AANDQ
case OOR_ | gc.TBOOL,
OOR_ | gc.TINT8,
OOR_ | gc.TUINT8:
a = x86.AORB
case OOR_ | gc.TINT16,
OOR_ | gc.TUINT16:
a = x86.AORW
case OOR_ | gc.TINT32,
OOR_ | gc.TUINT32,
OOR_ | gc.TPTR32:
a = x86.AORL
case OOR_ | gc.TINT64,
OOR_ | gc.TUINT64,
OOR_ | gc.TPTR64:
a = x86.AORQ
case OXOR_ | gc.TINT8,
OXOR_ | gc.TUINT8:
a = x86.AXORB
case OXOR_ | gc.TINT16,
OXOR_ | gc.TUINT16:
a = x86.AXORW
case OXOR_ | gc.TINT32,
OXOR_ | gc.TUINT32,
OXOR_ | gc.TPTR32:
a = x86.AXORL
case OXOR_ | gc.TINT64,
OXOR_ | gc.TUINT64,
OXOR_ | gc.TPTR64:
a = x86.AXORQ
case OLROT_ | gc.TINT8,
OLROT_ | gc.TUINT8:
a = x86.AROLB
case OLROT_ | gc.TINT16,
OLROT_ | gc.TUINT16:
a = x86.AROLW
case OLROT_ | gc.TINT32,
OLROT_ | gc.TUINT32,
OLROT_ | gc.TPTR32:
a = x86.AROLL
case OLROT_ | gc.TINT64,
OLROT_ | gc.TUINT64,
OLROT_ | gc.TPTR64:
a = x86.AROLQ
case OLSH_ | gc.TINT8,
OLSH_ | gc.TUINT8:
a = x86.ASHLB
case OLSH_ | gc.TINT16,
OLSH_ | gc.TUINT16:
a = x86.ASHLW
case OLSH_ | gc.TINT32,
OLSH_ | gc.TUINT32,
OLSH_ | gc.TPTR32:
a = x86.ASHLL
case OLSH_ | gc.TINT64,
OLSH_ | gc.TUINT64,
OLSH_ | gc.TPTR64:
a = x86.ASHLQ
case ORSH_ | gc.TUINT8:
a = x86.ASHRB
case ORSH_ | gc.TUINT16:
a = x86.ASHRW
case ORSH_ | gc.TUINT32,
ORSH_ | gc.TPTR32:
a = x86.ASHRL
case ORSH_ | gc.TUINT64,
ORSH_ | gc.TPTR64:
a = x86.ASHRQ
case ORSH_ | gc.TINT8:
a = x86.ASARB
case ORSH_ | gc.TINT16:
a = x86.ASARW
case ORSH_ | gc.TINT32:
a = x86.ASARL
case ORSH_ | gc.TINT64:
a = x86.ASARQ
case ORROTC_ | gc.TINT8,
ORROTC_ | gc.TUINT8:
a = x86.ARCRB
case ORROTC_ | gc.TINT16,
ORROTC_ | gc.TUINT16:
a = x86.ARCRW
case ORROTC_ | gc.TINT32,
ORROTC_ | gc.TUINT32:
a = x86.ARCRL
case ORROTC_ | gc.TINT64,
ORROTC_ | gc.TUINT64:
a = x86.ARCRQ
case OHMUL_ | gc.TINT8,
OMUL_ | gc.TINT8,
OMUL_ | gc.TUINT8:
a = x86.AIMULB
case OHMUL_ | gc.TINT16,
OMUL_ | gc.TINT16,
OMUL_ | gc.TUINT16:
a = x86.AIMULW
case OHMUL_ | gc.TINT32,
OMUL_ | gc.TINT32,
OMUL_ | gc.TUINT32,
OMUL_ | gc.TPTR32:
a = x86.AIMULL
case OHMUL_ | gc.TINT64,
OMUL_ | gc.TINT64,
OMUL_ | gc.TUINT64,
OMUL_ | gc.TPTR64:
a = x86.AIMULQ
case OHMUL_ | gc.TUINT8:
a = x86.AMULB
case OHMUL_ | gc.TUINT16:
a = x86.AMULW
case OHMUL_ | gc.TUINT32,
OHMUL_ | gc.TPTR32:
a = x86.AMULL
case OHMUL_ | gc.TUINT64,
OHMUL_ | gc.TPTR64:
a = x86.AMULQ
case OMUL_ | gc.TFLOAT32:
a = x86.AMULSS
case OMUL_ | gc.TFLOAT64:
a = x86.AMULSD
case ODIV_ | gc.TINT8,
OMOD_ | gc.TINT8:
a = x86.AIDIVB
case ODIV_ | gc.TUINT8,
OMOD_ | gc.TUINT8:
a = x86.ADIVB
case ODIV_ | gc.TINT16,
OMOD_ | gc.TINT16:
a = x86.AIDIVW
case ODIV_ | gc.TUINT16,
OMOD_ | gc.TUINT16:
a = x86.ADIVW
case ODIV_ | gc.TINT32,
OMOD_ | gc.TINT32:
a = x86.AIDIVL
case ODIV_ | gc.TUINT32,
ODIV_ | gc.TPTR32,
OMOD_ | gc.TUINT32,
OMOD_ | gc.TPTR32:
a = x86.ADIVL
case ODIV_ | gc.TINT64,
OMOD_ | gc.TINT64:
a = x86.AIDIVQ
case ODIV_ | gc.TUINT64,
ODIV_ | gc.TPTR64,
OMOD_ | gc.TUINT64,
OMOD_ | gc.TPTR64:
a = x86.ADIVQ
case OEXTEND_ | gc.TINT16:
a = x86.ACWD
case OEXTEND_ | gc.TINT32:
a = x86.ACDQ
case OEXTEND_ | gc.TINT64:
a = x86.ACQO
case ODIV_ | gc.TFLOAT32:
a = x86.ADIVSS
case ODIV_ | gc.TFLOAT64:
a = x86.ADIVSD
case OSQRT_ | gc.TFLOAT64:
a = x86.ASQRTSD
}
return a
}
/*
* return Axxx for Oxxx on type t.
*/
func optoas(op gc.Op, t *gc.Type) int {
if t == nil {
gc.Fatalf("optoas: t is nil")
}
// avoid constant conversions in switches below
const (
OMINUS_ = uint32(gc.OMINUS) << 16
OLSH_ = uint32(gc.OLSH) << 16
ORSH_ = uint32(gc.ORSH) << 16
OADD_ = uint32(gc.OADD) << 16
OSUB_ = uint32(gc.OSUB) << 16
OMUL_ = uint32(gc.OMUL) << 16
ODIV_ = uint32(gc.ODIV) << 16
OMOD_ = uint32(gc.OMOD) << 16
OOR_ = uint32(gc.OOR) << 16
OAND_ = uint32(gc.OAND) << 16
OXOR_ = uint32(gc.OXOR) << 16
OEQ_ = uint32(gc.OEQ) << 16
ONE_ = uint32(gc.ONE) << 16
OLT_ = uint32(gc.OLT) << 16
OLE_ = uint32(gc.OLE) << 16
OGE_ = uint32(gc.OGE) << 16
OGT_ = uint32(gc.OGT) << 16
OCMP_ = uint32(gc.OCMP) << 16
OPS_ = uint32(gc.OPS) << 16
OAS_ = uint32(gc.OAS) << 16
OSQRT_ = uint32(gc.OSQRT) << 16
)
a := obj.AXXX
switch uint32(op)<<16 | uint32(gc.Simtype[t.Etype]) {
default:
gc.Fatalf("optoas: no entry %v-%v etype %v simtype %v", gc.Oconv(int(op), 0), t, gc.Types[t.Etype], gc.Types[gc.Simtype[t.Etype]])
/* case CASE(OADDR, TPTR32):
a = ALEAL;
break;
case CASE(OADDR, TPTR64):
a = ALEAQ;
break;
*/
// TODO(kaib): make sure the conditional branches work on all edge cases
case OEQ_ | gc.TBOOL,
OEQ_ | gc.TINT8,
OEQ_ | gc.TUINT8,
OEQ_ | gc.TINT16,
OEQ_ | gc.TUINT16,
OEQ_ | gc.TINT32,
OEQ_ | gc.TUINT32,
OEQ_ | gc.TINT64,
OEQ_ | gc.TUINT64,
OEQ_ | gc.TPTR32,
OEQ_ | gc.TPTR64,
OEQ_ | gc.TFLOAT32,
OEQ_ | gc.TFLOAT64:
a = arm.ABEQ
case ONE_ | gc.TBOOL,
ONE_ | gc.TINT8,
ONE_ | gc.TUINT8,
ONE_ | gc.TINT16,
ONE_ | gc.TUINT16,
ONE_ | gc.TINT32,
ONE_ | gc.TUINT32,
ONE_ | gc.TINT64,
ONE_ | gc.TUINT64,
ONE_ | gc.TPTR32,
ONE_ | gc.TPTR64,
ONE_ | gc.TFLOAT32,
ONE_ | gc.TFLOAT64:
a = arm.ABNE
case OLT_ | gc.TINT8,
OLT_ | gc.TINT16,
OLT_ | gc.TINT32,
OLT_ | gc.TINT64,
OLT_ | gc.TFLOAT32,
OLT_ | gc.TFLOAT64:
a = arm.ABLT
case OLT_ | gc.TUINT8,
OLT_ | gc.TUINT16,
OLT_ | gc.TUINT32,
OLT_ | gc.TUINT64:
a = arm.ABLO
case OLE_ | gc.TINT8,
OLE_ | gc.TINT16,
OLE_ | gc.TINT32,
OLE_ | gc.TINT64,
OLE_ | gc.TFLOAT32,
OLE_ | gc.TFLOAT64:
a = arm.ABLE
case OLE_ | gc.TUINT8,
OLE_ | gc.TUINT16,
OLE_ | gc.TUINT32,
OLE_ | gc.TUINT64:
a = arm.ABLS
case OGT_ | gc.TINT8,
OGT_ | gc.TINT16,
OGT_ | gc.TINT32,
OGT_ | gc.TINT64,
OGT_ | gc.TFLOAT32,
OGT_ | gc.TFLOAT64:
a = arm.ABGT
case OGT_ | gc.TUINT8,
OGT_ | gc.TUINT16,
OGT_ | gc.TUINT32,
OGT_ | gc.TUINT64:
a = arm.ABHI
case OGE_ | gc.TINT8,
OGE_ | gc.TINT16,
OGE_ | gc.TINT32,
OGE_ | gc.TINT64,
OGE_ | gc.TFLOAT32,
OGE_ | gc.TFLOAT64:
a = arm.ABGE
case OGE_ | gc.TUINT8,
OGE_ | gc.TUINT16,
OGE_ | gc.TUINT32,
OGE_ | gc.TUINT64:
a = arm.ABHS
case OCMP_ | gc.TBOOL,
OCMP_ | gc.TINT8,
OCMP_ | gc.TUINT8,
OCMP_ | gc.TINT16,
OCMP_ | gc.TUINT16,
OCMP_ | gc.TINT32,
OCMP_ | gc.TUINT32,
OCMP_ | gc.TPTR32:
a = arm.ACMP
case OCMP_ | gc.TFLOAT32:
a = arm.ACMPF
case OCMP_ | gc.TFLOAT64:
a = arm.ACMPD
case OPS_ | gc.TFLOAT32,
OPS_ | gc.TFLOAT64:
a = arm.ABVS
case OAS_ | gc.TBOOL:
a = arm.AMOVB
case OAS_ | gc.TINT8:
a = arm.AMOVBS
case OAS_ | gc.TUINT8:
a = arm.AMOVBU
case OAS_ | gc.TINT16:
a = arm.AMOVHS
case OAS_ | gc.TUINT16:
a = arm.AMOVHU
case OAS_ | gc.TINT32,
OAS_ | gc.TUINT32,
OAS_ | gc.TPTR32:
a = arm.AMOVW
case OAS_ | gc.TFLOAT32:
a = arm.AMOVF
case OAS_ | gc.TFLOAT64:
a = arm.AMOVD
case OADD_ | gc.TINT8,
OADD_ | gc.TUINT8,
OADD_ | gc.TINT16,
OADD_ | gc.TUINT16,
OADD_ | gc.TINT32,
OADD_ | gc.TUINT32,
OADD_ | gc.TPTR32:
a = arm.AADD
case OADD_ | gc.TFLOAT32:
a = arm.AADDF
case OADD_ | gc.TFLOAT64:
a = arm.AADDD
case OSUB_ | gc.TINT8,
OSUB_ | gc.TUINT8,
OSUB_ | gc.TINT16,
OSUB_ | gc.TUINT16,
OSUB_ | gc.TINT32,
OSUB_ | gc.TUINT32,
OSUB_ | gc.TPTR32:
a = arm.ASUB
case OSUB_ | gc.TFLOAT32:
a = arm.ASUBF
case OSUB_ | gc.TFLOAT64:
a = arm.ASUBD
case OMINUS_ | gc.TINT8,
OMINUS_ | gc.TUINT8,
OMINUS_ | gc.TINT16,
OMINUS_ | gc.TUINT16,
OMINUS_ | gc.TINT32,
OMINUS_ | gc.TUINT32,
OMINUS_ | gc.TPTR32:
a = arm.ARSB
case OAND_ | gc.TINT8,
OAND_ | gc.TUINT8,
OAND_ | gc.TINT16,
OAND_ | gc.TUINT16,
OAND_ | gc.TINT32,
OAND_ | gc.TUINT32,
OAND_ | gc.TPTR32:
a = arm.AAND
case OOR_ | gc.TINT8,
OOR_ | gc.TUINT8,
OOR_ | gc.TINT16,
OOR_ | gc.TUINT16,
OOR_ | gc.TINT32,
OOR_ | gc.TUINT32,
OOR_ | gc.TPTR32:
a = arm.AORR
case OXOR_ | gc.TINT8,
OXOR_ | gc.TUINT8,
OXOR_ | gc.TINT16,
OXOR_ | gc.TUINT16,
OXOR_ | gc.TINT32,
OXOR_ | gc.TUINT32,
OXOR_ | gc.TPTR32:
a = arm.AEOR
case OLSH_ | gc.TINT8,
OLSH_ | gc.TUINT8,
OLSH_ | gc.TINT16,
OLSH_ | gc.TUINT16,
OLSH_ | gc.TINT32,
OLSH_ | gc.TUINT32,
OLSH_ | gc.TPTR32:
a = arm.ASLL
case ORSH_ | gc.TUINT8,
ORSH_ | gc.TUINT16,
ORSH_ | gc.TUINT32,
ORSH_ | gc.TPTR32:
a = arm.ASRL
case ORSH_ | gc.TINT8,
ORSH_ | gc.TINT16,
ORSH_ | gc.TINT32:
a = arm.ASRA
case OMUL_ | gc.TUINT8,
OMUL_ | gc.TUINT16,
OMUL_ | gc.TUINT32,
OMUL_ | gc.TPTR32:
a = arm.AMULU
case OMUL_ | gc.TINT8,
OMUL_ | gc.TINT16,
OMUL_ | gc.TINT32:
a = arm.AMUL
case OMUL_ | gc.TFLOAT32:
a = arm.AMULF
case OMUL_ | gc.TFLOAT64:
a = arm.AMULD
case ODIV_ | gc.TUINT8,
ODIV_ | gc.TUINT16,
ODIV_ | gc.TUINT32,
ODIV_ | gc.TPTR32:
a = arm.ADIVU
case ODIV_ | gc.TINT8,
ODIV_ | gc.TINT16,
ODIV_ | gc.TINT32:
a = arm.ADIV
case OMOD_ | gc.TUINT8,
OMOD_ | gc.TUINT16,
OMOD_ | gc.TUINT32,
OMOD_ | gc.TPTR32:
a = arm.AMODU
case OMOD_ | gc.TINT8,
OMOD_ | gc.TINT16,
OMOD_ | gc.TINT32:
a = arm.AMOD
// case CASE(OEXTEND, TINT16):
// a = ACWD;
// break;
// case CASE(OEXTEND, TINT32):
// a = ACDQ;
// break;
// case CASE(OEXTEND, TINT64):
// a = ACQO;
// break;
case ODIV_ | gc.TFLOAT32:
a = arm.ADIVF
case ODIV_ | gc.TFLOAT64:
a = arm.ADIVD
case OSQRT_ | gc.TFLOAT64:
a = arm.ASQRTD
}
return a
}
func foptoas(op gc.Op, t *gc.Type, flg int) int {
a := obj.AXXX
et := gc.Simtype[t.Etype]
// avoid constant conversions in switches below
const (
OCMP_ = uint32(gc.OCMP) << 16
OAS_ = uint32(gc.OAS) << 16
OADD_ = uint32(gc.OADD) << 16
OSUB_ = uint32(gc.OSUB) << 16
OMUL_ = uint32(gc.OMUL) << 16
ODIV_ = uint32(gc.ODIV) << 16
OMINUS_ = uint32(gc.OMINUS) << 16
)
if !gc.Thearch.Use387 {
switch uint32(op)<<16 | uint32(et) {
default:
gc.Fatalf("foptoas-sse: no entry %v-%v", gc.Oconv(int(op), 0), t)
case OCMP_ | gc.TFLOAT32:
a = x86.AUCOMISS
case OCMP_ | gc.TFLOAT64:
a = x86.AUCOMISD
case OAS_ | gc.TFLOAT32:
a = x86.AMOVSS
case OAS_ | gc.TFLOAT64:
a = x86.AMOVSD
case OADD_ | gc.TFLOAT32:
a = x86.AADDSS
case OADD_ | gc.TFLOAT64:
a = x86.AADDSD
case OSUB_ | gc.TFLOAT32:
a = x86.ASUBSS
case OSUB_ | gc.TFLOAT64:
a = x86.ASUBSD
case OMUL_ | gc.TFLOAT32:
a = x86.AMULSS
case OMUL_ | gc.TFLOAT64:
a = x86.AMULSD
case ODIV_ | gc.TFLOAT32:
a = x86.ADIVSS
case ODIV_ | gc.TFLOAT64:
a = x86.ADIVSD
}
return a
}
// If we need Fpop, it means we're working on
// two different floating-point registers, not memory.
// There the instruction only has a float64 form.
if flg&Fpop != 0 {
et = gc.TFLOAT64
}
// clear Frev if unneeded
switch op {
case gc.OADD,
gc.OMUL:
flg &^= Frev
}
switch uint32(op)<<16 | (uint32(et)<<8 | uint32(flg)) {
case OADD_ | (gc.TFLOAT32<<8 | 0):
return x86.AFADDF
case OADD_ | (gc.TFLOAT64<<8 | 0):
return x86.AFADDD
case OADD_ | (gc.TFLOAT64<<8 | Fpop):
return x86.AFADDDP
case OSUB_ | (gc.TFLOAT32<<8 | 0):
return x86.AFSUBF
case OSUB_ | (gc.TFLOAT32<<8 | Frev):
return x86.AFSUBRF
case OSUB_ | (gc.TFLOAT64<<8 | 0):
return x86.AFSUBD
case OSUB_ | (gc.TFLOAT64<<8 | Frev):
return x86.AFSUBRD
case OSUB_ | (gc.TFLOAT64<<8 | Fpop):
return x86.AFSUBDP
case OSUB_ | (gc.TFLOAT64<<8 | (Fpop | Frev)):
return x86.AFSUBRDP
case OMUL_ | (gc.TFLOAT32<<8 | 0):
return x86.AFMULF
case OMUL_ | (gc.TFLOAT64<<8 | 0):
return x86.AFMULD
case OMUL_ | (gc.TFLOAT64<<8 | Fpop):
return x86.AFMULDP
case ODIV_ | (gc.TFLOAT32<<8 | 0):
return x86.AFDIVF
case ODIV_ | (gc.TFLOAT32<<8 | Frev):
return x86.AFDIVRF
case ODIV_ | (gc.TFLOAT64<<8 | 0):
return x86.AFDIVD
case ODIV_ | (gc.TFLOAT64<<8 | Frev):
return x86.AFDIVRD
case ODIV_ | (gc.TFLOAT64<<8 | Fpop):
return x86.AFDIVDP
case ODIV_ | (gc.TFLOAT64<<8 | (Fpop | Frev)):
return x86.AFDIVRDP
case OCMP_ | (gc.TFLOAT32<<8 | 0):
return x86.AFCOMF
case OCMP_ | (gc.TFLOAT32<<8 | Fpop):
return x86.AFCOMFP
case OCMP_ | (gc.TFLOAT64<<8 | 0):
return x86.AFCOMD
case OCMP_ | (gc.TFLOAT64<<8 | Fpop):
return x86.AFCOMDP
case OCMP_ | (gc.TFLOAT64<<8 | Fpop2):
return x86.AFCOMDPP
case OMINUS_ | (gc.TFLOAT32<<8 | 0):
return x86.AFCHS
case OMINUS_ | (gc.TFLOAT64<<8 | 0):
return x86.AFCHS
}
gc.Fatalf("foptoas %v %v %#x", gc.Oconv(int(op), 0), t, flg)
return 0
}
/*
* attempt to generate 64-bit
* res = n
* return 1 on success, 0 if op not handled.
*/
func cgen64(n *gc.Node, res *gc.Node) {
if res.Op != gc.OINDREG && res.Op != gc.ONAME {
gc.Dump("n", n)
gc.Dump("res", res)
gc.Fatalf("cgen64 %v of %v", gc.Oconv(int(n.Op), 0), gc.Oconv(int(res.Op), 0))
}
switch n.Op {
default:
gc.Fatalf("cgen64 %v", gc.Oconv(int(n.Op), 0))
case gc.OMINUS:
gc.Cgen(n.Left, res)
var hi1 gc.Node
var lo1 gc.Node
split64(res, &lo1, &hi1)
gins(x86.ANEGL, nil, &lo1)
gins(x86.AADCL, ncon(0), &hi1)
gins(x86.ANEGL, nil, &hi1)
splitclean()
return
case gc.OCOM:
gc.Cgen(n.Left, res)
var lo1 gc.Node
var hi1 gc.Node
split64(res, &lo1, &hi1)
gins(x86.ANOTL, nil, &lo1)
gins(x86.ANOTL, nil, &hi1)
splitclean()
return
// binary operators.
// common setup below.
case gc.OADD,
gc.OSUB,
gc.OMUL,
gc.OLROT,
gc.OLSH,
gc.ORSH,
gc.OAND,
gc.OOR,
gc.OXOR:
break
}
l := n.Left
r := n.Right
if !l.Addable {
var t1 gc.Node
gc.Tempname(&t1, l.Type)
gc.Cgen(l, &t1)
l = &t1
}
if r != nil && !r.Addable {
var t2 gc.Node
gc.Tempname(&t2, r.Type)
gc.Cgen(r, &t2)
r = &t2
}
var ax gc.Node
gc.Nodreg(&ax, gc.Types[gc.TINT32], x86.REG_AX)
var cx gc.Node
gc.Nodreg(&cx, gc.Types[gc.TINT32], x86.REG_CX)
var dx gc.Node
gc.Nodreg(&dx, gc.Types[gc.TINT32], x86.REG_DX)
// Setup for binary operation.
var hi1 gc.Node
var lo1 gc.Node
split64(l, &lo1, &hi1)
var lo2 gc.Node
var hi2 gc.Node
if gc.Is64(r.Type) {
split64(r, &lo2, &hi2)
}
// Do op. Leave result in DX:AX.
switch n.Op {
// TODO: Constants
case gc.OADD:
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
gins(x86.AADDL, &lo2, &ax)
gins(x86.AADCL, &hi2, &dx)
// TODO: Constants.
case gc.OSUB:
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
gins(x86.ASUBL, &lo2, &ax)
gins(x86.ASBBL, &hi2, &dx)
case gc.OMUL:
// let's call the next three EX, FX and GX
var ex, fx, gx gc.Node
gc.Regalloc(&ex, gc.Types[gc.TPTR32], nil)
gc.Regalloc(&fx, gc.Types[gc.TPTR32], nil)
gc.Regalloc(&gx, gc.Types[gc.TPTR32], nil)
// load args into DX:AX and EX:GX.
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
gins(x86.AMOVL, &lo2, &gx)
gins(x86.AMOVL, &hi2, &ex)
// if DX and EX are zero, use 32 x 32 -> 64 unsigned multiply.
gins(x86.AMOVL, &dx, &fx)
gins(x86.AORL, &ex, &fx)
p1 := gc.Gbranch(x86.AJNE, nil, 0)
gins(x86.AMULL, &gx, nil) // implicit &ax
p2 := gc.Gbranch(obj.AJMP, nil, 0)
gc.Patch(p1, gc.Pc)
// full 64x64 -> 64, from 32x32 -> 64.
gins(x86.AIMULL, &gx, &dx)
gins(x86.AMOVL, &ax, &fx)
gins(x86.AIMULL, &ex, &fx)
gins(x86.AADDL, &dx, &fx)
gins(x86.AMOVL, &gx, &dx)
gins(x86.AMULL, &dx, nil) // implicit &ax
gins(x86.AADDL, &fx, &dx)
gc.Patch(p2, gc.Pc)
gc.Regfree(&ex)
gc.Regfree(&fx)
gc.Regfree(&gx)
// We only rotate by a constant c in [0,64).
// if c >= 32:
// lo, hi = hi, lo
// c -= 32
// if c == 0:
// no-op
// else:
// t = hi
// shld hi:lo, c
// shld lo:t, c
case gc.OLROT:
v := uint64(r.Int())
if v >= 32 {
// reverse during load to do the first 32 bits of rotate
v -= 32
gins(x86.AMOVL, &lo1, &dx)
gins(x86.AMOVL, &hi1, &ax)
} else {
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
}
if v == 0 {
} else // done
{
gins(x86.AMOVL, &dx, &cx)
p1 := gins(x86.ASHLL, ncon(uint32(v)), &dx)
p1.From.Index = x86.REG_AX // double-width shift
p1.From.Scale = 0
p1 = gins(x86.ASHLL, ncon(uint32(v)), &ax)
p1.From.Index = x86.REG_CX // double-width shift
p1.From.Scale = 0
}
case gc.OLSH:
if r.Op == gc.OLITERAL {
v := uint64(r.Int())
if v >= 64 {
if gc.Is64(r.Type) {
splitclean()
}
splitclean()
split64(res, &lo2, &hi2)
gins(x86.AMOVL, ncon(0), &lo2)
gins(x86.AMOVL, ncon(0), &hi2)
splitclean()
return
}
if v >= 32 {
if gc.Is64(r.Type) {
splitclean()
}
split64(res, &lo2, &hi2)
gmove(&lo1, &hi2)
if v > 32 {
gins(x86.ASHLL, ncon(uint32(v-32)), &hi2)
}
gins(x86.AMOVL, ncon(0), &lo2)
splitclean()
splitclean()
return
}
// general shift
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
p1 := gins(x86.ASHLL, ncon(uint32(v)), &dx)
p1.From.Index = x86.REG_AX // double-width shift
p1.From.Scale = 0
gins(x86.ASHLL, ncon(uint32(v)), &ax)
break
}
// load value into DX:AX.
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
// load shift value into register.
// if high bits are set, zero value.
var p1 *obj.Prog
if gc.Is64(r.Type) {
gins(x86.ACMPL, &hi2, ncon(0))
p1 = gc.Gbranch(x86.AJNE, nil, +1)
gins(x86.AMOVL, &lo2, &cx)
} else {
cx.Type = gc.Types[gc.TUINT32]
gmove(r, &cx)
}
// if shift count is >=64, zero value
gins(x86.ACMPL, &cx, ncon(64))
p2 := gc.Gbranch(optoas(gc.OLT, gc.Types[gc.TUINT32]), nil, +1)
if p1 != nil {
gc.Patch(p1, gc.Pc)
}
gins(x86.AXORL, &dx, &dx)
gins(x86.AXORL, &ax, &ax)
gc.Patch(p2, gc.Pc)
// if shift count is >= 32, zero low.
gins(x86.ACMPL, &cx, ncon(32))
p1 = gc.Gbranch(optoas(gc.OLT, gc.Types[gc.TUINT32]), nil, +1)
gins(x86.AMOVL, &ax, &dx)
gins(x86.ASHLL, &cx, &dx) // SHLL only uses bottom 5 bits of count
gins(x86.AXORL, &ax, &ax)
p2 = gc.Gbranch(obj.AJMP, nil, 0)
gc.Patch(p1, gc.Pc)
// general shift
p1 = gins(x86.ASHLL, &cx, &dx)
p1.From.Index = x86.REG_AX // double-width shift
p1.From.Scale = 0
gins(x86.ASHLL, &cx, &ax)
gc.Patch(p2, gc.Pc)
case gc.ORSH:
if r.Op == gc.OLITERAL {
v := uint64(r.Int())
if v >= 64 {
if gc.Is64(r.Type) {
splitclean()
}
splitclean()
split64(res, &lo2, &hi2)
if hi1.Type.Etype == gc.TINT32 {
gmove(&hi1, &lo2)
gins(x86.ASARL, ncon(31), &lo2)
gmove(&hi1, &hi2)
gins(x86.ASARL, ncon(31), &hi2)
} else {
gins(x86.AMOVL, ncon(0), &lo2)
gins(x86.AMOVL, ncon(0), &hi2)
}
splitclean()
return
}
if v >= 32 {
if gc.Is64(r.Type) {
splitclean()
}
split64(res, &lo2, &hi2)
gmove(&hi1, &lo2)
if v > 32 {
gins(optoas(gc.ORSH, hi1.Type), ncon(uint32(v-32)), &lo2)
}
if hi1.Type.Etype == gc.TINT32 {
gmove(&hi1, &hi2)
gins(x86.ASARL, ncon(31), &hi2)
} else {
gins(x86.AMOVL, ncon(0), &hi2)
}
splitclean()
splitclean()
return
}
// general shift
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
p1 := gins(x86.ASHRL, ncon(uint32(v)), &ax)
p1.From.Index = x86.REG_DX // double-width shift
p1.From.Scale = 0
gins(optoas(gc.ORSH, hi1.Type), ncon(uint32(v)), &dx)
break
}
// load value into DX:AX.
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
// load shift value into register.
// if high bits are set, zero value.
var p1 *obj.Prog
if gc.Is64(r.Type) {
gins(x86.ACMPL, &hi2, ncon(0))
p1 = gc.Gbranch(x86.AJNE, nil, +1)
gins(x86.AMOVL, &lo2, &cx)
} else {
cx.Type = gc.Types[gc.TUINT32]
gmove(r, &cx)
}
// if shift count is >=64, zero or sign-extend value
gins(x86.ACMPL, &cx, ncon(64))
p2 := gc.Gbranch(optoas(gc.OLT, gc.Types[gc.TUINT32]), nil, +1)
if p1 != nil {
gc.Patch(p1, gc.Pc)
}
if hi1.Type.Etype == gc.TINT32 {
gins(x86.ASARL, ncon(31), &dx)
gins(x86.AMOVL, &dx, &ax)
} else {
gins(x86.AXORL, &dx, &dx)
gins(x86.AXORL, &ax, &ax)
}
gc.Patch(p2, gc.Pc)
// if shift count is >= 32, sign-extend hi.
gins(x86.ACMPL, &cx, ncon(32))
p1 = gc.Gbranch(optoas(gc.OLT, gc.Types[gc.TUINT32]), nil, +1)
gins(x86.AMOVL, &dx, &ax)
if hi1.Type.Etype == gc.TINT32 {
gins(x86.ASARL, &cx, &ax) // SARL only uses bottom 5 bits of count
gins(x86.ASARL, ncon(31), &dx)
} else {
gins(x86.ASHRL, &cx, &ax)
gins(x86.AXORL, &dx, &dx)
}
p2 = gc.Gbranch(obj.AJMP, nil, 0)
gc.Patch(p1, gc.Pc)
// general shift
p1 = gins(x86.ASHRL, &cx, &ax)
p1.From.Index = x86.REG_DX // double-width shift
p1.From.Scale = 0
gins(optoas(gc.ORSH, hi1.Type), &cx, &dx)
gc.Patch(p2, gc.Pc)
// make constant the right side (it usually is anyway).
case gc.OXOR,
gc.OAND,
gc.OOR:
if lo1.Op == gc.OLITERAL {
nswap(&lo1, &lo2)
nswap(&hi1, &hi2)
}
if lo2.Op == gc.OLITERAL {
// special cases for constants.
lv := uint32(lo2.Int())
hv := uint32(hi2.Int())
splitclean() // right side
split64(res, &lo2, &hi2)
switch n.Op {
case gc.OXOR:
gmove(&lo1, &lo2)
gmove(&hi1, &hi2)
switch lv {
case 0:
break
case 0xffffffff:
gins(x86.ANOTL, nil, &lo2)
default:
gins(x86.AXORL, ncon(lv), &lo2)
}
switch hv {
case 0:
break
case 0xffffffff:
gins(x86.ANOTL, nil, &hi2)
default:
gins(x86.AXORL, ncon(hv), &hi2)
}
case gc.OAND:
switch lv {
case 0:
gins(x86.AMOVL, ncon(0), &lo2)
default:
gmove(&lo1, &lo2)
if lv != 0xffffffff {
gins(x86.AANDL, ncon(lv), &lo2)
}
}
switch hv {
case 0:
gins(x86.AMOVL, ncon(0), &hi2)
default:
gmove(&hi1, &hi2)
if hv != 0xffffffff {
gins(x86.AANDL, ncon(hv), &hi2)
}
}
case gc.OOR:
switch lv {
case 0:
gmove(&lo1, &lo2)
case 0xffffffff:
gins(x86.AMOVL, ncon(0xffffffff), &lo2)
default:
gmove(&lo1, &lo2)
gins(x86.AORL, ncon(lv), &lo2)
}
switch hv {
case 0:
gmove(&hi1, &hi2)
case 0xffffffff:
gins(x86.AMOVL, ncon(0xffffffff), &hi2)
default:
gmove(&hi1, &hi2)
gins(x86.AORL, ncon(hv), &hi2)
}
}
splitclean()
splitclean()
return
}
gins(x86.AMOVL, &lo1, &ax)
gins(x86.AMOVL, &hi1, &dx)
gins(optoas(n.Op, lo1.Type), &lo2, &ax)
gins(optoas(n.Op, lo1.Type), &hi2, &dx)
}
if gc.Is64(r.Type) {
splitclean()
}
splitclean()
split64(res, &lo1, &hi1)
gins(x86.AMOVL, &ax, &lo1)
gins(x86.AMOVL, &dx, &hi1)
splitclean()
}
func ginscmp(op gc.Op, t *gc.Type, n1, n2 *gc.Node, likely int) *obj.Prog {
if !gc.Isfloat[t.Etype] && (op == gc.OLT || op == gc.OGE) {
// swap nodes to fit SGT instruction
n1, n2 = n2, n1
}
if gc.Isfloat[t.Etype] && (op == gc.OLT || op == gc.OLE) {
// swap nodes to fit CMPGT, CMPGE instructions and reverse relation
n1, n2 = n2, n1
if op == gc.OLT {
op = gc.OGT
} else {
op = gc.OGE
}
}
var r1, r2, g1, g2 gc.Node
gc.Regalloc(&r1, t, n1)
gc.Regalloc(&g1, n1.Type, &r1)
gc.Cgen(n1, &g1)
gmove(&g1, &r1)
gc.Regalloc(&r2, t, n2)
gc.Regalloc(&g2, n1.Type, &r2)
gc.Cgen(n2, &g2)
gmove(&g2, &r2)
var p *obj.Prog
var ntmp gc.Node
gc.Nodreg(&ntmp, gc.Types[gc.TINT], mips.REGTMP)
switch gc.Simtype[t.Etype] {
case gc.TINT8,
gc.TINT16,
gc.TINT32,
gc.TINT64:
if op == gc.OEQ || op == gc.ONE {
p = ginsbranch(optoas(op, t), nil, &r1, &r2, likely)
} else {
gins3(mips.ASGT, &r1, &r2, &ntmp)
p = ginsbranch(optoas(op, t), nil, &ntmp, nil, likely)
}
case gc.TBOOL,
gc.TUINT8,
gc.TUINT16,
gc.TUINT32,
gc.TUINT64,
gc.TPTR32,
gc.TPTR64:
if op == gc.OEQ || op == gc.ONE {
p = ginsbranch(optoas(op, t), nil, &r1, &r2, likely)
} else {
gins3(mips.ASGTU, &r1, &r2, &ntmp)
p = ginsbranch(optoas(op, t), nil, &ntmp, nil, likely)
}
case gc.TFLOAT32:
switch op {
default:
gc.Fatalf("ginscmp: no entry for op=%v type=%v", gc.Oconv(int(op), 0), t)
case gc.OEQ,
gc.ONE:
gins3(mips.ACMPEQF, &r1, &r2, nil)
case gc.OGE:
gins3(mips.ACMPGEF, &r1, &r2, nil)
case gc.OGT:
gins3(mips.ACMPGTF, &r1, &r2, nil)
}
p = gc.Gbranch(optoas(op, t), nil, likely)
case gc.TFLOAT64:
switch op {
default:
gc.Fatalf("ginscmp: no entry for op=%v type=%v", gc.Oconv(int(op), 0), t)
case gc.OEQ,
gc.ONE:
gins3(mips.ACMPEQD, &r1, &r2, nil)
case gc.OGE:
gins3(mips.ACMPGED, &r1, &r2, nil)
case gc.OGT:
gins3(mips.ACMPGTD, &r1, &r2, nil)
}
p = gc.Gbranch(optoas(op, t), nil, likely)
}
gc.Regfree(&g2)
gc.Regfree(&r2)
gc.Regfree(&g1)
gc.Regfree(&r1)
return p
}
/*
* return Axxx for Oxxx on type t.
*/
func optoas(op gc.Op, t *gc.Type) int {
if t == nil {
gc.Fatalf("optoas: t is nil")
}
// avoid constant conversions in switches below
const (
OMINUS_ = uint32(gc.OMINUS) << 16
OLSH_ = uint32(gc.OLSH) << 16
ORSH_ = uint32(gc.ORSH) << 16
OADD_ = uint32(gc.OADD) << 16
OSUB_ = uint32(gc.OSUB) << 16
OMUL_ = uint32(gc.OMUL) << 16
ODIV_ = uint32(gc.ODIV) << 16
OOR_ = uint32(gc.OOR) << 16
OAND_ = uint32(gc.OAND) << 16
OXOR_ = uint32(gc.OXOR) << 16
OEQ_ = uint32(gc.OEQ) << 16
ONE_ = uint32(gc.ONE) << 16
OLT_ = uint32(gc.OLT) << 16
OLE_ = uint32(gc.OLE) << 16
OGE_ = uint32(gc.OGE) << 16
OGT_ = uint32(gc.OGT) << 16
OCMP_ = uint32(gc.OCMP) << 16
OAS_ = uint32(gc.OAS) << 16
OHMUL_ = uint32(gc.OHMUL) << 16
)
a := int(obj.AXXX)
switch uint32(op)<<16 | uint32(gc.Simtype[t.Etype]) {
default:
gc.Fatalf("optoas: no entry for op=%v type=%v", gc.Oconv(int(op), 0), t)
case OEQ_ | gc.TBOOL,
OEQ_ | gc.TINT8,
OEQ_ | gc.TUINT8,
OEQ_ | gc.TINT16,
OEQ_ | gc.TUINT16,
OEQ_ | gc.TINT32,
OEQ_ | gc.TUINT32,
OEQ_ | gc.TINT64,
OEQ_ | gc.TUINT64,
OEQ_ | gc.TPTR32,
OEQ_ | gc.TPTR64:
a = mips.ABEQ
case OEQ_ | gc.TFLOAT32, // ACMPEQF
OEQ_ | gc.TFLOAT64: // ACMPEQD
a = mips.ABFPT
case ONE_ | gc.TBOOL,
ONE_ | gc.TINT8,
ONE_ | gc.TUINT8,
ONE_ | gc.TINT16,
ONE_ | gc.TUINT16,
ONE_ | gc.TINT32,
ONE_ | gc.TUINT32,
ONE_ | gc.TINT64,
ONE_ | gc.TUINT64,
ONE_ | gc.TPTR32,
ONE_ | gc.TPTR64:
a = mips.ABNE
case ONE_ | gc.TFLOAT32, // ACMPEQF
ONE_ | gc.TFLOAT64: // ACMPEQD
a = mips.ABFPF
case OLT_ | gc.TINT8, // ASGT
OLT_ | gc.TINT16,
OLT_ | gc.TINT32,
OLT_ | gc.TINT64,
OLT_ | gc.TUINT8, // ASGTU
OLT_ | gc.TUINT16,
OLT_ | gc.TUINT32,
OLT_ | gc.TUINT64:
a = mips.ABNE
case OLT_ | gc.TFLOAT32, // ACMPGEF
OLT_ | gc.TFLOAT64: // ACMPGED
a = mips.ABFPT
case OLE_ | gc.TINT8, // ASGT
OLE_ | gc.TINT16,
OLE_ | gc.TINT32,
OLE_ | gc.TINT64,
OLE_ | gc.TUINT8, // ASGTU
OLE_ | gc.TUINT16,
OLE_ | gc.TUINT32,
OLE_ | gc.TUINT64:
a = mips.ABEQ
case OLE_ | gc.TFLOAT32, // ACMPGTF
OLE_ | gc.TFLOAT64: // ACMPGTD
a = mips.ABFPT
case OGT_ | gc.TINT8, // ASGT
OGT_ | gc.TINT16,
OGT_ | gc.TINT32,
OGT_ | gc.TINT64,
OGT_ | gc.TUINT8, // ASGTU
OGT_ | gc.TUINT16,
OGT_ | gc.TUINT32,
OGT_ | gc.TUINT64:
a = mips.ABNE
case OGT_ | gc.TFLOAT32, // ACMPGTF
OGT_ | gc.TFLOAT64: // ACMPGTD
a = mips.ABFPT
case OGE_ | gc.TINT8, // ASGT
OGE_ | gc.TINT16,
OGE_ | gc.TINT32,
OGE_ | gc.TINT64,
OGE_ | gc.TUINT8, // ASGTU
OGE_ | gc.TUINT16,
OGE_ | gc.TUINT32,
OGE_ | gc.TUINT64:
a = mips.ABEQ
case OGE_ | gc.TFLOAT32, // ACMPGEF
OGE_ | gc.TFLOAT64: // ACMPGED
a = mips.ABFPT
case OAS_ | gc.TBOOL,
OAS_ | gc.TINT8:
a = mips.AMOVB
case OAS_ | gc.TUINT8:
a = mips.AMOVBU
case OAS_ | gc.TINT16:
a = mips.AMOVH
case OAS_ | gc.TUINT16:
a = mips.AMOVHU
case OAS_ | gc.TINT32:
a = mips.AMOVW
case OAS_ | gc.TUINT32,
OAS_ | gc.TPTR32:
a = mips.AMOVWU
case OAS_ | gc.TINT64,
OAS_ | gc.TUINT64,
OAS_ | gc.TPTR64:
a = mips.AMOVV
case OAS_ | gc.TFLOAT32:
a = mips.AMOVF
case OAS_ | gc.TFLOAT64:
a = mips.AMOVD
case OADD_ | gc.TINT8,
OADD_ | gc.TUINT8,
OADD_ | gc.TINT16,
OADD_ | gc.TUINT16,
OADD_ | gc.TINT32,
OADD_ | gc.TUINT32,
OADD_ | gc.TPTR32:
a = mips.AADDU
case OADD_ | gc.TINT64,
OADD_ | gc.TUINT64,
OADD_ | gc.TPTR64:
a = mips.AADDVU
case OADD_ | gc.TFLOAT32:
a = mips.AADDF
case OADD_ | gc.TFLOAT64:
a = mips.AADDD
case OSUB_ | gc.TINT8,
OSUB_ | gc.TUINT8,
OSUB_ | gc.TINT16,
OSUB_ | gc.TUINT16,
OSUB_ | gc.TINT32,
OSUB_ | gc.TUINT32,
OSUB_ | gc.TPTR32:
a = mips.ASUBU
case OSUB_ | gc.TINT64,
OSUB_ | gc.TUINT64,
OSUB_ | gc.TPTR64:
a = mips.ASUBVU
case OSUB_ | gc.TFLOAT32:
a = mips.ASUBF
case OSUB_ | gc.TFLOAT64:
a = mips.ASUBD
case OMINUS_ | gc.TINT8,
OMINUS_ | gc.TUINT8,
OMINUS_ | gc.TINT16,
OMINUS_ | gc.TUINT16,
OMINUS_ | gc.TINT32,
OMINUS_ | gc.TUINT32,
OMINUS_ | gc.TPTR32,
OMINUS_ | gc.TINT64,
OMINUS_ | gc.TUINT64,
OMINUS_ | gc.TPTR64:
a = mips.ASUBVU
case OAND_ | gc.TINT8,
OAND_ | gc.TUINT8,
OAND_ | gc.TINT16,
OAND_ | gc.TUINT16,
OAND_ | gc.TINT32,
OAND_ | gc.TUINT32,
OAND_ | gc.TPTR32,
OAND_ | gc.TINT64,
OAND_ | gc.TUINT64,
OAND_ | gc.TPTR64:
a = mips.AAND
case OOR_ | gc.TINT8,
OOR_ | gc.TUINT8,
OOR_ | gc.TINT16,
OOR_ | gc.TUINT16,
OOR_ | gc.TINT32,
OOR_ | gc.TUINT32,
OOR_ | gc.TPTR32,
OOR_ | gc.TINT64,
OOR_ | gc.TUINT64,
OOR_ | gc.TPTR64:
a = mips.AOR
case OXOR_ | gc.TINT8,
OXOR_ | gc.TUINT8,
OXOR_ | gc.TINT16,
OXOR_ | gc.TUINT16,
OXOR_ | gc.TINT32,
OXOR_ | gc.TUINT32,
OXOR_ | gc.TPTR32,
OXOR_ | gc.TINT64,
OXOR_ | gc.TUINT64,
OXOR_ | gc.TPTR64:
a = mips.AXOR
// TODO(minux): handle rotates
//case CASE(OLROT, TINT8):
//case CASE(OLROT, TUINT8):
//case CASE(OLROT, TINT16):
//case CASE(OLROT, TUINT16):
//case CASE(OLROT, TINT32):
//case CASE(OLROT, TUINT32):
//case CASE(OLROT, TPTR32):
//case CASE(OLROT, TINT64):
//case CASE(OLROT, TUINT64):
//case CASE(OLROT, TPTR64):
// a = 0//???; RLDC?
// break;
case OLSH_ | gc.TINT8,
OLSH_ | gc.TUINT8,
OLSH_ | gc.TINT16,
OLSH_ | gc.TUINT16,
OLSH_ | gc.TINT32,
OLSH_ | gc.TUINT32,
OLSH_ | gc.TPTR32,
OLSH_ | gc.TINT64,
OLSH_ | gc.TUINT64,
OLSH_ | gc.TPTR64:
a = mips.ASLLV
case ORSH_ | gc.TUINT8,
ORSH_ | gc.TUINT16,
ORSH_ | gc.TUINT32,
ORSH_ | gc.TPTR32,
ORSH_ | gc.TUINT64,
ORSH_ | gc.TPTR64:
a = mips.ASRLV
case ORSH_ | gc.TINT8,
ORSH_ | gc.TINT16,
ORSH_ | gc.TINT32,
ORSH_ | gc.TINT64:
a = mips.ASRAV
// TODO(minux): handle rotates
//case CASE(ORROTC, TINT8):
//case CASE(ORROTC, TUINT8):
//case CASE(ORROTC, TINT16):
//case CASE(ORROTC, TUINT16):
//case CASE(ORROTC, TINT32):
//case CASE(ORROTC, TUINT32):
//case CASE(ORROTC, TINT64):
//case CASE(ORROTC, TUINT64):
// a = 0//??? RLDC??
// break;
case OHMUL_ | gc.TINT64:
a = mips.AMULV
case OHMUL_ | gc.TUINT64,
OHMUL_ | gc.TPTR64:
a = mips.AMULVU
case OMUL_ | gc.TINT8,
OMUL_ | gc.TINT16,
OMUL_ | gc.TINT32,
OMUL_ | gc.TINT64:
a = mips.AMULV
case OMUL_ | gc.TUINT8,
OMUL_ | gc.TUINT16,
OMUL_ | gc.TUINT32,
OMUL_ | gc.TPTR32,
OMUL_ | gc.TUINT64,
OMUL_ | gc.TPTR64:
a = mips.AMULVU
case OMUL_ | gc.TFLOAT32:
a = mips.AMULF
case OMUL_ | gc.TFLOAT64:
a = mips.AMULD
case ODIV_ | gc.TINT8,
ODIV_ | gc.TINT16,
ODIV_ | gc.TINT32,
ODIV_ | gc.TINT64:
a = mips.ADIVV
case ODIV_ | gc.TUINT8,
ODIV_ | gc.TUINT16,
ODIV_ | gc.TUINT32,
ODIV_ | gc.TPTR32,
ODIV_ | gc.TUINT64,
ODIV_ | gc.TPTR64:
a = mips.ADIVVU
case ODIV_ | gc.TFLOAT32:
a = mips.ADIVF
case ODIV_ | gc.TFLOAT64:
a = mips.ADIVD
}
return a
}
/*
* return Axxx for Oxxx on type t.
*/
func optoas(op gc.Op, t *gc.Type) int {
if t == nil {
gc.Fatalf("optoas: t is nil")
}
// avoid constant conversions in switches below
const (
OMINUS_ = uint32(gc.OMINUS) << 16
OLSH_ = uint32(gc.OLSH) << 16
ORSH_ = uint32(gc.ORSH) << 16
OADD_ = uint32(gc.OADD) << 16
OSUB_ = uint32(gc.OSUB) << 16
OMUL_ = uint32(gc.OMUL) << 16
ODIV_ = uint32(gc.ODIV) << 16
OOR_ = uint32(gc.OOR) << 16
OAND_ = uint32(gc.OAND) << 16
OXOR_ = uint32(gc.OXOR) << 16
OEQ_ = uint32(gc.OEQ) << 16
ONE_ = uint32(gc.ONE) << 16
OLT_ = uint32(gc.OLT) << 16
OLE_ = uint32(gc.OLE) << 16
OGE_ = uint32(gc.OGE) << 16
OGT_ = uint32(gc.OGT) << 16
OCMP_ = uint32(gc.OCMP) << 16
OAS_ = uint32(gc.OAS) << 16
OHMUL_ = uint32(gc.OHMUL) << 16
OSQRT_ = uint32(gc.OSQRT) << 16
)
a := int(obj.AXXX)
switch uint32(op)<<16 | uint32(gc.Simtype[t.Etype]) {
default:
gc.Fatalf("optoas: no entry for op=%v type=%v", gc.Oconv(int(op), 0), t)
case OEQ_ | gc.TBOOL,
OEQ_ | gc.TINT8,
OEQ_ | gc.TUINT8,
OEQ_ | gc.TINT16,
OEQ_ | gc.TUINT16,
OEQ_ | gc.TINT32,
OEQ_ | gc.TUINT32,
OEQ_ | gc.TINT64,
OEQ_ | gc.TUINT64,
OEQ_ | gc.TPTR32,
OEQ_ | gc.TPTR64,
OEQ_ | gc.TFLOAT32,
OEQ_ | gc.TFLOAT64:
a = arm64.ABEQ
case ONE_ | gc.TBOOL,
ONE_ | gc.TINT8,
ONE_ | gc.TUINT8,
ONE_ | gc.TINT16,
ONE_ | gc.TUINT16,
ONE_ | gc.TINT32,
ONE_ | gc.TUINT32,
ONE_ | gc.TINT64,
ONE_ | gc.TUINT64,
ONE_ | gc.TPTR32,
ONE_ | gc.TPTR64,
ONE_ | gc.TFLOAT32,
ONE_ | gc.TFLOAT64:
a = arm64.ABNE
case OLT_ | gc.TINT8,
OLT_ | gc.TINT16,
OLT_ | gc.TINT32,
OLT_ | gc.TINT64:
a = arm64.ABLT
case OLT_ | gc.TUINT8,
OLT_ | gc.TUINT16,
OLT_ | gc.TUINT32,
OLT_ | gc.TUINT64,
OLT_ | gc.TFLOAT32,
OLT_ | gc.TFLOAT64:
a = arm64.ABLO
case OLE_ | gc.TINT8,
OLE_ | gc.TINT16,
OLE_ | gc.TINT32,
OLE_ | gc.TINT64:
a = arm64.ABLE
case OLE_ | gc.TUINT8,
OLE_ | gc.TUINT16,
OLE_ | gc.TUINT32,
OLE_ | gc.TUINT64,
OLE_ | gc.TFLOAT32,
OLE_ | gc.TFLOAT64:
a = arm64.ABLS
case OGT_ | gc.TINT8,
OGT_ | gc.TINT16,
OGT_ | gc.TINT32,
OGT_ | gc.TINT64,
OGT_ | gc.TFLOAT32,
OGT_ | gc.TFLOAT64:
a = arm64.ABGT
case OGT_ | gc.TUINT8,
OGT_ | gc.TUINT16,
OGT_ | gc.TUINT32,
OGT_ | gc.TUINT64:
a = arm64.ABHI
case OGE_ | gc.TINT8,
OGE_ | gc.TINT16,
OGE_ | gc.TINT32,
OGE_ | gc.TINT64,
OGE_ | gc.TFLOAT32,
OGE_ | gc.TFLOAT64:
a = arm64.ABGE
case OGE_ | gc.TUINT8,
OGE_ | gc.TUINT16,
OGE_ | gc.TUINT32,
OGE_ | gc.TUINT64:
a = arm64.ABHS
case OCMP_ | gc.TBOOL,
OCMP_ | gc.TINT8,
OCMP_ | gc.TINT16,
OCMP_ | gc.TINT32,
OCMP_ | gc.TPTR32,
OCMP_ | gc.TINT64,
OCMP_ | gc.TUINT8,
OCMP_ | gc.TUINT16,
OCMP_ | gc.TUINT32,
OCMP_ | gc.TUINT64,
OCMP_ | gc.TPTR64:
a = arm64.ACMP
case OCMP_ | gc.TFLOAT32:
a = arm64.AFCMPS
case OCMP_ | gc.TFLOAT64:
a = arm64.AFCMPD
case OAS_ | gc.TBOOL,
OAS_ | gc.TINT8:
a = arm64.AMOVB
case OAS_ | gc.TUINT8:
a = arm64.AMOVBU
case OAS_ | gc.TINT16:
a = arm64.AMOVH
case OAS_ | gc.TUINT16:
a = arm64.AMOVHU
case OAS_ | gc.TINT32:
a = arm64.AMOVW
case OAS_ | gc.TUINT32,
OAS_ | gc.TPTR32:
a = arm64.AMOVWU
case OAS_ | gc.TINT64,
OAS_ | gc.TUINT64,
OAS_ | gc.TPTR64:
a = arm64.AMOVD
case OAS_ | gc.TFLOAT32:
a = arm64.AFMOVS
case OAS_ | gc.TFLOAT64:
a = arm64.AFMOVD
case OADD_ | gc.TINT8,
OADD_ | gc.TUINT8,
OADD_ | gc.TINT16,
OADD_ | gc.TUINT16,
OADD_ | gc.TINT32,
OADD_ | gc.TUINT32,
OADD_ | gc.TPTR32,
OADD_ | gc.TINT64,
OADD_ | gc.TUINT64,
OADD_ | gc.TPTR64:
a = arm64.AADD
case OADD_ | gc.TFLOAT32:
a = arm64.AFADDS
case OADD_ | gc.TFLOAT64:
a = arm64.AFADDD
case OSUB_ | gc.TINT8,
OSUB_ | gc.TUINT8,
OSUB_ | gc.TINT16,
OSUB_ | gc.TUINT16,
OSUB_ | gc.TINT32,
OSUB_ | gc.TUINT32,
OSUB_ | gc.TPTR32,
OSUB_ | gc.TINT64,
OSUB_ | gc.TUINT64,
OSUB_ | gc.TPTR64:
a = arm64.ASUB
case OSUB_ | gc.TFLOAT32:
a = arm64.AFSUBS
case OSUB_ | gc.TFLOAT64:
a = arm64.AFSUBD
case OMINUS_ | gc.TINT8,
OMINUS_ | gc.TUINT8,
OMINUS_ | gc.TINT16,
OMINUS_ | gc.TUINT16,
OMINUS_ | gc.TINT32,
OMINUS_ | gc.TUINT32,
OMINUS_ | gc.TPTR32,
OMINUS_ | gc.TINT64,
OMINUS_ | gc.TUINT64,
OMINUS_ | gc.TPTR64:
a = arm64.ANEG
case OMINUS_ | gc.TFLOAT32:
a = arm64.AFNEGS
case OMINUS_ | gc.TFLOAT64:
a = arm64.AFNEGD
case OAND_ | gc.TINT8,
OAND_ | gc.TUINT8,
OAND_ | gc.TINT16,
OAND_ | gc.TUINT16,
OAND_ | gc.TINT32,
OAND_ | gc.TUINT32,
OAND_ | gc.TPTR32,
OAND_ | gc.TINT64,
OAND_ | gc.TUINT64,
OAND_ | gc.TPTR64:
a = arm64.AAND
case OOR_ | gc.TINT8,
OOR_ | gc.TUINT8,
OOR_ | gc.TINT16,
OOR_ | gc.TUINT16,
OOR_ | gc.TINT32,
OOR_ | gc.TUINT32,
OOR_ | gc.TPTR32,
OOR_ | gc.TINT64,
OOR_ | gc.TUINT64,
OOR_ | gc.TPTR64:
a = arm64.AORR
case OXOR_ | gc.TINT8,
OXOR_ | gc.TUINT8,
OXOR_ | gc.TINT16,
OXOR_ | gc.TUINT16,
OXOR_ | gc.TINT32,
OXOR_ | gc.TUINT32,
OXOR_ | gc.TPTR32,
OXOR_ | gc.TINT64,
OXOR_ | gc.TUINT64,
OXOR_ | gc.TPTR64:
a = arm64.AEOR
// TODO(minux): handle rotates
//case CASE(OLROT, TINT8):
//case CASE(OLROT, TUINT8):
//case CASE(OLROT, TINT16):
//case CASE(OLROT, TUINT16):
//case CASE(OLROT, TINT32):
//case CASE(OLROT, TUINT32):
//case CASE(OLROT, TPTR32):
//case CASE(OLROT, TINT64):
//case CASE(OLROT, TUINT64):
//case CASE(OLROT, TPTR64):
// a = 0//???; RLDC?
// break;
case OLSH_ | gc.TINT8,
OLSH_ | gc.TUINT8,
OLSH_ | gc.TINT16,
OLSH_ | gc.TUINT16,
OLSH_ | gc.TINT32,
OLSH_ | gc.TUINT32,
OLSH_ | gc.TPTR32,
OLSH_ | gc.TINT64,
OLSH_ | gc.TUINT64,
OLSH_ | gc.TPTR64:
a = arm64.ALSL
case ORSH_ | gc.TUINT8,
ORSH_ | gc.TUINT16,
ORSH_ | gc.TUINT32,
ORSH_ | gc.TPTR32,
ORSH_ | gc.TUINT64,
ORSH_ | gc.TPTR64:
a = arm64.ALSR
case ORSH_ | gc.TINT8,
ORSH_ | gc.TINT16,
ORSH_ | gc.TINT32,
ORSH_ | gc.TINT64:
a = arm64.AASR
// TODO(minux): handle rotates
//case CASE(ORROTC, TINT8):
//case CASE(ORROTC, TUINT8):
//case CASE(ORROTC, TINT16):
//case CASE(ORROTC, TUINT16):
//case CASE(ORROTC, TINT32):
//case CASE(ORROTC, TUINT32):
//case CASE(ORROTC, TINT64):
//case CASE(ORROTC, TUINT64):
// a = 0//??? RLDC??
// break;
case OHMUL_ | gc.TINT64:
a = arm64.ASMULH
case OHMUL_ | gc.TUINT64,
OHMUL_ | gc.TPTR64:
a = arm64.AUMULH
case OMUL_ | gc.TINT8,
OMUL_ | gc.TINT16,
OMUL_ | gc.TINT32:
a = arm64.ASMULL
case OMUL_ | gc.TINT64:
a = arm64.AMUL
case OMUL_ | gc.TUINT8,
OMUL_ | gc.TUINT16,
OMUL_ | gc.TUINT32,
OMUL_ | gc.TPTR32:
// don't use word multiply, the high 32-bit are undefined.
a = arm64.AUMULL
case OMUL_ | gc.TUINT64,
OMUL_ | gc.TPTR64:
a = arm64.AMUL // for 64-bit multiplies, signedness doesn't matter.
case OMUL_ | gc.TFLOAT32:
a = arm64.AFMULS
case OMUL_ | gc.TFLOAT64:
a = arm64.AFMULD
case ODIV_ | gc.TINT8,
ODIV_ | gc.TINT16,
ODIV_ | gc.TINT32,
ODIV_ | gc.TINT64:
a = arm64.ASDIV
case ODIV_ | gc.TUINT8,
ODIV_ | gc.TUINT16,
ODIV_ | gc.TUINT32,
ODIV_ | gc.TPTR32,
ODIV_ | gc.TUINT64,
ODIV_ | gc.TPTR64:
a = arm64.AUDIV
case ODIV_ | gc.TFLOAT32:
a = arm64.AFDIVS
case ODIV_ | gc.TFLOAT64:
a = arm64.AFDIVD
case OSQRT_ | gc.TFLOAT64:
a = arm64.AFSQRTD
}
return a
}
func cgen_floatsse(n *gc.Node, res *gc.Node) {
var a int
nl := n.Left
nr := n.Right
switch n.Op {
default:
gc.Dump("cgen_floatsse", n)
gc.Fatalf("cgen_floatsse %v", gc.Oconv(int(n.Op), 0))
return
case gc.OMINUS,
gc.OCOM:
nr = gc.Nodintconst(-1)
gc.Convlit(&nr, n.Type)
a = foptoas(gc.OMUL, nl.Type, 0)
goto sbop
// symmetric binary
case gc.OADD,
gc.OMUL:
a = foptoas(n.Op, nl.Type, 0)
goto sbop
// asymmetric binary
case gc.OSUB,
gc.OMOD,
gc.ODIV:
a = foptoas(n.Op, nl.Type, 0)
goto abop
}
sbop: // symmetric binary
if nl.Ullman < nr.Ullman || nl.Op == gc.OLITERAL {
nl, nr = nr, nl
}
abop: // asymmetric binary
if nl.Ullman >= nr.Ullman {
var nt gc.Node
gc.Tempname(&nt, nl.Type)
gc.Cgen(nl, &nt)
var n2 gc.Node
gc.Mgen(nr, &n2, nil)
var n1 gc.Node
gc.Regalloc(&n1, nl.Type, res)
gmove(&nt, &n1)
gins(a, &n2, &n1)
gmove(&n1, res)
gc.Regfree(&n1)
gc.Mfree(&n2)
} else {
var n2 gc.Node
gc.Regalloc(&n2, nr.Type, res)
gc.Cgen(nr, &n2)
var n1 gc.Node
gc.Regalloc(&n1, nl.Type, nil)
gc.Cgen(nl, &n1)
gins(a, &n2, &n1)
gc.Regfree(&n2)
gmove(&n1, res)
gc.Regfree(&n1)
}
return
}