func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if cnt < int64(4*gc.Widthptr) {
for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
p = gc.Appendpp(p, ppc64.AMOVD, obj.TYPE_REG, ppc64.REGZERO, 0, obj.TYPE_MEM, ppc64.REGSP, gc.Ctxt.FixedFrameSize()+frame+lo+i)
}
} else if cnt <= int64(128*gc.Widthptr) {
p = gc.Appendpp(p, ppc64.AADD, obj.TYPE_CONST, 0, gc.Ctxt.FixedFrameSize()+frame+lo-8, obj.TYPE_REG, ppc64.REGRT1, 0)
p.Reg = ppc64.REGSP
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0)
gc.Naddr(&p.To, gc.Sysfunc("duffzero"))
p.To.Offset = 4 * (128 - cnt/int64(gc.Widthptr))
} else {
p = gc.Appendpp(p, ppc64.AMOVD, obj.TYPE_CONST, 0, gc.Ctxt.FixedFrameSize()+frame+lo-8, obj.TYPE_REG, ppc64.REGTMP, 0)
p = gc.Appendpp(p, ppc64.AADD, obj.TYPE_REG, ppc64.REGTMP, 0, obj.TYPE_REG, ppc64.REGRT1, 0)
p.Reg = ppc64.REGSP
p = gc.Appendpp(p, ppc64.AMOVD, obj.TYPE_CONST, 0, cnt, obj.TYPE_REG, ppc64.REGTMP, 0)
p = gc.Appendpp(p, ppc64.AADD, obj.TYPE_REG, ppc64.REGTMP, 0, obj.TYPE_REG, ppc64.REGRT2, 0)
p.Reg = ppc64.REGRT1
p = gc.Appendpp(p, ppc64.AMOVDU, obj.TYPE_REG, ppc64.REGZERO, 0, obj.TYPE_MEM, ppc64.REGRT1, int64(gc.Widthptr))
p1 := p
p = gc.Appendpp(p, ppc64.ACMP, obj.TYPE_REG, ppc64.REGRT1, 0, obj.TYPE_REG, ppc64.REGRT2, 0)
p = gc.Appendpp(p, ppc64.ABNE, obj.TYPE_NONE, 0, 0, obj.TYPE_BRANCH, 0, 0)
gc.Patch(p, p1)
}
return p
}
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64, ax *uint32) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if *ax == 0 {
p = gc.Appendpp(p, x86.AMOVL, obj.TYPE_CONST, 0, 0, obj.TYPE_REG, x86.REG_AX, 0)
*ax = 1
}
if cnt <= int64(4*gc.Widthreg) {
for i := int64(0); i < cnt; i += int64(gc.Widthreg) {
p = gc.Appendpp(p, x86.AMOVL, obj.TYPE_REG, x86.REG_AX, 0, obj.TYPE_MEM, x86.REG_SP, frame+lo+i)
}
} else if !gc.Nacl && cnt <= int64(128*gc.Widthreg) {
p = gc.Appendpp(p, x86.ALEAL, obj.TYPE_MEM, x86.REG_SP, frame+lo, obj.TYPE_REG, x86.REG_DI, 0)
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_ADDR, 0, 1*(128-cnt/int64(gc.Widthreg)))
p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
} else {
p = gc.Appendpp(p, x86.AMOVL, obj.TYPE_CONST, 0, cnt/int64(gc.Widthreg), obj.TYPE_REG, x86.REG_CX, 0)
p = gc.Appendpp(p, x86.ALEAL, obj.TYPE_MEM, x86.REG_SP, frame+lo, obj.TYPE_REG, x86.REG_DI, 0)
p = gc.Appendpp(p, x86.AREP, obj.TYPE_NONE, 0, 0, obj.TYPE_NONE, 0, 0)
p = gc.Appendpp(p, x86.ASTOSL, obj.TYPE_NONE, 0, 0, obj.TYPE_NONE, 0, 0)
}
return p
}
// TODO(mips): implement DUFFZERO
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if cnt < int64(4*gc.Widthptr) {
for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
p = gc.Appendpp(p, mips.AMOVW, obj.TYPE_REG, mips.REGZERO, 0, obj.TYPE_MEM, mips.REGSP, gc.Ctxt.FixedFrameSize()+frame+lo+i)
}
} else {
//fmt.Printf("zerorange frame:%v, lo: %v, hi:%v \n", frame ,lo, hi)
// ADD $(FIXED_FRAME+frame+lo-4), SP, r1
// ADD $cnt, r1, r2
// loop:
// MOVW R0, (Widthptr)r1
// ADD $Widthptr, r1
// BNE r1, r2, loop
p = gc.Appendpp(p, mips.AADD, obj.TYPE_CONST, 0, gc.Ctxt.FixedFrameSize()+frame+lo-4, obj.TYPE_REG, mips.REGRT1, 0)
p.Reg = mips.REGSP
p = gc.Appendpp(p, mips.AADD, obj.TYPE_CONST, 0, cnt, obj.TYPE_REG, mips.REGRT2, 0)
p.Reg = mips.REGRT1
p = gc.Appendpp(p, mips.AMOVW, obj.TYPE_REG, mips.REGZERO, 0, obj.TYPE_MEM, mips.REGRT1, int64(gc.Widthptr))
p1 := p
p = gc.Appendpp(p, mips.AADD, obj.TYPE_CONST, 0, int64(gc.Widthptr), obj.TYPE_REG, mips.REGRT1, 0)
p = gc.Appendpp(p, mips.ABNE, obj.TYPE_REG, mips.REGRT1, 0, obj.TYPE_BRANCH, 0, 0)
p.Reg = mips.REGRT2
gc.Patch(p, p1)
}
return p
}
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64, r0 *uint32) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if *r0 == 0 {
p = gc.Appendpp(p, arm.AMOVW, obj.TYPE_CONST, 0, 0, obj.TYPE_REG, arm.REG_R0, 0)
*r0 = 1
}
if cnt < int64(4*gc.Widthptr) {
for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
p = gc.Appendpp(p, arm.AMOVW, obj.TYPE_REG, arm.REG_R0, 0, obj.TYPE_MEM, arm.REGSP, 4+frame+lo+i)
}
} else if !gc.Nacl && (cnt <= int64(128*gc.Widthptr)) {
p = gc.Appendpp(p, arm.AADD, obj.TYPE_CONST, 0, 4+frame+lo, obj.TYPE_REG, arm.REG_R1, 0)
p.Reg = arm.REGSP
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0)
gc.Naddr(&p.To, gc.Sysfunc("duffzero"))
p.To.Offset = 4 * (128 - cnt/int64(gc.Widthptr))
} else {
p = gc.Appendpp(p, arm.AADD, obj.TYPE_CONST, 0, 4+frame+lo, obj.TYPE_REG, arm.REG_R1, 0)
p.Reg = arm.REGSP
p = gc.Appendpp(p, arm.AADD, obj.TYPE_CONST, 0, cnt, obj.TYPE_REG, arm.REG_R2, 0)
p.Reg = arm.REG_R1
p = gc.Appendpp(p, arm.AMOVW, obj.TYPE_REG, arm.REG_R0, 0, obj.TYPE_MEM, arm.REG_R1, 4)
p1 := p
p.Scond |= arm.C_PBIT
p = gc.Appendpp(p, arm.ACMP, obj.TYPE_REG, arm.REG_R1, 0, obj.TYPE_NONE, 0, 0)
p.Reg = arm.REG_R2
p = gc.Appendpp(p, arm.ABNE, obj.TYPE_NONE, 0, 0, obj.TYPE_BRANCH, 0, 0)
gc.Patch(p, p1)
}
return p
}
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if cnt < int64(4*gc.Widthptr) {
for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
p = gc.Appendpp(p, mips.AMOVV, obj.TYPE_REG, mips.REGZERO, 0, obj.TYPE_MEM, mips.REGSP, 8+frame+lo+i)
}
// TODO(dfc): https://golang.org/issue/12108
// If DUFFZERO is used inside a tail call (see genwrapper) it will
// overwrite the link register.
} else if false && cnt <= int64(128*gc.Widthptr) {
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, mips.REGRT1, 0)
p.Reg = mips.REGSP
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0)
gc.Naddr(&p.To, gc.Sysfunc("duffzero"))
p.To.Offset = 8 * (128 - cnt/int64(gc.Widthptr))
} else {
// ADDV $(8+frame+lo-8), SP, r1
// ADDV $cnt, r1, r2
// loop:
// MOVV R0, (Widthptr)r1
// ADDV $Widthptr, r1
// BNE r1, r2, loop
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, mips.REGRT1, 0)
p.Reg = mips.REGSP
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, cnt, obj.TYPE_REG, mips.REGRT2, 0)
p.Reg = mips.REGRT1
p = gc.Appendpp(p, mips.AMOVV, obj.TYPE_REG, mips.REGZERO, 0, obj.TYPE_MEM, mips.REGRT1, int64(gc.Widthptr))
p1 := p
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, int64(gc.Widthptr), obj.TYPE_REG, mips.REGRT1, 0)
p = gc.Appendpp(p, mips.ABNE, obj.TYPE_REG, mips.REGRT1, 0, obj.TYPE_BRANCH, 0, 0)
p.Reg = mips.REGRT2
gc.Patch(p, p1)
}
return p
}
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if cnt < int64(4*gc.Widthptr) {
for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
p = gc.Appendpp(p, mips.AMOVV, obj.TYPE_REG, mips.REGZERO, 0, obj.TYPE_MEM, mips.REGSP, 8+frame+lo+i)
}
} else if cnt <= int64(128*gc.Widthptr) {
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, mips.REGRT1, 0)
p.Reg = mips.REGSP
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0)
gc.Naddr(&p.To, gc.Sysfunc("duffzero"))
p.To.Offset = 8 * (128 - cnt/int64(gc.Widthptr))
} else {
// ADDV $(8+frame+lo-8), SP, r1
// ADDV $cnt, r1, r2
// loop:
// MOVV R0, (Widthptr)r1
// ADDV $Widthptr, r1
// BNE r1, r2, loop
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, mips.REGRT1, 0)
p.Reg = mips.REGSP
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, cnt, obj.TYPE_REG, mips.REGRT2, 0)
p.Reg = mips.REGRT1
p = gc.Appendpp(p, mips.AMOVV, obj.TYPE_REG, mips.REGZERO, 0, obj.TYPE_MEM, mips.REGRT1, int64(gc.Widthptr))
p1 := p
p = gc.Appendpp(p, mips.AADDV, obj.TYPE_CONST, 0, int64(gc.Widthptr), obj.TYPE_REG, mips.REGRT1, 0)
p = gc.Appendpp(p, mips.ABNE, obj.TYPE_REG, mips.REGRT1, 0, obj.TYPE_BRANCH, 0, 0)
p.Reg = mips.REGRT2
gc.Patch(p, p1)
}
return p
}
// zerorange clears the stack in the given range.
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
// Adjust the frame to account for LR.
frame += gc.Ctxt.FixedFrameSize()
offset := frame + lo
reg := int16(s390x.REGSP)
// If the offset cannot fit in a 12-bit unsigned displacement then we
// need to create a copy of the stack pointer that we can adjust.
// We also need to do this if we are going to loop.
if offset < 0 || offset > 4096-clearLoopCutoff || cnt > clearLoopCutoff {
p = gc.Appendpp(p, s390x.AADD, obj.TYPE_CONST, 0, offset, obj.TYPE_REG, s390x.REGRT1, 0)
p.Reg = int16(s390x.REGSP)
reg = s390x.REGRT1
offset = 0
}
// Generate a loop of large clears.
if cnt > clearLoopCutoff {
n := cnt - (cnt % 256)
end := int16(s390x.REGRT2)
p = gc.Appendpp(p, s390x.AADD, obj.TYPE_CONST, 0, offset+n, obj.TYPE_REG, end, 0)
p.Reg = reg
p = gc.Appendpp(p, s390x.AXC, obj.TYPE_MEM, reg, offset, obj.TYPE_MEM, reg, offset)
p.From3 = new(obj.Addr)
p.From3.Type = obj.TYPE_CONST
p.From3.Offset = 256
pl := p
p = gc.Appendpp(p, s390x.AADD, obj.TYPE_CONST, 0, 256, obj.TYPE_REG, reg, 0)
p = gc.Appendpp(p, s390x.ACMP, obj.TYPE_REG, reg, 0, obj.TYPE_REG, end, 0)
p = gc.Appendpp(p, s390x.ABNE, obj.TYPE_NONE, 0, 0, obj.TYPE_BRANCH, 0, 0)
gc.Patch(p, pl)
cnt -= n
}
// Generate remaining clear instructions without a loop.
for cnt > 0 {
n := cnt
// Can clear at most 256 bytes per instruction.
if n > 256 {
n = 256
}
switch n {
// Handle very small clears with move instructions.
case 8, 4, 2, 1:
ins := s390x.AMOVB
switch n {
case 8:
ins = s390x.AMOVD
case 4:
ins = s390x.AMOVW
case 2:
ins = s390x.AMOVH
}
p = gc.Appendpp(p, ins, obj.TYPE_CONST, 0, 0, obj.TYPE_MEM, reg, offset)
// Handle clears that would require multiple move instructions with XC.
default:
p = gc.Appendpp(p, s390x.AXC, obj.TYPE_MEM, reg, offset, obj.TYPE_MEM, reg, offset)
p.From3 = new(obj.Addr)
p.From3.Type = obj.TYPE_CONST
p.From3.Offset = n
}
cnt -= n
offset += n
}
return p
}
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64, ax *uint32, x0 *uint32) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if cnt%int64(gc.Widthreg) != 0 {
// should only happen with nacl
if cnt%int64(gc.Widthptr) != 0 {
gc.Fatalf("zerorange count not a multiple of widthptr %d", cnt)
}
if *ax == 0 {
p = gc.Appendpp(p, x86.AMOVQ, obj.TYPE_CONST, 0, 0, obj.TYPE_REG, x86.REG_AX, 0)
*ax = 1
}
p = gc.Appendpp(p, x86.AMOVL, obj.TYPE_REG, x86.REG_AX, 0, obj.TYPE_MEM, x86.REG_SP, frame+lo)
lo += int64(gc.Widthptr)
cnt -= int64(gc.Widthptr)
}
if cnt == 8 {
if *ax == 0 {
p = gc.Appendpp(p, x86.AMOVQ, obj.TYPE_CONST, 0, 0, obj.TYPE_REG, x86.REG_AX, 0)
*ax = 1
}
p = gc.Appendpp(p, x86.AMOVQ, obj.TYPE_REG, x86.REG_AX, 0, obj.TYPE_MEM, x86.REG_SP, frame+lo)
} else if !isPlan9 && cnt <= int64(8*gc.Widthreg) {
if *x0 == 0 {
p = gc.Appendpp(p, x86.AXORPS, obj.TYPE_REG, x86.REG_X0, 0, obj.TYPE_REG, x86.REG_X0, 0)
*x0 = 1
}
for i := int64(0); i < cnt/16; i++ {
p = gc.Appendpp(p, x86.AMOVUPS, obj.TYPE_REG, x86.REG_X0, 0, obj.TYPE_MEM, x86.REG_SP, frame+lo+i*16)
}
if cnt%16 != 0 {
p = gc.Appendpp(p, x86.AMOVUPS, obj.TYPE_REG, x86.REG_X0, 0, obj.TYPE_MEM, x86.REG_SP, frame+lo+cnt-int64(16))
}
} else if !gc.Nacl && !isPlan9 && (cnt <= int64(128*gc.Widthreg)) {
if *x0 == 0 {
p = gc.Appendpp(p, x86.AXORPS, obj.TYPE_REG, x86.REG_X0, 0, obj.TYPE_REG, x86.REG_X0, 0)
*x0 = 1
}
p = gc.Appendpp(p, leaptr, obj.TYPE_MEM, x86.REG_SP, frame+lo+dzDI(cnt), obj.TYPE_REG, x86.REG_DI, 0)
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_ADDR, 0, dzOff(cnt))
p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
if cnt%16 != 0 {
p = gc.Appendpp(p, x86.AMOVUPS, obj.TYPE_REG, x86.REG_X0, 0, obj.TYPE_MEM, x86.REG_DI, -int64(8))
}
} else {
if *ax == 0 {
p = gc.Appendpp(p, x86.AMOVQ, obj.TYPE_CONST, 0, 0, obj.TYPE_REG, x86.REG_AX, 0)
*ax = 1
}
p = gc.Appendpp(p, x86.AMOVQ, obj.TYPE_CONST, 0, cnt/int64(gc.Widthreg), obj.TYPE_REG, x86.REG_CX, 0)
p = gc.Appendpp(p, leaptr, obj.TYPE_MEM, x86.REG_SP, frame+lo, obj.TYPE_REG, x86.REG_DI, 0)
p = gc.Appendpp(p, x86.AREP, obj.TYPE_NONE, 0, 0, obj.TYPE_NONE, 0, 0)
p = gc.Appendpp(p, x86.ASTOSQ, obj.TYPE_NONE, 0, 0, obj.TYPE_NONE, 0, 0)
}
return p
}
func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
cnt := hi - lo
if cnt == 0 {
return p
}
if cnt < int64(4*gc.Widthptr) {
for i := int64(0); i < cnt; i += int64(gc.Widthptr) {
p = gc.Appendpp(p, arm64.AMOVD, obj.TYPE_REG, arm64.REGZERO, 0, obj.TYPE_MEM, arm64.REGSP, 8+frame+lo+i)
}
} else if cnt <= int64(128*gc.Widthptr) && !darwin { // darwin ld64 cannot handle BR26 reloc with non-zero addend
p = gc.Appendpp(p, arm64.AMOVD, obj.TYPE_REG, arm64.REGSP, 0, obj.TYPE_REG, arm64.REGRT1, 0)
p = gc.Appendpp(p, arm64.AADD, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, arm64.REGRT1, 0)
p.Reg = arm64.REGRT1
p = gc.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0)
gc.Naddr(&p.To, gc.Sysfunc("duffzero"))
p.To.Offset = 4 * (128 - cnt/int64(gc.Widthptr))
} else {
p = gc.Appendpp(p, arm64.AMOVD, obj.TYPE_CONST, 0, 8+frame+lo-8, obj.TYPE_REG, arm64.REGTMP, 0)
p = gc.Appendpp(p, arm64.AMOVD, obj.TYPE_REG, arm64.REGSP, 0, obj.TYPE_REG, arm64.REGRT1, 0)
p = gc.Appendpp(p, arm64.AADD, obj.TYPE_REG, arm64.REGTMP, 0, obj.TYPE_REG, arm64.REGRT1, 0)
p.Reg = arm64.REGRT1
p = gc.Appendpp(p, arm64.AMOVD, obj.TYPE_CONST, 0, cnt, obj.TYPE_REG, arm64.REGTMP, 0)
p = gc.Appendpp(p, arm64.AADD, obj.TYPE_REG, arm64.REGTMP, 0, obj.TYPE_REG, arm64.REGRT2, 0)
p.Reg = arm64.REGRT1
p = gc.Appendpp(p, arm64.AMOVD, obj.TYPE_REG, arm64.REGZERO, 0, obj.TYPE_MEM, arm64.REGRT1, int64(gc.Widthptr))
p.Scond = arm64.C_XPRE
p1 := p
p = gc.Appendpp(p, arm64.ACMP, obj.TYPE_REG, arm64.REGRT1, 0, obj.TYPE_NONE, 0, 0)
p.Reg = arm64.REGRT2
p = gc.Appendpp(p, arm64.ABNE, obj.TYPE_NONE, 0, 0, obj.TYPE_BRANCH, 0, 0)
gc.Patch(p, p1)
}
return p
}
