/*
 * forward copy loop
 */
forward:
	CMPQ	BX, $2048
	JLS	move_256through2048

	// If REP MOVSB isn't fast, don't use it
	CMPB	internal∕cpu·X86+const_offsetX86HasERMS(SB), $1 // enhanced REP MOVSB/STOSB
	JNE	fwdBy8

	// Check alignment
	MOVL	SI, AX
	ORL	DI, AX
	TESTL	$7, AX
	JEQ	fwdBy8

	// Do 1 byte at a time
	MOVQ	BX, CX
	REP;	MOVSB
	RET

fwdBy8:
	// Do 8 bytes at a time

	"њ":                            "\u045a",
	"⇍":                           "\u21cd",
	"≦̸":                             "\u2266\u0338",
	"↚":                           "\u219a",
	"‥":                            "\u2025",
	"≰":                             "\u2270",
	"↚":                      "\u219a",
	"↮":                 "\u21ae",

	MOVQ	m_vdsoSP(BX), DX
	MOVQ	CX, 0(SP)
	MOVQ	DX, 8(SP)

	LEAQ	ret+0(FP), DX
	MOVQ	-8(DX), CX
	MOVQ	CX, m_vdsoPC(BX)
	MOVQ	DX, m_vdsoSP(BX)

	CMPQ	R14, m_curg(BX)	// Only switch if on curg.
	JNE	noswitch

	MOVQ	m_g0(BX), DX
	MOVQ	(g_sched+gobuf_sp)(DX), SP	// Set SP to g0 stack

noswitch:
	SUBQ	$16, SP		// Space for results
	ANDQ	$~15, SP	// Align for C code

	MOVL	$1, DI // CLOCK_MONOTONIC

// isCondJmp records the conditional jumps.
var isCondJmp = [maxOp + 1]bool{
	JA:  true,
	JAE: true,
	JB:  true,
	JBE: true,
	JE:  true,
	JG:  true,
	JGE: true,
	JL:  true,
	JLE: true,
	JNE: true,
	JNO: true,
	JNP: true,
	JNS: true,
	JO:  true,
	JP:  true,
	JS:  true,
}

// isLoop records the loop operators.
var isLoop = [maxOp + 1]bool{
	LOOP:   true,

jcloud-ver-jpc.ik-server.com jcloud.ik-server.com jcloud.kz jdevcloud.com jdf.br je jed.wafaicloud.com jeep jefferson.museum jeju.kr jelastic.dogado.eu jelastic.regruhosting.ru jelastic.saveincloud.net jelastic.team jelastic.tsukaeru.net jele.cloud jele.club jele.host jele.io jele.site jelenia-gora.pl jellybean.jp jeonbuk.kr jeonnam.kr jerusalem.museum jessheim.no jetzt jevnaker.no jewelry jewelry.museum jewish.museum jewishart.museum jfk.museum jgora.pl jinsekikogen.hiroshima.jp jio jl.cn jll jls-sto1.elastx.net...

				"application/gzip",
				"application/x-hdf",
				"application/x-hwp",
				"application/x-ibooks+zip",
				"application/x-iso9660-image",
				"application/x-itunes-ipa",
				"application/x-java-jnlp-file",
				"application/x-java-pack200",
				"application/x-kdelnk",
				"application/x-latex",
				"application/x-mobipocket-ebook",
				"application/x-ms-application",
				"application/x-ms-wmd",

	LEAQ    (2*4*80+32)(SP), R15 \
	PRECALC \ // Precalc WK for first 2 blocks
	XCHGQ   R15, R14 \
loop: \  // this loops is unrolled
	CMPQ    R10, R8 \ // we use R8 value (set below) as a signal of a last block
	JNE	begin \
	VZEROUPPER \
	RET \
begin: \
	CALC_0 \
	CALC_1 \
	CALC_2 \
	CALC_3 \
	CALC_4 \
	CALC_5 \
	CALC_6 \
	CALC_7 \
	CALC_8 \
	CALC_9 \
	CALC_10 \

es-1.axarnet.cloud
diadem.cloud
vip.jelastic.cloud
jele.cloud
it1.eur.aruba.jenv-aruba.cloud
it1.jenv-aruba.cloud
keliweb.cloud
cs.keliweb.cloud
oxa.cloud
tn.oxa.cloud
uk.oxa.cloud
primetel.cloud
uk.primetel.cloud
ca.reclaim.cloud
uk.reclaim.cloud
us.reclaim.cloud
ch.trendhosting.cloud
de.trendhosting.cloud
jele.club
amscompute.com
clicketcloud.com

	// add current hash values with previously saved
	PADDD		abefSave, state0
	PADDD		cdghSave, state1

	// advance data pointer; loop until buffer empty
	ADDQ		$64, dataPtr
	CMPQ		numBytes, dataPtr
	JNE		roundLoop

	// write hash values back in the correct order
	PSHUFD		$0x1b, state0, state0		// FEBA
	PSHUFD		$0xb1, state1, state1		// DCHG
	VMOVDQA		state0, m4
	PBLENDW		$0xf0, state1, state0		// DCBA

	//TODO: JGE .+$0x11                     // 7d11
	//TODO: JL .+$0x11223344                // 0f8c44332211 or 480f8c44332211
	//TODO: JL .+$0x11                      // 7c11
	//TODO: JLE .+$0x11223344               // 0f8e44332211 or 480f8e44332211
	//TODO: JLE .+$0x11                     // 7e11
	JCS 2(PC)
	//TODO: JMPQ* (BX)                      // ff23
	JCS 2(PC)
	//TODO: JMPQ* (R11)                     // 41ff23
	JCS 2(PC)