JE dataTLS
	CMPQ autLen, $128
	JB startSinglesLoop
	JMP dataOctaLoop

dataTLS:
	MOVOU (16*14)(pTbl), T1
	MOVOU (16*15)(pTbl), T2
	PXOR B0, B0
	MOVQ (aut), B0
	PINSRD $2, 8(aut), B0
	PINSRB $12, 12(aut), B0
	XORQ autLen, autLen
	JMP dataMul

dataOctaLoop:
		CMPQ autLen, $128
		JB startSinglesLoop
		SUBQ $128, autLen

		MOVOU (16*0)(aut), X0

	// Check that addr is within [arenastart, arenaend) or within [racedatastart, racedataend).
	CMPQ	RARG1, runtime·racearenastart(SB)
	JB	data
	CMPQ	RARG1, runtime·racearenaend(SB)
	JB	call
data:
	CMPQ	RARG1, runtime·racedatastart(SB)
	JB	ret
	CMPQ	RARG1, runtime·racedataend(SB)
	JAE	ret
call:
	MOVQ	AX, AX		// w/o this 6a miscompiles this function
	JMP	racecall<>(SB)
ret:
	RET

		want string
	}{
		{Name(""), byte(7), "7"},
		{Name(""), jsonbyte(7), `{"JB":7}`},
		{Name(""), textbyte(4), `"TB:4"`},
		{Name(""), jsonint(5), `{"JI":5}`},
		{Name(""), textint(1), `"TI:1"`},
		{Name(""), []byte{0, 1}, `"AAE="`},
		{Name(""), []jsonbyte{0, 1}, `[{"JB":0},{"JB":1}]`},
		{Name(""), [][]jsonbyte{{0, 1}, {3}}, `[[{"JB":0},{"JB":1}],[{"JB":3}]]`},
		{Name(""), []textbyte{2, 3}, `["TB:2","TB:3"]`},

	VMOVDQU XDWORD3, XDWORD1

	CMPQ SRND, $4*4*32
	JB   avx2_loop2

	MOVQ dig+0(FP), CTX // d.h[8]
	MOVQ _INP(SP), INP

	addm(  0(CTX), a)
	addm(  4(CTX), b)
	addm(  8(CTX), c)
	addm( 12(CTX), d)
	addm( 16(CTX), e)
	addm( 20(CTX), f)
	addm( 24(CTX), g)
	addm( 28(CTX), h)

	CMPQ _INP_END(SP), INP
	JB   done_hash

	XORQ SRND, SRND

    //  it is explicitly spelled, e.g.
    //  Try<String> t = Try.successful("some")
    //  t.get().length()  // IDEA produces false positive "potential NPE" warning when JB annotations are used.
    //  Kotlin understands it correctly, though. JSpecify annotations of version 0.3 seem to work well too.
    private Try() {
    }

    /**

	MOVL R13, (5*4)(BP)
	ADDL (6*4)(BP), R14 // H6 = g + H6
	MOVL R14, (6*4)(BP)
	ADDL (7*4)(BP), R15 // H7 = h + H7
	MOVL R15, (7*4)(BP)

	ADDQ $64, SI
	CMPQ SI, 256(SP)
	JB   loop

end:
	RET

// shuffle byte order from LE to BE
DATA flip_mask<>+0x00(SB)/8, $0x0405060700010203
DATA flip_mask<>+0x08(SB)/8, $0x0c0d0e0f08090a0b
DATA flip_mask<>+0x10(SB)/8, $0x0405060700010203

	ROUND4(CX, DX, BP, AX, BX, 78)
	ROUND4(BX, CX, DX, BP, AX, 79)

	ADDL	R11, AX
	ADDL	R12, BX
	ADDL	R13, CX
	ADDL	R14, DX
	ADDL	R15, BP

	ADDQ	$64, SI
	CMPQ	SI, DI
	JB	loop

end:
	MOVQ	dig+0(FP), DI
	MOVL	AX, (0*4)(DI)
	MOVL	BX, (1*4)(DI)
	MOVL	CX, (2*4)(DI)
	MOVL	DX, (3*4)(DI)
	MOVL	BP, (4*4)(DI)
	RET

	MOVQ	R13, (5*8)(BP)
	ADDQ	(6*8)(BP), R14	// H6 = g + H6
	MOVQ	R14, (6*8)(BP)
	ADDQ	(7*8)(BP), R15	// H7 = h + H7
	MOVQ	R15, (7*8)(BP)

	ADDQ	$128, SI
	CMPQ	SI, 640(SP)
	JB	loop

end:
	RET

// Version below is based on "Fast SHA512 Implementations on Intel
// Architecture Processors" White-paper

	MOVO	X0, X1                      // save unscrambled seed
	PXOR	runtime·aeskeysched(SB), X0 // xor in per-process seed
	AESENC	X0, X0                      // scramble seed

	CMPL	BX, $16
	JB	aes0to15
	JE	aes16
	CMPL	BX, $32
	JBE	aes17to32
	CMPL	BX, $64
	JBE	aes33to64
	JMP	aes65plus

aes0to15:
	TESTL	BX, BX
	JE	aes0

	ADDL	$16, AX
	TESTW	$0xff0, AX
	JE	endofpage