ADDL	(6*4)(DI), CX
	MOVL	CX, (6*4)(DI)
	MOVL	CX, (6*4)(BP)
	MOVL	(7*4)(BP), DX		// H7 = h + H7
	ADDL	(7*4)(DI), DX
	MOVL	DX, (7*4)(DI)
	MOVL	DX, (7*4)(BP)

	ADDL	$64, SI
	CMPL	SI, 288(SP)
	JB	loop

end:

	MOVQ	AX, 32(SP)
	MOVQ	runtime·_TlsAlloc(SB), AX
	CALL	AX
	MOVQ	32(SP), SP

	MOVQ	AX, CX	// TLS index

	// Assert that slot is less than 64 so we can use _TEB->TlsSlots
	CMPQ	CX, $64
	JB	ok

	// Fallback to the TEB arbitrary pointer.
	// TODO: don't use the arbitrary pointer (see go.dev/issue/59824)
	MOVQ	$TEB_ArbitraryPtr, CX
	JMP	settls
ok:
	// Convert the TLS index at CX into

	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

y z I"},C:{"1":"2 3 4 5 6 7 8 9 AB BB CB DB EB FB GB HB IB JB KB LB MB NB OB PB QB RB SB TB UB VB WB XB YB ZB aB bB cB 0B dB 1B eB fB gB hB iB jB kB lB mB nB oB m pB qB rB sB tB P Q R 2B S T U V W X Y Z a b c d e f g h i j n o p q r s t u v w x y z I uB 3B 4B","2":"0 1 KC zB J K E F G A B C L M H N D O k l LC MC"},D:{"1":"8 9 AB BB CB DB EB FB GB HB IB JB KB LB MB NB OB PB QB RB SB TB UB VB WB XB YB ZB aB bB cB 0B dB 1B eB fB gB hB iB jB kB lB mB nB oB m pB qB rB sB tB P Q R S T U V W X Y Z a b c...

	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

	PSHUFHW $0, X0, X0			// repeat length 4 times total
	MOVO	X0, X1				// save unscrambled seed
	PXOR	runtime·aeskeysched(SB), X0	// xor in per-process seed
	AESENC	X0, X0				// scramble seed

	CMPQ	CX, $16
	JB	aes0to15
	JE	aes16
	CMPQ	CX, $32
	JBE	aes17to32
	CMPQ	CX, $64
	JBE	aes33to64
	CMPQ	CX, $128
	JBE	aes65to128
	JMP	aes129plus

aes0to15:
	TESTQ	CX, CX
	JE	aes0

rstart>=d)&&(a.lookahead=a.strstart-d,a.strstart=d,h(a,!1),0===a.strm.avail_out))return sb;if(a.strstart-a.block_start>=a.w_size-jb&&(h(a,!1),0===a.strm.avail_out))return sb}return a.insert=0,b===K?(h(a,!0),0===a.strm.avail_out?ub:vb):a.strstart>a.block_start&&(h(a,!1),0===a.strm.avail_out)?sb:sb}function o(a,b){for(var c,d;;){if(a.lookahead<jb){if(m(a),a.lookahead<jb&&b===H)return sb;if(0===a.lookahead)break}if(c=0,a.lookahead>=hb&&(a.ins_h=(a.ins_h<<a.hash_shift^a.window[a.strstart+hb-1])&a.ha...

eDIyPOCs2ce5AY0EXH8cdAEMALVFGVBG8Ve4cYqFOL8TFNaFs/vS38w5L/74CEsr
BSOqHCYCc2imxRO72w/fGM8vcSnONnreCVsVxSYC7RAA1LZhuJ9LPquJB1z/2zUQ
z9FdIvo0wXfjoXRbbCVCipfBv6byCtf15+JKdMCT2wWEUMSus5b/794z9MxdW+JB
YAchiY+z9dkRq0MjCyQ0J29Q8hU0pzTq2WpH0MdQLgJlKFjTnpXZa0qTDItcL4hy
4B/U/QdCyJx3xRk91jC7quxdNAZPFkV45AC0z0tVAuB1/18/L4zFUIjmq8OgYOjJ
M89XIVvjTOe+NWGBIwXlgVb+jzPccI0EZ1Jp1icmt0d/NSf4vErYQGvdeqVZd00B