alpha/lib/ev6-csum_ipv6_magic.S

17  *	E	- either cluster
36  * (we can't hide the 3-cycle latency of the unpkbw in the 6-instruction sequence)
47  * add the 3 low ushorts together, generating a uint
64 	ldq_u	$0,0($16)	# L : Latency: 3
66 	ldq_u	$1,8($16)	# L : Latency: 3
69 	and	$16,7,$6	# E : src misalignment
70 	ldq_u	$5,15($16)	# L : Latency: 3
72 	ldq_u	$2,0($17)	# L : U L U L : Latency: 3
76 	ldq_u	$3,8($17)	# L : Latency: 3
79 	cmoveq	$6,$31,$22	# E : src aligned?
80 	ldq_u	$23,15($17)	# L : Latency: 3
81 	inswl	$18,3,$18	# U : 000000CCDD000000
82 	addl	$19,$7,$19	# E : U L U L : <sign bits>bbaabb00
84 	or	$0,$22,$0	# E : 1st src word complete
86 	or	$18,$4,$18	# E : 000000CCDDAABBCC
89 	and	$17,7,$6	# E : dst misalignment
91 	or	$1,$5,$1	# E : 2nd src word complete
92 	extqh	$3,$6,$22	# U : L U L U :
94 	cmoveq	$6,$31,$22	# E : dst aligned?
95 	extql	$3,$6,$3	# U :
96 	addq	$20,$0,$20	# E : begin summing the words
100 	or	$2,$22,$2	# E : 1st dst word complete
102 	or	$3,$23,$3	# E : U L U L : 2nd dst word complete
104 	cmpult	$20,$0,$0	# E :
105 	addq	$20,$1,$20	# E :
109 	or	$18,$4,$18	# E : 00000000DDCCBBAA
110 	nop			# E :
111 	cmpult	$20,$1,$1	# E :
112 	addq	$20,$2,$20	# E : U L U L
114 	cmpult	$20,$2,$2	# E :
115 	addq	$20,$3,$20	# E :
116 	cmpult	$20,$3,$3	# E : (1 cycle stall on $20)
117 	addq	$20,$18,$20	# E : U L U L (1 cycle stall on $20)
119 	cmpult	$20,$18,$18	# E :
120 	addq	$20,$19,$20	# E : (1 cycle stall on $20)
121 	addq	$0,$1,$0	# E : merge the carries back into the csum
122 	addq	$2,$3,$2	# E :
124 	cmpult	$20,$19,$19	# E :
125 	addq	$18,$19,$18	# E : (1 cycle stall on $19)
126 	addq	$0,$2,$0	# E :
127 	addq	$20,$18,$20	# E : U L U L :
130 	addq	$0,$20,$0	# E :
132 	nop			# E :
135 	addq	$1,$0,$1	# E : Finished generating ulong
137 	zapnot	$1,3,$0		# U : ushort[0] (1 cycle stall on $1)
140 	addq	$0,$2,$0	# E
141 	addq	$0,$1,$3	# E : Finished generating uint
143 	extwl	$3,2,$1		# U : ushort[1] (1 cycle stall on $3)
144 	nop			# E : L U L U
146 	addq	$1,$3,$0	# E : Final carry
147 	not	$0,$4		# E : complement (1 cycle stall on $0)
148 	zapnot	$4,3,$0		# U : clear upper garbage bits