alpha/lib/ev6-copy_page.S

1 /* SPDX-License-Identifier: GPL-2.0 */
3  * arch/alpha/lib/ev6-copy_page.S
13    -----------------------------
28 	9 cycles but I was not able to get it to run that fast -- the initial
34    -------------------------------------
45    --------------------------------------
51 	forced me to add another cycle to the inner-most kernel - up to 11
68 	/* Prefetch 5 read cachelines; write-hint 10 cache lines.  */
69 	wh64	($16)
72 	lda	$1,1*64($16)
74 	wh64	($1)
77 	lda	$1,2*64($16)
79 	wh64	($1)
82 	lda	$1,3*64($16)
84 	wh64	($1)
86 	lda	$1,4*64($16)
87 	lda	$2,5*64($16)
89 	wh64	($1)
91 	lda	$1,6*64($16)
92 	lda	$2,7*64($16)
94 	wh64	($1)
96 	lda	$1,8*64($16)
97 	lda	$2,9*64($16)
99 	wh64	($1)
101 	lda	$19,10*64($16)
104 	/* Main prefetching/write-hinting loop.  */
105 1:	ldq	$0,0($17)
106 	ldq	$1,8($17)
112 	ldq	$2,16($17)
141 	stq	$0,0($16)
142 	subq	$18,1,$18
143 	stq	$1,8($16)
147 	stq	$2,16($16)
149 	stq	$3,24($16)
151 	stq	$4,32($16)
152 	stq	$5,40($16)
156 	stq	$6,48($16)
157 	stq	$7,56($16)
158 	addq	$16,64,$16
159 	bne	$18, 1b
172 	/* Non-prefetching, non-write-hinting cleanup loop for the
175 	ldq	$1,8($17)
176 	ldq	$2,16($17)
184 	stq	$0,0($16)
185 	subq	$18,1,$18
186 	stq	$1,8($16)
189 	stq	$2,16($16)
190 	stq	$3,24($16)
191 	stq	$4,32($16)
192 	stq	$5,40($16)
194 	stq	$6,48($16)
195 	stq	$7,56($16)
196 	addq	$16,64,$16