xtensa/lib/checksum.S

29  * This function assumes 2- or 4-byte alignment.  Other alignments will fail!
35 	bgeu	sum, val, 99f	; \
44 	 * is aligned on either a 2-byte or 4-byte boundary.
48 	bnez	a5, 8f		/* branch if 2-byte aligned */
49 	/* Fall-through on common case, 4-byte alignment */
53 	loopgtz	a5, 2f
55 	beqz	a5, 2f
61 	l32i	a7, a2, 4
76 	addi	a2, a2, 4*8
81 	extui	a5, a3, 2, 3	/* remaining 4-byte chunks */
83 	loopgtz	a5, 3f
85 	beqz	a5, 3f
87 	add	a5, a5, a2	/* a5 = end of last 4-byte chunk */
92 	addi	a2, a2, 4
97 	_bbci.l	a3, 1, 5f	/* remaining 2-byte chunk */
102 	_bbci.l	a3, 0, 7f	/* remaining 1-byte chunk */
118 	bnez	a5, 8f		/* branch if 1-byte aligned */
124 	j	1b		/* now buf is 4-byte aligned */
131 	srli	a5, a3, 2	/* 4-byte chunks */
133 	loopgtz	a5, 2f
135 	beqz	a5, 2f
137 	add	a5, a5, a2	/* a5 = end of last 4-byte chunk */
152 	addi	a2, a2, 4
157 	_bbci.l	a3, 1, 3f	/* remaining 2-byte chunk, still odd addr */
188     This function is optimized for 4-byte aligned addresses.  Other
198 	/* We optimize the following alignment tests for the 4-byte
206 	beqz	a9, 1f		/* branch if both are 4-byte aligned */
207 	bbsi.l	a10, 0, 5f	/* branch if one address is odd */
208 	j	3f		/* one address is 2-byte aligned */
210 /*	_bbsi.l	a10, 0, 5f */	/* branch if odd address */
211 /*	_bbsi.l	a10, 1, 3f */	/* branch if 2-byte-aligned address */
214 	/* src and dst are both 4-byte aligned */
217 	loopgtz	a10, 2f
219 	beqz	a10, 2f
224 EX(10f)	l32i	a9, a2, 0
225 EX(10f)	l32i	a8, a2, 4
226 EX(10f)	s32i	a9, a3, 0
227 EX(10f)	s32i	a8, a3, 4
230 EX(10f)	l32i	a9, a2, 8
231 EX(10f)	l32i	a8, a2, 12
232 EX(10f)	s32i	a9, a3, 8
233 EX(10f)	s32i	a8, a3, 12
236 EX(10f)	l32i	a9, a2, 16
237 EX(10f)	l32i	a8, a2, 20
238 EX(10f)	s32i	a9, a3, 16
239 EX(10f)	s32i	a8, a3, 20
242 EX(10f)	l32i	a9, a2, 24
243 EX(10f)	l32i	a8, a2, 28
244 EX(10f)	s32i	a9, a3, 24
245 EX(10f)	s32i	a8, a3, 28
254 	extui	a10, a4, 2, 3	/* remaining 4-byte chunks */
257 	loopgtz	a10, 3f
259 	beqz	a10, 3f
261 	add	a10, a10, a2	/* a10 = end of last 4-byte src chunk */
264 EX(10f)	l32i	a9, a2, 0
265 EX(10f)	s32i	a9, a3, 0
267 	addi	a2, a2, 4
268 	addi	a3, a3, 4
275 	to here from the 4-byte alignment case to process, at most,
279 	inefficient, so align your addresses to 4-byte boundaries.
288 	loopgtz	a10, 4f
290 	beqz	a10, 4f
295 EX(10f)	l16ui	a9, a2, 0
296 EX(10f)	s16i	a9, a3, 0
303 4:
305 	_bbci.l	a4, 0, 8f	/* 1-byte chunk */
306 EX(10f)	l8ui	a9, a2, 0
307 EX(10f)	s8i	a9, a3, 0
323 	loopgtz	a10, 6f
325 	beqz	a10, 6f
330 EX(10f)	l8ui	a9, a2, 0
331 EX(10f)	l8ui	a8, a2, 1
332 EX(10f)	s8i	a9, a3, 0
333 EX(10f)	s8i	a8, a3, 1
347 	j	4b		/* process the possible trailing odd byte */