alpha/lib/ev6-stxncpy.S

36  *	E	- either cluster
65 	/* Create the 1st output word and detect 0's in the 1st input word.  */
66 	lda	t2, -1		# E : build a mask against false zero
69 	ornot	t1, t2, t2	# E : (stall)
72 	cmpbge	zero, t2, t8	# E : bits set iff null found
73 	or	t0, t3, t0	# E : (stall)
87 	 *	limit of 1 bcond/quad to permit training
91 	addq	a0, 8, a0	# E :
92 	subq	a2, 1, a2	# E :
96 	addq	a1, 8, a1	# E :
97 	cmpbge	zero, t0, t8	# E :
113 	negq	t8, t12		# E : find low bit set
114 	and	t8, t12, t12	# E : (stall)
117 	and	t12, 0x80, t6	# E : (stall)
118 	bne	t6, 1f		# U : (stall)
123 	subq	t12, 1, t6	# E :
124 	or	t12, t6, t8	# E : (stall)
132 1:	stq_u	t0, 0(a0)	# L :
139 	or	t10, t8, t8	# E :
154 	xor	a0, a1, t1	# E :
155 	and	a0, 7, t0	# E : find dest misalignment
156 	and	t1, 7, t1	# E : (stall)
157 	addq	a2, t0, a2	# E : bias count by dest misalignment (stall)
159 	subq	a2, 1, a2	# E :
160 	and	a2, 7, t2	# E : (stall)
161 	srl	a2, 3, a2	# U : a2 = loop counter = (count - 1)/8 (stall)
162 	addq	zero, 1, t10	# E :
168 	addq	a1, 8, a1	# E :
195 	   t6 == bytemask that is -1 in dest word bytes */
198 	addq	a1, 8, a1	# E :
202 	or	t1, t4, t1	# E : first aligned src word complete (stall)
204 	or	t0, t1, t0	# E : first output word complete (stall)
205 	or	t0, t6, t6	# E : mask original data for zero test (stall)
207 	cmpbge	zero, t6, t8	# E :
209 	lda	t6, -1		# E :
215 	or      t6, t2, t2	# E : (stall)
217 	cmpbge	zero, t2, t8	# E : find nulls in second partial
218 	addq	a0, 8, a0	# E :
219 	subq	a2, 1, a2	# E :
227 	addq	a1, 8, a1	# E :
230 	cmpbge	zero, t2, t8	# E :
250 	or	t0, t1, t0	# E : current dst word now complete
251 	subq	a2, 1, a2	# E : decrement word count
253 	addq	a0, 8, a0	# E :
258 	addq	a1, 8, a1	# E :
261 	cmpbge	zero, t2, t8	# E : test new word for eos
275 	or	t0, t1, t0	# E : first (partial) source word complete
277 	cmpbge	zero, t0, t8	# E : is the null in this first bit? (stall)
281 	addq	a0, 8, a0	# E :
282 	subq	a2, 1, a2	# E :
287 	cmpbge	zero, t0, t8	# E :
288 	or	t8, t10, t6	# E : (stall)
289 	cmoveq	a2, t6, t8	# E : Latency=2, extra map slot (stall)
296 	negq	t8, t6		# E : isolate low bit set
297 	and	t6, t8, t12	# E : (stall)
298 	and	t12, 0x80, t6	# E : avoid dest word load if we can (stall)
299 	bne	t6, 1f		# U : (stall)
302 	subq	t12, 1, t6	# E :
303 	or	t6, t12, t8	# E : (stall)
307 	or	t0, t1, t0	# E : (stall)
311 1:	stq_u	t0, 0(a0)	# L :
318 	and	a1, 7, t6	# E : avoid final load if possible
320 	and	t6, 0xff, t6	# E : (stall)
321 	bne	t6, 1f		# U : (stall)
326 	or	t1, t0, t1	# E : (stall)
328 1:	cmpbge	zero, t1, t8	# E :
329 	mov	t1, t0		# E :
332 	or	t10, t8, t8	# E :
340 	and	a0, 7, t4	# E : find dest misalignment
341 	and	a1, 7, t5	# E : find src misalignment
344 	mov	zero, t0	# E :
346 	mov	zero, t6	# E :
347 	beq	t4, 1f		# U :
349 	lda	t6, -1		# E :
354 	subq	a1, t4, a1	# E : sub dest misalignment from src addr
359 1:	cmplt	t4, t5, t12	# E :
361 	lda	t2, -1		# E : for creating masks later
365 	cmpbge	zero, t1, t8	# E : is there a zero?
366 	andnot	t2, t6, t2	# E : dest mask for a single word copy
367 	or	t8, t10, t5	# E : test for end-of-count too
369 	cmpbge	zero, t2, t3	# E :
370 	cmoveq	a2, t5, t8	# E : Latency=2, extra map slot
371 	nop			# E : keep with cmoveq
372 	andnot	t8, t3, t8	# E : (stall)
380 	negq	t8, t6		# E : build bitmask of bytes <= zero
383 	and	t6, t8, t12	# E :
384 	subq	t12, 1, t6	# E : (stall)
385 	or	t6, t12, t8	# E : (stall)
389 	andnot	t0, t2, t0	# E : zero place for source to reside
390 	or	t0, t1, t0	# E : and put it there (stall both t0, t1)