alpha/lib/ev6-memset.S

19  * however the loop has been unrolled to enable better memory throughput,
21  * and __memset16 to permit better scheduling to eliminate the stalling
23  * A future enhancement might be to put in a byte store loop for really
26  * WARNING: Maintaining this is going to be more work than the above version,
27  * as fixes will need to be made in multiple places.  The performance gain
47 	 * Serious stalling happens.  The only way to mitigate this is to
48 	 * undertake a major re-write to interleave the constant materialization
58 	addq $18,$16,$6		# E : max address to write to
99 	 * We are now guaranteed to be quad aligned, with at least
100 	 * one partial quad to write.
103 	sra $18,3,$3		# U : Number of remaining quads to write
104 	and $18,7,$18		# E : Number of trailing bytes to write
109 	 * it's worth the effort to unroll this and use wh64 if possible
114 	 * $6	The max quadword address to write to
116 	 * $3	Number quads to write
120 	subq	$3, 16, $4	# E : Only try to unroll if > 128 bytes
126 	 * through unrolled loop.  Do a quad at a time to get us 0mod64
148 	 * $3 - number quads left to go
150 	 * $17 - mask of stuff to store
154 	 * Assumes the wh64 needs to be for 2 trips through the loop in the future
213 	stq $1,0($5)		# L : And back to memory
236 	 * This is the original body of code, prior to replication and
237 	 * rescheduling.  Leave it here, as there may be calls to this
246 	addq $18,$16,$6		# E : max address to write to
277 	 * We are now guaranteed to be quad aligned, with at least
278 	 * one partial quad to write.
281 	sra $18,3,$3		# U : Number of remaining quads to write
282 	and $18,7,$18		# E : Number of trailing bytes to write
287 	 * it's worth the effort to unroll this and use wh64 if possible
292 	 * $6	The max quadword address to write to
294 	 * $3	Number quads to write
298 	subq	$3, 16, $4	# E : Only try to unroll if > 128 bytes
304 	 * through unrolled loop.  Do a quad at a time to get us 0mod64
326 	 * $3 - number quads left to go
328 	 * $17 - mask of stuff to store
332 	 * Assumes the wh64 needs to be for 2 trips through the loop in the future
391 	stq $1,0($5)		# L : And back to memory
415 	 * to mask stalls.  Note that entry point names also had to change
427 	addq	$18,$16,$6	# E : max address to write to
465 	 * We are now guaranteed to be quad aligned, with at least
466 	 * one partial quad to write.
469 	sra $18,3,$3		# U : Number of remaining quads to write
470 	and $18,7,$18		# E : Number of trailing bytes to write
475 	 * it's worth the effort to unroll this and use wh64 if possible
480 	 * $6	The max quadword address to write to
482 	 * $3	Number quads to write
486 	subq	$3, 16, $4	# E : Only try to unroll if > 128 bytes
492 	 * through unrolled loop.  Do a quad at a time to get us 0mod64
514 	 * $3 - number quads left to go
516 	 * $17 - mask of stuff to store
520 	 * Assumes the wh64 needs to be for 2 trips through the loop in the future
579 	stq $1,0($5)		# L : And back to memory