Lines Matching +full:2 +full:- +full:5
1 /* SPDX-License-Identifier: GPL-2.0 */
3 * arch/alpha/lib/ev6-memset.S
8 * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com>
13 * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
15 * E - either cluster
16 * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1
17 * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1
41 .align 5
48 * undertake a major re-write to interleave the constant materialization
49 * with other parts of the fall-through code. This is important, even
54 insbl $17,1,$2 # U : 000000000000ch00
59 bis $1,$2,$17 # E : 000000000000chch
60 insbl $1,2,$3 # U : 0000000000ch0000
64 inswl $17,4,$5 # U : 0000chch00000000
66 inswl $17,6,$2 # U : chch000000000000
69 or $2,$5,$2 # E : chchchch00000000
73 or $17,$2,$17 # E : chchchchchchchch
82 bis $16,$16,$5 # E : Save the address
83 insql $17,$16,$2 # U : Insert new bytes
89 bis $2,$4,$1 # E : Final bytes
92 stq_u $1,0($5) # L : Store result
105 bis $16,$16,$5 # E : Save dest address
113 * $5 A copy of $16
119 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
121 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
136 stq $17, 0($5) # L :
139 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
143 addq $5, 8, $5 # E : Inc address
148 * $3 - number quads left to go
149 * $5 - target address (aligned 0mod64)
150 * $17 - mask of stuff to store
151 * Scratch registers available: $7, $2, $4, $1
154 * Assumes the wh64 needs to be for 2 trips through the loop in the future
155 * The wh64 is issued on for the starting destination address for trip +2
162 subq $3, 24, $2 # E : For determining future wh64 addresses
163 stq $17, 0($5) # L :
166 addq $5, 128, $4 # E : speculative target of next wh64
167 stq $17, 8($5) # L :
168 stq $17, 16($5) # L :
169 addq $5, 64, $7 # E : Fallback address for wh64 (== next trip addr)
171 stq $17, 24($5) # L :
172 stq $17, 32($5) # L :
173 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
176 stq $17, 40($5) # L :
177 stq $17, 48($5) # L :
178 subq $3, 16, $2 # E : Repeat the loop at least once more?
181 stq $17, 56($5) # L :
182 addq $5, 64, $5 # E :
184 bge $2, $do_wh64_b # U :
197 stq $17,0($5) # L :
199 addq $5,8,$5 # E : Inc address
208 ldq $7,0($5) # L :
209 mskqh $7,$6,$2 # U : Mask final quad
212 bis $2,$4,$1 # E : Put it all together
213 stq $1,0($5) # L : And back to memory
218 insql $17,$16,$2 # U : New bits
220 bis $2,$4,$2 # E : New result
222 mskql $2,$6,$4 # U :
223 mskqh $1,$6,$2 # U :
224 bis $2,$4,$1 # E :
260 bis $16,$16,$5 # E : Save the address
261 insql $17,$16,$2 # U : Insert new bytes
267 bis $2,$4,$1 # E : Final bytes
270 stq_u $1,0($5) # L : Store result
283 bis $16,$16,$5 # E : Save dest address
291 * $5 A copy of $16
297 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
299 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
314 stq $17, 0($5) # L :
317 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
321 addq $5, 8, $5 # E : Inc address
326 * $3 - number quads left to go
327 * $5 - target address (aligned 0mod64)
328 * $17 - mask of stuff to store
329 * Scratch registers available: $7, $2, $4, $1
332 * Assumes the wh64 needs to be for 2 trips through the loop in the future
333 * The wh64 is issued on for the starting destination address for trip +2
340 subq $3, 24, $2 # E : For determining future wh64 addresses
341 stq $17, 0($5) # L :
344 addq $5, 128, $4 # E : speculative target of next wh64
345 stq $17, 8($5) # L :
346 stq $17, 16($5) # L :
347 addq $5, 64, $7 # E : Fallback address for wh64 (== next trip addr)
349 stq $17, 24($5) # L :
350 stq $17, 32($5) # L :
351 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
354 stq $17, 40($5) # L :
355 stq $17, 48($5) # L :
356 subq $3, 16, $2 # E : Repeat the loop at least once more?
359 stq $17, 56($5) # L :
360 addq $5, 64, $5 # E :
362 bge $2, $do_wh64 # U :
375 stq $17,0($5) # L :
377 addq $5,8,$5 # E : Inc address
386 ldq $7,0($5) # L :
387 mskqh $7,$6,$2 # U : Mask final quad
390 bis $2,$4,$1 # E : Put it all together
391 stq $1,0($5) # L : And back to memory
396 insql $17,$16,$2 # U : New bits
398 bis $2,$4,$2 # E : New result
400 mskql $2,$6,$4 # U :
401 mskqh $1,$6,$2 # U :
402 bis $2,$4,$1 # E :
417 .align 5
424 inswl $17,0,$5 # U : 000000000000c1c2
425 inswl $17,2,$2 # U : 00000000c1c20000
434 or $2,$5,$2 # E : 00000000c1c2c1c2
439 or $17,$2,$17 # E : c1c2c1c2c1c2c1c2
448 bis $16,$16,$5 # E : Save the address
449 insql $17,$16,$2 # U : Insert new bytes
455 bis $2,$4,$1 # E : Final bytes
458 stq_u $1,0($5) # L : Store result
471 bis $16,$16,$5 # E : Save dest address
479 * $5 A copy of $16
485 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
487 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
502 stq $17, 0($5) # L :
505 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
509 addq $5, 8, $5 # E : Inc address
514 * $3 - number quads left to go
515 * $5 - target address (aligned 0mod64)
516 * $17 - mask of stuff to store
517 * Scratch registers available: $7, $2, $4, $1
520 * Assumes the wh64 needs to be for 2 trips through the loop in the future
521 * The wh64 is issued on for the starting destination address for trip +2
528 subq $3, 24, $2 # E : For determining future wh64 addresses
529 stq $17, 0($5) # L :
532 addq $5, 128, $4 # E : speculative target of next wh64
533 stq $17, 8($5) # L :
534 stq $17, 16($5) # L :
535 addq $5, 64, $7 # E : Fallback address for wh64 (== next trip addr)
537 stq $17, 24($5) # L :
538 stq $17, 32($5) # L :
539 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
542 stq $17, 40($5) # L :
543 stq $17, 48($5) # L :
544 subq $3, 16, $2 # E : Repeat the loop at least once more?
547 stq $17, 56($5) # L :
548 addq $5, 64, $5 # E :
550 bge $2, $do_wh64_w # U :
563 stq $17,0($5) # L :
565 addq $5,8,$5 # E : Inc address
574 ldq $7,0($5) # L :
575 mskqh $7,$6,$2 # U : Mask final quad
578 bis $2,$4,$1 # E : Put it all together
579 stq $1,0($5) # L : And back to memory
584 insql $17,$16,$2 # U : New bits
586 bis $2,$4,$2 # E : New result
588 mskql $2,$6,$4 # U :
589 mskqh $1,$6,$2 # U :
590 bis $2,$4,$1 # E :