Lines Matching +full:4 +full:- +full:16
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
48 * undertake a major re-write to interleave the constant materialization
49 * with other parts of the fall-through code. This is important, even
55 bis $16,$16,$0 # E : return value
58 addq $18,$16,$6 # E : max address to write to
61 insbl $1,3,$4 # U : 00000000ch000000
63 or $3,$4,$3 # E : 00000000chch0000
64 inswl $17,4,$5 # U : 0000chch00000000
65 xor $16,$6,$1 # E : will complete write be within one quadword?
71 and $16,7,$3 # E : Target addr misalignment
81 ldq_u $4,0($16) # L : Fetch first partial
82 bis $16,$16,$5 # E : Save the address
83 insql $17,$16,$2 # U : Insert new bytes
87 mskql $4,$16,$4 # U : clear relevant parts of the quad
88 subq $16,$3,$16 # E : $16 is new aligned destination
89 bis $2,$4,$1 # E : Final bytes
96 .align 4
105 bis $16,$16,$5 # E : Save dest address
112 * $16 Current destination address
113 * $5 A copy of $16
119 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
120 subq $3, 16, $4 # E : Only try to unroll if > 128 bytes
122 blt $4, loop_b # U :
125 * We know we've got at least 16 quads, minimum of one trip
139 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
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
161 wh64 ($4) # L1 : memory subsystem write hint
166 addq $5, 128, $4 # E : speculative target of next wh64
168 stq $17, 16($5) # L :
173 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
178 subq $3, 16, $2 # E : Repeat the loop at least once more?
191 .align 4
211 insqh $17,$6,$4 # U : New bits
212 bis $2,$4,$1 # E : Put it all together
217 ldq_u $1,0($16) # L :
218 insql $17,$16,$2 # U : New bits
219 mskql $1,$16,$4 # U : Clear old
220 bis $2,$4,$2 # E : New result
222 mskql $2,$6,$4 # U :
224 bis $2,$4,$1 # E :
225 stq_u $1,0($16) # L :
240 .align 4
246 addq $18,$16,$6 # E : max address to write to
247 bis $16,$16,$0 # E : return value
248 xor $16,$6,$1 # E : will complete write be within one quadword?
253 and $16,7,$3 # E : Target addr misalignment
259 ldq_u $4,0($16) # L : Fetch first partial
260 bis $16,$16,$5 # E : Save the address
261 insql $17,$16,$2 # U : Insert new bytes
265 mskql $4,$16,$4 # U : clear relevant parts of the quad
266 subq $16,$3,$16 # E : $16 is new aligned destination
267 bis $2,$4,$1 # E : Final bytes
274 .align 4
283 bis $16,$16,$5 # E : Save dest address
290 * $16 Current destination address
291 * $5 A copy of $16
297 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
298 subq $3, 16, $4 # E : Only try to unroll if > 128 bytes
300 blt $4, loop # U :
303 * We know we've got at least 16 quads, minimum of one trip
317 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
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
339 wh64 ($4) # L1 : memory subsystem write hint
344 addq $5, 128, $4 # E : speculative target of next wh64
346 stq $17, 16($5) # L :
351 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
356 subq $3, 16, $2 # E : Repeat the loop at least once more?
369 .align 4
389 insqh $17,$6,$4 # U : New bits
390 bis $2,$4,$1 # E : Put it all together
395 ldq_u $1,0($16) # L :
396 insql $17,$16,$2 # U : New bits
397 mskql $1,$16,$4 # U : Clear old
398 bis $2,$4,$2 # E : New result
400 mskql $2,$6,$4 # U :
402 bis $2,$4,$1 # E :
403 stq_u $1,0($16) # L :
426 bis $16,$16,$0 # E : return value
427 addq $18,$16,$6 # E : max address to write to
430 inswl $17,4,$3 # U : 0000c1c200000000
431 inswl $17,6,$4 # U : c1c2000000000000
432 xor $16,$6,$1 # E : will complete write be within one quadword?
435 or $3,$4,$17 # E : c1c2c1c200000000
437 and $16,7,$3 # E : Target addr misalignment
447 ldq_u $4,0($16) # L : Fetch first partial
448 bis $16,$16,$5 # E : Save the address
449 insql $17,$16,$2 # U : Insert new bytes
453 mskql $4,$16,$4 # U : clear relevant parts of the quad
454 subq $16,$3,$16 # E : $16 is new aligned destination
455 bis $2,$4,$1 # E : Final bytes
462 .align 4
471 bis $16,$16,$5 # E : Save dest address
478 * $16 Current destination address
479 * $5 A copy of $16
485 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
486 subq $3, 16, $4 # E : Only try to unroll if > 128 bytes
488 blt $4, loop_w # U :
491 * We know we've got at least 16 quads, minimum of one trip
505 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
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
527 wh64 ($4) # L1 : memory subsystem write hint
532 addq $5, 128, $4 # E : speculative target of next wh64
534 stq $17, 16($5) # L :
539 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
544 subq $3, 16, $2 # E : Repeat the loop at least once more?
557 .align 4
577 insqh $17,$6,$4 # U : New bits
578 bis $2,$4,$1 # E : Put it all together
583 ldq_u $1,0($16) # L :
584 insql $17,$16,$2 # U : New bits
585 mskql $1,$16,$4 # U : Clear old
586 bis $2,$4,$2 # E : New result
588 mskql $2,$6,$4 # U :
590 bis $2,$4,$1 # E :
591 stq_u $1,0($16) # L :