Lines Matching +full:1 +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
53 and $17,255,$1 # E : 00000000000000ch
54 insbl $17,1,$2 # U : 000000000000ch00
55 bis $16,$16,$0 # E : return value
58 addq $18,$16,$6 # E : max address to write to
59 bis $1,$2,$17 # E : 000000000000chch
60 insbl $1,2,$3 # U : 0000000000ch0000
61 insbl $1,3,$4 # U : 00000000ch000000
65 xor $16,$6,$1 # E : will complete write be within one quadword?
70 bic $1,7,$1 # E : fit within a single quadword?
71 and $16,7,$3 # E : Target addr misalignment
74 beq $1,within_quad_b # U :
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
92 stq_u $1,0($5) # L : Store result
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
121 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
125 * We know we've got at least 16 quads, minimum of one trip
133 beq $1, $bigalign_b # U :
137 subq $3, 1, $3 # E : For consistency later
138 addq $1, 8, $1 # E : Increment towards zero for alignment
144 blt $1, $alignmod64_b # U :
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
153 * CWG Section 3.7.6: do not expect a sustained store rate of > 1/cycle
168 stq $17, 16($5) # L :
178 subq $3, 16, $2 # E : Repeat the loop at least once more?
198 subq $3,1,$3 # E : Decrement number quads left
212 bis $2,$4,$1 # E : Put it all together
213 stq $1,0($5) # L : And back to memory
214 ret $31,($26),1 # L0 :
217 ldq_u $1,0($16) # L :
218 insql $17,$16,$2 # U : New bits
219 mskql $1,$16,$4 # U : Clear old
223 mskqh $1,$6,$2 # U :
224 bis $2,$4,$1 # E :
225 stq_u $1,0($16) # L :
231 ret $31,($26),1 # L0 :
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?
251 bic $1,7,$1 # E : fit within a single quadword
252 beq $1,within_one_quad # U :
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
270 stq_u $1,0($5) # L : Store result
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
299 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
303 * We know we've got at least 16 quads, minimum of one trip
311 beq $1, $bigalign # U :
315 subq $3, 1, $3 # E : For consistency later
316 addq $1, 8, $1 # E : Increment towards zero for alignment
322 blt $1, $alignmod64 # U :
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
331 * CWG Section 3.7.6: do not expect a sustained store rate of > 1/cycle
346 stq $17, 16($5) # L :
356 subq $3, 16, $2 # E : Repeat the loop at least once more?
376 subq $3,1,$3 # E : Decrement number quads left
390 bis $2,$4,$1 # E : Put it all together
391 stq $1,0($5) # L : And back to memory
392 ret $31,($26),1 # L0 :
395 ldq_u $1,0($16) # L :
396 insql $17,$16,$2 # U : New bits
397 mskql $1,$16,$4 # U : Clear old
401 mskqh $1,$6,$2 # U :
402 bis $2,$4,$1 # E :
403 stq_u $1,0($16) # L :
409 ret $31,($26),1 # L0 :
426 bis $16,$16,$0 # E : return value
427 addq $18,$16,$6 # E : max address to write to
432 xor $16,$6,$1 # E : will complete write be within one quadword?
436 bic $1,7,$1 # E : fit within a single quadword
437 and $16,7,$3 # E : Target addr misalignment
440 beq $1,within_quad_w # U :
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
458 stq_u $1,0($5) # L : Store result
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
487 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
491 * We know we've got at least 16 quads, minimum of one trip
499 beq $1, $bigalign_w # U :
503 subq $3, 1, $3 # E : For consistency later
504 addq $1, 8, $1 # E : Increment towards zero for alignment
510 blt $1, $alignmod64_w # U :
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
519 * CWG Section 3.7.6: do not expect a sustained store rate of > 1/cycle
534 stq $17, 16($5) # L :
544 subq $3, 16, $2 # E : Repeat the loop at least once more?
564 subq $3,1,$3 # E : Decrement number quads left
578 bis $2,$4,$1 # E : Put it all together
579 stq $1,0($5) # L : And back to memory
580 ret $31,($26),1 # L0 :
583 ldq_u $1,0($16) # L :
584 insql $17,$16,$2 # U : New bits
585 mskql $1,$16,$4 # U : Clear old
589 mskqh $1,$6,$2 # U :
590 bis $2,$4,$1 # E :
591 stq_u $1,0($16) # L :
597 ret $31,($26),1 # L0 :