Lines Matching refs:dst1
36 #define dst1 r25
68 mov dst1=dst // copy because of rotation
85 EX(.failure_out,(EPI) st1 [dst1]=val1[PIPE_DEPTH-1],1)
100 and dst2=0x7,dst1 // dst offset
103 // that we can reach 8-byte alignment for both src1 and dst1.
110 // to the dst1. Then copy zeros for the rest of dst1.
122 // Optimization. If dst1 is 8-byte aligned (quite common), we don't need
123 // to copy the head to dst1, to start 8-byte copy software pipeline.
167 EX(.failure_out,(EPI) st1 [dst1]=val1[PIPE_DEPTH-1],1)
184 // Now both src1 and dst1 point to an 8-byte aligned address. And
196 // 3 (EPI): Store tmp to dst1
210 EX(.failure_out,(EPI) st8 [dst1]=tmp,8); \
218 (EPI) st8 [dst1]=tmp,8; \
259 (p14) adds dst1=-8,dst1
260 (p15) sub dst1=dst1,t1
278 EX(.failure_out,(EPI) st1 [dst1]=val1[PIPE_DEPTH-1],1)
317 EX(.failure_out,(p6) st1 [dst1]=val1[0],1)
325 EX(.failure_out,(p7) st2 [dst1]=val1[1],2)
328 EX(.failure_out, (p8) st4 [dst1]=val2[0],4)
332 EX(.failure_out, (p9) st8 [dst1]=val2[1],8)
339 adds dst2=8,dst1
349 EX(.failure_out, (EPI) st8 [dst1]=val1[PIPE_DEPTH-1],16)
369 EX(.failure_out, (p6) st8 [dst1]=val1[0],8)
374 EX(.failure_out,(p7) st4 [dst1]=val1[1],4)
377 EX(.failure_out, (p8) st2 [dst1]=val2[0],2)
380 EX(.failure_out, (p9) st1 [dst1]=val2[1])
391 // In the same loop iteration, the dst1 pointer does not directly
412 (EPI) st1 [dst1]=val1[PIPE_DEPTH-1],1
429 (EPI) st1 [dst1]=val1[PIPE_DEPTH-1],1
432 sub len=enddst,dst1,1 // precompute len
523 st1 [dst1]=r0,1
538 // we MUST use src1/endsrc here and not dst1/enddst because
547 (EPI) st8 [dst1]=val1[PIPE_DEPTH-1],16
551 cmp.ne p6,p0=dst1,enddst // Do we need to finish the tail ?
552 sub len=enddst,dst1,1 // precompute len
562 cmp.ne p6,p0=dst1,enddst // Do we need to finish the tail ?
563 sub len=enddst,dst1,1 // precompute len
575 sub ret0=enddst,dst1