123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231 |
- #include <sysdep.h>
- #undef ret
- #define saved_lc r15
- #define saved_pr r16
- #define thresh r17
- #define dest r18
- #define dest2 r19
- #define src r20
- #define len r21
- #define asrc r22
- #define tmp r23
- #define pos r24
- #define w0 r25
- #define w1 r26
- #define c r27
- #define sh2 r28
- #define sh1 r29
- #define loopcnt r30
- #define value r31
- ENTRY(strncpy)
- .prologue
- alloc r2 = ar.pfs, 3, 0, 29, 32
- #define MEMLAT 2
- .rotr r[MEMLAT + 2]
- .rotp p[MEMLAT + 1]
- mov ret0 = in0
- .save pr, saved_pr
- mov saved_pr = pr
- .save ar.lc, saved_lc
- mov saved_lc = ar.lc
- mov ar.ec = 0
-
- .body
- cmp.geu p6, p5 = 24, in2
- (p6) br.cond.spnt .short_len
- sub tmp = r0, in0
- mov len = in2
- mov dest = in0
- mov src = in1
- and tmp = 7, tmp
- cmp.eq p6, p7 = tmp, r0
- adds loopcnt = -1, tmp
- (p6) br.cond.sptk .dest_aligned
- sub len = len, tmp
- mov ar.lc = loopcnt
- .l1:
- (p5) ld1 c = [src], 1
-
- st1 [dest] = c, 1
- cmp.ne p5, p7 = c, r0
- br.cloop.dptk .l1
- (p7) br.cond.dpnt .found0_align
- .dest_aligned:
- shr.u c = len, 3
- and sh1 = 7, src
- and asrc = -8, src
- adds c = (MEMLAT-1), c
- sub thresh = 8, sh1
- mov pr.rot = 1 << 16
- shl sh1 = sh1, 3
- mov ar.lc = c
- sub sh2 = 64, sh1
- cmp.eq p6, p0 = sh1, r0
- (p6) br.cond.sptk .src_aligned
- adds c = -(MEMLAT-1), c
- ld8 r[1] = [asrc],8
- mov ar.lc = c
- .align 32
- .l2:
- (p6) st8 [dest] = value, 8
- ld8.s r[0] = [asrc], 8
- shr.u value = r[1], sh1
- czx1.r pos = value
- cmp.lt p7, p0 = pos, thresh
- adds len = -8, len
- (p7) br.cond.dpnt .nonalign_found0
- chk.s r[0], .recovery2
- .back2:
- shl tmp = r[0], sh2
-
- or value = value, tmp
- czx1.r pos = value
- cmp.ne p7, p6 = 8, pos
- (p7) br.cond.dpnt .nonalign_found0
- br.ctop.dptk .l2
- adds len = 8, len
- br.cond.sptk .not_found0
- .nonalign_found0:
- cmp.gtu p6, p0 = -8, len
- (p6) br.cond.dptk .found0
- adds len = 8, len
- br.cond.sptk .not_found0
- .align 32
- .src_aligned:
- .l3:
- (p[0]) ld8.s r[0] = [src], 8
- (p[MEMLAT]) chk.s r[MEMLAT], .recovery3
- .back3:
- (p[MEMLAT]) mov value = r[MEMLAT]
- (p[MEMLAT]) czx1.r pos = r[MEMLAT]
- (p[MEMLAT]) cmp.ne p7, p0 = 8, pos
- (p[MEMLAT]) adds len = -8, len
- (p7) br.cond.dpnt .found0
- (p[MEMLAT]) st8 [dest] = r[MEMLAT], 8
- br.ctop.dptk .l3
- chk.s r[MEMLAT-1], .recovery4
- .back4:
- mov value = r[MEMLAT-1]
- .not_found0:
- cmp.eq p5, p6 = len, r0
- adds len = -1, len
- (p5) br.cond.dptk .restore_and_exit
- mov ar.lc = len
- .l4:
- (p6) extr.u c = value, 0, 8
- (p6) shr.u value = value, 8
- st1 [dest] = c, 1
- cmp.ne p6, p0 = c, r0
- br.cloop.dptk .l4
- br.cond.sptk .restore_and_exit
- .found0_align:
- mov pos = 0
- adds len = -8, len
- mov value = 0
- .found0:
- shl tmp = pos, 3
- shr.u loopcnt = len, 4
- mov c = -1
- cmp.eq p6, p0 = loopcnt, r0
- adds loopcnt = -1, loopcnt
- shl c = c, tmp
- and len = 0xf, len
- andcm value = value, c
- mov ar.lc = loopcnt
- cmp.le p7, p0 = 8, len
- adds dest2 = 16, dest
- st8 [dest] = value, 8
- and len = 0x7, len
- (p6) br.cond.dpnt .l6
- .l5:
- st8 [dest] = r0, 16
- st8 [dest2] = r0, 16
- br.cloop.dptk .l5
- .l6:
- (p7) st8 [dest] = r0, 8
- cmp.eq p5, p0 = len, r0
- adds len = -1, len
- (p5) br.cond.dptk .restore_and_exit
- mov ar.lc = len
- .l7:
- st1 [dest] = r0, 1
- br.cloop.dptk .l7
- .restore_and_exit:
- mov ar.lc = saved_lc
- mov pr = saved_pr, -1
- br.ret.sptk.many b0
- .short_len:
- cmp.eq p5, p0 = in2, r0
- adds loopcnt = -1, in2
- (p5) br.cond.spnt .restore_and_exit
- mov ar.lc = loopcnt
- .l8:
- (p6) ld1 c = [in1], 1
-
- st1 [in0] = c, 1
- (p6) cmp.ne p6, p0 = c, r0
- br.cloop.dptk .l8
-
- mov ar.lc = saved_lc
- mov pr = saved_pr, -1
- br.ret.sptk.many b0
- .recovery2:
- add c = 8, len
- add tmp = -8, asrc
- cmp.gtu p8, p5 = c, thresh
- (p8) ld8 r[0] = [tmp]
- (p5) mov r[0] = r0
- br.cond.sptk .back2
- .recovery3:
- add tmp = -(MEMLAT + 1) * 8, src
- ld8 r[MEMLAT] = [tmp]
- br.cond.sptk .back3
- .recovery4:
- cmp.eq p5, p6 = len, r0
- add tmp = -MEMLAT * 8, src
- (p6) ld8 r[MEMLAT - 1] = [tmp]
- (p5) mov r[MEMLAT - 1] = r0
- br.cond.sptk .back4
- END(strncpy)
- libc_hidden_def (strncpy)
|