sysdeps/s390/multiarch/memccpy-vx.S

   1 /* Vector optimized 32/64 bit S/390 version of memccpy.
   2    Copyright (C) 2015-2016 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, see
  17    <http://www.gnu.org/licenses/>.  */
  18
  19 #if defined HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc)
  20
  21 # include "sysdep.h"
  22 # include "asm-syntax.h"
  23
  24         .text
  25
  26 /* void *memccpy (void * dest, const void *src, int c, size_t n)
  27    Copies no more than n bytes from src to dest,
  28    stopping when the character c is found
  29    and returns pointer next to c in dest or null if c not found.
  30
  31    Register usage:
  32    -r0=tmp
  33    -r1=tmp
  34    -r2=dest
  35    -r3=src
  36    -r4=c
  37    -r5=n
  38    -r6=current_len
  39    -v16=part of s
  40    -v17=index of found c
  41    -v18=c replicated
  42    -v19=part #2 of s
  43    -v31=save area for r6
  44 */
  45 ENTRY(__memccpy_vx)
  46         .machine "z13"
  47         .machinemode "zarch_nohighgprs"
  48
  49 # if !defined __s390x__
  50         llgfr   %r5,%r5
  51 # endif /* !defined __s390x__ */
  52
  53         vlvgp   %v31,%r6,%r7    /* Save registers.  */
  54         clgije  %r5,0,.Lnf_end  /* If len == 0 then exit.  */
  55
  56         vlbb    %v16,0(%r3),6   /* Load s until next 4k-byte boundary.  */
  57         lcbb    %r0,0(%r3),6    /* Get bytes to 4k-byte boundary or 16.  */
  58         llgfr   %r0,%r0         /* Convert 32bit to 64bit.  */
  59
  60         vlvgb   %v18,%r4,0      /* Generate vector which elements are all c.
  61                                    if c > 255, c will be truncated.  */
  62         vrepb   %v18,%v18,0
  63         lghi    %r6,0           /* current_len = 0.  */
  64
  65         clgrjle %r5,%r0,.Lremaining_v16 /* If maxlen <= loaded-bytes
  66                                            -> Process remaining.  */
  67
  68         vfeebs  %v17,%v16,%v18  /* Find c.  */
  69         vlgvb   %r1,%v17,7      /* Load byte index of c.  */
  70         clgrjl  %r1,%r0,.Lfound_v16 /* Found c is within loaded bytes.  */
  71
  72         /* Align s to 16 byte.  */
  73         risbgn  %r1,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15.  */
  74         lghi    %r6,15          /* current_len = 15.  */
  75         slr     %r6,%r1         /* Compute highest index to 16byte boundary.  */
  76
  77         vstl    %v16,%r6,0(%r2) /* Store prcessed bytes */
  78         ahi     %r6,1
  79
  80 .Lpreloop1:
  81         /* Now we are 16byte aligned, so we can load
  82            a full vreg without page fault.  */
  83         vl      %v16,0(%r6,%r3) /* Load s.  */
  84         clgijl  %r5,17,.Lremaining_v16  /* If n <= 16,
  85                                            process remaining bytes.  */
  86         lgr     %r7,%r5
  87         slgfi   %r7,16          /* border_len = n - 16.  */
  88         j       .Lloop1
  89
  90 .Lloop2:
  91         vl      %v16,16(%r6,%r3)
  92         vst     %v19,0(%r6,%r2)
  93         aghi    %r6,16
  94
  95 .Lloop1:
  96         clgrjhe %r6,%r7,.Lremaining_v16 /* If current_len >= border
  97                                            then process remaining bytes.  */
  98         vfeebs  %v17,%v16,%v18  /* Find c.  */
  99         jl      .Lfound_v16     /* Jump away if c was found.  */
 100         vl      %v19,16(%r6,%r3) /* Load next s part.  */
 101         vst     %v16,0(%r6,%r2) /* Store previous part without c.  */
 102         aghi    %r6,16
 103
 104         clgrjhe %r6,%r7,.Lremaining_v19
 105         vfeebs  %v17,%v19,%v18
 106         jl      .Lfound_v19
 107         vl      %v16,16(%r6,%r3)
 108         vst     %v19,0(%r6,%r2)
 109         aghi    %r6,16
 110
 111         clgrjhe %r6,%r7,.Lremaining_v16
 112         vfeebs  %v17,%v16,%v18
 113         jl      .Lfound_v16
 114         vl      %v19,16(%r6,%r3)
 115         vst     %v16,0(%r6,%r2)
 116         aghi    %r6,16
 117
 118         clgrjhe %r6,%r7,.Lremaining_v19
 119         vfeebs  %v17,%v19,%v18
 120         jo      .Lloop2
 121
 122 .Lfound_v19:
 123         vlr     %v16,%v19
 124 .Lfound_v16:
 125         /* v16 contains c. Store remaining bytes to c. currlen hasn´t
 126            reached border, thus checking for maxlen is not needed! */
 127         vlgvb   %r1,%v17,7      /* Load byte index of c.  */
 128         la      %r2,0(%r6,%r2)  /* vstl has no support for index-register.  */
 129 .Lfound_v16_store:
 130         vstl    %v16,%r1,0(%r2) /* Copy bytes including c.  */
 131         la      %r2,1(%r1,%r2)  /* Return pointer next to c in dest.  */
 132         vlgvg   %r6,%v31,0
 133         vlgvg   %r7,%v31,1
 134         br      %r14
 135
 136 .Lremaining_v19:
 137         vlr     %v16,%v19
 138 .Lremaining_v16:
 139         /* v16 contains the remaining bytes [1...16].
 140            Check and store remaining bytes.  */
 141         vfeebs  %v17,%v16,%v18
 142         slgrk   %r7,%r5,%r6     /* Remaining bytes = maxlen - current_len.  */
 143         aghi    %r7,-1          /* vstl needs highest index.  */
 144         la      %r2,0(%r6,%r2)  /* vstl has no index register.  */
 145         vlgvb   %r1,%v17,7      /* Load index of c or 16 if not found.  */
 146         /* c in remaining bytes? -> Jump away (c-index <= max-index)  */
 147         clrjle  %r1,%r7,.Lfound_v16_store
 148         vstl    %v16,%r7,0(%r2) /* Store remaining bytes.  */
 149
 150 .Lnf_end:
 151         vlgvg   %r6,%v31,0
 152         vlgvg   %r7,%v31,1
 153         lghi    %r2,0           /* Return null.  */
 154         br      %r14
 155 END(__memccpy_vx)
 156 #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */