sysdeps/s390/multiarch/wcsnlen-vx.S

   1 /* Vector optimized 32/64 bit S/390 version of wcsnlen.
   2    Copyright (C) 2015-2018 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 /* size_t wcsnlen (const wchar_t *s, size_t maxlen)
  27    Returns the number of characters in s or at most maxlen.
  28
  29    Register usage:
  30    -r1=tmp
  31    -r2=address of string
  32    -r3=maxlen (number of characters to be read)
  33    -r4=tmp
  34    -r5=current_len and return_value
  35    -v16=part of s
  36 */
  37 ENTRY(__wcsnlen_vx)
  38
  39         .machine "z13"
  40         .machinemode "zarch_nohighgprs"
  41
  42 # if !defined __s390x__
  43         llgfr   %r3,%r3
  44 # endif /* !defined __s390x__ */
  45
  46         clgfi   %r3,0           /* if maxlen == 0, return 0.  */
  47         locgre  %r2,%r3
  48         ber     %r14
  49
  50         vlbb    %v16,0(%r2),6   /* Load s until next 4k-byte boundary.  */
  51         lcbb    %r1,0(%r2),6    /* Get bytes to 4k-byte boundary or 16.  */
  52         llgfr   %r1,%r1         /* Convert 32bit to 64bit.  */
  53
  54         tmll    %r2,3           /* Test if s is 4-byte aligned?  */
  55         jne     .Lfallback      /* And use common-code variant if not.  */
  56
  57         /* Check range of maxlen and convert to byte-count.  */
  58 # ifdef __s390x__
  59         tmhh    %r3,49152       /* Test bit 0 or 1 of maxlen.  */
  60         lghi    %r4,-4          /* Max byte-count is 18446744073709551612.  */
  61 # else
  62         tmlh    %r3,49152       /* Test bit 0 or 1 of maxlen.  */
  63         llilf   %r4,4294967292  /* Max byte-count is 4294967292.  */
  64 # endif /* !__s390x__ */
  65         sllg    %r3,%r3,2       /* Convert character-count to byte-count.  */
  66         locgrne %r3,%r4         /* Use max byte-count, if bit 0/1 was one.  */
  67
  68         vfenezf %v16,%v16,%v16  /* Find element not equal with zero search.  */
  69         clgr    %r1,%r3
  70         locgrh  %r1,%r3         /* loaded_byte_count
  71                                    = min (loaded_byte_count, maxlen)  */
  72
  73         vlgvb   %r5,%v16,7      /* Load zero index or 16 if not found.  */
  74         clrjl   %r5,%r1,.Lend   /* Found zero within loaded bytes -> return.  */
  75
  76         clgr    %r1,%r3         /* If loaded_byte_count == maxlen -> end.  */
  77         locgre  %r5,%r3
  78         je      .Lend
  79
  80         /* Align s to 16 byte.  */
  81         risbgn  %r4,%r2,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15.  */
  82         lghi    %r5,16          /* current_len = 16.  */
  83         slr     %r5,%r4         /* Compute bytes to 16bytes boundary.  */
  84
  85         lgr     %r1,%r5         /* If %r5 + 64 < maxlen? -> loop64.  */
  86         aghi    %r1,64
  87         clgrjl  %r1,%r3,.Lloop64
  88
  89         /* Find zero in max 64byte with aligned s.  */
  90 .Llt64:
  91         vl      %v16,0(%r5,%r2) /* Load s.  */
  92         vfenezfs %v16,%v16,%v16 /* Find element not equal with zero search.  */
  93         je      .Lfound         /* Jump away if zero was found.  */
  94         aghi    %r5,16
  95         clgrjhe %r5,%r3,.Lfound /* If current_len >= maxlen -> end.  */
  96         vl      %v16,0(%r5,%r2)
  97         vfenezfs %v16,%v16,%v16
  98         je      .Lfound
  99         aghi    %r5,16
 100         clgrjhe %r5,%r3,.Lfound
 101         vl      %v16,0(%r5,%r2)
 102         vfenezfs %v16,%v16,%v16
 103         je      .Lfound
 104         aghi    %r5,16
 105         clgrjhe %r5,%r3,.Lfound
 106         vl      %v16,0(%r5,%r2)
 107         vfenezfs %v16,%v16,%v16
 108         j       .Lfound
 109
 110 .Lfound48:
 111         aghi    %r5,16
 112 .Lfound32:
 113         aghi    %r5,16
 114 .Lfound16:
 115         aghi    %r5,16
 116 .Lfound:
 117         vlgvb   %r4,%v16,7      /* Load byte index of zero or 16 if no zero.  */
 118         algr    %r5,%r4
 119
 120         clgr    %r5,%r3
 121         locgrh  %r5,%r3         /* Return min (current_len, maxlen).  */
 122 .Lend:
 123         srlg    %r2,%r5,2       /* Convert byte-count to character-count.  */
 124         br      %r14
 125
 126         /* Find zero in 16byte aligned loop.  */
 127 .Lloop64:
 128         vl      %v16,0(%r5,%r2) /* Load s.  */
 129         vfenezfs %v16,%v16,%v16 /* Find element not equal with zero search.  */
 130         je      .Lfound         /* Jump away if zero was found.  */
 131         vl      %v16,16(%r5,%r2)
 132         vfenezfs %v16,%v16,%v16
 133         je      .Lfound16
 134         vl      %v16,32(%r5,%r2)
 135         vfenezfs %v16,%v16,%v16
 136         je      .Lfound32
 137         vl      %v16,48(%r5,%r2)
 138         vfenezfs %v16,%v16,%v16
 139         je      .Lfound48
 140
 141         aghi    %r5,64
 142         lgr     %r1,%r5         /* If %r5 + 64 < maxlen? -> loop64.  */
 143         aghi    %r1,64
 144         clgrjl  %r1,%r3,.Lloop64
 145
 146         j       .Llt64
 147
 148 .Lfallback:
 149         jg      __wcsnlen_c
 150 END(__wcsnlen_vx)
 151 #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */