sysdeps/s390/memmem-arch13.S

   1 /* Vector optimized 32/64 bit S/390 version of memmem.
   2    Copyright (C) 2019-2020 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    <https://www.gnu.org/licenses/>.  */
  18
  19 #include <ifunc-memmem.h>
  20 #if HAVE_MEMMEM_ARCH13
  21 # include "sysdep.h"
  22 # include "asm-syntax.h"
  23         .text
  24
  25 /* void *memmem(const void *haystack=r2, size_t haystacklen=r3,
  26                 const void *needle=r4, size_t needlelen=r5);
  27    Locate a substring.  */
  28 ENTRY(MEMMEM_ARCH13)
  29         .machine "arch13"
  30         .machinemode "zarch_nohighgprs"
  31 # if ! defined __s390x__
  32         llgfr   %r3,%r3
  33         llgfr   %r5,%r5
  34         llgfr   %r4,%r4
  35         llgfr   %r2,%r2
  36 # endif /* ! defined __s390x__ */
  37         clgrjl  %r3,%r5,.Lend_no_match  /* Haystack < needle?  */
  38
  39         /* Jump to fallback if needle > 9.  See also strstr-arch13.S.  */
  40 # if ! HAVE_MEMMEM_Z13
  41 #  error The arch13 variant of memmem needs the z13 variant of memmem!
  42 # endif
  43         clgfi   %r5,9
  44         jh      MEMMEM_Z13
  45
  46         aghik   %r0,%r5,-1              /* vll needs highest index.  */
  47         bc      4,0(%r14)               /* cc==1: return if needle-len == 0.  */
  48         vll     %v18,%r0,0(%r4)         /* Load needle.  */
  49         vlvgb   %v19,%r5,7              /* v19[7] contains length of needle.  */
  50
  51         clgijh  %r3,16,.Lhaystack_larger_16
  52 .Lhaystack_smaller_16_on_bb:
  53         aghik   %r0,%r3,-1              /* vll needs highest index.  */
  54         vll     %v16,%r0,0(%r2)         /* Load haystack.  */
  55 .Lhaystack_smaller_16:
  56         sgr     %r3,%r5                 /* r3 = largest valid match-index.  */
  57         jl      .Lend_no_match          /* Haystack-len < needle-len?  */
  58         vstrs   %v20,%v16,%v18,%v19,0,0
  59         /* Vector string search without zero search where v20 will contain
  60            the index of a partial/full match or 16 (index is named k).
  61            cc=0 (no match; k=16): .Lend_no_match
  62            cc=1 (only available with zero-search): Ignore
  63            cc=2 (full match; k<16): Needle found, but could be beyond haystack!
  64            cc=3 (partial match; k<16): Always at end of v16 and thus beyond!  */
  65         brc     9,.Lend_no_match        /* Jump away if cc == 0 || cc == 3.  */
  66         vlgvb   %r1,%v20,7
  67         /* Verify that the full-match (cc=2) is valid!  */
  68         clgrjh  %r1,%r3,.Lend_no_match  /* Jump away if match is beyond.  */
  69         la      %r2,0(%r1,%r2)
  70         br      %r14
  71 .Lend_no_match:
  72         lghi    %r2,0
  73         br      %r14
  74
  75 .Lhaystack_larger_16:
  76         vl      %v16,0(%r2)
  77         lghi    %r1,17
  78         lay     %r4,-16(%r3,%r2)        /* Boundary for loading with vl.  */
  79         lay     %r0,-64(%r3,%r2)        /* Boundary for loading with 4xvl.  */
  80         /* See also strstr-arch13.S:
  81            min-skip-partial-match-index = (16 - n_len) + 1  */
  82         sgr     %r1,%r5
  83         clgfi   %r3,64                  /* Set Boundary to zero ...  */
  84         la      %r3,0(%r3,%r2)
  85         locghil %r0,0                   /* ... if haystack < 64bytes.  */
  86         jh      .Lloop64
  87 .Lloop:
  88         la      %r2,16(%r2)
  89         /* Vector string search with zero search.  cc=0 => no match.  */
  90         vstrs   %v20,%v16,%v18,%v19,0,0
  91         jne     .Lloop_vstrs_nonzero_cc
  92         clgrjh  %r2,%r4,.Lhaystack_too_small
  93 .Lloop16:
  94         vl      %v16,0(%r2)
  95         la      %r2,16(%r2)
  96         vstrs   %v20,%v16,%v18,%v19,0,0
  97         jne     .Lloop_vstrs_nonzero_cc
  98         clgrjle %r2,%r4,.Lloop16
  99 .Lhaystack_too_small:
 100         sgr     %r3,%r2                 /* r3 = (haystack + len) - curr_pos  */
 101         je      .Lend_no_match          /* Remaining haystack is empty.  */
 102         lcbb    %r0,0(%r2),6
 103         jo      .Lhaystack_smaller_16_on_bb
 104         vl      %v16,0(%r2)             /* Load haystack.  */
 105         j       .Lhaystack_smaller_16
 106
 107 .Lend_match_found:
 108         vlgvb   %r4,%v20,7
 109         sgr     %r2,%r1
 110         la      %r2,0(%r4,%r2)
 111         br      %r14
 112
 113 .Lloop_vstrs_nonzero_cc32:
 114         la      %r2,16(%r2)
 115 .Lloop_vstrs_nonzero_cc16:
 116         la      %r2,16(%r2)
 117 .Lloop_vstrs_nonzero_cc0:
 118         la      %r2,16(%r2)
 119 .Lloop_vstrs_nonzero_cc:
 120         lay     %r2,-16(%r1,%r2)        /* Compute next load address.  */
 121         jh      .Lend_match_found       /* cc == 2 (full match)  */
 122         clgrjh  %r2,%r4,.Lhaystack_too_small
 123         vl      %v16,0(%r2)
 124 .Lloop_vstrs_nonzero_cc_loop:
 125         la      %r2,0(%r1,%r2)
 126         vstrs   %v20,%v16,%v18,%v19,0,0
 127         jh      .Lend_match_found
 128         clgrjh  %r2,%r4,.Lhaystack_too_small
 129         vl      %v16,0(%r2)             /* Next part of haystack.  */
 130         jo      .Lloop_vstrs_nonzero_cc_loop
 131         /* Case: no-match.  */
 132         clgrjh  %r2,%r0,.Lloop  /* Jump away if haystack has less than 64b.  */
 133 .Lloop64:
 134         vstrs   %v20,%v16,%v18,%v19,0,0
 135         jne     .Lloop_vstrs_nonzero_cc0
 136         vl      %v16,16(%r2)            /* Next part of haystack.  */
 137         vstrs   %v20,%v16,%v18,%v19,0,0
 138         jne     .Lloop_vstrs_nonzero_cc16
 139         vl      %v16,32(%r2)            /* Next part of haystack.  */
 140         vstrs   %v20,%v16,%v18,%v19,0,0
 141         jne     .Lloop_vstrs_nonzero_cc32
 142         vl      %v16,48(%r2)            /* Next part of haystack.  */
 143         la      %r2,64(%r2)
 144         vstrs   %v20,%v16,%v18,%v19,0,0
 145         jne     .Lloop_vstrs_nonzero_cc
 146         clgrjh  %r2,%r4,.Lhaystack_too_small
 147         vl      %v16,0(%r2)             /* Next part of haystack.  */
 148         clgrjle %r2,%r0,.Lloop64
 149         j       .Lloop
 150 END(MEMMEM_ARCH13)
 151
 152 # if ! HAVE_MEMMEM_IFUNC
 153 strong_alias (MEMMEM_ARCH13, __memmem)
 154 weak_alias (__memmem, memmem)
 155 # endif
 156
 157 # if MEMMEM_Z13_ONLY_USED_AS_FALLBACK && defined SHARED && IS_IN (libc)
 158 weak_alias (MEMMEM_ARCH13, __GI_memmem)
 159 strong_alias (MEMMEM_ARCH13, __GI___memmem)
 160 # endif
 161 #endif