OpenCores
URL https://opencores.org/ocsvn/openrisc_2011-10-31/openrisc_2011-10-31/trunk

Subversion Repositories openrisc_2011-10-31

[/] [openrisc/] [tags/] [gnu-src/] [newlib-1.18.0/] [newlib-1.18.0-or32-1.0rc1/] [newlib/] [libc/] [machine/] [cris/] [memset.c] - Diff between revs 207 and 345

Only display areas with differences | Details | Blame | View Log

Rev 207 Rev 345
/* A memset for CRIS.
/* A memset for CRIS.
   Copyright (C) 1999-2005 Axis Communications.
   Copyright (C) 1999-2005 Axis Communications.
   All rights reserved.
   All rights reserved.
 
 
   Redistribution and use in source and binary forms, with or without
   Redistribution and use in source and binary forms, with or without
   modification, are permitted provided that the following conditions
   modification, are permitted provided that the following conditions
   are met:
   are met:
 
 
   1. Redistributions of source code must retain the above copyright
   1. Redistributions of source code must retain the above copyright
      notice, this list of conditions and the following disclaimer.
      notice, this list of conditions and the following disclaimer.
 
 
   2. Neither the name of Axis Communications nor the names of its
   2. Neither the name of Axis Communications nor the names of its
      contributors may be used to endorse or promote products derived
      contributors may be used to endorse or promote products derived
      from this software without specific prior written permission.
      from this software without specific prior written permission.
 
 
   THIS SOFTWARE IS PROVIDED BY AXIS COMMUNICATIONS AND ITS CONTRIBUTORS
   THIS SOFTWARE IS PROVIDED BY AXIS COMMUNICATIONS AND ITS CONTRIBUTORS
   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL AXIS
   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL AXIS
   COMMUNICATIONS OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
   COMMUNICATIONS OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
   INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
   INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
   (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
   (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
   SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
   HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
   STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
   STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
   IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
   IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
   POSSIBILITY OF SUCH DAMAGE.  */
   POSSIBILITY OF SUCH DAMAGE.  */
 
 
/* FIXME: This file should really only be used for reference, as the
/* FIXME: This file should really only be used for reference, as the
   result is somewhat depending on gcc generating what we expect rather
   result is somewhat depending on gcc generating what we expect rather
   than what we describe.  An assembly file should be used instead.  */
   than what we describe.  An assembly file should be used instead.  */
 
 
/* Note the multiple occurrence of the expression "12*4", including the
/* Note the multiple occurrence of the expression "12*4", including the
   asm.  It is hard to get it into the asm in a good way.  Thus better to
   asm.  It is hard to get it into the asm in a good way.  Thus better to
   expose the problem everywhere: no macro.  */
   expose the problem everywhere: no macro.  */
 
 
/* Assuming one cycle per dword written or read (ok, not really true; the
/* Assuming one cycle per dword written or read (ok, not really true; the
   world is not ideal), and one cycle per instruction, then 43+3*(n/48-1)
   world is not ideal), and one cycle per instruction, then 43+3*(n/48-1)
   <= 24+24*(n/48-1) so n >= 45.7; n >= 0.9; we win on the first full
   <= 24+24*(n/48-1) so n >= 45.7; n >= 0.9; we win on the first full
   48-byte block to set.  */
   48-byte block to set.  */
 
 
#define MEMSET_BY_BLOCK_THRESHOLD (1 * 48)
#define MEMSET_BY_BLOCK_THRESHOLD (1 * 48)
 
 
/* No name ambiguities in this file.  */
/* No name ambiguities in this file.  */
__asm__ (".syntax no_register_prefix");
__asm__ (".syntax no_register_prefix");
 
 
void *memset(void *pdst, int c, unsigned int plen)
void *memset(void *pdst, int c, unsigned int plen)
{
{
  /* Now we want the parameters in special registers.  Make sure the
  /* Now we want the parameters in special registers.  Make sure the
     compiler does something usable with this.  */
     compiler does something usable with this.  */
 
 
  register char *return_dst __asm__ ("r10") = pdst;
  register char *return_dst __asm__ ("r10") = pdst;
  register int n __asm__ ("r12") = plen;
  register int n __asm__ ("r12") = plen;
  register int lc __asm__ ("r11") = c;
  register int lc __asm__ ("r11") = c;
 
 
  /* Most apps use memset sanely.  Memsetting about 3..4 bytes or less get
  /* Most apps use memset sanely.  Memsetting about 3..4 bytes or less get
     penalized here compared to the generic implementation.  */
     penalized here compared to the generic implementation.  */
 
 
  /* This is fragile performancewise at best.  Check with newer GCC
  /* This is fragile performancewise at best.  Check with newer GCC
     releases, if they compile cascaded "x |= x << 8" to sane code.  */
     releases, if they compile cascaded "x |= x << 8" to sane code.  */
  __asm__("movu.b %0,r13                                                \n\
  __asm__("movu.b %0,r13                                                \n\
           lslq 8,r13                                                   \n\
           lslq 8,r13                                                   \n\
           move.b %0,r13                                                \n\
           move.b %0,r13                                                \n\
           move.d r13,%0                                                \n\
           move.d r13,%0                                                \n\
           lslq 16,r13                                                  \n\
           lslq 16,r13                                                  \n\
           or.d r13,%0"
           or.d r13,%0"
          : "=r" (lc)           /* Inputs.  */
          : "=r" (lc)           /* Inputs.  */
          : "0" (lc)             /* Outputs.  */
          : "0" (lc)             /* Outputs.  */
          : "r13");             /* Trash.  */
          : "r13");             /* Trash.  */
 
 
  {
  {
    register char *dst __asm__ ("r13") = pdst;
    register char *dst __asm__ ("r13") = pdst;
 
 
    if (((unsigned long) pdst & 3) != 0
    if (((unsigned long) pdst & 3) != 0
        /* Oops! n = 0 must be a valid call, regardless of alignment.  */
        /* Oops! n = 0 must be a valid call, regardless of alignment.  */
        && n >= 3)
        && n >= 3)
      {
      {
        if ((unsigned long) dst & 1)
        if ((unsigned long) dst & 1)
          {
          {
            *dst = (char) lc;
            *dst = (char) lc;
            n--;
            n--;
            dst++;
            dst++;
          }
          }
 
 
        if ((unsigned long) dst & 2)
        if ((unsigned long) dst & 2)
          {
          {
            *(short *) dst = lc;
            *(short *) dst = lc;
            n -= 2;
            n -= 2;
            dst += 2;
            dst += 2;
          }
          }
      }
      }
 
 
    /* Decide which setting method to use.  */
    /* Decide which setting method to use.  */
    if (n >= MEMSET_BY_BLOCK_THRESHOLD)
    if (n >= MEMSET_BY_BLOCK_THRESHOLD)
      {
      {
        /* It is not optimal to tell the compiler about clobbering any
        /* It is not optimal to tell the compiler about clobbering any
           registers; that will move the saving/restoring of those registers
           registers; that will move the saving/restoring of those registers
           to the function prologue/epilogue, and make non-block sizes
           to the function prologue/epilogue, and make non-block sizes
           suboptimal.  */
           suboptimal.  */
        __asm__ volatile
        __asm__ volatile
          ("\
          ("\
           ;; GCC does promise correct register allocations, but let's  \n\
           ;; GCC does promise correct register allocations, but let's  \n\
           ;; make sure it keeps its promises.                          \n\
           ;; make sure it keeps its promises.                          \n\
           .ifnc %0-%1-%4,$r13-$r12-$r11                                \n\
           .ifnc %0-%1-%4,$r13-$r12-$r11                                \n\
           .error \"GCC reg alloc bug: %0-%1-%4 != $r13-$r12-$r11\"     \n\
           .error \"GCC reg alloc bug: %0-%1-%4 != $r13-$r12-$r11\"     \n\
           .endif                                                       \n\
           .endif                                                       \n\
                                                                        \n\
                                                                        \n\
           ;; Save the registers we'll clobber in the movem process     \n\
           ;; Save the registers we'll clobber in the movem process     \n\
           ;; on the stack.  Don't mention them to gcc, it will only be \n\
           ;; on the stack.  Don't mention them to gcc, it will only be \n\
           ;; upset.                                                    \n\
           ;; upset.                                                    \n\
           subq    11*4,sp                                              \n\
           subq    11*4,sp                                              \n\
           movem   r10,[sp]                                             \n\
           movem   r10,[sp]                                             \n\
                                                                        \n\
                                                                        \n\
           move.d  r11,r0                                               \n\
           move.d  r11,r0                                               \n\
           move.d  r11,r1                                               \n\
           move.d  r11,r1                                               \n\
           move.d  r11,r2                                               \n\
           move.d  r11,r2                                               \n\
           move.d  r11,r3                                               \n\
           move.d  r11,r3                                               \n\
           move.d  r11,r4                                               \n\
           move.d  r11,r4                                               \n\
           move.d  r11,r5                                               \n\
           move.d  r11,r5                                               \n\
           move.d  r11,r6                                               \n\
           move.d  r11,r6                                               \n\
           move.d  r11,r7                                               \n\
           move.d  r11,r7                                               \n\
           move.d  r11,r8                                               \n\
           move.d  r11,r8                                               \n\
           move.d  r11,r9                                               \n\
           move.d  r11,r9                                               \n\
           move.d  r11,r10                                              \n\
           move.d  r11,r10                                              \n\
                                                                        \n\
                                                                        \n\
           ;; Now we've got this:                                       \n\
           ;; Now we've got this:                                       \n\
           ;; r13 - dst                                                 \n\
           ;; r13 - dst                                                 \n\
           ;; r12 - n                                                   \n\
           ;; r12 - n                                                   \n\
                                                                        \n\
                                                                        \n\
           ;; Update n for the first loop                               \n\
           ;; Update n for the first loop                               \n\
           subq    12*4,r12                                             \n\
           subq    12*4,r12                                             \n\
0:                                                                      \n\
0:                                                                      \n\
"
"
#ifdef __arch_common_v10_v32
#ifdef __arch_common_v10_v32
           /* Cater to branch offset difference between v32 and v10.  We
           /* Cater to branch offset difference between v32 and v10.  We
              assume the branch below has an 8-bit offset.  */
              assume the branch below has an 8-bit offset.  */
"          setf\n"
"          setf\n"
#endif
#endif
"          subq   12*4,r12                                              \n\
"          subq   12*4,r12                                              \n\
           bge     0b                                                   \n\
           bge     0b                                                   \n\
           movem        r11,[r13+]                                      \n\
           movem        r11,[r13+]                                      \n\
                                                                        \n\
                                                                        \n\
           ;; Compensate for last loop underflowing n.                  \n\
           ;; Compensate for last loop underflowing n.                  \n\
           addq   12*4,r12                                              \n\
           addq   12*4,r12                                              \n\
                                                                        \n\
                                                                        \n\
           ;; Restore registers from stack.                             \n\
           ;; Restore registers from stack.                             \n\
           movem [sp+],r10"
           movem [sp+],r10"
 
 
           /* Outputs.  */
           /* Outputs.  */
           : "=r" (dst), "=r" (n)
           : "=r" (dst), "=r" (n)
 
 
           /* Inputs.  */
           /* Inputs.  */
           : "0" (dst), "1" (n), "r" (lc));
           : "0" (dst), "1" (n), "r" (lc));
      }
      }
 
 
    /* An ad-hoc unroll, used for 4*12-1..16 bytes. */
    /* An ad-hoc unroll, used for 4*12-1..16 bytes. */
    while (n >= 16)
    while (n >= 16)
      {
      {
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        n -= 16;
        n -= 16;
      }
      }
 
 
    switch (n)
    switch (n)
      {
      {
      case 0:
      case 0:
        break;
        break;
 
 
      case 1:
      case 1:
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 2:
      case 2:
        *(short *) dst = (short) lc;
        *(short *) dst = (short) lc;
        break;
        break;
 
 
      case 3:
      case 3:
        *(short *) dst = (short) lc; dst += 2;
        *(short *) dst = (short) lc; dst += 2;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 4:
      case 4:
        *(long *) dst = lc;
        *(long *) dst = lc;
        break;
        break;
 
 
      case 5:
      case 5:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 6:
      case 6:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(short *) dst = (short) lc;
        *(short *) dst = (short) lc;
        break;
        break;
 
 
      case 7:
      case 7:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(short *) dst = (short) lc; dst += 2;
        *(short *) dst = (short) lc; dst += 2;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 8:
      case 8:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc;
        *(long *) dst = lc;
        break;
        break;
 
 
      case 9:
      case 9:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 10:
      case 10:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(short *) dst = (short) lc;
        *(short *) dst = (short) lc;
        break;
        break;
 
 
      case 11:
      case 11:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(short *) dst = (short) lc; dst += 2;
        *(short *) dst = (short) lc; dst += 2;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 12:
      case 12:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc;
        *(long *) dst = lc;
        break;
        break;
 
 
      case 13:
      case 13:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
 
 
      case 14:
      case 14:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(short *) dst = (short) lc;
        *(short *) dst = (short) lc;
        break;
        break;
 
 
      case 15:
      case 15:
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(long *) dst = lc; dst += 4;
        *(short *) dst = (short) lc; dst += 2;
        *(short *) dst = (short) lc; dst += 2;
        *dst = (char) lc;
        *dst = (char) lc;
        break;
        break;
      }
      }
  }
  }
 
 
  return return_dst;
  return return_dst;
}
}
 
 

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.