/****************************************************************************
 * libs/libc/machine/xtensa/arch_memset.S
 *
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.  The
 * ASF licenses this file to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance with the
 * License.  You may obtain a copy of the License at
 *
 *   http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.  See the
 * License for the specific language governing permissions and limitations
 * under the License.
 *
 ****************************************************************************/

/****************************************************************************
 * Included Files
 ****************************************************************************/

#include "xtensa_asm.h"

#include <arch/chip/core-isa.h>
#include <arch/xtensa/xtensa_abi.h>

#include "libc.h"

#ifdef LIBC_BUILD_MEMMOVE

/****************************************************************************
 * Pre-processor Macros
 ****************************************************************************/

/* set to 1 when running on ISS (simulator) with the
   lint or ferret client, or 0 to save a few cycles */

#define SIM_CHECKS_ALIGNMENT  0

/****************************************************************************
 * Public Functions
 ****************************************************************************/
  .text
  .begin schedule
  .global memmove

/*
 * Byte by byte copy
 */
  .align  4
  .byte 0   # 1 mod 4 alignment for LOOPNEZ
        # (0 mod 4 alignment for LBEG)
.Lbytecopy:
#if XCHAL_HAVE_LOOPS
  loopnez a4, .Lbytecopydone
#else /* !XCHAL_HAVE_LOOPS */
  beqz  a4, .Lbytecopydone
  add a7, a3, a4  # a7 = end address for source
#endif /* !XCHAL_HAVE_LOOPS */
.Lnextbyte:
  l8ui  a6, a3, 0
  addi  a3, a3, 1
  s8i a6, a5, 0
  addi  a5, a5, 1
#if !XCHAL_HAVE_LOOPS
  bne a3, a7, .Lnextbyte # continue loop if $a3:src != $a7:src_end
#endif /* !XCHAL_HAVE_LOOPS */
.Lbytecopydone:
  RET(16)

/*
 * Destination is unaligned
 */

  .align  4
.Ldst1mod2: # dst is only byte aligned
  _bltui  a4, 7, .Lbytecopy # do short copies byte by byte

  # copy 1 byte
  l8ui  a6, a3,  0
  addi  a3, a3,  1
  addi  a4, a4, -1
  s8i a6, a5,  0
  addi  a5, a5,  1
  _bbci.l a5, 1, .Ldstaligned # if dst is now aligned, then
          # return to main algorithm
.Ldst2mod4: # dst 16-bit aligned
  # copy 2 bytes
  _bltui  a4, 6, .Lbytecopy # do short copies byte by byte
  l8ui  a6, a3,  0
  l8ui  a7, a3,  1
  addi  a3, a3,  2
  addi  a4, a4, -2
  s8i a6, a5,  0
  s8i a7, a5,  1
  addi  a5, a5,  2
  j .Ldstaligned  # dst is now aligned, return to main algorithm

.Lcommon:
  bbsi.l  a2, 0, .Ldst1mod2 # if dst is 1 mod 2
  bbsi.l  a2, 1, .Ldst2mod4 # if dst is 2 mod 4
.Ldstaligned: # return here from .Ldst?mod? once dst is aligned
  srli  a7, a4, 4 # number of loop iterations with 16B
        # per iteration
  movi  a8, 3   # if source is not aligned,
  bany  a3, a8, .Lsrcunaligned  # then use shifting copy
  /*
   * Destination and source are word-aligned, use word copy.
   */
  # copy 16 bytes per iteration for word-aligned dst and word-aligned src
#if XCHAL_HAVE_LOOPS
  loopnez a7, .Loop1done
#else /* !XCHAL_HAVE_LOOPS */
  beqz  a7, .Loop1done
  slli  a8, a7, 4
  add a8, a8, a3  # a8 = end of last 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.Loop1:
  l32i  a6, a3,  0
  l32i  a7, a3,  4
  s32i  a6, a5,  0
  l32i  a6, a3,  8
  s32i  a7, a5,  4
  l32i  a7, a3, 12
  s32i  a6, a5,  8
  addi  a3, a3, 16
  s32i  a7, a5, 12
  addi  a5, a5, 16
#if !XCHAL_HAVE_LOOPS
  bne a3, a8, .Loop1  # continue loop if a3:src != a8:src_end
#endif /* !XCHAL_HAVE_LOOPS */
.Loop1done:
  bbci.l  a4, 3, .L2
  # copy 8 bytes
  l32i  a6, a3,  0
  l32i  a7, a3,  4
  addi  a3, a3,  8
  s32i  a6, a5,  0
  s32i  a7, a5,  4
  addi  a5, a5,  8
.L2:
  bbsi.l  a4, 2, .L3
  bbsi.l  a4, 1, .L4
  bbsi.l  a4, 0, .L5
  RET(16)
.L3:
  # copy 4 bytes
  l32i  a6, a3,  0
  addi  a3, a3,  4
  s32i  a6, a5,  0
  addi  a5, a5,  4
  bbsi.l  a4, 1, .L4
  bbsi.l  a4, 0, .L5
  RET(16)
.L4:
  # copy 2 bytes
  l16ui a6, a3,  0
  addi  a3, a3,  2
  s16i  a6, a5,  0
  addi  a5, a5,  2
  bbsi.l  a4, 0, .L5
  RET(16)
.L5:
  # copy 1 byte
  l8ui  a6, a3,  0
  s8i a6, a5,  0
  RET(16)

/*
 * Destination is aligned, Source is unaligned
 */

  .align  4
.Lsrcunaligned:
  _beqz a4, .Ldone  # avoid loading anything for zero-length copies
  # copy 16 bytes per iteration for word-aligned dst and unaligned src
  ssa8  a3    # set shift amount from byte offset

#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
  and a11, a3, a8 # save unalignment offset for below
  sub a3, a3, a11 # align a3
#endif
  l32i  a6, a3, 0 # load first word
#if XCHAL_HAVE_LOOPS
  loopnez a7, .Loop2done
#else /* !XCHAL_HAVE_LOOPS */
  beqz  a7, .Loop2done
  slli  a10, a7, 4
  add a10, a10, a3  # a10 = end of last 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.Loop2:
  l32i  a7, a3,  4
  l32i  a8, a3,  8
  src_b a6, a6, a7
  s32i  a6, a5,  0
  l32i  a9, a3, 12
  src_b a7, a7, a8
  s32i  a7, a5,  4
  l32i  a6, a3, 16
  src_b a8, a8, a9
  s32i  a8, a5,  8
  addi  a3, a3, 16
  src_b a9, a9, a6
  s32i  a9, a5, 12
  addi  a5, a5, 16
#if !XCHAL_HAVE_LOOPS
  bne a3, a10, .Loop2 # continue loop if a3:src != a10:src_end
#endif /* !XCHAL_HAVE_LOOPS */
.Loop2done:
  bbci.l  a4, 3, .L12
  # copy 8 bytes
  l32i  a7, a3,  4
  l32i  a8, a3,  8
  src_b a6, a6, a7
  s32i  a6, a5,  0
  addi  a3, a3,  8
  src_b a7, a7, a8
  s32i  a7, a5,  4
  addi  a5, a5,  8
  mov a6, a8
.L12:
  bbci.l  a4, 2, .L13
  # copy 4 bytes
  l32i  a7, a3,  4
  addi  a3, a3,  4
  src_b a6, a6, a7
  s32i  a6, a5,  0
  addi  a5, a5,  4
  mov a6, a7
.L13:
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
  add a3, a3, a11 # readjust a3 with correct misalignment
#endif
  bbsi.l  a4, 1, .L14
  bbsi.l  a4, 0, .L15
.Ldone: RET(16)
.L14:
  # copy 2 bytes
  l8ui  a6, a3,  0
  l8ui  a7, a3,  1
  addi  a3, a3,  2
  s8i a6, a5,  0
  s8i a7, a5,  1
  addi  a5, a5,  2
  bbsi.l  a4, 0, .L15
  RET(16)
.L15:
  # copy 1 byte
  l8ui  a6, a3,  0
  s8i a6, a5,  0
  RET(16)

/*
 * Byte by byte copy
 */
  .align  4
  .byte 0   # 1 mod 4 alignment for LOOPNEZ
        # (0 mod 4 alignment for LBEG)
.Lbackbytecopy:
#if XCHAL_HAVE_LOOPS
  loopnez a4, .Lbackbytecopydone
#else /* !XCHAL_HAVE_LOOPS */
  beqz  a4, .Lbackbytecopydone
  sub a7, a3, a4  # a7 = start address for source
#endif /* !XCHAL_HAVE_LOOPS */
.Lbacknextbyte:
  addi  a3, a3, -1
  l8ui  a6, a3, 0
  addi  a5, a5, -1
  s8i a6, a5, 0
#if !XCHAL_HAVE_LOOPS
  bne a3, a7, .Lbacknextbyte # continue loop if
               # $a3:src != $a7:src_start
#endif /* !XCHAL_HAVE_LOOPS */
.Lbackbytecopydone:
  RET(16)

/*
 * Destination is unaligned
 */

  .align  4
.Lbackdst1mod2: # dst is only byte aligned
  _bltui  a4, 7, .Lbackbytecopy # do short copies byte by byte

  # copy 1 byte
  addi  a3, a3, -1
  l8ui  a6, a3,  0
  addi  a5, a5, -1
  s8i a6, a5,  0
  addi  a4, a4, -1
  _bbci.l a5, 1, .Lbackdstaligned # if dst is now aligned, then
          # return to main algorithm
.Lbackdst2mod4: # dst 16-bit aligned
  # copy 2 bytes
  _bltui  a4, 6, .Lbackbytecopy # do short copies byte by byte
  addi  a3, a3, -2
  l8ui  a6, a3,  0
  l8ui  a7, a3,  1
  addi  a5, a5, -2
  s8i a6, a5,  0
  s8i a7, a5,  1
  addi  a4, a4, -2
  j .Lbackdstaligned  # dst is now aligned,
          # return to main algorithm

  .align  4
memmove:

  ENTRY(16)
  # a2/ dst, a3/ src, a4/ len
  mov a5, a2    # copy dst so that a2 is return value
.Lmovecommon:
  sub a6, a5, a3
  bgeu  a6, a4, .Lcommon

  add a5, a5, a4
  add a3, a3, a4

  bbsi.l  a5, 0, .Lbackdst1mod2 # if dst is 1 mod 2
  bbsi.l  a5, 1, .Lbackdst2mod4 # if dst is 2 mod 4
.Lbackdstaligned: # return here from .Lbackdst?mod? once dst is aligned
  srli  a7, a4, 4 # number of loop iterations with 16B
        # per iteration
  movi  a8, 3   # if source is not aligned,
  bany  a3, a8, .Lbacksrcunaligned  # then use shifting copy
  /*
   * Destination and source are word-aligned, use word copy.
   */
  # copy 16 bytes per iteration for word-aligned dst and word-aligned src
#if XCHAL_HAVE_LOOPS
  loopnez a7, .backLoop1done
#else /* !XCHAL_HAVE_LOOPS */
  beqz  a7, .backLoop1done
  slli  a8, a7, 4
  sub a8, a3, a8  # a8 = start of first 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop1:
  addi  a3, a3, -16
  l32i  a7, a3, 12
  l32i  a6, a3,  8
  addi  a5, a5, -16
  s32i  a7, a5, 12
  l32i  a7, a3,  4
  s32i  a6, a5,  8
  l32i  a6, a3,  0
  s32i  a7, a5,  4
  s32i  a6, a5,  0
#if !XCHAL_HAVE_LOOPS
  bne a3, a8, .backLoop1  # continue loop if a3:src != a8:src_start
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop1done:
  bbci.l  a4, 3, .Lback2
  # copy 8 bytes
  addi  a3, a3, -8
  l32i  a6, a3,  0
  l32i  a7, a3,  4
  addi  a5, a5, -8
  s32i  a6, a5,  0
  s32i  a7, a5,  4
.Lback2:
  bbsi.l  a4, 2, .Lback3
  bbsi.l  a4, 1, .Lback4
  bbsi.l  a4, 0, .Lback5
  RET(16)
.Lback3:
  # copy 4 bytes
  addi  a3, a3, -4
  l32i  a6, a3,  0
  addi  a5, a5, -4
  s32i  a6, a5,  0
  bbsi.l  a4, 1, .Lback4
  bbsi.l  a4, 0, .Lback5
  RET(16)
.Lback4:
  # copy 2 bytes
  addi  a3, a3, -2
  l16ui a6, a3,  0
  addi  a5, a5, -2
  s16i  a6, a5,  0
  bbsi.l  a4, 0, .Lback5
  RET(16)
.Lback5:
  # copy 1 byte
  addi  a3, a3, -1
  l8ui  a6, a3,  0
  addi  a5, a5, -1
  s8i a6, a5,  0
  RET(16)

/*
 * Destination is aligned, Source is unaligned
 */

  .align  4
.Lbacksrcunaligned:
  _beqz a4, .Lbackdone  # avoid loading anything for zero-length copies
  # copy 16 bytes per iteration for word-aligned dst and unaligned src
  ssa8  a3    # set shift amount from byte offset
#undef SIM_CHECKS_ALIGNMENT
#define SIM_CHECKS_ALIGNMENT  1 /* set to 1 when running on ISS with
           * the lint or ferret client, or 0
           * to save a few cycles */
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
  and a11, a3, a8 # save unalignment offset for below
  sub a3, a3, a11 # align a3
#endif
  l32i  a6, a3, 0 # load first word
#if XCHAL_HAVE_LOOPS
  loopnez a7, .backLoop2done
#else /* !XCHAL_HAVE_LOOPS */
  beqz  a7, .backLoop2done
  slli  a10, a7, 4
  sub a10, a3, a10  # a10 = start of first 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop2:
  addi  a3, a3, -16
  l32i  a7, a3, 12
  l32i  a8, a3,  8
  addi  a5, a5, -16
  src_b a6, a7, a6
  s32i  a6, a5, 12
  l32i  a9, a3,  4
  src_b a7, a8, a7
  s32i  a7, a5,  8
  l32i  a6, a3,  0
  src_b a8, a9, a8
  s32i  a8, a5,  4
  src_b a9, a6, a9
  s32i  a9, a5,  0
#if !XCHAL_HAVE_LOOPS
  bne a3, a10, .backLoop2 # continue loop if a3:src != a10:src_start
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop2done:
  bbci.l  a4, 3, .Lback12
  # copy 8 bytes
  addi  a3, a3, -8
  l32i  a7, a3,  4
  l32i  a8, a3,  0
  addi  a5, a5, -8
  src_b a6, a7, a6
  s32i  a6, a5,  4
  src_b a7, a8, a7
  s32i  a7, a5,  0
  mov a6, a8
.Lback12:
  bbci.l  a4, 2, .Lback13
  # copy 4 bytes
  addi  a3, a3, -4
  l32i  a7, a3,  0
  addi  a5, a5, -4
  src_b a6, a7, a6
  s32i  a6, a5,  0
  mov a6, a7
.Lback13:
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
  add a3, a3, a11 # readjust a3 with correct misalignment
#endif
  bbsi.l  a4, 1, .Lback14
  bbsi.l  a4, 0, .Lback15
.Lbackdone:
  RET(16)
.Lback14:
  # copy 2 bytes
  addi  a3, a3, -2
  l8ui  a6, a3,  0
  l8ui  a7, a3,  1
  addi  a5, a5, -2
  s8i a6, a5,  0
  s8i a7, a5,  1
  bbsi.l  a4, 0, .Lback15
  RET(16)
.Lback15:
  # copy 1 byte
  addi  a3, a3, -1
  addi  a5, a5, -1
  l8ui  a6, a3,  0
  s8i a6, a5,  0
  RET(16)

  .end schedule
  .size memmove, . - memmove

#endif