/*
 * arch/xtensa/lib/hal/memcopy.S -- Core HAL library functions
 * xthal_memcpy and xthal_bcopy
 *
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 *
 * Copyright (C) 2002 - 2005 Tensilica Inc.
 */

#include <variant/core.h>

        .macro  src_b   r, w0, w1
#ifdef __XTENSA_EB__
        src     \r, \w0, \w1
#else
        src     \r, \w1, \w0
#endif
        .endm

        .macro  ssa8    r
#ifdef __XTENSA_EB__
        ssa8b   \r
#else
        ssa8l   \r
#endif
        .endm


/*
 * void *memcpy(void *dst, const void *src, size_t len);
 * void *memmove(void *dst, const void *src, size_t len);
 * void *bcopy(const void *src, void *dst, size_t len);
 *
 * This function is intended to do the same thing as the standard
 * library function memcpy() (or bcopy()) for most cases.
 * However, where the source and/or destination references
 * an instruction RAM or ROM or a data RAM or ROM, that
 * source and/or destination will always be accessed with
 * 32-bit load and store instructions (as required for these
 * types of devices).
 *
 * !!!!!!!  XTFIXME:
 * !!!!!!!  Handling of IRAM/IROM has not yet
 * !!!!!!!  been implemented.
 *
 * The bcopy version is provided here to avoid the overhead
 * of an extra call, for callers that require this convention.
 *
 * The (general case) algorithm is as follows:
 *   If destination is unaligned, align it by conditionally
 *     copying 1 and 2 bytes.
 *   If source is aligned,
 *     do 16 bytes with a loop, and then finish up with
 *     8, 4, 2, and 1 byte copies conditional on the length;
 *   else (if source is unaligned),
 *     do the same, but use SRC to align the source data.
 *   This code tries to use fall-through branches for the common
 *     case of aligned source and destination and multiple
 *     of 4 (or 8) length.
 *
 * Register use:
 *      a0/ return address
 *      a1/ stack pointer
 *      a2/ return value
 *      a3/ src
 *      a4/ length
 *      a5/ dst
 *      a6/ tmp
 *      a7/ tmp
 *      a8/ tmp
 *      a9/ tmp
 *      a10/ tmp
 *      a11/ tmp
 */

        .text
        .align  4
        .global bcopy
        .type   bcopy,@function
bcopy:
        entry   sp, 16          # minimal stack frame
        # a2=src, a3=dst, a4=len
        mov     a5, a3          # copy dst so that a2 is return value
        mov     a3, a2
        mov     a2, a5
        j       .Lcommon        # go to common code for memcpy+bcopy


/*
 * 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
        blt     a3, a7, .Lnextbyte
#endif /* !XCHAL_HAVE_LOOPS */
.Lbytecopydone:
        retw

/*
 * 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

        .align  4
        .global memcpy
        .type   memcpy,@function
memcpy:
        .global memmove
        .type   memmove,@function
memmove:

        entry   sp, 16          # minimal stack frame
        # a2/ dst, a3/ src, a4/ len
        mov     a5, a2          # copy dst so that a2 is return value
.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
        blt     a3, a8, .Loop1
#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
        retw
.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
        retw
.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
        retw
.L5:
        # copy 1 byte
        l8ui    a6, a3,  0
        s8i     a6, a5,  0
        retw

/*
 * 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
#define SIM_CHECKS_ALIGNMENT    1       /* set to 1 when running on ISS (simulator) 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, .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
        blt     a3, a10, .Loop2
#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: retw
.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
        retw
.L15:
        # copy 1 byte
        l8ui    a6, a3,  0
        s8i     a6, a5,  0
        retw

/*
 * Local Variables:
 * mode:fundamental
 * comment-start: "# "
 * comment-start-skip: "# *"
 * End:
 */