//----------------------------------------------------------------------------//
//                           GNU GPL OS/K                                     //
//                                                                            //
//  Authors:    spectral`                                                     //
//              NeoX                                                          //
//                                                                            //
//  Desc:       mem*() functions                                              //
//----------------------------------------------------------------------------//

#include <kaleid.h>

//------------------------------------------//
//            memset() family               //
//------------------------------------------//

//
// Set "bytes"-many bytes starting from ptr to val
//
void *memset(void *ptr, int val, size_t bytes)
{
    uchar *uptr = (uchar *)ptr;

    //
    // Deal with bytes before start of the first aligned qword
    //
    while (((ulong)uptr % QWORD_ALIGN) > 0 && bytes--) {
        *uptr++ = (uchar)val;
    }

    //
    // At this point we're qword-aligned
    //
    if (bytes > QWORD_SIZE) {
        const ulong uval = ((ulong)val << 56) | ((ulong)val << 48)
                         | ((ulong)val << 40) | ((ulong)val << 32)
                         | ((ulong)val << 24) | ((ulong)val << 16)
                         | ((ulong)val <<  8) | ((ulong)val);

        ulong *uqptr = (ulong *)ptr;

        //
        // Moving fast, qword by qword
        //
        while (bytes > QWORD_SIZE) {
            *uqptr++ = uval;
            bytes -= QWORD_SIZE;
        }

        uptr = (uchar *)(ulong)uqptr;
    }

    //
    // Deal with the few remaining bytes
    //
    while (bytes--) *uptr++ = (uchar)val;

    return ptr;
}

//
//  Set "words"-many words starting from ptr to val
//
void *memsetw(void *ptr, int val, size_t words)
{
    ushort *uptr = (ushort *)ptr;

    //
    // Check whether we can we do this an aligned way
    //
    if unlikely (((ulong)uptr % WORD_ALIGN) > 0) {
        //
        // We can't, so we write word by word all the way up
        //
        while (words--) *uptr++ = (ushort)val;
        return uptr;
    }

    while (((ulong)uptr % QWORD_ALIGN) > 0 && words--) {
        *uptr++ = (ushort)val;
    }

    if (words > QWORDS_TO_WORDS(1)) {
        const ulong uval = ((ulong)val << 48) | ((ulong)val << 32)
                         | ((ulong)val << 16) | ((ulong)val);

        ulong *uqptr = (ulong *)uptr;

        while (words > QWORDS_TO_WORDS(1)) {
            words -= QWORDS_TO_WORDS(1);
            *uqptr++ = uval;
        }

        uptr = (ushort *)(ulong)uqptr;
    }

    while (words--) *uptr++ = (ushort)val;

    return ptr;
}

//
//  Set "dwords"-many dwords starting from ptr to val
//  XXX unimplemented
//
void *memsetd(void *ptr, int val, size_t dwords)
{
    (void)val;
    (void)dwords;

    return ptr;
}

//
//  Set "qwords"-many qwords starting from ptr to val
//
void *memsetq(void *ptr, long val, size_t qwords)
{
    ulong *uptr = (ulong *)ptr;

    //
    // There's no need to check for alignment
    //
    while (qwords--) *uptr++ = (ulong)val;

    return ptr;
}

//------------------------------------------//
//          Other mem*() functions          //
//------------------------------------------//

//
//  Set "bytes"-many bytes starting from ptr to 0
//
void *memzero(void *ptr, size_t bytes)
{
    return memsetb(ptr, 0, bytes);
}

//
//  Copy "bytes"-many bytes of src to dst
//  Does not deal with overlapping blocks (memmove's job)
//
void *memcpy(void *restrict dst, const void *restrict src, size_t bytes)
{
    const ulong  *usrc = (const ulong  *)src;
    ulong        *udst = (ulong        *)dst;

    if unlikely (bytes == 0) return dst;

    //
    // Can align both src and dst at once at once?
    //
    if unlikely ((ulong)src % WORD_ALIGN == 1
              && (ulong)dst % WORD_ALIGN == 1) {
        const uchar *ubsrc = (const uchar *)usrc;
        uchar       *ubdst = (uchar       *)udst;

        //
        // Yes we can, we're guaranteed to be word-aligned now
        //
        *ubdst++ = *ubsrc++;
        bytes--;

        udst = (ulong *)ubdst;
        usrc = (ulong *)ubsrc;
    }

    const ushort *uwsrc = (const ushort *)usrc;
    ushort       *uwdst = (ushort       *)udst;

    //
    // Align either dst or src for qword access
    //
    while ((ulong)dst % QWORD_ALIGN > 0
        && (ulong)src % QWORD_ALIGN > 0
        && bytes > WORD_SIZE) {

        *uwdst++ = *uwsrc++;
        bytes -= WORD_SIZE;
    }

    udst = (ulong *)uwdst;
    usrc = (ulong *)uwsrc;

    //
    // This should be most of the job
    //
    while (bytes > QWORD_SIZE) {
        *udst++ = *usrc++;
        bytes -= QWORD_SIZE;
    }

    const uchar  *ubsrc = (const uchar  *)usrc;
    ushort       *ubdst = (ushort       *)udst;

    //
    // Deal with the few bytes left
    //
    while (bytes--) *ubdst ++ = *ubsrc++;

    return dst;
}

//
//  Move memory from src to dest, even if they overlap
//
void *memmove(void *dst, const void *src, size_t bytes)
{
    const uchar *usrc = src;
    uchar       *udst = dst;

    //
    // Can we use memcpy() safely?
    //
    if (udst < usrc) {
        return memcpy(dst, src, bytes);
    }

    //
    // No, so we go backwards
    //
    uchar *usrc_end = (uchar *)usrc + bytes - 1;
    uchar *udst_end = udst + bytes - 1;
    while (bytes--) *udst_end-- = *usrc_end--;

    return dst;
}

//
//  Compare memory areas
//
int memcmp(const void *ptr1, const void *ptr2, size_t bytes)
{
    const uchar *uptr1 = ptr1;
    const uchar *uptr2 = ptr2;

    while (bytes--) {
        if (*uptr1++ != *uptr2++) {
            return uptr1[-1] < uptr2[-1] ? -1 : 1;
        }
    }

    return 0;
}