// Function-wrapped Watcom pragmas
// by Jonathon Fowler (jf@jonof.id.au)
//
// These functions represent some of the more longer-winded pragmas
// from the original pragmas.h wrapped into functions for easier
// use since many jumps and whatnot make it harder to write macro-
// inline versions. I'll eventually convert these to macro-inline
// equivalents.		--Jonathon

#include "compat.h"
#include "pragmas.h"

libdivide_s64pad_t divtable64[DIVTABLESIZE];
libdivide_s32pad_t divtable32[DIVTABLESIZE];

uint32_t divideu32_noinline(uint32_t n, uint32_t d) { return divideu32(n, d); }
int32_t tabledivide32_noinline(int32_t n, int32_t d) { return tabledivide32(n, d); }
int32_t tabledivide64_noinline(int64_t n, int32_t d) { return tabledivide64(n, d); }

int32_t dmval;

#if defined(__GNUC__) && defined(__i386__) && !defined(NOASM)	// NOASM

//
// GCC Inline Assembler version
//

#define ASM __asm__ __volatile__

void clearbufbyte(void *D, int32_t c, int32_t a)
{
    ASM(
        "cmpl $4, %%ecx\n\t"
        "jae 1f\n\t"
        "testb $1, %%cl\n\t"
        "jz 0f\n\t"			// jz preskip
        "stosb\n\t"
        "0:\n\t"			// preskip:
        "shrl $1, %%ecx\n\t"
        "rep\n\t"
        "stosw\n\t"
        "jmp 5f\n\t"			// jmp endit
        "1:\n\t"			// intcopy:
        "testl $1, %%edi\n\t"
        "jz 2f\n\t"			// jz skip1
        "stosb\n\t"
        "decl %%ecx\n\t"
        "2:\n\t"			// skip1:
        "testl $2, %%edi\n\t"
        "jz 3f\n\t"			// jz skip2
        "stosw\n\t"
        "subl $2, %%ecx\n\t"
        "3:\n\t"			// skip2:
        "movl %%ecx, %%ebx\n\t"
        "shrl $2, %%ecx\n\t"
        "rep\n\t"
        "stosl\n\t"
        "testb $2, %%bl\n\t"
        "jz 4f\n\t"			// jz skip3
        "stosw\n\t"
        "4:\n\t"			// skip3:
        "testb $1, %%bl\n\t"
        "jz 5f\n\t"			// jz endit
        "stosb\n\t"
        "5:"				// endit
    : "+D"(D), "+c"(c), "+a"(a) :
            : "ebx", "memory", "cc"
        );
}

void copybufbyte(const void *S, void *D, int32_t c)
{
    ASM(
        "cmpl $4, %%ecx\n\t"		// cmp ecx, 4
        "jae 1f\n\t"
        "testb $1, %%cl\n\t"		// test cl, 1
        "jz 0f\n\t"
        "movsb\n\t"
        "0:\n\t"			// preskip:
        "shrl $1, %%ecx\n\t"		// shr ecx, 1
        "rep\n\t"
        "movsw\n\t"
        "jmp 5f\n\t"
        "1:\n\t"			// intcopy:
        "testl $1, %%edi\n\t"		// test edi, 1
        "jz 2f\n\t"
        "movsb\n\t"
        "decl %%ecx\n\t"
        "2:\n\t"			// skip1:
        "testl $2, %%edi\n\t"		// test edi, 2
        "jz 3f\n\t"
        "movsw\n\t"
        "subl $2, %%ecx\n\t"		// sub ecx, 2
        "3:\n\t"			// skip2:
        "movl %%ecx, %%ebx\n\t"		// mov ebx, ecx
        "shrl $2, %%ecx\n\t"		// shr ecx ,2
        "rep\n\t"
        "movsl\n\t"
        "testb $2, %%bl\n\t"		// test bl, 2
        "jz 4f\n\t"
        "movsw\n\t"
        "4:\n\t"			// skip3:
        "testb $1, %%bl\n\t"		// test bl, 1
        "jz 5f\n\t"
        "movsb\n\t"
        "5:"				// endit:
    : "+c"(c), "+S"(S), "+D"(D) :
            : "ebx", "memory", "cc"
        );
}

void copybufreverse(const void *S, void *D, int32_t c)
{
    ASM(
        "shrl $1, %%ecx\n\t"
        "jnc 0f\n\t"		// jnc skipit1
        "movb (%%esi), %%al\n\t"
        "decl %%esi\n\t"
        "movb %%al, (%%edi)\n\t"
        "incl %%edi\n\t"
        "0:\n\t"		// skipit1:
        "shrl $1, %%ecx\n\t"
        "jnc 1f\n\t"		// jnc skipit2
        "movw -1(%%esi), %%ax\n\t"
        "subl $2, %%esi\n\t"
        "rorw $8, %%ax\n\t"
        "movw %%ax, (%%edi)\n\t"
        "addl $2, %%edi\n\t"
        "1:\n\t"		// skipit2
        "testl %%ecx, %%ecx\n\t"
        "jz 3f\n\t"		// jz endloop
        "2:\n\t"		// begloop
        "movl -3(%%esi), %%eax\n\t"
        "subl $4, %%esi\n\t"
        "bswapl %%eax\n\t"
        "movl %%eax, (%%edi)\n\t"
        "addl $4, %%edi\n\t"
        "decl %%ecx\n\t"
        "jnz 2b\n\t"		// jnz begloop
        "3:"
    : "+S"(S), "+D"(D), "+c"(c) :
            : "eax", "memory", "cc"
        );
}

#elif defined(_MSC_VER) && !defined(NOASM)		// __GNUC__ && __i386__

//
// Microsoft C Inline Assembler version
//

#elif defined(__GNUC__) && defined(GEKKO)

void clearbufbyte(void *d, int32_t c, int32_t a)
{
    if (a==0) {
        uint8_t *dd = (uint8_t*)d;
        int32_t align = (32 - (int32_t)d) & 31;

        if (align && c >= align) {
            uint32_t izero = 0;
            double fzero = 0;
            c -= align;

            if (align&1) {
                *dd = izero;
                dd += 1;
            }
            if (align&2) {
                *(uint16_t*)dd = izero;
                dd += 2;
            }
            if (align&4) {
                *(uint32_t*)dd = izero;
                dd += 4;
            }
            if (align&8) {
                *(double*)dd = fzero;
                dd += 8;
            }
            if (align&16) {
                *(double*)dd = fzero;
                *(double*)(dd+8) = fzero;
                dd += 16;
            }
        }
        align = c >> 5;
        while (align) {
            __asm__ (
                " dcbz  0, %0\n"
                " addi %0, %0, 32\n"
                : "+r"(dd)
                :
                : "memory"
                );
            align--;
        }
        if ((c &= 31)) {
            while (c--) {
                *dd++ = 0;
            }
        }
        return;
    }
    __asm__ __volatile__(
        " add    %1, %1, %2\n"
        " neg.   %2, %2\n"
        " beq 2f\n"
        "1:\n"
        " stbx   %0, %1, %2\n"
        " addic. %2, %2, 1\n"
        " rotrwi %0, %0, 8\n"
        " bne 1b\n"
        "2:\n"
        : "+r"(a), "+b"(d), "+r"(c)
        :
        : "cc", "xer", "memory"
        );
}

#else

//
// Generic C version
//

void qinterpolatedown16(intptr_t bufptr, int32_t num, int32_t val, int32_t add)
{
    // gee, I wonder who could have provided this...
    int32_t i, *lptr = (int32_t *)bufptr;
    for (i=0; i<num; i++) { lptr[i] = (val>>16); val += add; }
}

void qinterpolatedown16short(intptr_t bufptr, int32_t num, int32_t val, int32_t add)
{
    // ...maybe the same person who provided this too?
    int32_t i; int16_t *sptr = (int16_t *)bufptr;
    for (i=0; i<num; i++) { sptr[i] = (int16_t)(val>>16); val += add; }
}

void clearbuf(void *d, int32_t c, int32_t a)
{
    int32_t *p = (int32_t *)d;

    while ((c--) > 0) *(p++) = a;
}

void copybuf(const void *s, void *d, int32_t c)
{
    const int32_t *p = (const int32_t *)s;
    int32_t *q = (int32_t *)d;

    while ((c--) > 0) *(q++) = *(p++);
}

void swapbuf4(void *a, void *b, int32_t c)
{
    int32_t *p = (int32_t *)a, *q = (int32_t *)b;
    int32_t x, y;
    while ((c--) > 0)
    {
        x = *q;
        y = *p;
        *(q++) = y;
        *(p++) = x;
    }
}

void clearbufbyte(void *D, int32_t c, int32_t a)
{
    // Cringe City
    int32_t const m[4] = { 0xffl, 0xff00l, 0xff0000l, (int32_t)0xff000000l };
    int32_t z = 0;
    char *p = (char *)D;

    while ((c--) > 0)
    {
        *(p++) = (uint8_t)((a & m[z])>>(z<<3));
        z=(z+1)&3;
    }
}

void copybufbyte(const void *S, void *D, int32_t c)
{
    const char *p = (const char *)S;
    char *q = (char *)D;

    while ((c--) > 0) *(q++) = *(p++);
}


// copybufreverse() is a special case: use the assembly version for GCC on x86
// *and* x86_64, and the C version otherwise.
// XXX: we don't honor NOASM in the x86_64 case.

#if defined(__GNUC__) && defined(__x86_64__)
// NOTE: Almost CODEDUP from x86 GCC assembly version, except that
// - %%esi -> %%rsi
// - %%edi -> %%rdi
// - (dec,inc,sub,add)l suffix removed where necessary
void copybufreverse(const void *S, void *D, int32_t c)
{
    __asm__ __volatile__(
        "shrl $1, %%ecx\n\t"
        "jnc 0f\n\t"		// jnc skipit1
        "movb (%%rsi), %%al\n\t"
        "dec %%rsi\n\t"
        "movb %%al, (%%rdi)\n\t"
        "inc %%rdi\n\t"
        "0:\n\t"		// skipit1:
        "shrl $1, %%ecx\n\t"
        "jnc 1f\n\t"		// jnc skipit2
        "movw -1(%%rsi), %%ax\n\t"
        "sub $2, %%rsi\n\t"
        "rorw $8, %%ax\n\t"
        "movw %%ax, (%%rdi)\n\t"
        "add $2, %%rdi\n\t"
        "1:\n\t"		// skipit2
        "testl %%ecx, %%ecx\n\t"
        "jz 3f\n\t"		// jz endloop
        "2:\n\t"		// begloop
        "movl -3(%%rsi), %%eax\n\t"
        "sub $4, %%rsi\n\t"
        "bswapl %%eax\n\t"
        "movl %%eax, (%%rdi)\n\t"
        "add $4, %%rdi\n\t"
        "decl %%ecx\n\t"
        "jnz 2b\n\t"		// jnz begloop
        "3:"
    : "+S"(S), "+D"(D), "+c"(c) :
            : "eax", "memory", "cc"
        );
}
#else
void copybufreverse(const void *S, void *D, int32_t c)
{
    const char *p = (const char *)S;
    char *q = (char *)D;

    while ((c--) > 0) *(q++) = *(p--);
}
#endif


#endif