/*

 * Copyright 2003 Sun Microsystems, Inc.  All Rights Reserved.

 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.

 *

 * This code is free software; you can redistribute it and/or modify it

 * under the terms of the GNU General Public License version 2 only, as

 * published by the Free Software Foundation.  Sun designates this

 * particular file as subject to the "Classpath" exception as provided

 * by Sun in the LICENSE file that accompanied this code.

 *

 * This code is distributed in the hope that it will be useful, but WITHOUT

 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or

 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License

 * version 2 for more details (a copy is included in the LICENSE file that

 * accompanied this code).

 *

 * You should have received a copy of the GNU General Public License version

 * 2 along with this work; if not, write to the Free Software Foundation,

 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.

 *

 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,

 * CA 95054 USA or visit www.sun.com if you need additional information or

 * have any questions.

 */

/*

 * FUNCTION

 *   Internal functions for mlib_ImageConv* on U8/S16/U16 types and

 *   MLIB_EDGE_DST_NO_WRITE mask

 */

#include "mlib_image.h"

#include "mlib_ImageConv.h"

#include "mlib_c_ImageConv.h"

/*

  This define switches between functions of different data types

*/

#define IMG_TYPE 1

/***************************************************************/

#if IMG_TYPE == 1

#define DTYPE             mlib_u8

#define CONV_FUNC(KERN)   mlib_c_conv##KERN##nw_u8

#define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_u8

#define DSCALE            (1 << 24)

#define FROM_S32(x)       (((x) >> 24) ^ 128)

#define S64TOS32(x)       (x)

#define SAT_OFF           -(1u << 31)

#elif IMG_TYPE == 2

#define DTYPE             mlib_s16

#define CONV_FUNC(KERN)   mlib_conv##KERN##nw_s16

#define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_s16

#define DSCALE            65536.0

#define FROM_S32(x)       ((x) >> 16)

#define S64TOS32(x)       ((x) & 0xffffffff)

#define SAT_OFF

#elif IMG_TYPE == 3

#define DTYPE             mlib_u16

#define CONV_FUNC(KERN)   mlib_conv##KERN##nw_u16

#define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_u16

#define DSCALE            65536.0

#define FROM_S32(x)       (((x) >> 16) ^ 0x8000)

#define S64TOS32(x)       (x)

#define SAT_OFF           -(1u << 31)

#endif /* IMG_TYPE == 1 */

/***************************************************************/

#define BUFF_SIZE   1600

#define CACHE_SIZE  (64*1024)

/***************************************************************/

#define FTYPE mlib_d64

#ifndef MLIB_USE_FTOI_CLAMPING

#define CLAMP_S32(x)                                            \

  (((x) <= MLIB_S32_MIN) ? MLIB_S32_MIN : (((x) >= MLIB_S32_MAX) ? MLIB_S32_MAX : (mlib_s32)(x)))

#else

#define CLAMP_S32(x) ((mlib_s32)(x))

#endif /* MLIB_USE_FTOI_CLAMPING */

/***************************************************************/

#define D2I(x) CLAMP_S32((x) SAT_OFF)

/***************************************************************/

#ifdef _LITTLE_ENDIAN

#define STORE2(res0, res1)                                      \

  dp[0    ] = res1;                                             \

  dp[chan1] = res0

#else

#define STORE2(res0, res1)                                      \

  dp[0    ] = res0;                                             \

  dp[chan1] = res1

#endif /* _LITTLE_ENDIAN */

/***************************************************************/

#ifdef _NO_LONGLONG

#define LOAD_BUFF(buff)                                         \

  buff[i    ] = sp[0];                                          \

  buff[i + 1] = sp[chan1]

#else /* _NO_LONGLONG */

#ifdef _LITTLE_ENDIAN

#define LOAD_BUFF(buff)                                         \

  *(mlib_s64*)(buff + i) = (((mlib_s64)sp[chan1]) << 32) | S64TOS32((mlib_s64)sp[0])

#else /* _LITTLE_ENDIAN */

#define LOAD_BUFF(buff)                                         \

  *(mlib_s64*)(buff + i) = (((mlib_s64)sp[0]) << 32) | S64TOS32((mlib_s64)sp[chan1])

#endif /* _LITTLE_ENDIAN */

#endif /* _NO_LONGLONG */

/***************************************************************/

typedef union {

  mlib_d64 d64;

  struct {

    mlib_s32 i0;

    mlib_s32 i1;

  } i32s;

  struct {

    mlib_s32 f0;

    mlib_s32 f1;

  } f32s;

} d64_2x32;

/***************************************************************/

#define BUFF_LINE 256

/***************************************************************/

#define DEF_VARS(type)                                          \

  type     *adr_src, *sl, *sp;                                  \

  type     *adr_dst, *dl, *dp;                                  \

  FTYPE    *pbuff = buff;                                       \

  mlib_s32 wid, hgt, sll, dll;                                  \

  mlib_s32 nchannel, chan1;                                     \

  mlib_s32 i, j, c

/***************************************************************/

#define LOAD_KERNEL3()                                                   \

  FTYPE    scalef = DSCALE;                                              \

  FTYPE    k0, k1, k2, k3, k4, k5, k6, k7, k8;                           \

  FTYPE    p00, p01, p02, p03,                                           \

           p10, p11, p12, p13,                                           \

           p20, p21, p22, p23;                                           \

                                                                         \

  while (scalef_expon > 30) {                                            \

    scalef /= (1 << 30);                                                 \

    scalef_expon -= 30;                                                  \

  }                                                                      \

                                                                         \

  scalef /= (1 << scalef_expon);                                         \

                                                                         \

  /* keep kernel in regs */                                              \

  k0 = scalef * kern[0];  k1 = scalef * kern[1];  k2 = scalef * kern[2]; \

  k3 = scalef * kern[3];  k4 = scalef * kern[4];  k5 = scalef * kern[5]; \

  k6 = scalef * kern[6];  k7 = scalef * kern[7];  k8 = scalef * kern[8]

/***************************************************************/

#define LOAD_KERNEL(SIZE)                                       \

  FTYPE    scalef = DSCALE;                                     \

                                                                \

  while (scalef_expon > 30) {                                   \

    scalef /= (1 << 30);                                        \

    scalef_expon -= 30;                                         \

  }                                                             \

                                                                \

  scalef /= (1 << scalef_expon);                                \

                                                                \

  for (j = 0; j < SIZE; j++) k[j] = scalef * kern[j]

/***************************************************************/

#define GET_SRC_DST_PARAMETERS(type)                            \

  hgt = mlib_ImageGetHeight(src);                               \

  wid = mlib_ImageGetWidth(src);                                \

  nchannel = mlib_ImageGetChannels(src);                        \

  sll = mlib_ImageGetStride(src) / sizeof(type);                \

  dll = mlib_ImageGetStride(dst) / sizeof(type);                \

  adr_src = (type *)mlib_ImageGetData(src);                     \

  adr_dst = (type *)mlib_ImageGetData(dst)

/***************************************************************/

#ifndef __sparc

#if IMG_TYPE == 1

/* Test for the presence of any "1" bit in bits

   8 to 31 of val. If present, then val is either

   negative or >255. If over/underflows of 8 bits

   are uncommon, then this technique can be a win,

   since only a single test, rather than two, is

   necessary to determine if clamping is needed.

   On the other hand, if over/underflows are common,

   it adds an extra test.

*/

#define CLAMP_STORE(dst, val)                                   \

  if (val & 0xffffff00) {                                       \

    if (val < MLIB_U8_MIN)                                      \

      dst = MLIB_U8_MIN;                                        \

    else                                                        \

      dst = MLIB_U8_MAX;                                        \

  } else {                                                      \

    dst = (mlib_u8)val;                                         \

  }

#elif IMG_TYPE == 2

#define CLAMP_STORE(dst, val)                                   \

  if (val >= MLIB_S16_MAX)                                      \

    dst = MLIB_S16_MAX;                                         \

  else if (val <= MLIB_S16_MIN)                                 \

    dst = MLIB_S16_MIN;                                         \

  else                                                          \

    dst = (mlib_s16)val

#elif IMG_TYPE == 3

#define CLAMP_STORE(dst, val)                                   \

  if (val >= MLIB_U16_MAX)                                      \

    dst = MLIB_U16_MAX;                                         \

  else if (val <= MLIB_U16_MIN)                                 \

    dst = MLIB_U16_MIN;                                         \

  else                                                          \

    dst = (mlib_u16)val

#endif /* IMG_TYPE == 1 */

#endif /* __sparc */

/***************************************************************/

#define KSIZE  3

mlib_status CONV_FUNC(3x3)(mlib_image       *dst,

                           const mlib_image *src,

                           const mlib_s32   *kern,

                           mlib_s32         scalef_expon,

                           mlib_s32         cmask)

{

  FTYPE    buff[(KSIZE + 2)*BUFF_LINE], *buff0, *buff1, *buff2, *buff3, *buffT;

  DEF_VARS(DTYPE);

  DTYPE *sl1;

  mlib_s32 chan2;

  mlib_s32 *buffo, *buffi;

  DTYPE *sl2;

#ifndef __sparc

  mlib_s32 d0, d1;

#endif /* __sparc */

  LOAD_KERNEL3();

  GET_SRC_DST_PARAMETERS(DTYPE);

  if (wid > BUFF_LINE) {

    pbuff = mlib_malloc((KSIZE + 2)*sizeof(FTYPE)*wid);

    if (pbuff == NULL) return MLIB_FAILURE;

  }

  buff0 = pbuff;

  buff1 = buff0 + wid;

  buff2 = buff1 + wid;

  buff3 = buff2 + wid;

  buffo = (mlib_s32*)(buff3 + wid);

  buffi = buffo + (wid &~ 1);

  chan1 = nchannel;

  chan2 = chan1 + chan1;

  wid -= (KSIZE - 1);

  hgt -= (KSIZE - 1);

  adr_dst += ((KSIZE - 1)/2)*(dll + chan1);

  for (c = 0; c < nchannel; c++) {

    if (!(cmask & (1 << (nchannel - 1 - c)))) continue;

    sl = adr_src + c;

    dl = adr_dst + c;

    sl1 = sl  + sll;

    sl2 = sl1 + sll;

#ifdef __SUNPRO_C

#pragma pipeloop(0)

#endif /* __SUNPRO_C */

    for (i = 0; i < wid + (KSIZE - 1); i++) {

      buff0[i] = (FTYPE)sl[i*chan1];

      buff1[i] = (FTYPE)sl1[i*chan1];

      buff2[i] = (FTYPE)sl2[i*chan1];

    }

    sl += KSIZE*sll;

    for (j = 0; j < hgt; j++) {

      FTYPE    s0, s1;

      p02 = buff0[0];

      p12 = buff1[0];

      p22 = buff2[0];

      p03 = buff0[1];

      p13 = buff1[1];

      p23 = buff2[1];

      s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;

      s1 = p03 * k0 + p13 * k3 + p23 * k6;

      sp = sl;

      dp = dl;

#ifdef __SUNPRO_C

#pragma pipeloop(0)

#endif /* __SUNPRO_C */

      for (i = 0; i <= (wid - 2); i += 2) {

#ifdef __sparc

#ifdef _NO_LONGLONG

        mlib_s32 o64_1, o64_2;

#else /* _NO_LONGLONG */

        mlib_s64 o64;

#endif /* _NO_LONGLONG */

#endif /* __sparc */

        d64_2x32 dd;

        p02 = buff0[i + 2]; p12 = buff1[i + 2]; p22 = buff2[i + 2];

        p03 = buff0[i + 3]; p13 = buff1[i + 3]; p23 = buff2[i + 3];

        LOAD_BUFF(buffi);

        dd.d64 = *(FTYPE   *)(buffi + i);

        buff3[i    ] = (FTYPE)dd.i32s.i0;

        buff3[i + 1] = (FTYPE)dd.i32s.i1;

#ifndef __sparc

        d0 = D2I(s0 + p02 * k2 + p12 * k5 + p22 * k8);

        d1 = D2I(s1 + p02 * k1 + p03 * k2 + p12 * k4 + p13 * k5 + p22 * k7 + p23 * k8);

        s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;

        s1 = p03 * k0 + p13 * k3 + p23 * k6;

        dp[0    ] = FROM_S32(d0);

        dp[chan1] = FROM_S32(d1);

#else /* __sparc */

        dd.i32s.i0 = D2I(s0 + p02 * k2 + p12 * k5 + p22 * k8);

        dd.i32s.i1 = D2I(s1 + p02 * k1 + p03 * k2 + p12 * k4 + p13 * k5 + p22 * k7 + p23 * k8);

        *(FTYPE   *)(buffo + i) = dd.d64;

        s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;

        s1 = p03 * k0 + p13 * k3 + p23 * k6;

#ifdef _NO_LONGLONG

        o64_1 = buffo[i];

        o64_2 = buffo[i+1];

#if IMG_TYPE != 1

        STORE2(FROM_S32(o64_1), FROM_S32(o64_2));

#else

        STORE2(o64_1 >> 24, o64_2 >> 24);

#endif /* IMG_TYPE != 1 */

#else /* _NO_LONGLONG */

        o64 = *(mlib_s64*)(buffo + i);

#if IMG_TYPE != 1

        STORE2(FROM_S32(o64 >> 32), FROM_S32(o64));

#else

        STORE2(o64 >> 56, o64 >> 24);

#endif /* IMG_TYPE != 1 */

#endif /* _NO_LONGLONG */

#endif /* __sparc */

        sp += chan2;

        dp += chan2;

      }

      for (; i < wid; i++) {

        p00 = buff0[i];     p10 = buff1[i];     p20 = buff2[i];

        p01 = buff0[i + 1]; p11 = buff1[i + 1]; p21 = buff2[i + 1];

        p02 = buff0[i + 2]; p12 = buff1[i + 2]; p22 = buff2[i + 2];

        buffi[i] = (mlib_s32)sp[0];

        buff3[i] = (FTYPE)buffi[i];

#ifndef __sparc

        d0 = D2I(p00 * k0 + p01 * k1 + p02 * k2 + p10 * k3 + p11 * k4 +

                 p12 * k5 + p20 * k6 + p21 * k7 + p22 * k8);

        dp[0] = FROM_S32(d0);

#else  /* __sparc */

        buffo[i] = D2I(p00 * k0 + p01 * k1 + p02 * k2 + p10 * k3 + p11 * k4 +

                       p12 * k5 + p20 * k6 + p21 * k7 + p22 * k8);

#if IMG_TYPE != 1

        dp[0] = FROM_S32(buffo[i]);

#else

        dp[0] = buffo[i] >> 24;

#endif /* IMG_TYPE != 1 */

#endif /* __sparc */

        sp += chan1;

        dp += chan1;

      }

      buffi[wid] = (mlib_s32)sp[0];

      buff3[wid] = (FTYPE)buffi[wid];

      buffi[wid + 1] = (mlib_s32)sp[chan1];

      buff3[wid + 1] = (FTYPE)buffi[wid + 1];

      sl += sll;

      dl += dll;

      buffT = buff0;

      buff0 = buff1;

      buff1 = buff2;

      buff2 = buff3;

      buff3 = buffT;

    }

  }

#ifdef __sparc

#if IMG_TYPE == 1

  {

    mlib_s32 amask = (1 << nchannel) - 1;

    if ((cmask & amask) != amask) {

      mlib_ImageXor80(adr_dst, wid, hgt, dll, nchannel, cmask);

    } else {

      mlib_ImageXor80_aa(adr_dst, wid*nchannel, hgt, dll);

    }

  }

#endif /* IMG_TYPE == 1 */

#endif /* __sparc */

  if (pbuff != buff) mlib_free(pbuff);

  return MLIB_SUCCESS;

}

/***************************************************************/

#ifndef __sparc /* for x86, using integer multiplies is faster */

mlib_status CONV_FUNC_I(3x3)(mlib_image       *dst,

                             const mlib_image *src,

                             const mlib_s32   *kern,

                             mlib_s32         scalef_expon,

                             mlib_s32         cmask)

{

  DTYPE    *adr_src, *sl, *sp0, *sp1, *sp2;

  DTYPE    *adr_dst, *dl, *dp;

  mlib_s32 wid, hgt, sll, dll;

  mlib_s32 nchannel, chan1, chan2;

  mlib_s32 i, j, c;

  mlib_s32 shift1, shift2;

  mlib_s32 k0, k1, k2, k3, k4, k5, k6, k7, k8;

  mlib_s32 p02, p03,

           p12, p13,

           p22, p23;

#if IMG_TYPE != 1

  shift1 = 16;

#else

  shift1 = 8;

#endif /* IMG_TYPE != 1 */

  shift2 = scalef_expon - shift1;

  /* keep kernel in regs */

  k0 = kern[0] >> shift1;  k1 = kern[1] >> shift1;  k2 = kern[2] >> shift1;

  k3 = kern[3] >> shift1;  k4 = kern[4] >> shift1;  k5 = kern[5] >> shift1;

  k6 = kern[6] >> shift1;  k7 = kern[7] >> shift1;  k8 = kern[8] >> shift1;

  GET_SRC_DST_PARAMETERS(DTYPE);

  chan1 = nchannel;

  chan2 = chan1 + chan1;

  wid -= (KSIZE - 1);

  hgt -= (KSIZE - 1);

  adr_dst += ((KSIZE - 1)/2)*(dll + chan1);

  for (c = 0; c < chan1; c++) {

    if (!(cmask & (1 << (chan1 - 1 - c)))) continue;

    sl = adr_src + c;

    dl = adr_dst + c;

    for (j = 0; j < hgt; j++) {

      mlib_s32 s0, s1;

      mlib_s32 pix0, pix1;

      dp  = dl;

      sp0 = sl;

      sp1 = sp0 + sll;

      sp2 = sp1 + sll;

      p02 = sp0[0];

      p12 = sp1[0];

      p22 = sp2[0];

      p03 = sp0[chan1];

      p13 = sp1[chan1];

      p23 = sp2[chan1];

      s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;

      s1 = p03 * k0 + p13 * k3 + p23 * k6;

      sp0 += chan2;

      sp1 += chan2;