0N/A/*
2362N/A * Copyright (c) 2000, 2003, Oracle and/or its affiliates. All rights reserved.
0N/A * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
0N/A *
0N/A * This code is free software; you can redistribute it and/or modify it
0N/A * under the terms of the GNU General Public License version 2 only, as
2362N/A * published by the Free Software Foundation. Oracle designates this
0N/A * particular file as subject to the "Classpath" exception as provided
2362N/A * by Oracle in the LICENSE file that accompanied this code.
0N/A *
0N/A * This code is distributed in the hope that it will be useful, but WITHOUT
0N/A * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
0N/A * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
0N/A * version 2 for more details (a copy is included in the LICENSE file that
0N/A * accompanied this code).
0N/A *
0N/A * You should have received a copy of the GNU General Public License version
0N/A * 2 along with this work; if not, write to the Free Software Foundation,
0N/A * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
0N/A *
2362N/A * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
2362N/A * or visit www.oracle.com if you need additional information or have any
2362N/A * questions.
0N/A */
0N/A
0N/A
0N/A/*
0N/A * FUNCTION
0N/A * Internal functions for mlib_ImageConv* on U8/S16/U16 types and
0N/A * MLIB_EDGE_DST_NO_WRITE mask
0N/A */
0N/A
0N/A#include "mlib_image.h"
0N/A#include "mlib_c_ImageConv.h"
0N/A
0N/A/*
0N/A This define switches between functions of different data types
0N/A*/
0N/A#define IMG_TYPE 2
0N/A
0N/A/***************************************************************/
0N/A#if IMG_TYPE == 1
0N/A
0N/A#define DTYPE mlib_u8
0N/A#define CONV_FUNC(KERN) mlib_c_conv##KERN##nw_u8
0N/A#define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_u8
0N/A#define DSCALE (1 << 24)
0N/A#define FROM_S32(x) (((x) >> 24) ^ 128)
0N/A#define S64TOS32(x) (x)
0N/A#define SAT_OFF -(1u << 31)
0N/A
0N/A#elif IMG_TYPE == 2
0N/A
0N/A#define DTYPE mlib_s16
0N/A#define CONV_FUNC(KERN) mlib_conv##KERN##nw_s16
0N/A#define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_s16
0N/A#define DSCALE 65536.0
0N/A#define FROM_S32(x) ((x) >> 16)
0N/A#define S64TOS32(x) ((x) & 0xffffffff)
0N/A#define SAT_OFF
0N/A
0N/A#elif IMG_TYPE == 3
0N/A
0N/A#define DTYPE mlib_u16
0N/A#define CONV_FUNC(KERN) mlib_conv##KERN##nw_u16
0N/A#define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_u16
0N/A#define DSCALE 65536.0
0N/A#define FROM_S32(x) (((x) >> 16) ^ 0x8000)
0N/A#define S64TOS32(x) (x)
0N/A#define SAT_OFF -(1u << 31)
0N/A
0N/A#endif /* IMG_TYPE == 1 */
0N/A
0N/A/***************************************************************/
0N/A#define BUFF_SIZE 1600
0N/A
0N/A#define CACHE_SIZE (64*1024)
0N/A
0N/A/***************************************************************/
0N/A#define FTYPE mlib_d64
0N/A
0N/A#ifndef MLIB_USE_FTOI_CLAMPING
0N/A
0N/A#define CLAMP_S32(x) \
0N/A (((x) <= MLIB_S32_MIN) ? MLIB_S32_MIN : (((x) >= MLIB_S32_MAX) ? MLIB_S32_MAX : (mlib_s32)(x)))
0N/A
0N/A#else
0N/A
0N/A#define CLAMP_S32(x) ((mlib_s32)(x))
0N/A
0N/A#endif /* MLIB_USE_FTOI_CLAMPING */
0N/A
0N/A/***************************************************************/
0N/A#define D2I(x) CLAMP_S32((x) SAT_OFF)
0N/A
0N/A/***************************************************************/
0N/A#ifdef _LITTLE_ENDIAN
0N/A
0N/A#define STORE2(res0, res1) \
0N/A dp[0 ] = res1; \
0N/A dp[chan1] = res0
0N/A
0N/A#else
0N/A
0N/A#define STORE2(res0, res1) \
0N/A dp[0 ] = res0; \
0N/A dp[chan1] = res1
0N/A
0N/A#endif /* _LITTLE_ENDIAN */
0N/A
0N/A/***************************************************************/
0N/A#ifdef _NO_LONGLONG
0N/A
0N/A#define LOAD_BUFF(buff) \
0N/A buff[i ] = sp[0]; \
0N/A buff[i + 1] = sp[chan1]
0N/A
0N/A#else /* _NO_LONGLONG */
0N/A
0N/A#ifdef _LITTLE_ENDIAN
0N/A
0N/A#define LOAD_BUFF(buff) \
0N/A *(mlib_s64*)(buff + i) = (((mlib_s64)sp[chan1]) << 32) | S64TOS32((mlib_s64)sp[0])
0N/A
0N/A#else /* _LITTLE_ENDIAN */
0N/A
0N/A#define LOAD_BUFF(buff) \
0N/A *(mlib_s64*)(buff + i) = (((mlib_s64)sp[0]) << 32) | S64TOS32((mlib_s64)sp[chan1])
0N/A
0N/A#endif /* _LITTLE_ENDIAN */
0N/A#endif /* _NO_LONGLONG */
0N/A
0N/A/***************************************************************/
0N/Atypedef union {
0N/A mlib_d64 d64;
0N/A struct {
0N/A mlib_s32 i0;
0N/A mlib_s32 i1;
0N/A } i32s;
0N/A struct {
0N/A mlib_s32 f0;
0N/A mlib_s32 f1;
0N/A } f32s;
0N/A} d64_2x32;
0N/A
0N/A/***************************************************************/
0N/A#define BUFF_LINE 256
0N/A
0N/A/***************************************************************/
0N/A#define DEF_VARS(type) \
3813N/A type *adr_src, *sl, *sp = NULL; \
3813N/A type *adr_dst, *dl, *dp = NULL; \
0N/A FTYPE *pbuff = buff; \
0N/A mlib_s32 wid, hgt, sll, dll; \
0N/A mlib_s32 nchannel, chan1; \
0N/A mlib_s32 i, j, c
0N/A
0N/A/***************************************************************/
0N/A#define LOAD_KERNEL3() \
0N/A FTYPE scalef = DSCALE; \
0N/A FTYPE k0, k1, k2, k3, k4, k5, k6, k7, k8; \
0N/A FTYPE p00, p01, p02, p03, \
0N/A p10, p11, p12, p13, \
0N/A p20, p21, p22, p23; \
0N/A \
0N/A while (scalef_expon > 30) { \
0N/A scalef /= (1 << 30); \
0N/A scalef_expon -= 30; \
0N/A } \
0N/A \
0N/A scalef /= (1 << scalef_expon); \
0N/A \
0N/A /* keep kernel in regs */ \
0N/A k0 = scalef * kern[0]; k1 = scalef * kern[1]; k2 = scalef * kern[2]; \
0N/A k3 = scalef * kern[3]; k4 = scalef * kern[4]; k5 = scalef * kern[5]; \
0N/A k6 = scalef * kern[6]; k7 = scalef * kern[7]; k8 = scalef * kern[8]
0N/A
0N/A/***************************************************************/
0N/A#define LOAD_KERNEL(SIZE) \
0N/A FTYPE scalef = DSCALE; \
0N/A \
0N/A while (scalef_expon > 30) { \
0N/A scalef /= (1 << 30); \
0N/A scalef_expon -= 30; \
0N/A } \
0N/A \
0N/A scalef /= (1 << scalef_expon); \
0N/A \
0N/A for (j = 0; j < SIZE; j++) k[j] = scalef * kern[j]
0N/A
0N/A/***************************************************************/
0N/A#define GET_SRC_DST_PARAMETERS(type) \
0N/A hgt = mlib_ImageGetHeight(src); \
0N/A wid = mlib_ImageGetWidth(src); \
0N/A nchannel = mlib_ImageGetChannels(src); \
0N/A sll = mlib_ImageGetStride(src) / sizeof(type); \
0N/A dll = mlib_ImageGetStride(dst) / sizeof(type); \
0N/A adr_src = (type *)mlib_ImageGetData(src); \
0N/A adr_dst = (type *)mlib_ImageGetData(dst)
0N/A
0N/A/***************************************************************/
0N/A#ifndef __sparc
0N/A
0N/A#if IMG_TYPE == 1
0N/A
0N/A/* Test for the presence of any "1" bit in bits
0N/A 8 to 31 of val. If present, then val is either
0N/A negative or >255. If over/underflows of 8 bits
0N/A are uncommon, then this technique can be a win,
0N/A since only a single test, rather than two, is
0N/A necessary to determine if clamping is needed.
0N/A On the other hand, if over/underflows are common,
0N/A it adds an extra test.
0N/A*/
0N/A#define CLAMP_STORE(dst, val) \
0N/A if (val & 0xffffff00) { \
0N/A if (val < MLIB_U8_MIN) \
0N/A dst = MLIB_U8_MIN; \
0N/A else \
0N/A dst = MLIB_U8_MAX; \
0N/A } else { \
0N/A dst = (mlib_u8)val; \
0N/A }
0N/A
0N/A#elif IMG_TYPE == 2
0N/A
0N/A#define CLAMP_STORE(dst, val) \
0N/A if (val >= MLIB_S16_MAX) \
0N/A dst = MLIB_S16_MAX; \
0N/A else if (val <= MLIB_S16_MIN) \
0N/A dst = MLIB_S16_MIN; \
0N/A else \
0N/A dst = (mlib_s16)val
0N/A
0N/A#elif IMG_TYPE == 3
0N/A
0N/A#define CLAMP_STORE(dst, val) \
0N/A if (val >= MLIB_U16_MAX) \
0N/A dst = MLIB_U16_MAX; \
0N/A else if (val <= MLIB_U16_MIN) \
0N/A dst = MLIB_U16_MIN; \
0N/A else \
0N/A dst = (mlib_u16)val
0N/A
0N/A#endif /* IMG_TYPE == 1 */
0N/A#endif /* __sparc */
0N/A
0N/A/***************************************************************/
0N/A#define KSIZE 3
0N/A
0N/Amlib_status CONV_FUNC(3x3)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kern,
0N/A mlib_s32 scalef_expon,
0N/A mlib_s32 cmask)
0N/A{
0N/A FTYPE buff[(KSIZE + 2)*BUFF_LINE], *buff0, *buff1, *buff2, *buff3, *buffT;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sl1;
0N/A mlib_s32 chan2;
0N/A mlib_s32 *buffo, *buffi;
0N/A DTYPE *sl2;
0N/A#ifndef __sparc
0N/A mlib_s32 d0, d1;
0N/A#endif /* __sparc */
0N/A LOAD_KERNEL3();
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A if (wid > BUFF_LINE) {
0N/A pbuff = mlib_malloc((KSIZE + 2)*sizeof(FTYPE)*wid);
0N/A
0N/A if (pbuff == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A buff0 = pbuff;
0N/A buff1 = buff0 + wid;
0N/A buff2 = buff1 + wid;
0N/A buff3 = buff2 + wid;
0N/A buffo = (mlib_s32*)(buff3 + wid);
0N/A buffi = buffo + (wid &~ 1);
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A adr_dst += ((KSIZE - 1)/2)*(dll + chan1);
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A sl1 = sl + sll;
0N/A sl2 = sl1 + sll;
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid + (KSIZE - 1); i++) {
0N/A buff0[i] = (FTYPE)sl[i*chan1];
0N/A buff1[i] = (FTYPE)sl1[i*chan1];
0N/A buff2[i] = (FTYPE)sl2[i*chan1];
0N/A }
0N/A
0N/A sl += KSIZE*sll;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A FTYPE s0, s1;
0N/A
0N/A p02 = buff0[0];
0N/A p12 = buff1[0];
0N/A p22 = buff2[0];
0N/A
0N/A p03 = buff0[1];
0N/A p13 = buff1[1];
0N/A p23 = buff2[1];
0N/A
0N/A s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;
0N/A s1 = p03 * k0 + p13 * k3 + p23 * k6;
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A#ifdef __sparc
0N/A#ifdef _NO_LONGLONG
0N/A mlib_s32 o64_1, o64_2;
0N/A#else /* _NO_LONGLONG */
0N/A mlib_s64 o64;
0N/A#endif /* _NO_LONGLONG */
0N/A#endif /* __sparc */
0N/A d64_2x32 dd;
0N/A
0N/A p02 = buff0[i + 2]; p12 = buff1[i + 2]; p22 = buff2[i + 2];
0N/A p03 = buff0[i + 3]; p13 = buff1[i + 3]; p23 = buff2[i + 3];
0N/A
0N/A LOAD_BUFF(buffi);
0N/A
0N/A dd.d64 = *(FTYPE *)(buffi + i);
0N/A buff3[i ] = (FTYPE)dd.i32s.i0;
0N/A buff3[i + 1] = (FTYPE)dd.i32s.i1;
0N/A
0N/A#ifndef __sparc
0N/A d0 = D2I(s0 + p02 * k2 + p12 * k5 + p22 * k8);
0N/A d1 = D2I(s1 + p02 * k1 + p03 * k2 + p12 * k4 + p13 * k5 + p22 * k7 + p23 * k8);
0N/A
0N/A s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;
0N/A s1 = p03 * k0 + p13 * k3 + p23 * k6;
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A#else /* __sparc */
0N/A
0N/A dd.i32s.i0 = D2I(s0 + p02 * k2 + p12 * k5 + p22 * k8);
0N/A dd.i32s.i1 = D2I(s1 + p02 * k1 + p03 * k2 + p12 * k4 + p13 * k5 + p22 * k7 + p23 * k8);
0N/A *(FTYPE *)(buffo + i) = dd.d64;
0N/A
0N/A s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;
0N/A s1 = p03 * k0 + p13 * k3 + p23 * k6;
0N/A
0N/A#ifdef _NO_LONGLONG
0N/A
0N/A o64_1 = buffo[i];
0N/A o64_2 = buffo[i+1];
0N/A#if IMG_TYPE != 1
0N/A STORE2(FROM_S32(o64_1), FROM_S32(o64_2));
0N/A#else
0N/A STORE2(o64_1 >> 24, o64_2 >> 24);
0N/A#endif /* IMG_TYPE != 1 */
0N/A
0N/A#else /* _NO_LONGLONG */
0N/A
0N/A o64 = *(mlib_s64*)(buffo + i);
0N/A#if IMG_TYPE != 1
0N/A STORE2(FROM_S32(o64 >> 32), FROM_S32(o64));
0N/A#else
0N/A STORE2(o64 >> 56, o64 >> 24);
0N/A#endif /* IMG_TYPE != 1 */
0N/A#endif /* _NO_LONGLONG */
0N/A#endif /* __sparc */
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A for (; i < wid; i++) {
0N/A p00 = buff0[i]; p10 = buff1[i]; p20 = buff2[i];
0N/A p01 = buff0[i + 1]; p11 = buff1[i + 1]; p21 = buff2[i + 1];
0N/A p02 = buff0[i + 2]; p12 = buff1[i + 2]; p22 = buff2[i + 2];
0N/A
0N/A buffi[i] = (mlib_s32)sp[0];
0N/A buff3[i] = (FTYPE)buffi[i];
0N/A
0N/A#ifndef __sparc
0N/A
0N/A d0 = D2I(p00 * k0 + p01 * k1 + p02 * k2 + p10 * k3 + p11 * k4 +
0N/A p12 * k5 + p20 * k6 + p21 * k7 + p22 * k8);
0N/A
0N/A dp[0] = FROM_S32(d0);
0N/A
0N/A#else /* __sparc */
0N/A
0N/A buffo[i] = D2I(p00 * k0 + p01 * k1 + p02 * k2 + p10 * k3 + p11 * k4 +
0N/A p12 * k5 + p20 * k6 + p21 * k7 + p22 * k8);
0N/A#if IMG_TYPE != 1
0N/A dp[0] = FROM_S32(buffo[i]);
0N/A#else
0N/A dp[0] = buffo[i] >> 24;
0N/A#endif /* IMG_TYPE != 1 */
0N/A#endif /* __sparc */
0N/A
0N/A sp += chan1;
0N/A dp += chan1;
0N/A }
0N/A
0N/A buffi[wid] = (mlib_s32)sp[0];
0N/A buff3[wid] = (FTYPE)buffi[wid];
0N/A buffi[wid + 1] = (mlib_s32)sp[chan1];
0N/A buff3[wid + 1] = (FTYPE)buffi[wid + 1];
0N/A
0N/A sl += sll;
0N/A dl += dll;
0N/A
0N/A buffT = buff0;
0N/A buff0 = buff1;
0N/A buff1 = buff2;
0N/A buff2 = buff3;
0N/A buff3 = buffT;
0N/A }
0N/A }
0N/A
0N/A#ifdef __sparc
0N/A#if IMG_TYPE == 1
0N/A {
0N/A mlib_s32 amask = (1 << nchannel) - 1;
0N/A
0N/A if ((cmask & amask) != amask) {
0N/A mlib_ImageXor80(adr_dst, wid, hgt, dll, nchannel, cmask);
0N/A } else {
0N/A mlib_ImageXor80_aa(adr_dst, wid*nchannel, hgt, dll);
0N/A }
0N/A }
0N/A
0N/A#endif /* IMG_TYPE == 1 */
0N/A#endif /* __sparc */
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/
0N/A#ifndef __sparc /* for x86, using integer multiplies is faster */
0N/A
0N/Amlib_status CONV_FUNC_I(3x3)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kern,
0N/A mlib_s32 scalef_expon,
0N/A mlib_s32 cmask)
0N/A{
0N/A DTYPE *adr_src, *sl, *sp0, *sp1, *sp2;
0N/A DTYPE *adr_dst, *dl, *dp;
0N/A mlib_s32 wid, hgt, sll, dll;
0N/A mlib_s32 nchannel, chan1, chan2;
0N/A mlib_s32 i, j, c;
0N/A mlib_s32 shift1, shift2;
0N/A mlib_s32 k0, k1, k2, k3, k4, k5, k6, k7, k8;
0N/A mlib_s32 p02, p03,
0N/A p12, p13,
0N/A p22, p23;
0N/A
0N/A#if IMG_TYPE != 1
0N/A shift1 = 16;
0N/A#else
0N/A shift1 = 8;
0N/A#endif /* IMG_TYPE != 1 */
0N/A
0N/A shift2 = scalef_expon - shift1;
0N/A
0N/A /* keep kernel in regs */
0N/A k0 = kern[0] >> shift1; k1 = kern[1] >> shift1; k2 = kern[2] >> shift1;
0N/A k3 = kern[3] >> shift1; k4 = kern[4] >> shift1; k5 = kern[5] >> shift1;
0N/A k6 = kern[6] >> shift1; k7 = kern[7] >> shift1; k8 = kern[8] >> shift1;
0N/A
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A adr_dst += ((KSIZE - 1)/2)*(dll + chan1);
0N/A
0N/A for (c = 0; c < chan1; c++) {
0N/A if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A mlib_s32 s0, s1;
0N/A mlib_s32 pix0, pix1;
0N/A
0N/A dp = dl;
0N/A sp0 = sl;
0N/A sp1 = sp0 + sll;
0N/A sp2 = sp1 + sll;
0N/A
0N/A p02 = sp0[0];
0N/A p12 = sp1[0];
0N/A p22 = sp2[0];
0N/A
0N/A p03 = sp0[chan1];
0N/A p13 = sp1[chan1];
0N/A p23 = sp2[chan1];
0N/A
0N/A s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;
0N/A s1 = p03 * k0 + p13 * k3 + p23 * k6;
0N/A
0N/A sp0 += chan2;
0N/A sp1 += chan2;
0N/A sp2 += chan2;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p02 = sp0[0]; p12 = sp1[0]; p22 = sp2[0];
0N/A p03 = sp0[chan1]; p13 = sp1[chan1]; p23 = sp2[chan1];
0N/A
0N/A pix0 = (s0 + p02 * k2 + p12 * k5 + p22 * k8) >> shift2;
0N/A pix1 = (s1 + p02 * k1 + p03 * k2 + p12 * k4 +
0N/A p13 * k5 + p22 * k7 + p23 * k8) >> shift2;
0N/A
0N/A CLAMP_STORE(dp[0], pix0);
0N/A CLAMP_STORE(dp[chan1], pix1);
0N/A
0N/A s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7;
0N/A s1 = p03 * k0 + p13 * k3 + p23 * k6;
0N/A
0N/A sp0 += chan2;
0N/A sp1 += chan2;
0N/A sp2 += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A if (wid & 1) {
0N/A p02 = sp0[0]; p12 = sp1[0]; p22 = sp2[0];
0N/A pix0 = (s0 + p02 * k2 + p12 * k5 + p22 * k8) >> shift2;
0N/A CLAMP_STORE(dp[0], pix0);
0N/A }
0N/A
0N/A sl += sll;
0N/A dl += dll;
0N/A }
0N/A }
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A#endif /* __sparc ( for x86, using integer multiplies is faster ) */
0N/A
0N/A/***************************************************************/
0N/A#undef KSIZE
0N/A#define KSIZE 4
0N/A
0N/Amlib_status CONV_FUNC(4x4)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kern,
0N/A mlib_s32 scalef_expon,
0N/A mlib_s32 cmask)
0N/A{
0N/A FTYPE buff[(KSIZE + 3)*BUFF_LINE];
0N/A FTYPE *buff0, *buff1, *buff2, *buff3, *buff4, *buffd, *buffT;
0N/A FTYPE k[KSIZE*KSIZE];
0N/A mlib_s32 d0, d1;
0N/A FTYPE k0, k1, k2, k3, k4, k5, k6, k7;
0N/A FTYPE p00, p01, p02, p03, p04,
0N/A p10, p11, p12, p13, p14,
0N/A p20, p21, p22, p23,
0N/A p30, p31, p32, p33;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sl1;
0N/A mlib_s32 chan2;
0N/A mlib_s32 *buffo, *buffi;
0N/A DTYPE *sl2, *sl3;
0N/A LOAD_KERNEL(KSIZE*KSIZE);
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A if (wid > BUFF_LINE) {
0N/A pbuff = mlib_malloc((KSIZE + 3)*sizeof(FTYPE)*wid);
0N/A
0N/A if (pbuff == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A buff0 = pbuff;
0N/A buff1 = buff0 + wid;
0N/A buff2 = buff1 + wid;
0N/A buff3 = buff2 + wid;
0N/A buff4 = buff3 + wid;
0N/A buffd = buff4 + wid;
0N/A buffo = (mlib_s32*)(buffd + wid);
0N/A buffi = buffo + (wid &~ 1);
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A adr_dst += ((KSIZE - 1)/2)*(dll + chan1);
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A sl1 = sl + sll;
0N/A sl2 = sl1 + sll;
0N/A sl3 = sl2 + sll;
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid + (KSIZE - 1); i++) {
0N/A buff0[i] = (FTYPE)sl[i*chan1];
0N/A buff1[i] = (FTYPE)sl1[i*chan1];
0N/A buff2[i] = (FTYPE)sl2[i*chan1];
0N/A buff3[i] = (FTYPE)sl3[i*chan1];
0N/A }
0N/A
0N/A sl += KSIZE*sll;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A d64_2x32 dd;
0N/A
0N/A /*
0N/A * First loop on two first lines of kernel
0N/A */
0N/A k0 = k[0]; k1 = k[1]; k2 = k[2]; k3 = k[3];
0N/A k4 = k[4]; k5 = k[5]; k6 = k[6]; k7 = k[7];
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p02 = buff0[0];
0N/A p12 = buff1[0];
0N/A p03 = buff0[1];
0N/A p13 = buff1[1];
0N/A p04 = buff0[2];
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = buff1[i + 2];
0N/A p03 = buff0[i + 3]; p13 = buff1[i + 3];
0N/A p04 = buff0[i + 4]; p14 = buff1[i + 4];
0N/A
0N/A LOAD_BUFF(buffi);
0N/A
0N/A dd.d64 = *(FTYPE *)(buffi + i);
0N/A buff4[i ] = (FTYPE)dd.i32s.i0;
0N/A buff4[i + 1] = (FTYPE)dd.i32s.i1;
0N/A
0N/A buffd[i ] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 +
0N/A p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7);
0N/A buffd[i + 1] = (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 +
0N/A p11 * k4 + p12 * k5 + p13 * k6 + p14 * k7);
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A /*
0N/A * Second loop on two last lines of kernel
0N/A */
0N/A k0 = k[ 8]; k1 = k[ 9]; k2 = k[10]; k3 = k[11];
0N/A k4 = k[12]; k5 = k[13]; k6 = k[14]; k7 = k[15];
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p02 = buff2[0];
0N/A p12 = buff3[0];
0N/A p03 = buff2[1];
0N/A p13 = buff3[1];
0N/A p04 = buff2[2];
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = buff3[i + 2];
0N/A p03 = buff2[i + 3]; p13 = buff3[i + 3];
0N/A p04 = buff2[i + 4]; p14 = buff3[i + 4];
0N/A
0N/A d0 = D2I(p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 +
0N/A p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7 + buffd[i]);
0N/A d1 = D2I(p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 +
0N/A p11 * k4 + p12 * k5 + p13 * k6 + p14 * k7 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A /* last pixels */
0N/A for (; i < wid; i++) {
0N/A p00 = buff0[i]; p10 = buff1[i]; p20 = buff2[i]; p30 = buff3[i];
0N/A p01 = buff0[i + 1]; p11 = buff1[i + 1]; p21 = buff2[i + 1]; p31 = buff3[i + 1];
0N/A p02 = buff0[i + 2]; p12 = buff1[i + 2]; p22 = buff2[i + 2]; p32 = buff3[i + 2];
0N/A p03 = buff0[i + 3]; p13 = buff1[i + 3]; p23 = buff2[i + 3]; p33 = buff3[i + 3];
0N/A
0N/A buff4[i] = (FTYPE)sp[0];
0N/A
0N/A buffo[i] = D2I(p00 * k[0] + p01 * k[1] + p02 * k[2] + p03 * k[3] +
0N/A p10 * k[4] + p11 * k[5] + p12 * k[6] + p13 * k[7] +
0N/A p20 * k[ 8] + p21 * k[ 9] + p22 * k[10] + p23 * k[11] +
0N/A p30 * k[12] + p31 * k[13] + p32 * k[14] + p33 * k[15]);
0N/A
0N/A dp[0] = FROM_S32(buffo[i]);
0N/A
0N/A sp += chan1;
0N/A dp += chan1;
0N/A }
0N/A
0N/A buff4[wid ] = (FTYPE)sp[0];
0N/A buff4[wid + 1] = (FTYPE)sp[chan1];
0N/A buff4[wid + 2] = (FTYPE)sp[chan2];
0N/A
0N/A /* next line */
0N/A sl += sll;
0N/A dl += dll;
0N/A
0N/A buffT = buff0;
0N/A buff0 = buff1;
0N/A buff1 = buff2;
0N/A buff2 = buff3;
0N/A buff3 = buff4;
0N/A buff4 = buffT;
0N/A }
0N/A }
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/
0N/A#undef KSIZE
0N/A#define KSIZE 5
0N/A
0N/Amlib_status CONV_FUNC(5x5)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kern,
0N/A mlib_s32 scalef_expon,
0N/A mlib_s32 cmask)
0N/A{
0N/A FTYPE buff[(KSIZE + 3)*BUFF_LINE];
0N/A FTYPE *buff0, *buff1, *buff2, *buff3, *buff4, *buff5, *buffd, *buffT;
0N/A FTYPE k[KSIZE*KSIZE];
0N/A mlib_s32 d0, d1;
0N/A FTYPE k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
0N/A FTYPE p00, p01, p02, p03, p04, p05,
0N/A p10, p11, p12, p13, p14, p15,
0N/A p20, p21, p22, p23, p24,
0N/A p30, p31, p32, p33, p34,
0N/A p40, p41, p42, p43, p44;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sl1;
0N/A mlib_s32 chan2;
0N/A mlib_s32 *buffo, *buffi;
0N/A DTYPE *sl2, *sl3, *sl4;
0N/A LOAD_KERNEL(KSIZE*KSIZE);
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A if (wid > BUFF_LINE) {
0N/A pbuff = mlib_malloc((KSIZE + 3)*sizeof(FTYPE)*wid);
0N/A
0N/A if (pbuff == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A buff0 = pbuff;
0N/A buff1 = buff0 + wid;
0N/A buff2 = buff1 + wid;
0N/A buff3 = buff2 + wid;
0N/A buff4 = buff3 + wid;
0N/A buff5 = buff4 + wid;
0N/A buffd = buff5 + wid;
0N/A buffo = (mlib_s32*)(buffd + wid);
0N/A buffi = buffo + (wid &~ 1);
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A adr_dst += ((KSIZE - 1)/2)*(dll + chan1);
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A sl1 = sl + sll;
0N/A sl2 = sl1 + sll;
0N/A sl3 = sl2 + sll;
0N/A sl4 = sl3 + sll;
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid + (KSIZE - 1); i++) {
0N/A buff0[i] = (FTYPE)sl[i*chan1];
0N/A buff1[i] = (FTYPE)sl1[i*chan1];
0N/A buff2[i] = (FTYPE)sl2[i*chan1];
0N/A buff3[i] = (FTYPE)sl3[i*chan1];
0N/A buff4[i] = (FTYPE)sl4[i*chan1];
0N/A }
0N/A
0N/A sl += KSIZE*sll;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A d64_2x32 dd;
0N/A
0N/A /*
0N/A * First loop
0N/A */
0N/A k0 = k[0]; k1 = k[1]; k2 = k[2]; k3 = k[3]; k4 = k[4];
0N/A k5 = k[5]; k6 = k[6]; k7 = k[7]; k8 = k[8]; k9 = k[9];
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p02 = buff0[0];
0N/A p12 = buff1[0];
0N/A p03 = buff0[1];
0N/A p13 = buff1[1];
0N/A p04 = buff0[2];
0N/A p14 = buff1[2];
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = p14;
0N/A
0N/A LOAD_BUFF(buffi);
0N/A
0N/A p03 = buff0[i + 3]; p13 = buff1[i + 3];
0N/A p04 = buff0[i + 4]; p14 = buff1[i + 4];
0N/A p05 = buff0[i + 5]; p15 = buff1[i + 5];
0N/A
0N/A buffd[i ] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A buffd[i + 1] = (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 +
0N/A p11 * k5 + p12 * k6 + p13 * k7 + p14 * k8 + p15 * k9);
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A /*
0N/A * Second loop
0N/A */
0N/A k0 = k[10]; k1 = k[11]; k2 = k[12]; k3 = k[13]; k4 = k[14];
0N/A k5 = k[15]; k6 = k[16]; k7 = k[17]; k8 = k[18]; k9 = k[19];
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p02 = buff2[0];
0N/A p12 = buff3[0];
0N/A p03 = buff2[1];
0N/A p13 = buff3[1];
0N/A p04 = buff2[2];
0N/A p14 = buff3[2];
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A
0N/A p02 = buff2[i + 2]; p12 = buff3[i + 2];
0N/A p03 = buff2[i + 3]; p13 = buff3[i + 3];
0N/A p04 = buff2[i + 4]; p14 = buff3[i + 4];
0N/A p05 = buff2[i + 5]; p15 = buff3[i + 5];
0N/A
0N/A dd.d64 = *(FTYPE *)(buffi + i);
0N/A buff5[i ] = (FTYPE)dd.i32s.i0;
0N/A buff5[i + 1] = (FTYPE)dd.i32s.i1;
0N/A
0N/A buffd[i ] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A buffd[i + 1] += (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 +
0N/A p11 * k5 + p12 * k6 + p13 * k7 + p14 * k8 + p15 * k9);
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A /*
0N/A * 3 loop
0N/A */
0N/A k0 = k[20]; k1 = k[21]; k2 = k[22]; k3 = k[23]; k4 = k[24];
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p02 = buff4[0];
0N/A p03 = buff4[1];
0N/A p04 = buff4[2];
0N/A p05 = buff4[3];
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p01 = p03; p02 = p04; p03 = p05;
0N/A
0N/A p04 = buff4[i + 4]; p05 = buff4[i + 5];
0N/A
0N/A d0 = D2I(p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 + buffd[i]);
0N/A d1 = D2I(p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A /* last pixels */
0N/A for (; i < wid; i++) {
0N/A p00 = buff0[i]; p10 = buff1[i]; p20 = buff2[i]; p30 = buff3[i];
0N/A p01 = buff0[i + 1]; p11 = buff1[i + 1]; p21 = buff2[i + 1]; p31 = buff3[i + 1];
0N/A p02 = buff0[i + 2]; p12 = buff1[i + 2]; p22 = buff2[i + 2]; p32 = buff3[i + 2];
0N/A p03 = buff0[i + 3]; p13 = buff1[i + 3]; p23 = buff2[i + 3]; p33 = buff3[i + 3];
0N/A p04 = buff0[i + 4]; p14 = buff1[i + 4]; p24 = buff2[i + 4]; p34 = buff3[i + 4];
0N/A
0N/A p40 = buff4[i]; p41 = buff4[i + 1]; p42 = buff4[i + 2];
0N/A p43 = buff4[i + 3]; p44 = buff4[i + 4];
0N/A
0N/A buff5[i] = (FTYPE)sp[0];
0N/A
0N/A buffo[i] = D2I(p00 * k[0] + p01 * k[1] + p02 * k[2] + p03 * k[3] + p04 * k[4] +
0N/A p10 * k[5] + p11 * k[6] + p12 * k[7] + p13 * k[8] + p14 * k[9] +
0N/A p20 * k[10] + p21 * k[11] + p22 * k[12] + p23 * k[13] + p24 * k[14] +
0N/A p30 * k[15] + p31 * k[16] + p32 * k[17] + p33 * k[18] + p34 * k[19] +
0N/A p40 * k[20] + p41 * k[21] + p42 * k[22] + p43 * k[23] + p44 * k[24]);
0N/A
0N/A dp[0] = FROM_S32(buffo[i]);
0N/A
0N/A sp += chan1;
0N/A dp += chan1;
0N/A }
0N/A
0N/A buff5[wid ] = (FTYPE)sp[0];
0N/A buff5[wid + 1] = (FTYPE)sp[chan1];
0N/A buff5[wid + 2] = (FTYPE)sp[chan2];
0N/A buff5[wid + 3] = (FTYPE)sp[chan2 + chan1];
0N/A
0N/A /* next line */
0N/A sl += sll;
0N/A dl += dll;
0N/A
0N/A buffT = buff0;
0N/A buff0 = buff1;
0N/A buff1 = buff2;
0N/A buff2 = buff3;
0N/A buff3 = buff4;
0N/A buff4 = buff5;
0N/A buff5 = buffT;
0N/A }
0N/A }
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/
0N/A#ifndef __sparc /* for x86, using integer multiplies is faster */
0N/A
0N/Amlib_status CONV_FUNC_I(5x5)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kern,
0N/A mlib_s32 scalef_expon,
0N/A mlib_s32 cmask)
0N/A{
0N/A mlib_s32 buff[BUFF_LINE];
0N/A mlib_s32 *buffd;
0N/A mlib_s32 k[KSIZE*KSIZE];
0N/A mlib_s32 shift1, shift2;
0N/A mlib_s32 k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
0N/A mlib_s32 p00, p01, p02, p03, p04, p05,
0N/A p10, p11, p12, p13, p14, p15;
0N/A DTYPE *adr_src, *sl, *sp0, *sp1;
0N/A DTYPE *adr_dst, *dl, *dp;
0N/A mlib_s32 *pbuff = buff;
0N/A mlib_s32 wid, hgt, sll, dll;
0N/A mlib_s32 nchannel, chan1, chan2, chan3, chan4;
0N/A mlib_s32 i, j, c;
0N/A
0N/A#if IMG_TYPE != 1
0N/A shift1 = 16;
0N/A#else
0N/A shift1 = 8;
0N/A#endif /* IMG_TYPE != 1 */
0N/A
0N/A shift2 = scalef_expon - shift1;
0N/A
0N/A for (j = 0; j < KSIZE*KSIZE; j++) k[j] = kern[j] >> shift1;
0N/A
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A if (wid > BUFF_LINE) {
0N/A pbuff = mlib_malloc(sizeof(mlib_s32)*wid);
0N/A
0N/A if (pbuff == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A buffd = pbuff;
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A chan3 = chan2 + chan1;
0N/A chan4 = chan3 + chan1;
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A adr_dst += ((KSIZE - 1)/2)*(dll + chan1);
0N/A
0N/A for (c = 0; c < chan1; c++) {
0N/A if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A mlib_s32 pix0, pix1;
0N/A /*
0N/A * First loop
0N/A */
0N/A sp0 = sl;
0N/A sp1 = sp0 + sll;
0N/A dp = dl;
0N/A
0N/A k0 = k[0]; k1 = k[1]; k2 = k[2]; k3 = k[3]; k4 = k[4];
0N/A k5 = k[5]; k6 = k[6]; k7 = k[7]; k8 = k[8]; k9 = k[9];
0N/A
0N/A p02 = sp0[0]; p12 = sp1[0];
0N/A p03 = sp0[chan1]; p13 = sp1[chan1];
0N/A p04 = sp0[chan2]; p14 = sp1[chan2];
0N/A p05 = sp0[chan3]; p15 = sp1[chan3];
0N/A
0N/A sp0 += chan4;
0N/A sp1 += chan4;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = p14;
0N/A p03 = p05; p13 = p15;
0N/A
0N/A p04 = sp0[0]; p14 = sp1[0];
0N/A p05 = sp0[chan1]; p15 = sp1[chan1];
0N/A
0N/A buffd[i ] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A buffd[i + 1] = (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 +
0N/A p11 * k5 + p12 * k6 + p13 * k7 + p14 * k8 + p15 * k9);
0N/A
0N/A sp0 += chan2;
0N/A sp1 += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A if (wid & 1) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = p14;
0N/A p03 = p05; p13 = p15;
0N/A
0N/A p04 = sp0[0]; p14 = sp1[0];
0N/A
0N/A buffd[i] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A }
0N/A
0N/A /*
0N/A * Second loop
0N/A */
0N/A sp0 = sl + 2*sll;
0N/A sp1 = sp0 + sll;
0N/A dp = dl;
0N/A
0N/A k0 = k[10]; k1 = k[11]; k2 = k[12]; k3 = k[13]; k4 = k[14];
0N/A k5 = k[15]; k6 = k[16]; k7 = k[17]; k8 = k[18]; k9 = k[19];
0N/A
0N/A p02 = sp0[0]; p12 = sp1[0];
0N/A p03 = sp0[chan1]; p13 = sp1[chan1];
0N/A p04 = sp0[chan2]; p14 = sp1[chan2];
0N/A p05 = sp0[chan3]; p15 = sp1[chan3];
0N/A
0N/A sp0 += chan4;
0N/A sp1 += chan4;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = p14;
0N/A p03 = p05; p13 = p15;
0N/A
0N/A p04 = sp0[0]; p14 = sp1[0];
0N/A p05 = sp0[chan1]; p15 = sp1[chan1];
0N/A
0N/A buffd[i ] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A buffd[i + 1] += (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 +
0N/A p11 * k5 + p12 * k6 + p13 * k7 + p14 * k8 + p15 * k9);
0N/A
0N/A sp0 += chan2;
0N/A sp1 += chan2;
0N/A dp += chan2;
0N/A }
0N/A
0N/A if (wid & 1) {
0N/A p00 = p02; p10 = p12;
0N/A p01 = p03; p11 = p13;
0N/A p02 = p04; p12 = p14;
0N/A p03 = p05; p13 = p15;
0N/A
0N/A p04 = sp0[0]; p14 = sp1[0];
0N/A
0N/A buffd[i] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A }
0N/A
0N/A /*
0N/A * 3 loop
0N/A */
0N/A dp = dl;
0N/A sp0 = sl + 4*sll;
0N/A
0N/A k0 = k[20]; k1 = k[21]; k2 = k[22]; k3 = k[23]; k4 = k[24];
0N/A
0N/A p02 = sp0[0];
0N/A p03 = sp0[chan1];
0N/A p04 = sp0[chan2];
0N/A p05 = sp0[chan3];
0N/A
0N/A sp0 += chan2 + chan2;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p00 = p02; p01 = p03; p02 = p04; p03 = p05;
0N/A
0N/A p04 = sp0[0]; p05 = sp0[chan1];
0N/A
0N/A pix0 = (buffd[i ] + p00 * k0 + p01 * k1 + p02 * k2 +
0N/A p03 * k3 + p04 * k4) >> shift2;
0N/A pix1 = (buffd[i + 1] + p01 * k0 + p02 * k1 + p03 * k2 +
0N/A p04 * k3 + p05 * k4) >> shift2;
0N/A
0N/A CLAMP_STORE(dp[0], pix0);
0N/A CLAMP_STORE(dp[chan1], pix1);
0N/A
0N/A dp += chan2;
0N/A sp0 += chan2;
0N/A }
0N/A
0N/A if (wid & 1) {
0N/A p00 = p02; p01 = p03; p02 = p04; p03 = p05;
0N/A
0N/A p04 = sp0[0];
0N/A
0N/A pix0 = (buffd[i ] + p00 * k0 + p01 * k1 + p02 * k2 +
0N/A p03 * k3 + p04 * k4) >> shift2;
0N/A CLAMP_STORE(dp[0], pix0);
0N/A }
0N/A
0N/A /* next line */
0N/A sl += sll;
0N/A dl += dll;
0N/A }
0N/A }
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A#endif /* __sparc ( for x86, using integer multiplies is faster ) */
0N/A
0N/A/***************************************************************/
0N/A#if IMG_TYPE == 1
0N/A
0N/A#undef KSIZE
0N/A#define KSIZE 7
0N/A
0N/Amlib_status CONV_FUNC(7x7)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kern,
0N/A mlib_s32 scalef_expon,
0N/A mlib_s32 cmask)
0N/A{
0N/A FTYPE buff[(KSIZE + 3)*BUFF_LINE], *buffs[2*(KSIZE + 1)], *buffd;
0N/A FTYPE k[KSIZE*KSIZE];
0N/A mlib_s32 l, m, buff_ind;
0N/A mlib_s32 d0, d1;
0N/A FTYPE k0, k1, k2, k3, k4, k5, k6;
0N/A FTYPE p0, p1, p2, p3, p4, p5, p6, p7;
0N/A DTYPE *sl2, *sl3, *sl4, *sl5, *sl6;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sl1;
0N/A mlib_s32 chan2;
0N/A mlib_s32 *buffo, *buffi;
0N/A LOAD_KERNEL(KSIZE*KSIZE);
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A if (wid > BUFF_LINE) {
0N/A pbuff = mlib_malloc((KSIZE + 3)*sizeof(FTYPE)*wid);
0N/A
0N/A if (pbuff == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A for (l = 0; l < KSIZE + 1; l++) buffs[l] = pbuff + l*wid;
0N/A for (l = 0; l < KSIZE + 1; l++) buffs[l + (KSIZE + 1)] = buffs[l];
0N/A buffd = buffs[KSIZE] + wid;
0N/A buffo = (mlib_s32*)(buffd + wid);
0N/A buffi = buffo + (wid &~ 1);
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A adr_dst += ((KSIZE - 1)/2)*(dll + chan1);
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A sl1 = sl + sll;
0N/A sl2 = sl1 + sll;
0N/A sl3 = sl2 + sll;
0N/A sl4 = sl3 + sll;
0N/A sl5 = sl4 + sll;
0N/A sl6 = sl5 + sll;
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid + (KSIZE - 1); i++) {
0N/A buffs[0][i] = (FTYPE)sl[i*chan1];
0N/A buffs[1][i] = (FTYPE)sl1[i*chan1];
0N/A buffs[2][i] = (FTYPE)sl2[i*chan1];
0N/A buffs[3][i] = (FTYPE)sl3[i*chan1];
0N/A buffs[4][i] = (FTYPE)sl4[i*chan1];
0N/A buffs[5][i] = (FTYPE)sl5[i*chan1];
0N/A buffs[6][i] = (FTYPE)sl6[i*chan1];
0N/A }
0N/A
0N/A buff_ind = 0;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid; i++) buffd[i] = 0.0;
0N/A
0N/A sl += KSIZE*sll;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A FTYPE **buffc = buffs + buff_ind;
0N/A FTYPE *buffn = buffc[KSIZE];
0N/A FTYPE *pk = k;
0N/A
0N/A for (l = 0; l < KSIZE; l++) {
0N/A FTYPE *buff = buffc[l];
0N/A d64_2x32 dd;
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
0N/A p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
0N/A
0N/A k0 = *pk++; k1 = *pk++; k2 = *pk++; k3 = *pk++;
0N/A k4 = *pk++; k5 = *pk++; k6 = *pk++;
0N/A
0N/A if (l < (KSIZE - 1)) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
0N/A
0N/A p6 = buff[i + 6]; p7 = buff[i + 7];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
0N/A
0N/A p6 = buff[i + 6]; p7 = buff[i + 7];
0N/A
0N/A LOAD_BUFF(buffi);
0N/A
0N/A dd.d64 = *(FTYPE *)(buffi + i);
0N/A buffn[i ] = (FTYPE)dd.i32s.i0;
0N/A buffn[i + 1] = (FTYPE)dd.i32s.i1;
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A }
0N/A
0N/A /* last pixels */
0N/A for (; i < wid; i++) {
0N/A FTYPE *pk = k, s = 0;
0N/A mlib_s32 d0;
0N/A
0N/A for (l = 0; l < KSIZE; l++) {
0N/A FTYPE *buff = buffc[l] + i;
0N/A
0N/A for (m = 0; m < KSIZE; m++) s += buff[m] * (*pk++);
0N/A }
0N/A
0N/A d0 = D2I(s);
0N/A dp[0] = FROM_S32(d0);
0N/A
0N/A buffn[i] = (FTYPE)sp[0];
0N/A
0N/A sp += chan1;
0N/A dp += chan1;
0N/A }
0N/A
0N/A for (l = 0; l < (KSIZE - 1); l++) buffn[wid + l] = sp[l*chan1];
0N/A
0N/A /* next line */
0N/A sl += sll;
0N/A dl += dll;
0N/A
0N/A buff_ind++;
0N/A
0N/A if (buff_ind >= KSIZE + 1) buff_ind = 0;
0N/A }
0N/A }
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A#endif /* IMG_TYPE == 1 */
0N/A
0N/A/***************************************************************/
0N/A#define MAX_KER 7
0N/A#define MAX_N 15
0N/A
0N/Astatic mlib_status mlib_ImageConv1xN(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_d64 *k,
0N/A mlib_s32 n,
0N/A mlib_s32 dn,
0N/A mlib_s32 cmask)
0N/A{
0N/A FTYPE buff[BUFF_SIZE];
0N/A mlib_s32 off, kh;
0N/A mlib_s32 d0, d1;
0N/A const FTYPE *pk;
0N/A FTYPE k0, k1, k2, k3;
0N/A FTYPE p0, p1, p2, p3, p4;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sl_c, *dl_c, *sl0;
0N/A mlib_s32 l, hsize, max_hsize;
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A hgt -= (n - 1);
0N/A adr_dst += dn*dll;
0N/A
0N/A max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll;
0N/A
0N/A if (!max_hsize) max_hsize = 1;
0N/A
0N/A if (max_hsize > BUFF_SIZE) {
0N/A pbuff = mlib_malloc(sizeof(FTYPE)*max_hsize);
0N/A }
0N/A
0N/A chan1 = nchannel;
0N/A
0N/A sl_c = adr_src;
0N/A dl_c = adr_dst;
0N/A
0N/A for (l = 0; l < hgt; l += hsize) {
0N/A hsize = hgt - l;
0N/A
0N/A if (hsize > max_hsize) hsize = max_hsize;
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
0N/A
0N/A sl = sl_c + c;
0N/A dl = dl_c + c;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (j = 0; j < hsize; j++) pbuff[j] = 0.0;
0N/A
0N/A for (i = 0; i < wid; i++) {
0N/A sl0 = sl;
0N/A
0N/A for (off = 0; off < (n - 4); off += 4) {
0N/A pk = k + off;
0N/A sp = sl0;
0N/A
0N/A k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
0N/A p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
0N/A sp += 3*sll;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (j = 0; j < hsize; j += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A p3 = sp[0];
0N/A p4 = sp[sll];
0N/A
0N/A pbuff[j ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
0N/A pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
0N/A
0N/A sp += 2*sll;
0N/A }
0N/A
0N/A sl0 += 4*sll;
0N/A }
0N/A
0N/A pk = k + off;
0N/A sp = sl0;
0N/A
0N/A k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
0N/A p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
0N/A
0N/A dp = dl;
0N/A kh = n - off;
0N/A
0N/A if (kh == 4) {
0N/A sp += 3*sll;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (j = 0; j <= (hsize - 2); j += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A p3 = sp[0];
0N/A p4 = sp[sll];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[dll] = FROM_S32(d1);
0N/A
0N/A pbuff[j] = 0;
0N/A pbuff[j + 1] = 0;
0N/A
0N/A sp += 2*sll;
0N/A dp += 2*dll;
0N/A }
0N/A
0N/A if (j < hsize) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A p3 = sp[0];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]);
0N/A
0N/A pbuff[j] = 0;
0N/A
0N/A dp[0] = FROM_S32(d0);
0N/A }
0N/A
0N/A } else if (kh == 3) {
0N/A sp += 2*sll;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (j = 0; j <= (hsize - 2); j += 2) {
0N/A p0 = p2; p1 = p3;
0N/A p2 = sp[0];
0N/A p3 = sp[sll];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + pbuff[j]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[dll] = FROM_S32(d1);
0N/A
0N/A pbuff[j] = 0;
0N/A pbuff[j + 1] = 0;
0N/A
0N/A sp += 2*sll;
0N/A dp += 2*dll;
0N/A }
0N/A
0N/A if (j < hsize) {
0N/A p0 = p2; p1 = p3;
0N/A p2 = sp[0];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + pbuff[j]);
0N/A
0N/A pbuff[j] = 0;
0N/A
0N/A dp[0] = FROM_S32(d0);
0N/A }
0N/A
0N/A } else if (kh == 2) {
0N/A sp += sll;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (j = 0; j <= (hsize - 2); j += 2) {
0N/A p0 = p2;
0N/A p1 = sp[0];
0N/A p2 = sp[sll];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + pbuff[j]);
0N/A d1 = D2I(p1*k0 + p2*k1 + pbuff[j + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[dll] = FROM_S32(d1);
0N/A
0N/A pbuff[j] = 0;
0N/A pbuff[j + 1] = 0;
0N/A
0N/A sp += 2*sll;
0N/A dp += 2*dll;
0N/A }
0N/A
0N/A if (j < hsize) {
0N/A p0 = p2;
0N/A p1 = sp[0];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + pbuff[j]);
0N/A
0N/A pbuff[j] = 0;
0N/A
0N/A dp[0] = FROM_S32(d0);
0N/A }
0N/A
0N/A } else /* if (kh == 1) */ {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (j = 0; j < hsize; j++) {
0N/A p0 = sp[0];
0N/A
0N/A d0 = D2I(p0*k0 + pbuff[j]);
0N/A
0N/A dp[0] = FROM_S32(d0);
0N/A
0N/A pbuff[j] = 0;
0N/A
0N/A sp += sll;
0N/A dp += dll;
0N/A }
0N/A }
0N/A
0N/A sl += chan1;
0N/A dl += chan1;
0N/A }
0N/A }
0N/A
0N/A sl_c += max_hsize*sll;
0N/A dl_c += max_hsize*dll;
0N/A }
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/
0N/Amlib_status CONV_FUNC(MxN)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kernel,
0N/A mlib_s32 m,
0N/A mlib_s32 n,
0N/A mlib_s32 dm,
0N/A mlib_s32 dn,
0N/A mlib_s32 scale,
0N/A mlib_s32 cmask)
0N/A{
0N/A FTYPE buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
0N/A FTYPE **buffs = buffs_arr, *buffd;
0N/A FTYPE akernel[256], *k = akernel, fscale = DSCALE;
0N/A mlib_s32 mn, l, off, kw, bsize, buff_ind;
0N/A mlib_s32 d0, d1;
0N/A FTYPE k0, k1, k2, k3, k4, k5, k6;
0N/A FTYPE p0, p1, p2, p3, p4, p5, p6, p7;
0N/A d64_2x32 dd;
0N/A DEF_VARS(DTYPE);
0N/A mlib_s32 chan2;
0N/A mlib_s32 *buffo, *buffi;
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A if (scale > 30) {
0N/A fscale *= 1.0/(1 << 30);
0N/A scale -= 30;
0N/A }
0N/A
0N/A fscale /= (1 << scale);
0N/A
0N/A mn = m*n;
0N/A
0N/A if (mn > 256) {
0N/A k = mlib_malloc(mn*sizeof(mlib_d64));
0N/A
0N/A if (k == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A for (i = 0; i < mn; i++) {
0N/A k[i] = kernel[i]*fscale;
0N/A }
0N/A
0N/A if (m == 1) return mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
0N/A
0N/A bsize = (n + 3)*wid;
0N/A
0N/A if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
0N/A pbuff = mlib_malloc(sizeof(FTYPE)*bsize + sizeof(FTYPE *)*2*(n + 1));
0N/A
0N/A if (pbuff == NULL) return MLIB_FAILURE;
0N/A buffs = (FTYPE **)(pbuff + bsize);
0N/A }
0N/A
0N/A for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*wid;
0N/A for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
0N/A buffd = buffs[n] + wid;
0N/A buffo = (mlib_s32*)(buffd + wid);
0N/A buffi = buffo + (wid &~ 1);
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (m - 1);
0N/A hgt -= (n - 1);
0N/A adr_dst += dn*dll + dm*nchannel;
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A for (l = 0; l < n; l++) {
0N/A FTYPE *buff = buffs[l];
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid + (m - 1); i++) {
0N/A buff[i] = (FTYPE)sl[i*chan1];
0N/A }
0N/A
0N/A sl += sll;
0N/A }
0N/A
0N/A buff_ind = 0;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid; i++) buffd[i] = 0.0;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A FTYPE **buffc = buffs + buff_ind;
0N/A FTYPE *buffn = buffc[n];
0N/A FTYPE *pk = k;
0N/A
0N/A for (l = 0; l < n; l++) {
0N/A FTYPE *buff_l = buffc[l];
0N/A
0N/A for (off = 0; off < m;) {
0N/A FTYPE *buff = buff_l + off;
0N/A
0N/A kw = m - off;
0N/A
0N/A if (kw > 2*MAX_KER) kw = MAX_KER; else
0N/A if (kw > MAX_KER) kw = kw/2;
0N/A off += kw;
0N/A
0N/A sp = sl;
0N/A dp = dl;
0N/A
0N/A p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
0N/A p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
0N/A
0N/A k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
0N/A k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
0N/A pk += kw;
0N/A
0N/A if (kw == 7) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
0N/A
0N/A p6 = buff[i + 6]; p7 = buff[i + 7];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
0N/A
0N/A p6 = buff[i + 6]; p7 = buff[i + 7];
0N/A
0N/A LOAD_BUFF(buffi);
0N/A
0N/A dd.d64 = *(FTYPE *)(buffi + i);
0N/A buffn[i ] = (FTYPE)dd.i32s.i0;
0N/A buffn[i + 1] = (FTYPE)dd.i32s.i1;
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 6) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
0N/A
0N/A p5 = buff[i + 5]; p6 = buff[i + 6];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
0N/A
0N/A p5 = buff[i + 5]; p6 = buff[i + 6];
0N/A
0N/A buffn[i ] = (FTYPE)sp[0];
0N/A buffn[i + 1] = (FTYPE)sp[chan1];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 5) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5;
0N/A
0N/A p4 = buff[i + 4]; p5 = buff[i + 5];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5;
0N/A
0N/A p4 = buff[i + 4]; p5 = buff[i + 5];
0N/A
0N/A buffn[i ] = (FTYPE)sp[0];
0N/A buffn[i + 1] = (FTYPE)sp[chan1];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 4) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A
0N/A p3 = buff[i + 3]; p4 = buff[i + 4];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A
0N/A p3 = buff[i + 3]; p4 = buff[i + 4];
0N/A
0N/A buffn[i ] = (FTYPE)sp[0];
0N/A buffn[i + 1] = (FTYPE)sp[chan1];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 3) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3;
0N/A
0N/A p2 = buff[i + 2]; p3 = buff[i + 3];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3;
0N/A
0N/A p2 = buff[i + 2]; p3 = buff[i + 3];
0N/A
0N/A buffn[i ] = (FTYPE)sp[0];
0N/A buffn[i + 1] = (FTYPE)sp[chan1];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else /*if (kw == 2)*/ {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2;
0N/A
0N/A p1 = buff[i + 1]; p2 = buff[i + 2];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1;
0N/A buffd[i + 1] += p1*k0 + p2*k1;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2;
0N/A
0N/A p1 = buff[i + 1]; p2 = buff[i + 2];
0N/A
0N/A buffn[i ] = (FTYPE)sp[0];
0N/A buffn[i + 1] = (FTYPE)sp[chan1];
0N/A
0N/A d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
0N/A d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
0N/A
0N/A dp[0 ] = FROM_S32(d0);
0N/A dp[chan1] = FROM_S32(d1);
0N/A
0N/A buffd[i ] = 0.0;
0N/A buffd[i + 1] = 0.0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A }
0N/A }
0N/A }
0N/A
0N/A /* last pixels */
0N/A for (; i < wid; i++) {
0N/A FTYPE *pk = k, s = 0;
0N/A mlib_s32 x, d0;
0N/A
0N/A for (l = 0; l < n; l++) {
0N/A FTYPE *buff = buffc[l] + i;
0N/A
0N/A for (x = 0; x < m; x++) s += buff[x] * (*pk++);
0N/A }
0N/A
0N/A d0 = D2I(s);
0N/A dp[0] = FROM_S32(d0);
0N/A
0N/A buffn[i] = (FTYPE)sp[0];
0N/A
0N/A sp += chan1;
0N/A dp += chan1;
0N/A }
0N/A
0N/A for (l = 0; l < (m - 1); l++) buffn[wid + l] = sp[l*chan1];
0N/A
0N/A /* next line */
0N/A sl += sll;
0N/A dl += dll;
0N/A
0N/A buff_ind++;
0N/A
0N/A if (buff_ind >= n + 1) buff_ind = 0;
0N/A }
0N/A }
0N/A
0N/A if (pbuff != buff) mlib_free(pbuff);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/
0N/A#ifndef __sparc /* for x86, using integer multiplies is faster */
0N/A
0N/A#define STORE_RES(res, x) \
0N/A x >>= shift2; \
0N/A CLAMP_STORE(res, x)
0N/A
0N/Amlib_status CONV_FUNC_I(MxN)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_s32 *kernel,
0N/A mlib_s32 m,
0N/A mlib_s32 n,
0N/A mlib_s32 dm,
0N/A mlib_s32 dn,
0N/A mlib_s32 scale,
0N/A mlib_s32 cmask)
0N/A{
0N/A mlib_s32 buff[BUFF_SIZE], *buffd = buff;
0N/A mlib_s32 l, off, kw;
0N/A mlib_s32 d0, d1, shift1, shift2;
0N/A mlib_s32 k0, k1, k2, k3, k4, k5, k6;
0N/A mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7;
3813N/A DTYPE *adr_src, *sl, *sp = NULL;
3813N/A DTYPE *adr_dst, *dl, *dp = NULL;
0N/A mlib_s32 wid, hgt, sll, dll;
0N/A mlib_s32 nchannel, chan1;
0N/A mlib_s32 i, j, c;
0N/A mlib_s32 chan2;
0N/A mlib_s32 k_locl[MAX_N*MAX_N], *k = k_locl;
0N/A GET_SRC_DST_PARAMETERS(DTYPE);
0N/A
0N/A#if IMG_TYPE != 1
0N/A shift1 = 16;
0N/A#else
0N/A shift1 = 8;
0N/A#endif /* IMG_TYPE != 1 */
0N/A shift2 = scale - shift1;
0N/A
0N/A chan1 = nchannel;
0N/A chan2 = chan1 + chan1;
0N/A
0N/A wid -= (m - 1);
0N/A hgt -= (n - 1);
0N/A adr_dst += dn*dll + dm*nchannel;
0N/A
0N/A if (wid > BUFF_SIZE) {
0N/A buffd = mlib_malloc(sizeof(mlib_s32)*wid);
0N/A
0N/A if (buffd == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A if (m*n > MAX_N*MAX_N) {
0N/A k = mlib_malloc(sizeof(mlib_s32)*(m*n));
0N/A
0N/A if (k == NULL) {
0N/A if (buffd != buff) mlib_free(buffd);
0N/A return MLIB_FAILURE;
0N/A }
0N/A }
0N/A
0N/A for (i = 0; i < m*n; i++) {
0N/A k[i] = kernel[i] >> shift1;
0N/A }
0N/A
0N/A for (c = 0; c < nchannel; c++) {
0N/A if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
0N/A
0N/A sl = adr_src + c;
0N/A dl = adr_dst + c;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i < wid; i++) buffd[i] = 0;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A mlib_s32 *pk = k;
0N/A
0N/A for (l = 0; l < n; l++) {
0N/A DTYPE *sp0 = sl + l*sll;
0N/A
0N/A for (off = 0; off < m;) {
0N/A sp = sp0 + off*chan1;
0N/A dp = dl;
0N/A
0N/A kw = m - off;
0N/A
0N/A if (kw > 2*MAX_KER) kw = MAX_KER; else
0N/A if (kw > MAX_KER) kw = kw/2;
0N/A off += kw;
0N/A
0N/A p2 = sp[0]; p3 = sp[chan1]; p4 = sp[chan2];
0N/A p5 = sp[chan2 + chan1]; p6 = sp[chan2 + chan2]; p7 = sp[5*chan1];
0N/A
0N/A k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
0N/A k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
0N/A pk += kw;
0N/A
0N/A sp += (kw - 1)*chan1;
0N/A
0N/A if (kw == 7) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
0N/A p6 = sp[0];
0N/A p7 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
0N/A p6 = sp[0];
0N/A p7 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
0N/A d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 6) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
0N/A p5 = sp[0];
0N/A p6 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
0N/A p5 = sp[0];
0N/A p6 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
0N/A d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 5) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5;
0N/A p4 = sp[0];
0N/A p5 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4; p3 = p5;
0N/A p4 = sp[0];
0N/A p5 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
0N/A d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 4) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A p3 = sp[0];
0N/A p4 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3; p2 = p4;
0N/A p3 = sp[0];
0N/A p4 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
0N/A d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 3) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3;
0N/A p2 = sp[0];
0N/A p3 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1 + p2*k2;
0N/A buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2; p1 = p3;
0N/A p2 = sp[0];
0N/A p3 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
0N/A d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 2) {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2;
0N/A p1 = sp[0];
0N/A p2 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0 + p1*k1;
0N/A buffd[i + 1] += p1*k0 + p2*k1;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = p2;
0N/A p1 = sp[0];
0N/A p2 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + p1*k1 + buffd[i ]);
0N/A d1 = (p1*k0 + p2*k1 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else /*if (kw == 1)*/ {
0N/A
0N/A if (l < (n - 1) || off < m) {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = sp[0];
0N/A p1 = sp[chan1];
0N/A
0N/A buffd[i ] += p0*k0;
0N/A buffd[i + 1] += p1*k0;
0N/A
0N/A sp += chan2;
0N/A }
0N/A
0N/A } else {
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 2); i += 2) {
0N/A p0 = sp[0];
0N/A p1 = sp[chan1];
0N/A
0N/A d0 = (p0*k0 + buffd[i ]);
0N/A d1 = (p1*k0 + buffd[i + 1]);
0N/A
0N/A STORE_RES(dp[0 ], d0);
0N/A STORE_RES(dp[chan1], d1);
0N/A
0N/A buffd[i ] = 0;
0N/A buffd[i + 1] = 0;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A }
0N/A }
0N/A }
0N/A
0N/A /* last pixels */
0N/A for (; i < wid; i++) {
0N/A mlib_s32 *pk = k, s = 0;
0N/A mlib_s32 x;
0N/A
0N/A for (l = 0; l < n; l++) {
0N/A sp = sl + l*sll + i*chan1;
0N/A
0N/A for (x = 0; x < m; x++) {
0N/A s += sp[0] * pk[0];
0N/A sp += chan1;
0N/A pk ++;
0N/A }
0N/A }
0N/A
0N/A STORE_RES(dp[0], s);
0N/A
0N/A sp += chan1;
0N/A dp += chan1;
0N/A }
0N/A
0N/A sl += sll;
0N/A dl += dll;
0N/A }
0N/A }
0N/A
0N/A if (buffd != buff) mlib_free(buffd);
0N/A if (k != k_locl) mlib_free(k);
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/
0N/A#endif /* __sparc ( for x86, using integer multiplies is faster ) */
0N/A
0N/A/***************************************************************/