0N/A/*
2362N/A * Copyright (c) 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 D64/F32 type and
0N/A * MLIB_EDGE_DST_NO_WRITE mask
0N/A *
0N/A */
0N/A
0N/A#include "mlib_image.h"
0N/A#include "mlib_ImageConv.h"
0N/A
0N/A/***************************************************************/
0N/A/*
0N/A This define switches between functions of MLIB_DOUBLE and MLIB_FLOAT types:
0N/A Files mlib_ImageConv_D64nw.c and mlib_ImageConv_F32nw.c
0N/A*/
0N/A
0N/A#define TYPE_DOUBLE
0N/A
0N/A/***************************************************************/
0N/A#ifdef TYPE_DOUBLE
0N/A
0N/A#define CONV_FUNC(KERN) mlib_conv##KERN##nw_d64
0N/A
0N/A#define DTYPE mlib_d64
0N/A
0N/A#else
0N/A
0N/A#define CONV_FUNC(KERN) mlib_conv##KERN##nw_f32
0N/A
0N/A#define DTYPE mlib_f32
0N/A
0N/A#endif /* TYPE_DOUBLE */
0N/A
0N/A/***************************************************************/
0N/A#define GET_SRC_DST_PARAMETERS(type) \
0N/A mlib_s32 hgt = mlib_ImageGetHeight(src); \
0N/A mlib_s32 wid = mlib_ImageGetWidth(src); \
0N/A mlib_s32 sll = mlib_ImageGetStride(src) / sizeof(type); \
0N/A mlib_s32 dll = mlib_ImageGetStride(dst) / sizeof(type); \
0N/A type* adr_src = mlib_ImageGetData(src); \
0N/A type* adr_dst = mlib_ImageGetData(dst); \
0N/A mlib_s32 chan1 = mlib_ImageGetChannels(src)
0N/A
0N/A/***************************************************************/
0N/A#define DEF_VARS(type) \
0N/A GET_SRC_DST_PARAMETERS(type); \
0N/A type *sl; \
3813N/A type *dl, *dp = NULL; \
3813N/A mlib_s32 i = 0, j, c
0N/A
0N/A/***************************************************************/
0N/A#undef KSIZE
0N/A#define KSIZE 2
0N/A
0N/Amlib_status CONV_FUNC(2x2)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_d64 *kern,
0N/A mlib_s32 cmask)
0N/A{
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sp0, *sp1;
0N/A mlib_s32 chan2 = chan1 + chan1;
0N/A mlib_s32 chan3 = chan1 + chan2;
0N/A mlib_s32 chan4 = chan3 + chan1;
0N/A DTYPE k0, k1, k2, k3;
0N/A DTYPE p00, p01, p02, p03, p04,
0N/A p10, p11, p12, p13, p14;
0N/A
0N/A /* keep kernel in regs */
0N/A k0 = (DTYPE)kern[0]; k1 = (DTYPE)kern[1];
0N/A k2 = (DTYPE)kern[2]; k3 = (DTYPE)kern[3];
0N/A
0N/A wid -= (KSIZE - 1);
0N/A hgt -= (KSIZE - 1);
0N/A
0N/A for (c = 0; c < chan1; c++) {
0N/A if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
0N/A
0N/A dl = adr_dst + c;
0N/A sl = adr_src + c;
0N/A
0N/A for (j = 0; j < hgt; j++) {
0N/A dp = dl;
0N/A sp0 = sl;
0N/A sp1 = sp0 + sll;
0N/A
0N/A p04 = sp0[0];
0N/A p14 = sp1[0];
0N/A
0N/A sp0 += chan1;
0N/A sp1 += chan1;
0N/A
0N/A#ifdef __SUNPRO_C
0N/A#pragma pipeloop(0)
0N/A#endif /* __SUNPRO_C */
0N/A for (i = 0; i <= (wid - 4); i += 4) {
0N/A p00 = p04; p10 = p14;
0N/A
0N/A p01 = sp0[0]; p11 = sp1[0];
0N/A p02 = sp0[chan1]; p12 = sp1[chan1];
0N/A p03 = sp0[chan2]; p13 = sp1[chan2];
0N/A p04 = sp0[chan3]; p14 = sp1[chan3];
0N/A
0N/A dp[0 ] = p00 * k0 + p01 * k1 + p10 * k2 + p11 * k3;
0N/A dp[chan1] = p01 * k0 + p02 * k1 + p11 * k2 + p12 * k3;
0N/A dp[chan2] = p02 * k0 + p03 * k1 + p12 * k2 + p13 * k3;
0N/A dp[chan3] = p03 * k0 + p04 * k1 + p13 * k2 + p14 * k3;
0N/A
0N/A dp += chan4;
0N/A sp0 += chan4;
0N/A sp1 += chan4;
0N/A }
0N/A
0N/A if (i < wid) {
0N/A p00 = p04; p10 = p14;
0N/A p01 = sp0[0]; p11 = sp1[0];
0N/A dp[0] = p00 * k0 + p01 * k1 + p10 * k2 + p11 * k3;
0N/A
0N/A if ((i + 1) < wid) {
0N/A p02 = sp0[chan1]; p12 = sp1[chan1];
0N/A dp[chan1] = p01 * k0 + p02 * k1 + p11 * k2 + p12 * k3;
0N/A
0N/A if ((i + 2) < wid) {
0N/A p03 = sp0[chan2]; p13 = sp1[chan2];
0N/A dp[chan2] = p02 * k0 + p03 * k1 + p12 * k2 + p13 * k3;
0N/A }
0N/A }
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/***************************************************************/
0N/A#undef KSIZE
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_d64 *kern,
0N/A mlib_s32 cmask)
0N/A{
0N/A DEF_VARS(DTYPE);
0N/A mlib_s32 chan2 = chan1 + chan1;
0N/A DTYPE *sp0, *sp1;
0N/A DTYPE *sp2;
0N/A DTYPE k0, k1, k2, k3, k4, k5, k6, k7, k8;
0N/A DTYPE p02, p03, p12, p13, p22, p23;
0N/A
0N/A /* keep kernel in regs */
0N/A k0 = (DTYPE)kern[0]; k1 = (DTYPE)kern[1]; k2 = (DTYPE)kern[2];
0N/A k3 = (DTYPE)kern[3]; k4 = (DTYPE)kern[4]; k5 = (DTYPE)kern[5];
0N/A k6 = (DTYPE)kern[6]; k7 = (DTYPE)kern[7]; k8 = (DTYPE)kern[8];
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 DTYPE s0, s1;
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 dp[0 ] = s0 + p02 * k2 + p12 * k5 + p22 * k8;
0N/A dp[chan1] = 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 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 dp[0] = s0 + p02 * k2 + p12 * k5 + p22 * k8;
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/***************************************************************/
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_d64 *k,
0N/A mlib_s32 cmask)
0N/A{
0N/A DTYPE k0, k1, k2, k3, k4, k5, k6, k7;
0N/A DTYPE p00, p01, p02, p03, p04,
0N/A p10, p11, p12, p13, p14;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sp0, *sp1;
0N/A mlib_s32 chan2 = chan1 + chan1;
0N/A mlib_s32 chan3 = chan1 + chan2;
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 /*
0N/A * First loop on two first lines of kernel
0N/A */
0N/A sp0 = sl;
0N/A sp1 = sp0 + sll;
0N/A dp = dl;
0N/A
0N/A k0 = (DTYPE)k[0]; k1 = (DTYPE)k[1]; k2 = (DTYPE)k[2]; k3 = (DTYPE)k[3];
0N/A k4 = (DTYPE)k[4]; k5 = (DTYPE)k[5]; k6 = (DTYPE)k[6]; k7 = (DTYPE)k[7];
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
0N/A sp0 += chan3;
0N/A sp1 += chan3;
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 p03 = sp0[0]; p13 = sp1[0];
0N/A p04 = sp0[chan1]; p14 = sp1[chan1];
0N/A
0N/A dp[0 ] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 +
0N/A p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7);
0N/A dp[chan1] = (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 +
0N/A p11 * k4 + p12 * k5 + p13 * k6 + p14 * k7);
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 = sp0[0]; p13 = sp1[0];
0N/A
0N/A dp[0] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 +
0N/A p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7);
0N/A }
0N/A
0N/A /*
0N/A * Second loop on two last lines of kernel
0N/A */
0N/A sp0 = sl + 2*sll;
0N/A sp1 = sp0 + sll;
0N/A dp = dl;
0N/A
0N/A k0 = (DTYPE)k[ 8]; k1 = (DTYPE)k[ 9]; k2 = (DTYPE)k[10]; k3 = (DTYPE)k[11];
0N/A k4 = (DTYPE)k[12]; k5 = (DTYPE)k[13]; k6 = (DTYPE)k[14]; k7 = (DTYPE)k[15];
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
0N/A sp0 += chan3;
0N/A sp1 += chan3;
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 p03 = sp0[0]; p13 = sp1[0];
0N/A p04 = sp0[chan1]; p14 = sp1[chan1];
0N/A
0N/A dp[0 ] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 +
0N/A p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7);
0N/A dp[chan1] += (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 +
0N/A p11 * k4 + p12 * k5 + p13 * k6 + p14 * k7);
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 = sp0[0]; p13 = sp1[0];
0N/A
0N/A dp[0] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 +
0N/A p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7);
0N/A }
0N/A
0N/A /* next line */
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/***************************************************************/
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_d64 *k,
0N/A mlib_s32 cmask)
0N/A{
0N/A DTYPE k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
0N/A DTYPE p00, p01, p02, p03, p04, p05,
0N/A p10, p11, p12, p13, p14, p15;
0N/A DEF_VARS(DTYPE);
0N/A DTYPE *sp0, *sp1;
0N/A mlib_s32 chan2 = chan1 + chan1;
0N/A mlib_s32 chan3 = chan1 + chan2;
0N/A mlib_s32 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 /*
0N/A * First loop
0N/A */
0N/A sp0 = sl;
0N/A sp1 = sp0 + sll;
0N/A dp = dl;
0N/A
0N/A k0 = (DTYPE)k[0]; k1 = (DTYPE)k[1]; k2 = (DTYPE)k[2]; k3 = (DTYPE)k[3]; k4 = (DTYPE)k[4];
0N/A k5 = (DTYPE)k[5]; k6 = (DTYPE)k[6]; k7 = (DTYPE)k[7]; k8 = (DTYPE)k[8]; k9 = (DTYPE)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 dp[ 0] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A dp[chan1] = (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 dp[0] = (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 = (DTYPE)k[10]; k1 = (DTYPE)k[11]; k2 = (DTYPE)k[12]; k3 = (DTYPE)k[13]; k4 = (DTYPE)k[14];
0N/A k5 = (DTYPE)k[15]; k6 = (DTYPE)k[16]; k7 = (DTYPE)k[17]; k8 = (DTYPE)k[18]; k9 = (DTYPE)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 dp[ 0] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 +
0N/A p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9);
0N/A dp[chan1] += (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 dp[0] += (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 = (DTYPE)k[20]; k1 = (DTYPE)k[21]; k2 = (DTYPE)k[22]; k3 = (DTYPE)k[23]; k4 = (DTYPE)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 dp[0 ] += p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4;
0N/A dp[chan1] += p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4;
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 dp[0] += p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4;
0N/A }
0N/A
0N/A /* next line */
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/***************************************************************/
0N/A#define BUFF_SIZE 1600
0N/A
0N/A#define CACHE_SIZE (64*1024)
0N/A
0N/Astatic mlib_status mlib_ImageConv1xN(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const DTYPE *k,
0N/A mlib_s32 n,
0N/A mlib_s32 dn,
0N/A mlib_s32 cmask)
0N/A{
0N/A DTYPE buff[BUFF_SIZE], *pbuff = buff;
0N/A const DTYPE *pk;
0N/A DTYPE k0, k1, k2, k3;
0N/A DTYPE p0, p1, p2, p3, p4;
0N/A DTYPE *sp, *sl_c, *dl_c, *sl0;
0N/A DEF_VARS(DTYPE);
0N/A mlib_s32 off, kh;
0N/A mlib_s32 l, hsize, max_hsize;
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(DTYPE)*max_hsize);
0N/A }
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 < chan1; 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 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
0N/A dp[dll] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1];
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 dp[0] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
0N/A
0N/A pbuff[j] = 0;
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 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
0N/A dp[dll] = p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1];
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 dp[0] = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
0N/A
0N/A pbuff[j] = 0;
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 dp[0 ] = p0*k0 + p1*k1 + pbuff[j];
0N/A dp[dll] = p1*k0 + p2*k1 + pbuff[j + 1];
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 dp[0] = p0*k0 + p1*k1 + pbuff[j];
0N/A
0N/A pbuff[j] = 0;
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 dp[0] = p0*k0 + pbuff[j];
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/A#define MAX_KER 7
0N/A#define MAX_NM 81
0N/A
0N/Amlib_status CONV_FUNC(MxN)(mlib_image *dst,
0N/A const mlib_image *src,
0N/A const mlib_d64 *ker,
0N/A mlib_s32 m,
0N/A mlib_s32 n,
0N/A mlib_s32 dm,
0N/A mlib_s32 dn,
0N/A mlib_s32 cmask)
0N/A{
0N/A DTYPE k0, k1, k2, k3, k4, k5, k6, *sp;
0N/A DTYPE p0, p1, p2, p3, p4, p5, p6, p7;
0N/A mlib_s32 l, off, kw;
0N/A DEF_VARS(DTYPE);
0N/A mlib_s32 chan2 = chan1 + chan1;
0N/A mlib_s32 chan3 = chan1 + chan2;
0N/A
0N/A#ifdef TYPE_DOUBLE
0N/A const mlib_d64 *k = ker;
0N/A#else
0N/A mlib_f32 k_arr[MAX_NM], *k = k_arr;
0N/A
0N/A if (n*m > MAX_NM) {
0N/A k = mlib_malloc(n*m*sizeof(mlib_f32));
0N/A
0N/A if (k == NULL) return MLIB_FAILURE;
0N/A }
0N/A
0N/A for (i = 0; i < n*m; i++) k[i] = (mlib_f32)ker[i];
0N/A#endif /* TYPE_DOUBLE */
0N/A
0N/A if (m == 1) return mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
0N/A
0N/A wid -= (m - 1);
0N/A hgt -= (n - 1);
0N/A adr_dst += dn*dll + dm*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 const DTYPE *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; off += kw, pk += kw, sp0 += chan1) {
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
0N/A p2 = sp0[0]; p3 = sp0[chan1]; p4 = sp0[chan2];
0N/A sp0 += chan3;
0N/A p5 = sp0[0]; p6 = sp0[chan1]; p7 = sp0[chan2];
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
0N/A dp = dl;
0N/A
0N/A if (kw == 7) {
0N/A sp = sp0 += chan3;
0N/A
0N/A if (pk == k) {
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 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1];
0N/A
0N/A dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
0N/A dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
0N/A
0N/A sp += chan2;
0N/A dp += 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
0N/A p5 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1];
0N/A
0N/A dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
0N/A dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 6) {
0N/A sp = sp0 += chan2;
0N/A
0N/A if (pk == k) {
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 = sp[0]; p6 = sp[chan1];
0N/A
0N/A dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
0N/A dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
0N/A
0N/A sp += chan2;
0N/A dp += 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
0N/A p5 = sp[0]; p6 = sp[chan1];
0N/A
0N/A dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
0N/A dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 5) {
0N/A sp = sp0 += chan1;
0N/A
0N/A if (pk == k) {
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 = sp[0]; p5 = sp[chan1];
0N/A
0N/A dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
0N/A dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
0N/A
0N/A sp += chan2;
0N/A dp += 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
0N/A p4 = sp[0]; p5 = sp[chan1];
0N/A
0N/A dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
0N/A dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
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 sp = sp0;
0N/A
0N/A if (pk == k) {
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 = sp[0]; p4 = sp[chan1];
0N/A
0N/A dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3;
0N/A dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3;
0N/A
0N/A sp += chan2;
0N/A dp += 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
0N/A p3 = sp[0]; p4 = sp[chan1];
0N/A
0N/A dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
0N/A dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else if (kw == 3) {
0N/A sp = sp0 -= chan1;
0N/A
0N/A if (pk == k) {
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 = sp[0]; p3 = sp[chan1];
0N/A
0N/A dp[0 ] = p0*k0 + p1*k1 + p2*k2;
0N/A dp[chan1] = p1*k0 + p2*k1 + p3*k2;
0N/A
0N/A sp += chan2;
0N/A dp += 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
0N/A p2 = sp[0]; p3 = sp[chan1];
0N/A
0N/A dp[0 ] += p0*k0 + p1*k1 + p2*k2;
0N/A dp[chan1] += p1*k0 + p2*k1 + p3*k2;
0N/A
0N/A sp += chan2;
0N/A dp += chan2;
0N/A }
0N/A }
0N/A
0N/A } else { /* kw == 2 */
0N/A sp = sp0 -= chan2;
0N/A
0N/A if (pk == k) {
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 = sp[0]; p2 = sp[chan1];
0N/A
0N/A dp[0 ] = p0*k0 + p1*k1;
0N/A dp[chan1] = p1*k0 + p2*k1;
0N/A
0N/A sp += chan2;
0N/A dp += 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
0N/A p1 = sp[0]; p2 = sp[chan1];
0N/A
0N/A dp[0 ] += p0*k0 + p1*k1;
0N/A dp[chan1] += p1*k0 + p2*k1;
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
0N/A if (wid & 1) {
0N/A DTYPE *sp0 = sl + i*chan1, s = 0;
0N/A const DTYPE *pk = k;
0N/A mlib_s32 x;
0N/A
0N/A for (l = 0; l < n; l++) {
0N/A DTYPE *sp = sp0 + l*sll;
0N/A
0N/A for (x = 0; x < m; x++) s += sp[x*chan1] * (*pk++);
0N/A }
0N/A
0N/A dp[0] = s;
0N/A }
0N/A
0N/A /* next line */
0N/A sl += sll;
0N/A dl += dll;
0N/A }
0N/A }
0N/A
0N/A#ifndef TYPE_DOUBLE
0N/A
0N/A if (k != k_arr) mlib_free(k);
0N/A#endif /* TYPE_DOUBLE */
0N/A
0N/A return MLIB_SUCCESS;
0N/A}
0N/A
0N/A/***************************************************************/