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#include "vis_proto.h"
0N/A#include "mlib_image.h"
0N/A#include "mlib_v_ImageLookUpFunc.h"
0N/A
0N/A/***************************************************************/
0N/Astatic void mlib_v_ImageLookUp_U8_U16_124_SrcOff0_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_124_SrcOff1_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_124_SrcOff2_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_124_SrcOff3_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_3_SrcOff0_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_3_SrcOff1_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_3_SrcOff2_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2);
0N/A
0N/Astatic void mlib_v_ImageLookUp_U8_U16_3_SrcOff3_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2);
0N/A
0N/A/***************************************************************/
0N/A#define VIS_LD_U16_I(X, Y) vis_ld_u16_i((void *)(X), (Y))
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_124_SrcOff0_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, acc0; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A
0N/A sa = (mlib_u32*)src;
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A i = 0;
0N/A
0N/A if (xsize >= 4) {
0N/A
0N/A s0 = *sa++;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 8; i+=4) {
0N/A t3 = VIS_LD_U16_I(table3, (s0 << 1) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = *sa++;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table3, (s0 << 1) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_124_SrcOff1_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, acc0; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A
0N/A sa = (mlib_u32*)(src - 1);
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A s0 = *sa++;
0N/A
0N/A if (xsize >= 4) {
0N/A
0N/A s1 = *sa++;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 8; i+=4) {
0N/A t3 = VIS_LD_U16_I(table3, (s1 >> 23) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 << 1) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 7) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 15) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = s1;
0N/A s1 = *sa++;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table3, (s1 >> 23) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 << 1) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 7) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 15) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = s1;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A sp -= 3;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_124_SrcOff2_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, acc0; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A
0N/A sa = (mlib_u32*)(src - 2);
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A s0 = *sa++;
0N/A
0N/A if (xsize >= 4) {
0N/A
0N/A s1 = *sa++;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 8; i+=4) {
0N/A t3 = VIS_LD_U16_I(table3, (s1 >> 15) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 23) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 << 1) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 7) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = s1;
0N/A s1 = *sa++;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table3, (s1 >> 15) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 23) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 << 1) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 7) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = s1;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A sp -= 2;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_124_SrcOff3_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2,
0N/A const mlib_u16 *table3)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, acc0; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A
0N/A sa = (mlib_u32*)(src - 3);
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A s0 = *sa++;
0N/A
0N/A if (xsize >= 4) {
0N/A
0N/A s1 = *sa++;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 8; i+=4) {
0N/A t3 = VIS_LD_U16_I(table3, (s1 >> 7) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = s1;
0N/A s1 = *sa++;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table3, (s1 >> 7) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A s0 = s1;
0N/A *dp++ = acc0;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A sp -= 1;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_1(const mlib_u8 *src,
0N/A mlib_s32 slb,
0N/A mlib_u16 *dst,
0N/A mlib_s32 dlb,
0N/A mlib_s32 xsize,
0N/A mlib_s32 ysize,
0N/A const mlib_u16 **table)
0N/A{
0N/A mlib_u8 *sl;
0N/A mlib_u16 *dl;
0N/A const mlib_u16 *tab = table[0];
0N/A mlib_s32 j, i;
0N/A
0N/A sl = (void *)src;
0N/A dl = dst;
0N/A
0N/A /* row loop */
0N/A for (j = 0; j < ysize; j ++) {
0N/A mlib_u8 *sp = sl;
0N/A mlib_u16 *dp = dl;
0N/A mlib_s32 off, size = xsize;
0N/A
0N/A off = ((8 - ((mlib_addr)dp & 7)) & 7) >> 1;
0N/A
0N/A off = (off < size) ? off : size;
0N/A
0N/A for (i = 0; i < off; i++) {
0N/A *dp++ = tab[(*sp++)];
0N/A size--;
0N/A }
0N/A
0N/A if (size > 0) {
0N/A
0N/A off = (mlib_addr)sp & 3;
0N/A
0N/A if (off == 0) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff0_D1(sp, dp, size, tab, tab, tab, tab);
0N/A } else if (off == 1) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff1_D1(sp, dp, size, tab, tab, tab, tab);
0N/A } else if (off == 2) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff2_D1(sp, dp, size, tab, tab, tab, tab);
0N/A } else {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff3_D1(sp, dp, size, tab, tab, tab, tab);
0N/A }
0N/A }
0N/A
0N/A sl = (mlib_u8 *) ((mlib_u8 *) sl + slb);
0N/A dl = (mlib_u16 *) ((mlib_u8 *) dl + dlb);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_2(const mlib_u8 *src,
0N/A mlib_s32 slb,
0N/A mlib_u16 *dst,
0N/A mlib_s32 dlb,
0N/A mlib_s32 xsize,
0N/A mlib_s32 ysize,
0N/A const mlib_u16 **table)
0N/A{
0N/A mlib_u8 *sl;
0N/A mlib_u16 *dl;
0N/A const mlib_u16 *tab;
0N/A mlib_s32 j, i;
0N/A
0N/A sl = (void *)src;
0N/A dl = dst;
0N/A
0N/A /* row loop */
0N/A for (j = 0; j < ysize; j ++) {
0N/A mlib_u8 *sp = sl;
0N/A mlib_u16 *dp = dl;
0N/A mlib_s32 off, size = xsize * 2;
0N/A const mlib_u16 *tab0 = table[0];
0N/A const mlib_u16 *tab1 = table[1];
0N/A
0N/A off = ((8 - ((mlib_addr)dp & 7)) & 7) >> 1;
0N/A
0N/A off = (off < size) ? off : size;
0N/A
0N/A for (i = 0; i < off - 1; i+=2) {
0N/A *dp++ = tab0[(*sp++)];
0N/A *dp++ = tab1[(*sp++)];
0N/A size-=2;
0N/A }
0N/A
0N/A if ((off & 1) != 0) {
0N/A *dp++ = tab0[(*sp++)];
0N/A size--;
0N/A tab = tab0; tab0 = tab1; tab1 = tab;
0N/A }
0N/A
0N/A if (size > 0) {
0N/A
0N/A off = (mlib_addr)sp & 3;
0N/A
0N/A if (off == 0) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff0_D1(sp, dp, size, tab0, tab1, tab0, tab1);
0N/A } else if (off == 1) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff1_D1(sp, dp, size, tab0, tab1, tab0, tab1);
0N/A } else if (off == 2) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff2_D1(sp, dp, size, tab0, tab1, tab0, tab1);
0N/A } else {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff3_D1(sp, dp, size, tab0, tab1, tab0, tab1);
0N/A }
0N/A }
0N/A
0N/A sl = (mlib_u8 *) ((mlib_u8 *) sl + slb);
0N/A dl = (mlib_u16 *) ((mlib_u8 *) dl + dlb);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_4(const mlib_u8 *src,
0N/A mlib_s32 slb,
0N/A mlib_u16 *dst,
0N/A mlib_s32 dlb,
0N/A mlib_s32 xsize,
0N/A mlib_s32 ysize,
0N/A const mlib_u16 **table)
0N/A{
0N/A mlib_u8 *sl;
0N/A mlib_u16 *dl;
0N/A const mlib_u16 *tab;
0N/A mlib_s32 j;
0N/A
0N/A sl = (void *)src;
0N/A dl = dst;
0N/A
0N/A /* row loop */
0N/A for (j = 0; j < ysize; j ++) {
0N/A mlib_u8 *sp = sl;
0N/A mlib_u16 *dp = dl;
0N/A const mlib_u16 *tab0 = table[0];
0N/A const mlib_u16 *tab1 = table[1];
0N/A const mlib_u16 *tab2 = table[2];
0N/A const mlib_u16 *tab3 = table[3];
0N/A mlib_s32 off, size = xsize * 4;
0N/A
0N/A off = ((8 - ((mlib_addr)dp & 7)) & 7) >> 1;
0N/A
0N/A off = (off < size) ? off : size;
0N/A
0N/A if (off == 1) {
0N/A *dp++ = tab0[(*sp++)];
0N/A tab = tab0; tab0 = tab1;
0N/A tab1 = tab2; tab2 = tab3; tab3 = tab;
0N/A size--;
0N/A } else if (off == 2) {
0N/A *dp++ = tab0[(*sp++)];
0N/A *dp++ = tab1[(*sp++)];
0N/A tab = tab0; tab0 = tab2; tab2 = tab;
0N/A tab = tab1; tab1 = tab3; tab3 = tab;
0N/A size-=2;
0N/A } else if (off == 3) {
0N/A *dp++ = tab0[(*sp++)];
0N/A *dp++ = tab1[(*sp++)];
0N/A *dp++ = tab2[(*sp++)];
0N/A tab = tab3; tab3 = tab2;
0N/A tab2 = tab1; tab1 = tab0; tab0 = tab;
0N/A size-=3;
0N/A }
0N/A
0N/A if (size > 0) {
0N/A
0N/A off = (mlib_addr)sp & 3;
0N/A
0N/A if (off == 0) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff0_D1(sp, dp, size, tab0, tab1, tab2, tab3);
0N/A } else if (off == 1) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff1_D1(sp, dp, size, tab0, tab1, tab2, tab3);
0N/A } else if (off == 2) {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff2_D1(sp, dp, size, tab0, tab1, tab2, tab3);
0N/A } else {
0N/A mlib_v_ImageLookUp_U8_U16_124_SrcOff3_D1(sp, dp, size, tab0, tab1, tab2, tab3);
0N/A }
0N/A }
0N/A
0N/A sl = (mlib_u8 *) ((mlib_u8 *) sl + slb);
0N/A dl = (mlib_u16 *) ((mlib_u8 *) dl + dlb);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_3_SrcOff0_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1, s2; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, t4, t5; /* destination data */
0N/A mlib_d64 t6, t7, t8; /* destination data */
0N/A mlib_d64 t9, t10, t11; /* destination data */
0N/A mlib_d64 acc0, acc1, acc2; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A const mlib_u16 *table;
0N/A
0N/A sa = (mlib_u32*)src;
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A i = 0;
0N/A
0N/A if (xsize >= 12) {
0N/A
0N/A s0 = sa[0];
0N/A s1 = sa[1];
0N/A s2 = sa[2];
0N/A sa += 3;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 24; i+=12, sa += 3, dp += 3) {
0N/A t3 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s1 << 1) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s2 << 1) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s2 >> 7) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s2 >> 15) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 >> 23) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A s0 = sa[0];
0N/A s1 = sa[1];
0N/A s2 = sa[2];
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s1 << 1) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s2 << 1) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s2 >> 7) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s2 >> 15) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 >> 23) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A dp += 3; i += 12;
0N/A }
0N/A
0N/A if (i <= xsize - 8) {
0N/A s0 = sa[0];
0N/A s1 = sa[1];
0N/A t3 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s1 << 1) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A table = table0; table0 = table2;
0N/A table2 = table1; table1 = table;
0N/A sa += 2; i += 8; dp += 2;
0N/A }
0N/A
0N/A if (i <= xsize - 4) {
0N/A s0 = sa[0];
0N/A t3 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A dp[0] = acc0;
0N/A table = table0; table0 = table1;
0N/A table1 = table2; table2 = table;
0N/A sa++; i += 4; dp++;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_3_SrcOff1_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1, s2, s3; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, t4, t5; /* destination data */
0N/A mlib_d64 t6, t7, t8; /* destination data */
0N/A mlib_d64 t9, t10, t11; /* destination data */
0N/A mlib_d64 acc0, acc1, acc2; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A const mlib_u16 *table;
0N/A
0N/A sa = (mlib_u32*)(src - 1);
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A i = 0;
0N/A
0N/A s0 = *sa++;
0N/A
0N/A if (xsize >= 12) {
0N/A
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A s3 = sa[2];
0N/A sa += 3;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 24; i+=12, sa += 3, dp += 3) {
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 23) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 << 1) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 7) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 15) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 23) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s1 << 1) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 >> 7) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 15) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s3 >> 23) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s2 << 1) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s2 >> 7) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 >> 15) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A s0 = s3;
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A s3 = sa[2];
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 23) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 << 1) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 7) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 15) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 23) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s1 << 1) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 >> 7) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 15) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s3 >> 23) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s2 << 1) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s2 >> 7) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 >> 15) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A s0 = s3;
0N/A dp += 3; i += 12;
0N/A }
0N/A
0N/A if (i <= xsize - 8) {
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 23) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 << 1) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 7) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 15) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 23) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s1 << 1) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 >> 7) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 15) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A table = table0; table0 = table2;
0N/A table2 = table1; table1 = table;
0N/A sa += 2; i += 8; dp += 2;
0N/A s0 = s2;
0N/A }
0N/A
0N/A if (i <= xsize - 4) {
0N/A s1 = sa[0];
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 23) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s0 << 1) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 >> 7) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 15) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A dp[0] = acc0;
0N/A table = table0; table0 = table1;
0N/A table1 = table2; table2 = table;
0N/A sa++; i += 4; dp++;
0N/A s0 = s1;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A sp -= 3;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_3_SrcOff2_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1, s2, s3; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, t4, t5; /* destination data */
0N/A mlib_d64 t6, t7, t8; /* destination data */
0N/A mlib_d64 t9, t10, t11; /* destination data */
0N/A mlib_d64 acc0, acc1, acc2; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A const mlib_u16 *table;
0N/A
0N/A sa = (mlib_u32*)(src - 2);
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A i = 0;
0N/A
0N/A s0 = *sa++;
0N/A
0N/A if (xsize >= 12) {
0N/A
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A s3 = sa[2];
0N/A sa += 3;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 24; i+=12, sa += 3, dp += 3) {
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 15) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 23) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 << 1) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 7) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 15) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s2 >> 23) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 << 1) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 7) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s3 >> 15) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s3 >> 23) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s2 << 1) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 >> 7) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A s0 = s3;
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A s3 = sa[2];
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 15) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 23) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 << 1) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 7) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 15) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s2 >> 23) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 << 1) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 7) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s3 >> 15) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s3 >> 23) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s2 << 1) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 >> 7) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A s0 = s3;
0N/A dp += 3; i += 12;
0N/A }
0N/A
0N/A if (i <= xsize - 8) {
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 15) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 23) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 << 1) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 7) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 15) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s2 >> 23) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s1 << 1) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 >> 7) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A table = table0; table0 = table2;
0N/A table2 = table1; table1 = table;
0N/A sa += 2; i += 8; dp += 2;
0N/A s0 = s2;
0N/A }
0N/A
0N/A if (i <= xsize - 4) {
0N/A s1 = sa[0];
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 15) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 23) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s0 << 1) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 >> 7) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A dp[0] = acc0;
0N/A table = table0; table0 = table1;
0N/A table1 = table2; table2 = table;
0N/A sa++; i += 4; dp++;
0N/A s0 = s1;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A sp -= 2;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_3_SrcOff3_D1(const mlib_u8 *src,
0N/A mlib_u16 *dst,
0N/A mlib_s32 xsize,
0N/A const mlib_u16 *table0,
0N/A const mlib_u16 *table1,
0N/A const mlib_u16 *table2)
0N/A{
0N/A mlib_u32 *sa; /* aligned pointer to source data */
0N/A mlib_u8 *sp; /* pointer to source data */
0N/A mlib_u32 s0, s1, s2, s3; /* source data */
0N/A mlib_u16 *dl; /* pointer to start of destination */
0N/A mlib_u16 *dend; /* pointer to end of destination */
0N/A mlib_d64 *dp; /* aligned pointer to destination */
0N/A mlib_d64 t0, t1, t2; /* destination data */
0N/A mlib_d64 t3, t4, t5; /* destination data */
0N/A mlib_d64 t6, t7, t8; /* destination data */
0N/A mlib_d64 t9, t10, t11; /* destination data */
0N/A mlib_d64 acc0, acc1, acc2; /* destination data */
0N/A mlib_s32 emask; /* edge mask */
0N/A mlib_s32 i, num; /* loop variable */
0N/A const mlib_u16 *table;
0N/A
0N/A sa = (mlib_u32*)(src - 3);
0N/A dl = dst;
0N/A dp = (mlib_d64 *) dl;
0N/A dend = dl + xsize - 1;
0N/A
0N/A vis_alignaddr((void *) 0, 6);
0N/A
0N/A i = 0;
0N/A
0N/A s0 = *sa++;
0N/A
0N/A if (xsize >= 12) {
0N/A
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A s3 = sa[2];
0N/A sa += 3;
0N/A
0N/A#pragma pipeloop(0)
0N/A for(i = 0; i <= xsize - 24; i+=12, sa += 3, dp += 3) {
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 7) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s2 >> 15) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s2 >> 23) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 << 1) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s3 >> 7) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s3 >> 15) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s3 >> 23) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 << 1) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A s0 = s3;
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A s3 = sa[2];
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A }
0N/A
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 7) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s2 >> 15) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s2 >> 23) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 << 1) & 0x1FE);
0N/A t11 = VIS_LD_U16_I(table2, (s3 >> 7) & 0x1FE);
0N/A t10 = VIS_LD_U16_I(table1, (s3 >> 15) & 0x1FE);
0N/A t9 = VIS_LD_U16_I(table0, (s3 >> 23) & 0x1FE);
0N/A t8 = VIS_LD_U16_I(table2, (s2 << 1) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A acc2 = vis_faligndata(t11, acc2);
0N/A acc2 = vis_faligndata(t10, acc2);
0N/A acc2 = vis_faligndata(t9, acc2);
0N/A acc2 = vis_faligndata(t8, acc2);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A dp[2] = acc2;
0N/A s0 = s3;
0N/A dp += 3; i += 12;
0N/A }
0N/A
0N/A if (i <= xsize - 8) {
0N/A s1 = sa[0];
0N/A s2 = sa[1];
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A t7 = VIS_LD_U16_I(table1, (s2 >> 7) & 0x1FE);
0N/A t6 = VIS_LD_U16_I(table0, (s2 >> 15) & 0x1FE);
0N/A t5 = VIS_LD_U16_I(table2, (s2 >> 23) & 0x1FE);
0N/A t4 = VIS_LD_U16_I(table1, (s1 << 1) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A acc1 = vis_faligndata(t7, acc1);
0N/A acc1 = vis_faligndata(t6, acc1);
0N/A acc1 = vis_faligndata(t5, acc1);
0N/A acc1 = vis_faligndata(t4, acc1);
0N/A dp[0] = acc0;
0N/A dp[1] = acc1;
0N/A table = table0; table0 = table2;
0N/A table2 = table1; table1 = table;
0N/A sa += 2; i += 8; dp += 2;
0N/A s0 = s2;
0N/A }
0N/A
0N/A if (i <= xsize - 4) {
0N/A s1 = sa[0];
0N/A t3 = VIS_LD_U16_I(table0, (s1 >> 7) & 0x1FE);
0N/A t2 = VIS_LD_U16_I(table2, (s1 >> 15) & 0x1FE);
0N/A t1 = VIS_LD_U16_I(table1, (s1 >> 23) & 0x1FE);
0N/A t0 = VIS_LD_U16_I(table0, (s0 << 1) & 0x1FE);
0N/A acc0 = vis_faligndata(t3, acc0);
0N/A acc0 = vis_faligndata(t2, acc0);
0N/A acc0 = vis_faligndata(t1, acc0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A dp[0] = acc0;
0N/A table = table0; table0 = table1;
0N/A table1 = table2; table2 = table;
0N/A sa++; i += 4; dp++;
0N/A s0 = s1;
0N/A }
0N/A
0N/A sp = (mlib_u8*)sa;
0N/A sp -= 1;
0N/A
0N/A if ((mlib_addr) dp <= (mlib_addr) dend) {
0N/A
0N/A num = (mlib_u16*) dend - (mlib_u16*) dp;
0N/A sp += num;
0N/A num ++;
0N/A
0N/A if (num == 1) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 2) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A } else if (num == 3) {
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table2, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table1, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A
0N/A s0 = (mlib_s32) *sp;
0N/A sp --;
0N/A
0N/A t0 = VIS_LD_U16_I(table0, 2*s0);
0N/A acc0 = vis_faligndata(t0, acc0);
0N/A }
0N/A
0N/A emask = vis_edge16(dp, dend);
0N/A vis_pst_16(acc0, dp, emask);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/
0N/Avoid mlib_v_ImageLookUp_U8_U16_3(const mlib_u8 *src,
0N/A mlib_s32 slb,
0N/A mlib_u16 *dst,
0N/A mlib_s32 dlb,
0N/A mlib_s32 xsize,
0N/A mlib_s32 ysize,
0N/A const mlib_u16 **table)
0N/A{
0N/A mlib_u8 *sl;
0N/A mlib_u16 *dl;
0N/A const mlib_u16 *tab;
0N/A mlib_s32 j, i;
0N/A
0N/A sl = (void *)src;
0N/A dl = dst;
0N/A
0N/A /* row loop */
0N/A for (j = 0; j < ysize; j ++) {
0N/A mlib_u8 *sp = sl;
0N/A mlib_u16 *dp = dl;
0N/A const mlib_u16 *tab0 = table[0];
0N/A const mlib_u16 *tab1 = table[1];
0N/A const mlib_u16 *tab2 = table[2];
0N/A mlib_s32 off, size = xsize * 3;
0N/A
0N/A off = ((8 - ((mlib_addr)dp & 7)) & 7) >> 1;
0N/A
0N/A off = (off < size) ? off : size;
0N/A
0N/A for (i = 0; i < off - 2; i += 3) {
0N/A *dp++ = tab0[(*sp++)];
0N/A *dp++ = tab1[(*sp++)];
0N/A *dp++ = tab2[(*sp++)];
0N/A size-=3;
0N/A }
0N/A
0N/A off -= i;
0N/A
0N/A if (off == 1) {
0N/A *dp++ = tab0[(*sp++)];
0N/A tab = tab0; tab0 = tab1;
0N/A tab1 = tab2; tab2 = tab;
0N/A size--;
0N/A } else if (off == 2) {
0N/A *dp++ = tab0[(*sp++)];
0N/A *dp++ = tab1[(*sp++)];
0N/A tab = tab2; tab2 = tab1;
0N/A tab1 = tab0; tab0 = tab;
0N/A size-=2;
0N/A }
0N/A
0N/A if (size > 0) {
0N/A
0N/A off = (mlib_addr)sp & 3;
0N/A
0N/A if (off == 0) {
0N/A mlib_v_ImageLookUp_U8_U16_3_SrcOff0_D1(sp, dp, size, tab0, tab1, tab2);
0N/A } else if (off == 1) {
0N/A mlib_v_ImageLookUp_U8_U16_3_SrcOff1_D1(sp, dp, size, tab0, tab1, tab2);
0N/A } else if (off == 2) {
0N/A mlib_v_ImageLookUp_U8_U16_3_SrcOff2_D1(sp, dp, size, tab0, tab1, tab2);
0N/A } else {
0N/A mlib_v_ImageLookUp_U8_U16_3_SrcOff3_D1(sp, dp, size, tab0, tab1, tab2);
0N/A }
0N/A }
0N/A
0N/A sl = (mlib_u8 *) ((mlib_u8 *) sl + slb);
0N/A dl = (mlib_u16 *) ((mlib_u8 *) dl + dlb);
0N/A }
0N/A}
0N/A
0N/A/***************************************************************/