rlm@1: /*
rlm@1:  * This file is part of the Advance project.
rlm@1:  *
rlm@1:  * Copyright (C) 2003 Andrea Mazzoleni
rlm@1:  *
rlm@1:  * This program is free software; you can redistribute it and/or modify
rlm@1:  * it under the terms of the GNU General Public License as published by
rlm@1:  * the Free Software Foundation; either version 2 of the License, or
rlm@1:  * (at your option) any later version.
rlm@1:  *
rlm@1:  * This program is distributed in the hope that it will be useful,
rlm@1:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
rlm@1:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
rlm@1:  * GNU General Public License for more details.
rlm@1:  *
rlm@1:  * You should have received a copy of the GNU General Public License
rlm@1:  * along with this program; if not, write to the Free Software
rlm@1:  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
rlm@1:  *
rlm@1:  * In addition, as a special exception, Andrea Mazzoleni
rlm@1:  * gives permission to link the code of this program with
rlm@1:  * the MAME library (or with modified versions of MAME that use the
rlm@1:  * same license as MAME), and distribute linked combinations including
rlm@1:  * the two.  You must obey the GNU General Public License in all
rlm@1:  * respects for all of the code used other than MAME.  If you modify
rlm@1:  * this file, you may extend this exception to your version of the
rlm@1:  * file, but you are not obligated to do so.  If you do not wish to
rlm@1:  * do so, delete this exception statement from your version.
rlm@1:  */
rlm@1: #include "../Port.h"
rlm@1: #include "interp.h"
rlm@1: 
rlm@1: unsigned interp_mask[2];
rlm@1: unsigned interp_bits_per_pixel;
rlm@1: 
rlm@1: /***************************************************************************/
rlm@1: /* HQ2x C implementation */
rlm@1: 
rlm@1: /*
rlm@1:  * This effect is a rewritten implementation of the hq2x effect made by Maxim Stepin
rlm@1:  */
rlm@1: 
rlm@1: static void hq2x_16_def(u16 *dst0, u16 *dst1, const u16 *src0, const u16 *src1, const u16 *src2, unsigned count)
rlm@1: {
rlm@1: 	unsigned i;
rlm@1: 
rlm@1: 	for (i = 0; i < count; ++i)
rlm@1: 	{
rlm@1: 		unsigned char mask;
rlm@1: 
rlm@1: 		u16 c[9];
rlm@1: 
rlm@1: 		c[1] = src0[0];
rlm@1: 		c[4] = src1[0];
rlm@1: 		c[7] = src2[0];
rlm@1: 
rlm@1: 		if (i > 0)
rlm@1: 		{
rlm@1: 			c[0] = src0[-1];
rlm@1: 			c[3] = src1[-1];
rlm@1: 			c[6] = src2[-1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[0] = c[1];
rlm@1: 			c[3] = c[4];
rlm@1: 			c[6] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		if (i < count - 1)
rlm@1: 		{
rlm@1: 			c[2] = src0[1];
rlm@1: 			c[5] = src1[1];
rlm@1: 			c[8] = src2[1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[2] = c[1];
rlm@1: 			c[5] = c[4];
rlm@1: 			c[8] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		mask = 0;
rlm@1: 
rlm@1: 		if (interp_16_diff(c[0], c[4]))
rlm@1: 			mask |= 1 << 0;
rlm@1: 		if (interp_16_diff(c[1], c[4]))
rlm@1: 			mask |= 1 << 1;
rlm@1: 		if (interp_16_diff(c[2], c[4]))
rlm@1: 			mask |= 1 << 2;
rlm@1: 		if (interp_16_diff(c[3], c[4]))
rlm@1: 			mask |= 1 << 3;
rlm@1: 		if (interp_16_diff(c[5], c[4]))
rlm@1: 			mask |= 1 << 4;
rlm@1: 		if (interp_16_diff(c[6], c[4]))
rlm@1: 			mask |= 1 << 5;
rlm@1: 		if (interp_16_diff(c[7], c[4]))
rlm@1: 			mask |= 1 << 6;
rlm@1: 		if (interp_16_diff(c[8], c[4]))
rlm@1: 			mask |= 1 << 7;
rlm@1: 
rlm@1: #define P0 dst0[0]
rlm@1: #define P1 dst0[1]
rlm@1: #define P2 dst1[0]
rlm@1: #define P3 dst1[1]
rlm@1: #define MUR interp_16_diff(c[1], c[5]) // top-right
rlm@1: #define MDR interp_16_diff(c[5], c[7]) // bottom-right
rlm@1: #define MDL interp_16_diff(c[7], c[3]) // bottom-left
rlm@1: #define MUL interp_16_diff(c[3], c[1]) // top-left
rlm@1: #define IC(p0) c[p0]
rlm@1: #define I11(p0, p1) interp_16_11(c[p0], c[p1])
rlm@1: #define I211(p0, p1, p2) interp_16_211(c[p0], c[p1], c[p2])
rlm@1: #define I31(p0, p1) interp_16_31(c[p0], c[p1])
rlm@1: #define I332(p0, p1, p2) interp_16_332(c[p0], c[p1], c[p2])
rlm@1: #define I431(p0, p1, p2) interp_16_431(c[p0], c[p1], c[p2])
rlm@1: #define I521(p0, p1, p2) interp_16_521(c[p0], c[p1], c[p2])
rlm@1: #define I53(p0, p1) interp_16_53(c[p0], c[p1])
rlm@1: #define I611(p0, p1, p2) interp_16_611(c[p0], c[p1], c[p2])
rlm@1: #define I71(p0, p1) interp_16_71(c[p0], c[p1])
rlm@1: #define I772(p0, p1, p2) interp_16_772(c[p0], c[p1], c[p2])
rlm@1: #define I97(p0, p1) interp_16_97(c[p0], c[p1])
rlm@1: #define I1411(p0, p1, p2) interp_16_1411(c[p0], c[p1], c[p2])
rlm@1: #define I151(p0, p1) interp_16_151(c[p0], c[p1])
rlm@1: 
rlm@1: 		switch (mask)
rlm@1: 		{
rlm@1: #include "hq2x.h"
rlm@1: 		}
rlm@1: 
rlm@1: #undef P0
rlm@1: #undef P1
rlm@1: #undef P2
rlm@1: #undef P3
rlm@1: #undef MUR
rlm@1: #undef MDR
rlm@1: #undef MDL
rlm@1: #undef MUL
rlm@1: #undef IC
rlm@1: #undef I11
rlm@1: #undef I211
rlm@1: #undef I31
rlm@1: #undef I332
rlm@1: #undef I431
rlm@1: #undef I521
rlm@1: #undef I53
rlm@1: #undef I611
rlm@1: #undef I71
rlm@1: #undef I772
rlm@1: #undef I97
rlm@1: #undef I1411
rlm@1: #undef I151
rlm@1: 
rlm@1: 		src0 += 1;
rlm@1: 		src1 += 1;
rlm@1: 		src2 += 1;
rlm@1: 		dst0 += 2;
rlm@1: 		dst1 += 2;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: static void hq2x_32_def(u32 *dst0, u32 *dst1, const u32 *src0, const u32 *src1, const u32 *src2, unsigned count)
rlm@1: {
rlm@1: 	unsigned i;
rlm@1: 
rlm@1: 	for (i = 0; i < count; ++i)
rlm@1: 	{
rlm@1: 		unsigned char mask;
rlm@1: 
rlm@1: 		u32 c[9];
rlm@1: 
rlm@1: 		c[1] = src0[0];
rlm@1: 		c[4] = src1[0];
rlm@1: 		c[7] = src2[0];
rlm@1: 
rlm@1: 		if (i > 0)
rlm@1: 		{
rlm@1: 			c[0] = src0[-1];
rlm@1: 			c[3] = src1[-1];
rlm@1: 			c[6] = src2[-1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[0] = c[1];
rlm@1: 			c[3] = c[4];
rlm@1: 			c[6] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		if (i < count - 1)
rlm@1: 		{
rlm@1: 			c[2] = src0[1];
rlm@1: 			c[5] = src1[1];
rlm@1: 			c[8] = src2[1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[2] = c[1];
rlm@1: 			c[5] = c[4];
rlm@1: 			c[8] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		mask = 0;
rlm@1: 
rlm@1: 		if (interp_32_diff(c[0], c[4]))
rlm@1: 			mask |= 1 << 0;
rlm@1: 		if (interp_32_diff(c[1], c[4]))
rlm@1: 			mask |= 1 << 1;
rlm@1: 		if (interp_32_diff(c[2], c[4]))
rlm@1: 			mask |= 1 << 2;
rlm@1: 		if (interp_32_diff(c[3], c[4]))
rlm@1: 			mask |= 1 << 3;
rlm@1: 		if (interp_32_diff(c[5], c[4]))
rlm@1: 			mask |= 1 << 4;
rlm@1: 		if (interp_32_diff(c[6], c[4]))
rlm@1: 			mask |= 1 << 5;
rlm@1: 		if (interp_32_diff(c[7], c[4]))
rlm@1: 			mask |= 1 << 6;
rlm@1: 		if (interp_32_diff(c[8], c[4]))
rlm@1: 			mask |= 1 << 7;
rlm@1: 
rlm@1: #define P0 dst0[0]
rlm@1: #define P1 dst0[1]
rlm@1: #define P2 dst1[0]
rlm@1: #define P3 dst1[1]
rlm@1: #define MUR interp_32_diff(c[1], c[5]) // top-right
rlm@1: #define MDR interp_32_diff(c[5], c[7]) // bottom-right
rlm@1: #define MDL interp_32_diff(c[7], c[3]) // bottom-left
rlm@1: #define MUL interp_32_diff(c[3], c[1]) // top-left
rlm@1: #define IC(p0) c[p0]
rlm@1: #define I11(p0, p1) interp_32_11(c[p0], c[p1])
rlm@1: #define I211(p0, p1, p2) interp_32_211(c[p0], c[p1], c[p2])
rlm@1: #define I31(p0, p1) interp_32_31(c[p0], c[p1])
rlm@1: #define I332(p0, p1, p2) interp_32_332(c[p0], c[p1], c[p2])
rlm@1: #define I431(p0, p1, p2) interp_32_431(c[p0], c[p1], c[p2])
rlm@1: #define I521(p0, p1, p2) interp_32_521(c[p0], c[p1], c[p2])
rlm@1: #define I53(p0, p1) interp_32_53(c[p0], c[p1])
rlm@1: #define I611(p0, p1, p2) interp_32_611(c[p0], c[p1], c[p2])
rlm@1: #define I71(p0, p1) interp_32_71(c[p0], c[p1])
rlm@1: #define I772(p0, p1, p2) interp_32_772(c[p0], c[p1], c[p2])
rlm@1: #define I97(p0, p1) interp_32_97(c[p0], c[p1])
rlm@1: #define I1411(p0, p1, p2) interp_32_1411(c[p0], c[p1], c[p2])
rlm@1: #define I151(p0, p1) interp_32_151(c[p0], c[p1])
rlm@1: 
rlm@1: 		switch (mask)
rlm@1: 		{
rlm@1: #include "hq2x.h"
rlm@1: 		}
rlm@1: 
rlm@1: #undef P0
rlm@1: #undef P1
rlm@1: #undef P2
rlm@1: #undef P3
rlm@1: #undef MUR
rlm@1: #undef MDR
rlm@1: #undef MDL
rlm@1: #undef MUL
rlm@1: #undef IC
rlm@1: #undef I11
rlm@1: #undef I211
rlm@1: #undef I31
rlm@1: #undef I332
rlm@1: #undef I431
rlm@1: #undef I521
rlm@1: #undef I53
rlm@1: #undef I611
rlm@1: #undef I71
rlm@1: #undef I772
rlm@1: #undef I97
rlm@1: #undef I1411
rlm@1: #undef I151
rlm@1: 
rlm@1: 		src0 += 1;
rlm@1: 		src1 += 1;
rlm@1: 		src2 += 1;
rlm@1: 		dst0 += 2;
rlm@1: 		dst1 += 2;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: /***************************************************************************/
rlm@1: /* HQ2xS C implementation */
rlm@1: 
rlm@1: /*
rlm@1:  * This effect is derived from the hq2x effect made by Maxim Stepin
rlm@1:  */
rlm@1: 
rlm@1: static void hq2xS_16_def(u16 *dst0, u16 *dst1, const u16 *src0, const u16 *src1, const u16 *src2, unsigned count)
rlm@1: {
rlm@1: 	unsigned i;
rlm@1: 
rlm@1: 	for (i = 0; i < count; ++i)
rlm@1: 	{
rlm@1: 		unsigned char mask;
rlm@1: 
rlm@1: 		u16 c[9];
rlm@1: 
rlm@1: 		c[1] = src0[0];
rlm@1: 		c[4] = src1[0];
rlm@1: 		c[7] = src2[0];
rlm@1: 
rlm@1: 		c[0] = src0[-1];
rlm@1: 		c[3] = src1[-1];
rlm@1: 		c[6] = src2[-1];
rlm@1: 
rlm@1: 		c[2] = src0[1];
rlm@1: 		c[5] = src1[1];
rlm@1: 		c[8] = src2[1];
rlm@1: 
rlm@1: 		mask = 0;
rlm@1: 
rlm@1: 		// hq2xS dynamic edge detection:
rlm@1: 		// simply comparing the center color against its surroundings will give bad results in many cases,
rlm@1: 		// so, instead, compare the center color relative to the max difference in brightness of this 3x3 block
rlm@1: 		int brightArray[9];
rlm@1: 		int maxBright = 0, minBright = 999999;
rlm@1: 		for (int j = 0; j < 9; j++)
rlm@1: 		{
rlm@1: 			int r, g, b;
rlm@1: 			if (interp_bits_per_pixel == 16)
rlm@1: 			{
rlm@1: 				b = (int)((c[j] & 0x1F)) << 3;
rlm@1: 				g = (int)((c[j] & 0x7E0)) >> 3;
rlm@1: 				r = (int)((c[j] & 0xF800)) >> 8;
rlm@1: 			}
rlm@1: 			else
rlm@1: 			{
rlm@1: 				b = (int)((c[j] & 0x1F)) << 3;
rlm@1: 				g = (int)((c[j] & 0x3E0)) >> 2;
rlm@1: 				r = (int)((c[j] & 0x7C00)) >> 7;
rlm@1: 			}
rlm@1: 			const int bright = r + r + r + g + g + g + b + b;
rlm@1: 			if (bright > maxBright) maxBright = bright;
rlm@1: 			if (bright < minBright) minBright = bright;
rlm@1: 
rlm@1: 			brightArray[j] = bright;
rlm@1: 		}
rlm@1: 		int diffBright = ((maxBright - minBright) * 7) >> 4;
rlm@1: 		if (diffBright > 7)
rlm@1: 		{
rlm@1: 		#define ABS(x) ((x) < 0 ? -(x) : (x))
rlm@1: 
rlm@1: 			const int centerBright = brightArray[4];
rlm@1: 			if (ABS(brightArray[0] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 0;
rlm@1: 			if (ABS(brightArray[1] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 1;
rlm@1: 			if (ABS(brightArray[2] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 2;
rlm@1: 			if (ABS(brightArray[3] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 3;
rlm@1: 			if (ABS(brightArray[5] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 4;
rlm@1: 			if (ABS(brightArray[6] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 5;
rlm@1: 			if (ABS(brightArray[7] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 6;
rlm@1: 			if (ABS(brightArray[8] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 7;
rlm@1: 		}
rlm@1: 
rlm@1: #define P0 dst0[0]
rlm@1: #define P1 dst0[1]
rlm@1: #define P2 dst1[0]
rlm@1: #define P3 dst1[1]
rlm@1: #define MUR false //(ABS(brightArray[1] - brightArray[5]) > diffBright) // top-right
rlm@1: #define MDR false //(ABS(brightArray[5] - brightArray[7]) > diffBright) // bottom-right
rlm@1: #define MDL false //(ABS(brightArray[7] - brightArray[3]) > diffBright) // bottom-left
rlm@1: #define MUL false //(ABS(brightArray[3] - brightArray[1]) > diffBright) // top-left
rlm@1: #define IC(p0) c[p0]
rlm@1: #define I11(p0, p1) interp_16_11(c[p0], c[p1])
rlm@1: #define I211(p0, p1, p2) interp_16_211(c[p0], c[p1], c[p2])
rlm@1: #define I31(p0, p1) interp_16_31(c[p0], c[p1])
rlm@1: #define I332(p0, p1, p2) interp_16_332(c[p0], c[p1], c[p2])
rlm@1: #define I431(p0, p1, p2) interp_16_431(c[p0], c[p1], c[p2])
rlm@1: #define I521(p0, p1, p2) interp_16_521(c[p0], c[p1], c[p2])
rlm@1: #define I53(p0, p1) interp_16_53(c[p0], c[p1])
rlm@1: #define I611(p0, p1, p2) interp_16_611(c[p0], c[p1], c[p2])
rlm@1: #define I71(p0, p1) interp_16_71(c[p0], c[p1])
rlm@1: #define I772(p0, p1, p2) interp_16_772(c[p0], c[p1], c[p2])
rlm@1: #define I97(p0, p1) interp_16_97(c[p0], c[p1])
rlm@1: #define I1411(p0, p1, p2) interp_16_1411(c[p0], c[p1], c[p2])
rlm@1: #define I151(p0, p1) interp_16_151(c[p0], c[p1])
rlm@1: 
rlm@1: 		switch (mask)
rlm@1: 		{
rlm@1: #include "hq2x.h"
rlm@1: 		}
rlm@1: 
rlm@1: #undef P0
rlm@1: #undef P1
rlm@1: #undef P2
rlm@1: #undef P3
rlm@1: #undef MUR
rlm@1: #undef MDR
rlm@1: #undef MDL
rlm@1: #undef MUL
rlm@1: #undef IC
rlm@1: #undef I11
rlm@1: #undef I211
rlm@1: #undef I31
rlm@1: #undef I332
rlm@1: #undef I431
rlm@1: #undef I521
rlm@1: #undef I53
rlm@1: #undef I611
rlm@1: #undef I71
rlm@1: #undef I772
rlm@1: #undef I97
rlm@1: #undef I1411
rlm@1: #undef I151
rlm@1: 
rlm@1: 		src0 += 1;
rlm@1: 		src1 += 1;
rlm@1: 		src2 += 1;
rlm@1: 		dst0 += 2;
rlm@1: 		dst1 += 2;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: static void hq2xS_32_def(u32 *dst0, u32 *dst1, const u32 *src0, const u32 *src1, const u32 *src2, unsigned count)
rlm@1: {
rlm@1: 	unsigned i;
rlm@1: 
rlm@1: 	for (i = 0; i < count; ++i)
rlm@1: 	{
rlm@1: 		unsigned char mask;
rlm@1: 
rlm@1: 		u32 c[9];
rlm@1: 
rlm@1: 		c[1] = src0[0];
rlm@1: 		c[4] = src1[0];
rlm@1: 		c[7] = src2[0];
rlm@1: 
rlm@1: 		c[0] = src0[-1];
rlm@1: 		c[3] = src1[-1];
rlm@1: 		c[6] = src2[-1];
rlm@1: 
rlm@1: 		c[2] = src0[1];
rlm@1: 		c[5] = src1[1];
rlm@1: 		c[8] = src2[1];
rlm@1: 
rlm@1: 		mask = 0;
rlm@1: 
rlm@1: 		// hq2xS dynamic edge detection:
rlm@1: 		// simply comparing the center color against its surroundings will give bad results in many cases,
rlm@1: 		// so, instead, compare the center color relative to the max difference in brightness of this 3x3 block
rlm@1: 		int brightArray[9];
rlm@1: 		int maxBright = 0, minBright = 999999;
rlm@1: 		for (int j = 0; j < 9; j++)
rlm@1: 		{
rlm@1: 			const int b		 = (int)((c[j] & 0xF8));
rlm@1: 			const int g		 = (int)((c[j] & 0xF800)) >> 8;
rlm@1: 			const int r		 = (int)((c[j] & 0xF80000)) >> 16;
rlm@1: 			const int bright = r + r + r + g + g + g + b + b;
rlm@1: 			if (bright > maxBright) maxBright = bright;
rlm@1: 			if (bright < minBright) minBright = bright;
rlm@1: 
rlm@1: 			brightArray[j] = bright;
rlm@1: 		}
rlm@1: 		int diffBright = ((maxBright - minBright) * 7) >> 4;
rlm@1: 		if (diffBright > 7)
rlm@1: 		{
rlm@1: 		#define ABS(x) ((x) < 0 ? -(x) : (x))
rlm@1: 
rlm@1: 			const int centerBright = brightArray[4];
rlm@1: 			if (ABS(brightArray[0] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 0;
rlm@1: 			if (ABS(brightArray[1] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 1;
rlm@1: 			if (ABS(brightArray[2] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 2;
rlm@1: 			if (ABS(brightArray[3] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 3;
rlm@1: 			if (ABS(brightArray[5] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 4;
rlm@1: 			if (ABS(brightArray[6] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 5;
rlm@1: 			if (ABS(brightArray[7] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 6;
rlm@1: 			if (ABS(brightArray[8] - centerBright) > diffBright)
rlm@1: 				mask |= 1 << 7;
rlm@1: 		}
rlm@1: 
rlm@1: #define P0 dst0[0]
rlm@1: #define P1 dst0[1]
rlm@1: #define P2 dst1[0]
rlm@1: #define P3 dst1[1]
rlm@1: #define MUR false //(ABS(brightArray[1] - brightArray[5]) > diffBright) // top-right
rlm@1: #define MDR false //(ABS(brightArray[5] - brightArray[7]) > diffBright) // bottom-right
rlm@1: #define MDL false //(ABS(brightArray[7] - brightArray[3]) > diffBright) // bottom-left
rlm@1: #define MUL false //(ABS(brightArray[3] - brightArray[1]) > diffBright) // top-left
rlm@1: #define IC(p0) c[p0]
rlm@1: #define I11(p0, p1) interp_32_11(c[p0], c[p1])
rlm@1: #define I211(p0, p1, p2) interp_32_211(c[p0], c[p1], c[p2])
rlm@1: #define I31(p0, p1) interp_32_31(c[p0], c[p1])
rlm@1: #define I332(p0, p1, p2) interp_32_332(c[p0], c[p1], c[p2])
rlm@1: #define I431(p0, p1, p2) interp_32_431(c[p0], c[p1], c[p2])
rlm@1: #define I521(p0, p1, p2) interp_32_521(c[p0], c[p1], c[p2])
rlm@1: #define I53(p0, p1) interp_32_53(c[p0], c[p1])
rlm@1: #define I611(p0, p1, p2) interp_32_611(c[p0], c[p1], c[p2])
rlm@1: #define I71(p0, p1) interp_32_71(c[p0], c[p1])
rlm@1: #define I772(p0, p1, p2) interp_32_772(c[p0], c[p1], c[p2])
rlm@1: #define I97(p0, p1) interp_32_97(c[p0], c[p1])
rlm@1: #define I1411(p0, p1, p2) interp_32_1411(c[p0], c[p1], c[p2])
rlm@1: #define I151(p0, p1) interp_32_151(c[p0], c[p1])
rlm@1: 
rlm@1: 		switch (mask)
rlm@1: 		{
rlm@1: #include "hq2x.h"
rlm@1: 		}
rlm@1: 
rlm@1: #undef P0
rlm@1: #undef P1
rlm@1: #undef P2
rlm@1: #undef P3
rlm@1: #undef MUR
rlm@1: #undef MDR
rlm@1: #undef MDL
rlm@1: #undef MUL
rlm@1: #undef IC
rlm@1: #undef I11
rlm@1: #undef I211
rlm@1: #undef I31
rlm@1: #undef I332
rlm@1: #undef I431
rlm@1: #undef I521
rlm@1: #undef I53
rlm@1: #undef I611
rlm@1: #undef I71
rlm@1: #undef I772
rlm@1: #undef I97
rlm@1: #undef I1411
rlm@1: #undef I151
rlm@1: 
rlm@1: 		src0 += 1;
rlm@1: 		src1 += 1;
rlm@1: 		src2 += 1;
rlm@1: 		dst0 += 2;
rlm@1: 		dst1 += 2;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: /***************************************************************************/
rlm@1: /* LQ2x C implementation */
rlm@1: 
rlm@1: /*
rlm@1:  * This effect is derived from the hq2x effect made by Maxim Stepin
rlm@1:  */
rlm@1: 
rlm@1: static void lq2x_16_def(u16 *dst0, u16 *dst1, const u16 *src0, const u16 *src1, const u16 *src2, unsigned count)
rlm@1: {
rlm@1: 	unsigned i;
rlm@1: 
rlm@1: 	for (i = 0; i < count; ++i)
rlm@1: 	{
rlm@1: 		unsigned char mask;
rlm@1: 
rlm@1: 		u16 c[9];
rlm@1: 
rlm@1: 		c[1] = src0[0];
rlm@1: 		c[4] = src1[0];
rlm@1: 		c[7] = src2[0];
rlm@1: 
rlm@1: 		if (i > 0)
rlm@1: 		{
rlm@1: 			c[0] = src0[-1];
rlm@1: 			c[3] = src1[-1];
rlm@1: 			c[6] = src2[-1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[0] = c[1];
rlm@1: 			c[3] = c[4];
rlm@1: 			c[6] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		if (i < count - 1)
rlm@1: 		{
rlm@1: 			c[2] = src0[1];
rlm@1: 			c[5] = src1[1];
rlm@1: 			c[8] = src2[1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[2] = c[1];
rlm@1: 			c[5] = c[4];
rlm@1: 			c[8] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		mask = 0;
rlm@1: 
rlm@1: 		if (c[0] != c[4])
rlm@1: 			mask |= 1 << 0;
rlm@1: 		if (c[1] != c[4])
rlm@1: 			mask |= 1 << 1;
rlm@1: 		if (c[2] != c[4])
rlm@1: 			mask |= 1 << 2;
rlm@1: 		if (c[3] != c[4])
rlm@1: 			mask |= 1 << 3;
rlm@1: 		if (c[5] != c[4])
rlm@1: 			mask |= 1 << 4;
rlm@1: 		if (c[6] != c[4])
rlm@1: 			mask |= 1 << 5;
rlm@1: 		if (c[7] != c[4])
rlm@1: 			mask |= 1 << 6;
rlm@1: 		if (c[8] != c[4])
rlm@1: 			mask |= 1 << 7;
rlm@1: 
rlm@1: #define P0 dst0[0]
rlm@1: #define P1 dst0[1]
rlm@1: #define P2 dst1[0]
rlm@1: #define P3 dst1[1]
rlm@1: #define MUR (c[1] != c[5])
rlm@1: #define MDR (c[5] != c[7])
rlm@1: #define MDL (c[7] != c[3])
rlm@1: #define MUL (c[3] != c[1])
rlm@1: #define IC(p0) c[p0]
rlm@1: #define I11(p0, p1) interp_16_11(c[p0], c[p1])
rlm@1: #define I211(p0, p1, p2) interp_16_211(c[p0], c[p1], c[p2])
rlm@1: #define I31(p0, p1) interp_16_31(c[p0], c[p1])
rlm@1: #define I332(p0, p1, p2) interp_16_332(c[p0], c[p1], c[p2])
rlm@1: #define I431(p0, p1, p2) interp_16_431(c[p0], c[p1], c[p2])
rlm@1: #define I521(p0, p1, p2) interp_16_521(c[p0], c[p1], c[p2])
rlm@1: #define I53(p0, p1) interp_16_53(c[p0], c[p1])
rlm@1: #define I611(p0, p1, p2) interp_16_611(c[p0], c[p1], c[p2])
rlm@1: #define I71(p0, p1) interp_16_71(c[p0], c[p1])
rlm@1: #define I772(p0, p1, p2) interp_16_772(c[p0], c[p1], c[p2])
rlm@1: #define I97(p0, p1) interp_16_97(c[p0], c[p1])
rlm@1: #define I1411(p0, p1, p2) interp_16_1411(c[p0], c[p1], c[p2])
rlm@1: #define I151(p0, p1) interp_16_151(c[p0], c[p1])
rlm@1: 
rlm@1: 		switch (mask)
rlm@1: 		{
rlm@1: #include "lq2x.h"
rlm@1: 		}
rlm@1: 
rlm@1: #undef P0
rlm@1: #undef P1
rlm@1: #undef P2
rlm@1: #undef P3
rlm@1: #undef MUR
rlm@1: #undef MDR
rlm@1: #undef MDL
rlm@1: #undef MUL
rlm@1: #undef IC
rlm@1: #undef I11
rlm@1: #undef I211
rlm@1: #undef I31
rlm@1: #undef I332
rlm@1: #undef I431
rlm@1: #undef I521
rlm@1: #undef I53
rlm@1: #undef I611
rlm@1: #undef I71
rlm@1: #undef I772
rlm@1: #undef I97
rlm@1: #undef I1411
rlm@1: #undef I151
rlm@1: 
rlm@1: 		src0 += 1;
rlm@1: 		src1 += 1;
rlm@1: 		src2 += 1;
rlm@1: 		dst0 += 2;
rlm@1: 		dst1 += 2;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: static void lq2x_32_def(u32 *dst0, u32 *dst1, const u32 *src0, const u32 *src1, const u32 *src2, unsigned count)
rlm@1: {
rlm@1: 	unsigned i;
rlm@1: 
rlm@1: 	for (i = 0; i < count; ++i)
rlm@1: 	{
rlm@1: 		unsigned char mask;
rlm@1: 
rlm@1: 		u32 c[9];
rlm@1: 
rlm@1: 		c[1] = src0[0];
rlm@1: 		c[4] = src1[0];
rlm@1: 		c[7] = src2[0];
rlm@1: 
rlm@1: 		if (i > 0)
rlm@1: 		{
rlm@1: 			c[0] = src0[-1];
rlm@1: 			c[3] = src1[-1];
rlm@1: 			c[6] = src2[-1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[0] = c[1];
rlm@1: 			c[3] = c[4];
rlm@1: 			c[6] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		if (i < count - 1)
rlm@1: 		{
rlm@1: 			c[2] = src0[1];
rlm@1: 			c[5] = src1[1];
rlm@1: 			c[8] = src2[1];
rlm@1: 		}
rlm@1: 		else
rlm@1: 		{
rlm@1: 			c[2] = c[1];
rlm@1: 			c[5] = c[4];
rlm@1: 			c[8] = c[7];
rlm@1: 		}
rlm@1: 
rlm@1: 		mask = 0;
rlm@1: 
rlm@1: 		if (c[0] != c[4])
rlm@1: 			mask |= 1 << 0;
rlm@1: 		if (c[1] != c[4])
rlm@1: 			mask |= 1 << 1;
rlm@1: 		if (c[2] != c[4])
rlm@1: 			mask |= 1 << 2;
rlm@1: 		if (c[3] != c[4])
rlm@1: 			mask |= 1 << 3;
rlm@1: 		if (c[5] != c[4])
rlm@1: 			mask |= 1 << 4;
rlm@1: 		if (c[6] != c[4])
rlm@1: 			mask |= 1 << 5;
rlm@1: 		if (c[7] != c[4])
rlm@1: 			mask |= 1 << 6;
rlm@1: 		if (c[8] != c[4])
rlm@1: 			mask |= 1 << 7;
rlm@1: 
rlm@1: #define P0 dst0[0]
rlm@1: #define P1 dst0[1]
rlm@1: #define P2 dst1[0]
rlm@1: #define P3 dst1[1]
rlm@1: #define MUR (c[1] != c[5])
rlm@1: #define MDR (c[5] != c[7])
rlm@1: #define MDL (c[7] != c[3])
rlm@1: #define MUL (c[3] != c[1])
rlm@1: #define IC(p0) c[p0]
rlm@1: #define I11(p0, p1) interp_32_11(c[p0], c[p1])
rlm@1: #define I211(p0, p1, p2) interp_32_211(c[p0], c[p1], c[p2])
rlm@1: #define I31(p0, p1) interp_32_31(c[p0], c[p1])
rlm@1: #define I332(p0, p1, p2) interp_32_332(c[p0], c[p1], c[p2])
rlm@1: #define I431(p0, p1, p2) interp_32_431(c[p0], c[p1], c[p2])
rlm@1: #define I521(p0, p1, p2) interp_32_521(c[p0], c[p1], c[p2])
rlm@1: #define I53(p0, p1) interp_32_53(c[p0], c[p1])
rlm@1: #define I611(p0, p1, p2) interp_32_611(c[p0], c[p1], c[p2])
rlm@1: #define I71(p0, p1) interp_32_71(c[p0], c[p1])
rlm@1: #define I772(p0, p1, p2) interp_32_772(c[p0], c[p1], c[p2])
rlm@1: #define I97(p0, p1) interp_32_97(c[p0], c[p1])
rlm@1: #define I1411(p0, p1, p2) interp_32_1411(c[p0], c[p1], c[p2])
rlm@1: #define I151(p0, p1) interp_32_151(c[p0], c[p1])
rlm@1: 
rlm@1: 		switch (mask)
rlm@1: 		{
rlm@1: #include "lq2x.h"
rlm@1: 		}
rlm@1: 
rlm@1: #undef P0
rlm@1: #undef P1
rlm@1: #undef P2
rlm@1: #undef P3
rlm@1: #undef MUR
rlm@1: #undef MDR
rlm@1: #undef MDL
rlm@1: #undef MUL
rlm@1: #undef IC
rlm@1: #undef I11
rlm@1: #undef I211
rlm@1: #undef I31
rlm@1: #undef I332
rlm@1: #undef I431
rlm@1: #undef I521
rlm@1: #undef I53
rlm@1: #undef I611
rlm@1: #undef I71
rlm@1: #undef I772
rlm@1: #undef I97
rlm@1: #undef I1411
rlm@1: #undef I151
rlm@1: 
rlm@1: 		src0 += 1;
rlm@1: 		src1 += 1;
rlm@1: 		src2 += 1;
rlm@1: 		dst0 += 2;
rlm@1: 		dst1 += 2;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: void hq2x(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:           u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u16 *dst0 = (u16 *)dstPtr;
rlm@1: 	u16 *dst1 = dst0 + (dstPitch >> 1);
rlm@1: 
rlm@1: 	u16 *src0 = (u16 *)srcPtr;
rlm@1: 	u16 *src1 = src0 + (srcPitch >> 1);
rlm@1: 	u16 *src2 = src1 + (srcPitch >> 1);
rlm@1: 
rlm@1: 	hq2x_16_def(dst0, dst1, src0, src0, src1, width);
rlm@1: 
rlm@1: 	int count = height;
rlm@1: 
rlm@1: 	count -= 2;
rlm@1: 	while (count)
rlm@1: 	{
rlm@1: 		dst0 += dstPitch;
rlm@1: 		dst1 += dstPitch;
rlm@1: 		hq2x_16_def(dst0, dst1, src0, src1, src2, width);
rlm@1: 		src0  = src1;
rlm@1: 		src1  = src2;
rlm@1: 		src2 += srcPitch >> 1;
rlm@1: 		--count;
rlm@1: 	}
rlm@1: 	dst0 += dstPitch;
rlm@1: 	dst1 += dstPitch;
rlm@1: 	hq2x_16_def(dst0, dst1, src0, src1, src1, width);
rlm@1: }
rlm@1: 
rlm@1: void hq2x32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:             u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u32 *dst0 = (u32 *)dstPtr;
rlm@1: 	u32 *dst1 = dst0 + (dstPitch >> 2);
rlm@1: 
rlm@1: 	u32 *src0 = (u32 *)srcPtr;
rlm@1: 	u32 *src1 = src0 + (srcPitch >> 2);
rlm@1: 	u32 *src2 = src1 + (srcPitch >> 2);
rlm@1: 	hq2x_32_def(dst0, dst1, src0, src0, src1, width);
rlm@1: 
rlm@1: 	int count = height;
rlm@1: 
rlm@1: 	count -= 2;
rlm@1: 	while (count)
rlm@1: 	{
rlm@1: 		dst0 += dstPitch >> 1;
rlm@1: 		dst1 += dstPitch >> 1;
rlm@1: 		hq2x_32_def(dst0, dst1, src0, src1, src2, width);
rlm@1: 		src0  = src1;
rlm@1: 		src1  = src2;
rlm@1: 		src2 += srcPitch >> 2;
rlm@1: 		--count;
rlm@1: 	}
rlm@1: 	dst0 += dstPitch >> 1;
rlm@1: 	dst1 += dstPitch >> 1;
rlm@1: 	hq2x_32_def(dst0, dst1, src0, src1, src1, width);
rlm@1: }
rlm@1: 
rlm@1: void hq2xS(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:            u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u16 *dst0 = (u16 *)dstPtr;
rlm@1: 	u16 *dst1 = dst0 + (dstPitch >> 1);
rlm@1: 
rlm@1: 	u16 *src0 = (u16 *)srcPtr;
rlm@1: 	u16 *src1 = src0 + (srcPitch >> 1);
rlm@1: 	u16 *src2 = src1 + (srcPitch >> 1);
rlm@1: 
rlm@1: 	hq2xS_16_def(dst0, dst1, src0, src0, src1, width);
rlm@1: 
rlm@1: 	int count = height;
rlm@1: 
rlm@1: 	count -= 2;
rlm@1: 	while (count)
rlm@1: 	{
rlm@1: 		dst0 += dstPitch;
rlm@1: 		dst1 += dstPitch;
rlm@1: 		hq2xS_16_def(dst0, dst1, src0, src1, src2, width);
rlm@1: 		src0  = src1;
rlm@1: 		src1  = src2;
rlm@1: 		src2 += srcPitch >> 1;
rlm@1: 		--count;
rlm@1: 	}
rlm@1: 	dst0 += dstPitch;
rlm@1: 	dst1 += dstPitch;
rlm@1: 	hq2xS_16_def(dst0, dst1, src0, src1, src1, width);
rlm@1: }
rlm@1: 
rlm@1: void hq2xS32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:              u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u32 *dst0 = (u32 *)dstPtr;
rlm@1: 	u32 *dst1 = dst0 + (dstPitch >> 2);
rlm@1: 
rlm@1: 	u32 *src0 = (u32 *)srcPtr;
rlm@1: 	u32 *src1 = src0 + (srcPitch >> 2);
rlm@1: 	u32 *src2 = src1 + (srcPitch >> 2);
rlm@1: 	hq2xS_32_def(dst0, dst1, src0, src0, src1, width);
rlm@1: 
rlm@1: 	int count = height;
rlm@1: 
rlm@1: 	count -= 2;
rlm@1: 	while (count)
rlm@1: 	{
rlm@1: 		dst0 += dstPitch >> 1;
rlm@1: 		dst1 += dstPitch >> 1;
rlm@1: 		hq2xS_32_def(dst0, dst1, src0, src1, src2, width);
rlm@1: 		src0  = src1;
rlm@1: 		src1  = src2;
rlm@1: 		src2 += srcPitch >> 2;
rlm@1: 		--count;
rlm@1: 	}
rlm@1: 	dst0 += dstPitch >> 1;
rlm@1: 	dst1 += dstPitch >> 1;
rlm@1: 	hq2xS_32_def(dst0, dst1, src0, src1, src1, width);
rlm@1: }
rlm@1: 
rlm@1: void lq2x(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:           u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u16 *dst0 = (u16 *)dstPtr;
rlm@1: 	u16 *dst1 = dst0 + (dstPitch >> 1);
rlm@1: 
rlm@1: 	u16 *src0 = (u16 *)srcPtr;
rlm@1: 	u16 *src1 = src0 + (srcPitch >> 1);
rlm@1: 	u16 *src2 = src1 + (srcPitch >> 1);
rlm@1: 
rlm@1: 	lq2x_16_def(dst0, dst1, src0, src0, src1, width);
rlm@1: 
rlm@1: 	int count = height;
rlm@1: 
rlm@1: 	count -= 2;
rlm@1: 	while (count)
rlm@1: 	{
rlm@1: 		dst0 += dstPitch;
rlm@1: 		dst1 += dstPitch;
rlm@1: 		lq2x_16_def(dst0, dst1, src0, src1, src2, width);
rlm@1: 		src0  = src1;
rlm@1: 		src1  = src2;
rlm@1: 		src2 += srcPitch >> 1;
rlm@1: 		--count;
rlm@1: 	}
rlm@1: 	dst0 += dstPitch;
rlm@1: 	dst1 += dstPitch;
rlm@1: 	lq2x_16_def(dst0, dst1, src0, src1, src1, width);
rlm@1: }
rlm@1: 
rlm@1: void lq2x32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:             u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u32 *dst0 = (u32 *)dstPtr;
rlm@1: 	u32 *dst1 = dst0 + (dstPitch >> 2);
rlm@1: 
rlm@1: 	u32 *src0 = (u32 *)srcPtr;
rlm@1: 	u32 *src1 = src0 + (srcPitch >> 2);
rlm@1: 	u32 *src2 = src1 + (srcPitch >> 2);
rlm@1: 	lq2x_32_def(dst0, dst1, src0, src0, src1, width);
rlm@1: 
rlm@1: 	int count = height;
rlm@1: 
rlm@1: 	count -= 2;
rlm@1: 	while (count)
rlm@1: 	{
rlm@1: 		dst0 += dstPitch >> 1;
rlm@1: 		dst1 += dstPitch >> 1;
rlm@1: 		lq2x_32_def(dst0, dst1, src0, src1, src2, width);
rlm@1: 		src0  = src1;
rlm@1: 		src1  = src2;
rlm@1: 		src2 += srcPitch >> 2;
rlm@1: 		--count;
rlm@1: 	}
rlm@1: 	dst0 += dstPitch >> 1;
rlm@1: 	dst1 += dstPitch >> 1;
rlm@1: 	lq2x_32_def(dst0, dst1, src0, src1, src1, width);
rlm@1: }
rlm@1: 
rlm@1: void hq2x_init(unsigned bits_per_pixel)
rlm@1: {
rlm@1: 	interp_set(bits_per_pixel);
rlm@1: }
rlm@1: