1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/filters/admame.cpp	Sat Mar 03 10:31:27 2012 -0600
     1.3 @@ -0,0 +1,1036 @@
     1.4 +/*
     1.5 + * This file is part of the Advance project.
     1.6 + *
     1.7 + * Copyright (C) 1999-2002 Andrea Mazzoleni
     1.8 + *
     1.9 + * This program is free software; you can redistribute it and/or modify
    1.10 + * it under the terms of the GNU General Public License as published by
    1.11 + * the Free Software Foundation; either version 2 of the License, or
    1.12 + * (at your option) any later version.
    1.13 + *
    1.14 + * This program is distributed in the hope that it will be useful,
    1.15 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
    1.16 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    1.17 + * GNU General Public License for more details.
    1.18 + *
    1.19 + * You should have received a copy of the GNU General Public License
    1.20 + * along with this program; if not, write to the Free Software
    1.21 + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
    1.22 + *
    1.23 + * In addition, as a special exception, Andrea Mazzoleni
    1.24 + * gives permission to link the code of this program with
    1.25 + * the MAME library (or with modified versions of MAME that use the
    1.26 + * same license as MAME), and distribute linked combinations including
    1.27 + * the two.  You must obey the GNU General Public License in all
    1.28 + * respects for all of the code used other than MAME.  If you modify
    1.29 + * this file, you may extend this exception to your version of the
    1.30 + * file, but you are not obligated to do so.  If you do not wish to
    1.31 + * do so, delete this exception statement from your version.
    1.32 + */
    1.33 +
    1.34 +/*
    1.35 + * Alternatively at the previous license terms, you are allowed to use this
    1.36 + * code in your program with these conditions:
    1.37 + * - the program is not used in commercial activities.
    1.38 + * - the whole source code of the program is released with the binary.
    1.39 + */
    1.40 +
    1.41 +#include "../Port.h"
    1.42 +
    1.43 +#ifdef MMX
    1.44 +extern "C" bool cpu_mmx;
    1.45 +#endif
    1.46 +
    1.47 +static void internal_scale2x_16_def(u16 *dst, const u16 *src0, const u16 *src1, const u16 *src2, unsigned count)
    1.48 +{
    1.49 +	/* first pixel */
    1.50 +	dst[0] = src1[0];
    1.51 +	if (src1[1] == src0[0] && src2[0] != src0[0])
    1.52 +		dst[1] = src0[0];
    1.53 +	else
    1.54 +		dst[1] = src1[0];
    1.55 +	++src0;
    1.56 +	++src1;
    1.57 +	++src2;
    1.58 +	dst += 2;
    1.59 +
    1.60 +	/* central pixels */
    1.61 +	count -= 2;
    1.62 +	while (count)
    1.63 +	{
    1.64 +		if (src0[0] != src2[0] && src1[-1] != src1[1])
    1.65 +		{
    1.66 +			dst[0] = src1[-1] == src0[0] ? src0[0] : src1[0];
    1.67 +			dst[1] = src1[1] == src0[0] ? src0[0] : src1[0];
    1.68 +		}
    1.69 +		else
    1.70 +		{
    1.71 +			dst[0] = src1[0];
    1.72 +			dst[1] = src1[0];
    1.73 +		}
    1.74 +
    1.75 +		++src0;
    1.76 +		++src1;
    1.77 +		++src2;
    1.78 +		dst += 2;
    1.79 +		--count;
    1.80 +	}
    1.81 +
    1.82 +	/* last pixel */
    1.83 +	if (src1[-1] == src0[0] && src2[0] != src0[0])
    1.84 +		dst[0] = src0[0];
    1.85 +	else
    1.86 +		dst[0] = src1[0];
    1.87 +	dst[1] = src1[0];
    1.88 +}
    1.89 +
    1.90 +static void internal_scale2x_32_def(u32 *dst,
    1.91 +                                    const u32 *src0,
    1.92 +                                    const u32 *src1,
    1.93 +                                    const u32 *src2,
    1.94 +                                    unsigned count)
    1.95 +{
    1.96 +	/* first pixel */
    1.97 +	dst[0] = src1[0];
    1.98 +	if (src1[1] == src0[0] && src2[0] != src0[0])
    1.99 +		dst[1] = src0[0];
   1.100 +	else
   1.101 +		dst[1] = src1[0];
   1.102 +	++src0;
   1.103 +	++src1;
   1.104 +	++src2;
   1.105 +	dst += 2;
   1.106 +
   1.107 +	/* central pixels */
   1.108 +	count -= 2;
   1.109 +	while (count)
   1.110 +	{
   1.111 +		if (src0[0] != src2[0] && src1[-1] != src1[1])
   1.112 +		{
   1.113 +			dst[0] = src1[-1] == src0[0] ? src0[0] : src1[0];
   1.114 +			dst[1] = src1[1] == src0[0] ? src0[0] : src1[0];
   1.115 +		}
   1.116 +		else
   1.117 +		{
   1.118 +			dst[0] = src1[0];
   1.119 +			dst[1] = src1[0];
   1.120 +		}
   1.121 +
   1.122 +		++src0;
   1.123 +		++src1;
   1.124 +		++src2;
   1.125 +		dst += 2;
   1.126 +		--count;
   1.127 +	}
   1.128 +
   1.129 +	/* last pixel */
   1.130 +	if (src1[-1] == src0[0] && src2[0] != src0[0])
   1.131 +		dst[0] = src0[0];
   1.132 +	else
   1.133 +		dst[0] = src1[0];
   1.134 +	dst[1] = src1[0];
   1.135 +}
   1.136 +
   1.137 +#ifdef MMX
   1.138 +static void internal_scale2x_16_mmx_single(u16 *dst, const u16 *src0, const u16 *src1, const u16 *src2, unsigned count)
   1.139 +{
   1.140 +	/* always do the first and last run */
   1.141 +	count -= 2 * 4;
   1.142 +
   1.143 +#ifdef __GNUC__
   1.144 +	__asm__ __volatile__ (
   1.145 +	    /* first run */
   1.146 +	    /* set the current, current_pre, current_next registers */
   1.147 +	    "movq 0(%1), %%mm0\n"
   1.148 +	    "movq 0(%1),%%mm7\n"
   1.149 +	    "movq 8(%1),%%mm1\n"
   1.150 +	    "psllq $48,%%mm0\n"
   1.151 +	    "psllq $48,%%mm1\n"
   1.152 +	    "psrlq $48, %%mm0\n"
   1.153 +	    "movq %%mm7,%%mm2\n"
   1.154 +	    "movq %%mm7,%%mm3\n"
   1.155 +	    "psllq $16,%%mm2\n"
   1.156 +	    "psrlq $16,%%mm3\n"
   1.157 +	    "por %%mm2,%%mm0\n"
   1.158 +	    "por %%mm3,%%mm1\n"
   1.159 +
   1.160 +	    /* current_upper */
   1.161 +	    "movq (%0),%%mm6\n"
   1.162 +
   1.163 +	    /* compute the upper-left pixel for dst on %%mm2 */
   1.164 +	    /* compute the upper-right pixel for dst on %%mm4 */
   1.165 +	    "movq %%mm0,%%mm2\n"
   1.166 +	    "movq %%mm1,%%mm4\n"
   1.167 +	    "movq %%mm0,%%mm3\n"
   1.168 +	    "movq %%mm1,%%mm5\n"
   1.169 +	    "pcmpeqw %%mm6,%%mm2\n"
   1.170 +	    "pcmpeqw %%mm6,%%mm4\n"
   1.171 +	    "pcmpeqw (%2),%%mm3\n"
   1.172 +	    "pcmpeqw (%2),%%mm5\n"
   1.173 +	    "pandn %%mm2,%%mm3\n"
   1.174 +	    "pandn %%mm4,%%mm5\n"
   1.175 +	    "movq %%mm0,%%mm2\n"
   1.176 +	    "movq %%mm1,%%mm4\n"
   1.177 +	    "pcmpeqw %%mm1,%%mm2\n"
   1.178 +	    "pcmpeqw %%mm0,%%mm4\n"
   1.179 +	    "pandn %%mm3,%%mm2\n"
   1.180 +	    "pandn %%mm5,%%mm4\n"
   1.181 +	    "movq %%mm2,%%mm3\n"
   1.182 +	    "movq %%mm4,%%mm5\n"
   1.183 +	    "pand %%mm6,%%mm2\n"
   1.184 +	    "pand %%mm6,%%mm4\n"
   1.185 +	    "pandn %%mm7,%%mm3\n"
   1.186 +	    "pandn %%mm7,%%mm5\n"
   1.187 +	    "por %%mm3,%%mm2\n"
   1.188 +	    "por %%mm5,%%mm4\n"
   1.189 +
   1.190 +	    /* set *dst */
   1.191 +	    "movq %%mm2,%%mm3\n"
   1.192 +	    "punpcklwd %%mm4,%%mm2\n"
   1.193 +	    "punpckhwd %%mm4,%%mm3\n"
   1.194 +	    "movq %%mm2,(%3)\n"
   1.195 +	    "movq %%mm3,8(%3)\n"
   1.196 +
   1.197 +	    /* next */
   1.198 +	    "addl $8,%0\n"
   1.199 +	    "addl $8,%1\n"
   1.200 +	    "addl $8,%2\n"
   1.201 +	    "addl $16,%3\n"
   1.202 +
   1.203 +	    /* central runs */
   1.204 +	    "shrl $2,%4\n"
   1.205 +	    "jz 1f\n"
   1.206 +
   1.207 +	    "0:\n"
   1.208 +
   1.209 +	    /* set the current, current_pre, current_next registers */
   1.210 +	    "movq -8(%1),%%mm0\n"
   1.211 +	    "movq (%1),%%mm7\n"
   1.212 +	    "movq 8(%1),%%mm1\n"
   1.213 +	    "psrlq $48,%%mm0\n"
   1.214 +	    "psllq $48,%%mm1\n"
   1.215 +	    "movq %%mm7,%%mm2\n"
   1.216 +	    "movq %%mm7,%%mm3\n"
   1.217 +	    "psllq $16,%%mm2\n"
   1.218 +	    "psrlq $16,%%mm3\n"
   1.219 +	    "por %%mm2,%%mm0\n"
   1.220 +	    "por %%mm3,%%mm1\n"
   1.221 +
   1.222 +	    /* current_upper */
   1.223 +	    "movq (%0),%%mm6\n"
   1.224 +
   1.225 +	    /* compute the upper-left pixel for dst on %%mm2 */
   1.226 +	    /* compute the upper-right pixel for dst on %%mm4 */
   1.227 +	    "movq %%mm0,%%mm2\n"
   1.228 +	    "movq %%mm1,%%mm4\n"
   1.229 +	    "movq %%mm0,%%mm3\n"
   1.230 +	    "movq %%mm1,%%mm5\n"
   1.231 +	    "pcmpeqw %%mm6,%%mm2\n"
   1.232 +	    "pcmpeqw %%mm6,%%mm4\n"
   1.233 +	    "pcmpeqw (%2),%%mm3\n"
   1.234 +	    "pcmpeqw (%2),%%mm5\n"
   1.235 +	    "pandn %%mm2,%%mm3\n"
   1.236 +	    "pandn %%mm4,%%mm5\n"
   1.237 +	    "movq %%mm0,%%mm2\n"
   1.238 +	    "movq %%mm1,%%mm4\n"
   1.239 +	    "pcmpeqw %%mm1,%%mm2\n"
   1.240 +	    "pcmpeqw %%mm0,%%mm4\n"
   1.241 +	    "pandn %%mm3,%%mm2\n"
   1.242 +	    "pandn %%mm5,%%mm4\n"
   1.243 +	    "movq %%mm2,%%mm3\n"
   1.244 +	    "movq %%mm4,%%mm5\n"
   1.245 +	    "pand %%mm6,%%mm2\n"
   1.246 +	    "pand %%mm6,%%mm4\n"
   1.247 +	    "pandn %%mm7,%%mm3\n"
   1.248 +	    "pandn %%mm7,%%mm5\n"
   1.249 +	    "por %%mm3,%%mm2\n"
   1.250 +	    "por %%mm5,%%mm4\n"
   1.251 +
   1.252 +	    /* set *dst */
   1.253 +	    "movq %%mm2,%%mm3\n"
   1.254 +	    "punpcklwd %%mm4,%%mm2\n"
   1.255 +	    "punpckhwd %%mm4,%%mm3\n"
   1.256 +	    "movq %%mm2,(%3)\n"
   1.257 +	    "movq %%mm3,8(%3)\n"
   1.258 +
   1.259 +	    /* next */
   1.260 +	    "addl $8,%0\n"
   1.261 +	    "addl $8,%1\n"
   1.262 +	    "addl $8,%2\n"
   1.263 +	    "addl $16,%3\n"
   1.264 +
   1.265 +	    "decl %4\n"
   1.266 +	    "jnz 0b\n"
   1.267 +	    "1:\n"
   1.268 +
   1.269 +	    /* final run */
   1.270 +	    /* set the current, current_pre, current_next registers */
   1.271 +	    "movq (%1),%%mm1\n"
   1.272 +	    "movq (%1),%%mm7\n"
   1.273 +	    "movq -8(%1),%%mm0\n"
   1.274 +	    "psrlq $48,%%mm1\n"
   1.275 +	    "psrlq $48,%%mm0\n"
   1.276 +	    "psllq $48,%%mm1\n"
   1.277 +	    "movq %%mm7,%%mm2\n"
   1.278 +	    "movq %%mm7,%%mm3\n"
   1.279 +	    "psllq $16,%%mm2\n"
   1.280 +	    "psrlq $16,%%mm3\n"
   1.281 +	    "por %%mm2,%%mm0\n"
   1.282 +	    "por %%mm3,%%mm1\n"
   1.283 +
   1.284 +	    /* current_upper */
   1.285 +	    "movq (%0),%%mm6\n"
   1.286 +
   1.287 +	    /* compute the upper-left pixel for dst on %%mm2 */
   1.288 +	    /* compute the upper-right pixel for dst on %%mm4 */
   1.289 +	    "movq %%mm0,%%mm2\n"
   1.290 +	    "movq %%mm1,%%mm4\n"
   1.291 +	    "movq %%mm0,%%mm3\n"
   1.292 +	    "movq %%mm1,%%mm5\n"
   1.293 +	    "pcmpeqw %%mm6,%%mm2\n"
   1.294 +	    "pcmpeqw %%mm6,%%mm4\n"
   1.295 +	    "pcmpeqw (%2),%%mm3\n"
   1.296 +	    "pcmpeqw (%2),%%mm5\n"
   1.297 +	    "pandn %%mm2,%%mm3\n"
   1.298 +	    "pandn %%mm4,%%mm5\n"
   1.299 +	    "movq %%mm0,%%mm2\n"
   1.300 +	    "movq %%mm1,%%mm4\n"
   1.301 +	    "pcmpeqw %%mm1,%%mm2\n"
   1.302 +	    "pcmpeqw %%mm0,%%mm4\n"
   1.303 +	    "pandn %%mm3,%%mm2\n"
   1.304 +	    "pandn %%mm5,%%mm4\n"
   1.305 +	    "movq %%mm2,%%mm3\n"
   1.306 +	    "movq %%mm4,%%mm5\n"
   1.307 +	    "pand %%mm6,%%mm2\n"
   1.308 +	    "pand %%mm6,%%mm4\n"
   1.309 +	    "pandn %%mm7,%%mm3\n"
   1.310 +	    "pandn %%mm7,%%mm5\n"
   1.311 +	    "por %%mm3,%%mm2\n"
   1.312 +	    "por %%mm5,%%mm4\n"
   1.313 +
   1.314 +	    /* set *dst */
   1.315 +	    "movq %%mm2,%%mm3\n"
   1.316 +	    "punpcklwd %%mm4,%%mm2\n"
   1.317 +	    "punpckhwd %%mm4,%%mm3\n"
   1.318 +	    "movq %%mm2,(%3)\n"
   1.319 +	    "movq %%mm3,8(%3)\n"
   1.320 +	    "emms\n"
   1.321 +
   1.322 +		: "+r" (src0), "+r" (src1), "+r" (src2), "+r" (dst), "+r" (count)
   1.323 +		:
   1.324 +		: "cc"
   1.325 +	    );
   1.326 +#else
   1.327 +	__asm {
   1.328 +		mov eax, src0;
   1.329 +		mov ebx, src1;
   1.330 +		mov ecx, src2;
   1.331 +		mov edx, dst;
   1.332 +		mov esi, count;
   1.333 +
   1.334 +		/* first run */
   1.335 +		/* set the current, current_pre, current_next registers */
   1.336 +		movq  mm0, qword ptr [ebx];
   1.337 +		movq  mm7, qword ptr [ebx];
   1.338 +		movq  mm1, qword ptr [ebx + 8];
   1.339 +		psllq mm0, 48;
   1.340 +		psllq mm1, 48;
   1.341 +		psrlq mm0, 48;
   1.342 +		movq  mm2, mm7;
   1.343 +		movq  mm3, mm7;
   1.344 +		psllq mm2, 16;
   1.345 +		psrlq mm3, 16;
   1.346 +		por	  mm0, mm2;
   1.347 +		por	  mm1, mm3;
   1.348 +
   1.349 +		/* current_upper */
   1.350 +		movq mm6, qword ptr [eax];
   1.351 +
   1.352 +		/* compute the upper-left pixel for dst on %%mm2 */
   1.353 +		/* compute the upper-right pixel for dst on %%mm4 */
   1.354 +		movq	mm2, mm0;
   1.355 +		movq	mm4, mm1;
   1.356 +		movq	mm3, mm0;
   1.357 +		movq	mm5, mm1;
   1.358 +		pcmpeqw mm2, mm6;
   1.359 +		pcmpeqw mm4, mm6;
   1.360 +		pcmpeqw mm3, qword ptr [ecx];
   1.361 +		pcmpeqw mm5, qword ptr [ecx];
   1.362 +		pandn	mm3, mm2;
   1.363 +		pandn	mm5, mm4;
   1.364 +		movq	mm2, mm0;
   1.365 +		movq	mm4, mm1;
   1.366 +		pcmpeqw mm2, mm1;
   1.367 +		pcmpeqw mm4, mm0;
   1.368 +		pandn	mm2, mm3;
   1.369 +		pandn	mm4, mm5;
   1.370 +		movq	mm3, mm2;
   1.371 +		movq	mm5, mm4;
   1.372 +		pand	mm2, mm6;
   1.373 +		pand	mm4, mm6;
   1.374 +		pandn	mm3, mm7;
   1.375 +		pandn	mm5, mm7;
   1.376 +		por		mm2, mm3;
   1.377 +		por		mm4, mm5;
   1.378 +
   1.379 +		/* set *dst0 */
   1.380 +		movq	   mm3, mm2;
   1.381 +		punpcklwd  mm2, mm4;
   1.382 +		punpckhwd  mm3, mm4;
   1.383 +		movq qword ptr [edx], mm2;
   1.384 +		movq qword ptr [edx + 8], mm3;
   1.385 +
   1.386 +		/* next */
   1.387 +		add eax, 8;
   1.388 +		add ebx, 8;
   1.389 +		add ecx, 8;
   1.390 +		add edx, 16;
   1.391 +
   1.392 +		/* central runs */
   1.393 +		shr esi, 2;
   1.394 +		jz	label1;
   1.395 +		align 4;
   1.396 +label0:
   1.397 +
   1.398 +		/* set the current, current_pre, current_next registers */
   1.399 +		movq mm0, qword ptr [ebx - 8];
   1.400 +		movq  mm7, qword ptr [ebx];
   1.401 +		movq  mm1, qword ptr [ebx + 8];
   1.402 +		psrlq mm0, 48;
   1.403 +		psllq mm1, 48;
   1.404 +		movq  mm2, mm7;
   1.405 +		movq  mm3, mm7;
   1.406 +		psllq mm2, 16;
   1.407 +		psrlq mm3, 16;
   1.408 +		por	  mm0, mm2;
   1.409 +		por	  mm1, mm3;
   1.410 +
   1.411 +		/* current_upper */
   1.412 +		movq mm6, qword ptr [eax];
   1.413 +
   1.414 +		/* compute the upper-left pixel for dst on %%mm2 */
   1.415 +		/* compute the upper-right pixel for dst on %%mm4 */
   1.416 +		movq	mm2, mm0;
   1.417 +		movq	mm4, mm1;
   1.418 +		movq	mm3, mm0;
   1.419 +		movq	mm5, mm1;
   1.420 +		pcmpeqw mm2, mm6;
   1.421 +		pcmpeqw mm4, mm6;
   1.422 +		pcmpeqw mm3, qword ptr [ecx];
   1.423 +		pcmpeqw mm5, qword ptr [ecx];
   1.424 +		pandn	mm3, mm2;
   1.425 +		pandn	mm5, mm4;
   1.426 +		movq	mm2, mm0;
   1.427 +		movq	mm4, mm1;
   1.428 +		pcmpeqw mm2, mm1;
   1.429 +		pcmpeqw mm4, mm0;
   1.430 +		pandn	mm2, mm3;
   1.431 +		pandn	mm4, mm5;
   1.432 +		movq	mm3, mm2;
   1.433 +		movq	mm5, mm4;
   1.434 +		pand	mm2, mm6;
   1.435 +		pand	mm4, mm6;
   1.436 +		pandn	mm3, mm7;
   1.437 +		pandn	mm5, mm7;
   1.438 +		por		mm2, mm3;
   1.439 +		por		mm4, mm5;
   1.440 +
   1.441 +		/* set *dst */
   1.442 +		movq	   mm3, mm2;
   1.443 +		punpcklwd  mm2, mm4;
   1.444 +		punpckhwd  mm3, mm4;
   1.445 +		movq qword ptr [edx], mm2;
   1.446 +		movq qword ptr [edx + 8], mm3;
   1.447 +
   1.448 +		/* next */
   1.449 +		add eax, 8;
   1.450 +		add ebx, 8;
   1.451 +		add ecx, 8;
   1.452 +		add edx, 16;
   1.453 +
   1.454 +		dec esi;
   1.455 +		jnz label0;
   1.456 +label1:
   1.457 +
   1.458 +		/* final run */
   1.459 +		/* set the current, current_pre, current_next registers */
   1.460 +		movq mm1, qword ptr [ebx];
   1.461 +		movq  mm7, qword ptr [ebx];
   1.462 +		movq  mm0, qword ptr [ebx - 8];
   1.463 +		psrlq mm1, 48;
   1.464 +		psrlq mm0, 48;
   1.465 +		psllq mm1, 48;
   1.466 +		movq  mm2, mm7;
   1.467 +		movq  mm3, mm7;
   1.468 +		psllq mm2, 16;
   1.469 +		psrlq mm3, 16;
   1.470 +		por	  mm0, mm2;
   1.471 +		por	  mm1, mm3;
   1.472 +
   1.473 +		/* current_upper */
   1.474 +		movq mm6, qword ptr [eax];
   1.475 +
   1.476 +		/* compute the upper-left pixel for dst on %%mm2 */
   1.477 +		/* compute the upper-right pixel for dst on %%mm4 */
   1.478 +		movq	mm2, mm0;
   1.479 +		movq	mm4, mm1;
   1.480 +		movq	mm3, mm0;
   1.481 +		movq	mm5, mm1;
   1.482 +		pcmpeqw mm2, mm6;
   1.483 +		pcmpeqw mm4, mm6;
   1.484 +		pcmpeqw mm3, qword ptr [ecx];
   1.485 +		pcmpeqw mm5, qword ptr [ecx];
   1.486 +		pandn	mm3, mm2;
   1.487 +		pandn	mm5, mm4;
   1.488 +		movq	mm2, mm0;
   1.489 +		movq	mm4, mm1;
   1.490 +		pcmpeqw mm2, mm1;
   1.491 +		pcmpeqw mm4, mm0;
   1.492 +		pandn	mm2, mm3;
   1.493 +		pandn	mm4, mm5;
   1.494 +		movq	mm3, mm2;
   1.495 +		movq	mm5, mm4;
   1.496 +		pand	mm2, mm6;
   1.497 +		pand	mm4, mm6;
   1.498 +		pandn	mm3, mm7;
   1.499 +		pandn	mm5, mm7;
   1.500 +		por		mm2, mm3;
   1.501 +		por		mm4, mm5;
   1.502 +
   1.503 +		/* set *dst */
   1.504 +		movq	   mm3, mm2;
   1.505 +		punpcklwd  mm2, mm4;
   1.506 +		punpckhwd  mm3, mm4;
   1.507 +		movq qword ptr [edx], mm2;
   1.508 +		movq qword ptr [edx + 8], mm3;
   1.509 +
   1.510 +		mov src0, eax;
   1.511 +		mov src1, ebx;
   1.512 +		mov src2, ecx;
   1.513 +		mov dst, edx;
   1.514 +		mov count, esi;
   1.515 +
   1.516 +		emms;
   1.517 +	}
   1.518 +#endif
   1.519 +}
   1.520 +
   1.521 +static void internal_scale2x_32_mmx_single(u32 *dst, const u32 *src0, const u32 *src1, const u32 *src2, unsigned count)
   1.522 +{
   1.523 +	/* always do the first and last run */
   1.524 +	count -= 2 * 2;
   1.525 +
   1.526 +#ifdef __GNUC__
   1.527 +	__asm__ __volatile__ (
   1.528 +	    /* first run */
   1.529 +	    /* set the current, current_pre, current_next registers */
   1.530 +	    "movq 0(%1),%%mm0\n"
   1.531 +	    "movq 0(%1),%%mm7\n"
   1.532 +	    "movq 8(%1),%%mm1\n"
   1.533 +	    "psllq $32,%%mm0\n"
   1.534 +	    "psllq $32,%%mm1\n"
   1.535 +	    "psrlq $32,%%mm0\n"
   1.536 +	    "movq %%mm7,%%mm2\n"
   1.537 +	    "movq %%mm7,%%mm3\n"
   1.538 +	    "psllq $32,%%mm2\n"
   1.539 +	    "psrlq $32,%%mm3\n"
   1.540 +	    "por %%mm2,%%mm0\n"
   1.541 +	    "por %%mm3,%%mm1\n"
   1.542 +
   1.543 +	    /* current_upper */
   1.544 +	    "movq (%0),%%mm6\n"
   1.545 +
   1.546 +	    /* compute the upper-left pixel for dst on %%mm2 */
   1.547 +	    /* compute the upper-right pixel for dst on %%mm4 */
   1.548 +	    "movq %%mm0,%%mm2\n"
   1.549 +	    "movq %%mm1,%%mm4\n"
   1.550 +	    "movq %%mm0,%%mm3\n"
   1.551 +	    "movq %%mm1,%%mm5\n"
   1.552 +	    "pcmpeqd %%mm6,%%mm2\n"
   1.553 +	    "pcmpeqd %%mm6,%%mm4\n"
   1.554 +	    "pcmpeqd (%2),%%mm3\n"
   1.555 +	    "pcmpeqd (%2),%%mm5\n"
   1.556 +	    "pandn %%mm2,%%mm3\n"
   1.557 +	    "pandn %%mm4,%%mm5\n"
   1.558 +	    "movq %%mm0,%%mm2\n"
   1.559 +	    "movq %%mm1,%%mm4\n"
   1.560 +	    "pcmpeqd %%mm1,%%mm2\n"
   1.561 +	    "pcmpeqd %%mm0,%%mm4\n"
   1.562 +	    "pandn %%mm3,%%mm2\n"
   1.563 +	    "pandn %%mm5,%%mm4\n"
   1.564 +	    "movq %%mm2,%%mm3\n"
   1.565 +	    "movq %%mm4,%%mm5\n"
   1.566 +	    "pand %%mm6,%%mm2\n"
   1.567 +	    "pand %%mm6,%%mm4\n"
   1.568 +	    "pandn %%mm7,%%mm3\n"
   1.569 +	    "pandn %%mm7,%%mm5\n"
   1.570 +	    "por %%mm3,%%mm2\n"
   1.571 +	    "por %%mm5,%%mm4\n"
   1.572 +
   1.573 +	    /* set *dst */
   1.574 +	    "movq %%mm2,%%mm3\n"
   1.575 +	    "punpckldq %%mm4,%%mm2\n"
   1.576 +	    "punpckhdq %%mm4,%%mm3\n"
   1.577 +	    "movq %%mm2,(%3)\n"
   1.578 +	    "movq %%mm3, 8(%3)\n"
   1.579 +
   1.580 +	    /* next */
   1.581 +	    "addl $8,%0\n"
   1.582 +	    "addl $8,%1\n"
   1.583 +	    "addl $8,%2\n"
   1.584 +	    "addl $16,%3\n"
   1.585 +
   1.586 +	    /* central runs */
   1.587 +	    "shrl $1,%4\n"
   1.588 +	    "jz 1f\n"
   1.589 +
   1.590 +	    "0:\n"
   1.591 +
   1.592 +	    /* set the current, current_pre, current_next registers */
   1.593 +	    "movq -8(%1),%%mm0\n"
   1.594 +	    "movq (%1),%%mm7\n"
   1.595 +	    "movq 8(%1),%%mm1\n"
   1.596 +	    "psrlq $32,%%mm0\n"
   1.597 +	    "psllq $32,%%mm1\n"
   1.598 +	    "movq %%mm7,%%mm2\n"
   1.599 +	    "movq %%mm7,%%mm3\n"
   1.600 +	    "psllq $32,%%mm2\n"
   1.601 +	    "psrlq $32,%%mm3\n"
   1.602 +	    "por %%mm2,%%mm0\n"
   1.603 +	    "por %%mm3,%%mm1\n"
   1.604 +
   1.605 +	    /* current_upper */
   1.606 +	    "movq (%0),%%mm6\n"
   1.607 +
   1.608 +	    /* compute the upper-left pixel for dst on %%mm2 */
   1.609 +	    /* compute the upper-right pixel for dst on %%mm4 */
   1.610 +	    "movq %%mm0,%%mm2\n"
   1.611 +	    "movq %%mm1,%%mm4\n"
   1.612 +	    "movq %%mm0,%%mm3\n"
   1.613 +	    "movq %%mm1,%%mm5\n"
   1.614 +	    "pcmpeqd %%mm6,%%mm2\n"
   1.615 +	    "pcmpeqd %%mm6,%%mm4\n"
   1.616 +	    "pcmpeqd (%2),%%mm3\n"
   1.617 +	    "pcmpeqd (%2),%%mm5\n"
   1.618 +	    "pandn %%mm2,%%mm3\n"
   1.619 +	    "pandn %%mm4,%%mm5\n"
   1.620 +	    "movq %%mm0,%%mm2\n"
   1.621 +	    "movq %%mm1,%%mm4\n"
   1.622 +	    "pcmpeqd %%mm1,%%mm2\n"
   1.623 +	    "pcmpeqd %%mm0,%%mm4\n"
   1.624 +	    "pandn %%mm3,%%mm2\n"
   1.625 +	    "pandn %%mm5,%%mm4\n"
   1.626 +	    "movq %%mm2,%%mm3\n"
   1.627 +	    "movq %%mm4,%%mm5\n"
   1.628 +	    "pand %%mm6,%%mm2\n"
   1.629 +	    "pand %%mm6,%%mm4\n"
   1.630 +	    "pandn %%mm7,%%mm3\n"
   1.631 +	    "pandn %%mm7,%%mm5\n"
   1.632 +	    "por %%mm3,%%mm2\n"
   1.633 +	    "por %%mm5,%%mm4\n"
   1.634 +
   1.635 +	    /* set *dst */
   1.636 +	    "movq %%mm2,%%mm3\n"
   1.637 +	    "punpckldq %%mm4,%%mm2\n"
   1.638 +	    "punpckhdq %%mm4,%%mm3\n"
   1.639 +	    "movq %%mm2,(%3)\n"
   1.640 +	    "movq %%mm3,8(%3)\n"
   1.641 +
   1.642 +	    /* next */
   1.643 +	    "addl $8,%0\n"
   1.644 +	    "addl $8,%1\n"
   1.645 +	    "addl $8,%2\n"
   1.646 +	    "addl $16,%3\n"
   1.647 +
   1.648 +	    "decl %4\n"
   1.649 +	    "jnz 0b\n"
   1.650 +	    "1:\n"
   1.651 +
   1.652 +	    /* final run */
   1.653 +	    /* set the current, current_pre, current_next registers */
   1.654 +	    "movq (%1),%%mm1\n"
   1.655 +	    "movq (%1),%%mm7\n"
   1.656 +	    "movq -8(%1), %%mm0\n"
   1.657 +	    "psrlq $32,%%mm1\n"
   1.658 +	    "psrlq $32,%%mm0\n"
   1.659 +	    "psllq $32,%%mm1\n"
   1.660 +	    "movq %%mm7,%%mm2\n"
   1.661 +	    "movq %%mm7,%%mm3\n"
   1.662 +	    "psllq $32,%%mm2\n"
   1.663 +	    "psrlq $32,%%mm3\n"
   1.664 +	    "por %%mm2,%%mm0\n"
   1.665 +	    "por %%mm3,%%mm1\n"
   1.666 +
   1.667 +	    /* current_upper */
   1.668 +	    "movq (%0),%%mm6\n"
   1.669 +
   1.670 +	    /* compute the upper-left pixel for dst on %%mm2 */
   1.671 +	    /* compute the upper-right pixel for dst on %%mm4 */
   1.672 +	    "movq %%mm0,%%mm2\n"
   1.673 +	    "movq %%mm1,%%mm4\n"
   1.674 +	    "movq %%mm0,%%mm3\n"
   1.675 +	    "movq %%mm1,%%mm5\n"
   1.676 +	    "pcmpeqd %%mm6,%%mm2\n"
   1.677 +	    "pcmpeqd %%mm6,%%mm4\n"
   1.678 +	    "pcmpeqd (%2),%%mm3\n"
   1.679 +	    "pcmpeqd (%2),%%mm5\n"
   1.680 +	    "pandn %%mm2,%%mm3\n"
   1.681 +	    "pandn %%mm4,%%mm5\n"
   1.682 +	    "movq %%mm0,%%mm2\n"
   1.683 +	    "movq %%mm1,%%mm4\n"
   1.684 +	    "pcmpeqd %%mm1,%%mm2\n"
   1.685 +	    "pcmpeqd %%mm0,%%mm4\n"
   1.686 +	    "pandn %%mm3,%%mm2\n"
   1.687 +	    "pandn %%mm5,%%mm4\n"
   1.688 +	    "movq %%mm2,%%mm3\n"
   1.689 +	    "movq %%mm4,%%mm5\n"
   1.690 +	    "pand %%mm6,%%mm2\n"
   1.691 +	    "pand %%mm6,%%mm4\n"
   1.692 +	    "pandn %%mm7,%%mm3\n"
   1.693 +	    "pandn %%mm7,%%mm5\n"
   1.694 +	    "por %%mm3,%%mm2\n"
   1.695 +	    "por %%mm5,%%mm4\n"
   1.696 +
   1.697 +	    /* set *dst */
   1.698 +	    "movq %%mm2,%%mm3\n"
   1.699 +	    "punpckldq %%mm4,%%mm2\n"
   1.700 +	    "punpckhdq %%mm4,%%mm3\n"
   1.701 +	    "movq %%mm2,(%3)\n"
   1.702 +	    "movq %%mm3,8(%3)\n"
   1.703 +	    "emms\n"
   1.704 +
   1.705 +		: "+r" (src0), "+r" (src1), "+r" (src2), "+r" (dst), "+r" (count)
   1.706 +		:
   1.707 +		: "cc"
   1.708 +	    );
   1.709 +#else
   1.710 +	__asm {
   1.711 +		mov eax, src0;
   1.712 +		mov ebx, src1;
   1.713 +		mov ecx, src2;
   1.714 +		mov edx, dst;
   1.715 +		mov esi, count;
   1.716 +
   1.717 +		/* first run */
   1.718 +		/* set the current, current_pre, current_next registers */
   1.719 +		movq  mm0, qword ptr [ebx];
   1.720 +		movq  mm7, qword ptr [ebx];
   1.721 +		movq  mm1, qword ptr [ebx + 8];
   1.722 +		psllq mm0, 32;
   1.723 +		psllq mm1, 32;
   1.724 +		psrlq mm0, 32;
   1.725 +		movq  mm2, mm7;
   1.726 +		movq  mm3, mm7;
   1.727 +		psllq mm2, 32;
   1.728 +		psrlq mm3, 32;
   1.729 +		por	  mm0, mm2;
   1.730 +		por	  mm1, mm3;
   1.731 +
   1.732 +		/* current_upper */
   1.733 +		movq mm6, qword ptr [eax];
   1.734 +
   1.735 +		/* compute the upper-left pixel for dst on %%mm2 */
   1.736 +		/* compute the upper-right pixel for dst on %%mm4 */
   1.737 +		movq	mm2, mm0;
   1.738 +		movq	mm4, mm1;
   1.739 +		movq	mm3, mm0;
   1.740 +		movq	mm5, mm1;
   1.741 +		pcmpeqd mm2, mm6;
   1.742 +		pcmpeqd mm4, mm6;
   1.743 +		pcmpeqd mm3, qword ptr [ecx];
   1.744 +		pcmpeqd mm5, qword ptr [ecx];
   1.745 +		pandn	mm3, mm2;
   1.746 +		pandn	mm5, mm4;
   1.747 +		movq	mm2, mm0;
   1.748 +		movq	mm4, mm1;
   1.749 +		pcmpeqd mm2, mm1;
   1.750 +		pcmpeqd mm4, mm0;
   1.751 +		pandn	mm2, mm3;
   1.752 +		pandn	mm4, mm5;
   1.753 +		movq	mm3, mm2;
   1.754 +		movq	mm5, mm4;
   1.755 +		pand	mm2, mm6;
   1.756 +		pand	mm4, mm6;
   1.757 +		pandn	mm3, mm7;
   1.758 +		pandn	mm5, mm7;
   1.759 +		por		mm2, mm3;
   1.760 +		por		mm4, mm5;
   1.761 +
   1.762 +		/* set *dst */
   1.763 +		movq	   mm3, mm2;
   1.764 +		punpckldq  mm2, mm4;
   1.765 +		punpckhdq  mm3, mm4;
   1.766 +		movq qword ptr [edx], mm2;
   1.767 +		movq qword ptr [edx + 8], mm3;
   1.768 +
   1.769 +		/* next */
   1.770 +		add eax, 8;
   1.771 +		add ebx, 8;
   1.772 +		add ecx, 8;
   1.773 +		add edx, 16;
   1.774 +
   1.775 +		/* central runs */
   1.776 +		shr esi, 1;
   1.777 +		jz	label1;
   1.778 +label0:
   1.779 +
   1.780 +		/* set the current, current_pre, current_next registers */
   1.781 +		movq mm0, qword ptr [ebx - 8];
   1.782 +		movq  mm7, qword ptr [ebx];
   1.783 +		movq  mm1, qword ptr [ebx + 8];
   1.784 +		psrlq mm0, 32;
   1.785 +		psllq mm1, 32;
   1.786 +		movq  mm2, mm7;
   1.787 +		movq  mm3, mm7;
   1.788 +		psllq mm2, 32;
   1.789 +		psrlq mm3, 32;
   1.790 +		por	  mm0, mm2;
   1.791 +		por	  mm1, mm3;
   1.792 +
   1.793 +		/* current_upper */
   1.794 +		movq mm6, qword ptr[eax];
   1.795 +
   1.796 +		/* compute the upper-left pixel for dst on %%mm2 */
   1.797 +		/* compute the upper-right pixel for dst on %%mm4 */
   1.798 +		movq	mm2, mm0;
   1.799 +		movq	mm4, mm1;
   1.800 +		movq	mm3, mm0;
   1.801 +		movq	mm5, mm1;
   1.802 +		pcmpeqd mm2, mm6;
   1.803 +		pcmpeqd mm4, mm6;
   1.804 +		pcmpeqd mm3, qword ptr[ecx];
   1.805 +		pcmpeqd mm5, qword ptr[ecx];
   1.806 +		pandn	mm3, mm2;
   1.807 +		pandn	mm5, mm4;
   1.808 +		movq	mm2, mm0;
   1.809 +		movq	mm4, mm1;
   1.810 +		pcmpeqd mm2, mm1;
   1.811 +		pcmpeqd mm4, mm0;
   1.812 +		pandn	mm2, mm3;
   1.813 +		pandn	mm4, mm5;
   1.814 +		movq	mm3, mm2;
   1.815 +		movq	mm5, mm4;
   1.816 +		pand	mm2, mm6;
   1.817 +		pand	mm4, mm6;
   1.818 +		pandn	mm3, mm7;
   1.819 +		pandn	mm5, mm7;
   1.820 +		por		mm2, mm3;
   1.821 +		por		mm4, mm5;
   1.822 +
   1.823 +		/* set *dst */
   1.824 +		movq	   mm3, mm2;
   1.825 +		punpckldq  mm2, mm4;
   1.826 +		punpckhdq  mm3, mm4;
   1.827 +		movq qword ptr [edx], mm2;
   1.828 +		movq qword ptr [edx + 8], mm3;
   1.829 +
   1.830 +		/* next */
   1.831 +		add eax, 8;
   1.832 +		add ebx, 8;
   1.833 +		add ecx, 8;
   1.834 +		add edx, 16;
   1.835 +
   1.836 +		dec esi;
   1.837 +		jnz label0;
   1.838 +label1:
   1.839 +
   1.840 +		/* final run */
   1.841 +		/* set the current, current_pre, current_next registers */
   1.842 +		movq mm1, qword ptr [ebx];
   1.843 +		movq  mm7, qword ptr [ebx];
   1.844 +		movq  mm0, qword ptr [ebx - 8];
   1.845 +		psrlq mm1, 32;
   1.846 +		psrlq mm0, 32;
   1.847 +		psllq mm1, 32;
   1.848 +		movq  mm2, mm7;
   1.849 +		movq  mm3, mm7;
   1.850 +		psllq mm2, 32;
   1.851 +		psrlq mm3, 32;
   1.852 +		por	  mm0, mm2;
   1.853 +		por	  mm1, mm3;
   1.854 +
   1.855 +		/* current_upper */
   1.856 +		movq mm6, qword ptr [eax];
   1.857 +
   1.858 +		/* compute the upper-left pixel for dst on %%mm2 */
   1.859 +		/* compute the upper-right pixel for dst on %%mm4 */
   1.860 +		movq	mm2, mm0;
   1.861 +		movq	mm4, mm1;
   1.862 +		movq	mm3, mm0;
   1.863 +		movq	mm5, mm1;
   1.864 +		pcmpeqd mm2, mm6;
   1.865 +		pcmpeqd mm4, mm6;
   1.866 +		pcmpeqd mm3, qword ptr [ecx];
   1.867 +		pcmpeqd mm5, qword ptr [ecx];
   1.868 +		pandn	mm3, mm2;
   1.869 +		pandn	mm5, mm4;
   1.870 +		movq	mm2, mm0;
   1.871 +		movq	mm4, mm1;
   1.872 +		pcmpeqd mm2, mm1;
   1.873 +		pcmpeqd mm4, mm0;
   1.874 +		pandn	mm2, mm3;
   1.875 +		pandn	mm4, mm5;
   1.876 +		movq	mm3, mm2;
   1.877 +		movq	mm5, mm4;
   1.878 +		pand	mm2, mm6;
   1.879 +		pand	mm4, mm6;
   1.880 +		pandn	mm3, mm7;
   1.881 +		pandn	mm5, mm7;
   1.882 +		por		mm2, mm3;
   1.883 +		por		mm4, mm5;
   1.884 +
   1.885 +		/* set *dst */
   1.886 +		movq	   mm3, mm2;
   1.887 +		punpckldq  mm2, mm4;
   1.888 +		punpckhdq  mm3, mm4;
   1.889 +		movq qword ptr [edx], mm2;
   1.890 +		movq qword ptr [edx + 8], mm3;
   1.891 +
   1.892 +		mov src0, eax;
   1.893 +		mov src1, ebx;
   1.894 +		mov src2, ecx;
   1.895 +		mov dst, edx;
   1.896 +		mov count, esi;
   1.897 +
   1.898 +		emms;
   1.899 +	}
   1.900 +#endif
   1.901 +}
   1.902 +
   1.903 +static void internal_scale2x_16_mmx(u16 *dst0, u16 *dst1, const u16 *src0, const u16 *src1, const u16 *src2, unsigned count)
   1.904 +{
   1.905 +	//	assert( count >= 2*4 );
   1.906 +	internal_scale2x_16_mmx_single(dst0, src0, src1, src2, count);
   1.907 +	internal_scale2x_16_mmx_single(dst1, src2, src1, src0, count);
   1.908 +}
   1.909 +
   1.910 +static void internal_scale2x_32_mmx(u32 *dst0, u32 *dst1, const u32 *src0, const u32 *src1, const u32 *src2, unsigned count)
   1.911 +{
   1.912 +	//	assert( count >= 2*2 );
   1.913 +	internal_scale2x_32_mmx_single(dst0, src0, src1, src2, count);
   1.914 +	internal_scale2x_32_mmx_single(dst1, src2, src1, src0, count);
   1.915 +}
   1.916 +
   1.917 +#endif
   1.918 +
   1.919 +void AdMame2x(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
   1.920 +              u8 *dstPtr, u32 dstPitch, int width, int height)
   1.921 +{
   1.922 +	u16 *dst0 = (u16 *)dstPtr;
   1.923 +	u16 *dst1 = dst0 + (dstPitch >> 1);
   1.924 +
   1.925 +	u16 *src0 = (u16 *)srcPtr;
   1.926 +	u16 *src1 = src0 + (srcPitch >> 1);
   1.927 +	u16 *src2 = src1 + (srcPitch >> 1);
   1.928 +#ifdef MMX
   1.929 +	if (cpu_mmx)
   1.930 +	{
   1.931 +		internal_scale2x_16_mmx(dst0, dst1, src0, src0, src1, width);
   1.932 +
   1.933 +		int count = height;
   1.934 +
   1.935 +		count -= 2;
   1.936 +		while (count)
   1.937 +		{
   1.938 +			dst0 += dstPitch;
   1.939 +			dst1 += dstPitch;
   1.940 +			internal_scale2x_16_mmx(dst0, dst1, src0, src1, src2, width);
   1.941 +			src0  = src1;
   1.942 +			src1  = src2;
   1.943 +			src2 += srcPitch >> 1;
   1.944 +			--count;
   1.945 +		}
   1.946 +		dst0 += dstPitch;
   1.947 +		dst1 += dstPitch;
   1.948 +		internal_scale2x_16_mmx(dst0, dst1, src0, src1, src1, width);
   1.949 +	}
   1.950 +	else
   1.951 +	{
   1.952 +#endif
   1.953 +	internal_scale2x_16_def(dst0, src0, src0, src1, width);
   1.954 +	internal_scale2x_16_def(dst1, src1, src0, src0, width);
   1.955 +
   1.956 +	int count = height;
   1.957 +
   1.958 +	count -= 2;
   1.959 +	while (count)
   1.960 +	{
   1.961 +		dst0 += dstPitch;
   1.962 +		dst1 += dstPitch;
   1.963 +		internal_scale2x_16_def(dst0, src0, src1, src2, width);
   1.964 +		internal_scale2x_16_def(dst1, src2, src1, src0, width);
   1.965 +		src0  = src1;
   1.966 +		src1  = src2;
   1.967 +		src2 += srcPitch >> 1;
   1.968 +		--count;
   1.969 +	}
   1.970 +	dst0 += dstPitch;
   1.971 +	dst1 += dstPitch;
   1.972 +	internal_scale2x_16_def(dst0, src0, src1, src1, width);
   1.973 +	internal_scale2x_16_def(dst1, src1, src1, src0, width);
   1.974 +#ifdef MMX
   1.975 +}
   1.976 +
   1.977 +#endif
   1.978 +}
   1.979 +
   1.980 +void AdMame2x32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
   1.981 +                u8 *dstPtr, u32 dstPitch, int width, int height)
   1.982 +{
   1.983 +	u32 *dst0 = (u32 *)dstPtr;
   1.984 +	u32 *dst1 = dst0 + (dstPitch >> 2);
   1.985 +
   1.986 +	u32 *src0 = (u32 *)srcPtr;
   1.987 +	u32 *src1 = src0 + (srcPitch >> 2);
   1.988 +	u32 *src2 = src1 + (srcPitch >> 2);
   1.989 +#ifdef MMX
   1.990 +	if (cpu_mmx)
   1.991 +	{
   1.992 +		internal_scale2x_32_mmx(dst0, dst1, src0, src0, src1, width);
   1.993 +
   1.994 +		int count = height;
   1.995 +
   1.996 +		count -= 2;
   1.997 +		while (count)
   1.998 +		{
   1.999 +			dst0 += dstPitch >> 1;
  1.1000 +			dst1 += dstPitch >> 1;
  1.1001 +			internal_scale2x_32_mmx(dst0, dst1, src0, src1, src2, width);
  1.1002 +			src0  = src1;
  1.1003 +			src1  = src2;
  1.1004 +			src2 += srcPitch >> 2;
  1.1005 +			--count;
  1.1006 +		}
  1.1007 +		dst0 += dstPitch >> 1;
  1.1008 +		dst1 += dstPitch >> 1;
  1.1009 +		internal_scale2x_32_mmx(dst0, dst1, src0, src1, src1, width);
  1.1010 +	}
  1.1011 +	else
  1.1012 +	{
  1.1013 +#endif
  1.1014 +	internal_scale2x_32_def(dst0, src0, src0, src1, width);
  1.1015 +	internal_scale2x_32_def(dst1, src1, src0, src0, width);
  1.1016 +
  1.1017 +	int count = height;
  1.1018 +
  1.1019 +	count -= 2;
  1.1020 +	while (count)
  1.1021 +	{
  1.1022 +		dst0 += dstPitch >> 1;
  1.1023 +		dst1 += dstPitch >> 1;
  1.1024 +		internal_scale2x_32_def(dst0, src0, src1, src2, width);
  1.1025 +		internal_scale2x_32_def(dst1, src2, src1, src0, width);
  1.1026 +		src0  = src1;
  1.1027 +		src1  = src2;
  1.1028 +		src2 += srcPitch >> 2;
  1.1029 +		--count;
  1.1030 +	}
  1.1031 +	dst0 += dstPitch >> 1;
  1.1032 +	dst1 += dstPitch >> 1;
  1.1033 +	internal_scale2x_32_def(dst0, src0, src1, src1, width);
  1.1034 +	internal_scale2x_32_def(dst1, src1, src1, src0, width);
  1.1035 +#ifdef MMX
  1.1036 +}
  1.1037 +
  1.1038 +#endif
  1.1039 +}