view libmpeg2/libmpeg-0.4.1.diff @ 25729:9a96fd14d1c5

Add experimental unsharp-mask OpenGL scaler. Certainly not yet perfect.

--- include/attributes.h	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/attributes.h	2006-06-16 20:12:50.000000000 +0200
@@ -25,7 +29,7 @@
 #ifdef ATTRIBUTE_ALIGNED_MAX
 #define ATTR_ALIGN(align) __attribute__ ((__aligned__ ((ATTRIBUTE_ALIGNED_MAX < align) ? ATTRIBUTE_ALIGNED_MAX : align)))
 #else
-#define ATTR_ALIGN(align)
+#define ATTR_ALIGN(align) __attribute__ ((__aligned__ ((16 < align) ? 16 : align)))
 #endif
 
 #ifdef HAVE_BUILTIN_EXPECT
--- libmpeg2/cpu_accel.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/cpu_accel.c	2006-06-16 20:12:50.000000000 +0200
@@ -22,6 +26,7 @@
  */
 
 #include "config.h"
+#include "cpudetect.h"
 
 #include <inttypes.h>
 
@@ -30,9 +35,17 @@
 #include "mpeg2_internal.h"
 
 #ifdef ACCEL_DETECT
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+
+/* MPlayer imports libmpeg2 as decoder, which detects MMX / 3DNow! 
+ * instructions via assembly. However, it is regarded as duplicaed work
+ * in MPlayer, so that we enforce to use MPlayer's implementation.
+ */
+#define USE_MPLAYER_CPUDETECT
+
 static inline uint32_t arch_accel (void)
 {
+#if !defined(USE_MPLAYER_CPUDETECT)
     uint32_t eax, ebx, ecx, edx;
     int AMD;
     uint32_t caps;
@@ -105,10 +118,24 @@
 	caps |= MPEG2_ACCEL_X86_MMXEXT;
 
     return caps;
+#else /* USE_MPLAYER_CPUDETECT: Use MPlayer's cpu capability property */
+    caps = 0;
+    if (gCpuCaps.hasMMX)
+        caps |= MPEG2_ACCEL_X86_MMX;
+    if (gCpuCaps.hasSSE2)
+	caps |= MPEG2_ACCEL_X86_SSE2;
+    if (gCpuCaps.hasMMX2)
+	caps |= MPEG2_ACCEL_X86_MMXEXT;
+    if (gCpuCaps.has3DNow)
+	caps |= MPEG2_ACCEL_X86_3DNOW;
+
+    return caps;
+
+#endif /* USE_MPLAYER_CPUDETECT */
 }
-#endif /* ARCH_X86 */
+#endif /* ARCH_X86 || ARCH_X86_64 */
 
-#if defined(ARCH_PPC) || defined(ARCH_SPARC)
+#if defined(ARCH_PPC) || (defined(ARCH_SPARC) && defined(HAVE_VIS))
 #include <signal.h>
 #include <setjmp.h>
 
@@ -166,10 +166,10 @@
 
     canjump = 1;
 
-#ifdef HAVE_ALTIVEC_H	/* gnu */
-#define VAND(a,b,c) "vand " #a "," #b "," #c "\n\t"
-#else			/* apple */
+#if defined( __APPLE_CC__ ) && defined( __APPLE_ALTIVEC__ ) /* apple */
 #define VAND(a,b,c) "vand v" #a ",v" #b ",v" #c "\n\t"
+#else			/* gnu */
+#define VAND(a,b,c) "vand " #a "," #b "," #c "\n\t"
 #endif
     asm volatile ("mtspr 256, %0\n\t"
 		  VAND (0, 0, 0)
@@ -195,6 +222,7 @@
 #ifdef ARCH_ALPHA
 static inline uint32_t arch_accel (void)
 {
+#ifdef CAN_COMPILE_ALPHA_MVI
     uint64_t no_mvi;
 
     asm volatile ("amask %1, %0"
@@ -202,6 +230,9 @@
 		  : "rI" (256));	/* AMASK_MVI */
     return no_mvi ? MPEG2_ACCEL_ALPHA : (MPEG2_ACCEL_ALPHA |
 					 MPEG2_ACCEL_ALPHA_MVI);
+#else
+    return MPEG2_ACCEL_ALPHA;
+#endif
 }
 #endif /* ARCH_ALPHA */
 #endif /* ACCEL_DETECT */
@@ -212,7 +243,7 @@
 
     accel = 0;
 #ifdef ACCEL_DETECT
-#if defined (ARCH_X86) || defined (ARCH_PPC) || defined (ARCH_ALPHA) || defined (ARCH_SPARC)
+#if defined (ARCH_X86) || defined (ARCH_X86_64) || defined (ARCH_PPC) || defined (ARCH_ALPHA) || defined (ARCH_SPARC)
     accel = arch_accel ();
 #endif
 #endif
--- libmpeg2/cpu_state.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/cpu_state.c	2006-06-16 20:12:50.000000000 +0200
@@ -29,14 +33,14 @@
 #include "mpeg2.h"
 #include "attributes.h"
 #include "mpeg2_internal.h"
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
 #include "mmx.h"
 #endif
 
 void (* mpeg2_cpu_state_save) (cpu_state_t * state) = NULL;
 void (* mpeg2_cpu_state_restore) (cpu_state_t * state) = NULL;
 
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
 static void state_restore_mmx (cpu_state_t * state)
 {
     emms ();
@@ -48,18 +48,18 @@
 #endif
 
-#ifdef ARCH_PPC
+#if defined(ARCH_PPC) && defined(HAVE_ALTIVEC)
-#ifdef HAVE_ALTIVEC_H	/* gnu */
-#define LI(a,b) "li " #a "," #b "\n\t"
-#define STVX0(a,b,c) "stvx " #a ",0," #c "\n\t"
-#define STVX(a,b,c) "stvx " #a "," #b "," #c "\n\t"
-#define LVX0(a,b,c) "lvx " #a ",0," #c "\n\t"
-#define LVX(a,b,c) "lvx " #a "," #b "," #c "\n\t"
-#else			/* apple */
+#if defined( __APPLE_CC__ ) && defined( __APPLE_ALTIVEC__ )	/* apple */
 #define LI(a,b) "li r" #a "," #b "\n\t"
 #define STVX0(a,b,c) "stvx v" #a ",0,r" #c "\n\t"
 #define STVX(a,b,c) "stvx v" #a ",r" #b ",r" #c "\n\t"
 #define LVX0(a,b,c) "lvx v" #a ",0,r" #c "\n\t"
 #define LVX(a,b,c) "lvx v" #a ",r" #b ",r" #c "\n\t"
+#else			/* gnu */
+#define LI(a,b) "li " #a "," #b "\n\t"
+#define STVX0(a,b,c) "stvx " #a ",0," #c "\n\t"
+#define STVX(a,b,c) "stvx " #a "," #b "," #c "\n\t"
+#define LVX0(a,b,c) "lvx " #a ",0," #c "\n\t"
+#define LVX(a,b,c) "lvx " #a "," #b "," #c "\n\t"
 #endif
 
 static void state_save_altivec (cpu_state_t * state)
@@ -115,12 +119,12 @@
 
 void mpeg2_cpu_state_init (uint32_t accel)
 {
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
     if (accel & MPEG2_ACCEL_X86_MMX) {
 	mpeg2_cpu_state_restore = state_restore_mmx;
     }
 #endif
-#ifdef ARCH_PPC
+#if defined(ARCH_PPC) && defined(HAVE_ALTIVEC)
     if (accel & MPEG2_ACCEL_PPC_ALTIVEC) {
 	mpeg2_cpu_state_save = state_save_altivec;
 	mpeg2_cpu_state_restore = state_restore_altivec;
--- libmpeg2/decode.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/decode.c	2006-06-16 20:12:50.000000000 +0200
@@ -351,6 +355,15 @@
     fbuf->buf[1] = buf[1];
     fbuf->buf[2] = buf[2];
     fbuf->id = id;
+    // HACK! FIXME! At first I frame, copy pointers to prediction frame too!
+    if (mpeg2dec->custom_fbuf && !mpeg2dec->fbuf[1]->buf[0]){
+	mpeg2dec->fbuf[1]->buf[0]=buf[0];
+	mpeg2dec->fbuf[1]->buf[1]=buf[1];
+	mpeg2dec->fbuf[1]->buf[2]=buf[2];
+	mpeg2dec->fbuf[1]->id=NULL;
+    }
+//        printf("libmpeg2: FBUF 0:%p 1:%p 2:%p\n",
+//	    mpeg2dec->fbuf[0]->buf[0],mpeg2dec->fbuf[1]->buf[0],mpeg2dec->fbuf[2]->buf[0]);
 }
 
 void mpeg2_custom_fbuf (mpeg2dec_t * mpeg2dec, int custom_fbuf)
--- libmpeg2/header.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/header.c	2006-06-16 20:12:50.000000000 +0200
@@ -100,6 +104,9 @@
     mpeg2dec->decoder.convert = NULL;
     mpeg2dec->decoder.convert_id = NULL;
     mpeg2dec->picture = mpeg2dec->pictures;
+    memset(&mpeg2dec->fbuf_alloc[0].fbuf, 0, sizeof(mpeg2_fbuf_t));
+    memset(&mpeg2dec->fbuf_alloc[1].fbuf, 0, sizeof(mpeg2_fbuf_t));
+    memset(&mpeg2dec->fbuf_alloc[2].fbuf, 0, sizeof(mpeg2_fbuf_t));
     mpeg2dec->fbuf[0] = &mpeg2dec->fbuf_alloc[0].fbuf;
     mpeg2dec->fbuf[1] = &mpeg2dec->fbuf_alloc[1].fbuf;
     mpeg2dec->fbuf[2] = &mpeg2dec->fbuf_alloc[2].fbuf;
@@ -551,6 +558,7 @@
 	if (!(mpeg2dec->sequence.flags & SEQ_FLAG_PROGRESSIVE_SEQUENCE)) {
 	    picture->nb_fields = (buffer[3] & 2) ? 3 : 2;
 	    flags |= (buffer[3] & 128) ? PIC_FLAG_TOP_FIELD_FIRST : 0;
+	    flags |= (buffer[3] &   2) ? PIC_FLAG_REPEAT_FIRST_FIELD : 0;
 	} else
 	    picture->nb_fields = (buffer[3]&2) ? ((buffer[3]&128) ? 6 : 4) : 2;
 	break;
@@ -799,6 +807,7 @@
 	mpeg2dec->scaled[index] = mpeg2dec->q_scale_type;
 	for (i = 0; i < 32; i++) {
 	    k = mpeg2dec->q_scale_type ? non_linear_scale[i] : (i << 1);
+	    decoder->quantizer_scales[i] = k;
 	    for (j = 0; j < 64; j++)
 		decoder->quantizer_prescale[index][i][j] =
 		    k * mpeg2dec->quantizer_matrix[index][j];
--- libmpeg2/idct.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/idct.c	2006-06-16 20:12:50.000000000 +0200
@@ -239,12 +239,15 @@
 
 void mpeg2_idct_init (uint32_t accel)
 {
-#ifdef ARCH_X86
+#ifdef HAVE_MMX2
     if (accel & MPEG2_ACCEL_X86_MMXEXT) {
 	mpeg2_idct_copy = mpeg2_idct_copy_mmxext;
 	mpeg2_idct_add = mpeg2_idct_add_mmxext;
 	mpeg2_idct_mmx_init ();
-    } else if (accel & MPEG2_ACCEL_X86_MMX) {
+    } else
+#endif
+#ifdef HAVE_MMX
+    if (accel & MPEG2_ACCEL_X86_MMX) {
 	mpeg2_idct_copy = mpeg2_idct_copy_mmx;
 	mpeg2_idct_add = mpeg2_idct_add_mmx;
 	mpeg2_idct_mmx_init ();
@@ -254,11 +261,14 @@
     } else
 #endif
 #ifdef ARCH_ALPHA
+#ifdef CAN_COMPILE_ALPHA_MVI
     if (accel & MPEG2_ACCEL_ALPHA_MVI) {
 	mpeg2_idct_copy = mpeg2_idct_copy_mvi;
 	mpeg2_idct_add = mpeg2_idct_add_mvi;
 	mpeg2_idct_alpha_init ();
-    } else if (accel & MPEG2_ACCEL_ALPHA) {
+    } else
+#endif
+    if (accel & MPEG2_ACCEL_ALPHA) {
 	int i;
 
 	mpeg2_idct_copy = mpeg2_idct_copy_alpha;
--- libmpeg2/idct_alpha.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/idct_alpha.c	2006-06-16 20:12:50.000000000 +0200
@@ -157,6 +161,7 @@
     block[8*7] = (a0 - b0) >> 17;
 }
 
+#ifdef CAN_COMPILE_ALPHA_MVI
 void mpeg2_idct_copy_mvi (int16_t * block, uint8_t * dest, const int stride)
 {
     uint64_t clampmask;
@@ -289,6 +294,7 @@
 	stq (p7, dest + 7 * stride);
     }
 }
+#endif
 
 void mpeg2_idct_copy_alpha (int16_t * block, uint8_t * dest, const int stride)
 {
--- libmpeg2/idct_mmx.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/idct_mmx.c	2006-06-16 20:12:50.000000000 +0200
@@ -23,7 +27,7 @@
 
 #include "config.h"
 
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
 
 #include <inttypes.h>
 
--- libmpeg2/motion_comp.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/motion_comp.c	2006-06-16 20:12:50.000000000 +0200
@@ -37,16 +37,22 @@
 
 void mpeg2_mc_init (uint32_t accel)
 {
-#ifdef ARCH_X86
+#ifdef HAVE_MMX2
     if (accel & MPEG2_ACCEL_X86_MMXEXT)
 	mpeg2_mc = mpeg2_mc_mmxext;
-    else if (accel & MPEG2_ACCEL_X86_3DNOW)
+    else
+#endif
+#ifdef HAVE_3DNOW
+    if (accel & MPEG2_ACCEL_X86_3DNOW)
 	mpeg2_mc = mpeg2_mc_3dnow;
-    else if (accel & MPEG2_ACCEL_X86_MMX)
+    else
+#endif
+#ifdef HAVE_MMX
+    if (accel & MPEG2_ACCEL_X86_MMX)
 	mpeg2_mc = mpeg2_mc_mmx;
     else
 #endif
-#ifdef ARCH_PPC
+#if defined(ARCH_PPC) && defined(HAVE_ALTIVEC)
     if (accel & MPEG2_ACCEL_PPC_ALTIVEC)
 	mpeg2_mc = mpeg2_mc_altivec;
     else
@@ -52,7 +62,7 @@
 	mpeg2_mc = mpeg2_mc_alpha;
     else
 #endif
-#ifdef ARCH_SPARC
+#if defined(ARCH_SPARC) && defined(HAVE_VIS)
     if (accel & MPEG2_ACCEL_SPARC_VIS)
 	mpeg2_mc = mpeg2_mc_vis;
     else
@@ -67,6 +67,16 @@
 	mpeg2_mc = mpeg2_mc_vis;
     else
 #endif
+#ifdef ARCH_ARM
+    if (1 /*accel & MPEG2_ACCEL_ARM*/) {
+#ifdef HAVE_IWMMXT
+	if (1 /*accel & MPEG2_ACCEL_ARM_IWMMXT*/)
+	    mpeg2_mc = mpeg2_mc_iwmmxt;
+	else
+#endif
+	    mpeg2_mc = mpeg2_mc_arm;
+    } else
+#endif
 	mpeg2_mc = mpeg2_mc_c;
 }
 
--- libmpeg2/motion_comp_mmx.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/motion_comp_mmx.c	2006-06-16 20:12:50.000000000 +0200
@@ -23,7 +27,7 @@
 
 #include "config.h"
 
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
 
 #include <inttypes.h>
 
--- include/mpeg2.h	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/mpeg2.h	2006-06-16 20:12:50.000000000 +0200
@@ -82,6 +86,7 @@
 #define PIC_FLAG_COMPOSITE_DISPLAY 32
 #define PIC_FLAG_SKIP 64
 #define PIC_FLAG_TAGS 128
+#define PIC_FLAG_REPEAT_FIRST_FIELD 256
 #define PIC_MASK_COMPOSITE_DISPLAY 0xfffff000
 
 typedef struct mpeg2_picture_s {
@@ -154,6 +159,7 @@
 #define MPEG2_ACCEL_X86_MMX 1
 #define MPEG2_ACCEL_X86_3DNOW 2
 #define MPEG2_ACCEL_X86_MMXEXT 4
+#define MPEG2_ACCEL_X86_SSE2 8
 #define MPEG2_ACCEL_PPC_ALTIVEC 1
 #define MPEG2_ACCEL_ALPHA 1
 #define MPEG2_ACCEL_ALPHA_MVI 2
--- libmpeg2/mpeg2_internal.h	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/mpeg2_internal.h	2006-06-16 20:12:50.000000000 +0200
@@ -144,6 +148,12 @@
     int second_field;
 
     int mpeg1;
+
+    /* for MPlayer: */
+    int quantizer_scales[32];
+    int quantizer_scale;
+    char* quant_store;
+    int quant_stride;
 };
 
 typedef struct {
@@ -214,6 +224,10 @@
     int8_t q_scale_type, scaled[4];
     uint8_t quantizer_matrix[4][64];
     uint8_t new_quantizer_matrix[4][64];
+
+    /* for MPlayer: */
+    unsigned char *pending_buffer;
+    int pending_length;
 };
 
 typedef struct {
@@ -312,3 +312,5 @@
 extern mpeg2_mc_t mpeg2_mc_altivec;
 extern mpeg2_mc_t mpeg2_mc_alpha;
 extern mpeg2_mc_t mpeg2_mc_vis;
+extern mpeg2_mc_t mpeg2_mc_arm;
+extern mpeg2_mc_t mpeg2_mc_iwmmxt;
--- libmpeg2/slice.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/slice.c	2006-06-16 20:12:50.000000000 +0200
@@ -142,6 +146,7 @@
 
     quantizer_scale_code = UBITS (bit_buf, 5);
     DUMPBITS (bit_buf, bits, 5);
+    decoder->quantizer_scale = decoder->quantizer_scales[quantizer_scale_code];
 
     decoder->quantizer_matrix[0] =
 	decoder->quantizer_prescale[0][quantizer_scale_code];
@@ -1568,6 +1569,18 @@
 
 #define NEXT_MACROBLOCK							\
 do {									\
+    if(decoder->quant_store) {                                          \
+       if (decoder->picture_structure == TOP_FIELD)                     \
+        decoder->quant_store[2*decoder->quant_stride*(decoder->v_offset>>4) \
+                    +(decoder->offset>>4)] = decoder->quantizer_scale;  \
+       else if (decoder->picture_structure == BOTTOM_FIELD)             \
+        decoder->quant_store[2*decoder->quant_stride*(decoder->v_offset>>4) \
+	            + decoder->quant_stride                             \
+                    +(decoder->offset>>4)] = decoder->quantizer_scale;  \
+       else                                                             \
+        decoder->quant_store[decoder->quant_stride*(decoder->v_offset>>4) \
+                    +(decoder->offset>>4)] = decoder->quantizer_scale;  \
+    }                                                                   \
     decoder->offset += 16;						\
     if (decoder->offset == decoder->width) {				\
 	do { /* just so we can use the break statement */		\
@@ -1604,6 +1604,12 @@
     }									\
 } while (0)
 
+static void motion_dummy (mpeg2_decoder_t * const decoder,
+                          motion_t * const motion,
+                          mpeg2_mc_fct * const * const table)
+{
+}
+
 void mpeg2_init_fbuf (mpeg2_decoder_t * decoder, uint8_t * current_fbuf[3],
 		      uint8_t * forward_fbuf[3], uint8_t * backward_fbuf[3])
 {
@@ -1661,7 +1667,9 @@
 
     if (decoder->mpeg1) {
 	decoder->motion_parser[0] = motion_zero_420;
+        decoder->motion_parser[MC_FIELD] = motion_dummy;
 	decoder->motion_parser[MC_FRAME] = motion_mp1;
+        decoder->motion_parser[MC_DMV] = motion_dummy;
 	decoder->motion_parser[4] = motion_reuse_420;
     } else if (decoder->picture_structure == FRAME_PICTURE) {
 	if (decoder->chroma_format == 0) {
--- libmpeg2/idct.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/idct.c	2006-06-16 20:12:50.000000000 +0200
@@ -253,7 +253,7 @@
 	mpeg2_idct_mmx_init ();
     } else
 #endif
-#ifdef ARCH_PPC
+#if defined(ARCH_PPC) && defined(HAVE_ALTIVEC)
     if (accel & MPEG2_ACCEL_PPC_ALTIVEC) {
 	mpeg2_idct_copy = mpeg2_idct_copy_altivec;
 	mpeg2_idct_add = mpeg2_idct_add_altivec;
--- libmpeg2/idct_altivec.c	2004/08/02 11:26:43	12933
+++ libmpeg2/idct_altivec.c	2005/05/15 20:11:34	15484
@@ -41,7 +41,7 @@
 typedef vector signed int vector_s32_t;
 typedef vector unsigned int vector_u32_t;
 
-#if defined(HAVE_ALTIVEC_H) && (__GNUC__ * 100 + __GNUC_MINOR__ < 303)
+#if defined( HAVE_ALTIVEC_H ) && !defined( __APPLE_ALTIVEC__ ) && (__GNUC__ * 100 + __GNUC_MINOR__ < 303)
 /* work around gcc <3.3 vec_mergel bug */
 static inline vector_s16_t my_vec_mergel (vector_s16_t const A,
 					  vector_s16_t const B)
@@ -56,10 +56,10 @@
 #define vec_mergel my_vec_mergel
 #endif
 
-#ifdef HAVE_ALTIVEC_H	/* gnu */
-#define VEC_S16(a,b,c,d,e,f,g,h) {a, b, c, d, e, f, g, h}
-#else			/* apple */
+#if defined( __APPLE_CC__ ) && defined( __APPLE_ALTIVEC__ ) /* apple */
 #define VEC_S16(a,b,c,d,e,f,g,h) (vector_s16_t) (a, b, c, d, e, f, g, h)
+#else			/* gnu */
+#define VEC_S16(a,b,c,d,e,f,g,h) {a, b, c, d, e, f, g, h}
 #endif
 
 static const vector_s16_t constants ATTR_ALIGN(16) =
Index: libmpeg2/motion_comp_arm.c
===================================================================
--- libmpeg2/motion_comp_arm.c	(revision 0)
+++ libmpeg2/motion_comp_arm.c	(revision 0)
@@ -0,0 +1,187 @@
+/*
+ * motion_comp_arm.c
+ * Copyright (C) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+ *
+ * This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
+ * See http://libmpeg2.sourceforge.net/ for updates.
+ *
+ * mpeg2dec is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * mpeg2dec is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "config.h"
+
+#ifdef ARCH_ARM
+
+#include <inttypes.h>
+
+#include "mpeg2.h"
+#include "attributes.h"
+#include "mpeg2_internal.h"
+
+#define avg2(a,b) ((a+b+1)>>1)
+#define avg4(a,b,c,d) ((a+b+c+d+2)>>2)
+
+#define predict_o(i) (ref[i])
+#define predict_x(i) (avg2 (ref[i], ref[i+1]))
+#define predict_y(i) (avg2 (ref[i], (ref+stride)[i]))
+#define predict_xy(i) (avg4 (ref[i], ref[i+1], \
+			     (ref+stride)[i], (ref+stride)[i+1]))
+
+#define put(predictor,i) dest[i] = predictor (i)
+#define avg(predictor,i) dest[i] = avg2 (predictor (i), dest[i])
+
+/* mc function template */
+
+#define MC_FUNC(op,xy)							\
+static void inline MC_##op##_##xy##_16_c (uint8_t * dest, const uint8_t * ref,	\
+				   const int stride, int height)	\
+{									\
+    do {								\
+	op (predict_##xy, 0);						\
+	op (predict_##xy, 1);						\
+	op (predict_##xy, 2);						\
+	op (predict_##xy, 3);						\
+	op (predict_##xy, 4);						\
+	op (predict_##xy, 5);						\
+	op (predict_##xy, 6);						\
+	op (predict_##xy, 7);						\
+	op (predict_##xy, 8);						\
+	op (predict_##xy, 9);						\
+	op (predict_##xy, 10);						\
+	op (predict_##xy, 11);						\
+	op (predict_##xy, 12);						\
+	op (predict_##xy, 13);						\
+	op (predict_##xy, 14);						\
+	op (predict_##xy, 15);						\
+	ref += stride;							\
+	dest += stride;							\
+    } while (--height);							\
+}									\
+static void MC_##op##_##xy##_8_c (uint8_t * dest, const uint8_t * ref,	\
+				  const int stride, int height)		\
+{									\
+    do {								\
+	op (predict_##xy, 0);						\
+	op (predict_##xy, 1);						\
+	op (predict_##xy, 2);						\
+	op (predict_##xy, 3);						\
+	op (predict_##xy, 4);						\
+	op (predict_##xy, 5);						\
+	op (predict_##xy, 6);						\
+	op (predict_##xy, 7);						\
+	ref += stride;							\
+	dest += stride;							\
+    } while (--height);							\
+}									\
+/* definitions of the actual mc functions */
+
+MC_FUNC (put,o)
+MC_FUNC (avg,o)
+MC_FUNC (put,x)
+MC_FUNC (avg,x)
+MC_FUNC (put,y)
+MC_FUNC (avg,y)
+MC_FUNC (put,xy)
+MC_FUNC (avg,xy)
+
+
+extern void MC_put_o_16_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height);
+
+extern void MC_put_x_16_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height);
+
+
+static void MC_put_y_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_put_y_16_c(dest, ref, stride, height);
+}
+
+static void MC_put_xy_16_arm (uint8_t * dest, const uint8_t * ref,
+			       int stride, int height)
+{
+    MC_put_xy_16_c(dest, ref, stride, height);
+}
+
+extern void MC_put_o_8_arm (uint8_t * dest, const uint8_t * ref,
+				int stride, int height);
+
+extern void MC_put_x_8_arm (uint8_t * dest, const uint8_t * ref,
+			    int stride, int height);
+
+static void MC_put_y_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_put_y_8_c(dest, ref, stride, height);
+}
+
+static void MC_put_xy_8_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_put_xy_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_o_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_o_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_x_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_x_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_y_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_y_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_xy_16_arm (uint8_t * dest, const uint8_t * ref,
+			       int stride, int height)
+{
+    MC_avg_xy_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_o_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_avg_o_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_x_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_avg_x_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_y_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_avg_y_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_xy_8_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_xy_8_c(dest, ref, stride, height);
+}
+
+MPEG2_MC_EXTERN (arm)
+
+#endif
Index: libmpeg2/motion_comp_arm_s.S
===================================================================
--- libmpeg2/motion_comp_arm_s.S	(revision 0)
+++ libmpeg2/motion_comp_arm_s.S	(revision 0)
@@ -0,0 +1,322 @@
+@ motion_comp_arm_s.S
+@ Copyright (C) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+@
+@ This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
+@ See http://libmpeg2.sourceforge.net/ for updates.
+@
+@ mpeg2dec is free software; you can redistribute it and/or modify
+@ it under the terms of the GNU General Public License as published by
+@ the Free Software Foundation; either version 2 of the License, or
+@ (at your option) any later version.
+@
+@ mpeg2dec is distributed in the hope that it will be useful,
+@ but WITHOUT ANY WARRANTY; without even the implied warranty of
+@ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+@ GNU General Public License for more details.
+@
+@ You should have received a copy of the GNU General Public License
+@ along with this program; if not, write to the Free Software
+@ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+	.text
+	
+@ ----------------------------------------------------------------
+	.align
+	.global MC_put_o_16_arm
+MC_put_o_16_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r11, lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_o_16_arm_align_jt
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5]
+
+MC_put_o_16_arm_align0:
+	ldmia r1, {r4-r7}
+	add r1, r1, r2
+	pld [r1]
+	stmia r0, {r4-r7}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne MC_put_o_16_arm_align0
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+
+.macro	PROC shift
+	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	mov r9, r4, lsr #(\shift)
+	pld [r1]
+	mov r10, r5, lsr #(\shift)
+	orr r9, r9, r5, lsl #(32-\shift)
+	mov r11, r6, lsr #(\shift)
+	orr r10, r10, r6, lsl #(32-\shift)
+	mov r12, r7, lsr #(\shift)
+	orr r11, r11, r7, lsl #(32-\shift)
+	orr r12, r12, r8, lsl #(32-\shift)
+	stmia r0, {r9-r12}
+	subs r3, r3, #1
+	add r0, r0, r2
+.endm
+
+MC_put_o_16_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC(8)
+	bne 1b
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+MC_put_o_16_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC(16)
+	bne 1b
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+MC_put_o_16_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC(24)
+	bne 1b
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+MC_put_o_16_arm_align_jt:
+	.word MC_put_o_16_arm_align0
+	.word MC_put_o_16_arm_align1
+	.word MC_put_o_16_arm_align2
+	.word MC_put_o_16_arm_align3
+
+@ ----------------------------------------------------------------
+	.align
+	.global MC_put_o_8_arm
+MC_put_o_8_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r10, lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_o_8_arm_align_jt
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5]
+MC_put_o_8_arm_align0:
+	ldmia r1, {r4-r5}
+	add r1, r1, r2
+	pld [r1]
+	stmia r0, {r4-r5}
+	add r0, r0, r2
+	subs r3, r3, #1
+	bne MC_put_o_8_arm_align0
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+.macro	PROC8 shift
+	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	mov r9, r4, lsr #(\shift)
+	pld [r1]
+	mov r10, r5, lsr #(\shift)
+	orr r9, r9, r5, lsl #(32-\shift)
+	orr r10, r10, r6, lsl #(32-\shift)
+	stmia r0, {r9-r10}
+	subs r3, r3, #1
+	add r0, r0, r2
+.endm
+
+MC_put_o_8_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC8(8)
+	bne 1b
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+MC_put_o_8_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC8(16)
+	bne 1b
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+MC_put_o_8_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC8(24)
+	bne 1b
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+MC_put_o_8_arm_align_jt:
+	.word MC_put_o_8_arm_align0
+	.word MC_put_o_8_arm_align1
+	.word MC_put_o_8_arm_align2
+	.word MC_put_o_8_arm_align3
+
+@ ----------------------------------------------------------------
+.macro	AVG_PW rW1, rW2
+	mov \rW2, \rW2, lsl #24
+	orr \rW2, \rW2, \rW1, lsr #8
+	eor r9, \rW1, \rW2
+	and \rW2, \rW1, \rW2
+	and r10, r9, r12
+	add \rW2, \rW2, r10, lsr #1
+	and r10, r9, r11
+	add \rW2, \rW2, r10
+.endm
+
+	.align
+	.global MC_put_x_16_arm
+MC_put_x_16_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_x_16_arm_align_jt
+	ldr r11, [r5]
+	mvn r12, r11
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5, #4]
+
+.macro	ADJ_ALIGN_QW shift, R0, R1, R2, R3, R4
+	mov \R0, \R0, lsr #(\shift)
+	orr \R0, \R0, \R1, lsl #(32 - \shift)
+	mov \R1, \R1, lsr #(\shift)
+	orr \R1, \R1, \R2, lsl #(32 - \shift)
+	mov \R2, \R2, lsr #(\shift)
+	orr \R2, \R2, \R3, lsl #(32 - \shift)
+	mov \R3, \R3, lsr #(\shift)
+	orr \R3, \R3, \R4, lsl #(32 - \shift)
+	mov \R4, \R4, lsr #(\shift)
+@	and \R4, \R4, #0xFF
+.endm
+
+MC_put_x_16_arm_align0:
+	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne MC_put_x_16_arm_align0
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_QW 8, r4, r5, r6, r7, r8
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_QW 16, r4, r5, r6, r7, r8
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_QW 24, r4, r5, r6, r7, r8
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align_jt:
+	.word 0x01010101
+	.word MC_put_x_16_arm_align0
+	.word MC_put_x_16_arm_align1
+	.word MC_put_x_16_arm_align2
+	.word MC_put_x_16_arm_align3
+
+@ ----------------------------------------------------------------
+	.align
+	.global MC_put_x_8_arm
+MC_put_x_8_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_x_8_arm_align_jt
+	ldr r11, [r5]
+	mvn r12, r11
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5, #4]
+
+.macro	ADJ_ALIGN_DW shift, R0, R1, R2
+	mov \R0, \R0, lsr #(\shift)
+	orr \R0, \R0, \R1, lsl #(32 - \shift)
+	mov \R1, \R1, lsr #(\shift)
+	orr \R1, \R1, \R2, lsl #(32 - \shift)
+	mov \R2, \R2, lsr #(\shift)
+@	and \R4, \R4, #0xFF
+.endm
+
+MC_put_x_8_arm_align0:
+	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne MC_put_x_8_arm_align0
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_DW 8, r4, r5, r6
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_DW 16, r4, r5, r6
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_DW 24, r4, r5, r6
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align_jt:
+	.word 0x01010101
+	.word MC_put_x_8_arm_align0
+	.word MC_put_x_8_arm_align1
+	.word MC_put_x_8_arm_align2
+	.word MC_put_x_8_arm_align3
Index: libmpeg2/motion_comp_iwmmxt.c
===================================================================
--- libmpeg2/motion_comp_iwmmxt.c	(revision 0)
+++ libmpeg2/motion_comp_iwmmxt.c	(revision 0)
@@ -0,0 +1,61 @@
+/*
+ * motion_comp_iwmmxt.c
+ * Copyright (C) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+ *
+ * This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
+ * See http://libmpeg2.sourceforge.net/ for updates.
+ *
+ * mpeg2dec is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * mpeg2dec is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "config.h"
+
+#ifdef ARCH_ARM
+#ifdef HAVE_IWMMXT
+
+#include <inttypes.h>
+
+#include "mpeg2.h"
+#include "attributes.h"
+#include "mpeg2_internal.h"
+
+/* defined in libavcodec */
+
+extern void put_pixels16_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels16_x2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels16_y2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels16_xy2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels8_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels8_x2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels8_y2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void put_pixels8_xy2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels16_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels16_x2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels16_y2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels16_xy2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels8_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels8_x2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels8_y2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+extern void avg_pixels8_xy2_iwmmxt(uint8_t * dest, const uint8_t * ref, const int stride, int height);
+
+mpeg2_mc_t mpeg2_mc_iwmmxt = {
+    {put_pixels16_iwmmxt, put_pixels16_x2_iwmmxt, put_pixels16_y2_iwmmxt, put_pixels16_xy2_iwmmxt,
+     put_pixels8_iwmmxt, put_pixels8_x2_iwmmxt,  put_pixels8_y2_iwmmxt,  put_pixels8_xy2_iwmmxt}, \
+    {avg_pixels16_iwmmxt, avg_pixels16_x2_iwmmxt, avg_pixels16_y2_iwmmxt, avg_pixels16_xy2_iwmmxt,
+     avg_pixels8_iwmmxt, avg_pixels8_x2_iwmmxt,  avg_pixels8_y2_iwmmxt,  avg_pixels8_xy2_iwmmxt}, \
+};
+
+#endif
+#endif