view libmpeg2/libmpeg-0.4.0.diff @ 20874:b6d87b58754f

Partial fix for semitransparent glyph outlines. This fix removes semitransparent area (less then pixel width) between glyph and it's outline. Instead, it makes them overlap a little. It usually looks much better this way. Complete fix seems impossible with the current output format (single color alpha bitmaps). The right way is to blend both glyph and outline into one bitmap so that 2 pixels with 50% transparency produce a fully solid one. This requires RGBA bitmap output from libass.

--- 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>
 
@@ -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 ();
@@ -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
@@ -66,7 +70,7 @@
 } while (0)
 #endif
 
-static void inline idct_row (int16_t * const block)
+static inline void idct_row (int16_t * const block)
 {
     int d0, d1, d2, d3;
     int a0, a1, a2, a3, b0, b1, b2, b3;
@@ -119,7 +123,7 @@
     block[7] = (a0 - b0) >> 12;
 }
 
-static void inline idct_col (int16_t * const block)
+static inline void idct_col (int16_t * const block)
 {
     int d0, d1, d2, d3;
     int a0, a1, a2, a3, b0, b1, b2, b3;
@@ -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
@@ -59,7 +63,7 @@
 } while (0)
 #endif
 
-static void inline idct_row (int16_t * const block)
+static inline void idct_row (int16_t * const block)
 {
     uint64_t l, r;
     int_fast32_t d0, d1, d2, d3;
@@ -116,7 +120,7 @@
     block[7] = (a0 - b0) >> 12;
 }
 
-static void inline idct_col (int16_t * const block)
+static inline void idct_col (int16_t * const block)
 {
     int_fast32_t d0, d1, d2, d3;
     int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3;
@@ -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;
-    elseo
+    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
--- 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 {
--- 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 */		\