view libmpeg2/libmpeg2_changes.diff @ 27645:83d915449a10

Remove IWMMXT optimizations through libavcodec from libmpeg2. According to Siarhei Siamashka libavcodec is faster on ARM so it is better to use it directly instead of creating this hackish mix of two libraries. Plus, these local changes would never be acceptable upstream, so no good reason for keeping it in our local patchset remains.

--- libmpeg2/cpu_accel.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/cpu_accel.c	2006-06-16 20:12:50.000000000 +0200
@@ -29,9 +33,13 @@
 #include "attributes.h"
 #include "mpeg2_internal.h"
 
+#include "cpudetect.h"
+
 #if defined(ARCH_X86) || defined(ARCH_X86_64)
 static inline uint32_t arch_accel (uint32_t accel)
 {
+/* Use MPlayer CPU detection instead of libmpeg2 variant. */
+#if 0
     if (accel & (MPEG2_ACCEL_X86_3DNOW | MPEG2_ACCEL_X86_MMXEXT))
 	accel |= MPEG2_ACCEL_X86_MMX;
 	
@@ -124,6 +132,21 @@
 #endif /* ACCEL_DETECT */
 
     return accel;
+
+#else /* 0 */
+    accel = 0;
+    if (gCpuCaps.hasMMX)
+        accel |= MPEG2_ACCEL_X86_MMX;
+    if (gCpuCaps.hasSSE2)
+	accel |= MPEG2_ACCEL_X86_SSE2;
+    if (gCpuCaps.hasMMX2)
+	accel |= MPEG2_ACCEL_X86_MMXEXT;
+    if (gCpuCaps.has3DNow)
+	accel |= MPEG2_ACCEL_X86_3DNOW;
+
+    return accel;
+
+#endif /* 0 */
 }
 #endif /* ARCH_X86 || ARCH_X86_64 */
 
--- libmpeg2/decode.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/decode.c	2006-06-16 20:12:50.000000000 +0200
@@ -345,6 +349,15 @@
     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;
@@ -872,6 +879,7 @@
 	mpeg2dec->scaled[idx] = decoder->q_scale_type;
 	for (i = 0; i < 32; i++) {
 	    k = decoder->q_scale_type ? non_linear_scale[i] : (i << 1);
+	    decoder->quantizer_scales[i] = k;
 	    for (j = 0; j < 64; j++)
 		decoder->quantizer_prescale[idx][i][j] =
 		    k * mpeg2dec->quantizer_matrix[idx][j];
--- libmpeg2/idct.c	(revision 26652)
+++ libmpeg2/idct.c	(working copy)
@@ -235,34 +239,40 @@
 
 void mpeg2_idct_init (uint32_t accel)
 {
-#ifdef ARCH_X86
+#ifdef HAVE_SSE2
     if (accel & MPEG2_ACCEL_X86_SSE2) {
 	mpeg2_idct_copy = mpeg2_idct_copy_sse2;
 	mpeg2_idct_add = mpeg2_idct_add_sse2;
 	mpeg2_idct_mmx_init ();
-    } else if (accel & MPEG2_ACCEL_X86_MMXEXT) {
+    } else
+#elif 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
+#elif 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 ();
     } else
 #endif
-#ifdef ARCH_PPC
+#ifdef HAVE_ALTIVEC
     if (accel & MPEG2_ACCEL_PPC_ALTIVEC) {
 	mpeg2_idct_copy = mpeg2_idct_copy_altivec;
 	mpeg2_idct_add = mpeg2_idct_add_altivec;
 	mpeg2_idct_altivec_init ();
     } else
 #endif
-#ifdef ARCH_ALPHA
+#ifdef HAVE_VIS
     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
+#elif ARCH_ALPHA
+    if (accel & MPEG2_ACCEL_ALPHA) {
 	int i;
 
 	mpeg2_idct_copy = mpeg2_idct_copy_alpha;
--- libmpeg2/motion_comp.c	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/motion_comp.c	2006-06-16 20:12:50.000000000 +0200
@@ -33,16 +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
+#ifdef HAVE_ALTIVEC
     if (accel & MPEG2_ACCEL_PPC_ALTIVEC)
 	mpeg2_mc = mpeg2_mc_altivec;
     else
@@ -52,15 +62,15 @@
 	mpeg2_mc = mpeg2_mc_alpha;
     else
 #endif
-#ifdef ARCH_SPARC
+#ifdef HAVE_VIS
     if (accel & MPEG2_ACCEL_SPARC_VIS)
 	mpeg2_mc = mpeg2_mc_vis;
     else
 #endif
 #ifdef ARCH_ARM
-    if (accel & MPEG2_ACCEL_ARM) {
+    if (accel & MPEG2_ACCEL_ARM)
 	mpeg2_mc = mpeg2_mc_arm;
-    } else
+    else
 #endif
 	mpeg2_mc = mpeg2_mc_c;
 }
--- libmpeg2/mpeg2_internal.h	2006-06-16 20:12:26.000000000 +0200
+++ libmpeg2/mpeg2_internal.h	2006-06-16 20:12:50.000000000 +0200
@@ -152,6 +156,11 @@
 
     /* XXX: stuff due to xine shit */
     int8_t q_scale_type;
+
+    int quantizer_scales[32];
+    int quantizer_scale;
+    char* quant_store;
+    int quant_stride;
 };
 
 typedef struct {
@@ -223,6 +232,9 @@
     //int8_t q_scale_type, scaled[4];
     uint8_t quantizer_matrix[4][64];
     uint8_t new_quantizer_matrix[4][64];
+
+    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];
@@ -1564,6 +1569,24 @@
 
 #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 */		\