Sync with mplayer's stuff

Originally committed as revision 14 to svn://svn.ffmpeg.org/ffmpeg/trunk

3 files changed, 257 insertions, 16 deletions

diff --git a/libavcodec/i386/dsputil_mmx.c b/libavcodec/i386/dsputil_mmx.c
index 18cb03e0e8..d4a07c3a77 100644
--- a/libavcodec/i386/dsputil_mmx.c
+++ b/libavcodec/i386/dsputil_mmx.c
@@ -30,8 +30,10 @@ int pix_abs16x16_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
 int pix_abs16x16_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
 
 /* pixel operations */
-static const unsigned short mm_wone[4] __attribute__ ((aligned(8))) = { 0x1, 0x1, 0x1, 0x1 };
-static const unsigned short mm_wtwo[4] __attribute__ ((aligned(8))) = { 0x2, 0x2, 0x2, 0x2 };
+static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001;
+static const unsigned long long int mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002;
+//static const unsigned short mm_wone[4] __attribute__ ((aligned(8))) = { 0x1, 0x1, 0x1, 0x1 };
+//static const unsigned short mm_wtwo[4] __attribute__ ((aligned(8))) = { 0x2, 0x2, 0x2, 0x2 };
 
 /***********************************/
 /* 3Dnow specific */
@@ -215,7 +217,7 @@ static void put_pixels_x2_mmx(UINT8 *block, const UINT8 *pixels, int line_size,
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm4\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
@@ -250,7 +252,7 @@ static void put_pixels_y2_mmx(UINT8 *block, const UINT8 *pixels, int line_size,
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm4\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
@@ -287,7 +289,7 @@ static void put_pixels_xy2_mmx(UINT8 *block, const UINT8 *pixels, int line_size,
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wtwo[0]):"memory");
+	::"m"(mm_wtwo):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
@@ -399,7 +401,7 @@ static void   put_no_rnd_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
@@ -448,7 +450,7 @@ static void avg_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%0, %%mm0\n\t"
@@ -485,7 +487,7 @@ static void   avg_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_si
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm1\n\t"
@@ -531,7 +533,7 @@ static void   avg_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_si
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm1\n\t"
@@ -577,7 +579,7 @@ static void   avg_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_s
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wtwo[0]):"memory");
+	::"m"(mm_wtwo):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
@@ -621,7 +623,7 @@ static void   avg_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_s
 	"movq	%%mm0, %0\n\t"
 	:"=m"(*p)
 	:"m"(*pix),
-	 "m"(*(pix+line_size)), "m"(mm_wone[0])
+	 "m"(*(pix+line_size)), "m"(mm_wone)
 	:"memory");
    pix += line_size;
    p +=   line_size ;
@@ -748,7 +750,7 @@ static void   avg_no_rnd_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wone[0]):"memory");
+	::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
@@ -832,7 +834,7 @@ static void sub_pixels_x2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_si
   __asm __volatile(
       "pxor	%%mm7, %%mm7\n\t"
       "movq	%0, %%mm6"
-      ::"m"(mm_wone[0]):"memory");
+      ::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%0, %%mm0\n\t"
@@ -872,7 +874,7 @@ static void sub_pixels_y2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_si
   __asm __volatile(
       "pxor	%%mm7, %%mm7\n\t"
       "movq	%0, %%mm6"
-      ::"m"(mm_wone[0]):"memory");
+      ::"m"(mm_wone):"memory");
   do {
     __asm __volatile(
 	"movq	%0, %%mm0\n\t"
@@ -912,7 +914,7 @@ static void   sub_pixels_xy2_mmx( DCTELEM  *block, const UINT8 *pixels, int line
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wtwo[0]):"memory");
+	::"m"(mm_wtwo):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
diff --git a/libavcodec/i386/dsputil_mmx_avg.h b/libavcodec/i386/dsputil_mmx_avg.h
index 99e806a714..5cd640f713 100644
--- a/libavcodec/i386/dsputil_mmx_avg.h
+++ b/libavcodec/i386/dsputil_mmx_avg.h
@@ -243,7 +243,7 @@ static void DEF(avg_pixels_xy2)( UINT8  *block, const UINT8 *pixels, int line_si
   __asm __volatile(
 	"pxor	%%mm7, %%mm7\n\t"
 	"movq	%0, %%mm6\n\t"
-	::"m"(mm_wtwo[0]):"memory");
+	::"m"(mm_wtwo):"memory");
   do {
     __asm __volatile(
 	"movq	%1, %%mm0\n\t"
diff --git a/libavcodec/i386/mpegvideo.c b/libavcodec/i386/mpegvideo.c
new file mode 100644
index 0000000000..d9e5e5c114
--- /dev/null
+++ b/libavcodec/i386/mpegvideo.c
@@ -0,0 +1,239 @@
+/*
+ * The simplest mpeg encoder (well, it was the simplest!)
+ * Copyright (c) 2000,2001 Gerard Lantau.
+ *
+ * This program 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.
+ *
+ * This program 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., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * Optimized for ia32 cpus by Nick Kurshev <nickols_k@mail.ru>
+ */
+
+void MPV_frame_start(MpegEncContext *s)
+{
+    if (s->pict_type == B_TYPE) {
+	__asm __volatile(
+	    "movl	(%1), %%eax\n\t"
+	    "movl	4(%1), %%edx\n\t"
+	    "movl	8(%1), %%ecx\n\t"
+	    "movl	%%eax, (%0)\n\t"
+	    "movl	%%edx, 4(%0)\n\t"
+	    "movl	%%ecx, 8(%0)\n\t"
+	    :
+	    :"r"(s->current_picture), "r"(s->aux_picture)
+	    :"eax","edx","ecx","memory");
+    } else {
+            /* swap next and last */
+	__asm __volatile(
+	    "movl	(%1), %%eax\n\t"
+	    "movl	4(%1), %%edx\n\t"
+	    "movl	8(%1), %%ecx\n\t"
+	    "xchgl	(%0), %%eax\n\t"
+	    "xchgl	4(%0), %%edx\n\t"
+	    "xchgl	8(%0), %%ecx\n\t"
+	    "movl	%%eax, (%1)\n\t"
+	    "movl	%%edx, 4(%1)\n\t"
+	    "movl	%%ecx, 8(%1)\n\t"
+	    "movl	%%eax, (%2)\n\t"
+	    "movl	%%edx, 4(%2)\n\t"
+	    "movl	%%ecx, 8(%2)\n\t"
+	    :
+	    :"r"(s->last_picture), "r"(s->next_picture), "r"(s->current_picture)
+	    :"eax","edx","ecx","memory");
+    }
+}
+
+static void dct_unquantize(MpegEncContext *s, DCTELEM *block, int n, int qscale);
+
+#ifdef HAVE_MMX 
+static const unsigned long long int mm_wabs __attribute__ ((aligned(8))) = 0xffffffffffffffffULL;
+static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
+
+/*
+  NK:
+  Note: looking at PARANOID:
+  "enable all paranoid tests for rounding, overflows, etc..."
+
+#ifdef PARANOID
+                if (level < -2048 || level > 2047)
+                    fprintf(stderr, "unquant error %d %d\n", i, level);
+#endif
+  We can suppose that result of two multiplications can't be greate of 0xFFFF
+  i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
+  a complex multiplication.
+=====================================================
+ Full formula for multiplication of 2 integer numbers
+ which are represent as high:low words:
+ input: value1 = high1:low1
+        value2 = high2:low2
+ output: value3 = value1*value2
+ value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
+ this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
+ but this algorithm will compute only 0x66cb0ce4
+ this limited by 16-bit size of operands
+ ---------------------------------
+ tlow1 = high1*low2
+ tlow2 = high2*low1
+ tlow1 = tlow1 + tlow2
+ high3:low3 = low1*low2
+ high3 += tlow1
+*/
+#ifdef BIN_PORTABILITY
+static void dct_unquantize_mmx
+#else
+#define HAVE_DCT_UNQUANTIZE 1
+static void dct_unquantize
+#endif
+(MpegEncContext *s,DCTELEM *block, int n, int qscale)
+{
+    int i, level;
+    const UINT16 *quant_matrix;
+    if (s->mb_intra) {
+        if (n < 4) 
+            block[0] = block[0] * s->y_dc_scale;
+        else
+            block[0] = block[0] * s->c_dc_scale;
+        if (s->out_format == FMT_H263) {
+            i = 1;
+            goto unquant_even;
+        }
+        /* XXX: only mpeg1 */
+        quant_matrix = s->intra_matrix;
+	i=1;
+	/* Align on 4 elements boundary */
+	while(i&3)
+	{
+            level = block[i];
+            if (level) {
+                if (level < 0) level = -level;
+                    level = (int)(level * qscale * quant_matrix[i]) >> 3;
+                    level = (level - 1) | 1;
+                if (block[i] < 0) level = -level;
+                block[i] = level;
+            }
+	    i++;
+	}
+	__asm __volatile(
+	"movd	%0, %%mm6\n\t"       /* mm6 = qscale | 0  */
+	"punpckldq %%mm6, %%mm6\n\t" /* mm6 = qscale | qscale */
+	"movq	%2, %%mm4\n\t"
+	"movq	%%mm6, %%mm7\n\t"
+	"movq	%1, %%mm5\n\t"
+	"packssdw %%mm6, %%mm7\n\t" /* mm7 = qscale | qscale | qscale | qscale */
+	"pxor	%%mm6, %%mm6\n\t"
+	::"g"(qscale),"m"(mm_wone),"m"(mm_wabs):"memory");
+        for(;i<64;i+=4) {
+		__asm __volatile(
+			"movq	%1, %%mm0\n\t"
+			"movq	%%mm7, %%mm1\n\t"
+			"movq	%%mm0, %%mm2\n\t"
+			"movq	%%mm0, %%mm3\n\t"
+			"pcmpgtw %%mm6, %%mm2\n\t"
+			"pmullw	%2, %%mm1\n\t"
+			"pandn	%%mm4, %%mm2\n\t"
+			"por	%%mm5, %%mm2\n\t"  
+			"pmullw	%%mm2, %%mm0\n\t" /* mm0 = abs(block[i]). */
+			
+			"pcmpeqw %%mm6, %%mm3\n\t"
+			"pmullw	%%mm0, %%mm1\n\t"
+			"psraw	$3, %%mm1\n\t"
+			"psubw	%%mm5, %%mm1\n\t"   /* block[i] --; */
+			"pandn	%%mm4, %%mm3\n\t"  /* fake of pcmpneqw : mm0 != 0 then mm1 = -1 */
+			"por	%%mm5, %%mm1\n\t"   /* block[i] |= 1 */
+			"pmullw %%mm2, %%mm1\n\t"   /* change signs again */
+
+			"pand	%%mm3, %%mm1\n\t" /* nullify if was zero */
+			"movq	%%mm1, %0"
+			:"=m"(block[i])
+			:"m"(block[i]), "m"(quant_matrix[i])
+			:"memory");
+        }
+    } else {
+        i = 0;
+    unquant_even:
+        quant_matrix = s->non_intra_matrix;
+	/* Align on 4 elements boundary */
+	while(i&3)
+	{
+	    level = block[i];
+            if (level) {
+                if (level < 0) level = -level;
+                    level = (((level << 1) + 1) * qscale *
+                             ((int) quant_matrix[i])) >> 4;
+                    level = (level - 1) | 1;
+                if(block[i] < 0) level = -level;
+                block[i] = level;
+	    }
+	    i++;
+	}
+	__asm __volatile(
+	"movd	%0, %%mm6\n\t"       /* mm6 = qscale | 0  */
+	"punpckldq %%mm6, %%mm6\n\t" /* mm6 = qscale | qscale */
+	"movq	%2, %%mm4\n\t"
+	"movq	%%mm6, %%mm7\n\t"
+	"movq	%1, %%mm5\n\t"
+	"packssdw %%mm6, %%mm7\n\t" /* mm7 = qscale | qscale | qscale | qscale */
+	"pxor	%%mm6, %%mm6\n\t"
+	::"g"(qscale),"m"(mm_wone),"m"(mm_wabs):"memory");
+        for(;i<64;i+=4) {
+		__asm __volatile(
+			"movq	%1, %%mm0\n\t"
+			"movq	%%mm7, %%mm1\n\t"
+			"movq	%%mm0, %%mm2\n\t"
+			"movq	%%mm0, %%mm3\n\t"
+			"pcmpgtw %%mm6, %%mm2\n\t"
+			"pmullw	%2, %%mm1\n\t"
+			"pandn	%%mm4, %%mm2\n\t"
+			"por	%%mm5, %%mm2\n\t"  
+			"pmullw	%%mm2, %%mm0\n\t" /* mm0 = abs(block[i]). */
+			"psllw	$1, %%mm0\n\t" /* block[i] <<= 1 */
+			"paddw	%%mm5, %%mm0\n\t" /* block[i] ++ */
+			
+			"pmullw	%%mm0, %%mm1\n\t"
+			"psraw	$4, %%mm1\n\t"
+			"pcmpeqw %%mm6, %%mm3\n\t"
+			"psubw	%%mm5, %%mm1\n\t"   /* block[i] --; */
+			"pandn	%%mm4, %%mm3\n\t"  /* fake of pcmpneqw : mm0 != 0 then mm1 = -1 */
+			"por	%%mm5, %%mm1\n\t"   /* block[i] |= 1 */
+			"pmullw %%mm2, %%mm1\n\t"   /* change signs again */
+
+			"pand	%%mm3, %%mm1\n\t" /* nullify if was zero */
+			"movq	%%mm1, %0"
+			:"=m"(block[i])
+			:"m"(block[i]), "m"(quant_matrix[i])
+			:"memory");
+        }
+    }
+}
+
+#ifdef BIN_PORTABILITY
+static void (*dct_unquantize_ptr)(MpegEncContext *s, 
+                		    DCTELEM *block, int n, int qscale);
+
+void MPV_common_init_mmx(void)
+{
+  int mm_flags;
+  mm_flags = mm_support();
+  if (mm_flags & MM_MMX) {
+    dct_unquantize_ptr = dct_unquantize_mmx;
+  }
+  else {
+    dct_unquantize_ptr = dct_unquantize;
+  }
+}
+
+#define DCT_UNQUANTIZE(a,b,c,d) (*dct_unquantize_ptr)(a,b,c,d)
+#else
+#define DCT_UNQUANTIZE(a,b,c,d) dct_unquantize(a,b,c,d)
+#endif /* BIN_PORTABILITY */
+#endif /* HAVE_MMX */