1. ffmpeg
  2. Commit 826f429ae9990632a04b06bd375afa54ffe54b76
Browse code

qpel in mmx2/3dnow qpel refinement quality parameter

Originally committed as revision 1393 to svn://svn.ffmpeg.org/ffmpeg/trunk

Michael Niedermayer authored on 2003/01/06 00:57:10
Showing 11 changed files
libavcodec/avcodec.h
History View file @ 826f429
... ... 
@@ -5,8 +5,8 @@
5 5
6 6 
 #define LIBAVCODEC_VERSION_INT 0x000406
7 7 
 #define LIBAVCODEC_VERSION     "0.4.6"
8 
-#define LIBAVCODEC_BUILD       4651
9 
-#define LIBAVCODEC_BUILD_STR   "4651"
8 
+#define LIBAVCODEC_BUILD       4652
9 
+#define LIBAVCODEC_BUILD_STR   "4652"
10 10
11 11 
 enum CodecID {
12 12 
     CODEC_ID_NONE,
... ... 
@@ -909,7 +909,7 @@ typedef struct AVCodecContext {
909 909 
      * decoding: unused
910 910 
      */
911 911 
     int me_pre_cmp;
912 
-
912 
+
913 913 
     /**
914 914 
      * ME pre pass diamond size & shape
915 915 
      * encoding: set by user.
... ... 
@@ -917,6 +917,13 @@ typedef struct AVCodecContext {
917 917 
      */
918 918 
     int pre_dia_size;
919 919
920 
+    /**
921 
+     * subpel ME quality
922 
+     * encoding: set by user.
923 
+     * decoding: unused
924 
+     */
925 
+    int me_subpel_quality;
926 
+
920 927 
 } AVCodecContext;
921 928
922 929 
 typedef struct AVCodec {
libavcodec/dsputil.c
History View file @ 826f429
... ... 
@@ -781,6 +781,7 @@ static inline void copy_block9(UINT8 *dst, UINT8 *src, int dstStride, int srcStr
781 781 
     }
782 782 
 }
783 783
784 
+
784 785 
 #define QPEL_MC(r, OPNAME, RND, OP) \
785 786 
 static void OPNAME ## mpeg4_qpel8_h_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int h){\
786 787 
     UINT8 *cm = cropTbl + MAX_NEG_CROP;\
... ... 
@@ -830,6 +831,7 @@ static void OPNAME ## mpeg4_qpel8_v_lowpass(UINT8 *dst, UINT8 *src, int dstStrid
830 830 
 static void OPNAME ## mpeg4_qpel16_h_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int h){\
831 831 
     UINT8 *cm = cropTbl + MAX_NEG_CROP;\
832 832 
     int i;\
833 
+    \
833 834 
     for(i=0; i<h; i++)\
834 835 
     {\
835 836 
         OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\
... ... 
@@ -853,9 +855,10 @@ static void OPNAME ## mpeg4_qpel16_h_lowpass(UINT8 *dst, UINT8 *src, int dstStri
853 853 
     }\
854 854 
 }\
855 855 
 \
856 
-static void OPNAME ## mpeg4_qpel16_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int w){\
856 
+static void OPNAME ## mpeg4_qpel16_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride){\
857 857 
     UINT8 *cm = cropTbl + MAX_NEG_CROP;\
858 858 
     int i;\
859 
+    const int w=16;\
859 860 
     for(i=0; i<w; i++)\
860 861 
     {\
861 862 
         const int src0= src[0*srcStride];\
... ... 
@@ -1046,21 +1049,21 @@ static void OPNAME ## qpel16_mc01_c(UINT8 *dst, UINT8 *src, int stride){\
1046 1046 
     UINT8 full[24*17];\
1047 1047 
     UINT8 half[256];\
1048 1048 
     copy_block17(full, src, 24, stride, 17);\
1049 
-    put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24, 16);\
1049 
+    put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
1050 1050 
     OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\
1051 1051 
 }\
1052 1052 
 \
1053 1053 
 static void OPNAME ## qpel16_mc02_c(UINT8 *dst, UINT8 *src, int stride){\
1054 1054 
     UINT8 full[24*17];\
1055 1055 
     copy_block17(full, src, 24, stride, 17);\
1056 
-    OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24, 16);\
1056 
+    OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\
1057 1057 
 }\
1058 1058 
 \
1059 1059 
 static void OPNAME ## qpel16_mc03_c(UINT8 *dst, UINT8 *src, int stride){\
1060 1060 
     UINT8 full[24*17];\
1061 1061 
     UINT8 half[256];\
1062 1062 
     copy_block17(full, src, 24, stride, 17);\
1063 
-    put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24, 16);\
1063 
+    put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
1064 1064 
     OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\
1065 1065 
 }\
1066 1066 
 static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
... ... 
@@ -1070,8 +1073,8 @@ static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
1070 1070 
     UINT8 halfHV[256];\
1071 1071 
     copy_block17(full, src, 24, stride, 17);\
1072 1072 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
1073 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24, 16);\
1074 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1073 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
1074 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1075 1075 
     OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
1076 1076 
 }\
1077 1077 
 static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
... ... 
@@ -1081,8 +1084,8 @@ static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
1081 1081 
     UINT8 halfHV[256];\
1082 1082 
     copy_block17(full, src, 24, stride, 17);\
1083 1083 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
1084 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24, 16);\
1085 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1084 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
1085 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1086 1086 
     OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
1087 1087 
 }\
1088 1088 
 static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
... ... 
@@ -1092,8 +1095,8 @@ static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
1092 1092 
     UINT8 halfHV[256];\
1093 1093 
     copy_block17(full, src, 24, stride, 17);\
1094 1094 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
1095 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24, 16);\
1096 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1095 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
1096 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1097 1097 
     OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
1098 1098 
 }\
1099 1099 
 static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
... ... 
@@ -1103,22 +1106,22 @@ static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
1103 1103 
     UINT8 halfHV[256];\
1104 1104 
     copy_block17(full, src, 24, stride, 17);\
1105 1105 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full  , 16, 24, 17);\
1106 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24, 16);\
1107 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1106 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
1107 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1108 1108 
     OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
1109 1109 
 }\
1110 1110 
 static void OPNAME ## qpel16_mc21_c(UINT8 *dst, UINT8 *src, int stride){\
1111 1111 
     UINT8 halfH[272];\
1112 1112 
     UINT8 halfHV[256];\
1113 1113 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
1114 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1114 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1115 1115 
     OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\
1116 1116 
 }\
1117 1117 
 static void OPNAME ## qpel16_mc23_c(UINT8 *dst, UINT8 *src, int stride){\
1118 1118 
     UINT8 halfH[272];\
1119 1119 
     UINT8 halfHV[256];\
1120 1120 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
1121 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1121 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1122 1122 
     OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\
1123 1123 
 }\
1124 1124 
 static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
... ... 
@@ -1128,8 +1131,8 @@ static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
1128 1128 
     UINT8 halfHV[256];\
1129 1129 
     copy_block17(full, src, 24, stride, 17);\
1130 1130 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
1131 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24, 16);\
1132 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1131 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
1132 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1133 1133 
     OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\
1134 1134 
 }\
1135 1135 
 static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
... ... 
@@ -1139,14 +1142,14 @@ static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
1139 1139 
     UINT8 halfHV[256];\
1140 1140 
     copy_block17(full, src, 24, stride, 17);\
1141 1141 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
1142 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24, 16);\
1143 
-    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16, 16);\
1142 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
1143 
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
1144 1144 
     OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\
1145 1145 
 }\
1146 1146 
 static void OPNAME ## qpel16_mc22_c(UINT8 *dst, UINT8 *src, int stride){\
1147 1147 
     UINT8 halfH[272];\
1148 1148 
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
1149 
-    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16, 16);\
1149 
+    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
1150 1150 
 }
1151 1151
1152 1152 
 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1)
libavcodec/dsputil.h
History View file @ 826f429
... ... 
@@ -102,6 +102,7 @@ typedef struct DSPContext {
102 102 
     me_cmp_func quant_psnr[2];
103 103 
     int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
104 104
105 
+    me_cmp_func me_pre_cmp[11];
105 106 
     me_cmp_func me_cmp[11];
106 107 
     me_cmp_func me_sub_cmp[11];
107 108 
     me_cmp_func mb_cmp[11];
libavcodec/i386/dsputil_mmx.c
History View file @ 826f429
... ... 
@@ -53,6 +53,11 @@ static const uint64_t mm_bone __attribute__ ((aligned(8))) = 0x0101010101010101U
53 53 
 static const uint64_t mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
54 54 
 static const uint64_t mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002ULL;
55 55
56 
+static const uint64_t ff_pw_20 __attribute__ ((aligned(8))) = 0x0014001400140014ULL;
57 
+static const uint64_t ff_pw_3  __attribute__ ((aligned(8))) = 0x0003000300030003ULL;
58 
+static const uint64_t ff_pw_16 __attribute__ ((aligned(8))) = 0x0010001000100010ULL;
59 
+static const uint64_t ff_pw_15 __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;
60 
+
56 61 
 #define JUMPALIGN() __asm __volatile (".balign 8"::)
57 62 
 #define MOVQ_ZERO(regd)  __asm __volatile ("pxor %%" #regd ", %%" #regd ::)
58 63
... ... 
@@ -646,10 +651,698 @@ static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride)
646 646
647 647 
 WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)
648 648
649 
+#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
650 
+        "paddw " #m4 ", " #m3 "		\n\t" /* x1 */\
651 
+        "movq " #pw_20 ", %%mm4		\n\t" /* 20 */\
652 
+        "pmullw " #m3 ", %%mm4		\n\t" /* 20x1 */\
653 
+        "movq "#in7", " #m3 "		\n\t" /* d */\
654 
+        "movq "#in0", %%mm5		\n\t" /* D */\
655 
+        "paddw " #m3 ", %%mm5		\n\t" /* x4 */\
656 
+        "psubw %%mm5, %%mm4		\n\t" /* 20x1 - x4 */\
657 
+        "movq "#in1", %%mm5		\n\t" /* C */\
658 
+        "movq "#in2", %%mm6		\n\t" /* B */\
659 
+        "paddw " #m6 ", %%mm5		\n\t" /* x3 */\
660 
+        "paddw " #m5 ", %%mm6		\n\t" /* x2 */\
661 
+        "paddw %%mm6, %%mm6		\n\t" /* 2x2 */\
662 
+        "psubw %%mm6, %%mm5		\n\t" /* -2x2 + x3 */\
663 
+        "pmullw " #pw_3 ", %%mm5	\n\t" /* -6x2 + 3x3 */\
664 
+        "paddw " #rnd ", %%mm4		\n\t" /* x2 */\
665 
+        "paddw %%mm4, %%mm5		\n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
666 
+        "psraw $5, %%mm5		\n\t"\
667 
+        "packuswb %%mm5, %%mm5		\n\t"\
668 
+        OP(%%mm5, out, %%mm7, d)
669 
+
670 
+#define QPEL_BASE(OPNAME, ROUNDER, RND, OP)\
671 
+void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
672 
+    uint64_t temp;\
673 
+\
674 
+    asm volatile(\
675 
+        "pxor %%mm7, %%mm7		\n\t"\
676 
+        "1:				\n\t"\
677 
+        "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\
678 
+        "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\
679 
+        "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\
680 
+        "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\
681 
+        "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\
682 
+        "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\
683 
+        "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\
684 
+        "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\
685 
+        "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\
686 
+        "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\
687 
+        "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\
688 
+        "psllq $24, %%mm4		\n\t" /* 000ABCDE */\
689 
+        "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\
690 
+        "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\
691 
+        "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\
692 
+        "paddw %%mm3, %%mm5		\n\t" /* b */\
693 
+        "paddw %%mm2, %%mm6		\n\t" /* c */\
694 
+        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
695 
+        "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\
696 
+        "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\
697 
+        "pmullw %6, %%mm6		\n\t" /* 3c - 6b */\
698 
+        "paddw %%mm4, %%mm0		\n\t" /* a */\
699 
+        "paddw %%mm1, %%mm5		\n\t" /* d */\
700 
+        "pmullw %5, %%mm0		\n\t" /* 20a */\
701 
+        "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\
702 
+        "paddw %8, %%mm6		\n\t"\
703 
+        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
704 
+        "psraw $5, %%mm0		\n\t"\
705 
+        "movq %%mm0, %7			\n\t"\
706 
+        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
707 
+        \
708 
+        "movq 5(%0), %%mm0		\n\t" /* FGHIJKLM */\
709 
+        "movq %%mm0, %%mm5		\n\t" /* FGHIJKLM */\
710 
+        "movq %%mm0, %%mm6		\n\t" /* FGHIJKLM */\
711 
+        "psrlq $8, %%mm0		\n\t" /* GHIJKLM0 */\
712 
+        "psrlq $16, %%mm5		\n\t" /* HIJKLM00 */\
713 
+        "punpcklbw %%mm7, %%mm0		\n\t" /* 0G0H0I0J */\
714 
+        "punpcklbw %%mm7, %%mm5		\n\t" /* 0H0I0J0K */\
715 
+        "paddw %%mm0, %%mm2		\n\t" /* b */\
716 
+        "paddw %%mm5, %%mm3		\n\t" /* c */\
717 
+        "paddw %%mm2, %%mm2		\n\t" /* 2b */\
718 
+        "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\
719 
+        "movq %%mm6, %%mm2		\n\t" /* FGHIJKLM */\
720 
+        "psrlq $24, %%mm6		\n\t" /* IJKLM000 */\
721 
+        "punpcklbw %%mm7, %%mm2		\n\t" /* 0F0G0H0I */\
722 
+        "punpcklbw %%mm7, %%mm6		\n\t" /* 0I0J0K0L */\
723 
+        "pmullw %6, %%mm3		\n\t" /* 3c - 6b */\
724 
+        "paddw %%mm2, %%mm1		\n\t" /* a */\
725 
+        "paddw %%mm6, %%mm4		\n\t" /* d */\
726 
+        "pmullw %5, %%mm1		\n\t" /* 20a */\
727 
+        "psubw %%mm4, %%mm3		\n\t" /* - 6b +3c - d */\
728 
+        "paddw %8, %%mm1		\n\t"\
729 
+        "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b +3c - d */\
730 
+        "psraw $5, %%mm3		\n\t"\
731 
+        "movq %7, %%mm1			\n\t"\
732 
+        "packuswb %%mm3, %%mm1		\n\t"\
733 
+        OP(%%mm1, (%1),%%mm4, q)\
734 
+        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
735 
+        \
736 
+        "movq 9(%0), %%mm1		\n\t" /* JKLMNOPQ */\
737 
+        "movq %%mm1, %%mm4		\n\t" /* JKLMNOPQ */\
738 
+        "movq %%mm1, %%mm3		\n\t" /* JKLMNOPQ */\
739 
+        "psrlq $8, %%mm1		\n\t" /* KLMNOPQ0 */\
740 
+        "psrlq $16, %%mm4		\n\t" /* LMNOPQ00 */\
741 
+        "punpcklbw %%mm7, %%mm1		\n\t" /* 0K0L0M0N */\
742 
+        "punpcklbw %%mm7, %%mm4		\n\t" /* 0L0M0N0O */\
743 
+        "paddw %%mm1, %%mm5		\n\t" /* b */\
744 
+        "paddw %%mm4, %%mm0		\n\t" /* c */\
745 
+        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
746 
+        "psubw %%mm5, %%mm0		\n\t" /* c - 2b */\
747 
+        "movq %%mm3, %%mm5		\n\t" /* JKLMNOPQ */\
748 
+        "psrlq $24, %%mm3		\n\t" /* MNOPQ000 */\
749 
+        "pmullw %6, %%mm0		\n\t" /* 3c - 6b */\
750 
+        "punpcklbw %%mm7, %%mm3		\n\t" /* 0M0N0O0P */\
751 
+        "paddw %%mm3, %%mm2		\n\t" /* d */\
752 
+        "psubw %%mm2, %%mm0		\n\t" /* -6b + 3c - d */\
753 
+        "movq %%mm5, %%mm2		\n\t" /* JKLMNOPQ */\
754 
+        "punpcklbw %%mm7, %%mm2		\n\t" /* 0J0K0L0M */\
755 
+        "punpckhbw %%mm7, %%mm5		\n\t" /* 0N0O0P0Q */\
756 
+        "paddw %%mm2, %%mm6		\n\t" /* a */\
757 
+        "pmullw %5, %%mm6		\n\t" /* 20a */\
758 
+        "paddw %8, %%mm0		\n\t"\
759 
+        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
760 
+        "psraw $5, %%mm0		\n\t"\
761 
+        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
762 
+        \
763 
+        "paddw %%mm5, %%mm3		\n\t" /* a */\
764 
+        "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0O0P0Q0Q */\
765 
+        "paddw %%mm4, %%mm6		\n\t" /* b */\
766 
+        "pshufw $0xBE, %%mm5, %%mm4	\n\t" /* 0P0Q0Q0P */\
767 
+        "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0Q0Q0P0O */\
768 
+        "paddw %%mm1, %%mm4		\n\t" /* c */\
769 
+        "paddw %%mm2, %%mm5		\n\t" /* d */\
770 
+        "paddw %%mm6, %%mm6		\n\t" /* 2b */\
771 
+        "psubw %%mm6, %%mm4		\n\t" /* c - 2b */\
772 
+        "pmullw %5, %%mm3		\n\t" /* 20a */\
773 
+        "pmullw %6, %%mm4		\n\t" /* 3c - 6b */\
774 
+        "psubw %%mm5, %%mm3		\n\t" /* -6b + 3c - d */\
775 
+        "paddw %8, %%mm4		\n\t"\
776 
+        "paddw %%mm3, %%mm4		\n\t" /* 20a - 6b + 3c - d */\
777 
+        "psraw $5, %%mm4		\n\t"\
778 
+        "packuswb %%mm4, %%mm0		\n\t"\
779 
+        OP(%%mm0, 8(%1), %%mm4, q)\
780 
+        \
781 
+        "addl %3, %0			\n\t"\
782 
+        "addl %4, %1			\n\t"\
783 
+        "decl %2			\n\t"\
784 
+        " jnz 1b				\n\t"\
785 
+        : "+r"(src), "+r"(dst), "+g"(h)\
786 
+        : "r"(srcStride), "r"(dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(temp), "m"(ROUNDER)\
787 
+    );\
788 
+}\
789 
+\
790 
+static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
791 
+    int i;\
792 
+    int16_t temp[16];\
793 
+    /* quick HACK, XXX FIXME MUST be optimized */\
794 
+    for(i=0; i<h; i++)\
795 
+    {\
796 
+        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
797 
+        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
798 
+        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
799 
+        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
800 
+        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
801 
+        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
802 
+        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
803 
+        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
804 
+        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
805 
+        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
806 
+        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
807 
+        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
808 
+        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
809 
+        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
810 
+        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
811 
+        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
812 
+        asm volatile(\
813 
+            "movq (%0), %%mm0		\n\t"\
814 
+            "movq 8(%0), %%mm1		\n\t"\
815 
+            "paddw %2, %%mm0		\n\t"\
816 
+            "paddw %2, %%mm1		\n\t"\
817 
+            "psraw $5, %%mm0		\n\t"\
818 
+            "psraw $5, %%mm1		\n\t"\
819 
+            "packuswb %%mm1, %%mm0	\n\t"\
820 
+            OP(%%mm0, (%1), %%mm1, q)\
821 
+            "movq 16(%0), %%mm0		\n\t"\
822 
+            "movq 24(%0), %%mm1		\n\t"\
823 
+            "paddw %2, %%mm0		\n\t"\
824 
+            "paddw %2, %%mm1		\n\t"\
825 
+            "psraw $5, %%mm0		\n\t"\
826 
+            "psraw $5, %%mm1		\n\t"\
827 
+            "packuswb %%mm1, %%mm0	\n\t"\
828 
+            OP(%%mm0, 8(%1), %%mm1, q)\
829 
+            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
830 
+        );\
831 
+        dst+=dstStride;\
832 
+        src+=srcStride;\
833 
+    }\
834 
+}\
835 
+\
836 
+void OPNAME ## mpeg4_qpel16_v_lowpass_mmx(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
837 
+    uint64_t temp[17*4];\
838 
+    uint64_t *temp_ptr= temp;\
839 
+    int count= 17;\
840 
+\
841 
+    /*FIXME unroll */\
842 
+    asm volatile(\
843 
+        "pxor %%mm7, %%mm7		\n\t"\
844 
+        "1:				\n\t"\
845 
+        "movq (%0), %%mm0		\n\t"\
846 
+        "movq (%0), %%mm1		\n\t"\
847 
+        "movq 8(%0), %%mm2		\n\t"\
848 
+        "movq 8(%0), %%mm3		\n\t"\
849 
+        "punpcklbw %%mm7, %%mm0		\n\t"\
850 
+        "punpckhbw %%mm7, %%mm1		\n\t"\
851 
+        "punpcklbw %%mm7, %%mm2		\n\t"\
852 
+        "punpckhbw %%mm7, %%mm3		\n\t"\
853 
+        "movq %%mm0, (%1)		\n\t"\
854 
+        "movq %%mm1, 17*8(%1)		\n\t"\
855 
+        "movq %%mm2, (%1, %4)		\n\t"\
856 
+        "movq %%mm3, (%1, %5)		\n\t"\
857 
+        "addl $8, %1			\n\t"\
858 
+        "addl %3, %0			\n\t"\
859 
+        "decl %2			\n\t"\
860 
+        " jnz 1b			\n\t"\
861 
+        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
862 
+        : "r" (srcStride), "r"(2*8*17), "r"(3*8*17)\
863 
+    );\
864 
+    \
865 
+    temp_ptr= temp;\
866 
+    count=4;\
867 
+    \
868 
+/*FIXME reorder for speed */\
869 
+    asm volatile(\
870 
+        /*"pxor %%mm7, %%mm7		\n\t"*/\
871 
+        "1:				\n\t"\
872 
+        "movq (%0), %%mm0		\n\t"\
873 
+        "movq 8(%0), %%mm1		\n\t"\
874 
+        "movq 16(%0), %%mm2		\n\t"\
875 
+        "movq 24(%0), %%mm3		\n\t"\
876 
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
877 
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
878 
+        "addl %4, %1			\n\t"\
879 
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
880 
+        \
881 
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
882 
+        "addl %4, %1			\n\t"\
883 
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
884 
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
885 
+        "addl %4, %1			\n\t"\
886 
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
887 
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
888 
+        "addl %4, %1			\n\t"\
889 
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
890 
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
891 
+        "addl %4, %1			\n\t"\
892 
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
893 
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
894 
+        "addl %4, %1			\n\t"\
895 
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
896 
+        \
897 
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
898 
+        "addl %4, %1			\n\t"  \
899 
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
900 
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
901 
+        \
902 
+        "addl $136, %0			\n\t"\
903 
+        "addl %8, %1			\n\t"\
904 
+        "decl %2			\n\t"\
905 
+        " jnz 1b			\n\t"\
906 
+         \
907 
+        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
908 
+        : "r"(dstStride), "r"(2*dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(ROUNDER), "g"(4-14*dstStride)\
909 
+    );\
910 
+}\
911 
+void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
912 
+    uint64_t temp;\
913 
+\
914 
+    asm volatile(\
915 
+        "pxor %%mm7, %%mm7		\n\t"\
916 
+        "1:				\n\t"\
917 
+        "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\
918 
+        "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\
919 
+        "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\
920 
+        "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\
921 
+        "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\
922 
+        "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\
923 
+        "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\
924 
+        "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\
925 
+        "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\
926 
+        "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\
927 
+        "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\
928 
+        "psllq $24, %%mm4		\n\t" /* 000ABCDE */\
929 
+        "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\
930 
+        "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\
931 
+        "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\
932 
+        "paddw %%mm3, %%mm5		\n\t" /* b */\
933 
+        "paddw %%mm2, %%mm6		\n\t" /* c */\
934 
+        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
935 
+        "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\
936 
+        "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\
937 
+        "pmullw %6, %%mm6		\n\t" /* 3c - 6b */\
938 
+        "paddw %%mm4, %%mm0		\n\t" /* a */\
939 
+        "paddw %%mm1, %%mm5		\n\t" /* d */\
940 
+        "pmullw %5, %%mm0		\n\t" /* 20a */\
941 
+        "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\
942 
+        "paddw %8, %%mm6		\n\t"\
943 
+        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
944 
+        "psraw $5, %%mm0		\n\t"\
945 
+        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
946 
+        \
947 
+        "movd 5(%0), %%mm5		\n\t" /* FGHI */\
948 
+        "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\
949 
+        "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\
950 
+        "paddw %%mm5, %%mm1		\n\t" /* a */\
951 
+        "paddw %%mm6, %%mm2		\n\t" /* b */\
952 
+        "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\
953 
+        "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\
954 
+        "paddw %%mm6, %%mm3		\n\t" /* c */\
955 
+        "paddw %%mm5, %%mm4		\n\t" /* d */\
956 
+        "paddw %%mm2, %%mm2		\n\t" /* 2b */\
957 
+        "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\
958 
+        "pmullw %5, %%mm1		\n\t" /* 20a */\
959 
+        "pmullw %6, %%mm3		\n\t" /* 3c - 6b */\
960 
+        "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\
961 
+        "paddw %8, %%mm1		\n\t"\
962 
+        "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\
963 
+        "psraw $5, %%mm3		\n\t"\
964 
+        "packuswb %%mm3, %%mm0		\n\t"\
965 
+        OP(%%mm0, (%1), %%mm4, q)\
966 
+        \
967 
+        "addl %3, %0			\n\t"\
968 
+        "addl %4, %1			\n\t"\
969 
+        "decl %2			\n\t"\
970 
+        " jnz 1b				\n\t"\
971 
+        : "+r"(src), "+r"(dst), "+g"(h)\
972 
+        : "r"(srcStride), "r"(dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(temp), "m"(ROUNDER)\
973 
+    );\
974 
+}\
975 
+\
976 
+static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
977 
+    int i;\
978 
+    int16_t temp[8];\
979 
+    /* quick HACK, XXX FIXME MUST be optimized */\
980 
+    for(i=0; i<h; i++)\
981 
+    {\
982 
+        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
983 
+        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
984 
+        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
985 
+        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
986 
+        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
987 
+        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
988 
+        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
989 
+        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
990 
+        asm volatile(\
991 
+            "movq (%0), %%mm0		\n\t"\
992 
+            "movq 8(%0), %%mm1		\n\t"\
993 
+            "paddw %2, %%mm0		\n\t"\
994 
+            "paddw %2, %%mm1		\n\t"\
995 
+            "psraw $5, %%mm0		\n\t"\
996 
+            "psraw $5, %%mm1		\n\t"\
997 
+            "packuswb %%mm1, %%mm0	\n\t"\
998 
+            OP(%%mm0, (%1), %%mm1, q)\
999 
+            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1000 
+        );\
1001 
+        dst+=dstStride;\
1002 
+        src+=srcStride;\
1003 
+    }\
1004 
+}\
1005 
+\
1006 
+void OPNAME ## mpeg4_qpel8_v_lowpass_mmx(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1007 
+    uint64_t temp[9*4];\
1008 
+    uint64_t *temp_ptr= temp;\
1009 
+    int count= 9;\
1010 
+\
1011 
+    /*FIXME unroll */\
1012 
+    asm volatile(\
1013 
+        "pxor %%mm7, %%mm7		\n\t"\
1014 
+        "1:				\n\t"\
1015 
+        "movq (%0), %%mm0		\n\t"\
1016 
+        "movq (%0), %%mm1		\n\t"\
1017 
+        "punpcklbw %%mm7, %%mm0		\n\t"\
1018 
+        "punpckhbw %%mm7, %%mm1		\n\t"\
1019 
+        "movq %%mm0, (%1)		\n\t"\
1020 
+        "movq %%mm1, 9*8(%1)		\n\t"\
1021 
+        "addl $8, %1			\n\t"\
1022 
+        "addl %3, %0			\n\t"\
1023 
+        "decl %2			\n\t"\
1024 
+        " jnz 1b			\n\t"\
1025 
+        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1026 
+        : "r" (srcStride)\
1027 
+    );\
1028 
+    \
1029 
+    temp_ptr= temp;\
1030 
+    count=2;\
1031 
+    \
1032 
+/*FIXME reorder for speed */\
1033 
+    asm volatile(\
1034 
+        /*"pxor %%mm7, %%mm7		\n\t"*/\
1035 
+        "1:				\n\t"\
1036 
+        "movq (%0), %%mm0		\n\t"\
1037 
+        "movq 8(%0), %%mm1		\n\t"\
1038 
+        "movq 16(%0), %%mm2		\n\t"\
1039 
+        "movq 24(%0), %%mm3		\n\t"\
1040 
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1041 
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1042 
+        "addl %4, %1			\n\t"\
1043 
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1044 
+        \
1045 
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1046 
+        "addl %4, %1			\n\t"\
1047 
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1048 
+        \
1049 
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1050 
+        "addl %4, %1			\n\t"\
1051 
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1052 
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1053 
+                \
1054 
+        "addl $72, %0			\n\t"\
1055 
+        "addl %8, %1			\n\t"\
1056 
+        "decl %2			\n\t"\
1057 
+        " jnz 1b			\n\t"\
1058 
+         \
1059 
+        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1060 
+        : "r"(dstStride), "r"(2*dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(ROUNDER), "g"(4-6*dstStride)\
1061 
+    );\
1062 
+}
1063 
+
1064 
+#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1065 
+\
1066 
+static void OPNAME ## qpel8_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
1067 
+    put_pixels8_mmx(dst, src, stride, 8);\
1068 
+}\
1069 
+\
1070 
+static void OPNAME ## qpel8_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1071 
+    uint64_t temp[32];\
1072 
+    uint8_t * const half= (uint8_t*)temp;\
1073 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1074 
+    OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
1075 
+}\
1076 
+\
1077 
+static void OPNAME ## qpel8_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1078 
+    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1079 
+}\
1080 
+\
1081 
+static void OPNAME ## qpel8_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1082 
+    uint64_t temp[32];\
1083 
+    uint8_t * const half= (uint8_t*)temp;\
1084 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1085 
+    OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\
1086 
+}\
1087 
+\
1088 
+static void OPNAME ## qpel8_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1089 
+    uint64_t temp[32];\
1090 
+    uint8_t * const half= (uint8_t*)temp;\
1091 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(half, src, 8, stride);\
1092 
+    OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
1093 
+}\
1094 
+\
1095 
+static void OPNAME ## qpel8_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1096 
+    OPNAME ## mpeg4_qpel8_v_lowpass_mmx(dst, src, stride, stride);\
1097 
+}\
1098 
+\
1099 
+static void OPNAME ## qpel8_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1100 
+    uint64_t temp[32];\
1101 
+    uint8_t * const half= (uint8_t*)temp;\
1102 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(half, src, 8, stride);\
1103 
+    OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\
1104 
+}\
1105 
+static void OPNAME ## qpel8_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1106 
+    uint64_t half[8*2 + 8*2 + 18*2];\
1107 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*64 + 8;\
1108 
+    uint8_t * const halfV= ((uint8_t*)half);\
1109 
+    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1110 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1111 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src, 8, stride);\
1112 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1113 
+    OPNAME ## pixels8_l4_mmx(dst, src, (uint8_t*)half, stride, 8);\
1114 
+}\
1115 
+static void OPNAME ## qpel8_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1116 
+    uint64_t half[8*2 + 8*2 + 18*2];\
1117 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*64 + 8;\
1118 
+    uint8_t * const halfV= ((uint8_t*)half);\
1119 
+    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1120 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1121 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src+1, 8, stride);\
1122 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1123 
+    OPNAME ## pixels8_l4_mmx(dst, src+1, (uint8_t*)half, stride, 8);\
1124 
+}\
1125 
+static void OPNAME ## qpel8_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1126 
+    uint64_t half[8*2 + 8*2 + 9*2];\
1127 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1128 
+    uint8_t * const halfV= ((uint8_t*)half);\
1129 
+    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1130 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1131 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src, 8, stride);\
1132 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1133 
+    OPNAME ## pixels8_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 8);\
1134 
+}\
1135 
+static void OPNAME ## qpel8_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1136 
+    uint64_t half[8*2 + 8*2 + 9*2];\
1137 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1138 
+    uint8_t * const halfV= ((uint8_t*)half);\
1139 
+    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1140 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src  , 8, stride, 9);\
1141 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src+1, 8, stride);\
1142 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1143 
+    OPNAME ## pixels8_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 8);\
1144 
+}\
1145 
+static void OPNAME ## qpel8_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1146 
+    uint64_t half[8*2 + 9*2];\
1147 
+    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1148 
+    uint8_t * const halfHV= ((uint8_t*)half);\
1149 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1150 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1151 
+    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
1152 
+}\
1153 
+static void OPNAME ## qpel8_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1154 
+    uint64_t half[8*2 + 9*2];\
1155 
+    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1156 
+    uint8_t * const halfHV= ((uint8_t*)half);\
1157 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1158 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1159 
+    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
1160 
+}\
1161 
+static void OPNAME ## qpel8_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1162 
+    uint64_t half[8*2 + 8*2 + 9*2];\
1163 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1164 
+    uint8_t * const halfV= ((uint8_t*)half);\
1165 
+    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1166 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1167 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src, 8, stride);\
1168 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1169 
+    OPNAME ## pixels8_l2_mmx(dst, halfV, halfHV, stride, 8, 8);\
1170 
+}\
1171 
+static void OPNAME ## qpel8_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1172 
+    uint64_t half[8*2 + 8*2 + 9*2];\
1173 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1174 
+    uint8_t * const halfV= ((uint8_t*)half);\
1175 
+    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1176 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1177 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src+1, 8, stride);\
1178 
+    put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1179 
+    OPNAME ## pixels8_l2_mmx(dst, halfV, halfHV, stride, 8, 8);\
1180 
+}\
1181 
+static void OPNAME ## qpel8_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1182 
+    uint64_t half[9*2];\
1183 
+    uint8_t * const halfH= ((uint8_t*)half);\
1184 
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1185 
+    OPNAME ## mpeg4_qpel8_v_lowpass_mmx(dst, halfH, stride, 8);\
1186 
+}\
1187 
+static void OPNAME ## qpel16_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
1188 
+    put_pixels16_mmx(dst, src, stride, 16);\
1189 
+}\
1190 
+\
1191 
+static void OPNAME ## qpel16_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1192 
+    uint64_t temp[32];\
1193 
+    uint8_t * const half= (uint8_t*)temp;\
1194 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1195 
+    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
1196 
+}\
1197 
+\
1198 
+static void OPNAME ## qpel16_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1199 
+    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1200 
+}\
1201 
+\
1202 
+static void OPNAME ## qpel16_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1203 
+    uint64_t temp[32];\
1204 
+    uint8_t * const half= (uint8_t*)temp;\
1205 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1206 
+    OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\
1207 
+}\
1208 
+\
1209 
+static void OPNAME ## qpel16_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1210 
+    uint64_t temp[32];\
1211 
+    uint8_t * const half= (uint8_t*)temp;\
1212 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(half, src, 16, stride);\
1213 
+    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
1214 
+}\
1215 
+\
1216 
+static void OPNAME ## qpel16_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1217 
+    OPNAME ## mpeg4_qpel16_v_lowpass_mmx(dst, src, stride, stride);\
1218 
+}\
1219 
+\
1220 
+static void OPNAME ## qpel16_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1221 
+    uint64_t temp[32];\
1222 
+    uint8_t * const half= (uint8_t*)temp;\
1223 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(half, src, 16, stride);\
1224 
+    OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\
1225 
+}\
1226 
+static void OPNAME ## qpel16_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1227 
+    uint64_t half[16*2 + 16*2 + 18*2];\
1228 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*256 + 16;\
1229 
+    uint8_t * const halfV= ((uint8_t*)half);\
1230 
+    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1231 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1232 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src, 16, stride);\
1233 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1234 
+    OPNAME ## pixels16_l4_mmx(dst, src, (uint8_t*)half, stride, 16);\
1235 
+}\
1236 
+static void OPNAME ## qpel16_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1237 
+    uint64_t half[16*2 + 16*2 + 18*2];\
1238 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*256 + 16;\
1239 
+    uint8_t * const halfV= ((uint8_t*)half);\
1240 
+    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1241 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1242 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src+1, 16, stride);\
1243 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1244 
+    OPNAME ## pixels16_l4_mmx(dst, src+1, (uint8_t*)half, stride, 16);\
1245 
+}\
1246 
+static void OPNAME ## qpel16_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1247 
+    uint64_t half[16*2 + 16*2 + 17*2];\
1248 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1249 
+    uint8_t * const halfV= ((uint8_t*)half);\
1250 
+    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1251 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1252 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src, 16, stride);\
1253 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1254 
+    OPNAME ## pixels16_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 16);\
1255 
+}\
1256 
+static void OPNAME ## qpel16_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1257 
+    uint64_t half[16*2 + 16*2 + 17*2];\
1258 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1259 
+    uint8_t * const halfV= ((uint8_t*)half);\
1260 
+    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1261 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src  , 16, stride, 17);\
1262 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src+1, 16, stride);\
1263 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1264 
+    OPNAME ## pixels16_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 16);\
1265 
+}\
1266 
+static void OPNAME ## qpel16_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1267 
+    uint64_t half[16*2 + 17*2];\
1268 
+    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1269 
+    uint8_t * const halfHV= ((uint8_t*)half);\
1270 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1271 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1272 
+    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
1273 
+}\
1274 
+static void OPNAME ## qpel16_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1275 
+    uint64_t half[16*2 + 17*2];\
1276 
+    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1277 
+    uint8_t * const halfHV= ((uint8_t*)half);\
1278 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1279 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1280 
+    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
1281 
+}\
1282 
+static void OPNAME ## qpel16_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1283 
+    uint64_t half[16*2 + 16*2 + 17*2];\
1284 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1285 
+    uint8_t * const halfV= ((uint8_t*)half);\
1286 
+    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1287 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1288 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src, 16, stride);\
1289 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1290 
+    OPNAME ## pixels16_l2_mmx(dst, halfV, halfHV, stride, 16, 16);\
1291 
+}\
1292 
+static void OPNAME ## qpel16_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1293 
+    uint64_t half[16*2 + 16*2 + 17*2];\
1294 
+    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1295 
+    uint8_t * const halfV= ((uint8_t*)half);\
1296 
+    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1297 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1298 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src+1, 16, stride);\
1299 
+    put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1300 
+    OPNAME ## pixels16_l2_mmx(dst, halfV, halfHV, stride, 16, 16);\
1301 
+}\
1302 
+static void OPNAME ## qpel16_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1303 
+    uint64_t half[17*2];\
1304 
+    uint8_t * const halfH= ((uint8_t*)half);\
1305 
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1306 
+    OPNAME ## mpeg4_qpel16_v_lowpass_mmx(dst, halfH, stride, 16);\
1307 
+}
1308 
+
1309 
+
1310 
+#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"
1311 
+#define AVG_OP(a,b,temp, size) \
1312 
+"mov" #size " " #b ", " #temp "	\n\t"\
1313 
+"pavgusb " #temp ", " #a "	\n\t"\
1314 
+"mov" #size " " #a ", " #b "	\n\t"
1315 
+
1316 
+QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP)
1317 
+QPEL_BASE(avg_       , ff_pw_16, _       , AVG_OP)
1318 
+QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP)
1319 
+QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
1320 
+QPEL_OP(avg_       , ff_pw_16, _       , AVG_OP, 3dnow)
1321 
+QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1322 
+
1323 
+#undef AVG_OP
1324 
+#define AVG_OP(a,b,temp, size) \
1325 
+"mov" #size " " #b ", " #temp "	\n\t"\
1326 
+"pavgb " #temp ", " #a "	\n\t"\
1327 
+"mov" #size " " #a ", " #b "	\n\t"
1328 
+QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
1329 
+QPEL_OP(avg_       , ff_pw_16, _       , AVG_OP, mmx2)
1330 
+QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1331 
+
649 1332 
 #if 0
650 1333 
 static void just_return() { return; }
651 1334 
 #endif
652 1335
1336 
+#define SET_QPEL_FUNC(postfix1, postfix2) \
1337 
+    c->put_ ## postfix1 = put_ ## postfix2;\
1338 
+    c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
1339 
+    c->avg_ ## postfix1 = avg_ ## postfix2;
1340 
+
653 1341 
 void dsputil_init_mmx(DSPContext* c, unsigned mask)
654 1342 
 {
655 1343 
     mm_flags = mm_support();
... ... 
@@ -724,7 +1417,7 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
724 724 
         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
725 725 
         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
726 726 
         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
727 
-
727 
+
728 728 
         c->add_bytes= add_bytes_mmx;
729 729 
         c->diff_bytes= diff_bytes_mmx;
730 730
... ... 
@@ -767,6 +1460,38 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
767 767 
             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
768 768 
             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
769 769 
             c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
770 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
771 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
772 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
773 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
774 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
775 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
776 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
777 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
778 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
779 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
780 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
781 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)
782 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)
783 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)
784 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)
785 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)
786 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)
787 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)
788 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)
789 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)
790 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)
791 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)
792 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)
793 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)
794 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)
795 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)
796 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)
797 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)
798 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)
799 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)
800 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)
801 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)
770 802 
         } else if (mm_flags & MM_3DNOW) {
771 803 
             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
772 804 
             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
... ... 
@@ -787,6 +1512,39 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
787 787 
             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
788 788 
             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
789 789 
             c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
790 
+
791 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
792 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
793 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)
794 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)
795 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)
796 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)
797 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)
798 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)
799 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)
800 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)
801 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)
802 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)
803 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)
804 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)
805 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)
806 
+            SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)
807 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)
808 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)
809 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)
810 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)
811 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)
812 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)
813 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)
814 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)
815 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)
816 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)
817 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)
818 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)
819 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)
820 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)
821 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
822 
+            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
790 823 
         }
791 824 
     }
792 825
libavcodec/i386/dsputil_mmx_avg.h
History View file @ 826f429
... ... 
@@ -53,6 +53,38 @@ static void DEF(put_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size
53 53 
 	:"%eax", "memory");
54 54 
 }
55 55
56 
+static void DEF(put_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
57 
+{
58 
+    __asm __volatile(
59 
+	"1:				\n\t"
60 
+	"movq	(%1), %%mm0		\n\t"
61 
+	"addl	%4, %1			\n\t"
62 
+	"movq	(%1), %%mm1		\n\t"
63 
+	"addl	%4, %1			\n\t"
64 
+	PAVGB" (%2), %%mm0		\n\t"
65 
+	PAVGB" 8(%2), %%mm1		\n\t"
66 
+	"movq	%%mm0, (%3)		\n\t"
67 
+	"addl	%5, %3			\n\t"
68 
+	"movq	%%mm1, (%3)		\n\t"
69 
+	"addl	%5, %3			\n\t"
70 
+	"movq	(%1), %%mm0		\n\t"
71 
+	"addl	%4, %1			\n\t"
72 
+	"movq	(%1), %%mm1		\n\t"
73 
+	"addl	%4, %1			\n\t"
74 
+	PAVGB" 16(%2), %%mm0		\n\t"
75 
+	PAVGB" 24(%2), %%mm1		\n\t"
76 
+	"movq	%%mm0, (%3)		\n\t"
77 
+	"addl	%5, %3			\n\t"
78 
+	"movq	%%mm1, (%3)		\n\t"
79 
+	"addl	%5, %3			\n\t"
80 
+        "addl	$32, %2			\n\t"
81 
+	"subl	$4, %0			\n\t"
82 
+	"jnz	1b			\n\t"
83 
+	:"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
84 
+	:"r"(src1Stride), "r"(dstStride)
85 
+	:"memory");
86 
+}
87 
+
56 88 
 static void DEF(put_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
57 89 
 {
58 90 
     __asm __volatile(
... ... 
@@ -92,6 +124,34 @@ static void DEF(put_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_siz
92 92 
 	:"r" (line_size)
93 93 
 	:"%eax", "memory");
94 94 
 }
95 
+
96 
+static void DEF(put_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
97 
+{
98 
+    __asm __volatile(
99 
+	"1:				\n\t"
100 
+	"movq	(%1), %%mm0		\n\t"
101 
+	"movq	8(%1), %%mm1		\n\t"
102 
+	"addl	%4, %1			\n\t"
103 
+	PAVGB" (%2), %%mm0		\n\t"
104 
+	PAVGB" 8(%2), %%mm1		\n\t"
105 
+	"movq	%%mm0, (%3)		\n\t"
106 
+	"movq	%%mm1, 8(%3)		\n\t"
107 
+	"addl	%5, %3			\n\t"
108 
+	"movq	(%1), %%mm0		\n\t"
109 
+	"movq	8(%1), %%mm1		\n\t"
110 
+	"addl	%4, %1			\n\t"
111 
+	PAVGB" 16(%2), %%mm0		\n\t"
112 
+	PAVGB" 24(%2), %%mm1		\n\t"
113 
+	"movq	%%mm0, (%3)		\n\t"
114 
+	"movq	%%mm1, 8(%3)		\n\t"
115 
+	"addl	%5, %3			\n\t"
116 
+        "addl	$32, %2			\n\t"
117 
+	"subl	$2, %0			\n\t"
118 
+	"jnz	1b			\n\t"
119 
+	:"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
120 
+	:"r"(src1Stride), "r"(dstStride)
121 
+	:"memory");
122 
+}
95 123
96 124 
 /* GL: this function does incorrect rounding if overflow */
97 125 
 static void DEF(put_no_rnd_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
libavcodec/i386/dsputil_mmx_rnd.h
History View file @ 826f429
... ... 
@@ -54,6 +54,42 @@ static void DEF(put, pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_siz
54 54 
 	:"eax", "memory");
55 55 
 }
56 56
57 
+static void DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
58 
+{
59 
+    MOVQ_BFE(mm6);
60 
+    __asm __volatile(
61 
+	".balign 8			\n\t"
62 
+	"1:				\n\t"
63 
+	"movq	(%1), %%mm0		\n\t"
64 
+	"movq	(%2), %%mm1		\n\t"
65 
+	"addl	%4, %1			\n\t"
66 
+	"movq	(%1), %%mm2		\n\t"
67 
+	"movq	8(%2), %%mm3		\n\t"
68 
+	"addl	%4, %1			\n\t"
69 
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
70 
+	"movq	%%mm4, (%3)		\n\t"
71 
+	"addl	%5, %3			\n\t"
72 
+	"movq	%%mm5, (%3)		\n\t"
73 
+	"addl	%5, %3			\n\t"
74 
+	"movq	(%1), %%mm0		\n\t"
75 
+	"movq	16(%2), %%mm1		\n\t"
76 
+	"addl	%4, %1			\n\t"
77 
+	"movq	(%1), %%mm2		\n\t"
78 
+	"movq	24(%2), %%mm3		\n\t"
79 
+	"addl	%4, %1			\n\t"
80 
+	"addl	$32, %2			\n\t"
81 
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
82 
+	"movq	%%mm4, (%3)		\n\t"
83 
+	"addl	%5, %3			\n\t"
84 
+	"movq	%%mm5, (%3)		\n\t"
85 
+	"addl	%5, %3			\n\t"
86 
+	"subl	$4, %0			\n\t"
87 
+	"jnz	1b			\n\t"
88 
+	:"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
89 
+	:"r"(src1Stride), "r"(dstStride)
90 
+	:"memory");
91 
+}
92 
+
57 93 
 static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
58 94 
 {
59 95 
     MOVQ_BFE(mm6);
... ... 
@@ -90,7 +126,7 @@ static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si
90 90 
 	"movq	9(%1, %3), %%mm3	\n\t"
91 91 
 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
92 92 
 	"movq	%%mm4, 8(%2)		\n\t"
93 
-	"movq	%%mm5, 8(%2, %3)		\n\t"
93 
+	"movq	%%mm5, 8(%2, %3)	\n\t"
94 94 
 	"addl	%%eax, %1		\n\t"
95 95 
 	"addl	%%eax, %2		\n\t"
96 96 
 	"subl	$4, %0			\n\t"
... ... 
@@ -100,6 +136,38 @@ static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si
100 100 
 	:"eax", "memory");
101 101 
 }
102 102
103 
+static void DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
104 
+{
105 
+    MOVQ_BFE(mm6);
106 
+    __asm __volatile(
107 
+	".balign 8			\n\t"
108 
+	"1:				\n\t"
109 
+	"movq	(%1), %%mm0		\n\t"
110 
+	"movq	(%2), %%mm1		\n\t"
111 
+	"movq	8(%1), %%mm2		\n\t"
112 
+	"movq	8(%2), %%mm3		\n\t"
113 
+	"addl	%4, %1			\n\t"
114 
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
115 
+	"movq	%%mm4, (%3)		\n\t"
116 
+	"movq	%%mm5, 8(%3)		\n\t"
117 
+	"addl	%5, %3			\n\t"
118 
+	"movq	(%1), %%mm0		\n\t"
119 
+	"movq	16(%2), %%mm1		\n\t"
120 
+	"movq	8(%1), %%mm2		\n\t"
121 
+	"movq	24(%2), %%mm3		\n\t"
122 
+	"addl	%4, %1			\n\t"
123 
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
124 
+	"movq	%%mm4, (%3)		\n\t"
125 
+	"movq	%%mm5, 8(%3)		\n\t"
126 
+	"addl	%5, %3			\n\t"
127 
+	"addl	$32, %2			\n\t"
128 
+	"subl	$2, %0			\n\t"
129 
+	"jnz	1b			\n\t"
130 
+	:"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
131 
+	:"r"(src1Stride), "r"(dstStride)
132 
+	:"memory");
133 
+}
134 
+
103 135 
 static void DEF(put, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
104 136 
 {
105 137 
     MOVQ_BFE(mm6);
... ... 
@@ -195,6 +263,124 @@ static void DEF(put, pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_si
195 195 
 	:"eax", "memory");
196 196 
 }
197 197
198 
+static void DEF(put, pixels8_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
199 
+{
200 
+    MOVQ_ZERO(mm7);
201 
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
202 
+    __asm __volatile(
203 
+	".balign 8      		\n\t"
204 
+	"1:				\n\t"
205 
+	"movq	(%1), %%mm0		\n\t"
206 
+	"movq	(%2), %%mm1		\n\t"
207 
+	"movq	64(%2), %%mm2		\n\t"
208 
+	"movq	136(%2), %%mm3		\n\t"
209 
+	"punpcklbw %%mm7, %%mm0		\n\t"
210 
+	"punpcklbw %%mm7, %%mm1		\n\t"
211 
+	"punpcklbw %%mm7, %%mm2		\n\t"
212 
+	"punpcklbw %%mm7, %%mm3		\n\t"
213 
+	"paddusw %%mm6, %%mm0		\n\t"
214 
+	"paddusw %%mm0, %%mm1		\n\t"
215 
+	"paddusw %%mm2, %%mm3		\n\t"
216 
+	"paddusw %%mm1, %%mm3		\n\t"
217 
+	"psrlw	$2, %%mm3		\n\t"
218 
+	"movq	(%1), %%mm0		\n\t"
219 
+	"movq	(%2), %%mm1		\n\t"
220 
+	"movq	64(%2), %%mm2		\n\t"
221 
+	"movq	136(%2), %%mm4		\n\t"
222 
+	"punpckhbw %%mm7, %%mm0		\n\t"
223 
+	"punpckhbw %%mm7, %%mm1		\n\t"
224 
+	"punpckhbw %%mm7, %%mm2		\n\t"
225 
+	"punpckhbw %%mm7, %%mm4		\n\t"
226 
+	"paddusw %%mm6, %%mm0		\n\t"
227 
+	"paddusw %%mm0, %%mm1		\n\t"
228 
+	"paddusw %%mm2, %%mm4		\n\t"
229 
+	"paddusw %%mm1, %%mm4		\n\t"
230 
+	"psrlw	$2, %%mm4		\n\t"
231 
+	"packuswb  %%mm4, %%mm3		\n\t"
232 
+	"movq	%%mm3, (%0)		\n\t"
233 
+        "addl	%4, %0			\n\t"
234 
+        "addl	%4, %1			\n\t"
235 
+        "addl	$8, %2			\n\t"
236 
+        "decl	%3			\n\t"
237 
+	"jnz	1b			\n\t"
238 
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
239 
+	:"r"(stride)
240 
+	:"memory");
241 
+}
242 
+
243 
+static void DEF(put, pixels16_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
244 
+{
245 
+    MOVQ_ZERO(mm7);
246 
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
247 
+    __asm __volatile(
248 
+	".balign 8      		\n\t"
249 
+	"1:				\n\t"
250 
+	"movq	(%1), %%mm0		\n\t"
251 
+	"movq	(%2), %%mm1		\n\t"
252 
+	"movq	256(%2), %%mm2		\n\t"
253 
+	"movq	528(%2), %%mm3		\n\t"
254 
+	"punpcklbw %%mm7, %%mm0		\n\t"
255 
+	"punpcklbw %%mm7, %%mm1		\n\t"
256 
+	"punpcklbw %%mm7, %%mm2		\n\t"
257 
+	"punpcklbw %%mm7, %%mm3		\n\t"
258 
+	"paddusw %%mm6, %%mm0		\n\t"
259 
+	"paddusw %%mm0, %%mm1		\n\t"
260 
+	"paddusw %%mm2, %%mm3		\n\t"
261 
+	"paddusw %%mm1, %%mm3		\n\t"
262 
+	"psrlw	$2, %%mm3		\n\t"
263 
+	"movq	(%1), %%mm0		\n\t"
264 
+	"movq	(%2), %%mm1		\n\t"
265 
+	"movq	256(%2), %%mm2		\n\t"
266 
+	"movq	528(%2), %%mm4		\n\t"
267 
+	"punpckhbw %%mm7, %%mm0		\n\t"
268 
+	"punpckhbw %%mm7, %%mm1		\n\t"
269 
+	"punpckhbw %%mm7, %%mm2		\n\t"
270 
+	"punpckhbw %%mm7, %%mm4		\n\t"
271 
+	"paddusw %%mm6, %%mm0		\n\t"
272 
+	"paddusw %%mm0, %%mm1		\n\t"
273 
+	"paddusw %%mm2, %%mm4		\n\t"
274 
+	"paddusw %%mm1, %%mm4		\n\t"
275 
+	"psrlw	$2, %%mm4		\n\t"
276 
+	"packuswb  %%mm4, %%mm3		\n\t"
277 
+	"movq	%%mm3, (%0)		\n\t"
278 
+	"movq	8(%1), %%mm0		\n\t"
279 
+	"movq	8(%2), %%mm1		\n\t"
280 
+	"movq	264(%2), %%mm2		\n\t"
281 
+	"movq	536(%2), %%mm3		\n\t"
282 
+	"punpcklbw %%mm7, %%mm0		\n\t"
283 
+	"punpcklbw %%mm7, %%mm1		\n\t"
284 
+	"punpcklbw %%mm7, %%mm2		\n\t"
285 
+	"punpcklbw %%mm7, %%mm3		\n\t"
286 
+	"paddusw %%mm6, %%mm0		\n\t"
287 
+	"paddusw %%mm0, %%mm1		\n\t"
288 
+	"paddusw %%mm2, %%mm3		\n\t"
289 
+	"paddusw %%mm1, %%mm3		\n\t"
290 
+	"psrlw	$2, %%mm3		\n\t"
291 
+	"movq	8(%1), %%mm0		\n\t"
292 
+	"movq	8(%2), %%mm1		\n\t"
293 
+	"movq	264(%2), %%mm2		\n\t"
294 
+	"movq	536(%2), %%mm4		\n\t"
295 
+	"punpckhbw %%mm7, %%mm0		\n\t"
296 
+	"punpckhbw %%mm7, %%mm1		\n\t"
297 
+	"punpckhbw %%mm7, %%mm2		\n\t"
298 
+	"punpckhbw %%mm7, %%mm4		\n\t"
299 
+	"paddusw %%mm6, %%mm0		\n\t"
300 
+	"paddusw %%mm0, %%mm1		\n\t"
301 
+	"paddusw %%mm2, %%mm4		\n\t"
302 
+	"paddusw %%mm1, %%mm4		\n\t"
303 
+	"psrlw	$2, %%mm4		\n\t"
304 
+	"packuswb  %%mm4, %%mm3		\n\t"
305 
+	"movq	%%mm3, 8(%0)		\n\t"
306 
+        "addl	%4, %0			\n\t"
307 
+        "addl	%4, %1			\n\t"
308 
+        "addl	$16, %2			\n\t"
309 
+        "decl	%3			\n\t"
310 
+	"jnz	1b			\n\t"
311 
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
312 
+	:"r"(stride)
313 
+	:"memory");
314 
+}
315 
+
198 316 
 // avg_pixels
199 317 
 // in case more speed is needed - unroling would certainly help
200 318 
 static void DEF(avg, pixels8)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
... ... 
@@ -259,6 +445,27 @@ static void DEF(avg, pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_siz
259 259 
     } while (--h);
260 260 
 }
261 261
262 
+static void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
263 
+{
264 
+    MOVQ_BFE(mm6);
265 
+    JUMPALIGN();
266 
+    do {
267 
+	__asm __volatile(
268 
+	    "movq  %1, %%mm0		\n\t"
269 
+	    "movq  %2, %%mm1		\n\t"
270 
+	    "movq  %0, %%mm3		\n\t"
271 
+	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
272 
+	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
273 
+	    "movq  %%mm0, %0		\n\t"
274 
+	    :"+m"(*dst)
275 
+	    :"m"(*src1), "m"(*src2)
276 
+	    :"memory");
277 
+	dst += dstStride;
278 
+        src1 += src1Stride;
279 
+        src2 += 8;
280 
+    } while (--h);
281 
+}
282 
+
262 283 
 static void DEF(avg, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
263 284 
 {
264 285 
     MOVQ_BFE(mm6);
... ... 
@@ -285,6 +492,33 @@ static void DEF(avg, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si
285 285 
     } while (--h);
286 286 
 }
287 287
288 
+static void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
289 
+{
290 
+    MOVQ_BFE(mm6);
291 
+    JUMPALIGN();
292 
+    do {
293 
+	__asm __volatile(
294 
+	    "movq  %1, %%mm0		\n\t"
295 
+	    "movq  %2, %%mm1		\n\t"
296 
+	    "movq  %0, %%mm3		\n\t"
297 
+	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
298 
+	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
299 
+	    "movq  %%mm0, %0		\n\t"
300 
+	    "movq  8%1, %%mm0		\n\t"
301 
+	    "movq  8%2, %%mm1		\n\t"
302 
+	    "movq  8%0, %%mm3		\n\t"
303 
+	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
304 
+	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
305 
+	    "movq  %%mm0, 8%0		\n\t"
306 
+	    :"+m"(*dst)
307 
+	    :"m"(*src1), "m"(*src2)
308 
+	    :"memory");
309 
+	dst += dstStride;
310 
+        src1 += src1Stride;
311 
+        src2 += 16;
312 
+    } while (--h);
313 
+}
314 
+
288 315 
 static void DEF(avg, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
289 316 
 {
290 317 
     MOVQ_BFE(mm6);
... ... 
@@ -399,6 +633,133 @@ static void DEF(avg, pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_si
399 399 
 	:"eax", "memory");
400 400 
 }
401 401
402 
+static void DEF(avg, pixels8_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
403 
+{
404 
+    MOVQ_ZERO(mm7);
405 
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
406 
+    MOVQ_BFE(mm5);
407 
+    __asm __volatile(
408 
+	".balign 8      		\n\t"
409 
+	"1:				\n\t"
410 
+	"movq	(%1), %%mm0		\n\t"
411 
+	"movq	(%2), %%mm1		\n\t"
412 
+	"movq	64(%2), %%mm2		\n\t"
413 
+	"movq	136(%2), %%mm3		\n\t"
414 
+	"punpcklbw %%mm7, %%mm0		\n\t"
415 
+	"punpcklbw %%mm7, %%mm1		\n\t"
416 
+	"punpcklbw %%mm7, %%mm2		\n\t"
417 
+	"punpcklbw %%mm7, %%mm3		\n\t"
418 
+	"paddusw %%mm6, %%mm0		\n\t"
419 
+	"paddusw %%mm0, %%mm1		\n\t"
420 
+	"paddusw %%mm2, %%mm3		\n\t"
421 
+	"paddusw %%mm1, %%mm3		\n\t"
422 
+	"psrlw	$2, %%mm3		\n\t"
423 
+	"movq	(%1), %%mm0		\n\t"
424 
+	"movq	(%2), %%mm1		\n\t"
425 
+	"movq	64(%2), %%mm2		\n\t"
426 
+	"movq	136(%4), %%mm4		\n\t"
427 
+	"punpckhbw %%mm7, %%mm0		\n\t"
428 
+	"punpckhbw %%mm7, %%mm1		\n\t"
429 
+	"punpckhbw %%mm7, %%mm2		\n\t"
430 
+	"punpckhbw %%mm7, %%mm4		\n\t"
431 
+	"paddusw %%mm6, %%mm0		\n\t"
432 
+	"paddusw %%mm0, %%mm1		\n\t"
433 
+	"paddusw %%mm2, %%mm4		\n\t"
434 
+	"paddusw %%mm1, %%mm4		\n\t"
435 
+	"psrlw	$2, %%mm4		\n\t"
436 
+	"packuswb  %%mm4, %%mm3		\n\t"
437 
+	"movq	(%0), %%mm4		\n\t"
438 
+        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
439 
+	"movq	%%mm3, (%0)		\n\t"
440 
+        "addl	%4, %0			\n\t"
441 
+        "addl	%4, %1			\n\t"
442 
+        "addl	$8, %2			\n\t"
443 
+        "decl	%3			\n\t"
444 
+	"jnz	1b			\n\t"
445 
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
446 
+	:"r"(stride)
447 
+	:"memory");
448 
+}
449 
+
450 
+static void DEF(avg, pixels16_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
451 
+{
452 
+    MOVQ_ZERO(mm7);
453 
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
454 
+    MOVQ_BFE(mm5);
455 
+    __asm __volatile(
456 
+	".balign 8      		\n\t"
457 
+	"1:				\n\t"
458 
+	"movq	(%1), %%mm0		\n\t"
459 
+	"movq	(%2), %%mm1		\n\t"
460 
+	"movq	256(%2), %%mm2		\n\t"
461 
+	"movq	528(%2), %%mm3		\n\t"
462 
+	"punpcklbw %%mm7, %%mm0		\n\t"
463 
+	"punpcklbw %%mm7, %%mm1		\n\t"
464 
+	"punpcklbw %%mm7, %%mm2		\n\t"
465 
+	"punpcklbw %%mm7, %%mm3		\n\t"
466 
+	"paddusw %%mm6, %%mm0		\n\t"
467 
+	"paddusw %%mm0, %%mm1		\n\t"
468 
+	"paddusw %%mm2, %%mm3		\n\t"
469 
+	"paddusw %%mm1, %%mm3		\n\t"
470 
+	"psrlw	$2, %%mm3		\n\t"
471 
+	"movq	(%1), %%mm0		\n\t"
472 
+	"movq	(%2), %%mm1		\n\t"
473 
+	"movq	256(%2), %%mm2		\n\t"
474 
+	"movq	528(%4), %%mm4		\n\t"
475 
+	"punpckhbw %%mm7, %%mm0		\n\t"
476 
+	"punpckhbw %%mm7, %%mm1		\n\t"
477 
+	"punpckhbw %%mm7, %%mm2		\n\t"
478 
+	"punpckhbw %%mm7, %%mm4		\n\t"
479 
+	"paddusw %%mm6, %%mm0		\n\t"
480 
+	"paddusw %%mm0, %%mm1		\n\t"
481 
+	"paddusw %%mm2, %%mm4		\n\t"
482 
+	"paddusw %%mm1, %%mm4		\n\t"
483 
+	"psrlw	$2, %%mm4		\n\t"
484 
+	"packuswb  %%mm4, %%mm3		\n\t"
485 
+	"movq	(%0), %%mm4		\n\t"
486 
+        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
487 
+	"movq	%%mm3, (%0)		\n\t"
488 
+	"movq	8(%1), %%mm0		\n\t"
489 
+	"movq	8(%2), %%mm1		\n\t"
490 
+	"movq	264(%2), %%mm2		\n\t"
491 
+	"movq	536(%2), %%mm3		\n\t"
492 
+	"punpcklbw %%mm7, %%mm0		\n\t"
493 
+	"punpcklbw %%mm7, %%mm1		\n\t"
494 
+	"punpcklbw %%mm7, %%mm2		\n\t"
495 
+	"punpcklbw %%mm7, %%mm3		\n\t"
496 
+	"paddusw %%mm6, %%mm0		\n\t"
497 
+	"paddusw %%mm0, %%mm1		\n\t"
498 
+	"paddusw %%mm2, %%mm3		\n\t"
499 
+	"paddusw %%mm1, %%mm3		\n\t"
500 
+	"psrlw	$2, %%mm3		\n\t"
501 
+	"movq	8(%1), %%mm0		\n\t"
502 
+	"movq	8(%2), %%mm1		\n\t"
503 
+	"movq	264(%2), %%mm2		\n\t"
504 
+	"movq	536(%4), %%mm4		\n\t"
505 
+	"punpckhbw %%mm7, %%mm0		\n\t"
506 
+	"punpckhbw %%mm7, %%mm1		\n\t"
507 
+	"punpckhbw %%mm7, %%mm2		\n\t"
508 
+	"punpckhbw %%mm7, %%mm4		\n\t"
509 
+	"paddusw %%mm6, %%mm0		\n\t"
510 
+	"paddusw %%mm0, %%mm1		\n\t"
511 
+	"paddusw %%mm2, %%mm4		\n\t"
512 
+	"paddusw %%mm1, %%mm4		\n\t"
513 
+	"psrlw	$2, %%mm4		\n\t"
514 
+	"packuswb  %%mm4, %%mm3		\n\t"
515 
+	"movq	8(%0), %%mm4		\n\t"
516 
+        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
517 
+	"movq	%%mm3, 8(%0)		\n\t"
518 
+        "addl	%4, %0			\n\t"
519 
+        "addl	%4, %1			\n\t"
520 
+        "addl	$16, %2			\n\t"
521 
+        "decl	%3			\n\t"
522 
+	"jnz	1b			\n\t"
523 
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
524 
+	:"r"(stride)
525 
+	:"memory");
526 
+}
527 
+
528 
+
402 529 
 //FIXME optimize
403 530 
 static void DEF(put, pixels16_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){
404 531 
     DEF(put, pixels8_y2)(block  , pixels  , line_size, h);
libavcodec/motion_est.c
History View file @ 826f429
... ... 
@@ -311,6 +311,7 @@ static inline int get_penalty_factor(MpegEncContext *s, int type){
311 311 
 }
312 312
313 313 
 void ff_init_me(MpegEncContext *s){
314 
+    set_cmp(s, s->dsp.me_pre_cmp, s->avctx->me_pre_cmp);
314 315 
     set_cmp(s, s->dsp.me_cmp, s->avctx->me_cmp);
315 316 
     set_cmp(s, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
316 317 
     set_cmp(s, s->dsp.mb_cmp, s->avctx->mb_cmp);
... ... 
@@ -336,6 +337,12 @@ void ff_init_me(MpegEncContext *s){
336 336 
         s->me.motion_search[0]= simple_epzs_motion_search;
337 337 
         s->me.motion_search[1]= simple_epzs_motion_search4;
338 338 
     }
339 
+
340 
+    if(s->avctx->me_pre_cmp&FF_CMP_CHROMA){
341 
+        s->me.pre_motion_search= simple_chroma_epzs_motion_search;
342 
+    }else{
343 
+        s->me.pre_motion_search= simple_epzs_motion_search;
344 
+    }
339 345 
 }
340 346
341 347 
 static int pix_dev(UINT8 * pix, int line_size, int mean)
... ... 
@@ -1037,7 +1044,7 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1037 1037
1038 1038 
     assert(s->quarter_sample==0 || s->quarter_sample==1);
1039 1039
1040 
-    s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);
1040 
+    s->me.pre_penalty_factor    = get_penalty_factor(s, s->avctx->me_pre_cmp);
1041 1041
1042 1042 
     get_limits(s, &range, &xmin, &ymin, &xmax, &ymax, s->f_code);
1043 1043 
     rel_xmin= xmin - mb_x*16;
... ... 
@@ -1072,8 +1079,8 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1072 1072 
         pred_x = P_MEDIAN[0];
1073 1073 
         pred_y = P_MEDIAN[1];
1074 1074 
     }
1075 
-    dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
1076 
-                                  &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);
1075 
+    dmin = s->me.pre_motion_search(s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
1076 
+                                   &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);
1077 1077
1078 1078 
     s->p_mv_table[xy][0] = mx<<shift;
1079 1079 
     s->p_mv_table[xy][1] = my<<shift;
libavcodec/motion_est_template.c
History View file @ 826f429
... ... 
@@ -268,6 +268,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
268 268 
     const int my = *my_ptr;
269 269 
     const int penalty_factor= s->me.sub_penalty_factor;
270 270 
     const int map_generation= s->me.map_generation;
271 
+    const int subpel_quality= s->avctx->me_subpel_quality;
271 272 
     uint32_t *map= s->me.map;
272 273 
     me_cmp_func cmp, chroma_cmp;
273 274 
     me_cmp_func cmp_sub, chroma_cmp_sub;
... ... 
@@ -309,7 +310,7 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
309 309
310 310 
         memset(best, 64, sizeof(int)*8);
311 311 
 #if 1
312 
-        if(s->avctx->dia_size>=2){
312 
+        if(s->me.dia_size>=2){
313 313 
             const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
314 314 
             const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
315 315 
             const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
... ... 
@@ -388,24 +389,34 @@ static int RENAME(qpel_motion_search)(MpegEncContext * s,
388 388 
                 }
389 389 
             }
390 390 
         }
391 
-        for(i=0; i<8; i++){
391 
+        for(i=0; i<subpel_quality; i++){
392 392 
             nx= best_pos[i][0];
393 393 
             ny= best_pos[i][1];
394 394 
             CHECK_QUARTER_MV(nx&3, ny&3, nx>>2, ny>>2)
395 395 
         }
396 
+
396 397 
 #if 0
397 
-            nx= FFMAX(4*mx - bx, bx - 4*mx);
398 
-            ny= FFMAX(4*my - by, by - 4*my);
398 
+            const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
399 
+            const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
400 
+            const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
401 
+            const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
402 
+//            if(l < r && l < t && l < b && l < tl && l < bl && l < tr && l < br && bl < tl){
403 
+            if(tl<br){
404 
+
405 
+//            nx= FFMAX(4*mx - bx, bx - 4*mx);
406 
+//            ny= FFMAX(4*my - by, by - 4*my);
399 407
400 
-            static int stats[4][4];
401 
-            stats[nx][ny]++;
402 
-            if(256*256*256*64 % (stats[0][0]+1) ==0){
403 
-                for(i=0; i<16; i++){
404 
-                    if((i&3)==0) printf("\n");
408 
+            static int stats[7][7], count;
409 
+            count++;
410 
+            stats[4*mx - bx + 3][4*my - by + 3]++;
411 
+            if(256*256*256*64 % count ==0){
412 
+                for(i=0; i<49; i++){
413 
+                    if((i%7)==0) printf("\n");
405 414 
                     printf("%6d ", stats[0][i]);
406 415 
                 }
407 416 
                 printf("\n");
408 417 
             }
418 
+            }
409 419 
 #endif
410 420 
 #else
411 421
... ... 
@@ -659,7 +670,7 @@ static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int
659 659 
 {
660 660 
     me_cmp_func cmp, chroma_cmp;
661 661 
     Minima minima[MAX_SAB_SIZE];
662 
-    const int minima_count= ABS(s->avctx->dia_size);
662 
+    const int minima_count= ABS(s->me.dia_size);
663 663 
     int i, j;
664 664 
     LOAD_COMMON(s->mb_x*16, s->mb_y*16);
665 665
... ... 
@@ -744,7 +755,7 @@ static inline int RENAME(var_diamond_search)(MpegEncContext * s, int *best, int
744 744 
     cmp= s->dsp.me_cmp[size];
745 745 
     chroma_cmp= s->dsp.me_cmp[size+1];
746 746
747 
-    for(dia_size=1; dia_size<=s->avctx->dia_size; dia_size++){
747 
+    for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){
748 748 
         int dir, start, end;
749 749 
         const int x= best[0];
750 750 
         const int y= best[1];
... ... 
@@ -893,15 +904,15 @@ static int RENAME(epzs_motion_search)(MpegEncContext * s, int block,
893 893 
     }
894 894
895 895 
 //check(best[0],best[1],0, b0)
896 
-    if(s->avctx->dia_size==-1)
896 
+    if(s->me.dia_size==-1)
897 897 
         dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture,
898 898 
                                    pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
899 899 
 				   shift, map, map_generation, size, mv_penalty);
900 
-    else if(s->avctx->dia_size<-1)
900 
+    else if(s->me.dia_size<-1)
901 901 
         dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture,
902 902 
                                    pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
903 903 
 				   shift, map, map_generation, size, mv_penalty);
904 
-    else if(s->avctx->dia_size<2)
904 
+    else if(s->me.dia_size<2)
905 905 
         dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture,
906 906 
                                    pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
907 907 
 				   shift, map, map_generation, size, mv_penalty);
... ... 
@@ -969,15 +980,15 @@ static int RENAME(epzs_motion_search4)(MpegEncContext * s, int block,
969 969 
                         (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
970 970 
     }
971 971
972 
-    if(s->avctx->dia_size==-1)
972 
+    if(s->me.dia_size==-1)
973 973 
         dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture,
974 974 
                                    pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
975 975 
 				   shift, map, map_generation, size, mv_penalty);
976 
-    else if(s->avctx->dia_size<-1)
976 
+    else if(s->me.dia_size<-1)
977 977 
         dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture,
978 978 
                                    pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
979 979 
 				   shift, map, map_generation, size, mv_penalty);
980 
-    else if(s->avctx->dia_size<2)
980 
+    else if(s->me.dia_size<2)
981 981 
         dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture,
982 982 
                                    pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
983 983 
 				   shift, map, map_generation, size, mv_penalty);
libavcodec/mpegvideo.c
History View file @ 826f429
... ... 
@@ -2786,12 +2786,12 @@ static void encode_picture(MpegEncContext *s, int picture_number)
2786 2786 
         else if(s->pict_type!=B_TYPE)
2787 2787 
             s->no_rounding ^= 1;
2788 2788 
     }
2789 
-
2790 2789 
     /* Estimate motion for every MB */
2791 2790 
     if(s->pict_type != I_TYPE){
2792 2791 
         if(s->pict_type != B_TYPE){
2793 2792 
             if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){
2794 2793 
                 s->me.pre_pass=1;
2794 
+                s->me.dia_size= s->avctx->pre_dia_size;
2795 2795
2796 2796 
                 for(mb_y=s->mb_height-1; mb_y >=0 ; mb_y--) {
2797 2797 
                     for(mb_x=s->mb_width-1; mb_x >=0 ; mb_x--) {
... ... 
@@ -2804,6 +2804,7 @@ static void encode_picture(MpegEncContext *s, int picture_number)
2804 2804 
             }
2805 2805 
         }
2806 2806
2807 
+        s->me.dia_size= s->avctx->dia_size;
2807 2808 
         for(mb_y=0; mb_y < s->mb_height; mb_y++) {
2808 2809 
             s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
2809 2810 
             s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
... ... 
@@ -2816,7 +2817,7 @@ static void encode_picture(MpegEncContext *s, int picture_number)
2816 2816 
                 s->block_index[1]+=2;
2817 2817 
                 s->block_index[2]+=2;
2818 2818 
                 s->block_index[3]+=2;
2819 
-
2819 
+
2820 2820 
                 /* compute motion vector & mb_type and store in context */
2821 2821 
                 if(s->pict_type==B_TYPE)
2822 2822 
                     ff_estimate_b_frame_motion(s, mb_x, mb_y);
libavcodec/mpegvideo.h
History View file @ 826f429
... ... 
@@ -139,9 +139,11 @@ typedef struct MotionEstContext{
139 139 
     uint32_t *map;                     /* map to avoid duplicate evaluations */
140 140 
     uint32_t *score_map;               /* map to store the scores */
141 141 
     int map_generation;
142 
+    int pre_penalty_factor;
142 143 
     int penalty_factor;
143 144 
     int sub_penalty_factor;
144 145 
     int pre_pass;                      /* = 1 for the pre pass */
146 
+    int dia_size;
145 147 
     UINT16 (*mv_penalty)[MAX_MV*2+1];  /* amount of bits needed to encode a MV */
146 148 
     int (*sub_motion_search)(struct MpegEncContext * s,
147 149 
 				  int *mx_ptr, int *my_ptr, int dmin,
... ... 
@@ -153,6 +155,11 @@ typedef struct MotionEstContext{
153 153 
                              int P[10][2], int pred_x, int pred_y,
154 154 
                              int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2],
155 155 
                              int ref_mv_scale, uint16_t * const mv_penalty);
156 
+    int (*pre_motion_search)(struct MpegEncContext * s, int block,
157 
+                             int *mx_ptr, int *my_ptr,
158 
+                             int P[10][2], int pred_x, int pred_y,
159 
+                             int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2],
160 
+                             int ref_mv_scale, uint16_t * const mv_penalty);
156 161 
 }MotionEstContext;
157 162
158 163 
 typedef struct MpegEncContext {
libavcodec/utils.c
History View file @ 826f429
... ... 
@@ -234,6 +234,7 @@ void avcodec_get_context_defaults(AVCodecContext *s){
234 234 
     s->me_method= ME_EPZS;
235 235 
     s->get_buffer= avcodec_default_get_buffer;
236 236 
     s->release_buffer= avcodec_default_release_buffer;
237 
+    s->me_subpel_quality=8;
237 238 
 }
238 239
239 240 
 /**