Move PS2 MMI code below the mips subdirectory, where it belongs.

Also give a more suitable name to the MMI-optimized IDCT;
it is not PS2-specific, as the name currently suggests.

5 files changed, 792 insertions, 3 deletions

diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile
index 3f4da68ec9..006617687e 100644
--- a/libavcodec/mips/Makefile
+++ b/libavcodec/mips/Makefile
@@ -1,3 +1,3 @@
-OBJS-$(HAVE_MMI)                       += ps2/dsputil_mmi.o             \
-                                          ps2/idct_mmi.o                \
-                                          ps2/mpegvideo_mmi.o           \
+OBJS-$(HAVE_MMI) += mips/dsputil_mmi.o                                  \
+                    mips/idct_mmi.o                                     \
+                    mips/mpegvideo_mmi.o                                \
diff --git a/libavcodec/mips/dsputil_mmi.c b/libavcodec/mips/dsputil_mmi.c
new file mode 100644
index 0000000000..20a435a065
--- /dev/null
+++ b/libavcodec/mips/dsputil_mmi.c
@@ -0,0 +1,162 @@
+/*
+ * MMI optimized DSP utils
+ * Copyright (c) 2000, 2001 Fabrice Bellard
+ *
+ * MMI optimization by Leon van Stuivenberg
+ * clear_blocks_mmi() by BroadQ
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavcodec/dsputil.h"
+#include "mmi.h"
+
+static void clear_blocks_mmi(DCTELEM * blocks)
+{
+        __asm__ volatile(
+        ".set noreorder    \n"
+        "addiu $9, %0, 768 \n"
+        "nop               \n"
+        "1:                \n"
+        "sq $0, 0(%0)      \n"
+        "move $8, %0       \n"
+        "addi %0, %0, 64   \n"
+        "sq $0, 16($8)     \n"
+        "slt $10, %0, $9   \n"
+        "sq $0, 32($8)     \n"
+        "bnez $10, 1b      \n"
+        "sq $0, 48($8)     \n"
+        ".set reorder      \n"
+        : "+r" (blocks) ::  "$8", "$9", "memory" );
+}
+
+
+static void get_pixels_mmi(DCTELEM *block, const uint8_t *pixels, int line_size)
+{
+        __asm__ volatile(
+        ".set   push            \n\t"
+        ".set   mips3           \n\t"
+        "ld     $8, 0(%0)       \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $9, 0(%0)       \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $10, 0(%0)      \n\t"
+        "pextlb $8, $0, $8      \n\t"
+        "sq     $8, 0(%1)       \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $8, 0(%0)       \n\t"
+        "pextlb $9, $0, $9      \n\t"
+        "sq     $9, 16(%1)      \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $9, 0(%0)       \n\t"
+        "pextlb $10, $0, $10    \n\t"
+        "sq     $10, 32(%1)     \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $10, 0(%0)      \n\t"
+        "pextlb $8, $0, $8      \n\t"
+        "sq     $8, 48(%1)      \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $8, 0(%0)       \n\t"
+        "pextlb $9, $0, $9      \n\t"
+        "sq     $9, 64(%1)      \n\t"
+        "add    %0, %0, %2      \n\t"
+        "ld     $9, 0(%0)       \n\t"
+        "pextlb $10, $0, $10    \n\t"
+        "sq     $10, 80(%1)     \n\t"
+        "pextlb $8, $0, $8      \n\t"
+        "sq     $8, 96(%1)      \n\t"
+        "pextlb $9, $0, $9      \n\t"
+        "sq     $9, 112(%1)     \n\t"
+        ".set   pop             \n\t"
+        : "+r" (pixels) : "r" (block), "r" (line_size) : "$8", "$9", "$10", "memory" );
+}
+
+
+static void put_pixels8_mmi(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+        __asm__ volatile(
+        ".set   push            \n\t"
+        ".set   mips3           \n\t"
+        "1:                     \n\t"
+        "ldr    $8, 0(%1)       \n\t"
+        "addiu  %2, %2, -1      \n\t"
+        "ldl    $8, 7(%1)       \n\t"
+        "add    %1, %1, %3      \n\t"
+        "sd     $8, 0(%0)       \n\t"
+        "add    %0, %0, %3      \n\t"
+        "bgtz   %2, 1b          \n\t"
+        ".set   pop             \n\t"
+        : "+r" (block), "+r" (pixels), "+r" (h) : "r" (line_size)
+        : "$8", "memory" );
+}
+
+
+static void put_pixels16_mmi(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+        __asm__ volatile (
+        ".set   push            \n\t"
+        ".set   mips3           \n\t"
+        "1:                     \n\t"
+        "ldr    $8, 0(%1)       \n\t"
+        "add    $11, %1, %3     \n\t"
+        "ldl    $8, 7(%1)       \n\t"
+        "add    $10, %0, %3     \n\t"
+        "ldr    $9, 8(%1)       \n\t"
+        "ldl    $9, 15(%1)      \n\t"
+        "ldr    $12, 0($11)     \n\t"
+        "add    %1, $11, %3     \n\t"
+        "ldl    $12, 7($11)     \n\t"
+        "pcpyld $8, $9, $8      \n\t"
+        "sq     $8, 0(%0)       \n\t"
+        "ldr    $13, 8($11)     \n\t"
+        "addiu  %2, %2, -2      \n\t"
+        "ldl    $13, 15($11)    \n\t"
+        "add    %0, $10, %3     \n\t"
+        "pcpyld $12, $13, $12   \n\t"
+        "sq     $12, 0($10)     \n\t"
+        "bgtz   %2, 1b          \n\t"
+        ".set   pop             \n\t"
+        : "+r" (block), "+r" (pixels), "+r" (h) : "r" (line_size)
+        : "$8", "$9", "$10", "$11", "$12", "$13", "memory" );
+}
+
+
+void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx)
+{
+    const int idct_algo= avctx->idct_algo;
+    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
+
+    if (!high_bit_depth) {
+    c->clear_blocks = clear_blocks_mmi;
+
+    c->put_pixels_tab[1][0] = put_pixels8_mmi;
+    c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmi;
+
+    c->put_pixels_tab[0][0] = put_pixels16_mmi;
+    c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmi;
+
+    c->get_pixels = get_pixels_mmi;
+    }
+
+    if (avctx->bits_per_raw_sample <= 8 &&
+        (idct_algo == FF_IDCT_AUTO || idct_algo == FF_IDCT_MMI)) {
+        c->idct_put= ff_mmi_idct_put;
+        c->idct_add= ff_mmi_idct_add;
+        c->idct    = ff_mmi_idct;
+        c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
+    }
+}
diff --git a/libavcodec/mips/idct_mmi.c b/libavcodec/mips/idct_mmi.c
new file mode 100644
index 0000000000..9b9033ae44
--- /dev/null
+++ b/libavcodec/mips/idct_mmi.c
@@ -0,0 +1,361 @@
+/*
+ * Originally provided by Intel at Application Note AP-922.
+ *
+ * Column code adapted from Peter Gubanov.
+ * Copyright (c) 2000-2001 Peter Gubanov <peter@elecard.net.ru>
+ * http://www.elecard.com/peter/idct.shtml
+ * rounding trick copyright (c) 2000 Michel Lespinasse <walken@zoy.org>
+ *
+ * MMI port and (c) 2002 by Leon van Stuivenberg
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/common.h"
+#include "libavcodec/dsputil.h"
+#include "mmi.h"
+
+#define BITS_INV_ACC    5       // 4 or 5 for IEEE
+#define SHIFT_INV_ROW   (16 - BITS_INV_ACC)
+#define SHIFT_INV_COL   (1 + BITS_INV_ACC)
+
+#define TG1             6518
+#define TG2             13573
+#define TG3             21895
+#define CS4             23170
+
+#define ROUNDER_0       0
+#define ROUNDER_1       16
+
+#define TAB_i_04        (32+0)
+#define TAB_i_17        (32+64)
+#define TAB_i_26        (32+128)
+#define TAB_i_35        (32+192)
+
+#define TG_1_16         (32+256+0)
+#define TG_2_16         (32+256+16)
+#define TG_3_16         (32+256+32)
+#define COS_4_16        (32+256+48)
+
+#define CLIPMAX         (32+256+64+0)
+
+static short consttable[] align16 = {
+/* rounder 0*/  // assume SHIFT_INV_ROW == 11
+ 0x3ff, 1, 0x3ff, 1, 0x3ff, 1, 0x3ff, 1,
+/* rounder 1*/
+ 0x3ff, 0, 0x3ff, 0, 0x3ff, 0, 0x3ff, 0,
+/* row 0/4*/
+ 16384,  21407, -16384, -21407,  22725,  19266, -22725, -12873,
+  8867,  16384,   8867,  16384,   4520,  12873,  -4520,  19266,
+ 16384,  -8867,  16384,  -8867,  12873, -22725,  19266, -22725,
+ 21407, -16384, -21407,  16384,  19266,   4520, -12873,   4520,
+/* row 1/7*/
+ 22725,  29692, -22725, -29692,  31521,  26722, -31521, -17855,
+ 12299,  22725,  12299,  22725,   6270,  17855,  -6270,  26722,
+ 22725, -12299,  22725, -12299,  17855, -31521,  26722, -31521,
+ 29692, -22725, -29692,  22725,  26722,   6270, -17855,   6270,
+/* row 2/6*/
+ 21407,  27969, -21407, -27969,  29692,  25172, -29692, -16819,
+ 11585,  21407,  11585,  21407,   5906,  16819,  -5906,  25172,
+ 21407, -11585,  21407, -11585,  16819, -29692,  25172, -29692,
+ 27969, -21407, -27969,  21407,  25172,   5906, -16819,   5906,
+/*row 3/5*/
+ 19266,  25172, -19266, -25172,  26722,  22654, -26722, -15137,
+ 10426,  19266,  10426,  19266,   5315,  15137,  -5315,  22654,
+ 19266, -10426,  19266, -10426,  15137, -26722,  22654, -26722,
+ 25172, -19266, -25172,  19266,  22654,   5315, -15137,   5315,
+/*column constants*/
+ TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG1,
+ TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG2,
+ TG3, TG3, TG3, TG3, TG3, TG3, TG3, TG3,
+ CS4, CS4, CS4, CS4, CS4, CS4, CS4, CS4,
+/* clamp */
+ 255, 255, 255, 255, 255, 255, 255, 255
+};
+
+
+#define DCT_8_INV_ROW1(blk, rowoff, taboff, rnd, outreg) { \
+        lq(blk, rowoff, $16);   /* r16 = x7  x5  x3  x1  x6  x4  x2  x0 */ \
+        /*slot*/ \
+        lq($24, 0+taboff, $17); /* r17 = w */ \
+        /*delay slot $16*/ \
+        lq($24, 16+taboff, $18);/* r18 = w */ \
+        prevh($16, $2);         /* r2  = x1  x3  x5  x7  x0  x2  x4  x6 */ \
+        lq($24, 32+taboff, $19);/* r19 = w */ \
+        phmadh($17, $16, $17);  /* r17 = b1"b0'a1"a0' */ \
+        lq($24, 48+taboff, $20);/* r20 = w */ \
+        phmadh($18, $2, $18);   /* r18 = b1'b0"a1'a0" */ \
+        phmadh($19, $16, $19);  /* r19 = b3"b2'a3"a2' */ \
+        phmadh($20, $2, $20);   /* r20 = b3'b2"a3'a2" */ \
+        paddw($17, $18, $17);   /* r17 = (b1)(b0)(a1)(a0) */ \
+        paddw($19, $20, $19);   /* r19 = (b3)(b2)(a3)(a2) */ \
+        pcpyld($19, $17, $18);  /* r18 = (a3)(a2)(a1)(a0) */ \
+        pcpyud($17, $19, $20);  /* r20 = (b3)(b2)(b1)(b0) */ \
+        paddw($18, rnd, $18);   /* r18 = (a3)(a2)(a1)(a0) */\
+        paddw($18, $20, $17);   /* r17 = ()()()(a0+b0) */ \
+        psubw($18, $20, $20);   /* r20 = ()()()(a0-b0) */ \
+        psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \
+        psraw($20, SHIFT_INV_ROW, $20); /* r20 = (y4 y5 y6 y7) */ \
+        ppach($20, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0  Note order */ \
+\
+        prevh(outreg, $2);        \
+        pcpyud($2, $2, $2);        \
+        pcpyld($2, outreg, outreg);        \
+}
+
+
+#define DCT_8_INV_COL8() \
+\
+        lq($24, TG_3_16, $2);   /* r2  = tn3 */         \
+\
+        pmulth($11, $2, $17);   /* r17 = x3 * tn3 (6420) */ \
+        psraw($17, 15, $17);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $17, $17);   /* r17 = x3 * tn3 */    \
+        psubh($17, $13, $17);   /* r17 = tm35 */        \
+\
+        pmulth($13, $2, $18);   /* r18 = x5 * tn3 (6420) */ \
+        psraw($18, 15, $18);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $18, $18);   /* r18 = x5 * tn3 */    \
+        paddh($18, $11, $18);   /* r18 = tp35 */        \
+\
+        lq($24, TG_1_16, $2);   /* r2  = tn1 */         \
+\
+        pmulth($15, $2, $19);   /* r19 = x7 * tn1 (6420) */ \
+        psraw($19, 15, $19);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $19, $19);   /* r19 = x7 * tn1 */    \
+        paddh($19, $9, $19);    /* r19 = tp17 */        \
+\
+        pmulth($9, $2, $20);    /* r20 = x1 * tn1 (6420) */ \
+        psraw($20, 15, $20);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $20, $20);   /* r20 = x1 * tn1 */    \
+        psubh($20, $15, $20);   /* r20 = tm17 */        \
+\
+        psubh($19, $18, $3);    /* r3  = t1 */          \
+        paddh($20, $17, $16);   /* r16 = t2 */          \
+        psubh($20, $17, $23);   /* r23 = b3 */          \
+        paddh($19, $18, $20);   /* r20 = b0 */          \
+\
+        lq($24, COS_4_16, $2);  /* r2  = cs4 */         \
+\
+        paddh($3, $16, $21);    /* r21 = t1+t2 */       \
+        psubh($3, $16, $22);    /* r22 = t1-t2 */       \
+\
+        pmulth($21, $2, $21);   /* r21 = cs4 * (t1+t2) 6420 */ \
+        psraw($21, 15, $21);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $21, $21);   /* r21 = b1 */          \
+\
+        pmulth($22, $2, $22);   /* r22 = cs4 * (t1-t2) 6420 */ \
+        psraw($22, 15, $22);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $22, $22);   /* r22 = b2 */          \
+\
+        lq($24, TG_2_16, $2);   /* r2  = tn2 */         \
+\
+        pmulth($10, $2, $17);   /* r17 = x2 * tn2 (6420) */ \
+        psraw($17, 15, $17);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $17, $17);   /* r17 = x3 * tn3 */    \
+        psubh($17, $14, $17);   /* r17 = tm26 */        \
+\
+        pmulth($14, $2, $18);   /* r18 = x6 * tn2 (6420) */ \
+        psraw($18, 15, $18);    \
+        pmfhl_uw($3);           /* r3  = 7531 */        \
+        psraw($3, 15, $3);      \
+        pinteh($3, $18, $18);   /* r18 = x6 * tn2 */    \
+        paddh($18, $10, $18);   /* r18 = tp26 */        \
+\
+        paddh($8, $12, $2);     /* r2  = tp04 */        \
+        psubh($8, $12, $3);     /* r3  = tm04 */        \
+\
+        paddh($2, $18, $16);    /* r16 = a0 */          \
+        psubh($2, $18, $19);    /* r19 = a3 */          \
+        psubh($3, $17, $18);    /* r18 = a2 */          \
+        paddh($3, $17, $17);    /* r17 = a1 */
+
+
+#define DCT_8_INV_COL8_STORE(blk) \
+\
+        paddh($16, $20, $2);    /* y0  a0+b0 */ \
+        psubh($16, $20, $16);   /* y7  a0-b0 */ \
+        psrah($2, SHIFT_INV_COL, $2);           \
+        psrah($16, SHIFT_INV_COL, $16);         \
+        sq($2, 0, blk);                         \
+        sq($16, 112, blk);                      \
+\
+        paddh($17, $21, $3);    /* y1  a1+b1 */ \
+        psubh($17, $21, $17);   /* y6  a1-b1 */ \
+        psrah($3, SHIFT_INV_COL, $3);           \
+        psrah($17, SHIFT_INV_COL, $17);         \
+        sq($3, 16, blk);                        \
+        sq($17, 96, blk);                       \
+\
+        paddh($18, $22, $2);    /* y2  a2+b2 */ \
+        psubh($18, $22, $18);   /* y5  a2-b2 */ \
+        psrah($2, SHIFT_INV_COL, $2);           \
+        psrah($18, SHIFT_INV_COL, $18);         \
+        sq($2, 32, blk);                        \
+        sq($18, 80, blk);                       \
+\
+        paddh($19, $23, $3);    /* y3  a3+b3 */ \
+        psubh($19, $23, $19);   /* y4  a3-b3 */ \
+        psrah($3, SHIFT_INV_COL, $3);           \
+        psrah($19, SHIFT_INV_COL, $19);         \
+        sq($3, 48, blk);                        \
+        sq($19, 64, blk);
+
+
+
+#define DCT_8_INV_COL8_PMS() \
+        paddh($16, $20, $2);    /* y0  a0+b0 */ \
+        psubh($16, $20, $20);   /* y7  a0-b0 */ \
+        psrah($2, SHIFT_INV_COL, $16);          \
+        psrah($20, SHIFT_INV_COL, $20);         \
+\
+        paddh($17, $21, $3);    /* y1  a1+b1 */ \
+        psubh($17, $21, $21);   /* y6  a1-b1 */ \
+        psrah($3, SHIFT_INV_COL, $17);          \
+        psrah($21, SHIFT_INV_COL, $21);         \
+\
+        paddh($18, $22, $2);    /* y2  a2+b2 */ \
+        psubh($18, $22, $22);   /* y5  a2-b2 */ \
+        psrah($2, SHIFT_INV_COL, $18);          \
+        psrah($22, SHIFT_INV_COL, $22);         \
+\
+        paddh($19, $23, $3);    /* y3  a3+b3 */ \
+        psubh($19, $23, $23);   /* y4  a3-b3 */ \
+        psrah($3, SHIFT_INV_COL, $19);          \
+        psrah($23, SHIFT_INV_COL, $23);
+
+#define PUT(rs)                 \
+        pminh(rs, $11, $2);     \
+        pmaxh($2, $0, $2);      \
+        ppacb($0, $2, $2);      \
+        sd3(2, 0, 4);           \
+        __asm__ volatile ("add $4, $5, $4");
+
+#define DCT_8_INV_COL8_PUT() \
+        PUT($16);        \
+        PUT($17);        \
+        PUT($18);        \
+        PUT($19);        \
+        PUT($23);        \
+        PUT($22);        \
+        PUT($21);        \
+        PUT($20);
+
+#define ADD(rs)          \
+        ld3(4, 0, 2);        \
+        pextlb($0, $2, $2);  \
+        paddh($2, rs, $2);   \
+        pminh($2, $11, $2);  \
+        pmaxh($2, $0, $2);   \
+        ppacb($0, $2, $2);   \
+        sd3(2, 0, 4); \
+        __asm__ volatile ("add $4, $5, $4");
+
+/*fixme: schedule*/
+#define DCT_8_INV_COL8_ADD() \
+        ADD($16);        \
+        ADD($17);        \
+        ADD($18);        \
+        ADD($19);        \
+        ADD($23);        \
+        ADD($22);        \
+        ADD($21);        \
+        ADD($20);
+
+
+void ff_mmi_idct(int16_t * block)
+{
+        /* $4 = block */
+        __asm__ volatile("la $24, %0"::"m"(consttable[0]));
+        lq($24, ROUNDER_0, $8);
+        lq($24, ROUNDER_1, $7);
+        DCT_8_INV_ROW1($4, 0, TAB_i_04, $8, $8);
+        DCT_8_INV_ROW1($4, 16, TAB_i_17, $7, $9);
+        DCT_8_INV_ROW1($4, 32, TAB_i_26, $7, $10);
+        DCT_8_INV_ROW1($4, 48, TAB_i_35, $7, $11);
+        DCT_8_INV_ROW1($4, 64, TAB_i_04, $7, $12);
+        DCT_8_INV_ROW1($4, 80, TAB_i_35, $7, $13);
+        DCT_8_INV_ROW1($4, 96, TAB_i_26, $7, $14);
+        DCT_8_INV_ROW1($4, 112, TAB_i_17, $7, $15);
+        DCT_8_INV_COL8();
+        DCT_8_INV_COL8_STORE($4);
+
+        //let savedtemp regs be saved
+        __asm__ volatile(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
+}
+
+
+void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+        /* $4 = dest, $5 = line_size, $6 = block */
+        __asm__ volatile("la $24, %0"::"m"(consttable[0]));
+        lq($24, ROUNDER_0, $8);
+        lq($24, ROUNDER_1, $7);
+        DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
+        DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
+        DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
+        DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
+        DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
+        DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
+        DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
+        DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
+        DCT_8_INV_COL8();
+        lq($24, CLIPMAX, $11);
+        DCT_8_INV_COL8_PMS();
+        DCT_8_INV_COL8_PUT();
+
+        //let savedtemp regs be saved
+        __asm__ volatile(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
+}
+
+
+void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+        /* $4 = dest, $5 = line_size, $6 = block */
+        __asm__ volatile("la $24, %0"::"m"(consttable[0]));
+        lq($24, ROUNDER_0, $8);
+        lq($24, ROUNDER_1, $7);
+        DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
+        DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
+        DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
+        DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
+        DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
+        DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
+        DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
+        DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
+        DCT_8_INV_COL8();
+        lq($24, CLIPMAX, $11);
+        DCT_8_INV_COL8_PMS();
+        DCT_8_INV_COL8_ADD();
+
+        //let savedtemp regs be saved
+        __asm__ volatile(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
+}
diff --git a/libavcodec/mips/mmi.h b/libavcodec/mips/mmi.h
new file mode 100644
index 0000000000..48f2778809
--- /dev/null
+++ b/libavcodec/mips/mmi.h
@@ -0,0 +1,179 @@
+/*
+ * copyright (c) 2002 Leon van Stuivenberg
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_MIPS_MMI_H
+#define AVCODEC_MIPS_MMI_H
+
+#include <stdint.h>
+
+#include "libavcodec/dsputil.h"
+
+void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block);
+void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block);
+void ff_mmi_idct(DCTELEM *block);
+
+#define align16 __attribute__ ((aligned (16)))
+
+/*
+#define r0 $zero
+#define r1 $at          //assembler!
+#define r2 $v0          //return
+#define r3 $v1          //return
+#define r4 $a0          //arg
+#define r5 $a1          //arg
+#define r6 $a2          //arg
+#define r7 $a3          //arg
+#define r8 $t0          //temp
+#define r9 $t1          //temp
+#define r10 $t2         //temp
+#define r11 $t3         //temp
+#define r12 $t4         //temp
+#define r13 $t5         //temp
+#define r14 $t6         //temp
+#define r15 $t7         //temp
+#define r16 $s0         //saved temp
+#define r17 $s1         //saved temp
+#define r18 $s2         //saved temp
+#define r19 $s3         //saved temp
+#define r20 $s4         //saved temp
+#define r21 $s5         //saved temp
+#define r22 $s6         //saved temp
+#define r23 $s7         //saved temp
+#define r24 $t8         //temp
+#define r25 $t9         //temp
+#define r26 $k0         //kernel
+#define r27 $k1         //kernel
+#define r28 $gp         //global ptr
+#define r29 $sp         //stack ptr
+#define r30 $fp         //frame ptr
+#define r31 $ra         //return addr
+*/
+
+
+#define         lq(base, off, reg)        \
+        __asm__ volatile ("lq " #reg ", %0("#base ")" : : "i" (off) )
+
+#define         lq2(mem, reg)        \
+        __asm__ volatile ("lq " #reg ", %0" : : "r" (mem))
+
+#define         sq(reg, off, base)        \
+        __asm__ volatile ("sq " #reg ", %0("#base ")" : : "i" (off) )
+
+/*
+#define         ld(base, off, reg)        \
+        __asm__ volatile ("ld " #reg ", " #off "("#base ")")
+*/
+
+#define         ld3(base, off, reg)        \
+        __asm__ volatile (".word %0" : : "i" ( 0xdc000000 | (base<<21) | (reg<<16) | (off)))
+
+#define         ldr3(base, off, reg)        \
+        __asm__ volatile (".word %0" : : "i" ( 0x6c000000 | (base<<21) | (reg<<16) | (off)))
+
+#define         ldl3(base, off, reg)        \
+        __asm__ volatile (".word %0" : : "i" ( 0x68000000 | (base<<21) | (reg<<16) | (off)))
+
+/*
+#define         sd(reg, off, base)        \
+        __asm__ volatile ("sd " #reg ", " #off "("#base ")")
+*/
+//seems assembler has bug encoding mnemonic 'sd', so DIY
+#define         sd3(reg, off, base)        \
+        __asm__ volatile (".word %0" : : "i" ( 0xfc000000 | (base<<21) | (reg<<16) | (off)))
+
+#define         sw(reg, off, base)        \
+        __asm__ volatile ("sw " #reg ", " #off "("#base ")")
+
+#define         sq2(reg, mem)        \
+        __asm__ volatile ("sq " #reg ", %0" : : "m" (*(mem)))
+
+#define         pinth(rs, rt, rd) \
+        __asm__ volatile ("pinth  " #rd ", " #rs ", " #rt )
+
+#define         phmadh(rs, rt, rd) \
+        __asm__ volatile ("phmadh " #rd ", " #rs ", " #rt )
+
+#define         pcpyud(rs, rt, rd) \
+        __asm__ volatile ("pcpyud " #rd ", " #rs ", " #rt )
+
+#define         pcpyld(rs, rt, rd) \
+        __asm__ volatile ("pcpyld " #rd ", " #rs ", " #rt )
+
+#define         pcpyh(rt, rd) \
+        __asm__ volatile ("pcpyh  " #rd ", " #rt )
+
+#define         paddw(rs, rt, rd) \
+        __asm__ volatile ("paddw  " #rd ", " #rs ", " #rt )
+
+#define         pextlw(rs, rt, rd) \
+        __asm__ volatile ("pextlw " #rd ", " #rs ", " #rt )
+
+#define         pextuw(rs, rt, rd) \
+        __asm__ volatile ("pextuw " #rd ", " #rs ", " #rt )
+
+#define         pextlh(rs, rt, rd) \
+        __asm__ volatile ("pextlh " #rd ", " #rs ", " #rt )
+
+#define         pextuh(rs, rt, rd) \
+        __asm__ volatile ("pextuh " #rd ", " #rs ", " #rt )
+
+#define         psubw(rs, rt, rd) \
+        __asm__ volatile ("psubw  " #rd ", " #rs ", " #rt )
+
+#define         psraw(rt, sa, rd) \
+        __asm__ volatile ("psraw  " #rd ", " #rt ", %0" : : "i"(sa) )
+
+#define         ppach(rs, rt, rd) \
+        __asm__ volatile ("ppach  " #rd ", " #rs ", " #rt )
+
+#define         ppacb(rs, rt, rd) \
+        __asm__ volatile ("ppacb  " #rd ", " #rs ", " #rt )
+
+#define         prevh(rt, rd) \
+        __asm__ volatile ("prevh  " #rd ", " #rt )
+
+#define         pmulth(rs, rt, rd) \
+        __asm__ volatile ("pmulth " #rd ", " #rs ", " #rt )
+
+#define         pmaxh(rs, rt, rd) \
+        __asm__ volatile ("pmaxh " #rd ", " #rs ", " #rt )
+
+#define         pminh(rs, rt, rd) \
+        __asm__ volatile ("pminh " #rd ", " #rs ", " #rt )
+
+#define         pinteh(rs, rt, rd) \
+        __asm__ volatile ("pinteh  " #rd ", " #rs ", " #rt )
+
+#define         paddh(rs, rt, rd) \
+        __asm__ volatile ("paddh  " #rd ", " #rs ", " #rt )
+
+#define         psubh(rs, rt, rd) \
+        __asm__ volatile ("psubh  " #rd ", " #rs ", " #rt )
+
+#define         psrah(rt, sa, rd) \
+        __asm__ volatile ("psrah  " #rd ", " #rt ", %0" : : "i"(sa) )
+
+#define         pmfhl_uw(rd) \
+        __asm__ volatile ("pmfhl.uw  " #rd)
+
+#define         pextlb(rs, rt, rd) \
+        __asm__ volatile ("pextlb  " #rd ", " #rs ", " #rt )
+
+#endif /* AVCODEC_MIPS_MMI_H */
diff --git a/libavcodec/mips/mpegvideo_mmi.c b/libavcodec/mips/mpegvideo_mmi.c
new file mode 100644
index 0000000000..623e35af33
--- /dev/null
+++ b/libavcodec/mips/mpegvideo_mmi.c
@@ -0,0 +1,87 @@
+/*
+ * Copyright (c) 2000,2001 Fabrice Bellard
+ *
+ * MMI optimization by Leon van Stuivenberg
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/dsputil.h"
+#include "libavcodec/mpegvideo.h"
+
+static void dct_unquantize_h263_mmi(MpegEncContext *s,
+                                  DCTELEM *block, int n, int qscale)
+{
+    int level=0, qmul, qadd;
+    int nCoeffs;
+
+    assert(s->block_last_index[n]>=0);
+
+    qadd = (qscale - 1) | 1;
+    qmul = qscale << 1;
+
+    if (s->mb_intra) {
+        if (!s->h263_aic) {
+            if (n < 4)
+                level = block[0] * s->y_dc_scale;
+            else
+                level = block[0] * s->c_dc_scale;
+        }else {
+            qadd = 0;
+            level = block[0];
+        }
+        nCoeffs= 63; //does not always use zigzag table
+    } else {
+        nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
+    }
+
+    __asm__ volatile(
+        "add    $14, $0, %3     \n\t"
+        "pcpyld $8, %0, %0      \n\t"
+        "pcpyh  $8, $8          \n\t"   //r8 = qmul
+        "pcpyld $9, %1, %1      \n\t"
+        "pcpyh  $9, $9          \n\t"   //r9 = qadd
+        ".p2align 2             \n\t"
+        "1:                     \n\t"
+        "lq     $10, 0($14)     \n\t"   //r10 = level
+        "addi   $14, $14, 16    \n\t"   //block+=8
+        "addi   %2, %2, -8      \n\t"
+        "pcgth  $11, $0, $10    \n\t"   //r11 = level < 0 ? -1 : 0
+        "pcgth  $12, $10, $0    \n\t"   //r12 = level > 0 ? -1 : 0
+        "por    $12, $11, $12   \n\t"
+        "pmulth $10, $10, $8    \n\t"
+        "paddh  $13, $9, $11    \n\t"
+        "pxor   $13, $13, $11   \n\t"   //r13 = level < 0 ? -qadd : qadd
+        "pmfhl.uw $11           \n\t"
+        "pinteh $10, $11, $10   \n\t"   //r10 = level * qmul
+        "paddh  $10, $10, $13   \n\t"
+        "pand   $10, $10, $12   \n\t"
+        "sq     $10, -16($14)   \n\t"
+        "bgez   %2, 1b          \n\t"
+        :: "r"(qmul), "r" (qadd), "r" (nCoeffs), "r" (block) : "$8", "$9", "$10", "$11", "$12", "$13", "$14", "memory" );
+
+    if(s->mb_intra)
+        block[0]= level;
+}
+
+
+void MPV_common_init_mmi(MpegEncContext *s)
+{
+    s->dct_unquantize_h263_intra =
+    s->dct_unquantize_h263_inter = dct_unquantize_h263_mmi;
+}