[FFmpeg-devel] [PATCH v2] avcodec/mips: [loongson] mmi optimizations for VP9 put and avg functions

gxw guxiwei-hf at loongson.cn
Thu Feb 21 14:39:28 EET 2019


VP9 decoding speed improved about 60.5%(from 38fps to 61fps, tested on loongson 3A3000).
---
 libavcodec/mips/Makefile           |   1 +
 libavcodec/mips/vp9_mc_mmi.c       | 692 +++++++++++++++++++++++++++++++++++++
 libavcodec/mips/vp9dsp_init_mips.c |  42 +++
 libavcodec/mips/vp9dsp_mips.h      |  50 +++
 libavutil/mips/mmiutils.h          |  15 +
 5 files changed, 800 insertions(+)
 create mode 100644 libavcodec/mips/vp9_mc_mmi.c

diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile
index c827649..c5b54d5 100644
--- a/libavcodec/mips/Makefile
+++ b/libavcodec/mips/Makefile
@@ -88,3 +88,4 @@ MMI-OBJS-$(CONFIG_VC1_DECODER)            += mips/vc1dsp_mmi.o
 MMI-OBJS-$(CONFIG_WMV2DSP)                += mips/wmv2dsp_mmi.o
 MMI-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_mmi.o
 MMI-OBJS-$(CONFIG_VP3DSP)                 += mips/vp3dsp_idct_mmi.o
+MMI-OBJS-$(CONFIG_VP9_DECODER)            += mips/vp9_mc_mmi.o
diff --git a/libavcodec/mips/vp9_mc_mmi.c b/libavcodec/mips/vp9_mc_mmi.c
new file mode 100644
index 0000000..58a920b
--- /dev/null
+++ b/libavcodec/mips/vp9_mc_mmi.c
@@ -0,0 +1,692 @@
+/*
+ * Copyright (c) 2019 gxw <guxiwei-hf at loongson.cn>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavcodec/vp9dsp.h"
+#include "libavutil/mips/mmiutils.h"
+#include "vp9dsp_mips.h"
+
+#define GET_DATA_H_MMI                                       \
+    "pmaddhw    %[ftmp4],    %[ftmp4],   %[filter1]    \n\t" \
+    "pmaddhw    %[ftmp5],    %[ftmp5],   %[filter2]    \n\t" \
+    "paddw      %[ftmp4],    %[ftmp4],   %[ftmp5]      \n\t" \
+    "punpckhwd  %[ftmp5],    %[ftmp4],   %[ftmp0]      \n\t" \
+    "paddw      %[ftmp4],    %[ftmp4],   %[ftmp5]      \n\t" \
+    "pmaddhw    %[ftmp6],    %[ftmp6],   %[filter1]    \n\t" \
+    "pmaddhw    %[ftmp7],    %[ftmp7],   %[filter2]    \n\t" \
+    "paddw      %[ftmp6],    %[ftmp6],   %[ftmp7]      \n\t" \
+    "punpckhwd  %[ftmp7],    %[ftmp6],   %[ftmp0]      \n\t" \
+    "paddw      %[ftmp6],    %[ftmp6],   %[ftmp7]      \n\t" \
+    "punpcklwd  %[srcl],     %[ftmp4],   %[ftmp6]      \n\t" \
+    "pmaddhw    %[ftmp8],    %[ftmp8],   %[filter1]    \n\t" \
+    "pmaddhw    %[ftmp9],    %[ftmp9],   %[filter2]    \n\t" \
+    "paddw      %[ftmp8],    %[ftmp8],   %[ftmp9]      \n\t" \
+    "punpckhwd  %[ftmp9],    %[ftmp8],   %[ftmp0]      \n\t" \
+    "paddw      %[ftmp8],    %[ftmp8],   %[ftmp9]      \n\t" \
+    "pmaddhw    %[ftmp10],   %[ftmp10],  %[filter1]    \n\t" \
+    "pmaddhw    %[ftmp11],   %[ftmp11],  %[filter2]    \n\t" \
+    "paddw      %[ftmp10],   %[ftmp10],  %[ftmp11]     \n\t" \
+    "punpckhwd  %[ftmp11],   %[ftmp10],  %[ftmp0]      \n\t" \
+    "paddw      %[ftmp10],   %[ftmp10],  %[ftmp11]     \n\t" \
+    "punpcklwd  %[srch],     %[ftmp8],   %[ftmp10]     \n\t"
+
+#define GET_DATA_V_MMI                                       \
+    "punpcklhw  %[srcl],     %[ftmp4],   %[ftmp5]      \n\t" \
+    "pmaddhw    %[srcl],     %[srcl],    %[filter10]   \n\t" \
+    "punpcklhw  %[ftmp12],   %[ftmp6],   %[ftmp7]      \n\t" \
+    "pmaddhw    %[ftmp12],   %[ftmp12],  %[filter32]   \n\t" \
+    "paddw      %[srcl],     %[srcl],    %[ftmp12]     \n\t" \
+    "punpcklhw  %[ftmp12],   %[ftmp8],   %[ftmp9]      \n\t" \
+    "pmaddhw    %[ftmp12],   %[ftmp12],  %[filter54]   \n\t" \
+    "paddw      %[srcl],     %[srcl],    %[ftmp12]     \n\t" \
+    "punpcklhw  %[ftmp12],   %[ftmp10],  %[ftmp11]     \n\t" \
+    "pmaddhw    %[ftmp12],   %[ftmp12],  %[filter76]   \n\t" \
+    "paddw      %[srcl],     %[srcl],    %[ftmp12]     \n\t" \
+    "punpckhhw  %[srch],     %[ftmp4],   %[ftmp5]      \n\t" \
+    "pmaddhw    %[srch],     %[srch],    %[filter10]   \n\t" \
+    "punpckhhw  %[ftmp12],   %[ftmp6],   %[ftmp7]      \n\t" \
+    "pmaddhw    %[ftmp12],   %[ftmp12],  %[filter32]   \n\t" \
+    "paddw      %[srch],     %[srch],    %[ftmp12]     \n\t" \
+    "punpckhhw  %[ftmp12],   %[ftmp8],   %[ftmp9]      \n\t" \
+    "pmaddhw    %[ftmp12],   %[ftmp12],  %[filter54]   \n\t" \
+    "paddw      %[srch],     %[srch],    %[ftmp12]     \n\t" \
+    "punpckhhw  %[ftmp12],   %[ftmp10],  %[ftmp11]     \n\t" \
+    "pmaddhw    %[ftmp12],   %[ftmp12],  %[filter76]   \n\t" \
+    "paddw      %[srch],     %[srch],    %[ftmp12]     \n\t"
+
+static void convolve_horiz_mmi(const uint8_t *src, int32_t src_stride,
+                               uint8_t *dst, int32_t dst_stride,
+                               const uint16_t *filter_x, int32_t w,
+                               int32_t h)
+{
+    double ftmp[15];
+    uint32_t tmp[2];
+    src -= 3;
+    src_stride -= w;
+    dst_stride -= w;
+    __asm__ volatile (
+        "move       %[tmp1],    %[width]                   \n\t"
+        "xor        %[ftmp0],   %[ftmp0],    %[ftmp0]      \n\t"
+        "gsldlc1    %[filter1], 0x03(%[filter])            \n\t"
+        "gsldrc1    %[filter1], 0x00(%[filter])            \n\t"
+        "gsldlc1    %[filter2], 0x0b(%[filter])            \n\t"
+        "gsldrc1    %[filter2], 0x08(%[filter])            \n\t"
+        "li         %[tmp0],    0x07                       \n\t"
+        "dmtc1      %[tmp0],    %[ftmp13]                  \n\t"
+        "punpcklwd  %[ftmp13],  %[ftmp13],   %[ftmp13]     \n\t"
+        "1:                                                \n\t"
+        /* Get 8 data per row */
+        "gsldlc1    %[ftmp5],   0x07(%[src])               \n\t"
+        "gsldrc1    %[ftmp5],   0x00(%[src])               \n\t"
+        "gsldlc1    %[ftmp7],   0x08(%[src])               \n\t"
+        "gsldrc1    %[ftmp7],   0x01(%[src])               \n\t"
+        "gsldlc1    %[ftmp9],   0x09(%[src])               \n\t"
+        "gsldrc1    %[ftmp9],   0x02(%[src])               \n\t"
+        "gsldlc1    %[ftmp11],  0x0A(%[src])               \n\t"
+        "gsldrc1    %[ftmp11],  0x03(%[src])               \n\t"
+        "punpcklbh  %[ftmp4],   %[ftmp5],    %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp5],   %[ftmp5],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp6],   %[ftmp7],    %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp7],   %[ftmp7],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp8],   %[ftmp9],    %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp9],   %[ftmp9],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp10],  %[ftmp11],   %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp11],  %[ftmp11],   %[ftmp0]      \n\t"
+        PTR_ADDIU  "%[width],   %[width],    -0x04         \n\t"
+        /* Get raw data */
+        GET_DATA_H_MMI
+        ROUND_POWER_OF_TWO_MMI(%[srcl], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        ROUND_POWER_OF_TWO_MMI(%[srch], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        "packsswh   %[srcl],    %[srcl],     %[srch]       \n\t"
+        "packushb   %[ftmp12],  %[srcl],     %[ftmp0]      \n\t"
+        "swc1       %[ftmp12],  0x00(%[dst])               \n\t"
+        PTR_ADDIU  "%[dst],     %[dst],      0x04          \n\t"
+        PTR_ADDIU  "%[src],     %[src],      0x04          \n\t"
+        /* Loop count */
+        "bnez       %[width],   1b                         \n\t"
+        "move       %[width],   %[tmp1]                    \n\t"
+        PTR_ADDU   "%[src],     %[src],      %[src_stride] \n\t"
+        PTR_ADDU   "%[dst],     %[dst],      %[dst_stride] \n\t"
+        PTR_ADDIU  "%[height],  %[height],   -0x01         \n\t"
+        "bnez       %[height],  1b                         \n\t"
+        : [srcl]"=&f"(ftmp[0]),     [srch]"=&f"(ftmp[1]),
+          [filter1]"=&f"(ftmp[2]),  [filter2]"=&f"(ftmp[3]),
+          [ftmp0]"=&f"(ftmp[4]),    [ftmp4]"=&f"(ftmp[5]),
+          [ftmp5]"=&f"(ftmp[6]),    [ftmp6]"=&f"(ftmp[7]),
+          [ftmp7]"=&f"(ftmp[8]),    [ftmp8]"=&f"(ftmp[9]),
+          [ftmp9]"=&f"(ftmp[10]),   [ftmp10]"=&f"(ftmp[11]),
+          [ftmp11]"=&f"(ftmp[12]),  [ftmp12]"=&f"(ftmp[13]),
+          [tmp0]"=&r"(tmp[0]),      [tmp1]"=&r"(tmp[1]),
+          [src]"+&r"(src),          [width]"+&r"(w),
+          [dst]"+&r"(dst),          [height]"+&r"(h),
+          [ftmp13]"=&f"(ftmp[14])
+        : [filter]"r"(filter_x),
+          [src_stride]"r"((mips_reg)src_stride),
+          [dst_stride]"r"((mips_reg)dst_stride)
+        : "memory"
+    );
+}
+
+static void convolve_vert_mmi(const uint8_t *src, int32_t src_stride,
+                              uint8_t *dst, int32_t dst_stride,
+                              const int16_t *filter_y, int32_t w,
+                              int32_t h)
+{
+    double ftmp[17];
+    uint32_t tmp[1];
+    ptrdiff_t addr = src_stride;
+    src_stride -= w;
+    dst_stride -= w;
+
+    __asm__ volatile (
+        "xor        %[ftmp0],    %[ftmp0],   %[ftmp0]      \n\t"
+        "gsldlc1    %[ftmp4],    0x03(%[filter])           \n\t"
+        "gsldrc1    %[ftmp4],    0x00(%[filter])           \n\t"
+        "gsldlc1    %[ftmp5],    0x0b(%[filter])           \n\t"
+        "gsldrc1    %[ftmp5],    0x08(%[filter])           \n\t"
+        "punpcklwd  %[filter10], %[ftmp4],   %[ftmp4]      \n\t"
+        "punpckhwd  %[filter32], %[ftmp4],   %[ftmp4]      \n\t"
+        "punpcklwd  %[filter54], %[ftmp5],   %[ftmp5]      \n\t"
+        "punpckhwd  %[filter76], %[ftmp5],   %[ftmp5]      \n\t"
+        "li         %[tmp0],     0x07                      \n\t"
+        "dmtc1      %[tmp0],     %[ftmp13]                 \n\t"
+        "punpcklwd  %[ftmp13],   %[ftmp13],  %[ftmp13]     \n\t"
+        "1:                                                \n\t"
+        /* Get 8 data per column */
+        "gsldlc1    %[ftmp4],    0x07(%[src])              \n\t"
+        "gsldrc1    %[ftmp4],    0x00(%[src])              \n\t"
+        PTR_ADDU   "%[tmp0],     %[src],     %[addr]       \n\t"
+        "gsldlc1    %[ftmp5],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp5],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp6],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp6],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp7],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp7],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp8],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp8],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp9],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp9],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp10],   0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp10],   0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp11],   0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp11],   0x00(%[tmp0])             \n\t"
+        "punpcklbh  %[ftmp4],    %[ftmp4],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp5],    %[ftmp5],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp6],    %[ftmp6],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp7],    %[ftmp7],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp8],    %[ftmp8],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp9],    %[ftmp9],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp10],   %[ftmp10],  %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp11],   %[ftmp11],  %[ftmp0]      \n\t"
+        PTR_ADDIU  "%[width],    %[width],   -0x04         \n\t"
+        /* Get raw data */
+        GET_DATA_V_MMI
+        ROUND_POWER_OF_TWO_MMI(%[srcl], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        ROUND_POWER_OF_TWO_MMI(%[srch], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        "packsswh   %[srcl],     %[srcl],    %[srch]       \n\t"
+        "packushb   %[ftmp12],   %[srcl],    %[ftmp0]      \n\t"
+        "swc1       %[ftmp12],   0x00(%[dst])              \n\t"
+        PTR_ADDIU  "%[dst],      %[dst],      0x04         \n\t"
+        PTR_ADDIU  "%[src],      %[src],      0x04         \n\t"
+        /* Loop count */
+        "bnez       %[width],    1b                        \n\t"
+        PTR_SUBU   "%[width],    %[addr],    %[src_stride] \n\t"
+        PTR_ADDU   "%[src],      %[src],     %[src_stride] \n\t"
+        PTR_ADDU   "%[dst],      %[dst],     %[dst_stride] \n\t"
+        PTR_ADDIU  "%[height],   %[height],  -0x01         \n\t"
+        "bnez       %[height],   1b                        \n\t"
+        : [srcl]"=&f"(ftmp[0]),     [srch]"=&f"(ftmp[1]),
+          [filter10]"=&f"(ftmp[2]), [filter32]"=&f"(ftmp[3]),
+          [filter54]"=&f"(ftmp[4]), [filter76]"=&f"(ftmp[5]),
+          [ftmp0]"=&f"(ftmp[6]),    [ftmp4]"=&f"(ftmp[7]),
+          [ftmp5]"=&f"(ftmp[8]),    [ftmp6]"=&f"(ftmp[9]),
+          [ftmp7]"=&f"(ftmp[10]),   [ftmp8]"=&f"(ftmp[11]),
+          [ftmp9]"=&f"(ftmp[12]),   [ftmp10]"=&f"(ftmp[13]),
+          [ftmp11]"=&f"(ftmp[14]),  [ftmp12]"=&f"(ftmp[15]),
+          [src]"+&r"(src),          [dst]"+&r"(dst),
+          [width]"+&r"(w),          [height]"+&r"(h),
+          [tmp0]"=&r"(tmp[0]),      [ftmp13]"=&f"(ftmp[16])
+        : [filter]"r"(filter_y),
+          [src_stride]"r"((mips_reg)src_stride),
+          [dst_stride]"r"((mips_reg)dst_stride),
+          [addr]"r"((mips_reg)addr)
+        : "memory"
+    );
+}
+
+static void convolve_avg_horiz_mmi(const uint8_t *src, int32_t src_stride,
+                                   uint8_t *dst, int32_t dst_stride,
+                                   const uint16_t *filter_x, int32_t w,
+                                   int32_t h)
+{
+    double ftmp[15];
+    uint32_t tmp[2];
+    src -= 3;
+    src_stride -= w;
+    dst_stride -= w;
+
+    __asm__ volatile (
+        "move       %[tmp1],    %[width]                   \n\t"
+        "xor        %[ftmp0],   %[ftmp0],    %[ftmp0]      \n\t"
+        "gsldlc1    %[filter1], 0x03(%[filter])            \n\t"
+        "gsldrc1    %[filter1], 0x00(%[filter])            \n\t"
+        "gsldlc1    %[filter2], 0x0b(%[filter])            \n\t"
+        "gsldrc1    %[filter2], 0x08(%[filter])            \n\t"
+        "li         %[tmp0],    0x07                       \n\t"
+        "dmtc1      %[tmp0],    %[ftmp13]                  \n\t"
+        "punpcklwd  %[ftmp13],  %[ftmp13],   %[ftmp13]     \n\t"
+        "1:                                                \n\t"
+        /* Get 8 data per row */
+        "gsldlc1    %[ftmp5],   0x07(%[src])               \n\t"
+        "gsldrc1    %[ftmp5],   0x00(%[src])               \n\t"
+        "gsldlc1    %[ftmp7],   0x08(%[src])               \n\t"
+        "gsldrc1    %[ftmp7],   0x01(%[src])               \n\t"
+        "gsldlc1    %[ftmp9],   0x09(%[src])               \n\t"
+        "gsldrc1    %[ftmp9],   0x02(%[src])               \n\t"
+        "gsldlc1    %[ftmp11],  0x0A(%[src])               \n\t"
+        "gsldrc1    %[ftmp11],  0x03(%[src])               \n\t"
+        "punpcklbh  %[ftmp4],   %[ftmp5],    %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp5],   %[ftmp5],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp6],   %[ftmp7],    %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp7],   %[ftmp7],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp8],   %[ftmp9],    %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp9],   %[ftmp9],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp10],  %[ftmp11],   %[ftmp0]      \n\t"
+        "punpckhbh  %[ftmp11],  %[ftmp11],   %[ftmp0]      \n\t"
+        PTR_ADDIU  "%[width],   %[width],    -0x04         \n\t"
+        /* Get raw data */
+        GET_DATA_H_MMI
+        ROUND_POWER_OF_TWO_MMI(%[srcl], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        ROUND_POWER_OF_TWO_MMI(%[srch], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        "packsswh   %[srcl],    %[srcl],     %[srch]       \n\t"
+        "packushb   %[ftmp12],  %[srcl],     %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp12],  %[ftmp12],   %[ftmp0]      \n\t"
+        "gsldlc1    %[ftmp4],   0x07(%[dst])               \n\t"
+        "gsldrc1    %[ftmp4],   0x00(%[dst])               \n\t"
+        "punpcklbh  %[ftmp4],   %[ftmp4],    %[ftmp0]      \n\t"
+        "paddh      %[ftmp12],  %[ftmp12],   %[ftmp4]      \n\t"
+        "li         %[tmp0],    0x10001                    \n\t"
+        "dmtc1      %[tmp0],    %[ftmp5]                   \n\t"
+        "punpcklhw  %[ftmp5],   %[ftmp5],    %[ftmp5]      \n\t"
+        "paddh      %[ftmp12],  %[ftmp12],   %[ftmp5]      \n\t"
+        "psrah      %[ftmp12],  %[ftmp12],   %[ftmp5]      \n\t"
+        "packushb   %[ftmp12],  %[ftmp12],   %[ftmp0]      \n\t"
+        "swc1       %[ftmp12],  0x00(%[dst])               \n\t"
+        PTR_ADDIU  "%[dst],     %[dst],      0x04          \n\t"
+        PTR_ADDIU  "%[src],     %[src],      0x04          \n\t"
+        /* Loop count */
+        "bnez       %[width],   1b                         \n\t"
+        "move       %[width],   %[tmp1]                    \n\t"
+        PTR_ADDU   "%[src],     %[src],      %[src_stride] \n\t"
+        PTR_ADDU   "%[dst],     %[dst],      %[dst_stride] \n\t"
+        PTR_ADDIU  "%[height],  %[height],   -0x01         \n\t"
+        "bnez       %[height],  1b                         \n\t"
+        : [srcl]"=&f"(ftmp[0]),     [srch]"=&f"(ftmp[1]),
+          [filter1]"=&f"(ftmp[2]),  [filter2]"=&f"(ftmp[3]),
+          [ftmp0]"=&f"(ftmp[4]),    [ftmp4]"=&f"(ftmp[5]),
+          [ftmp5]"=&f"(ftmp[6]),    [ftmp6]"=&f"(ftmp[7]),
+          [ftmp7]"=&f"(ftmp[8]),    [ftmp8]"=&f"(ftmp[9]),
+          [ftmp9]"=&f"(ftmp[10]),   [ftmp10]"=&f"(ftmp[11]),
+          [ftmp11]"=&f"(ftmp[12]),  [ftmp12]"=&f"(ftmp[13]),
+          [tmp0]"=&r"(tmp[0]),      [tmp1]"=&r"(tmp[1]),
+          [src]"+&r"(src),          [width]"+&r"(w),
+          [dst]"+&r"(dst),          [height]"+&r"(h),
+          [ftmp13]"=&f"(ftmp[14])
+        : [filter]"r"(filter_x),
+          [src_stride]"r"((mips_reg)src_stride),
+          [dst_stride]"r"((mips_reg)dst_stride)
+        : "memory"
+    );
+}
+
+static void convolve_avg_vert_mmi(const uint8_t *src, int32_t src_stride,
+                                  uint8_t *dst, int32_t dst_stride,
+                                  const int16_t *filter_y, int32_t w,
+                                  int32_t h)
+{
+    double ftmp[17];
+    uint32_t tmp[1];
+    ptrdiff_t addr = src_stride;
+    src_stride -= w;
+    dst_stride -= w;
+
+    __asm__ volatile (
+        "xor        %[ftmp0],    %[ftmp0],   %[ftmp0]      \n\t"
+        "gsldlc1    %[ftmp4],    0x03(%[filter])           \n\t"
+        "gsldrc1    %[ftmp4],    0x00(%[filter])           \n\t"
+        "gsldlc1    %[ftmp5],    0x0b(%[filter])           \n\t"
+        "gsldrc1    %[ftmp5],    0x08(%[filter])           \n\t"
+        "punpcklwd  %[filter10], %[ftmp4],   %[ftmp4]      \n\t"
+        "punpckhwd  %[filter32], %[ftmp4],   %[ftmp4]      \n\t"
+        "punpcklwd  %[filter54], %[ftmp5],   %[ftmp5]      \n\t"
+        "punpckhwd  %[filter76], %[ftmp5],   %[ftmp5]      \n\t"
+        "li         %[tmp0],     0x07                      \n\t"
+        "dmtc1      %[tmp0],     %[ftmp13]                 \n\t"
+        "punpcklwd  %[ftmp13],   %[ftmp13],  %[ftmp13]     \n\t"
+        "1:                                                \n\t"
+        /* Get 8 data per column */
+        "gsldlc1    %[ftmp4],    0x07(%[src])              \n\t"
+        "gsldrc1    %[ftmp4],    0x00(%[src])              \n\t"
+        PTR_ADDU   "%[tmp0],     %[src],     %[addr]       \n\t"
+        "gsldlc1    %[ftmp5],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp5],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp6],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp6],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp7],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp7],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp8],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp8],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp9],    0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp9],    0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp10],   0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp10],   0x00(%[tmp0])             \n\t"
+        PTR_ADDU   "%[tmp0],     %[tmp0],    %[addr]       \n\t"
+        "gsldlc1    %[ftmp11],   0x07(%[tmp0])             \n\t"
+        "gsldrc1    %[ftmp11],   0x00(%[tmp0])             \n\t"
+        "punpcklbh  %[ftmp4],    %[ftmp4],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp5],    %[ftmp5],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp6],    %[ftmp6],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp7],    %[ftmp7],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp8],    %[ftmp8],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp9],    %[ftmp9],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp10],   %[ftmp10],  %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp11],   %[ftmp11],  %[ftmp0]      \n\t"
+        PTR_ADDIU  "%[width],    %[width],   -0x04         \n\t"
+        /* Get raw data */
+        GET_DATA_V_MMI
+        ROUND_POWER_OF_TWO_MMI(%[srcl], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        ROUND_POWER_OF_TWO_MMI(%[srch], %[ftmp13], %[ftmp5],
+                               %[ftmp6], %[tmp0])
+        "packsswh   %[srcl],     %[srcl],    %[srch]       \n\t"
+        "packushb   %[ftmp12],   %[srcl],    %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp12],   %[ftmp12],  %[ftmp0]      \n\t"
+        "gsldlc1    %[ftmp4],    0x07(%[dst])              \n\t"
+        "gsldrc1    %[ftmp4],    0x00(%[dst])              \n\t"
+        "punpcklbh  %[ftmp4],    %[ftmp4],   %[ftmp0]      \n\t"
+        "paddh      %[ftmp12],   %[ftmp12],  %[ftmp4]      \n\t"
+        "li         %[tmp0],     0x10001                   \n\t"
+        "dmtc1      %[tmp0],     %[ftmp5]                  \n\t"
+        "punpcklhw  %[ftmp5],    %[ftmp5],   %[ftmp5]      \n\t"
+        "paddh      %[ftmp12],   %[ftmp12],  %[ftmp5]      \n\t"
+        "psrah      %[ftmp12],   %[ftmp12],  %[ftmp5]      \n\t"
+        "packushb   %[ftmp12],   %[ftmp12],  %[ftmp0]      \n\t"
+        "swc1       %[ftmp12],   0x00(%[dst])              \n\t"
+        PTR_ADDIU  "%[dst],      %[dst],     0x04          \n\t"
+        PTR_ADDIU  "%[src],      %[src],     0x04          \n\t"
+        /* Loop count */
+        "bnez       %[width],    1b                        \n\t"
+        PTR_SUBU   "%[width],    %[addr],    %[src_stride] \n\t"
+        PTR_ADDU   "%[src],      %[src],     %[src_stride] \n\t"
+        PTR_ADDU   "%[dst],      %[dst],     %[dst_stride] \n\t"
+        PTR_ADDIU  "%[height],   %[height],  -0x01         \n\t"
+        "bnez       %[height],   1b                        \n\t"
+        : [srcl]"=&f"(ftmp[0]),     [srch]"=&f"(ftmp[1]),
+          [filter10]"=&f"(ftmp[2]), [filter32]"=&f"(ftmp[3]),
+          [filter54]"=&f"(ftmp[4]), [filter76]"=&f"(ftmp[5]),
+          [ftmp0]"=&f"(ftmp[6]),    [ftmp4]"=&f"(ftmp[7]),
+          [ftmp5]"=&f"(ftmp[8]),    [ftmp6]"=&f"(ftmp[9]),
+          [ftmp7]"=&f"(ftmp[10]),   [ftmp8]"=&f"(ftmp[11]),
+          [ftmp9]"=&f"(ftmp[12]),   [ftmp10]"=&f"(ftmp[13]),
+          [ftmp11]"=&f"(ftmp[14]),  [ftmp12]"=&f"(ftmp[15]),
+          [src]"+&r"(src),          [dst]"+&r"(dst),
+          [width]"+&r"(w),          [height]"+&r"(h),
+          [tmp0]"=&r"(tmp[0]),      [ftmp13]"=&f"(ftmp[16])
+        : [filter]"r"(filter_y),
+          [src_stride]"r"((mips_reg)src_stride),
+          [dst_stride]"r"((mips_reg)dst_stride),
+          [addr]"r"((mips_reg)addr)
+        : "memory"
+    );
+}
+
+static void convolve_avg_mmi(const uint8_t *src, int32_t src_stride,
+                             uint8_t *dst, int32_t dst_stride,
+                             int32_t w, int32_t h)
+{
+    double ftmp[4];
+    uint32_t tmp[2];
+    src_stride -= w;
+    dst_stride -= w;
+
+    __asm__ volatile (
+        "move       %[tmp1],    %[width]                  \n\t"
+        "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]      \n\t"
+        "li         %[tmp0],    0x10001                   \n\t"
+        "dmtc1      %[tmp0],    %[ftmp3]                  \n\t"
+        "punpcklhw  %[ftmp3],   %[ftmp3],   %[ftmp3]      \n\t"
+        "1:                                               \n\t"
+        "gsldlc1    %[ftmp1],   0x07(%[src])              \n\t"
+        "gsldrc1    %[ftmp1],   0x00(%[src])              \n\t"
+        "gsldlc1    %[ftmp2],   0x07(%[dst])              \n\t"
+        "gsldrc1    %[ftmp2],   0x00(%[dst])              \n\t"
+        "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]      \n\t"
+        "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]      \n\t"
+        "paddh      %[ftmp1],   %[ftmp1],   %[ftmp2]      \n\t"
+        "paddh      %[ftmp1],   %[ftmp1],   %[ftmp3]      \n\t"
+        "psrah      %[ftmp1],   %[ftmp1],   %[ftmp3]      \n\t"
+        "packushb   %[ftmp1],   %[ftmp1],   %[ftmp0]      \n\t"
+        "swc1       %[ftmp1],   0x00(%[dst])              \n\t"
+        PTR_ADDIU  "%[width],   %[width],   -0x04         \n\t"
+        PTR_ADDIU  "%[dst],     %[dst],     0x04          \n\t"
+        PTR_ADDIU  "%[src],     %[src],     0x04          \n\t"
+        "bnez       %[width],   1b                        \n\t"
+        "move       %[width],   %[tmp1]                   \n\t"
+        PTR_ADDU   "%[dst],     %[dst],     %[dst_stride] \n\t"
+        PTR_ADDU   "%[src],     %[src],     %[src_stride] \n\t"
+        PTR_ADDIU  "%[height],  %[height],  -0x01         \n\t"
+        "bnez       %[height],  1b                        \n\t"
+        : [ftmp0]"=&f"(ftmp[0]),  [ftmp1]"=&f"(ftmp[1]),
+          [ftmp2]"=&f"(ftmp[2]),  [ftmp3]"=&f"(ftmp[3]),
+          [tmp0]"=&r"(tmp[0]),    [tmp1]"=&r"(tmp[1]),
+          [src]"+&r"(src),        [dst]"+&r"(dst),
+          [width]"+&r"(w),        [height]"+&r"(h)
+        : [src_stride]"r"((mips_reg)src_stride),
+          [dst_stride]"r"((mips_reg)dst_stride)
+        : "memory"
+    );
+}
+
+#define VP9_COMMON_MIPS_MMI_FUNC(SIZE)                                          \
+static void common_hz_8t_##SIZE##w_mmi(const uint8_t *src, int32_t src_stride,  \
+                                       uint8_t *dst, int32_t dst_stride,        \
+                                       const int16_t *filter, int32_t h)        \
+{                                                                               \
+    convolve_horiz_mmi(src, src_stride, dst, dst_stride, filter, SIZE, h);      \
+}                                                                               \
+                                                                                \
+static void common_vt_8t_##SIZE##w_mmi(const uint8_t *src, int32_t src_stride,  \
+                                       uint8_t *dst, int32_t dst_stride,        \
+                                       const int16_t *filter, int32_t h)        \
+{                                                                               \
+    src -= (3 * src_stride);                                                    \
+    convolve_vert_mmi(src, src_stride, dst, dst_stride, filter, SIZE, h);       \
+}                                                                               \
+                                                                                \
+static void common_hv_8ht_8vt_##SIZE##w_mmi(const uint8_t *src,                 \
+                                            int32_t src_stride, uint8_t *dst,   \
+                                            int32_t dst_stride,                 \
+                                            const int16_t *filter_horiz,        \
+                                            const int16_t *filter_vert,         \
+                                            int32_t h)                          \
+{                                                                               \
+    int tmp_h = h + 7;                                                          \
+    uint8_t temp[64 * 71];                                                      \
+    src -= (3 * src_stride);                                                    \
+    convolve_horiz_mmi(src, src_stride, temp, 64, filter_horiz, SIZE, tmp_h);   \
+    convolve_vert_mmi(temp, 64, dst, dst_stride, filter_vert, SIZE, h);         \
+}                                                                               \
+                                                                                \
+static void common_hz_8t_and_aver_dst_##SIZE##w_mmi(const uint8_t *src,         \
+                                                    int32_t src_stride,         \
+                                                    uint8_t *dst,               \
+                                                    int32_t dst_stride,         \
+                                                    const int16_t *filter,      \
+                                                    int32_t h)                  \
+{                                                                               \
+    convolve_avg_horiz_mmi(src, src_stride, dst, dst_stride, filter, SIZE, h);  \
+}                                                                               \
+                                                                                \
+static void common_vt_8t_and_aver_dst_##SIZE##w_mmi(const uint8_t *src,         \
+                                                    int32_t src_stride,         \
+                                                    uint8_t *dst,               \
+                                                    int32_t dst_stride,         \
+                                                    const int16_t *filter,      \
+                                                    int32_t h)                  \
+{                                                                               \
+    src -= (3 * src_stride);                                                    \
+    convolve_avg_vert_mmi(src, src_stride, dst, dst_stride, filter, SIZE, h);   \
+}                                                                               \
+                                                                                \
+static void common_hv_8ht_8vt_and_aver_dst_##SIZE##w_mmi(const uint8_t *src,    \
+                                                   int32_t src_stride,          \
+                                                   uint8_t *dst,                \
+                                                   int32_t dst_stride,          \
+                                                   const int16_t *filter_horiz, \
+                                                   const int16_t *filter_vert,  \
+                                                   int32_t h)                   \
+{                                                                               \
+    uint8_t temp[64 * 64];                                                      \
+    common_hv_8ht_8vt_##SIZE##w_mmi(src, src_stride, temp, 64,                  \
+                                    filter_horiz, filter_vert, h);              \
+    convolve_avg_mmi(temp, 64, dst, dst_stride, SIZE, h);                       \
+}
+
+VP9_COMMON_MIPS_MMI_FUNC(64);
+VP9_COMMON_MIPS_MMI_FUNC(32);
+VP9_COMMON_MIPS_MMI_FUNC(16);
+VP9_COMMON_MIPS_MMI_FUNC(8);
+VP9_COMMON_MIPS_MMI_FUNC(4);
+
+static const int16_t vp9_subpel_filters_mmi[3][15][8] = {
+    [FILTER_8TAP_REGULAR] = {
+         {0, 1, -5, 126, 8, -3, 1, 0},
+         {-1, 3, -10, 122, 18, -6, 2, 0},
+         {-1, 4, -13, 118, 27, -9, 3, -1},
+         {-1, 4, -16, 112, 37, -11, 4, -1},
+         {-1, 5, -18, 105, 48, -14, 4, -1},
+         {-1, 5, -19, 97, 58, -16, 5, -1},
+         {-1, 6, -19, 88, 68, -18, 5, -1},
+         {-1, 6, -19, 78, 78, -19, 6, -1},
+         {-1, 5, -18, 68, 88, -19, 6, -1},
+         {-1, 5, -16, 58, 97, -19, 5, -1},
+         {-1, 4, -14, 48, 105, -18, 5, -1},
+         {-1, 4, -11, 37, 112, -16, 4, -1},
+         {-1, 3, -9, 27, 118, -13, 4, -1},
+         {0, 2, -6, 18, 122, -10, 3, -1},
+         {0, 1, -3, 8, 126, -5, 1, 0},
+    }, [FILTER_8TAP_SHARP] = {
+        {-1, 3, -7, 127, 8, -3, 1, 0},
+        {-2, 5, -13, 125, 17, -6, 3, -1},
+        {-3, 7, -17, 121, 27, -10, 5, -2},
+        {-4, 9, -20, 115, 37, -13, 6, -2},
+        {-4, 10, -23, 108, 48, -16, 8, -3},
+        {-4, 10, -24, 100, 59, -19, 9, -3},
+        {-4, 11, -24, 90, 70, -21, 10, -4},
+        {-4, 11, -23, 80, 80, -23, 11, -4},
+        {-4, 10, -21, 70, 90, -24, 11, -4},
+        {-3, 9, -19, 59, 100, -24, 10, -4},
+        {-3, 8, -16, 48, 108, -23, 10, -4},
+        {-2, 6, -13, 37, 115, -20, 9, -4},
+        {-2, 5, -10, 27, 121, -17, 7, -3},
+        {-1, 3, -6, 17, 125, -13, 5, -2},
+        {0, 1, -3, 8, 127, -7, 3, -1},
+    }, [FILTER_8TAP_SMOOTH] = {
+        {-3, -1, 32, 64, 38, 1, -3, 0},
+        {-2, -2, 29, 63, 41, 2, -3, 0},
+        {-2, -2, 26, 63, 43, 4, -4, 0},
+        {-2, -3, 24, 62, 46, 5, -4, 0},
+        {-2, -3, 21, 60, 49, 7, -4, 0},
+        {-1, -4, 18, 59, 51, 9, -4, 0},
+        {-1, -4, 16, 57, 53, 12, -4, -1},
+        {-1, -4, 14, 55, 55, 14, -4, -1},
+        {-1, -4, 12, 53, 57, 16, -4, -1},
+        {0, -4, 9, 51, 59, 18, -4, -1},
+        {0, -4, 7, 49, 60, 21, -3, -2},
+        {0, -4, 5, 46, 62, 24, -3, -2},
+        {0, -4, 4, 43, 63, 26, -2, -2},
+        {0, -3, 2, 41, 63, 29, -2, -2},
+        {0, -3, 1, 38, 64, 32, -1, -3},
+    }
+};
+
+#define VP9_8TAP_MIPS_MMI_FUNC(SIZE, TYPE, TYPE_IDX)                           \
+void ff_put_8tap_##TYPE##_##SIZE##h_mmi(uint8_t *dst, ptrdiff_t dststride,     \
+                                        const uint8_t *src,                    \
+                                        ptrdiff_t srcstride,                   \
+                                        int h, int mx, int my)                 \
+{                                                                              \
+    const int16_t *filter = vp9_subpel_filters_mmi[TYPE_IDX][mx-1];            \
+                                                                               \
+    common_hz_8t_##SIZE##w_mmi(src, srcstride, dst, dststride, filter, h);     \
+}                                                                              \
+                                                                               \
+void ff_put_8tap_##TYPE##_##SIZE##v_mmi(uint8_t *dst, ptrdiff_t dststride,     \
+                                        const uint8_t *src,                    \
+                                        ptrdiff_t srcstride,                   \
+                                        int h, int mx, int my)                 \
+{                                                                              \
+    const int16_t *filter = vp9_subpel_filters_mmi[TYPE_IDX][my-1];            \
+                                                                               \
+    common_vt_8t_##SIZE##w_mmi(src, srcstride, dst, dststride, filter, h);     \
+}                                                                              \
+                                                                               \
+void ff_put_8tap_##TYPE##_##SIZE##hv_mmi(uint8_t *dst, ptrdiff_t dststride,    \
+                                         const uint8_t *src,                   \
+                                         ptrdiff_t srcstride,                  \
+                                         int h, int mx, int my)                \
+{                                                                              \
+    const uint16_t *hfilter = vp9_subpel_filters_mmi[TYPE_IDX][mx-1];          \
+    const uint16_t *vfilter = vp9_subpel_filters_mmi[TYPE_IDX][my-1];          \
+                                                                               \
+    common_hv_8ht_8vt_##SIZE##w_mmi(src, srcstride, dst, dststride, hfilter,   \
+                                    vfilter, h);                               \
+}                                                                              \
+                                                                               \
+void ff_avg_8tap_##TYPE##_##SIZE##h_mmi(uint8_t *dst, ptrdiff_t dststride,     \
+                                        const uint8_t *src,                    \
+                                        ptrdiff_t srcstride,                   \
+                                        int h, int mx, int my)                 \
+{                                                                              \
+    const int16_t *filter = vp9_subpel_filters_mmi[TYPE_IDX][mx-1];            \
+                                                                               \
+    common_hz_8t_and_aver_dst_##SIZE##w_mmi(src, srcstride, dst,               \
+                                            dststride, filter, h);             \
+}                                                                              \
+                                                                               \
+void ff_avg_8tap_##TYPE##_##SIZE##v_mmi(uint8_t *dst, ptrdiff_t dststride,     \
+                                        const uint8_t *src,                    \
+                                        ptrdiff_t srcstride,                   \
+                                        int h, int mx, int my)                 \
+{                                                                              \
+    const int16_t *filter = vp9_subpel_filters_mmi[TYPE_IDX][my-1];            \
+                                                                               \
+    common_vt_8t_and_aver_dst_##SIZE##w_mmi(src, srcstride, dst, dststride,    \
+                                            filter, h);                        \
+}                                                                              \
+                                                                               \
+void ff_avg_8tap_##TYPE##_##SIZE##hv_mmi(uint8_t *dst, ptrdiff_t dststride,    \
+                                         const uint8_t *src,                   \
+                                         ptrdiff_t srcstride,                  \
+                                         int h, int mx, int my)                \
+{                                                                              \
+    const uint16_t *hfilter = vp9_subpel_filters_mmi[TYPE_IDX][mx-1];          \
+    const uint16_t *vfilter = vp9_subpel_filters_mmi[TYPE_IDX][my-1];          \
+                                                                               \
+    common_hv_8ht_8vt_and_aver_dst_##SIZE##w_mmi(src, srcstride, dst,          \
+                                                 dststride, hfilter,           \
+                                                 vfilter, h);                  \
+}
+
+VP9_8TAP_MIPS_MMI_FUNC(64, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(32, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(16, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(8, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(4, regular, FILTER_8TAP_REGULAR);
+
+VP9_8TAP_MIPS_MMI_FUNC(64, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(32, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(16, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(8, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(4, sharp, FILTER_8TAP_SHARP);
+
+VP9_8TAP_MIPS_MMI_FUNC(64, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(32, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(16, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(8, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(4, smooth, FILTER_8TAP_SMOOTH);
+
+#undef VP9_8TAP_MIPS_MMI_FUNC
diff --git a/libavcodec/mips/vp9dsp_init_mips.c b/libavcodec/mips/vp9dsp_init_mips.c
index c8a4890..5990fa6 100644
--- a/libavcodec/mips/vp9dsp_init_mips.c
+++ b/libavcodec/mips/vp9dsp_init_mips.c
@@ -168,8 +168,50 @@ static av_cold void vp9dsp_init_msa(VP9DSPContext *dsp, int bpp)
 }
 #endif  // #if HAVE_MSA
 
+#if HAVE_MMI
+static av_cold void vp9dsp_mc_init_mmi(VP9DSPContext *dsp)
+{
+#define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type)  \
+    dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] =   \
+        ff_##type##_8tap_smooth_##sz##dir##_mmi;             \
+    dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] =   \
+        ff_##type##_8tap_regular_##sz##dir##_mmi;            \
+    dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][idxh][idxv] =   \
+        ff_##type##_8tap_sharp_##sz##dir##_mmi;
+
+#define init_subpel2(idx, idxh, idxv, dir, type)      \
+    init_subpel1(0, idx, idxh, idxv, 64, dir, type);  \
+    init_subpel1(1, idx, idxh, idxv, 32, dir, type);  \
+    init_subpel1(2, idx, idxh, idxv, 16, dir, type);  \
+    init_subpel1(3, idx, idxh, idxv,  8, dir, type);  \
+    init_subpel1(4, idx, idxh, idxv,  4, dir, type)
+
+#define init_subpel3(idx, type)         \
+    init_subpel2(idx, 1, 1, hv, type);  \
+    init_subpel2(idx, 0, 1, v, type);   \
+    init_subpel2(idx, 1, 0, h, type)
+
+    init_subpel3(0, put);
+    init_subpel3(1, avg);
+
+#undef init_subpel1
+#undef init_subpel2
+#undef init_subpel3
+}
+
+static av_cold void vp9dsp_init_mmi(VP9DSPContext *dsp, int bpp)
+{
+    if (bpp == 8) {
+        vp9dsp_mc_init_mmi(dsp);
+    }
+}
+#endif  // #if HAVE_MMI
+
 av_cold void ff_vp9dsp_init_mips(VP9DSPContext *dsp, int bpp)
 {
+#if HAVE_MMI
+    vp9dsp_init_mmi(dsp, bpp);
+#endif  // #if HAVE_MMI
 #if HAVE_MSA
     vp9dsp_init_msa(dsp, bpp);
 #endif  // #if HAVE_MSA
diff --git a/libavcodec/mips/vp9dsp_mips.h b/libavcodec/mips/vp9dsp_mips.h
index 4d73038..0b6ce7c 100644
--- a/libavcodec/mips/vp9dsp_mips.h
+++ b/libavcodec/mips/vp9dsp_mips.h
@@ -234,4 +234,54 @@ void ff_tm_16x16_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
 void ff_tm_32x32_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
                      const uint8_t *top);
 
+#define VP9_8TAP_MIPS_MMI_FUNC(SIZE, type, type_idx)                         \
+void ff_put_8tap_##type##_##SIZE##h_mmi(uint8_t *dst, ptrdiff_t dststride,   \
+                                        const uint8_t *src,                  \
+                                        ptrdiff_t srcstride,                 \
+                                        int h, int mx, int my);              \
+                                                                             \
+void ff_put_8tap_##type##_##SIZE##v_mmi(uint8_t *dst, ptrdiff_t dststride,   \
+                                        const uint8_t *src,                  \
+                                        ptrdiff_t srcstride,                 \
+                                        int h, int mx, int my);              \
+                                                                             \
+void ff_put_8tap_##type##_##SIZE##hv_mmi(uint8_t *dst, ptrdiff_t dststride,  \
+                                         const uint8_t *src,                 \
+                                         ptrdiff_t srcstride,                \
+                                         int h, int mx, int my);             \
+                                                                             \
+void ff_avg_8tap_##type##_##SIZE##h_mmi(uint8_t *dst, ptrdiff_t dststride,   \
+                                        const uint8_t *src,                  \
+                                        ptrdiff_t srcstride,                 \
+                                        int h, int mx, int my);              \
+                                                                             \
+void ff_avg_8tap_##type##_##SIZE##v_mmi(uint8_t *dst, ptrdiff_t dststride,   \
+                                        const uint8_t *src,                  \
+                                        ptrdiff_t srcstride,                 \
+                                        int h, int mx, int my);              \
+                                                                             \
+void ff_avg_8tap_##type##_##SIZE##hv_mmi(uint8_t *dst, ptrdiff_t dststride,  \
+                                         const uint8_t *src,                 \
+                                         ptrdiff_t srcstride,                \
+                                         int h, int mx, int my);
+
+VP9_8TAP_MIPS_MMI_FUNC(64, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(32, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(16, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(8, regular, FILTER_8TAP_REGULAR);
+VP9_8TAP_MIPS_MMI_FUNC(4, regular, FILTER_8TAP_REGULAR);
+
+VP9_8TAP_MIPS_MMI_FUNC(64, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(32, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(16, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(8, sharp, FILTER_8TAP_SHARP);
+VP9_8TAP_MIPS_MMI_FUNC(4, sharp, FILTER_8TAP_SHARP);
+
+VP9_8TAP_MIPS_MMI_FUNC(64, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(32, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(16, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(8, smooth, FILTER_8TAP_SMOOTH);
+VP9_8TAP_MIPS_MMI_FUNC(4, smooth, FILTER_8TAP_SMOOTH);
+#undef VP9_8TAP_MIPS_MMI_FUNC
+
 #endif  // #ifndef AVCODEC_MIPS_VP9DSP_MIPS_H
diff --git a/libavutil/mips/mmiutils.h b/libavutil/mips/mmiutils.h
index 5fc1908..05f6b31 100644
--- a/libavutil/mips/mmiutils.h
+++ b/libavutil/mips/mmiutils.h
@@ -345,5 +345,20 @@
         PSRAH_4_MMI(fp1, fp2, fp3, fp4, shift)                            \
         PSRAH_4_MMI(fp5, fp6, fp7, fp8, shift)
 
+/**
+ * brief: (((value) + (1 << ((n) - 1))) >> (n))
+ * fr_i0: src & dst
+ * fr_i1: Operand number
+ * fr_t0, fr_t1: temporary FPR
+ * gr_t0: temporary GPR
+ */
+#define ROUND_POWER_OF_TWO_MMI(fr_i0, fr_i1, fr_t0, fr_t1, gr_t0)         \
+        "li         "#gr_t0",     0x01                              \n\t" \
+        "dmtc1      "#gr_t0",     "#fr_t0"                          \n\t" \
+        "punpcklwd  "#fr_t0",     "#fr_t0",    "#fr_t0"             \n\t" \
+        "psubw      "#fr_t1",     "#fr_i1",    "#fr_t0"             \n\t" \
+        "psllw      "#fr_t1",     "#fr_t0",    "#fr_t1"             \n\t" \
+        "paddw      "#fr_i0",     "#fr_i0",    "#fr_t1"             \n\t" \
+        "psraw      "#fr_i0",     "#fr_i0",    "#fr_i1"             \n\t"
 
 #endif /* AVUTILS_MIPS_MMIUTILS_H */
-- 
2.1.0




More information about the ffmpeg-devel mailing list