avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for AVC qpel functions

This patch adds MSA (MIPS-SIMD-Arch) optimizations for AVC qpel functions in new file h264qpel_msa.c
Adds new generic macros (needed for this patch) in libavutil/mips/generic_macros_msa.h

Added const to local static array.

Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/h264qpel.c

@@ -104,4 +104,6 @@ av_cold void ff_h264qpel_init(H264QpelContext *c, int bit_depth)
         ff_h264qpel_init_ppc(c, bit_depth);
     if (ARCH_X86)
         ff_h264qpel_init_x86(c, bit_depth);
+    if (ARCH_MIPS)
+        ff_h264qpel_init_mips(c, bit_depth);
 }

libavcodec/h264qpel.h

@@ -35,5 +35,6 @@ void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth);
 void ff_h264qpel_init_arm(H264QpelContext *c, int bit_depth);
 void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth);
 void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth);
+void ff_h264qpel_init_mips(H264QpelContext *c, int bit_depth);
 
 #endif /* AVCODEC_H264QPEL_H */

libavcodec/mips/Makefile

@@ -21,6 +21,7 @@ MIPSFPU-OBJS-$(CONFIG_AAC_ENCODER)        += mips/iirfilter_mips.o
 OBJS-$(CONFIG_HEVC_DECODER)               += mips/hevcdsp_init_mips.o      \
                                              mips/hevcpred_init_mips.o
 OBJS-$(CONFIG_H264DSP)                    += mips/h264dsp_init_mips.o
+OBJS-$(CONFIG_H264QPEL)                   += mips/h264qpel_init_mips.o
 OBJS-$(CONFIG_H264CHROMA)                 += mips/h264chroma_init_mips.o
 OBJS-$(CONFIG_H264PRED)                   += mips/h264pred_init_mips.o
 MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
@@ -33,6 +34,7 @@ MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
                                              mips/hevcpred_msa.o
 MSA-OBJS-$(CONFIG_H264DSP)                += mips/h264dsp_msa.o            \
                                              mips/h264idct_msa.o
+MSA-OBJS-$(CONFIG_H264QPEL)               += mips/h264qpel_msa.o
 MSA-OBJS-$(CONFIG_H264CHROMA)             += mips/h264chroma_msa.o
 MSA-OBJS-$(CONFIG_H264PRED)               += mips/h264pred_msa.o
 LOONGSON3-OBJS-$(CONFIG_H264DSP)          += mips/h264dsp_mmi.o

libavcodec/mips/h264dsp_mips.h

@@ -92,6 +92,204 @@ void ff_weight_h264_pixels8_8_msa(uint8_t *src, int stride, int height,
 void ff_weight_h264_pixels4_8_msa(uint8_t *src, int stride, int height,
                                   int log2_denom, int weight, int offset);
 
+void ff_put_h264_qpel16_mc00_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc10_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc20_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc30_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc01_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc11_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc21_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc31_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc02_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc12_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc22_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc32_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc03_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc13_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc23_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_put_h264_qpel16_mc33_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+
+void ff_put_h264_qpel8_mc00_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc10_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc20_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc30_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc01_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc11_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc21_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc31_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc02_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc12_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc22_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc32_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc03_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc13_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc23_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel8_mc33_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+
+void ff_put_h264_qpel4_mc00_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc10_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc20_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc30_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc01_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc11_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc21_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc31_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc02_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc12_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc22_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc32_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc03_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc13_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc23_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_put_h264_qpel4_mc33_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+
+void ff_avg_h264_qpel16_mc00_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc10_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc20_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc30_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc01_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc11_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc21_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc31_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc02_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc12_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc22_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc32_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc03_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc13_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc23_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+void ff_avg_h264_qpel16_mc33_msa(uint8_t *dst, const uint8_t *src,
+                                 ptrdiff_t dst_stride);
+
+void ff_avg_h264_qpel8_mc00_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc10_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc20_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc30_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc01_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc11_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc21_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc31_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc02_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc12_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc22_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc32_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc03_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc13_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc23_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel8_mc33_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+
+void ff_avg_h264_qpel4_mc00_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc10_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc20_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc30_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc01_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc11_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc21_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc31_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc02_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc12_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc22_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc32_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc03_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc13_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc23_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+void ff_avg_h264_qpel4_mc33_msa(uint8_t *dst, const uint8_t *src,
+                                ptrdiff_t dst_stride);
+
 void ff_h264_intra_predict_plane_8x8_msa(uint8_t *src, ptrdiff_t stride);
 void ff_h264_intra_predict_dc_4blk_8x8_msa(uint8_t *src, ptrdiff_t stride);
 void ff_h264_intra_predict_hor_dc_8x8_msa(uint8_t *src, ptrdiff_t stride);

libavcodec/mips/h264qpel_init_mips.c

@@ -0,0 +1,136 @@
+/*
+ * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar@imgtec.com)
+ *
+ * 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 "h264dsp_mips.h"
+
+#if HAVE_MSA
+static av_cold void h264qpel_init_msa(H264QpelContext *c, int bit_depth)
+{
+    if (8 == bit_depth) {
+        c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_msa;
+        c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_msa;
+        c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_msa;
+        c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_msa;
+        c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_msa;
+        c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_msa;
+        c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_msa;
+        c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_msa;
+        c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_msa;
+        c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_msa;
+        c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_msa;
+        c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_msa;
+        c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_msa;
+        c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_msa;
+        c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_msa;
+        c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_msa;
+
+        c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_msa;
+        c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_msa;
+        c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_msa;
+        c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_msa;
+        c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_msa;
+        c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_msa;
+        c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_msa;
+        c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_msa;
+        c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_msa;
+        c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_msa;
+        c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_msa;
+        c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_msa;
+        c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_msa;
+        c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_msa;
+        c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_msa;
+        c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_msa;
+
+        c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_msa;
+        c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_msa;
+        c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_msa;
+        c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_msa;
+        c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_msa;
+        c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_msa;
+        c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_msa;
+        c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_msa;
+        c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_msa;
+        c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_msa;
+        c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_msa;
+        c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_msa;
+        c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_msa;
+        c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_msa;
+        c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_msa;
+
+        c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_msa;
+        c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_msa;
+        c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_msa;
+        c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_msa;
+        c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_msa;
+        c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_msa;
+        c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_msa;
+        c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_msa;
+        c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_msa;
+        c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_msa;
+        c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_msa;
+        c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_msa;
+        c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_msa;
+        c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_msa;
+        c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_msa;
+        c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_msa;
+
+        c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_msa;
+        c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_msa;
+        c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_msa;
+        c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_msa;
+        c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_msa;
+        c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_msa;
+        c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_msa;
+        c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_msa;
+        c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_msa;
+        c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_msa;
+        c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_msa;
+        c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_msa;
+        c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_msa;
+        c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_msa;
+        c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_msa;
+        c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_msa;
+
+        c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_msa;
+        c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_msa;
+        c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_msa;
+        c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_msa;
+        c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_msa;
+        c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_msa;
+        c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_msa;
+        c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_msa;
+        c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_msa;
+        c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_msa;
+        c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_msa;
+        c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_msa;
+        c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_msa;
+        c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_msa;
+        c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_msa;
+        c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_msa;
+    }
+}
+#endif  // #if HAVE_MSA
+
+av_cold void ff_h264qpel_init_mips(H264QpelContext *c, int bit_depth)
+{
+#if HAVE_MSA
+    h264qpel_init_msa(c, bit_depth);
+#endif  // #if HAVE_MSA
+}

libavutil/mips/generic_macros_msa.h

@@ -883,6 +883,14 @@
 }
 #define VSHF_H2_SH(...) VSHF_H2(v8i16, __VA_ARGS__)
 
+#define VSHF_H3(RTYPE, in0, in1, in2, in3, in4, in5, mask0, mask1, mask2,  \
+                out0, out1, out2)                                          \
+{                                                                          \
+    VSHF_H2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1);          \
+    out2 = (RTYPE) __msa_vshf_h((v8i16) mask2, (v8i16) in5, (v8i16) in4);  \
+}
+#define VSHF_H3_SH(...) VSHF_H3(v8i16, __VA_ARGS__)
+
 /* Description : Shuffle byte vector elements as per mask vector
    Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
                  Outputs - out0, out1
@@ -1107,6 +1115,29 @@
     out_m;                                                \
 } )
 
+/* Description : Horizontal addition of signed byte vector elements
+   Arguments   : Inputs  - in0, in1
+                 Outputs - out0, out1
+                 Return Type - as per RTYPE
+   Details     : Each signed odd byte element from 'in0' is added to
+                 even signed byte element from 'in0' (pairwise) and the
+                 halfword result is stored in 'out0'
+*/
+#define HADD_SB2(RTYPE, in0, in1, out0, out1)                 \
+{                                                             \
+    out0 = (RTYPE) __msa_hadd_s_h((v16i8) in0, (v16i8) in0);  \
+    out1 = (RTYPE) __msa_hadd_s_h((v16i8) in1, (v16i8) in1);  \
+}
+#define HADD_SB2_SH(...) HADD_SB2(v8i16, __VA_ARGS__)
+
+#define HADD_SB4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3)  \
+{                                                                    \
+    HADD_SB2(RTYPE, in0, in1, out0, out1);                           \
+    HADD_SB2(RTYPE, in2, in3, out2, out3);                           \
+}
+#define HADD_SB4_UH(...) HADD_SB4(v8u16, __VA_ARGS__)
+#define HADD_SB4_SH(...) HADD_SB4(v8i16, __VA_ARGS__)
+
 /* Description : Horizontal addition of unsigned byte vector elements
    Arguments   : Inputs  - in0, in1
                  Outputs - out0, out1
@@ -1212,6 +1243,8 @@
     out1 = (RTYPE) __msa_ilvev_h((v8i16) in3, (v8i16) in2);  \
 }
 #define ILVEV_H2_UB(...) ILVEV_H2(v16u8, __VA_ARGS__)
+#define ILVEV_H2_SH(...) ILVEV_H2(v8i16, __VA_ARGS__)
+#define ILVEV_H2_SW(...) ILVEV_H2(v4i32, __VA_ARGS__)
 
 /* Description : Interleave even word elements from vectors
    Arguments   : Inputs  - in0, in1, in2, in3
@@ -1738,6 +1771,24 @@
 }
 #define PCKEV_D4_UB(...) PCKEV_D4(v16u8, __VA_ARGS__)
 
+/* Description : Pack odd double word elements of vector pairs
+   Arguments   : Inputs  - in0, in1
+                 Outputs - out0, out1
+                 Return Type - as per RTYPE
+   Details     : As operation is on same input 'in0' vector, index 1 double word
+                 element is overwritten to index 0 and result is written to out0
+                 As operation is on same input 'in1' vector, index 1 double word
+                 element is overwritten to index 0 and result is written to out1
+*/
+#define PCKOD_D2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                            \
+    out0 = (RTYPE) __msa_pckod_d((v2i64) in0, (v2i64) in1);  \
+    out1 = (RTYPE) __msa_pckod_d((v2i64) in2, (v2i64) in3);  \
+}
+#define PCKOD_D2_UB(...) PCKOD_D2(v16u8, __VA_ARGS__)
+#define PCKOD_D2_SH(...) PCKOD_D2(v8i16, __VA_ARGS__)
+#define PCKOD_D2_SD(...) PCKOD_D2(v2i64, __VA_ARGS__)
+
 /* Description : Each byte element is logically xor'ed with immediate 128
    Arguments   : Inputs  - in0, in1
                  Outputs - in0, in1 (in-place)
@@ -2336,6 +2387,37 @@
     out3 = (v4i32) __msa_ilvl_d((v2i64) s3_m, (v2i64) s1_m);            \
 }
 
+/* Description : Average rounded byte elements from pair of vectors and store
+                 8x4 byte block in destination memory
+   Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
+                 Outputs -
+                 Return Type -
+   Details     : Each byte element from input vector pair 'in0' and 'in1' are
+                 average rounded (a + b + 1)/2 and stored in 'tmp0_m'
+                 Each byte element from input vector pair 'in2' and 'in3' are
+                 average rounded (a + b + 1)/2 and stored in 'tmp1_m'
+                 Each byte element from input vector pair 'in4' and 'in5' are
+                 average rounded (a + b + 1)/2 and stored in 'tmp2_m'
+                 Each byte element from input vector pair 'in6' and 'in7' are
+                 average rounded (a + b + 1)/2 and stored in 'tmp3_m'
+                 The half vector results from all 4 vectors are stored in
+                 destination memory as 8x4 byte block
+*/
+#define AVER_ST8x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride)  \
+{                                                                            \
+    uint64_t out0_m, out1_m, out2_m, out3_m;                                 \
+    v16u8 tp0_m, tp1_m, tp2_m, tp3_m;                                        \
+                                                                             \
+    AVER_UB4_UB(in0, in1, in2, in3, in4, in5, in6, in7,                      \
+                tp0_m, tp1_m, tp2_m, tp3_m);                                 \
+                                                                             \
+    out0_m = __msa_copy_u_d((v2i64) tp0_m, 0);                               \
+    out1_m = __msa_copy_u_d((v2i64) tp1_m, 0);                               \
+    out2_m = __msa_copy_u_d((v2i64) tp2_m, 0);                               \
+    out3_m = __msa_copy_u_d((v2i64) tp3_m, 0);                               \
+    SD4(out0_m, out1_m, out2_m, out3_m, pdst, stride);                       \
+}
+
 /* Description : Add block 4x4
    Arguments   : Inputs  - in0, in1, in2, in3, pdst, stride
                  Outputs -
@@ -2368,6 +2450,30 @@
     SW4(out0_m, out1_m, out2_m, out3_m, pdst, stride);            \
 }
 
+/* Description : Dot product and addition of 3 signed halfword input vectors
+   Arguments   : Inputs  - in0, in1, in2, coeff0, coeff1, coeff2
+                 Outputs - out0_m
+                 Return Type - signed halfword
+   Details     : Dot product of 'in0' with 'coeff0'
+                 Dot product of 'in1' with 'coeff1'
+                 Dot product of 'in2' with 'coeff2'
+                 Addition of all the 3 vector results
+
+                 out0_m = (in0 * coeff0) + (in1 * coeff1) + (in2 * coeff2)
+*/
+#define DPADD_SH3_SH(in0, in1, in2, coeff0, coeff1, coeff2)         \
+( {                                                                 \
+    v8i16 tmp1_m;                                                   \
+    v8i16 out0_m;                                                   \
+                                                                    \
+    out0_m = __msa_dotp_s_h((v16i8) in0, (v16i8) coeff0);           \
+    out0_m = __msa_dpadd_s_h(out0_m, (v16i8) in1, (v16i8) coeff1);  \
+    tmp1_m = __msa_dotp_s_h((v16i8) in2, (v16i8) coeff2);           \
+    out0_m = __msa_adds_s_h(out0_m, tmp1_m);                        \
+                                                                    \
+    out0_m;                                                         \
+} )
+
 /* Description : Pack even elements of input vectors & xor with 128
    Arguments   : Inputs  - in0, in1
                  Outputs - out_m
@@ -2384,6 +2490,24 @@
     out_m;                                                    \
 } )
 
+/* Description : Converts inputs to unsigned bytes, interleave, average & store
+                 as 8x4 unsigned byte block
+   Arguments   : Inputs  - in0, in1, in2, in3, dst0, dst1, dst2, dst3,
+                           pdst, stride
+*/
+#define CONVERT_UB_AVG_ST8x4_UB(in0, in1, in2, in3,                    \
+                                dst0, dst1, dst2, dst3, pdst, stride)  \
+{                                                                      \
+    v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m;                              \
+    uint8_t *pdst_m = (uint8_t *) (pdst);                              \
+                                                                       \
+    tmp0_m = PCKEV_XORI128_UB(in0, in1);                               \
+    tmp1_m = PCKEV_XORI128_UB(in2, in3);                               \
+    ILVR_D2_UB(dst1, dst0, dst3, dst2, tmp2_m, tmp3_m);                \
+    AVER_UB2_UB(tmp0_m, tmp2_m, tmp1_m, tmp3_m, tmp0_m, tmp1_m);       \
+    ST8x4_UB(tmp0_m, tmp1_m, pdst_m, stride);                          \
+}
+
 /* Description : Pack even byte elements, extract 0 & 2 index words from pair
                  of results and store 4 words in destination memory as per
                  stride