avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for HEVC uniw mc functions
authorShivraj Patil <shivraj.patil@imgtec.com>
Tue, 2 Jun 2015 08:38:09 +0000 (14:08 +0530)
committerMichael Niedermayer <michaelni@gmx.at>
Wed, 3 Jun 2015 11:46:47 +0000 (13:46 +0200)
This patch adds MSA (MIPS-SIMD-Arch) optimizations for HEVC uniw mc functions (qpel as well as epel) in new file hevc_mc_uniw_msa.c
Adds new generic macros (needed for this patch) in libavutil/mips/generic_macros_msa.h

Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
libavcodec/mips/Makefile
libavcodec/mips/hevc_mc_uniw_msa.c [new file with mode: 0644]
libavcodec/mips/hevcdsp_init_mips.c
libavcodec/mips/hevcdsp_mips.h
libavutil/mips/generic_macros_msa.h

index 5999fa7..9d8bf1d 100644 (file)
@@ -23,6 +23,7 @@ OBJS-$(CONFIG_H264DSP)                    += mips/h264dsp_init_mips.o
 OBJS-$(CONFIG_H264CHROMA)                 += mips/h264chroma_init_mips.o
 MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
                                              mips/hevc_mc_uni_msa.o        \
+                                             mips/hevc_mc_uniw_msa.o       \
                                              mips/hevc_mc_bi_msa.o
 MSA-OBJS-$(CONFIG_H264DSP)                += mips/h264dsp_msa.o
 LOONGSON3-OBJS-$(CONFIG_H264DSP)          += mips/h264dsp_mmi.o
diff --git a/libavcodec/mips/hevc_mc_uniw_msa.c b/libavcodec/mips/hevc_mc_uniw_msa.c
new file mode 100644 (file)
index 0000000..9079630
--- /dev/null
@@ -0,0 +1,4790 @@
+/*
+ * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@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 "libavutil/mips/generic_macros_msa.h"
+#include "libavcodec/mips/hevcdsp_mips.h"
+#include "libavcodec/mips/hevc_macros_msa.h"
+
+#define HEVC_HV_UNIW_RND_CLIP4(in0, in1, in2, in3, wgt, offset, rnd,       \
+                               out0, out1, out2, out3)                     \
+{                                                                          \
+    MUL4(in0, wgt, in1, wgt, in2, wgt, in3, wgt, out0, out1, out2, out3);  \
+    SRAR_W4_SW(out0, out1, out2, out3, rnd);                               \
+    ADD4(out0, offset, out1, offset, out2, offset, out3, offset,           \
+         out0, out1, out2, out3);                                          \
+    out0 = CLIP_SW_0_255(out0);                                            \
+    out1 = CLIP_SW_0_255(out1);                                            \
+    out2 = CLIP_SW_0_255(out2);                                            \
+    out3 = CLIP_SW_0_255(out3);                                            \
+}
+
+#define HEVC_UNIW_RND_CLIP2(in0, in1, wgt, offset, rnd,              \
+                            out0_r, out1_r, out0_l, out1_l)          \
+{                                                                    \
+    ILVR_H2_SW(in0, in0, in1, in1, out0_r, out1_r);                  \
+    ILVL_H2_SW(in0, in0, in1, in1, out0_l, out1_l);                  \
+    DOTP_SH4_SW(out0_r, out1_r, out0_l, out1_l, wgt, wgt, wgt, wgt,  \
+                out0_r, out1_r, out0_l, out1_l);                     \
+    SRAR_W4_SW(out0_r, out1_r, out0_l, out1_l, rnd);                 \
+    ADD4(out0_r, offset, out1_r, offset,                             \
+         out0_l, offset, out1_l, offset,                             \
+         out0_r, out1_r, out0_l, out1_l);                            \
+    out0_r = CLIP_SW_0_255(out0_r);                                  \
+    out1_r = CLIP_SW_0_255(out1_r);                                  \
+    out0_l = CLIP_SW_0_255(out0_l);                                  \
+    out1_l = CLIP_SW_0_255(out1_l);                                  \
+}
+
+#define HEVC_UNIW_RND_CLIP4(in0, in1, in2, in3, wgt, offset, rnd,  \
+                            out0_r, out1_r, out2_r, out3_r,        \
+                            out0_l, out1_l, out2_l, out3_l)        \
+{                                                                  \
+    HEVC_UNIW_RND_CLIP2(in0, in1, wgt, offset, rnd,                \
+                        out0_r, out1_r, out0_l, out1_l);           \
+    HEVC_UNIW_RND_CLIP2(in2, in3, wgt, offset, rnd,                \
+                        out2_r, out3_r, out2_l, out3_l);           \
+}
+
+static void hevc_uniwgt_copy_4w_msa(uint8_t *src,
+                                    int32_t src_stride,
+                                    uint8_t *dst,
+                                    int32_t dst_stride,
+                                    int32_t height,
+                                    int32_t weight,
+                                    int32_t offset,
+                                    int32_t rnd_val)
+{
+    v16i8 zero = { 0 };
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    if (2 == height) {
+        v16i8 src0, src1;
+        v8i16 dst0;
+        v4i32 dst0_r, dst0_l;
+
+        LD_SB2(src, src_stride, src0, src1);
+        src0 = (v16i8) __msa_ilvr_w((v4i32) src1, (v4i32) src0);
+        dst0 = (v8i16) __msa_ilvr_b(zero, src0);
+        dst0 <<= 6;
+
+        ILVRL_H2_SW(dst0, dst0, dst0_r, dst0_l);
+        DOTP_SH2_SW(dst0_r, dst0_l, weight_vec, weight_vec, dst0_r, dst0_l);
+        SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
+        ADD2(dst0_r, offset_vec, dst0_l, offset_vec, dst0_r, dst0_l);
+        dst0_r = CLIP_SW_0_255(dst0_r);
+        dst0_l = CLIP_SW_0_255(dst0_l);
+
+        HEVC_PCK_SW_SB2(dst0_l, dst0_r, dst0_r);
+        ST4x2_UB(dst0_r, dst, dst_stride);
+    } else if (4 == height) {
+        v16i8 src0, src1, src2, src3;
+        v8i16 dst0, dst1;
+        v4i32 dst0_r, dst1_r;
+        v4i32 dst0_l, dst1_l;
+
+        LD_SB4(src, src_stride, src0, src1, src2, src3);
+        ILVR_W2_SB(src1, src0, src3, src2, src0, src1);
+        ILVR_B2_SH(zero, src0, zero, src1, dst0, dst1);
+        dst0 <<= 6;
+        dst1 <<= 6;
+
+        HEVC_UNIW_RND_CLIP2(dst0, dst1, weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst0_l, dst1_l);
+
+        HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+        ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
+    } else if (0 == height % 8) {
+        uint32_t loop_cnt;
+        v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
+        v8i16 dst0, dst1, dst2, dst3;
+        v4i32 dst0_r, dst1_r, dst2_r, dst3_r;
+        v4i32 dst0_l, dst1_l, dst2_l, dst3_l;
+
+        for (loop_cnt = (height >> 3); loop_cnt--;) {
+            LD_SB8(src, src_stride,
+                   src0, src1, src2, src3, src4, src5, src6, src7);
+            src += (8 * src_stride);
+            ILVR_W4_SB(src1, src0, src3, src2, src5, src4, src7, src6,
+                       src0, src1, src2, src3);
+            ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3,
+                       dst0, dst1, dst2, dst3);
+
+            SLLI_4V(dst0, dst1, dst2, dst3, 6);
+            HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                                weight_vec, offset_vec, rnd_vec,
+                                dst0_r, dst1_r, dst2_r, dst3_r,
+                                dst0_l, dst1_l, dst2_l, dst3_l);
+
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                            dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+            ST4x8_UB(dst0_r, dst1_r, dst, dst_stride);
+            dst += (8 * dst_stride);
+        }
+    }
+}
+
+static void hevc_uniwgt_copy_6w_msa(uint8_t *src,
+                                    int32_t src_stride,
+                                    uint8_t *dst,
+                                    int32_t dst_stride,
+                                    int32_t height,
+                                    int32_t weight,
+                                    int32_t offset,
+                                    int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 zero = { 0 };
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    for (loop_cnt = (height >> 3); loop_cnt--;) {
+        LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+        src += (8 * src_stride);
+        ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3,
+                   dst0, dst1, dst2, dst3);
+        ILVR_B4_SH(zero, src4, zero, src5, zero, src6, zero, src7,
+                   dst4, dst5, dst6, dst7);
+
+        SLLI_4V(dst0, dst1, dst2, dst3, 6);
+        SLLI_4V(dst4, dst5, dst6, dst7, 6);
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+
+        HEVC_UNIW_RND_CLIP4(dst4, dst5, dst6, dst7,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_uniwgt_copy_8w_msa(uint8_t *src,
+                                    int32_t src_stride,
+                                    uint8_t *dst,
+                                    int32_t dst_stride,
+                                    int32_t height,
+                                    int32_t weight,
+                                    int32_t offset,
+                                    int32_t rnd_val)
+{
+    v16i8 zero = { 0 };
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    if (2 == height) {
+        v16i8 src0, src1;
+        v8i16 dst0, dst1;
+        v4i32 dst0_r, dst1_r, dst0_l, dst1_l;
+
+        LD_SB2(src, src_stride, src0, src1);
+        ILVR_B2_SH(zero, src0, zero, src1, dst0, dst1);
+
+        dst0 <<= 6;
+        dst1 <<= 6;
+        HEVC_UNIW_RND_CLIP2(dst0, dst1, weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst0_l, dst1_l);
+
+        HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+        ST8x2_UB(dst0_r, dst, dst_stride);
+    } else if (6 == height) {
+        v16i8 src0, src1, src2, src3, src4, src5;
+        v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+        v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+        v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+
+        LD_SB6(src, src_stride, src0, src1, src2, src3, src4, src5);
+        ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3,
+                   dst0, dst1, dst2, dst3);
+        ILVR_B2_SH(zero, src4, zero, src5, dst4, dst5);
+
+        SLLI_4V(dst0, dst1, dst2, dst3, 6);
+        dst4 <<= 6;
+        dst5 <<= 6;
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(dst4, dst5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                         dst2_l, dst2_r, dst3_l, dst3_r,
+                         dst4_l, dst4_r, dst5_l, dst5_r,
+                         dst0_r, dst1_r, dst2_r);
+        ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+        ST8x2_UB(dst2_r, dst, dst_stride);
+    } else if (0 == height % 4) {
+        uint32_t loop_cnt;
+        v16i8 src0, src1, src2, src3;
+        v8i16 dst0, dst1, dst2, dst3;
+        v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+
+        for (loop_cnt = (height >> 2); loop_cnt--;) {
+            LD_SB4(src, src_stride, src0, src1, src2, src3);
+            src += (4 * src_stride);
+            ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3,
+                       dst0, dst1, dst2, dst3);
+
+            SLLI_4V(dst0, dst1, dst2, dst3, 6);
+            HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                                weight_vec, offset_vec, rnd_vec,
+                                dst0_r, dst1_r, dst2_r, dst3_r,
+                                dst0_l, dst1_l, dst2_l, dst3_l);
+
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                            dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+            ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+            dst += (4 * dst_stride);
+        }
+    }
+}
+
+static void hevc_uniwgt_copy_12w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+    v16i8 zero = { 0 };
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src0, src1, src2, src3);
+        src += (4 * src_stride);
+        ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3,
+                   dst0, dst1, dst2, dst3);
+
+        SLLI_4V(dst0, dst1, dst2, dst3, 6);
+        ILVL_W2_SB(src1, src0, src3, src2, src0, src1);
+        ILVR_B2_SH(zero, src0, zero, src1, dst4, dst5);
+        dst4 <<= 6;
+        dst5 <<= 6;
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(dst4, dst5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                         dst2_l, dst2_r, dst3_l, dst3_r,
+                         dst4_l, dst4_r, dst5_l, dst5_r,
+                         dst0_r, dst1_r, dst2_r);
+        ST12x4_UB(dst0_r, dst1_r, dst2_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_uniwgt_copy_16multx4mult_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val,
+                                              int32_t width)
+{
+    uint32_t loop_cnt, cnt;
+    uint8_t *src_tmp;
+    uint8_t *dst_tmp;
+    v16i8 src0, src1, src2, src3;
+    v8i16 tmp0, tmp1, tmp2, tmp3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v16i8 zero = { 0 };
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    for (cnt = width >> 4; cnt--;) {
+        src_tmp = src;
+        dst_tmp = dst;
+
+        for (loop_cnt = height >> 2; loop_cnt--;) {
+            LD_SB4(src_tmp, src_stride, src0, src1, src2, src3);
+            src_tmp += (4 * src_stride);
+            ILVR_B2_SH(zero, src0, zero, src1, tmp0, tmp1);
+            ILVL_B2_SH(zero, src0, zero, src1, tmp2, tmp3);
+
+            SLLI_4V(tmp0, tmp1, tmp2, tmp3, 6);
+            HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                                weight_vec, offset_vec, rnd_vec,
+                                dst0_r, dst1_r, dst2_r, dst3_r,
+                                dst0_l, dst1_l, dst2_l, dst3_l);
+
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                            dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+
+            ST_SW2(dst0_r, dst1_r, dst_tmp, dst_stride);
+            dst_tmp += (2 * dst_stride);
+
+            ILVR_B2_SH(zero, src2, zero, src3, tmp0, tmp1);
+            ILVL_B2_SH(zero, src2, zero, src3, tmp2, tmp3);
+
+            SLLI_4V(tmp0, tmp1, tmp2, tmp3, 6);
+            HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                                weight_vec, offset_vec, rnd_vec,
+                                dst0_r, dst1_r, dst2_r, dst3_r,
+                                dst0_l, dst1_l, dst2_l, dst3_l);
+
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                            dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+
+            ST_SW2(dst0_r, dst1_r, dst_tmp, dst_stride);
+            dst_tmp += (2 * dst_stride);
+        }
+
+        src += 16;
+        dst += 16;
+    }
+}
+
+static void hevc_uniwgt_copy_16w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    hevc_uniwgt_copy_16multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      height, weight, offset, rnd_val, 16);
+}
+
+static void hevc_uniwgt_copy_24w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    hevc_uniwgt_copy_16multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      height, weight, offset, rnd_val, 16);
+
+    hevc_uniwgt_copy_8w_msa(src + 16, src_stride, dst + 16, dst_stride,
+                            height, weight, offset, rnd_val);
+}
+
+static void hevc_uniwgt_copy_32w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    hevc_uniwgt_copy_16multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      height, weight, offset, rnd_val, 32);
+}
+
+static void hevc_uniwgt_copy_48w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    hevc_uniwgt_copy_16multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      height, weight, offset, rnd_val, 48);
+}
+
+static void hevc_uniwgt_copy_64w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    hevc_uniwgt_copy_16multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      height, weight, offset, rnd_val, 64);
+}
+
+static void hevc_hz_uniwgt_8t_4w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
+    v16i8 mask1, mask2, mask3;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+
+    src -= 3;
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+
+    for (loop_cnt = (height >> 3); loop_cnt--;) {
+        LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+        src += (8 * src_stride);
+        XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
+
+        VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        VSHF_B4_SB(src2, src3, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        VSHF_B4_SB(src4, src5, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        VSHF_B4_SB(src6, src7, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST4x8_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (8 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_8t_8w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 mask1, mask2, mask3;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= 3;
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src0, src1, src2, src3);
+        src += (4 * src_stride);
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_8t_12w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hz_uniwgt_8t_8w_msa(src, src_stride, dst, dst_stride,
+                             filter, height, weight, offset, rnd_val);
+    hevc_hz_uniwgt_8t_4w_msa(src + 8, src_stride, dst + 8, dst_stride,
+                             filter, height, weight, offset, rnd_val);
+}
+
+static void hevc_hz_uniwgt_8t_16w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 mask1, mask2, mask3;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= 3;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+
+    for (loop_cnt = (height >> 1); loop_cnt--;) {
+        LD_SB2(src, src_stride, src0, src2);
+        LD_SB2(src + 8, src_stride, src1, src3);
+        src += (2 * src_stride);
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_8t_24w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= 3;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+    mask4 = mask0 + 8;
+    mask5 = mask0 + 10;
+    mask6 = mask0 + 12;
+    mask7 = mask0 + 14;
+
+    for (loop_cnt = (height >> 1); loop_cnt--;) {
+        LD_SB2(src, 16, src0, src1);
+        src += src_stride;
+        LD_SB2(src, 16, src2, src3);
+        src += src_stride;
+        XORI_B4_128_SB(src0, src1, src2, src3);
+        VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
+                   vec0, vec1, vec2, vec3);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+        VSHF_B4_SB(src2, src3, mask4, mask5, mask6, mask7,
+                   vec0, vec1, vec2, vec3);
+        dst4 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst4, dst4, dst4, dst4);
+        VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst5 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst5, dst5, dst5, dst5);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(dst4, dst5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst3_l, dst3_r, dst4_l, dst4_r, dst0_r, dst1_r);
+        HEVC_PCK_SW_SB4(dst2_l, dst2_r, dst5_l, dst5_r, dst2_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        ST8x2_UB(dst2_r, dst + 16, dst_stride);
+        dst += (2 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_8t_32w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= 3;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+    mask4 = mask0 + 8;
+    mask5 = mask0 + 10;
+    mask6 = mask0 + 12;
+    mask7 = mask0 + 14;
+
+    for (loop_cnt = height; loop_cnt--;) {
+        LD_SB2(src, 16, src0, src1);
+        src2 = LD_SB(src + 24);
+        src += src_stride;
+        XORI_B3_128_SB(src0, src1, src2);
+
+        VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
+                   vec0, vec1, vec2, vec3);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, 16);
+        dst += dst_stride;
+    }
+}
+
+static void hevc_hz_uniwgt_8t_48w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= 3;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+    mask4 = mask0 + 8;
+    mask5 = mask0 + 10;
+    mask6 = mask0 + 12;
+    mask7 = mask0 + 14;
+
+    for (loop_cnt = height; loop_cnt--;) {
+        LD_SB3(src, 16, src0, src1, src2);
+        src3 = LD_SB(src + 40);
+        src += src_stride;
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
+                   vec0, vec1, vec2, vec3);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        VSHF_B4_SB(src1, src2, mask4, mask5, mask6, mask7,
+                   vec0, vec1, vec2, vec3);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+        VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst4 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst4, dst4, dst4, dst4);
+        VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst5 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst5, dst5, dst5, dst5);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_UNIW_RND_CLIP2(dst4, dst5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                         dst2_l, dst2_r, dst3_l, dst3_r,
+                         dst4_l, dst4_r, dst5_l, dst5_r,
+                         dst0_r, dst1_r, dst2_r);
+        ST_SW2(dst0_r, dst1_r, dst, 16);
+        ST_SW(dst2_r, dst + 32);
+        dst += dst_stride;
+    }
+}
+
+static void hevc_hz_uniwgt_8t_64w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint8_t *src_tmp;
+    uint8_t *dst_tmp;
+    uint32_t loop_cnt, cnt;
+    v16i8 src0, src1, src2;
+    v8i16 filt0, filt1, filt2, filt3;
+    v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7;
+    v16i8 vec0, vec1, vec2, vec3;
+    v8i16 dst0, dst1, dst2, dst3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= 3;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+    mask4 = mask0 + 8;
+    mask5 = mask0 + 10;
+    mask6 = mask0 + 12;
+    mask7 = mask0 + 14;
+
+    for (loop_cnt = height; loop_cnt--;) {
+        src_tmp = src;
+        dst_tmp = dst;
+
+        for (cnt = 2; cnt--;) {
+            LD_SB2(src_tmp, 16, src0, src1);
+            src2 = LD_SB(src_tmp + 24);
+            src_tmp += 32;
+            XORI_B3_128_SB(src0, src1, src2);
+
+            VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                       vec0, vec1, vec2, vec3);
+            dst0 = const_vec;
+            DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                         dst0, dst0, dst0, dst0);
+            VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
+                       vec0, vec1, vec2, vec3);
+            dst1 = const_vec;
+            DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                         dst1, dst1, dst1, dst1);
+            VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                       vec0, vec1, vec2, vec3);
+            dst2 = const_vec;
+            DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                         dst2, dst2, dst2, dst2);
+            VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                       vec0, vec1, vec2, vec3);
+            dst3 = const_vec;
+            DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                         dst3, dst3, dst3, dst3);
+
+            HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                                weight_vec, offset_vec, rnd_vec,
+                                dst0_r, dst1_r, dst2_r, dst3_r,
+                                dst0_l, dst1_l, dst2_l, dst3_l);
+
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                            dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+            ST_SW2(dst0_r, dst1_r, dst_tmp, 16);
+            dst_tmp += 32;
+        }
+
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+static void hevc_vt_uniwgt_8t_4w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v16i8 src9, src10, src11, src12, src13, src14;
+    v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
+    v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
+    v16i8 src1110_r, src1211_r, src1312_r, src1413_r;
+    v16i8 src2110, src4332, src6554, src8776, src10998;
+    v16i8 src12111110, src14131312;
+    v8i16 dst10, dst32, dst54, dst76;
+    v8i16 filt0, filt1, filt2, filt3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (3 * src_stride);
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+    src += (7 * src_stride);
+
+    ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
+               src10_r, src32_r, src54_r, src21_r);
+
+    ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+
+    ILVR_D3_SB(src21_r, src10_r, src43_r,
+               src32_r, src65_r, src54_r, src2110, src4332, src6554);
+
+    XORI_B3_128_SB(src2110, src4332, src6554);
+
+    for (loop_cnt = (height >> 3); loop_cnt--;) {
+        LD_SB8(src, src_stride,
+               src7, src8, src9, src10, src11, src12, src13, src14);
+        src += (8 * src_stride);
+        ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
+                   src76_r, src87_r, src98_r, src109_r);
+        ILVR_B4_SB(src11, src10, src12, src11, src13, src12, src14, src13,
+                   src1110_r, src1211_r, src1312_r, src1413_r);
+        ILVR_D4_SB(src87_r, src76_r, src109_r, src98_r, src1211_r, src1110_r,
+                   src1413_r, src1312_r,
+                   src8776, src10998, src12111110, src14131312);
+        XORI_B4_128_SB(src8776, src10998, src12111110, src14131312);
+
+        dst10 = const_vec;
+        DPADD_SB4_SH(src2110, src4332, src6554, src8776, filt0, filt1,
+                     filt2, filt3, dst10, dst10, dst10, dst10);
+        dst32 = const_vec;
+        DPADD_SB4_SH(src4332, src6554, src8776, src10998,
+                     filt0, filt1, filt2, filt3, dst32, dst32, dst32, dst32);
+        dst54 = const_vec;
+        DPADD_SB4_SH(src6554, src8776, src10998, src12111110,
+                     filt0, filt1, filt2, filt3, dst54, dst54, dst54, dst54);
+        dst76 = const_vec;
+        DPADD_SB4_SH(src8776, src10998, src12111110, src14131312,
+                     filt0, filt1, filt2, filt3, dst76, dst76, dst76, dst76);
+
+        HEVC_UNIW_RND_CLIP4(dst10, dst32, dst54, dst76,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST4x8_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (8 * dst_stride);
+
+        src2110 = src10998;
+        src4332 = src12111110;
+        src6554 = src14131312;
+        src6 = src14;
+    }
+}
+
+static void hevc_vt_uniwgt_8t_8w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+    v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
+    v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3;
+    v8i16 filt0, filt1, filt2, filt3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (3 * src_stride);
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+    src += (7 * src_stride);
+    XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
+    ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
+               src10_r, src32_r, src54_r, src21_r);
+    ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src7, src8, src9, src10);
+        src += (4 * src_stride);
+        XORI_B4_128_SB(src7, src8, src9, src10);
+        ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
+                   src76_r, src87_r, src98_r, src109_r);
+
+        tmp0 = const_vec;
+        DPADD_SB4_SH(src10_r, src32_r, src54_r, src76_r,
+                     filt0, filt1, filt2, filt3, tmp0, tmp0, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB4_SH(src21_r, src43_r, src65_r, src87_r,
+                     filt0, filt1, filt2, filt3, tmp1, tmp1, tmp1, tmp1);
+        tmp2 = const_vec;
+        DPADD_SB4_SH(src32_r, src54_r, src76_r, src98_r,
+                     filt0, filt1, filt2, filt3, tmp2, tmp2, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB4_SH(src43_r, src65_r, src87_r, src109_r,
+                     filt0, filt1, filt2, filt3, tmp3, tmp3, tmp3, tmp3);
+
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+
+        src10_r = src54_r;
+        src32_r = src76_r;
+        src54_r = src98_r;
+        src21_r = src65_r;
+        src43_r = src87_r;
+        src65_r = src109_r;
+        src6 = src10;
+    }
+}
+
+static void hevc_vt_uniwgt_8t_12w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+    v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
+    v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+    v16i8 src10_l, src32_l, src54_l, src76_l, src98_l;
+    v16i8 src21_l, src43_l, src65_l, src87_l, src109_l;
+    v16i8 src2110, src4332, src6554, src8776, src10998;
+    v8i16 filt0, filt1, filt2, filt3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (3 * src_stride);
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+    src += (7 * src_stride);
+    XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
+    ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
+               src10_r, src32_r, src54_r, src21_r);
+    ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+    ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
+               src10_l, src32_l, src54_l, src21_l);
+    ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
+    ILVR_D3_SB(src21_l, src10_l, src43_l, src32_l, src65_l, src54_l,
+               src2110, src4332, src6554);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src7, src8, src9, src10);
+        src += (4 * src_stride);
+        XORI_B4_128_SB(src7, src8, src9, src10);
+
+        ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
+                   src76_r, src87_r, src98_r, src109_r);
+        ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
+                   src76_l, src87_l, src98_l, src109_l);
+        ILVR_D2_SB(src87_l, src76_l, src109_l, src98_l, src8776, src10998);
+
+        tmp0 = const_vec;
+        DPADD_SB4_SH(src10_r, src32_r, src54_r, src76_r,
+                     filt0, filt1, filt2, filt3, tmp0, tmp0, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB4_SH(src21_r, src43_r, src65_r, src87_r,
+                     filt0, filt1, filt2, filt3, tmp1, tmp1, tmp1, tmp1);
+        tmp2 = const_vec;
+        DPADD_SB4_SH(src32_r, src54_r, src76_r, src98_r,
+                     filt0, filt1, filt2, filt3, tmp2, tmp2, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB4_SH(src43_r, src65_r, src87_r, src109_r,
+                     filt0, filt1, filt2, filt3, tmp3, tmp3, tmp3, tmp3);
+        tmp4 = const_vec;
+        DPADD_SB4_SH(src2110, src4332, src6554, src8776,
+                     filt0, filt1, filt2, filt3, tmp4, tmp4, tmp4, tmp4);
+        tmp5 = const_vec;
+        DPADD_SB4_SH(src4332, src6554, src8776, src10998,
+                     filt0, filt1, filt2, filt3, tmp5, tmp5, tmp5, tmp5);
+
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(tmp4, tmp5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                         dst2_l, dst2_r, dst3_l, dst3_r,
+                         dst4_l, dst4_r, dst5_l, dst5_r,
+                         dst0_r, dst1_r, dst2_r);
+        ST12x4_UB(dst0_r, dst1_r, dst2_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+
+        src10_r = src54_r;
+        src32_r = src76_r;
+        src54_r = src98_r;
+        src21_r = src65_r;
+        src43_r = src87_r;
+        src65_r = src109_r;
+        src2110 = src6554;
+        src4332 = src8776;
+        src6554 = src10998;
+        src6 = src10;
+    }
+}
+
+static void hevc_vt_uniwgt_8t_16multx2mult_msa(uint8_t *src,
+                                               int32_t src_stride,
+                                               uint8_t *dst,
+                                               int32_t dst_stride,
+                                               const int8_t *filter,
+                                               int32_t height,
+                                               int32_t weight,
+                                               int32_t offset,
+                                               int32_t rnd_val,
+                                               int32_t width)
+{
+    uint8_t *src_tmp;
+    uint8_t *dst_tmp;
+    int32_t loop_cnt, cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v16i8 src10_r, src32_r, src54_r, src76_r;
+    v16i8 src21_r, src43_r, src65_r, src87_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3;
+    v16i8 src10_l, src32_l, src54_l, src76_l;
+    v16i8 src21_l, src43_l, src65_l, src87_l;
+    v8i16 filt0, filt1, filt2, filt3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (3 * src_stride);
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight = weight & 0x0000FFFF;
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    for (cnt = (width >> 4); cnt--;) {
+        src_tmp = src;
+        dst_tmp = dst;
+
+        LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
+        src_tmp += (7 * src_stride);
+        XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+        ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
+                   src10_r, src32_r, src54_r, src21_r);
+        ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+        ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
+                   src10_l, src32_l, src54_l, src21_l);
+        ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
+
+        for (loop_cnt = (height >> 1); loop_cnt--;) {
+            LD_SB2(src_tmp, src_stride, src7, src8);
+            src_tmp += (2 * src_stride);
+            XORI_B2_128_SB(src7, src8);
+            ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
+            ILVL_B2_SB(src7, src6, src8, src7, src76_l, src87_l);
+
+            tmp0 = const_vec;
+            DPADD_SB4_SH(src10_r, src32_r, src54_r, src76_r,
+                         filt0, filt1, filt2, filt3, tmp0, tmp0, tmp0, tmp0);
+            tmp1 = const_vec;
+            DPADD_SB4_SH(src21_r, src43_r, src65_r, src87_r,
+                         filt0, filt1, filt2, filt3, tmp1, tmp1, tmp1, tmp1);
+            tmp2 = const_vec;
+            DPADD_SB4_SH(src10_l, src32_l, src54_l, src76_l,
+                         filt0, filt1, filt2, filt3, tmp2, tmp2, tmp2, tmp2);
+            tmp3 = const_vec;
+            DPADD_SB4_SH(src21_l, src43_l, src65_l, src87_l,
+                         filt0, filt1, filt2, filt3, tmp3, tmp3, tmp3, tmp3);
+
+            HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                                weight_vec, offset_vec, rnd_vec,
+                                dst0_r, dst1_r, dst2_r, dst3_r,
+                                dst0_l, dst1_l, dst2_l, dst3_l);
+
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                            dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+            ST_SW2(dst0_r, dst1_r, dst_tmp, dst_stride);
+            dst_tmp += (2 * dst_stride);
+
+            src10_r = src32_r;
+            src32_r = src54_r;
+            src54_r = src76_r;
+            src21_r = src43_r;
+            src43_r = src65_r;
+            src65_r = src87_r;
+            src10_l = src32_l;
+            src32_l = src54_l;
+            src54_l = src76_l;
+            src21_l = src43_l;
+            src43_l = src65_l;
+            src65_l = src87_l;
+            src6 = src8;
+        }
+
+        src += 16;
+        dst += 16;
+    }
+}
+
+static void hevc_vt_uniwgt_8t_16w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_vt_uniwgt_8t_16multx2mult_msa(src, src_stride, dst, dst_stride,
+                                       filter, height, weight,
+                                       offset, rnd_val, 16);
+}
+
+static void hevc_vt_uniwgt_8t_24w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_vt_uniwgt_8t_16multx2mult_msa(src, src_stride, dst, dst_stride,
+                                       filter, height, weight,
+                                       offset, rnd_val, 16);
+
+    hevc_vt_uniwgt_8t_8w_msa(src + 16, src_stride, dst + 16, dst_stride,
+                             filter, height, weight, offset, rnd_val);
+}
+
+static void hevc_vt_uniwgt_8t_32w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_vt_uniwgt_8t_16multx2mult_msa(src, src_stride, dst, dst_stride,
+                                       filter, height, weight,
+                                       offset, rnd_val, 32);
+}
+
+static void hevc_vt_uniwgt_8t_48w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_vt_uniwgt_8t_16multx2mult_msa(src, src_stride, dst, dst_stride,
+                                       filter, height, weight,
+                                       offset, rnd_val, 48);
+}
+
+static void hevc_vt_uniwgt_8t_64w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_vt_uniwgt_8t_16multx2mult_msa(src, src_stride, dst, dst_stride,
+                                       filter, height, weight,
+                                       offset, rnd_val, 64);
+}
+
+static void hevc_hv_uniwgt_8t_4w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter_x,
+                                     const int8_t *filter_y,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v8i16 filt0, filt1, filt2, filt3;
+    v4i32 filt_h0, filt_h1, filt_h2, filt_h3;
+    v16i8 mask1, mask2, mask3;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+    v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
+    v8i16 dst30, dst41, dst52, dst63, dst66, dst87;
+    v4i32 dst0_r, dst1_r, weight_vec, offset_vec, rnd_vec;
+    v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
+    v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+    v8u16 mask4 = { 0, 4, 1, 5, 2, 6, 3, 7 };
+
+    src -= ((3 * src_stride) + 3);
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W4_SW(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+    src += (7 * src_stride);
+    XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
+    /* row 0 row 1 row 2 row 3 */
+    VSHF_B4_SB(src0, src3, mask0, mask1, mask2, mask3, vec0, vec1, vec2, vec3);
+    VSHF_B4_SB(src1, src4, mask0, mask1, mask2, mask3, vec4, vec5, vec6, vec7);
+    VSHF_B4_SB(src2, src5, mask0, mask1, mask2, mask3,
+               vec8, vec9, vec10, vec11);
+    VSHF_B4_SB(src3, src6, mask0, mask1, mask2, mask3,
+               vec12, vec13, vec14, vec15);
+    dst30 = const_vec;
+    DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                 dst30, dst30, dst30, dst30);
+    dst41 = const_vec;
+    DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2, filt3,
+                 dst41, dst41, dst41, dst41);
+    dst52 = const_vec;
+    DPADD_SB4_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2, filt3,
+                 dst52, dst52, dst52, dst52);
+    dst63 = const_vec;
+    DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt0, filt1, filt2, filt3,
+                 dst63, dst63, dst63, dst63);
+
+    ILVR_H3_SH(dst41, dst30, dst52, dst41, dst63, dst52,
+               dst10_r, dst21_r, dst32_r);
+
+    dst43_r = __msa_ilvl_h(dst41, dst30);
+    dst54_r = __msa_ilvl_h(dst52, dst41);
+    dst65_r = __msa_ilvl_h(dst63, dst52);
+
+    dst66 = (v8i16) __msa_splati_d((v2i64) dst63, 1);
+
+    for (loop_cnt = height >> 1; loop_cnt--;) {
+        LD_SB2(src, src_stride, src7, src8);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src7, src8);
+
+        VSHF_B4_SB(src7, src8, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        dst87 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst87, dst87, dst87, dst87);
+        dst76_r = __msa_ilvr_h(dst87, dst66);
+        dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r,
+                                filt_h0, filt_h1, filt_h2, filt_h3);
+        dst87_r = __msa_vshf_h((v8i16) mask4, dst87, dst87);
+        dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r,
+                                filt_h0, filt_h1, filt_h2, filt_h3);
+
+        dst0_r >>= 6;
+        dst1_r >>= 6;
+        MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
+        SRAR_W2_SW(dst0_r, dst1_r, rnd_vec);
+        ADD2(dst0_r, offset_vec, dst1_r, offset_vec, dst0_r, dst1_r);
+        dst0_r = CLIP_SW_0_255(dst0_r);
+        dst1_r = CLIP_SW_0_255(dst1_r);
+
+        HEVC_PCK_SW_SB2(dst1_r, dst0_r, dst0_r);
+        ST4x2_UB(dst0_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+
+        dst10_r = dst32_r;
+        dst32_r = dst54_r;
+        dst54_r = dst76_r;
+        dst21_r = dst43_r;
+        dst43_r = dst65_r;
+        dst65_r = dst87_r;
+        dst66 = (v8i16) __msa_splati_d((v2i64) dst87, 1);
+    }
+}
+
+static void hevc_hv_uniwgt_8t_8multx2mult_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter_x,
+                                              const int8_t *filter_y,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val,
+                                              int32_t width)
+{
+    uint32_t loop_cnt, cnt;
+    uint8_t *src_tmp;
+    uint8_t *dst_tmp;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v8i16 filt0, filt1, filt2, filt3;
+    v4i32 filt_h0, filt_h1, filt_h2, filt_h3;
+    v16i8 mask1, mask2, mask3;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+    v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
+    v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
+    v8i16 dst10_l, dst32_l, dst54_l, dst76_l;
+    v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
+    v8i16 dst21_l, dst43_l, dst65_l, dst87_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+
+    src -= ((3 * src_stride) + 3);
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+    SPLATI_W4_SW(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 4;
+    mask3 = mask0 + 6;
+
+    for (cnt = width >> 3; cnt--;) {
+        src_tmp = src;
+        dst_tmp = dst;
+
+        LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
+        src_tmp += (7 * src_stride);
+        XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
+        VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
+                   vec4, vec5, vec6, vec7);
+        VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
+                   vec8, vec9, vec10, vec11);
+        VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
+                   vec12, vec13, vec14, vec15);
+        dst0 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst0, dst0, dst0, dst0);
+        dst1 = const_vec;
+        DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2, filt3,
+                     dst1, dst1, dst1, dst1);
+        dst2 = const_vec;
+        DPADD_SB4_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2, filt3,
+                     dst2, dst2, dst2, dst2);
+        dst3 = const_vec;
+        DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt0, filt1, filt2, filt3,
+                     dst3, dst3, dst3, dst3);
+
+        VSHF_B4_SB(src4, src4, mask0, mask1, mask2, mask3,
+                   vec0, vec1, vec2, vec3);
+        VSHF_B4_SB(src5, src5, mask0, mask1, mask2, mask3,
+                   vec4, vec5, vec6, vec7);
+        VSHF_B4_SB(src6, src6, mask0, mask1, mask2, mask3,
+                   vec8, vec9, vec10, vec11);
+        dst4 = const_vec;
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                     dst4, dst4, dst4, dst4);
+        dst5 = const_vec;
+        DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2, filt3,
+                     dst5, dst5, dst5, dst5);
+        dst6 = const_vec;
+        DPADD_SB4_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2, filt3,
+                     dst6, dst6, dst6, dst6);
+
+        ILVR_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst2, dst1,
+                   dst10_r, dst32_r, dst54_r, dst21_r);
+        ILVR_H2_SH(dst4, dst3, dst6, dst5, dst43_r, dst65_r);
+        ILVL_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst2, dst1,
+                   dst10_l, dst32_l, dst54_l, dst21_l);
+        ILVL_H2_SH(dst4, dst3, dst6, dst5, dst43_l, dst65_l);
+
+        for (loop_cnt = height >> 1; loop_cnt--;) {
+            LD_SB2(src_tmp, src_stride, src7, src8);
+            src_tmp += 2 * src_stride;
+            XORI_B2_128_SB(src7, src8);
+
+            VSHF_B4_SB(src7, src7, mask0, mask1, mask2, mask3,
+                       vec0, vec1, vec2, vec3);
+            dst7 = const_vec;
+            DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                         dst7, dst7, dst7, dst7);
+
+            ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
+            dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r,
+                                    filt_h0, filt_h1, filt_h2, filt_h3);
+            dst0_l = HEVC_FILT_8TAP(dst10_l, dst32_l, dst54_l, dst76_l,
+                                    filt_h0, filt_h1, filt_h2, filt_h3);
+            dst0_r >>= 6;
+            dst0_l >>= 6;
+
+            /* row 8 */
+            VSHF_B4_SB(src8, src8, mask0, mask1, mask2, mask3,
+                       vec0, vec1, vec2, vec3);
+            dst8 = const_vec;
+            DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3,
+                         dst8, dst8, dst8, dst8);
+
+            ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l);
+            dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r,
+                                    filt_h0, filt_h1, filt_h2, filt_h3);
+            dst1_l = HEVC_FILT_8TAP(dst21_l, dst43_l, dst65_l, dst87_l,
+                                    filt_h0, filt_h1, filt_h2, filt_h3);
+            dst1_r >>= 6;
+            dst1_l >>= 6;
+
+            HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst0_l, dst1_l,
+                                   weight_vec, offset_vec, rnd_vec,
+                                   dst0_r, dst1_r, dst0_l, dst1_l);
+
+            HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+            ST8x2_UB(dst0_r, dst_tmp, dst_stride);
+            dst_tmp += (2 * dst_stride);
+
+            dst10_r = dst32_r;
+            dst32_r = dst54_r;
+            dst54_r = dst76_r;
+            dst10_l = dst32_l;
+            dst32_l = dst54_l;
+            dst54_l = dst76_l;
+            dst21_r = dst43_r;
+            dst43_r = dst65_r;
+            dst65_r = dst87_r;
+            dst21_l = dst43_l;
+            dst43_l = dst65_l;
+            dst65_l = dst87_l;
+            dst6 = dst8;
+        }
+
+        src += 8;
+        dst += 8;
+    }
+}
+
+static void hevc_hv_uniwgt_8t_8w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter_x,
+                                     const int8_t *filter_y,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 8);
+}
+
+static void hevc_hv_uniwgt_8t_12w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 8);
+    hevc_hv_uniwgt_8t_4w_msa(src + 8, src_stride, dst + 8, dst_stride,
+                             filter_x, filter_y, height, weight, offset,
+                             rnd_val);
+}
+
+static void hevc_hv_uniwgt_8t_16w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 16);
+}
+
+static void hevc_hv_uniwgt_8t_24w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 24);
+}
+
+static void hevc_hv_uniwgt_8t_32w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 32);
+}
+
+static void hevc_hv_uniwgt_8t_48w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 48);
+}
+
+static void hevc_hv_uniwgt_8t_64w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 64);
+}
+
+static void hevc_hz_uniwgt_4t_4x2_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, vec0, vec1;
+    v16i8 mask1;
+    v8i16 dst0;
+    v4i32 dst0_r, dst0_l;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    mask1 = mask0 + 2;
+
+    weight = weight & 0x0000FFFF;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB2(src, src_stride, src0, src1);
+    XORI_B2_128_SB(src0, src1);
+
+    VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+    ILVRL_H2_SW(dst0, dst0, dst0_r, dst0_l);
+    DOTP_SH2_SW(dst0_r, dst0_l, weight_vec, weight_vec, dst0_r, dst0_l);
+    SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
+    ADD2(dst0_r, offset_vec, dst0_l, offset_vec, dst0_r, dst0_l);
+    dst0_r = CLIP_SW_0_255(dst0_r);
+    dst0_l = CLIP_SW_0_255(dst0_l);
+
+    HEVC_PCK_SW_SB2(dst0_l, dst0_r, dst0_r);
+    ST4x2_UB(dst0_r, dst, dst_stride);
+    dst += (4 * dst_stride);
+}
+
+static void hevc_hz_uniwgt_4t_4x4_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, src2, src3;
+    v16i8 mask1, vec0, vec1;
+    v8i16 dst0, dst1;
+    v4i32 dst0_r, dst1_r, dst0_l, dst1_l;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+
+    src -= 1;
+
+    /* rearranging filter */
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    mask1 = mask0 + 2;
+
+    weight = weight & 0x0000FFFF;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB4(src, src_stride, src0, src1, src2, src3);
+    XORI_B4_128_SB(src0, src1, src2, src3);
+
+    VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+    VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec0, vec1);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+    HEVC_UNIW_RND_CLIP2(dst0, dst1, weight_vec, offset_vec, rnd_vec,
+                        dst0_r, dst1_r, dst0_l, dst1_l);
+
+    HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+    ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
+    dst += (4 * dst_stride);
+}
+
+static void hevc_hz_uniwgt_4t_4x8multiple_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
+    v16i8 mask1, vec0, vec1;
+    v8i16 dst0, dst1, dst2, dst3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+
+    for (loop_cnt = (height >> 3); loop_cnt--;) {
+        LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+        src += (8 * src_stride);
+
+        XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
+
+        VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src4, src5, src4, src5, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src6, src7, src6, src7, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST4x8_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (8 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_4w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    if (2 == height) {
+        hevc_hz_uniwgt_4t_4x2_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else if (4 == height) {
+        hevc_hz_uniwgt_4t_4x4_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else if (8 == height || 16 == height) {
+        hevc_hz_uniwgt_4t_4x8multiple_msa(src, src_stride, dst, dst_stride,
+                                          filter, height, weight,
+                                          offset, rnd_val);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_6w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, src2, src3;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v16i8 vec0, vec1;
+    v8i16 dst0, dst1, dst2, dst3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src0, src1, src2, src3);
+        src += (4 * src_stride);
+
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+
+        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_8x2_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v8i16 filt0, filt1, dst0, dst1;
+    v16i8 src0, src1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v16i8 vec0, vec1;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst0_l, dst1_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+
+    LD_SB2(src, src_stride, src0, src1);
+    XORI_B2_128_SB(src0, src1);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+    HEVC_UNIW_RND_CLIP2(dst0, dst1, weight_vec, offset_vec, rnd_vec,
+                        dst0_r, dst1_r, dst0_l, dst1_l);
+
+    HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+    ST8x2_UB(dst0_r, dst, dst_stride);
+}
+
+static void hevc_hz_uniwgt_4t_8x6_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, src2, src3, src4, src5;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v16i8 vec0, vec1;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+
+    LD_SB6(src, src_stride, src0, src1, src2, src3, src4, src5);
+    LD_SB6(src, src_stride, src0, src1, src2, src3, src4, src5);
+    XORI_B6_128_SB(src0, src1, src2, src3, src4, src5);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+    dst3 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+    dst4 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+
+    VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+    dst5 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+
+    HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                        weight_vec, offset_vec, rnd_vec,
+                        dst0_r, dst1_r, dst2_r, dst3_r,
+                        dst0_l, dst1_l, dst2_l, dst3_l);
+
+    HEVC_UNIW_RND_CLIP2(dst4, dst5, weight_vec, offset_vec, rnd_vec,
+                        dst4_r, dst5_r, dst4_l, dst5_l);
+
+    HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                     dst2_l, dst2_r, dst3_l, dst3_r,
+                     dst4_l, dst4_r, dst5_l, dst5_r, dst0_r, dst1_r, dst2_r);
+
+    ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+    dst += (4 * dst_stride);
+    ST8x2_UB(dst2_r, dst, dst_stride);
+}
+
+static void hevc_hz_uniwgt_4t_8x4multiple_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, src2, src3;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v16i8 vec0, vec1;
+    v8i16 dst0, dst1, dst2, dst3;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src0, src1, src2, src3);
+        src += (4 * src_stride);
+
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+
+        ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_8w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    if (2 == height) {
+        hevc_hz_uniwgt_4t_8x2_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else if (6 == height) {
+        hevc_hz_uniwgt_4t_8x6_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else {
+        hevc_hz_uniwgt_4t_8x4multiple_msa(src, src_stride, dst, dst_stride,
+                                          filter, height, weight, offset,
+                                          rnd_val);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_12w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v8i16 filt0, filt1;
+    v16i8 src0, src1, src2, src3;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask2 = { 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
+    };
+    v16i8 mask1;
+    v16i8 vec0, vec1;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v8i16 filter_vec, const_vec;
+    v16i8 mask3;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+    mask3 = mask2 + 2;
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src0, src1, src2, src3);
+        src += (4 * src_stride);
+
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
+        dst4 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+
+        VSHF_B2_SB(src2, src3, src2, src3, mask2, mask3, vec0, vec1);
+        dst5 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_UNIW_RND_CLIP2(dst4, dst5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                         dst2_l, dst2_r, dst3_l, dst3_r,
+                         dst4_l, dst4_r, dst5_l, dst5_r,
+                         dst0_r, dst1_r, dst2_r);
+
+        ST12x4_UB(dst0_r, dst1_r, dst2_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_16w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
+    v8i16 filt0, filt1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+    v16i8 vec0, vec1;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB4(src, src_stride, src0, src2, src4, src6);
+        LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+        src += (4 * src_stride);
+
+        XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+        dst4 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+
+        VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+        dst5 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+
+        VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
+        dst6 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6);
+
+        VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
+        dst7 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+
+        HEVC_UNIW_RND_CLIP4(dst4, dst5, dst6, dst7,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_24w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    uint8_t *dst_tmp = dst + 16;
+    v16i8 src0, src1, src2, src3;
+    v8i16 filt0, filt1;
+    v8i16 dst0, dst1, dst2, dst3;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1, mask2, mask3;
+    v16i8 vec0, vec1;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 8;
+    mask3 = mask0 + 10;
+
+    for (loop_cnt = (height >> 1); loop_cnt--;) {
+        /* 16 width */
+        LD_SB2(src, src_stride, src0, src2);
+        LD_SB2(src + 16, src_stride, src1, src3);
+        src += (2 * src_stride);
+
+        XORI_B4_128_SB(src0, src1, src2, src3);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src2, src3, src2, src3, mask2, mask3, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+
+        /* 8 width */
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        HEVC_UNIW_RND_CLIP2(dst0, dst1, weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst0_l, dst1_l);
+
+        HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+        ST8x2_UB(dst0_r, dst_tmp, dst_stride);
+        dst_tmp += (2 * dst_stride);
+    }
+}
+
+static void hevc_hz_uniwgt_4t_32w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2;
+    v8i16 filt0, filt1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1, mask2, mask3;
+    v8i16 dst0, dst1, dst2, dst3;
+    v16i8 vec0, vec1;
+    v8i16 filter_vec, const_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= 1;
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    weight = weight & 0x0000FFFF;
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    mask1 = mask0 + 2;
+    mask2 = mask0 + 8;
+    mask3 = mask0 + 10;
+
+    for (loop_cnt = (height >> 1); loop_cnt--;) {
+        LD_SB2(src, 16, src0, src1);
+        src2 = LD_SB(src + 24);
+        src += src_stride;
+
+        XORI_B3_128_SB(src0, src1, src2);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, 16);
+        dst += dst_stride;
+
+        LD_SB2(src, 16, src0, src1);
+        src2 = LD_SB(src + 24);
+        src += src_stride;
+
+        XORI_B3_128_SB(src0, src1, src2);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+
+        VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1);
+
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+        HEVC_UNIW_RND_CLIP4(dst0, dst1, dst2, dst3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, 16);
+        dst += dst_stride;
+    }
+}
+
+static void hevc_vt_uniwgt_4t_4x2_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4;
+    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v16i8 src2110, src4332;
+    v8i16 dst10;
+    v4i32 dst0_r, dst0_l;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r);
+    src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128);
+    LD_SB2(src, src_stride, src3, src4);
+    ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+    src4332 = (v16i8) __msa_ilvr_d((v2i64) src43_r, (v2i64) src32_r);
+    src4332 = (v16i8) __msa_xori_b((v16u8) src4332, 128);
+
+    dst10 = const_vec;
+    DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10);
+
+    ILVRL_H2_SW(dst10, dst10, dst0_r, dst0_l);
+    DOTP_SH2_SW(dst0_r, dst0_l, weight_vec, weight_vec, dst0_r, dst0_l);
+    SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
+    ADD2(dst0_r, offset_vec, dst0_l, offset_vec, dst0_r, dst0_l);
+    dst0_r = CLIP_SW_0_255(dst0_r);
+    dst0_l = CLIP_SW_0_255(dst0_l);
+
+    HEVC_PCK_SW_SB2(dst0_l, dst0_r, dst0_r);
+    ST4x2_UB(dst0_r, dst, dst_stride);
+}
+
+static void hevc_vt_uniwgt_4t_4x4_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4, src5, src6;
+    v16i8 src10_r, src32_r, src54_r, src21_r, src43_r, src65_r;
+    v16i8 src2110, src4332, src6554;
+    v8i16 dst10, dst32;
+    v4i32 dst0_r, dst1_r, dst0_l, dst1_l;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r);
+    src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128);
+
+    LD_SB4(src, src_stride, src3, src4, src5, src6);
+    ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5,
+               src32_r, src43_r, src54_r, src65_r);
+    ILVR_D2_SB(src43_r, src32_r, src65_r, src54_r, src4332, src6554);
+    XORI_B2_128_SB(src4332, src6554);
+
+    dst10 = const_vec;
+    DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10);
+    dst32 = const_vec;
+    DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32);
+    HEVC_UNIW_RND_CLIP2(dst10, dst32, weight_vec, offset_vec, rnd_vec,
+                        dst0_r, dst1_r, dst0_l, dst1_l);
+
+    HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+    ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
+    dst += (4 * dst_stride);
+}
+
+static void hevc_vt_uniwgt_4t_4x8multiple_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+    v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
+    v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
+    v16i8 src2110, src4332, src6554, src8776;
+    v8i16 dst10, dst32, dst54, dst76;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r);
+    src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128);
+
+    for (loop_cnt = (height >> 3); loop_cnt--;) {
+        LD_SB6(src, src_stride, src3, src4, src5, src6, src7, src8);
+        src += (6 * src_stride);
+        ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5,
+                   src32_r, src43_r, src54_r, src65_r);
+        ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
+        ILVR_D3_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r,
+                   src4332, src6554, src8776);
+        XORI_B3_128_SB(src4332, src6554, src8776);
+
+        dst10 = const_vec;
+        DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10);
+        dst32 = const_vec;
+        DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32);
+        dst54 = const_vec;
+        DPADD_SB2_SH(src6554, src8776, filt0, filt1, dst54, dst54);
+
+        LD_SB2(src, src_stride, src9, src2);
+        src += (2 * src_stride);
+        ILVR_B2_SB(src9, src8, src2, src9, src98_r, src109_r);
+        src2110 = (v16i8) __msa_ilvr_d((v2i64) src109_r, (v2i64) src98_r);
+        src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128);
+
+        dst76 = const_vec;
+        DPADD_SB2_SH(src8776, src2110, filt0, filt1, dst76, dst76);
+        HEVC_UNIW_RND_CLIP4(dst10, dst32, dst54, dst76,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST4x8_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (8 * dst_stride);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_4w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    if (2 == height) {
+        hevc_vt_uniwgt_4t_4x2_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else if (4 == height) {
+        hevc_vt_uniwgt_4t_4x4_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else if (0 == (height % 8)) {
+        hevc_vt_uniwgt_4t_4x8multiple_msa(src, src_stride, dst, dst_stride,
+                                          filter, height, weight, offset,
+                                          rnd_val);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_6w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4;
+    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB2(src, src_stride, src3, src4);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src3, src4);
+        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+
+        LD_SB2(src, src_stride, src1, src2);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src1, src2);
+        ILVR_B2_SB(src1, src4, src2, src1, src10_r, src21_r);
+
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, tmp3, tmp3);
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+
+        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_8x2_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4;
+    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v8i16 tmp0, tmp1;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst0_l, dst1_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    LD_SB2(src, src_stride, src3, src4);
+    XORI_B2_128_SB(src3, src4);
+    ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+
+    tmp0 = const_vec;
+    DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+    tmp1 = const_vec;
+    DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+    HEVC_UNIW_RND_CLIP2(tmp0, tmp1, weight_vec, offset_vec, rnd_vec,
+                        dst0_r, dst1_r, dst0_l, dst1_l);
+
+    HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+    ST8x2_UB(dst0_r, dst, dst_stride);
+}
+
+static void hevc_vt_uniwgt_4t_8x6_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v16i8 src10_r, src32_r, src54_r, src76_r;
+    v16i8 src21_r, src43_r, src65_r, src87_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+
+    LD_SB6(src, src_stride, src3, src4, src5, src6, src7, src8);
+    XORI_B6_128_SB(src3, src4, src5, src6, src7, src8);
+    ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5,
+               src32_r, src43_r, src54_r, src65_r);
+    ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
+
+    tmp0 = const_vec;
+    DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+    tmp1 = const_vec;
+    DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+    tmp2 = const_vec;
+    DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, tmp2, tmp2);
+    tmp3 = const_vec;
+    DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, tmp3, tmp3);
+    tmp4 = const_vec;
+    DPADD_SB2_SH(src54_r, src76_r, filt0, filt1, tmp4, tmp4);
+    tmp5 = const_vec;
+    DPADD_SB2_SH(src65_r, src87_r, filt0, filt1, tmp5, tmp5);
+    HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                        weight_vec, offset_vec, rnd_vec,
+                        dst0_r, dst1_r, dst2_r, dst3_r,
+                        dst0_l, dst1_l, dst2_l, dst3_l);
+    HEVC_UNIW_RND_CLIP2(tmp4, tmp5, weight_vec, offset_vec, rnd_vec,
+                        dst4_r, dst5_r, dst4_l, dst5_l);
+
+    HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                     dst2_l, dst2_r, dst3_l, dst3_r,
+                     dst4_l, dst4_r, dst5_l, dst5_r, dst0_r, dst1_r, dst2_r);
+    ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+    dst += (4 * dst_stride);
+    ST8x2_UB(dst2_r, dst, dst_stride);
+}
+
+static void hevc_vt_uniwgt_4t_8x4multiple_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4;
+    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB2(src, src_stride, src3, src4);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src3, src4);
+        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+
+        LD_SB2(src, src_stride, src1, src2);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src1, src2);
+        ILVR_B2_SB(src1, src4, src2, src1, src10_r, src21_r);
+
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, tmp3, tmp3);
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_8w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    if (2 == height) {
+        hevc_vt_uniwgt_4t_8x2_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else if (6 == height) {
+        hevc_vt_uniwgt_4t_8x6_msa(src, src_stride, dst, dst_stride,
+                                  filter, height, weight, offset, rnd_val);
+    } else {
+        hevc_vt_uniwgt_4t_8x4multiple_msa(src, src_stride, dst, dst_stride,
+                                          filter, height, weight, offset,
+                                          rnd_val);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_12w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5;
+    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+    v16i8 src10_l, src32_l, src54_l, src21_l, src43_l, src65_l;
+    v16i8 src2110, src4332;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+
+    src -= (1 * src_stride);
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
+    src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_l, (v2i64) src10_l);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB2(src, src_stride, src3, src4);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src3, src4);
+        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+        ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l);
+        src4332 = (v16i8) __msa_ilvr_d((v2i64) src43_l, (v2i64) src32_l);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+        tmp4 = const_vec;
+        DPADD_SB2_SH(src2110, src4332, filt0, filt1, tmp4, tmp4);
+
+        LD_SB2(src, src_stride, src5, src2);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src5, src2);
+        ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r);
+        ILVL_B2_SB(src5, src4, src2, src5, src54_l, src65_l);
+        src2110 = (v16i8) __msa_ilvr_d((v2i64) src65_l, (v2i64) src54_l);
+
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, tmp3, tmp3);
+        tmp5 = const_vec;
+        DPADD_SB2_SH(src4332, src2110, filt0, filt1, tmp5, tmp5);
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(tmp4, tmp5, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                         dst2_l, dst2_r, dst3_l, dst3_r,
+                         dst4_l, dst4_r, dst5_l, dst5_r,
+                         dst0_r, dst1_r, dst2_r);
+        ST12x4_UB(dst0_r, dst1_r, dst2_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_16w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5;
+    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v16i8 src10_l, src32_l, src21_l, src43_l;
+    v8i16 tmp0, tmp1, tmp2, tmp3;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB2(src, src_stride, src3, src4);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src3, src4);
+        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+        ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src10_l, src32_l, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src21_l, src43_l, filt0, filt1, tmp3, tmp3);
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                        dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+
+        LD_SB2(src, src_stride, src5, src2);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src5, src2);
+        ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r);
+        ILVL_B2_SB(src5, src4, src2, src5, src10_l, src21_l);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, tmp0, tmp0);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, tmp1, tmp1);
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src32_l, src10_l, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src43_l, src21_l, filt0, filt1, tmp3, tmp3);
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp2, tmp3,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                        dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_24w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5;
+    v16i8 src6, src7, src8, src9, src10, src11;
+    v16i8 src10_r, src32_r, src76_r, src98_r;
+    v16i8 src21_r, src43_r, src87_r, src109_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+    v16i8 src10_l, src32_l, src21_l, src43_l;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
+
+    LD_SB3(src + 16, src_stride, src6, src7, src8);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src6, src7, src8);
+    ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
+
+    for (loop_cnt = (height >> 2); loop_cnt--;) {
+        LD_SB2(src, src_stride, src3, src4);
+        XORI_B2_128_SB(src3, src4);
+        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+        ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l);
+        LD_SB2(src + 16, src_stride, src9, src10);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src9, src10);
+        ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+        tmp4 = const_vec;
+        DPADD_SB2_SH(src10_l, src32_l, filt0, filt1, tmp4, tmp4);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+        tmp5 = const_vec;
+        DPADD_SB2_SH(src21_l, src43_l, filt0, filt1, tmp5, tmp5);
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src76_r, src98_r, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src87_r, src109_r, filt0, filt1, tmp3, tmp3);
+
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp4, tmp5,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(tmp2, tmp3, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                        dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        HEVC_PCK_SW_SB4(dst4_l, dst4_r, dst5_l, dst5_r, dst4_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        ST8x2_UB(dst4_r, dst + 16, dst_stride);
+        dst += (2 * dst_stride);
+
+        LD_SB2(src, src_stride, src5, src2);
+        XORI_B2_128_SB(src5, src2);
+        ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r);
+        ILVL_B2_SB(src5, src4, src2, src5, src10_l, src21_l);
+        LD_SB2(src + 16, src_stride, src11, src8);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src11, src8);
+        ILVR_B2_SB(src11, src10, src8, src11, src76_r, src87_r);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, tmp0, tmp0);
+        tmp4 = const_vec;
+        DPADD_SB2_SH(src32_l, src10_l, filt0, filt1, tmp4, tmp4);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, tmp1, tmp1);
+        tmp5 = const_vec;
+        DPADD_SB2_SH(src43_l, src21_l, filt0, filt1, tmp5, tmp5);
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src98_r, src76_r, filt0, filt1, tmp2, tmp2);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src109_r, src87_r, filt0, filt1, tmp3, tmp3);
+
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp4, tmp5,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_UNIW_RND_CLIP2(tmp2, tmp3, weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst4_l, dst5_l);
+
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                        dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        HEVC_PCK_SW_SB4(dst4_l, dst4_r, dst5_l, dst5_r, dst4_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        ST8x2_UB(dst4_r, dst + 16, dst_stride);
+        dst += (2 * dst_stride);
+    }
+}
+
+static void hevc_vt_uniwgt_4t_32w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    int32_t loop_cnt;
+    uint8_t *dst_tmp = dst + 16;
+    v16i8 src0, src1, src2, src3, src4, src6, src7, src8, src9, src10;
+    v16i8 src10_r, src32_r, src76_r, src98_r;
+    v16i8 src21_r, src43_r, src87_r, src109_r;
+    v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+    v16i8 src10_l, src32_l, src76_l, src98_l;
+    v16i8 src21_l, src43_l, src87_l, src109_l;
+    v8i16 filt0, filt1;
+    v8i16 filter_vec, const_vec;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r;
+    v4i32 dst0_l, dst1_l, dst2_l, dst3_l, dst4_l, dst5_l, dst6_l, dst7_l;
+
+    src -= src_stride;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+    weight = weight & 0x0000FFFF;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    filter_vec = LD_SH(filter);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    XORI_B3_128_SB(src0, src1, src2);
+    ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
+
+    LD_SB3(src + 16, src_stride, src6, src7, src8);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src6, src7, src8);
+    ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
+    ILVL_B2_SB(src7, src6, src8, src7, src76_l, src87_l);
+
+    for (loop_cnt = (height >> 1); loop_cnt--;) {
+        LD_SB2(src, src_stride, src3, src4);
+        XORI_B2_128_SB(src3, src4);
+        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+        ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l);
+
+        tmp0 = const_vec;
+        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, tmp0, tmp0);
+        tmp4 = const_vec;
+        DPADD_SB2_SH(src10_l, src32_l, filt0, filt1, tmp4, tmp4);
+        tmp1 = const_vec;
+        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, tmp1, tmp1);
+        tmp5 = const_vec;
+        DPADD_SB2_SH(src21_l, src43_l, filt0, filt1, tmp5, tmp5);
+
+        HEVC_UNIW_RND_CLIP4(tmp0, tmp1, tmp4, tmp5,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst0_r, dst1_r, dst2_r, dst3_r,
+                            dst0_l, dst1_l, dst2_l, dst3_l);
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst2_l, dst2_r,
+                        dst1_l, dst1_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST_SW2(dst0_r, dst1_r, dst, dst_stride);
+        dst += (2 * dst_stride);
+
+        src10_r = src32_r;
+        src21_r = src43_r;
+        src10_l = src32_l;
+        src21_l = src43_l;
+        src2 = src4;
+
+        LD_SB2(src + 16, src_stride, src9, src10);
+        src += (2 * src_stride);
+        XORI_B2_128_SB(src9, src10);
+        ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r);
+        ILVL_B2_SB(src9, src8, src10, src9, src98_l, src109_l);
+
+        tmp2 = const_vec;
+        DPADD_SB2_SH(src76_r, src98_r, filt0, filt1, tmp2, tmp2);
+        tmp6 = const_vec;
+        DPADD_SB2_SH(src76_l, src98_l, filt0, filt1, tmp6, tmp6);
+        tmp3 = const_vec;
+        DPADD_SB2_SH(src87_r, src109_r, filt0, filt1, tmp3, tmp3);
+        tmp7 = const_vec;
+        DPADD_SB2_SH(src87_l, src109_l, filt0, filt1, tmp7, tmp7);
+
+        HEVC_UNIW_RND_CLIP4(tmp2, tmp3, tmp6, tmp7,
+                            weight_vec, offset_vec, rnd_vec,
+                            dst4_r, dst5_r, dst6_r, dst7_r,
+                            dst4_l, dst5_l, dst6_l, dst7_l);
+
+        HEVC_PCK_SW_SB8(dst4_l, dst4_r, dst6_l, dst6_r,
+                        dst5_l, dst5_r, dst7_l, dst7_r, dst4_r, dst5_r);
+        ST_SW2(dst4_r, dst5_r, dst_tmp, dst_stride);
+        dst_tmp += (2 * dst_stride);
+
+        src76_r = src98_r;
+        src87_r = src109_r;
+        src76_l = src98_l;
+        src87_l = src109_l;
+        src8 = src10;
+    }
+}
+
+static void hevc_hv_uniwgt_4t_4x2_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4;
+    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v4i32 dst0_r, dst1_r;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+
+    ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r);
+    LD_SB2(src, src_stride, src3, src4);
+    XORI_B2_128_SB(src3, src4);
+
+    /* row 3 */
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+    dst3 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+
+    dst32_r = __msa_ilvr_h(dst3, dst2);
+    dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+    dst0_r >>= 6;
+
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+    dst4 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+
+    dst43_r = __msa_ilvr_h(dst4, dst3);
+    dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+    dst1_r >>= 6;
+
+    MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
+    SRAR_W2_SW(dst0_r, dst1_r, rnd_vec);
+    ADD2(dst0_r, offset_vec, dst1_r, offset_vec, dst0_r, dst1_r);
+    dst0_r = CLIP_SW_0_255(dst0_r);
+    dst1_r = CLIP_SW_0_255(dst1_r);
+
+    HEVC_PCK_SW_SB2(dst1_r, dst0_r, dst0_r);
+    ST4x2_UB(dst0_r, dst, dst_stride);
+}
+
+static void hevc_hv_uniwgt_4t_4x4_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4, src5, src6;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r;
+    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+
+    ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r);
+
+    LD_SB4(src, src_stride, src3, src4, src5, src6);
+    XORI_B4_128_SB(src3, src4, src5, src6);
+
+    /* row 3 */
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+    dst3 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+    dst32_r = __msa_ilvr_h(dst3, dst2);
+    dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+    dst0_r >>= 6;
+
+    /* row 4 */
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+    dst4 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+    dst43_r = __msa_ilvr_h(dst4, dst3);
+    dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+    dst1_r >>= 6;
+
+    /* row 5 */
+    VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+    dst5 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+    dst10_r = __msa_ilvr_h(dst5, dst4);
+    dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
+    dst2_r >>= 6;
+
+    /* row 6 */
+    VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+    dst21_r = __msa_ilvr_h(dst2, dst5);
+    dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
+    dst3_r >>= 6;
+
+    HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst2_r, dst3_r,
+                           weight_vec, offset_vec, rnd_vec,
+                           dst0_r, dst1_r, dst2_r, dst3_r);
+    HEVC_PCK_SW_SB4(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r);
+    ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void hevc_hv_uniwgt_4t_4multx8mult_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter_x,
+                                              const int8_t *filter_y,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8, dst9;
+    v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r;
+    v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
+    v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+    ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r);
+
+    for (loop_cnt = height >> 3; loop_cnt--;) {
+        LD_SB8(src, src_stride,
+               src3, src4, src5, src6, src7, src8, src9, src10);
+        src += (8 * src_stride);
+        XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+        dst32_r = __msa_ilvr_h(dst3, dst2);
+        dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+        dst0_r >>= 6;
+
+        VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+        dst4 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+        dst43_r = __msa_ilvr_h(dst4, dst3);
+        dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+        dst1_r >>= 6;
+
+        VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+        dst5 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+        dst54_r = __msa_ilvr_h(dst5, dst4);
+        dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+        dst2_r >>= 6;
+
+        VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
+        dst6 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6);
+        dst65_r = __msa_ilvr_h(dst6, dst5);
+        dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+        dst3_r >>= 6;
+
+        VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
+        dst7 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7);
+        dst76_r = __msa_ilvr_h(dst7, dst6);
+        dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
+        dst4_r >>= 6;
+
+        VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec0, vec1);
+        dst8 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst8, dst8);
+        dst87_r = __msa_ilvr_h(dst8, dst7);
+        dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
+        dst5_r >>= 6;
+
+        VSHF_B2_SB(src9, src9, src9, src9, mask0, mask1, vec0, vec1);
+        dst9 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst9, dst9);
+        dst10_r = __msa_ilvr_h(dst9, dst8);
+        dst6_r = HEVC_FILT_4TAP(dst76_r, dst10_r, filt_h0, filt_h1);
+        dst6_r >>= 6;
+
+        VSHF_B2_SB(src10, src10, src10, src10, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+        dst21_r = __msa_ilvr_h(dst2, dst9);
+        dst7_r = HEVC_FILT_4TAP(dst87_r, dst21_r, filt_h0, filt_h1);
+        dst7_r >>= 6;
+
+        HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst2_r, dst3_r,
+                               weight_vec, offset_vec, rnd_vec,
+                               dst0_r, dst1_r, dst2_r, dst3_r);
+        HEVC_PCK_SW_SB4(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r);
+        ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
+        dst += (4 * dst_stride);
+
+        HEVC_HV_UNIW_RND_CLIP4(dst4_r, dst5_r, dst6_r, dst7_r,
+                               weight_vec, offset_vec, rnd_vec,
+                               dst4_r, dst5_r, dst6_r, dst7_r);
+        HEVC_PCK_SW_SB4(dst5_r, dst4_r, dst7_r, dst6_r, dst0_r);
+        ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_hv_uniwgt_4t_4w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter_x,
+                                     const int8_t *filter_y,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    if (2 == height) {
+        hevc_hv_uniwgt_4t_4x2_msa(src, src_stride, dst, dst_stride,
+                                  filter_x, filter_y, height, weight,
+                                  offset, rnd_val);
+    } else if (4 == height) {
+        hevc_hv_uniwgt_4t_4x4_msa(src, src_stride, dst, dst_stride,
+                                  filter_x, filter_y, height, weight,
+                                  offset, rnd_val);
+    } else if (0 == (height % 8)) {
+        hevc_hv_uniwgt_4t_4multx8mult_msa(src, src_stride, dst, dst_stride,
+                                          filter_x, filter_y, height, weight,
+                                          offset, rnd_val);
+    }
+}
+
+static void hevc_hv_uniwgt_4t_6w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter_x,
+                                     const int8_t *filter_y,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+    uint32_t loop_cnt;
+    v16i8 src0, src1, src2, src3, src4, src5, src6;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst2_r, dst2_l, dst3_r, dst3_l;
+    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+
+    ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
+    ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
+
+    for (loop_cnt = height >> 2; loop_cnt--;) {
+        LD_SB4(src, src_stride, src3, src4, src5, src6);
+        src += (4 * src_stride);
+        XORI_B4_128_SB(src3, src4, src5, src6);
+
+        /* row 3 */
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        dst3 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+        ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+        dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+        dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
+        dst0_r >>= 6;
+        dst0_l >>= 6;
+
+        /* row 4 */
+        VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+        dst4 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+        ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+        dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+        dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
+        dst1_r >>= 6;
+        dst1_l >>= 6;
+
+        /* row 5 */
+        VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+        dst5 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+        ILVRL_H2_SH(dst5, dst4, dst10_r, dst10_l);
+        dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
+        dst2_l = HEVC_FILT_4TAP(dst32_l, dst10_l, filt_h0, filt_h1);
+        dst2_r >>= 6;
+        dst2_l >>= 6;
+
+        /* row 6 */
+        VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+        ILVRL_H2_SH(dst2, dst5, dst21_r, dst21_l);
+        dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
+        dst3_l = HEVC_FILT_4TAP(dst43_l, dst21_l, filt_h0, filt_h1);
+        dst3_r >>= 6;
+        dst3_l >>= 6;
+
+        HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst0_l, dst1_l,
+                               weight_vec, offset_vec, rnd_vec,
+                               dst0_r, dst1_r, dst0_l, dst1_l);
+        HEVC_HV_UNIW_RND_CLIP4(dst2_r, dst3_r, dst2_l, dst3_l,
+                               weight_vec, offset_vec, rnd_vec,
+                               dst2_r, dst3_r, dst2_l, dst3_l);
+        HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                        dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+        dst += (4 * dst_stride);
+    }
+}
+
+static void hevc_hv_uniwgt_4t_8x2_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
+    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+
+    ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
+    ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
+
+    LD_SB2(src, src_stride, src3, src4);
+    src += (2 * src_stride);
+    XORI_B2_128_SB(src3, src4);
+
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+    dst3 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+    ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+    dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+    dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
+    dst0_r >>= 6;
+    dst0_l >>= 6;
+
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+    dst4 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+    ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+    dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+    dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
+    dst1_r >>= 6;
+    dst1_l >>= 6;
+
+    HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst0_l, dst1_l,
+                           weight_vec, offset_vec, rnd_vec,
+                           dst0_r, dst1_r, dst0_l, dst1_l);
+    HEVC_PCK_SW_SB4(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r);
+    ST8x2_UB(dst0_r, dst, dst_stride);
+    dst += (2 * dst_stride);
+}
+
+static void hevc_hv_uniwgt_4t_8x6_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+    v4i32 dst4_r, dst4_l, dst5_r, dst5_l;
+    v8i16 dst10_r, dst32_r, dst10_l, dst32_l;
+    v8i16 dst21_r, dst43_r, dst21_l, dst43_l;
+    v8i16 dst54_r, dst54_l, dst65_r, dst65_l;
+    v8i16 dst76_r, dst76_l, dst87_r, dst87_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    LD_SB3(src, src_stride, src0, src1, src2);
+    src += (3 * src_stride);
+
+    XORI_B3_128_SB(src0, src1, src2);
+
+    VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+    dst0 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+    dst1 = const_vec;
+    DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+    dst2 = const_vec;
+    DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+
+    ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
+    ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
+
+    LD_SB2(src, src_stride, src3, src4);
+    src += (2 * src_stride);
+    XORI_B2_128_SB(src3, src4);
+
+    /* row 3 */
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+    dst3 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+    ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+    dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+    dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
+    dst0_r >>= 6;
+    dst0_l >>= 6;
+
+    /* row 4 */
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+    dst4 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+    ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+    dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+    dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
+    dst1_r >>= 6;
+    dst1_l >>= 6;
+
+    LD_SB2(src, src_stride, src5, src6);
+    src += (2 * src_stride);
+    XORI_B2_128_SB(src5, src6);
+
+    /* row 5 */
+    VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+    dst5 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+    ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+    dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+    dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
+    dst2_r >>= 6;
+    dst2_l >>= 6;
+
+    /* row 6 */
+    VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
+    dst6 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6);
+    ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
+    dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+    dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
+    dst3_r >>= 6;
+    dst3_l >>= 6;
+
+    LD_SB2(src, src_stride, src7, src8);
+    src += (2 * src_stride);
+    XORI_B2_128_SB(src7, src8);
+
+    /* row 7 */
+    VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
+    dst7 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7);
+    ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
+    dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
+    dst4_l = HEVC_FILT_4TAP(dst54_l, dst76_l, filt_h0, filt_h1);
+
+    dst4_r >>= 6;
+    dst4_l >>= 6;
+
+    /* row 8 */
+    VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec0, vec1);
+    dst8 = const_vec;
+    DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst8, dst8);
+    ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l);
+    dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
+    dst5_l = HEVC_FILT_4TAP(dst65_l, dst87_l, filt_h0, filt_h1);
+    dst5_r >>= 6;
+    dst5_l >>= 6;
+
+    HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst0_l, dst1_l,
+                           weight_vec, offset_vec, rnd_vec,
+                           dst0_r, dst1_r, dst0_l, dst1_l);
+    HEVC_HV_UNIW_RND_CLIP4(dst2_r, dst3_r, dst2_l, dst3_l,
+                           weight_vec, offset_vec, rnd_vec,
+                           dst2_r, dst3_r, dst2_l, dst3_l);
+    HEVC_HV_UNIW_RND_CLIP4(dst4_r, dst5_r, dst4_l, dst5_l,
+                           weight_vec, offset_vec, rnd_vec,
+                           dst4_r, dst5_r, dst4_l, dst5_l);
+    HEVC_PCK_SW_SB12(dst0_l, dst0_r, dst1_l, dst1_r,
+                     dst2_l, dst2_r, dst3_l, dst3_r,
+                     dst4_l, dst4_r, dst5_l, dst5_r, dst0_r, dst1_r, dst2_r);
+    ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+    dst += (4 * dst_stride);
+    ST8x2_UB(dst2_r, dst, dst_stride);
+}
+
+static void hevc_hv_uniwgt_4t_8multx4mult_msa(uint8_t *src,
+                                              int32_t src_stride,
+                                              uint8_t *dst,
+                                              int32_t dst_stride,
+                                              const int8_t *filter_x,
+                                              const int8_t *filter_y,
+                                              int32_t height,
+                                              int32_t weight,
+                                              int32_t offset,
+                                              int32_t rnd_val,
+                                              int32_t width)
+{
+    uint32_t loop_cnt, cnt;
+    uint8_t *src_tmp;
+    uint8_t *dst_tmp;
+    v16i8 src0, src1, src2, src3, src4, src5, src6;
+    v8i16 filt0, filt1;
+    v4i32 filt_h0, filt_h1;
+    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask1;
+    v8i16 filter_vec, const_vec;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
+    v4i32 weight_vec, offset_vec, rnd_vec;
+    v4i32 dst2_r, dst2_l, dst3_r, dst3_l;
+    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
+
+    src -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
+    filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+
+    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    weight_vec = __msa_fill_w(weight);
+    offset_vec = __msa_fill_w(offset);
+    rnd_vec = __msa_fill_w(rnd_val);
+
+    for (cnt = width >> 3; cnt--;) {
+        src_tmp = src;
+        dst_tmp = dst;
+
+        LD_SB3(src_tmp, src_stride, src0, src1, src2);
+        src_tmp += (3 * src_stride);
+        XORI_B3_128_SB(src0, src1, src2);
+
+        VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+        VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+        VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+        dst0 = const_vec;
+        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
+        dst1 = const_vec;
+        DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
+        dst2 = const_vec;
+        DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+
+        ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
+        ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
+
+        for (loop_cnt = height >> 2; loop_cnt--;) {
+            LD_SB4(src_tmp, src_stride, src3, src4, src5, src6);
+            src_tmp += (4 * src_stride);
+            XORI_B4_128_SB(src3, src4, src5, src6);
+
+            VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+            dst3 = const_vec;
+            DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+            ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+            dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+            dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
+            dst0_r >>= 6;
+            dst0_l >>= 6;
+
+            VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+            dst4 = const_vec;
+            DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+            ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+            dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+            dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
+            dst1_r >>= 6;
+            dst1_l >>= 6;
+
+            VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
+            dst5 = const_vec;
+            DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
+            ILVRL_H2_SH(dst5, dst4, dst10_r, dst10_l);
+            dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
+            dst2_l = HEVC_FILT_4TAP(dst32_l, dst10_l, filt_h0, filt_h1);
+            dst2_r >>= 6;
+            dst2_l >>= 6;
+
+            VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
+            dst2 = const_vec;
+            DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+            ILVRL_H2_SH(dst2, dst5, dst21_r, dst21_l);
+            dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
+            dst3_l = HEVC_FILT_4TAP(dst43_l, dst21_l, filt_h0, filt_h1);
+            dst3_r >>= 6;
+            dst3_l >>= 6;
+
+            HEVC_HV_UNIW_RND_CLIP4(dst0_r, dst1_r, dst0_l, dst1_l,
+                                   weight_vec, offset_vec, rnd_vec,
+                                   dst0_r, dst1_r, dst0_l, dst1_l);
+            HEVC_HV_UNIW_RND_CLIP4(dst2_r, dst3_r, dst2_l, dst3_l,
+                                   weight_vec, offset_vec, rnd_vec,
+                                   dst2_r, dst3_r, dst2_l, dst3_l);
+            HEVC_PCK_SW_SB8(dst0_l, dst0_r, dst1_l, dst1_r,
+                            dst2_l, dst2_r, dst3_l, dst3_r, dst0_r, dst1_r);
+            ST8x4_UB(dst0_r, dst1_r, dst_tmp, dst_stride);
+            dst_tmp += (4 * dst_stride);
+        }
+
+        src += 8;
+        dst += 8;
+    }
+}
+
+static void hevc_hv_uniwgt_4t_8w_msa(uint8_t *src,
+                                     int32_t src_stride,
+                                     uint8_t *dst,
+                                     int32_t dst_stride,
+                                     const int8_t *filter_x,
+                                     const int8_t *filter_y,
+                                     int32_t height,
+                                     int32_t weight,
+                                     int32_t offset,
+                                     int32_t rnd_val)
+{
+
+    if (2 == height) {
+        hevc_hv_uniwgt_4t_8x2_msa(src, src_stride, dst, dst_stride,
+                                  filter_x, filter_y, height, weight,
+                                  offset, rnd_val);
+    } else if (6 == height) {
+        hevc_hv_uniwgt_4t_8x6_msa(src, src_stride, dst, dst_stride,
+                                  filter_x, filter_y, height, weight,
+                                  offset, rnd_val);
+    } else if (0 == (height % 4)) {
+        hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+                                          filter_x, filter_y, height, weight,
+                                          offset, rnd_val, 8);
+    }
+}
+
+static void hevc_hv_uniwgt_4t_12w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 8);
+    hevc_hv_uniwgt_4t_4w_msa(src + 8, src_stride, dst + 8, dst_stride,
+                             filter_x, filter_y, height, weight,
+                             offset, rnd_val);
+}
+
+static void hevc_hv_uniwgt_4t_16w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 16);
+}
+
+static void hevc_hv_uniwgt_4t_24w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 24);
+}
+
+static void hevc_hv_uniwgt_4t_32w_msa(uint8_t *src,
+                                      int32_t src_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t height,
+                                      int32_t weight,
+                                      int32_t offset,
+                                      int32_t rnd_val)
+{
+    hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+                                      filter_x, filter_y, height, weight,
+                                      offset, rnd_val, 32);
+}
+
+#define UNIWGT_MC_COPY(WIDTH)                                                \
+void ff_hevc_put_hevc_uni_w_pel_pixels##WIDTH##_8_msa(uint8_t *dst,          \
+                                                      ptrdiff_t dst_stride,  \
+                                                      uint8_t *src,          \
+                                                      ptrdiff_t src_stride,  \
+                                                      int height,            \
+                                                      int denom,             \
+                                                      int weight,            \
+                                                      int offset,            \
+                                                      intptr_t mx,           \
+                                                      intptr_t my,           \
+                                                      int width)             \
+{                                                                            \
+    int shift = denom + 14 - 8;                                              \
+    hevc_uniwgt_copy_##WIDTH##w_msa(src, src_stride, dst, dst_stride,        \
+                                    height, weight, offset, shift);          \
+}
+
+UNIWGT_MC_COPY(4);
+UNIWGT_MC_COPY(6);
+UNIWGT_MC_COPY(8);
+UNIWGT_MC_COPY(12);
+UNIWGT_MC_COPY(16);
+UNIWGT_MC_COPY(24);
+UNIWGT_MC_COPY(32);
+UNIWGT_MC_COPY(48);
+UNIWGT_MC_COPY(64);
+
+#undef UNIWGT_MC_COPY
+
+#define UNI_W_MC(PEL, DIR, WIDTH, TAP, DIR1, FILT_DIR)                        \
+void ff_hevc_put_hevc_uni_w_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst,      \
+                                                           ptrdiff_t          \
+                                                           dst_stride,        \
+                                                           uint8_t *src,      \
+                                                           ptrdiff_t          \
+                                                           src_stride,        \
+                                                           int height,        \
+                                                           int denom,         \
+                                                           int weight,        \
+                                                           int offset,        \
+                                                           intptr_t mx,       \
+                                                           intptr_t my,       \
+                                                           int width)         \
+{                                                                             \
+    const int8_t *filter = ff_hevc_##PEL##_filters[FILT_DIR - 1];             \
+    int shift = denom + 14 - 8;                                               \
+                                                                              \
+    hevc_##DIR1##_uniwgt_##TAP##t_##WIDTH##w_msa(src, src_stride, dst,        \
+                                                 dst_stride, filter, height,  \
+                                                 weight, offset, shift);      \
+}
+
+UNI_W_MC(qpel, h, 4, 8, hz, mx);
+UNI_W_MC(qpel, h, 8, 8, hz, mx);
+UNI_W_MC(qpel, h, 12, 8, hz, mx);
+UNI_W_MC(qpel, h, 16, 8, hz, mx);
+UNI_W_MC(qpel, h, 24, 8, hz, mx);
+UNI_W_MC(qpel, h, 32, 8, hz, mx);
+UNI_W_MC(qpel, h, 48, 8, hz, mx);
+UNI_W_MC(qpel, h, 64, 8, hz, mx);
+
+UNI_W_MC(qpel, v, 4, 8, vt, my);
+UNI_W_MC(qpel, v, 8, 8, vt, my);
+UNI_W_MC(qpel, v, 12, 8, vt, my);
+UNI_W_MC(qpel, v, 16, 8, vt, my);
+UNI_W_MC(qpel, v, 24, 8, vt, my);
+UNI_W_MC(qpel, v, 32, 8, vt, my);
+UNI_W_MC(qpel, v, 48, 8, vt, my);
+UNI_W_MC(qpel, v, 64, 8, vt, my);
+
+UNI_W_MC(epel, h, 4, 4, hz, mx);
+UNI_W_MC(epel, h, 6, 4, hz, mx);
+UNI_W_MC(epel, h, 8, 4, hz, mx);
+UNI_W_MC(epel, h, 12, 4, hz, mx);
+UNI_W_MC(epel, h, 16, 4, hz, mx);
+UNI_W_MC(epel, h, 24, 4, hz, mx);
+UNI_W_MC(epel, h, 32, 4, hz, mx);
+
+UNI_W_MC(epel, v, 4, 4, vt, my);
+UNI_W_MC(epel, v, 6, 4, vt, my);
+UNI_W_MC(epel, v, 8, 4, vt, my);
+UNI_W_MC(epel, v, 12, 4, vt, my);
+UNI_W_MC(epel, v, 16, 4, vt, my);
+UNI_W_MC(epel, v, 24, 4, vt, my);
+UNI_W_MC(epel, v, 32, 4, vt, my);
+
+#undef UNI_W_MC
+
+#define UNI_W_MC_HV(PEL, DIR, WIDTH, TAP, DIR1)                              \
+void ff_hevc_put_hevc_uni_w_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst,     \
+                                                           ptrdiff_t         \
+                                                           dst_stride,       \
+                                                           uint8_t *src,     \
+                                                           ptrdiff_t         \
+                                                           src_stride,       \
+                                                           int height,       \
+                                                           int denom,        \
+                                                           int weight,       \
+                                                           int offset,       \
+                                                           intptr_t mx,      \
+                                                           intptr_t my,      \
+                                                           int width)        \
+{                                                                            \
+    const int8_t *filter_x = ff_hevc_##PEL##_filters[mx - 1];                \
+    const int8_t *filter_y = ff_hevc_##PEL##_filters[my - 1];                \
+    int shift = denom + 14 - 8;                                              \
+                                                                             \
+    hevc_##DIR1##_uniwgt_##TAP##t_##WIDTH##w_msa(src, src_stride, dst,       \
+                                                 dst_stride, filter_x,       \
+                                                 filter_y,  height, weight,  \
+                                                 offset, shift);             \
+}
+
+UNI_W_MC_HV(qpel, hv, 4, 8, hv);
+UNI_W_MC_HV(qpel, hv, 8, 8, hv);
+UNI_W_MC_HV(qpel, hv, 12, 8, hv);
+UNI_W_MC_HV(qpel, hv, 16, 8, hv);
+UNI_W_MC_HV(qpel, hv, 24, 8, hv);
+UNI_W_MC_HV(qpel, hv, 32, 8, hv);
+UNI_W_MC_HV(qpel, hv, 48, 8, hv);
+UNI_W_MC_HV(qpel, hv, 64, 8, hv);
+
+UNI_W_MC_HV(epel, hv, 4, 4, hv);
+UNI_W_MC_HV(epel, hv, 6, 4, hv);
+UNI_W_MC_HV(epel, hv, 8, 4, hv);
+UNI_W_MC_HV(epel, hv, 12, 4, hv);
+UNI_W_MC_HV(epel, hv, 16, 4, hv);
+UNI_W_MC_HV(epel, hv, 24, 4, hv);
+UNI_W_MC_HV(epel, hv, 32, 4, hv);
+
+#undef UNI_W_MC_HV
index 1d8ed9a..5451f96 100644 (file)
@@ -97,6 +97,99 @@ static av_cold void hevc_dsp_init_msa(HEVCDSPContext *c,
         c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_msa;
         c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_msa;
 
+        c->put_hevc_qpel_uni_w[1][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels4_8_msa;
+        c->put_hevc_qpel_uni_w[3][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels8_8_msa;
+        c->put_hevc_qpel_uni_w[4][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels12_8_msa;
+        c->put_hevc_qpel_uni_w[5][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels16_8_msa;
+        c->put_hevc_qpel_uni_w[6][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels24_8_msa;
+        c->put_hevc_qpel_uni_w[7][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels32_8_msa;
+        c->put_hevc_qpel_uni_w[8][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels48_8_msa;
+        c->put_hevc_qpel_uni_w[9][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels64_8_msa;
+
+        c->put_hevc_qpel_uni_w[1][0][1] = ff_hevc_put_hevc_uni_w_qpel_h4_8_msa;
+        c->put_hevc_qpel_uni_w[3][0][1] = ff_hevc_put_hevc_uni_w_qpel_h8_8_msa;
+        c->put_hevc_qpel_uni_w[4][0][1] = ff_hevc_put_hevc_uni_w_qpel_h12_8_msa;
+        c->put_hevc_qpel_uni_w[5][0][1] = ff_hevc_put_hevc_uni_w_qpel_h16_8_msa;
+        c->put_hevc_qpel_uni_w[6][0][1] = ff_hevc_put_hevc_uni_w_qpel_h24_8_msa;
+        c->put_hevc_qpel_uni_w[7][0][1] = ff_hevc_put_hevc_uni_w_qpel_h32_8_msa;
+        c->put_hevc_qpel_uni_w[8][0][1] = ff_hevc_put_hevc_uni_w_qpel_h48_8_msa;
+        c->put_hevc_qpel_uni_w[9][0][1] = ff_hevc_put_hevc_uni_w_qpel_h64_8_msa;
+
+        c->put_hevc_qpel_uni_w[1][1][0] = ff_hevc_put_hevc_uni_w_qpel_v4_8_msa;
+        c->put_hevc_qpel_uni_w[3][1][0] = ff_hevc_put_hevc_uni_w_qpel_v8_8_msa;
+        c->put_hevc_qpel_uni_w[4][1][0] = ff_hevc_put_hevc_uni_w_qpel_v12_8_msa;
+        c->put_hevc_qpel_uni_w[5][1][0] = ff_hevc_put_hevc_uni_w_qpel_v16_8_msa;
+        c->put_hevc_qpel_uni_w[6][1][0] = ff_hevc_put_hevc_uni_w_qpel_v24_8_msa;
+        c->put_hevc_qpel_uni_w[7][1][0] = ff_hevc_put_hevc_uni_w_qpel_v32_8_msa;
+        c->put_hevc_qpel_uni_w[8][1][0] = ff_hevc_put_hevc_uni_w_qpel_v48_8_msa;
+        c->put_hevc_qpel_uni_w[9][1][0] = ff_hevc_put_hevc_uni_w_qpel_v64_8_msa;
+
+        c->put_hevc_qpel_uni_w[1][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv4_8_msa;
+        c->put_hevc_qpel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv8_8_msa;
+        c->put_hevc_qpel_uni_w[4][1][1] =
+            ff_hevc_put_hevc_uni_w_qpel_hv12_8_msa;
+        c->put_hevc_qpel_uni_w[5][1][1] =
+            ff_hevc_put_hevc_uni_w_qpel_hv16_8_msa;
+        c->put_hevc_qpel_uni_w[6][1][1] =
+            ff_hevc_put_hevc_uni_w_qpel_hv24_8_msa;
+        c->put_hevc_qpel_uni_w[7][1][1] =
+            ff_hevc_put_hevc_uni_w_qpel_hv32_8_msa;
+        c->put_hevc_qpel_uni_w[8][1][1] =
+            ff_hevc_put_hevc_uni_w_qpel_hv48_8_msa;
+        c->put_hevc_qpel_uni_w[9][1][1] =
+            ff_hevc_put_hevc_uni_w_qpel_hv64_8_msa;
+
+        c->put_hevc_epel_uni_w[1][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels4_8_msa;
+        c->put_hevc_epel_uni_w[2][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels6_8_msa;
+        c->put_hevc_epel_uni_w[3][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels8_8_msa;
+        c->put_hevc_epel_uni_w[4][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels12_8_msa;
+        c->put_hevc_epel_uni_w[5][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels16_8_msa;
+        c->put_hevc_epel_uni_w[6][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels24_8_msa;
+        c->put_hevc_epel_uni_w[7][0][0] =
+            ff_hevc_put_hevc_uni_w_pel_pixels32_8_msa;
+
+        c->put_hevc_epel_uni_w[1][0][1] = ff_hevc_put_hevc_uni_w_epel_h4_8_msa;
+        c->put_hevc_epel_uni_w[2][0][1] = ff_hevc_put_hevc_uni_w_epel_h6_8_msa;
+        c->put_hevc_epel_uni_w[3][0][1] = ff_hevc_put_hevc_uni_w_epel_h8_8_msa;
+        c->put_hevc_epel_uni_w[4][0][1] = ff_hevc_put_hevc_uni_w_epel_h12_8_msa;
+        c->put_hevc_epel_uni_w[5][0][1] = ff_hevc_put_hevc_uni_w_epel_h16_8_msa;
+        c->put_hevc_epel_uni_w[6][0][1] = ff_hevc_put_hevc_uni_w_epel_h24_8_msa;
+        c->put_hevc_epel_uni_w[7][0][1] = ff_hevc_put_hevc_uni_w_epel_h32_8_msa;
+
+        c->put_hevc_epel_uni_w[1][1][0] = ff_hevc_put_hevc_uni_w_epel_v4_8_msa;
+        c->put_hevc_epel_uni_w[2][1][0] = ff_hevc_put_hevc_uni_w_epel_v6_8_msa;
+        c->put_hevc_epel_uni_w[3][1][0] = ff_hevc_put_hevc_uni_w_epel_v8_8_msa;
+        c->put_hevc_epel_uni_w[4][1][0] = ff_hevc_put_hevc_uni_w_epel_v12_8_msa;
+        c->put_hevc_epel_uni_w[5][1][0] = ff_hevc_put_hevc_uni_w_epel_v16_8_msa;
+        c->put_hevc_epel_uni_w[6][1][0] = ff_hevc_put_hevc_uni_w_epel_v24_8_msa;
+        c->put_hevc_epel_uni_w[7][1][0] = ff_hevc_put_hevc_uni_w_epel_v32_8_msa;
+
+        c->put_hevc_epel_uni_w[1][1][1] = ff_hevc_put_hevc_uni_w_epel_hv4_8_msa;
+        c->put_hevc_epel_uni_w[2][1][1] = ff_hevc_put_hevc_uni_w_epel_hv6_8_msa;
+        c->put_hevc_epel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_epel_hv8_8_msa;
+        c->put_hevc_epel_uni_w[4][1][1] =
+            ff_hevc_put_hevc_uni_w_epel_hv12_8_msa;
+        c->put_hevc_epel_uni_w[5][1][1] =
+            ff_hevc_put_hevc_uni_w_epel_hv16_8_msa;
+        c->put_hevc_epel_uni_w[6][1][1] =
+            ff_hevc_put_hevc_uni_w_epel_hv24_8_msa;
+        c->put_hevc_epel_uni_w[7][1][1] =
+            ff_hevc_put_hevc_uni_w_epel_hv32_8_msa;
+
         c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa;
         c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa;
         c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa;
index d4f43e4..9695e73 100644 (file)
@@ -117,6 +117,90 @@ UNI_MC(qpel, hv, 64);
 
 #undef UNI_MC
 
+#define UNI_W_MC(PEL, DIR, WIDTH)                                         \
+void ff_hevc_put_hevc_uni_w_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst,  \
+                                                           ptrdiff_t      \
+                                                           dst_stride,    \
+                                                           uint8_t *src,  \
+                                                           ptrdiff_t      \
+                                                           src_stride,    \
+                                                           int height,    \
+                                                           int denom,     \
+                                                           int weight,    \
+                                                           int offset,    \
+                                                           intptr_t mx,   \
+                                                           intptr_t my,   \
+                                                           int width)
+
+UNI_W_MC(pel, pixels, 4);
+UNI_W_MC(pel, pixels, 6);
+UNI_W_MC(pel, pixels, 8);
+UNI_W_MC(pel, pixels, 12);
+UNI_W_MC(pel, pixels, 16);
+UNI_W_MC(pel, pixels, 24);
+UNI_W_MC(pel, pixels, 32);
+UNI_W_MC(pel, pixels, 48);
+UNI_W_MC(pel, pixels, 64);
+
+UNI_W_MC(qpel, h, 4);
+UNI_W_MC(qpel, h, 8);
+UNI_W_MC(qpel, h, 12);
+UNI_W_MC(qpel, h, 16);
+UNI_W_MC(qpel, h, 24);
+UNI_W_MC(qpel, h, 32);
+UNI_W_MC(qpel, h, 48);
+UNI_W_MC(qpel, h, 64);
+
+UNI_W_MC(qpel, v, 4);
+UNI_W_MC(qpel, v, 8);
+UNI_W_MC(qpel, v, 12);
+UNI_W_MC(qpel, v, 16);
+UNI_W_MC(qpel, v, 24);
+UNI_W_MC(qpel, v, 32);
+UNI_W_MC(qpel, v, 48);
+UNI_W_MC(qpel, v, 64);
+
+UNI_W_MC(qpel, hv, 4);
+UNI_W_MC(qpel, hv, 8);
+UNI_W_MC(qpel, hv, 12);
+UNI_W_MC(qpel, hv, 16);
+UNI_W_MC(qpel, hv, 24);
+UNI_W_MC(qpel, hv, 32);
+UNI_W_MC(qpel, hv, 48);
+UNI_W_MC(qpel, hv, 64);
+
+UNI_W_MC(epel, h, 4);
+UNI_W_MC(epel, h, 6);
+UNI_W_MC(epel, h, 8);
+UNI_W_MC(epel, h, 12);
+UNI_W_MC(epel, h, 16);
+UNI_W_MC(epel, h, 24);
+UNI_W_MC(epel, h, 32);
+UNI_W_MC(epel, h, 48);
+UNI_W_MC(epel, h, 64);
+
+UNI_W_MC(epel, v, 4);
+UNI_W_MC(epel, v, 6);
+UNI_W_MC(epel, v, 8);
+UNI_W_MC(epel, v, 12);
+UNI_W_MC(epel, v, 16);
+UNI_W_MC(epel, v, 24);
+UNI_W_MC(epel, v, 32);
+UNI_W_MC(epel, v, 48);
+UNI_W_MC(epel, v, 64);
+
+UNI_W_MC(epel, hv, 4);
+UNI_W_MC(epel, hv, 6);
+UNI_W_MC(epel, hv, 8);
+UNI_W_MC(epel, hv, 12);
+UNI_W_MC(epel, hv, 16);
+UNI_W_MC(epel, hv, 24);
+UNI_W_MC(epel, hv, 32);
+UNI_W_MC(epel, hv, 48);
+UNI_W_MC(epel, hv, 64);
+
+#undef UNI_W_MC
+
 #define BI_MC(PEL, DIR, WIDTH)                                                 \
 void ff_hevc_put_hevc_bi_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst,          \
                                                         ptrdiff_t dst_stride,  \
index c4c549d..17f5c05 100644 (file)
 }
 #define DOTP_SB4_SH(...) DOTP_SB4(v8i16, __VA_ARGS__)
 
+/* Description : Dot product of halfword vector elements
+   Arguments   : Inputs  - mult0, mult1
+                           cnst0, cnst1
+                 Outputs - out0, out1
+                 Return Type - signed word
+   Details     : Signed halfword elements from mult0 are multiplied with
+                 signed halfword elements from cnst0 producing a result
+                 twice the size of input i.e. signed word.
+                 Then this multiplication results of adjacent odd-even elements
+                 are added together and stored to the out vector
+                 (2 signed word results)
+*/
+#define DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1)   \
+{                                                                 \
+    out0 = (RTYPE) __msa_dotp_s_w((v8i16) mult0, (v8i16) cnst0);  \
+    out1 = (RTYPE) __msa_dotp_s_w((v8i16) mult1, (v8i16) cnst1);  \
+}
+#define DOTP_SH2_SW(...) DOTP_SH2(v4i32, __VA_ARGS__)
+
+#define DOTP_SH4(RTYPE, mult0, mult1, mult2, mult3,           \
+                 cnst0, cnst1, cnst2, cnst3,                  \
+                 out0, out1, out2, out3)                      \
+{                                                             \
+    DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1);  \
+    DOTP_SH2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3);  \
+}
+#define DOTP_SH4_SW(...) DOTP_SH4(v4i32, __VA_ARGS__)
+
 /* Description : Dot product & addition of byte vector elements
    Arguments   : Inputs  - mult0, mult1
                            cnst0, cnst1
     out1 = (RTYPE) __msa_ilvl_h((v8i16) in2, (v8i16) in3);  \
 }
 #define ILVL_H2_SH(...) ILVL_H2(v8i16, __VA_ARGS__)
+#define ILVL_H2_SW(...) ILVL_H2(v4i32, __VA_ARGS__)
 
 #define ILVL_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
                 out0, out1, out2, out3)                         \
     out1 = (RTYPE) __msa_ilvr_h((v8i16) in2, (v8i16) in3);  \
 }
 #define ILVR_H2_SH(...) ILVR_H2(v8i16, __VA_ARGS__)
+#define ILVR_H2_SW(...) ILVR_H2(v4i32, __VA_ARGS__)
 
 #define ILVR_H3(RTYPE, in0, in1, in2, in3, in4, in5, out0, out1, out2)  \
 {                                                                       \
 #define SRAR_H4_UH(...) SRAR_H4(v8u16, __VA_ARGS__)
 #define SRAR_H4_SH(...) SRAR_H4(v8i16, __VA_ARGS__)
 
+/* Description : Shift right arithmetic rounded words
+   Arguments   : Inputs  - in0, in1, shift
+                 Outputs - in0, in1, (in place)
+                 Return Type - as per RTYPE
+   Details     : Each element of vector 'in0' is shifted right arithmetic by
+                 number of bits respective element holds in vector 'shift'.
+                 The last discarded bit is added to shifted value for rounding
+                 and the result is in place written to 'in0'
+                 Here, 'shift' is a vector passed in
+                 Similar for other pairs
+*/
+#define SRAR_W2(RTYPE, in0, in1, shift)                      \
+{                                                            \
+    in0 = (RTYPE) __msa_srar_w((v4i32) in0, (v4i32) shift);  \
+    in1 = (RTYPE) __msa_srar_w((v4i32) in1, (v4i32) shift);  \
+}
+#define SRAR_W2_SW(...) SRAR_W2(v4i32, __VA_ARGS__)
+
+#define SRAR_W4(RTYPE, in0, in1, in2, in3, shift)  \
+{                                                  \
+    SRAR_W2(RTYPE, in0, in1, shift)                \
+    SRAR_W2(RTYPE, in2, in3, shift)                \
+}
+#define SRAR_W4_SW(...) SRAR_W4(v4i32, __VA_ARGS__)
+
 /* Description : Shift right arithmetic rounded (immediate)
    Arguments   : Inputs  - in0, in1, in2, in3, shift
                  Outputs - in0, in1, in2, in3 (in place)
     MUL2(in4, in5, in6, in7, out2, out3);                                     \
 }
 
+/* Description : Addition of 2 pairs of vectors
+   Arguments   : Inputs  - in0, in1, in2, in3
+                 Outputs - out0, out1
+   Details     : Each element from 2 pairs vectors is added and 2 results are
+                 produced
+*/
+#define ADD2(in0, in1, in2, in3, out0, out1)  \
+{                                             \
+    out0 = in0 + in1;                         \
+    out1 = in2 + in3;                         \
+}
+#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3)  \
+{                                                                             \
+    ADD2(in0, in1, in2, in3, out0, out1);                                     \
+    ADD2(in4, in5, in6, in7, out2, out3);                                     \
+}
+
 /* Description : Zero extend unsigned byte elements to halfword elements
    Arguments   : Inputs  - in           (1 input unsigned byte vector)
                  Outputs - out0, out1   (unsigned 2 halfword vectors)