Merge commit 'accb06120c13a4ead442464d96f2fa318fa07a4e'

[ffmpeg] / libavcodec / mips / hevc_mc_bi_msa.c

diff --git a/libavcodec/mips/hevc_mc_bi_msa.c b/libavcodec/mips/hevc_mc_bi_msa.c
index 9c03ef8c5ae785c873ea97f26884c1841ac48e66..b555517b27d7e12d11c6473b69cb9c07b9498f88 100644 (file)
--- a/libavcodec/mips/hevc_mc_bi_msa.c
+++ b/libavcodec/mips/hevc_mc_bi_msa.c
@@ -2183,7 +2183,7 @@ static void hevc_hz_bi_4t_4x2_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1, dst0, vec0, vec1;
     v8i16 in0, in1;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
     v16i8 mask1;
     v8i16 tmp0;
     v8i16 filter_vec, const_vec;
@@ -2226,7 +2226,8 @@ static void hevc_hz_bi_4t_4x4_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1, src2, src3, dst0, vec0, vec1;
     v8i16 in0, in1, in2, in3;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
+    v16i8 vec2, vec3;
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
     v16i8 mask1;
     v8i16 tmp0, tmp1;
     v8i16 filter_vec, const_vec;
@@ -2247,12 +2248,12 @@ static void hevc_hz_bi_4t_4x4_msa(uint8_t *src0_ptr,
     ILVR_D2_SH(in1, in0, in3, in2, in0, in1);
     XORI_B4_128_SB(src0, src1, src2, src3);
 
-    VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
     tmp0 = const_vec;
-    DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp0, tmp0);
-    VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec0, vec1);
     tmp1 = const_vec;
-    DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp1, tmp1);
+    VSHF_B2_SB(src0, src1, src2, src3, mask0, mask0, vec0, vec1);
+    VSHF_B2_SB(src0, src1, src2, src3, mask1, mask1, vec2, vec3);
+    DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt1, filt1, tmp0, tmp1,
+                 tmp0, tmp1);
     HEVC_BI_RND_CLIP2(in0, in1, tmp0, tmp1, 7, tmp0, tmp1);
     dst0 = __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0);
 
@@ -2273,8 +2274,8 @@ static void hevc_hz_bi_4t_4x8multiple_msa(uint8_t *src0_ptr,
     v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
     v16i8 dst0, dst1;
     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 };
-    v16i8 mask1, vec0, vec1;
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
+    v16i8 mask1, vec0, vec1, vec2, vec3;
     v8i16 tmp0, tmp1, tmp2, tmp3;
     v8i16 filter_vec, const_vec;
 
@@ -2300,18 +2301,18 @@ static void hevc_hz_bi_4t_4x8multiple_msa(uint8_t *src0_ptr,
         ILVR_D2_SH(in5, in4, in7, in6, in2, in3);
         XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
 
-        VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
         tmp0 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp0, tmp0);
-        VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec0, vec1);
         tmp1 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp1, tmp1);
-        VSHF_B2_SB(src4, src5, src4, src5, mask0, mask1, vec0, vec1);
         tmp2 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp2, tmp2);
-        VSHF_B2_SB(src6, src7, src6, src7, mask0, mask1, vec0, vec1);
         tmp3 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp3, tmp3);
+        VSHF_B2_SB(src0, src1, src2, src3, mask0, mask0, vec0, vec1);
+        VSHF_B2_SB(src4, src5, src6, src7, mask0, mask0, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0,
+                     tmp1, tmp2, tmp3);
+        VSHF_B2_SB(src0, src1, src2, src3, mask1, mask1, vec0, vec1);
+        VSHF_B2_SB(src4, src5, src6, src7, mask1, mask1, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, tmp0,
+                     tmp1, tmp2, tmp3);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3);
@@ -2357,9 +2358,9 @@ static void hevc_hz_bi_4t_6w_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1, src2, src3;
     v8i16 in0, in1, in2, in3;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
     v16i8 mask1;
-    v16i8 vec0, vec1;
+    v16i8 vec0, vec1, vec2, vec3;
     v8i16 dst0, dst1, dst2, dst3;
     v8i16 filter_vec, const_vec;
 
@@ -2380,18 +2381,18 @@ static void hevc_hz_bi_4t_6w_msa(uint8_t *src0_ptr,
         src1_ptr += (4 * src2_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, src0, src1, src1, mask0, mask0, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -2414,8 +2415,8 @@ static void hevc_hz_bi_4t_8x2_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1;
     v8i16 in0, in1;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
-    v16i8 mask1, vec0, vec1;
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
+    v16i8 mask1, vec0, vec1, vec2, vec3;
     v8i16 dst0, dst1;
     v8i16 filter_vec, const_vec;
 
@@ -2433,12 +2434,12 @@ static void hevc_hz_bi_4t_8x2_msa(uint8_t *src0_ptr,
     LD_SH2(src1_ptr, src2_stride, in0, in1);
     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);
+    VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0, vec1);
+    VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec2, vec3);
+    DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt1, filt1, dst0, dst1,
+                 dst0, dst1);
     HEVC_BI_RND_CLIP2(in0, in1, dst0, dst1, 7, dst0, dst1);
 
     dst0 = (v8i16) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
@@ -2457,9 +2458,9 @@ static void hevc_hz_bi_4t_8x6_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1, src2, src3, src4, src5;
     v8i16 in0, in1, in2, in3, in4, in5;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
     v16i8 mask1;
-    v16i8 vec0, vec1;
+    v16i8 vec0, vec1, vec2, vec3;
     v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
     v8i16 filter_vec, const_vec;
 
@@ -2479,24 +2480,25 @@ static void hevc_hz_bi_4t_8x6_msa(uint8_t *src0_ptr,
     LD_SH2(src1_ptr, src2_stride, in4, in5);
     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);
+    VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0, vec1);
+    VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3);
+    DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0, dst1,
+                 dst2, dst3);
+    VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1);
+    VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3);
+    DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0, dst1,
+                 dst2, dst3);
     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(src4, src4, src5, src5, mask0, mask0, vec0, vec1);
+    VSHF_B2_SB(src4, src4, src5, src5, mask1, mask1, vec2, vec3);
+    DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt1, filt1, dst4, dst5,
+                 dst4, dst5);
 
     HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                       dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -2522,9 +2524,9 @@ static void hevc_hz_bi_4t_8x4multiple_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1, src2, src3;
     v8i16 in0, in1, in2, in3;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
     v16i8 mask1;
-    v16i8 vec0, vec1;
+    v16i8 vec0, vec1, vec2, vec3;
     v8i16 dst0, dst1, dst2, dst3;
     v8i16 filter_vec, const_vec;
 
@@ -2545,18 +2547,18 @@ static void hevc_hz_bi_4t_8x4multiple_msa(uint8_t *src0_ptr,
         src1_ptr += (4 * src2_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, src0, src1, src1, mask0, mask0, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -2602,12 +2604,12 @@ static void hevc_hz_bi_4t_12w_msa(uint8_t *src0_ptr,
     v8i16 filt0, filt1;
     v16i8 src0, src1, src2, src3;
     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
     v16i8 mask2 = {
         8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
     };
     v16i8 mask1, mask3;
-    v16i8 vec0, vec1;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
     v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
     v8i16 filter_vec, const_vec;
 
@@ -2632,24 +2634,24 @@ static void hevc_hz_bi_4t_12w_msa(uint8_t *src0_ptr,
         ILVR_D2_SH(in5, in4, in7, in6, in4, in5);
         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);
+        VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3);
+        VSHF_B2_SB(src0, src1, src2, src3, mask2, mask2, vec4, vec5);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        DPADD_SB2_SH(vec4, vec5, filt0, filt0, dst4, dst5);
+        VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3);
+        VSHF_B2_SB(src0, src1, src2, src3, mask3, mask3, vec4, vec5);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
+        DPADD_SB2_SH(vec4, vec5, filt1, filt1, dst4, dst5);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -2672,13 +2674,11 @@ static void hevc_hz_bi_4t_16w_msa(uint8_t *src0_ptr,
                                   int32_t height)
 {
     uint32_t loop_cnt;
-    v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
-    v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+    v16i8 src0, src1, src2, src3, vec0, vec1, vec2, vec3;
+    v8i16 in0, in1, in2, in3, dst0, dst1, dst2, dst3;
     v8i16 filt0, filt1;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
     v16i8 mask1;
-    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
-    v16i8 vec0, vec1;
     v8i16 filter_vec, const_vec;
 
     src0_ptr -= 1;
@@ -2691,49 +2691,36 @@ static void hevc_hz_bi_4t_16w_msa(uint8_t *src0_ptr,
 
     mask1 = mask0 + 2;
 
-    for (loop_cnt = (height >> 2); loop_cnt--;) {
-        LD_SB4(src0_ptr, src_stride, src0, src2, src4, src6);
-        LD_SB4(src0_ptr + 8, src_stride, src1, src3, src5, src7);
-        src0_ptr += (4 * src_stride);
-        LD_SH4(src1_ptr, src2_stride, in0, in2, in4, in6);
-        LD_SH4(src1_ptr + 8, src2_stride, in1, in3, in5, in7);
-        src1_ptr += (4 * src2_stride);
-        XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
+    for (loop_cnt = (height >> 1); loop_cnt--;) {
+        LD_SB2(src0_ptr, src_stride, src0, src2);
+        LD_SB2(src0_ptr + 8, src_stride, src1, src3);
+        src0_ptr += (2 * src_stride);
+        LD_SH2(src1_ptr, src2_stride, in0, in2);
+        LD_SH2(src1_ptr + 8, src2_stride, in1, in3);
+        src1_ptr += (2 * src2_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(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);
+
+        VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
-        HEVC_BI_RND_CLIP4(in4, in5, in6, in7,
-                          dst4, dst5, dst6, dst7, 7, dst4, dst5, dst6, dst7);
 
-        PCKEV_B4_SH(dst1, dst0, dst3, dst2,
-                    dst5, dst4, dst7, dst6, dst0, dst1, dst2, dst3);
-        ST_SH4(dst0, dst1, dst2, dst3, dst, dst_stride);
-        dst += (4 * dst_stride);
+        PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1);
+        ST_SH2(dst0, dst1, dst, dst_stride);
+        dst += (2 * dst_stride);
     }
 }
 
@@ -2752,9 +2739,9 @@ static void hevc_hz_bi_4t_24w_msa(uint8_t *src0_ptr,
     v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
     v8i16 filt0, filt1;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
     v16i8 mask1, mask2, mask3;
-    v16i8 vec0, vec1;
+    v16i8 vec0, vec1, vec2, vec3;
     v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
     v8i16 filter_vec, const_vec;
 
@@ -2782,30 +2769,31 @@ static void hevc_hz_bi_4t_24w_msa(uint8_t *src0_ptr,
         src1_ptr += (4 * src2_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(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);
-        VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
+        VSHF_B2_SB(src0, src0, src0, src1, mask0, mask2, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src2, src3, mask0, mask2, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        VSHF_B2_SB(src0, src0, src0, src1, mask1, mask3, vec0, vec1);
+        VSHF_B2_SB(src2, src2, src2, src3, mask1, mask3, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
+
         dst4 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-        VSHF_B2_SB(src4, src5, src4, src5, mask2, mask3, 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(src6, src7, src6, src7, mask2, mask3, vec0, vec1);
         dst7 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7);
+        VSHF_B2_SB(src4, src4, src4, src5, mask0, mask2, vec0, vec1);
+        VSHF_B2_SB(src6, src6, src6, src7, mask0, mask2, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst4,
+                     dst5, dst6, dst7);
+        VSHF_B2_SB(src4, src4, src4, src5, mask1, mask3, vec0, vec1);
+        VSHF_B2_SB(src6, src6, src6, src7, mask1, mask3, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst4,
+                     dst5, dst6, dst7);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -2820,18 +2808,18 @@ static void hevc_hz_bi_4t_24w_msa(uint8_t *src0_ptr,
         LD_SH4(src1_ptr_tmp, src2_stride, in0, in1, in2, in3);
         src1_ptr_tmp += (4 * src2_stride);
 
-        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);
-        VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
         dst2 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
-        VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
         dst3 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+        VSHF_B2_SB(src1, src1, src3, src3, mask0, mask0, vec0, vec1);
+        VSHF_B2_SB(src5, src5, src7, src7, mask0, mask0, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        VSHF_B2_SB(src1, src1, src3, src3, mask1, mask1, vec0, vec1);
+        VSHF_B2_SB(src5, src5, src7, src7, mask1, mask1, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -2855,10 +2843,10 @@ static void hevc_hz_bi_4t_32w_msa(uint8_t *src0_ptr,
     v16i8 src0, src1, src2;
     v8i16 in0, in1, in2, in3;
     v8i16 filt0, filt1;
-    v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+    v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
     v16i8 mask1, mask2, mask3;
     v8i16 dst0, dst1, dst2, dst3;
-    v16i8 vec0, vec1;
+    v16i8 vec0, vec1, vec2, vec3;
     v8i16 filter_vec, const_vec;
 
     src0_ptr -= 1;
@@ -2873,34 +2861,7 @@ static void hevc_hz_bi_4t_32w_msa(uint8_t *src0_ptr,
     mask2 = mask0 + 8;
     mask3 = mask0 + 10;
 
-    for (loop_cnt = (height >> 1); loop_cnt--;) {
-        LD_SB2(src0_ptr, 16, src0, src1);
-        src2 = LD_SB(src0_ptr + 24);
-        src0_ptr += src_stride;
-        LD_SH4(src1_ptr, 8, in0, in1, in2, in3);
-        src1_ptr += src2_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_BI_RND_CLIP4(in0, in1, in2, in3,
-                          dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
-
-        PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1);
-        ST_SH2(dst0, dst1, dst, 16);
-        dst += dst_stride;
-
+    for (loop_cnt = height; loop_cnt--;) {
         LD_SB2(src0_ptr, 16, src0, src1);
         src2 = LD_SB(src0_ptr + 24);
         src0_ptr += src_stride;
@@ -2908,18 +2869,18 @@ static void hevc_hz_bi_4t_32w_msa(uint8_t *src0_ptr,
         src1_ptr += src2_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);
+        VSHF_B2_SB(src0, src0, src0, src1, mask0, mask2, vec0, vec1);
+        VSHF_B2_SB(src1, src1, src2, src2, mask0, mask0, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0,
+                     dst1, dst2, dst3);
+        VSHF_B2_SB(src0, src0, src0, src1, mask1, mask3, vec0, vec1);
+        VSHF_B2_SB(src1, src1, src2, src2, mask1, mask1, vec2, vec3);
+        DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0,
+                     dst1, dst2, dst3);
 
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3);
@@ -3131,10 +3092,9 @@ static void hevc_vt_bi_4t_6w_msa(uint8_t *src0_ptr,
                                  const int8_t *filter,
                                  int32_t height)
 {
-    int32_t loop_cnt;
-    v16i8 src0, src1, src2, src3, src4, src5;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
     v8i16 in0, in1, in2, in3;
-    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v16i8 src10_r, src32_r, src21_r, src43_r, src54_r, src65_r;
     v8i16 dst0_r, dst1_r, dst2_r, dst3_r;
     v8i16 filt0, filt1;
     v8i16 filter_vec, const_vec;
@@ -3149,40 +3109,70 @@ static void hevc_vt_bi_4t_6w_msa(uint8_t *src0_ptr,
 
     LD_SB3(src0_ptr, src_stride, src0, src1, src2);
     src0_ptr += (3 * src_stride);
+    LD_SB2(src0_ptr, src_stride, src3, src4);
+    src0_ptr += (2 * src_stride);
+    LD_SB2(src0_ptr, src_stride, src5, src6);
+    src0_ptr += (2 * src_stride);
+    LD_SB2(src0_ptr, src_stride, src7, src8);
+    src0_ptr += (2 * src_stride);
+    LD_SB2(src0_ptr, src_stride, src9, src10);
+    src0_ptr += (2 * src_stride);
+
+    LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
+    src1_ptr += (4 * src2_stride);
+
     XORI_B3_128_SB(src0, src1, src2);
+    XORI_B2_128_SB(src3, src4);
+    XORI_B2_128_SB(src5, src6);
+    XORI_B2_128_SB(src7, src8);
+    XORI_B2_128_SB(src9, src10);
+
     ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
+    ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
 
-    for (loop_cnt = (height >> 2); loop_cnt--;) {
-        LD_SB2(src0_ptr, src_stride, src3, src4);
-        src0_ptr += (2 * src_stride);
-        LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
-        src1_ptr += (4 * src2_stride);
-        XORI_B2_128_SB(src3, src4);
-        ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
+    dst0_r = const_vec;
+    DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r);
+    dst1_r = const_vec;
+    DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r);
 
-        dst0_r = const_vec;
-        DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r);
-        dst1_r = const_vec;
-        DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r);
+    ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r);
 
-        LD_SB2(src0_ptr, src_stride, src5, src2);
-        src0_ptr += (2 * src_stride);
-        XORI_B2_128_SB(src5, src2);
-        ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r);
+    dst2_r = const_vec;
+    DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, dst2_r, dst2_r);
+    dst3_r = const_vec;
+    DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, dst3_r, dst3_r);
 
-        dst2_r = const_vec;
-        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst2_r, dst2_r);
-        dst3_r = const_vec;
-        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst3_r, dst3_r);
+    HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
+                      dst0_r, dst1_r, dst2_r, dst3_r, 7,
+                      dst0_r, dst1_r, dst2_r, dst3_r);
 
-        HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
-                          dst0_r, dst1_r, dst2_r, dst3_r, 7,
-                          dst0_r, dst1_r, dst2_r, dst3_r);
+    PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r);
+    ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+    dst += (4 * dst_stride);
 
-        PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r);
-        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
-        dst += (4 * dst_stride);
-    }
+    LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
+    src1_ptr += (4 * src2_stride);
+    ILVR_B2_SB(src7, src6, src8, src7, src32_r, src43_r);
+
+    dst0_r = const_vec;
+    DPADD_SB2_SH(src54_r, src32_r, filt0, filt1, dst0_r, dst0_r);
+    dst1_r = const_vec;
+    DPADD_SB2_SH(src65_r, src43_r, filt0, filt1, dst1_r, dst1_r);
+
+    ILVR_B2_SB(src9, src8, src10, src9, src54_r, src65_r);
+
+    dst2_r = const_vec;
+    DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, dst2_r, dst2_r);
+    dst3_r = const_vec;
+    DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, dst3_r, dst3_r);
+
+    HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
+                      dst0_r, dst1_r, dst2_r, dst3_r, 7,
+                      dst0_r, dst1_r, dst2_r, dst3_r);
+
+    PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r);
+    ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
+    dst += (4 * dst_stride);
 }
 
 static void hevc_vt_bi_4t_8x2_msa(uint8_t *src0_ptr,
@@ -3384,12 +3374,12 @@ static void hevc_vt_bi_4t_12w_msa(uint8_t *src0_ptr,
                                   int32_t height)
 {
     int32_t loop_cnt;
-    v16i8 src0, src1, src2, src3, src4, src5;
+    v16i8 src0, src1, src2, src3, src4, src5, src6;
     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
-    v16i8 src10_r, src32_r, src21_r, src43_r;
+    v16i8 src10_r, src32_r, src21_r, src43_r, src54_r, src65_r;
     v8i16 dst0_r, dst1_r, dst2_r, dst3_r;
     v16i8 src10_l, src32_l, src54_l, src21_l, src43_l, src65_l;
-    v16i8 src2110, src4332;
+    v16i8 src2110, src4332, src6554;
     v8i16 dst0_l, dst1_l, filt0, filt1;
     v8i16 filter_vec, const_vec;
 
@@ -3411,15 +3401,21 @@ static void hevc_vt_bi_4t_12w_msa(uint8_t *src0_ptr,
     for (loop_cnt = (height >> 2); loop_cnt--;) {
         LD_SB2(src0_ptr, src_stride, src3, src4);
         src0_ptr += (2 * src_stride);
+        LD_SB2(src0_ptr, src_stride, src5, src6);
+        src0_ptr += (2 * src_stride);
         LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
         LD_SH4((src1_ptr + 8), src2_stride, in4, in5, in6, in7);
         src1_ptr += (4 * src2_stride);
         ILVR_D2_SH(in5, in4, in7, in6, in4, in5);
         XORI_B2_128_SB(src3, src4);
+        XORI_B2_128_SB(src5, src6);
 
         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);
+        ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r);
+        ILVL_B2_SB(src5, src4, src6, src5, src54_l, src65_l);
+        src6554 = (v16i8) __msa_ilvr_d((v2i64) src65_l, (v2i64) src54_l);
 
         dst0_r = const_vec;
         DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r);
@@ -3427,21 +3423,12 @@ static void hevc_vt_bi_4t_12w_msa(uint8_t *src0_ptr,
         DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r);
         dst0_l = const_vec;
         DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst0_l, dst0_l);
-
-        LD_SB2(src0_ptr, src_stride, src5, src2);
-        src0_ptr += (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);
-
         dst2_r = const_vec;
-        DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst2_r, dst2_r);
+        DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, dst2_r, dst2_r);
         dst3_r = const_vec;
-        DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst3_r, dst3_r);
+        DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, dst3_r, dst3_r);
         dst1_l = const_vec;
-        DPADD_SB2_SH(src4332, src2110, filt0, filt1, dst1_l, dst1_l);
+        DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst1_l, dst1_l);
         HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
                           dst0_r, dst1_r, dst2_r, dst3_r, 7,
                           dst0_r, dst1_r, dst2_r, dst3_r);
@@ -3451,6 +3438,11 @@ static void hevc_vt_bi_4t_12w_msa(uint8_t *src0_ptr,
         dst0_l = (v8i16) __msa_pckev_b((v16i8) dst1_l, (v16i8) dst0_l);
         ST12x4_UB(dst0_r, dst1_r, dst0_l, dst, dst_stride);
         dst += (4 * dst_stride);
+
+        src2 = src6;
+        src10_r = src54_r;
+        src21_r = src65_r;
+        src2110 = src6554;
     }
 }
 
@@ -3780,20 +3772,20 @@ static void hevc_hv_bi_4t_4x2_msa(uint8_t *src0_ptr,
                                   uint8_t *dst,
                                   int32_t dst_stride,
                                   const int8_t *filter_x,
-                                  const int8_t *filter_y,
-                                  int32_t height)
+                                  const int8_t *filter_y)
 {
-    v8i16 in0, in1;
+    uint64_t tp0, tp1;
+    v16u8 out;
+    v8i16 in0 = { 0 };
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
     v16i8 mask1;
     v8i16 filter_vec, const_vec;
     v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
-    v8i16 dst0, dst1, dst2, dst3, dst4;
-    v4i32 dst0_r, dst1_r;
-    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v8i16 dst20, dst31, dst42, dst10, dst32, dst21, dst43, tmp;
+    v4i32 dst0, dst1;
 
     src0_ptr -= (src_stride + 1);
 
@@ -3801,57 +3793,44 @@ static void hevc_hv_bi_4t_4x2_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
     const_vec = __msa_ldi_h(128);
     const_vec <<= 6;
 
-    LD_SB3(src0_ptr, src_stride, src0, src1, src2);
-    src0_ptr += (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(src0_ptr, src_stride, src3, src4);
-    LD_SH2(src1_ptr, src2_stride, in0, in1);
-    in0 = (v8i16) __msa_ilvr_d((v2i64) in1, (v2i64) in0);
-    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;
-    /* 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;
-    dst0_r = (v4i32) __msa_pckev_h((v8i16) dst1_r, (v8i16) dst0_r);
-    dst0_r = (v4i32) __msa_adds_s_h((v8i16) dst0_r, in0);
-    dst0_r = (v4i32) __msa_srari_h((v8i16) dst0_r, 7);
-    dst0_r = (v4i32) CLIP_SH_0_255(dst0_r);
-
-    dst0_r = (v4i32) __msa_pckev_b((v16i8) dst0_r, (v16i8) dst0_r);
-    ST4x2_UB(dst0_r, dst, dst_stride);
-}
+    LD_SB5(src0_ptr, src_stride, src0, src1, src2, src3, src4);
+    XORI_B5_128_SB(src0, src1, src2, src3, src4);
+
+    LD2(src1_ptr, src2_stride, tp0, tp1);
+    INSERT_D2_SH(tp0, tp1, in0);
+    in0 = __msa_adds_s_h(in0, const_vec);
+
+    VSHF_B2_SB(src0, src2, src0, src2, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src3, src1, src3, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src4, src2, src4, mask0, mask1, vec4, vec5);
+
+    dst20 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dst31 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dst42 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+
+    ILVRL_H2_SH(dst31, dst20, dst10, dst32);
+    ILVRL_H2_SH(dst42, dst31, dst21, dst43);
+
+    dst0 = HEVC_FILT_4TAP(dst10, dst32, filt_h0, filt_h1);
+    dst1 = HEVC_FILT_4TAP(dst21, dst43, filt_h0, filt_h1);
+    dst0 >>= 6;
+    dst1 >>= 6;
+    tmp = __msa_pckev_h((v8i16) dst1, (v8i16) dst0);
+    tmp = __msa_adds_s_h(tmp, in0);
+    tmp = __msa_srari_h(tmp, 7);
+    tmp = CLIP_SH_0_255_MAX_SATU(tmp);
+    out = (v16u8) __msa_pckev_b((v16i8) tmp, (v16i8) tmp);
+    ST4x2_UB(out, dst, dst_stride);
+}
 
 static void hevc_hv_bi_4t_4x4_msa(uint8_t *src0_ptr,
                                   int32_t src_stride,
@@ -3860,21 +3839,22 @@ static void hevc_hv_bi_4t_4x4_msa(uint8_t *src0_ptr,
                                   uint8_t *dst,
                                   int32_t dst_stride,
                                   const int8_t *filter_x,
-                                  const int8_t *filter_y,
-                                  int32_t height)
+                                  const int8_t *filter_y)
 {
-    v8i16 in0, in1, in2, in3;
+    uint64_t tp0, tp1;
+    v16u8 out;
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
     v16i8 mask1;
     v8i16 filter_vec, const_vec;
-    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
-    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
-    v8i16 dst0_r, dst1_r;
-    v4i32 tmp0, tmp1, tmp2, tmp3;
-    v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+    v8i16 tmp0, tmp1;
+    v8i16 in0 = { 0 }, in1 = { 0 };
+    v8i16 dst30, dst41, dst52, dst63;
+    v8i16 dst10, dst32, dst54, dst21, dst43, dst65;
+    v4i32 dst0, dst1, dst2, dst3;
 
     src0_ptr -= (src_stride + 1);
 
@@ -3882,69 +3862,50 @@ static void hevc_hv_bi_4t_4x4_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
+    LD_SB7(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5, src6);
+    XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
     const_vec = __msa_ldi_h(128);
     const_vec <<= 6;
 
-    LD_SB3(src0_ptr, src_stride, src0, src1, src2);
-    src0_ptr += (3 * src_stride);
-    XORI_B3_128_SB(src0, src1, src2);
+    LD2(src1_ptr, src2_stride, tp0, tp1);
+    src1_ptr += 2 * src2_stride;
+    INSERT_D2_SH(tp0, tp1, in0);
+    LD2(src1_ptr, src2_stride, tp0, tp1);
+    INSERT_D2_SH(tp0, tp1, in1);
 
-    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);
+    ADDS_SH2_SH(in0, const_vec, in1, const_vec, in0, in1);
 
-    LD_SB4(src0_ptr, src_stride, src3, src4, src5, src6);
-    LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
-    ILVR_D2_SH(in1, in0, in3, in2, in0, in1);
-    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);
-    tmp0 = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
-    tmp0 >>= 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);
-    tmp1 = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
-    tmp1 >>= 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);
-    tmp2 = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
-    tmp2 >>= 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);
-    tmp3 = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
-    tmp3 >>= 6;
-    PCKEV_H2_SH(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r);
-    HEVC_BI_RND_CLIP2(in0, in1, dst0_r, dst1_r, 7, dst0_r, dst1_r);
+    VSHF_B2_SB(src0, src3, src0, src3, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src4, src1, src4, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src2, src5, src2, src5, mask0, mask1, vec4, vec5);
+    VSHF_B2_SB(src3, src6, src3, src6, mask0, mask1, vec6, vec7);
 
-    dst0_r = (v8i16) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst0_r);
-    ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride);
-    dst += (4 * dst_stride);
+    dst30 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dst41 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dst52 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+    dst63 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+    ILVRL_H2_SH(dst41, dst30, dst10, dst43);
+    ILVRL_H2_SH(dst52, dst41, dst21, dst54);
+    ILVRL_H2_SH(dst63, dst52, dst32, dst65);
+    dst0 = HEVC_FILT_4TAP(dst10, dst32, filt_h0, filt_h1);
+    dst1 = HEVC_FILT_4TAP(dst21, dst43, filt_h0, filt_h1);
+    dst2 = HEVC_FILT_4TAP(dst32, dst54, filt_h0, filt_h1);
+    dst3 = HEVC_FILT_4TAP(dst43, dst65, filt_h0, filt_h1);
+    SRA_4V(dst0, dst1, dst2, dst3, 6);
+    PCKEV_H2_SH(dst1, dst0, dst3, dst2, tmp0, tmp1);
+    ADDS_SH2_SH(tmp0, in0, tmp1, in1, tmp0, tmp1);
+    SRARI_H2_SH(tmp0, tmp1, 7);
+    CLIP_SH2_0_255_MAX_SATU(tmp0, tmp1);
+    out = (v16u8) __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0);
+    ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
 }
 
 static void hevc_hv_bi_4t_4multx8mult_msa(uint8_t *src0_ptr,
@@ -3958,18 +3919,21 @@ static void hevc_hv_bi_4t_4multx8mult_msa(uint8_t *src0_ptr,
                                           int32_t height)
 {
     uint32_t loop_cnt;
-    v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+    uint64_t tp0, tp1;
+    v16u8 out0, out1;
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
     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;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
     v8i16 tmp0, tmp1, tmp2, tmp3;
+    v8i16 dst10, dst21, dst22, dst73, dst84, dst95, dst106;
     v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
     v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
+    v8i16 dst98_r, dst109_r;
+    v8i16 in0 = { 0 }, in1 = { 0 }, in2 = { 0 }, in3 = { 0 };
     v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r;
 
     src0_ptr -= (src_stride + 1);
@@ -3978,10 +3942,9 @@ static void hevc_hv_bi_4t_4multx8mult_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
@@ -3992,90 +3955,75 @@ static void hevc_hv_bi_4t_4multx8mult_msa(uint8_t *src0_ptr,
     src0_ptr += (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);
+    VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src1, src2, src1, src2, mask0, mask1, vec2, vec3);
+    dst10 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dst21 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    ILVRL_H2_SH(dst21, dst10, dst10_r, dst21_r);
+    dst22 = (v8i16) __msa_splati_d((v2i64) dst21, 1);
+
 
     for (loop_cnt = height >> 3; loop_cnt--;) {
         LD_SB8(src0_ptr, src_stride,
                src3, src4, src5, src6, src7, src8, src9, src10);
         src0_ptr += (8 * src_stride);
-        LD_SH8(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5, in6, in7);
-        src1_ptr += (8 * src2_stride);
-        ILVR_D2_SH(in1, in0, in3, in2, in0, in1);
-        ILVR_D2_SH(in5, in4, in7, in6, in2, in3);
         XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
-        /* 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);
+        VSHF_B2_SB(src3, src7, src3, src7, mask0, mask1, vec0, vec1);
+        VSHF_B2_SB(src4, src8, src4, src8, mask0, mask1, vec2, vec3);
+        VSHF_B2_SB(src5, src9, src5, src9, mask0, mask1, vec4, vec5);
+        VSHF_B2_SB(src6, src10, src6, src10, mask0, mask1, vec6, vec7);
+
+        dst73 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+        dst84 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+        dst95 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+        dst106 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+        dst32_r = __msa_ilvr_h(dst73, dst22);
+        ILVRL_H2_SH(dst84, dst73, dst43_r, dst87_r);
+        ILVRL_H2_SH(dst95, dst84, dst54_r, dst98_r);
+        ILVRL_H2_SH(dst106, dst95, dst65_r, dst109_r);
+        dst22 = (v8i16) __msa_splati_d((v2i64) dst73, 1);
+        dst76_r = __msa_ilvr_h(dst22, dst106);
+
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in0);
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in1);
+
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in2);
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in3);
+
+        ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3,
+                    const_vec, in0, in1, in2, in3);
         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);
-        dst54_r = __msa_ilvr_h(dst5, dst4);
         dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
-        dst2_r >>= 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);
-        dst65_r = __msa_ilvr_h(dst6, dst5);
         dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
-        dst3_r >>= 6;
-        /* 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);
-        dst76_r = __msa_ilvr_h(dst7, dst6);
         dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
-        dst4_r >>= 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);
-        dst87_r = __msa_ilvr_h(dst8, dst7);
         dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
-        dst5_r >>= 6;
-        /* row 9 */
-        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;
-        /* row 10 */
-        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;
+        dst6_r = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
+        dst7_r = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
+        SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
+        SRA_4V(dst4_r, dst5_r, dst6_r, dst7_r, 6);
         PCKEV_H4_SH(dst1_r, dst0_r, dst3_r, dst2_r,
                     dst5_r, dst4_r, dst7_r, dst6_r, tmp0, tmp1, tmp2, tmp3);
-        HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
-                          tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3);
-
-        PCKEV_B2_SH(tmp1, tmp0, tmp3, tmp2, tmp0, tmp1);
-        ST4x8_UB(tmp0, tmp1, dst, dst_stride);
+        ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3, tmp0, tmp1,
+                    tmp2, tmp3);
+        SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+        CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+        PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+        ST4x8_UB(out0, out1, dst, dst_stride);
         dst += (8 * dst_stride);
+
+        dst10_r = dst98_r;
+        dst21_r = dst109_r;
+        dst22 = (v8i16) __msa_splati_d((v2i64) dst106, 1);
     }
 }
 
@@ -4091,10 +4039,10 @@ static void hevc_hv_bi_4t_4w_msa(uint8_t *src0_ptr,
 {
     if (2 == height) {
         hevc_hv_bi_4t_4x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
-                              dst, dst_stride, filter_x, filter_y, height);
+                              dst, dst_stride, filter_x, filter_y);
     } else if (4 == height) {
         hevc_hv_bi_4t_4x4_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
-                              dst, dst_stride, filter_x, filter_y, height);
+                              dst, dst_stride, filter_x, filter_y);
     } else if (0 == (height % 8)) {
         hevc_hv_bi_4t_4multx8mult_msa(src0_ptr, src_stride,
                                       src1_ptr, src2_stride,
@@ -4113,20 +4061,28 @@ static void hevc_hv_bi_4t_6w_msa(uint8_t *src0_ptr,
                                  const int8_t *filter_y,
                                  int32_t height)
 {
-    uint32_t loop_cnt;
-    v16i8 src0, src1, src2, src3, src4, src5, src6;
-    v8i16 in0, in1, in2, in3;
+    uint32_t tpw0, tpw1, tpw2, tpw3;
+    uint64_t tp0, tp1;
+    v16u8 out0, out1, out2;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
     v16i8 mask1;
     v8i16 filter_vec, const_vec;
-    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
-    v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+    v8i16 dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6, dsth7, dsth8, dsth9;
+    v8i16 dsth10, tmp4, tmp5;
     v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+    v4i32 dst4_r, dst5_r, dst6_r, dst7_r;
     v8i16 tmp0, tmp1, tmp2, tmp3;
     v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
     v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
+    v8i16 dst54_r, dst76_r, dst98_r, dst65_r, dst87_r, dst109_r;
+    v8i16 dst54_l, dst76_l, dst98_l, dst65_l, dst87_l, dst109_l;
+    v8i16 dst1021_l, dst3243_l, dst5465_l, dst7687_l, dst98109_l;
+    v8i16 in0 = { 0 }, in1 = { 0 }, in2 = { 0 }, in3 = { 0 };
+    v8i16 in4 = { 0 }, in5 = { 0 };
 
     src0_ptr -= (src_stride + 1);
 
@@ -4134,10 +4090,9 @@ static void hevc_hv_bi_4t_6w_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
@@ -4151,71 +4106,101 @@ static void hevc_hv_bi_4t_6w_msa(uint8_t *src0_ptr,
     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);
+    dsth0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dsth1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dsth2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
 
-    for (loop_cnt = height >> 2; loop_cnt--;) {
-        LD_SB4(src0_ptr, src_stride, src3, src4, src5, src6);
-        src0_ptr += (4 * src_stride);
-        LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
-        src1_ptr += (4 * src2_stride);
-        XORI_B4_128_SB(src3, src4, src5, src6);
+    ILVRL_H2_SH(dsth1, dsth0, dst10_r, dst10_l);
+    ILVRL_H2_SH(dsth2, dsth1, dst21_r, dst21_l);
 
-        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
-        dst3 = const_vec;
-        DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+    LD_SB8(src0_ptr, src_stride,
+           src3, src4, src5, src6, src7, src8, src9, src10);
+    XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
 
-        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);
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+    VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
 
-        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;
+    dsth3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dsth4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dsth5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+    dsth6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
 
-        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;
-        PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r,
-                    dst2_l, dst2_r, dst3_l, dst3_r, tmp0, tmp1, tmp2, tmp3);
-        HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
-                          tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3);
+    VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
+    VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec2, vec3);
+    VSHF_B2_SB(src9, src9, src9, src9, mask0, mask1, vec4, vec5);
+    VSHF_B2_SB(src10, src10, src10, src10, mask0, mask1, vec6, vec7);
+
+    dsth7 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dsth8 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dsth9 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+    dsth10 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+    ILVRL_H2_SH(dsth3, dsth2, dst32_r, dst32_l);
+    ILVRL_H2_SH(dsth4, dsth3, dst43_r, dst43_l);
+    ILVRL_H2_SH(dsth5, dsth4, dst54_r, dst54_l);
+    ILVRL_H2_SH(dsth6, dsth5, dst65_r, dst65_l);
+    ILVRL_H2_SH(dsth7, dsth6, dst76_r, dst76_l);
+    ILVRL_H2_SH(dsth8, dsth7, dst87_r, dst87_l);
+    ILVRL_H2_SH(dsth9, dsth8, dst98_r, dst98_l);
+    ILVRL_H2_SH(dsth10, dsth9, dst109_r, dst109_l);
+    PCKEV_D2_SH(dst21_l, dst10_l, dst43_l, dst32_l, dst1021_l, dst3243_l);
+    PCKEV_D2_SH(dst65_l, dst54_l, dst87_l, dst76_l, dst5465_l, dst7687_l);
+    dst98109_l = (v8i16) __msa_pckev_d((v2i64) dst109_l, (v2i64) dst98_l);
 
-        PCKEV_B2_SW(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r);
-        ST6x4_UB(dst0_r, dst1_r, dst, dst_stride);
-        dst += (4 * dst_stride);
-    }
+    dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+    dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+    dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+    dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+    dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
+    dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
+    dst6_r = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
+    dst7_r = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
+    dst0_l = HEVC_FILT_4TAP(dst1021_l, dst3243_l, filt_h0, filt_h1);
+    dst1_l = HEVC_FILT_4TAP(dst3243_l, dst5465_l, filt_h0, filt_h1);
+    dst2_l = HEVC_FILT_4TAP(dst5465_l, dst7687_l, filt_h0, filt_h1);
+    dst3_l = HEVC_FILT_4TAP(dst7687_l, dst98109_l, filt_h0, filt_h1);
+    SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
+    SRA_4V(dst4_r, dst5_r, dst6_r, dst7_r, 6);
+    SRA_4V(dst0_l, dst1_l, dst2_l, dst3_l, 6);
+    PCKEV_H2_SH(dst1_r, dst0_r, dst3_r, dst2_r, tmp0, tmp1);
+    PCKEV_H2_SH(dst5_r, dst4_r, dst7_r, dst6_r, tmp2, tmp3);
+    PCKEV_H2_SH(dst1_l, dst0_l, dst3_l, dst2_l, tmp4, tmp5);
+
+    LD2(src1_ptr, src2_stride, tp0, tp1);
+    INSERT_D2_SH(tp0, tp1, in0);
+    LD2(src1_ptr + 2 * src2_stride, src2_stride, tp0, tp1);
+    INSERT_D2_SH(tp0, tp1, in1);
+
+    LD2(src1_ptr + 4 * src2_stride, src2_stride, tp0, tp1);
+    INSERT_D2_SH(tp0, tp1, in2);
+    LD2(src1_ptr + 6 * src2_stride, src2_stride, tp0, tp1);
+    INSERT_D2_SH(tp0, tp1, in3);
+
+    ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3, const_vec,
+                in0, in1, in2, in3);
+    ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3, tmp0, tmp1, tmp2,
+                tmp3);
+    SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+    CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+    PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+    ST4x8_UB(out0, out1, dst, dst_stride);
+
+    LW4(src1_ptr + 4, src2_stride, tpw0, tpw1, tpw2, tpw3);
+    src1_ptr += (4 * src2_stride);
+    INSERT_W4_SH(tpw0, tpw1, tpw2, tpw3, in4);
+    LW4(src1_ptr + 4, src2_stride, tpw0, tpw1, tpw2, tpw3);
+    INSERT_W4_SH(tpw0, tpw1, tpw2, tpw3, in5);
+    ADDS_SH2_SH(in4, const_vec, in5, const_vec, in4, in5);
+    ADDS_SH2_SH(in4, tmp4, in5, tmp5, tmp4, tmp5);
+    SRARI_H2_SH(tmp4, tmp5, 7);
+    CLIP_SH2_0_255_MAX_SATU(tmp4, tmp5);
+    out2 = (v16u8) __msa_pckev_b((v16i8) tmp5, (v16i8) tmp4);
+    ST2x4_UB(out2, 0, dst + 4, dst_stride);
+    dst += 4 * dst_stride;
+    ST2x4_UB(out2, 4, dst + 4, dst_stride);
 }
 
 static void hevc_hv_bi_4t_8x2_msa(uint8_t *src0_ptr,
@@ -4225,16 +4210,16 @@ static void hevc_hv_bi_4t_8x2_msa(uint8_t *src0_ptr,
                                   uint8_t *dst,
                                   int32_t dst_stride,
                                   const int8_t *filter_x,
-                                  const int8_t *filter_y,
-                                  int32_t height)
+                                  const int8_t *filter_y)
 {
+    v16u8 out;
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
     v16i8 mask1;
     v8i16 filter_vec, const_vec;
-    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9;
     v8i16 dst0, dst1, dst2, dst3, dst4;
     v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
     v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
@@ -4248,62 +4233,144 @@ static void hevc_hv_bi_4t_8x2_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
     const_vec = __msa_ldi_h(128);
     const_vec <<= 6;
 
-    LD_SB3(src0_ptr, src_stride, src0, src1, src2);
-    src0_ptr += (3 * src_stride);
-    XORI_B3_128_SB(src0, src1, src2);
+    LD_SB5(src0_ptr, src_stride, src0, src1, src2, src3, src4);
+    XORI_B5_128_SB(src0, src1, src2, src3, src4);
+
+    LD_SH2(src1_ptr, src2_stride, in0, in1);
+    ADDS_SH2_SH(in0, const_vec, in1, const_vec, in0, in1);
 
     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);
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
+
+    dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+    dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+    dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
 
     ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
     ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
-
-    LD_SB2(src0_ptr, src_stride, src3, src4);
-    LD_SH2(src1_ptr, src2_stride, in0, in1);
-    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);
+    ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_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;
-
+    SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
     PCKEV_H2_SH(dst0_l, dst0_r, dst1_l, dst1_r, tmp0, tmp1);
-    HEVC_BI_RND_CLIP2(in0, in1, tmp0, tmp1, 7, tmp0, tmp1);
+    ADDS_SH2_SH(in0, tmp0, in1, tmp1, tmp0, tmp1);
+    SRARI_H2_SH(tmp0, tmp1, 7);
+    CLIP_SH2_0_255_MAX_SATU(tmp0, tmp1);
+    out = (v16u8) __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0);
+    ST8x2_UB(out, dst, dst_stride);
+}
 
-    dst0_r = (v4i32) __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0);
-    ST8x2_UB(dst0_r, dst, dst_stride);
+static void hevc_hv_bi_4t_8multx4_msa(uint8_t *src0_ptr,
+                                      int32_t src_stride,
+                                      int16_t *src1_ptr,
+                                      int32_t src2_stride,
+                                      uint8_t *dst,
+                                      int32_t dst_stride,
+                                      const int8_t *filter_x,
+                                      const int8_t *filter_y,
+                                      int32_t width8mult)
+{
+    uint32_t cnt;
+    v16u8 out0, out1;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, mask0, mask1;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+    v8i16 filt0, filt1, filt_h0, filt_h1, filter_vec, const_vec;
+    v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, tmp0, tmp1, tmp2, tmp3;
+    v8i16 in0, in1, in2, in3;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+    v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
+    v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
+
+    src0_ptr -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
+
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
+
+    mask0 = LD_SB(ff_hevc_mask_arr);
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    for (cnt = width8mult; cnt--;) {
+        LD_SB7(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5, src6);
+        src0_ptr += 8;
+        XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
+        LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3);
+        src1_ptr += 8;
+        ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3,
+                    const_vec, in0, in1, in2, in3);
+
+        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 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+        dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+        dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+
+        ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
+        ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+        VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+        VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
+
+        dst3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+        dst4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+        dst5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+        dst6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+        ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+        ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+        ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+        ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_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);
+        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);
+        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);
+        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);
+
+        SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+        SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+        PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
+                    dst3_r, tmp0, tmp1, tmp2, tmp3);
+        ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3,
+                    tmp0, tmp1, tmp2, tmp3);
+        SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+        CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+        PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+        ST8x4_UB(out0, out1, dst, dst_stride);
+        dst += 8;
+    }
 }
 
 static void hevc_hv_bi_4t_8x6_msa(uint8_t *src0_ptr,
@@ -4313,17 +4380,18 @@ static void hevc_hv_bi_4t_8x6_msa(uint8_t *src0_ptr,
                                   uint8_t *dst,
                                   int32_t dst_stride,
                                   const int8_t *filter_x,
-                                  const int8_t *filter_y,
-                                  int32_t height)
+                                  const int8_t *filter_y)
 {
+    v16u8 out0, out1, out2;
     v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
     v8i16 in0, in1, in2, in3, in4, in5;
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
     v16i8 mask1;
     v8i16 filter_vec, const_vec;
-    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9;
+    v16i8 vec10, vec11, vec12, vec13, vec14, vec15, vec16, vec17;
     v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
     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;
@@ -4339,120 +4407,87 @@ static void hevc_hv_bi_4t_8x6_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
     const_vec = __msa_ldi_h(128);
     const_vec <<= 6;
 
-    LD_SB3(src0_ptr, src_stride, src0, src1, src2);
-    src0_ptr += (3 * src_stride);
-    XORI_B3_128_SB(src0, src1, src2);
+    LD_SB5(src0_ptr, src_stride, src0, src1, src2, src3, src4);
+    src0_ptr += (5 * src_stride);
+    LD_SB4(src0_ptr, src_stride, src5, src6, src7, src8);
+
+    XORI_B5_128_SB(src0, src1, src2, src3, src4);
+    XORI_B4_128_SB(src5, src6, src7, src8);
+
+    LD_SH6(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5);
+    ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3, const_vec,
+                in0, in1, in2, in3);
+    ADDS_SH2_SH(in4, const_vec, in5, const_vec, in4, in5);
+
     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);
+    VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
+    VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
+    VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec10, vec11);
+    VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec12, vec13);
+    VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec14, vec15);
+    VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec16, vec17);
+
+    dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+    dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+    dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
+    dst5 = HEVC_FILT_4TAP_SH(vec10, vec11, filt0, filt1);
+    dst6 = HEVC_FILT_4TAP_SH(vec12, vec13, filt0, filt1);
+    dst7 = HEVC_FILT_4TAP_SH(vec14, vec15, filt0, filt1);
+    dst8 = HEVC_FILT_4TAP_SH(vec16, vec17, filt0, filt1);
 
     ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
     ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
-
-    LD_SB2(src0_ptr, src_stride, src3, src4);
-    src0_ptr += (2 * src_stride);
-    XORI_B2_128_SB(src3, src4);
-    LD_SH6(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5);
-    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);
+    ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+    ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+    ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
+    ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
+    ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_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;
-    tmp0 = __msa_pckev_h((v8i16) dst0_l, (v8i16) dst0_r);
-
-    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;
-    tmp1 = __msa_pckev_h((v8i16) dst1_l, (v8i16) dst1_r);
-
-    LD_SB2(src0_ptr, src_stride, src5, src6);
-    src0_ptr += (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;
-    tmp2 = __msa_pckev_h((v8i16) dst2_l, (v8i16) dst2_r);
-
-    /* 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;
-    tmp3 = __msa_pckev_h((v8i16) dst3_l, (v8i16) dst3_r);
-
-    LD_SB2(src0_ptr, src_stride, src7, src8);
-    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;
-    tmp4 = __msa_pckev_h((v8i16) dst4_l, (v8i16) dst4_r);
-    /* 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;
-    tmp5 = __msa_pckev_h((v8i16) dst5_l, (v8i16) dst5_r);
 
-    HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
-                      tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3);
-    HEVC_BI_RND_CLIP2(in4, in5, tmp4, tmp5, 7, tmp4, tmp5);
-
-    PCKEV_B2_SW(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r);
-    dst2_r = (v4i32) __msa_pckev_b((v16i8) tmp5, (v16i8) tmp4);
-    ST8x4_UB(dst0_r, dst1_r, dst, dst_stride);
+    SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+    SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+    SRA_4V(dst4_r, dst4_l, dst5_r, dst5_l, 6);
+    PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l, dst3_r,
+                tmp0, tmp1, tmp2, tmp3);
+    PCKEV_H2_SH(dst4_l, dst4_r, dst5_l, dst5_r, tmp4, tmp5);
+    ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3,
+                tmp0, tmp1, tmp2, tmp3);
+    ADDS_SH2_SH(in4, tmp4, in5, tmp5, tmp4, tmp5);
+    SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+    SRARI_H2_SH(tmp4, tmp5, 7);
+    CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+    CLIP_SH2_0_255_MAX_SATU(tmp4, tmp5);
+    PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+    out2 = (v16u8) __msa_pckev_b((v16i8) tmp5, (v16i8) tmp4);
+    ST8x4_UB(out0, out1, dst, dst_stride);
     dst += (4 * dst_stride);
-    ST8x2_UB(dst2_r, dst, dst_stride);
+    ST8x2_UB(out2, dst, dst_stride);
 }
 
 static void hevc_hv_bi_4t_8multx4mult_msa(uint8_t *src0_ptr,
@@ -4470,19 +4505,21 @@ static void hevc_hv_bi_4t_8multx4mult_msa(uint8_t *src0_ptr,
     uint8_t *src0_ptr_tmp;
     int16_t *src1_ptr_tmp;
     uint8_t *dst_tmp;
+    v16u8 out0, out1;
     v16i8 src0, src1, src2, src3, src4, src5, src6;
     v8i16 in0, in1, in2, in3;
     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 };
+    v8i16 filt_h0, filt_h1;
+    v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
     v16i8 mask1;
     v8i16 filter_vec, const_vec;
-    v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
     v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
     v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
     v8i16 tmp0, tmp1, tmp2, tmp3;
     v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
     v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
+    v8i16 dst54_r, dst54_l, dst65_r, dst65_l, dst6;
 
     src0_ptr -= (src_stride + 1);
 
@@ -4490,10 +4527,9 @@ static void hevc_hv_bi_4t_8multx4mult_msa(uint8_t *src0_ptr,
     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);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
 
-    SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
 
     mask1 = mask0 + 2;
 
@@ -4512,12 +4548,10 @@ static void hevc_hv_bi_4t_8multx4mult_msa(uint8_t *src0_ptr,
         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);
+
+        dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+        dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+        dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
 
         ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
         ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
@@ -4529,55 +4563,50 @@ static void hevc_hv_bi_4t_8multx4mult_msa(uint8_t *src0_ptr,
             src1_ptr_tmp += (4 * src2_stride);
             XORI_B4_128_SB(src3, src4, src5, src6);
 
+            ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3,
+                        const_vec, in0, in1, in2, in3);
+
             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, vec2, vec3);
+            VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+            VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
+
+            dst3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+            dst4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+            dst5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+            dst6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
 
             ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+            ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+            ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+            ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_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;
-
-            PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r,
-                        dst2_l, dst2_r, dst3_l, dst3_r, tmp0, tmp1, tmp2, tmp3);
-            HEVC_BI_RND_CLIP4(in0, in1, in2, in3,
-                              tmp0, tmp1, tmp2, tmp3, 7,
-                              tmp0, tmp1, tmp2, tmp3);
-
-            PCKEV_B2_SW(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r);
-            ST8x4_UB(dst0_r, dst1_r, dst_tmp, dst_stride);
+            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);
+            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);
+
+            SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+            SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+            PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
+                        dst3_r, tmp0, tmp1, tmp2, tmp3);
+            ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3,
+                        tmp0, tmp1, tmp2, tmp3);
+            SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+            CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+            PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+            ST8x4_UB(out0, out1, dst_tmp, dst_stride);
             dst_tmp += (4 * dst_stride);
+
+            dst10_r = dst54_r;
+            dst10_l = dst54_l;
+            dst21_r = dst65_r;
+            dst21_l = dst65_l;
+            dst2 = dst6;
         }
 
         src0_ptr += 8;
@@ -4598,10 +4627,13 @@ static void hevc_hv_bi_4t_8w_msa(uint8_t *src0_ptr,
 {
     if (2 == height) {
         hevc_hv_bi_4t_8x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
-                              dst, dst_stride, filter_x, filter_y, height);
+                              dst, dst_stride, filter_x, filter_y);
+    } else if (4 == height) {
+        hevc_hv_bi_4t_8multx4_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
+                                  dst, dst_stride, filter_x, filter_y, 1);
     } else if (6 == height) {
         hevc_hv_bi_4t_8x6_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
-                              dst, dst_stride, filter_x, filter_y, height);
+                              dst, dst_stride, filter_x, filter_y);
     } else {
         hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride,
                                       src1_ptr, src2_stride,
@@ -4620,11 +4652,196 @@ static void hevc_hv_bi_4t_12w_msa(uint8_t *src0_ptr,
                                   const int8_t *filter_y,
                                   int32_t height)
 {
-    hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
-                                  dst, dst_stride, filter_x, filter_y,
-                                  height, 8);
-    hevc_hv_bi_4t_4w_msa(src0_ptr + 8, src_stride, src1_ptr + 8, src2_stride,
-                         dst + 8, dst_stride, filter_x, filter_y, height);
+    uint32_t loop_cnt;
+    uint64_t tp0, tp1;
+    uint8_t *src0_ptr_tmp, *dst_tmp;
+    int16_t *src1_ptr_tmp;
+    v16u8 out0, out1;
+    v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+    v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+    v16i8 mask0, mask1, mask2, mask3;
+    v8i16 filt0, filt1, filt_h0, filt_h1, filter_vec, tmp0, tmp1, tmp2, tmp3;
+    v8i16 dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6, const_vec;
+    v8i16 dst10, dst21, dst22, dst73, dst84, dst95, dst106;
+    v8i16 dst76_r, dst98_r, dst87_r, dst109_r;
+    v8i16 in0 = { 0 }, in1 = { 0 }, in2 = { 0 }, in3 = { 0 };
+    v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
+    v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
+    v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+    v4i32 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+
+    src0_ptr -= (src_stride + 1);
+
+    filter_vec = LD_SH(filter_x);
+    SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+    filter_vec = LD_SH(filter_y);
+    UNPCK_R_SB_SH(filter_vec, filter_vec);
+
+    SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
+
+    mask0 = LD_SB(ff_hevc_mask_arr);
+    mask1 = mask0 + 2;
+
+    const_vec = __msa_ldi_h(128);
+    const_vec <<= 6;
+
+    src0_ptr_tmp = src0_ptr;
+    dst_tmp = dst;
+    src1_ptr_tmp = src1_ptr;
+
+    LD_SB3(src0_ptr_tmp, src_stride, src0, src1, src2);
+    src0_ptr_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);
+
+    dsth0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dsth1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+    dsth2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+
+    ILVRL_H2_SH(dsth1, dsth0, dst10_r, dst10_l);
+    ILVRL_H2_SH(dsth2, dsth1, dst21_r, dst21_l);
+
+    for (loop_cnt = 4; loop_cnt--;) {
+        LD_SB4(src0_ptr_tmp, src_stride, src3, src4, src5, src6);
+        src0_ptr_tmp += (4 * src_stride);
+        XORI_B4_128_SB(src3, src4, src5, src6);
+
+        LD_SH4(src1_ptr_tmp, src2_stride, in0, in1, in2, in3);
+        src1_ptr_tmp += (4 * src2_stride);
+        ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3,
+                    const_vec, in0, in1, in2, in3);
+
+        VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+        VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+        VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+        VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
+
+        dsth3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+        dsth4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+        dsth5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+        dsth6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+        ILVRL_H2_SH(dsth3, dsth2, dst32_r, dst32_l);
+        ILVRL_H2_SH(dsth4, dsth3, dst43_r, dst43_l);
+        ILVRL_H2_SH(dsth5, dsth4, dst54_r, dst54_l);
+        ILVRL_H2_SH(dsth6, dsth5, dst65_r, dst65_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);
+        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);
+        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);
+        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);
+
+        SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+        SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+        PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
+                    dst3_r, tmp0, tmp1, tmp2, tmp3);
+        ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3,
+                    tmp0, tmp1, tmp2, tmp3);
+        SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+        CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+        PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+        ST8x4_UB(out0, out1, dst_tmp, dst_stride);
+        dst_tmp += (4 * dst_stride);
+
+        dst10_r = dst54_r;
+        dst10_l = dst54_l;
+        dst21_r = dst65_r;
+        dst21_l = dst65_l;
+        dsth2 = dsth6;
+    }
+
+    src0_ptr += 8;
+    dst += 8;
+    src1_ptr += 8;
+
+    mask2 = LD_SB(ff_hevc_mask_arr + 16);
+    mask3 = mask2 + 2;
+
+    LD_SB3(src0_ptr, src_stride, src0, src1, src2);
+    src0_ptr += (3 * src_stride);
+    XORI_B3_128_SB(src0, src1, src2);
+    VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
+    VSHF_B2_SB(src1, src2, src1, src2, mask2, mask3, vec2, vec3);
+
+    dst10 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+    dst21 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+
+    ILVRL_H2_SH(dst21, dst10, dst10_r, dst21_r);
+    dst22 = (v8i16) __msa_splati_d((v2i64) dst21, 1);
+
+    for (loop_cnt = 2; loop_cnt--;) {
+        LD_SB8(src0_ptr, src_stride,
+               src3, src4, src5, src6, src7, src8, src9, src10);
+        src0_ptr += (8 * src_stride);
+        XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
+        VSHF_B2_SB(src3, src7, src3, src7, mask2, mask3, vec0, vec1);
+        VSHF_B2_SB(src4, src8, src4, src8, mask2, mask3, vec2, vec3);
+        VSHF_B2_SB(src5, src9, src5, src9, mask2, mask3, vec4, vec5);
+        VSHF_B2_SB(src6, src10, src6, src10, mask2, mask3, vec6, vec7);
+
+        dst73 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+        dst84 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+        dst95 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+        dst106 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+        dst32_r = __msa_ilvr_h(dst73, dst22);
+        ILVRL_H2_SH(dst84, dst73, dst43_r, dst87_r);
+        ILVRL_H2_SH(dst95, dst84, dst54_r, dst98_r);
+        ILVRL_H2_SH(dst106, dst95, dst65_r, dst109_r);
+        dst22 = (v8i16) __msa_splati_d((v2i64) dst73, 1);
+        dst76_r = __msa_ilvr_h(dst22, dst106);
+
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in0);
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in1);
+
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in2);
+        LD2(src1_ptr, src2_stride, tp0, tp1);
+        src1_ptr += 2 * src2_stride;
+        INSERT_D2_SH(tp0, tp1, in3);
+
+        ADDS_SH4_SH(in0, const_vec, in1, const_vec, in2, const_vec, in3,
+                    const_vec, in0, in1, in2, in3);
+
+        dst0 = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+        dst1 = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+        dst2 = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+        dst3 = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+        dst4 = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
+        dst5 = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
+        dst6 = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
+        dst7 = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
+
+        SRA_4V(dst0, dst1, dst2, dst3, 6);
+        SRA_4V(dst4, dst5, dst6, dst7, 6);
+        PCKEV_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6,
+                    tmp0, tmp1, tmp2, tmp3);
+        ADDS_SH4_SH(in0, tmp0, in1, tmp1, in2, tmp2, in3, tmp3,
+                    tmp0, tmp1, tmp2, tmp3);
+        SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+        CLIP_SH4_0_255_MAX_SATU(tmp0, tmp1, tmp2, tmp3);
+        PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+        ST4x8_UB(out0, out1, dst, dst_stride);
+        dst += (8 * dst_stride);
+
+        dst10_r = dst98_r;
+        dst21_r = dst109_r;
+        dst22 = (v8i16) __msa_splati_d((v2i64) dst106, 1);
+    }
 }
 
 static void hevc_hv_bi_4t_16w_msa(uint8_t *src0_ptr,
@@ -4637,9 +4854,14 @@ static void hevc_hv_bi_4t_16w_msa(uint8_t *src0_ptr,
                                   const int8_t *filter_y,
                                   int32_t height)
 {
-    hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
-                                  dst, dst_stride, filter_x, filter_y,
-                                  height, 16);
+    if (4 == height) {
+        hevc_hv_bi_4t_8multx4_msa(src0_ptr, src_stride, src1_ptr, src2_stride,
+                                  dst, dst_stride, filter_x, filter_y, 2);
+    } else {
+        hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr,
+                                      src2_stride, dst, dst_stride, filter_x,
+                                      filter_y, height, 16);
+    }
 }
 
 static void hevc_hv_bi_4t_24w_msa(uint8_t *src0_ptr,
@@ -4664,7 +4886,7 @@ static void hevc_hv_bi_4t_32w_msa(uint8_t *src0_ptr,
                                   uint8_t *dst,
                                   int32_t dst_stride,
                                   const int8_t *filter_x,
-                                  const const int8_t *filter_y,
+                                  const int8_t *filter_y,
                                   int32_t height)
 {
     hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride,