avcodec/mips: Improve avc chroma avg hv mc msa functions

Replace generic with block size specific function.
Load the specific destination bytes instead of MSA load and pack.

Signed-off-by: Kaustubh Raste <kaustubh.raste@imgtec.com>
Reviewed-by: Manojkumar Bhosale <Manojkumar.Bhosale@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

1 files changed, 238 insertions, 200 deletions

diff --git a/libavcodec/mips/h264chroma_msa.c b/libavcodec/mips/h264chroma_msa.c
index a5c333471b..4c25761194 100644
--- a/libavcodec/mips/h264chroma_msa.c
+++ b/libavcodec/mips/h264chroma_msa.c
@@ -1408,15 +1408,15 @@ static void avc_chroma_vt_and_aver_dst_8w_msa(uint8_t *src, uint8_t *dst,
     }
 }
 
-static void avc_chroma_hv_and_aver_dst_2x2_msa(uint8_t *src, int32_t src_stride,
-                                               uint8_t *dst, int32_t dst_stride,
+static void avc_chroma_hv_and_aver_dst_2x2_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
                                                uint32_t coef_hor0,
                                                uint32_t coef_hor1,
                                                uint32_t coef_ver0,
                                                uint32_t coef_ver1)
 {
     uint16_t out0, out1;
-    v16u8 dst0, dst1;
+    v16u8 dst0 = { 0 };
     v16u8 src0, src1, src2;
     v8u16 res_hz0, res_hz1, res_vt0, res_vt1;
     v16i8 res, mask;
@@ -1428,8 +1428,11 @@ static void avc_chroma_hv_and_aver_dst_2x2_msa(uint8_t *src, int32_t src_stride,
 
     mask = LD_SB(&chroma_mask_arr[48]);
 
-    LD_UB3(src, src_stride, src0, src1, src2);
-    LD_UB2(dst, dst_stride, dst0, dst1);
+    LD_UB3(src, stride, src0, src1, src2);
+    out0 = LH(dst);
+    out1 = LH(dst + stride);
+    dst0 = (v16u8) __msa_insert_h((v8i16) dst0, 0, out0);
+    dst0 = (v16u8) __msa_insert_h((v8i16) dst0, 1, out1);
     VSHF_B2_UB(src0, src1, src1, src2, mask, mask, src0, src1);
     DOTP_UB2_UH(src0, src1, coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1);
     MUL2(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_vt0, res_vt1);
@@ -1438,67 +1441,26 @@ static void avc_chroma_hv_and_aver_dst_2x2_msa(uint8_t *src, int32_t src_stride,
     res_vt0 = (v8u16) __msa_srari_h((v8i16) res_vt0, 6);
     res_vt0 = __msa_sat_u_h(res_vt0, 7);
     res = __msa_pckev_b((v16i8) res_vt0, (v16i8) res_vt0);
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 1, (v8i16) dst1);
     dst0 = __msa_aver_u_b((v16u8) res, dst0);
     out0 = __msa_copy_u_h((v8i16) dst0, 0);
     out1 = __msa_copy_u_h((v8i16) dst0, 1);
 
     SH(out0, dst);
-    dst += dst_stride;
+    dst += stride;
     SH(out1, dst);
 }
 
-static void avc_chroma_hv_and_aver_dst_2x4_msa(uint8_t *src, int32_t src_stride,
-                                               uint8_t *dst, int32_t dst_stride,
+static void avc_chroma_hv_and_aver_dst_2x4_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
                                                uint32_t coef_hor0,
                                                uint32_t coef_hor1,
                                                uint32_t coef_ver0,
                                                uint32_t coef_ver1)
 {
+    uint16_t tp0, tp1, tp2, tp3;
     v16u8 src0, src1, src2, src3, src4;
     v16u8 tmp0, tmp1, tmp2, tmp3;
-    v16u8 dst0, dst1, dst2, dst3;
-    v8u16 res_hz0, res_hz1, res_vt0, res_vt1;
-    v16i8 res, mask;
-    v16i8 coeff_hz_vec0 = __msa_fill_b(coef_hor0);
-    v16i8 coeff_hz_vec1 = __msa_fill_b(coef_hor1);
-    v16u8 coeff_hz_vec = (v16u8) __msa_ilvr_b(coeff_hz_vec0, coeff_hz_vec1);
-    v8u16 coeff_vt_vec0 = (v8u16) __msa_fill_h(coef_ver0);
-    v8u16 coeff_vt_vec1 = (v8u16) __msa_fill_h(coef_ver1);
-
-    mask = LD_SB(&chroma_mask_arr[48]);
-
-    LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
-    LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-    VSHF_B2_UB(src0, src1, src2, src3, mask, mask, tmp0, tmp1);
-    VSHF_B2_UB(src1, src2, src3, src4, mask, mask, tmp2, tmp3);
-    ILVR_D2_UB(tmp1, tmp0, tmp3, tmp2, src0, src1);
-    DOTP_UB2_UH(src0, src1, coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1);
-    MUL2(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_vt0, res_vt1);
-
-    res_vt0 += res_vt1;
-    res_vt0 = (v8u16) __msa_srari_h((v8i16) res_vt0, 6);
-    res_vt0 = __msa_sat_u_h(res_vt0, 7);
-    res = __msa_pckev_b((v16i8) res_vt0, (v16i8) res_vt0);
-
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 1, (v8i16) dst1);
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 2, (v8i16) dst2);
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 3, (v8i16) dst3);
-    dst0 = __msa_aver_u_b((v16u8) res, dst0);
-
-    ST2x4_UB(dst0, 0, dst, dst_stride);
-}
-
-static void avc_chroma_hv_and_aver_dst_2x8_msa(uint8_t *src, int32_t src_stride,
-                                               uint8_t *dst, int32_t dst_stride,
-                                               uint32_t coef_hor0,
-                                               uint32_t coef_hor1,
-                                               uint32_t coef_ver0,
-                                               uint32_t coef_ver1)
-{
-    v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
-    v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
-    v16u8 tmp0, tmp1, tmp2, tmp3;
+    v16u8 dst0 = { 0 };
     v8u16 res_hz0, res_hz1, res_vt0, res_vt1;
     v16i8 res, mask;
     v16i8 coeff_hz_vec0 = __msa_fill_b(coef_hor0);
@@ -1509,26 +1471,18 @@ static void avc_chroma_hv_and_aver_dst_2x8_msa(uint8_t *src, int32_t src_stride,
 
     mask = LD_SB(&chroma_mask_arr[48]);
 
-    LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
-    src += (5 * src_stride);
-    LD_UB4(src, src_stride, src5, src6, src7, src8);
-
-    LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
-
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 1, (v8i16) dst1);
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 2, (v8i16) dst2);
-    dst0 = (v16u8) __msa_insve_h((v8i16) dst0, 3, (v8i16) dst3);
-
-    dst4 = (v16u8) __msa_insve_h((v8i16) dst4, 1, (v8i16) dst5);
-    dst4 = (v16u8) __msa_insve_h((v8i16) dst4, 2, (v8i16) dst6);
-    dst4 = (v16u8) __msa_insve_h((v8i16) dst4, 3, (v8i16) dst7);
-
+    LD_UB5(src, stride, src0, src1, src2, src3, src4);
+    tp0 = LH(dst);
+    tp1 = LH(dst + stride);
+    tp2 = LH(dst + 2 * stride);
+    tp3 = LH(dst + 3 * stride);
+    dst0 = (v16u8) __msa_insert_h((v8i16) dst0, 0, tp0);
+    dst0 = (v16u8) __msa_insert_h((v8i16) dst0, 1, tp1);
+    dst0 = (v16u8) __msa_insert_h((v8i16) dst0, 2, tp2);
+    dst0 = (v16u8) __msa_insert_h((v8i16) dst0, 3, tp3);
     VSHF_B2_UB(src0, src1, src2, src3, mask, mask, tmp0, tmp1);
     VSHF_B2_UB(src1, src2, src3, src4, mask, mask, tmp2, tmp3);
     ILVR_D2_UB(tmp1, tmp0, tmp3, tmp2, src0, src1);
-    VSHF_B2_UB(src4, src5, src6, src7, mask, mask, tmp0, tmp1);
-    VSHF_B2_UB(src5, src6, src7, src8, mask, mask, tmp2, tmp3);
-    ILVR_D2_UB(tmp1, tmp0, tmp3, tmp2, src4, src5);
     DOTP_UB2_UH(src0, src1, coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1);
     MUL2(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_vt0, res_vt1);
 
@@ -1538,23 +1492,11 @@ static void avc_chroma_hv_and_aver_dst_2x8_msa(uint8_t *src, int32_t src_stride,
     res = __msa_pckev_b((v16i8) res_vt0, (v16i8) res_vt0);
     dst0 = __msa_aver_u_b((v16u8) res, dst0);
 
-    ST2x4_UB(dst0, 0, dst, dst_stride);
-    dst += (4 * dst_stride);
-
-    DOTP_UB2_UH(src4, src5, coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1);
-    MUL2(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_vt0, res_vt1);
-
-    res_vt0 += res_vt1;
-    res_vt0 = (v8u16) __msa_srari_h((v8i16) res_vt0, 6);
-    res_vt0 = __msa_sat_u_h(res_vt0, 7);
-    res = __msa_pckev_b((v16i8) res_vt0, (v16i8) res_vt0);
-    dst4 = __msa_aver_u_b((v16u8) res, dst4);
-
-    ST2x4_UB(dst4, 0, dst, dst_stride);
+    ST2x4_UB(dst0, 0, dst, stride);
 }
 
-static void avc_chroma_hv_and_aver_dst_2w_msa(uint8_t *src, int32_t src_stride,
-                                              uint8_t *dst, int32_t dst_stride,
+static void avc_chroma_hv_and_aver_dst_2w_msa(uint8_t *src, uint8_t *dst,
+                                              int32_t stride,
                                               uint32_t coef_hor0,
                                               uint32_t coef_hor1,
                                               uint32_t coef_ver0,
@@ -1562,31 +1504,26 @@ static void avc_chroma_hv_and_aver_dst_2w_msa(uint8_t *src, int32_t src_stride,
                                               int32_t height)
 {
     if (2 == height) {
-        avc_chroma_hv_and_aver_dst_2x2_msa(src, src_stride, dst, dst_stride,
-                                           coef_hor0, coef_hor1,
-                                           coef_ver0, coef_ver1);
+        avc_chroma_hv_and_aver_dst_2x2_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
     } else if (4 == height) {
-        avc_chroma_hv_and_aver_dst_2x4_msa(src, src_stride, dst, dst_stride,
-                                           coef_hor0, coef_hor1,
-                                           coef_ver0, coef_ver1);
-    } else if (8 == height) {
-        avc_chroma_hv_and_aver_dst_2x8_msa(src, src_stride, dst, dst_stride,
-                                           coef_hor0, coef_hor1,
-                                           coef_ver0, coef_ver1);
+        avc_chroma_hv_and_aver_dst_2x4_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
     }
 }
 
-static void avc_chroma_hv_and_aver_dst_4x2_msa(uint8_t *src, int32_t src_stride,
-                                               uint8_t *dst, int32_t dst_stride,
+static void avc_chroma_hv_and_aver_dst_4x2_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
                                                uint32_t coef_hor0,
                                                uint32_t coef_hor1,
                                                uint32_t coef_ver0,
                                                uint32_t coef_ver1)
 {
+    uint32_t tp0, tp1;
     v16u8 src0, src1, src2;
-    v16u8 dst0, dst1;
+    v16u8 dst0, dst_data = { 0 };
     v8u16 res_hz0, res_hz1, res_vt0, res_vt1;
-    v16i8 res, mask;
+    v16i8 mask;
     v16i8 coeff_hz_vec0 = __msa_fill_b(coef_hor0);
     v16i8 coeff_hz_vec1 = __msa_fill_b(coef_hor1);
     v16u8 coeff_hz_vec = (v16u8) __msa_ilvr_b(coeff_hz_vec0, coeff_hz_vec1);
@@ -1595,8 +1532,9 @@ static void avc_chroma_hv_and_aver_dst_4x2_msa(uint8_t *src, int32_t src_stride,
 
     mask = LD_SB(&chroma_mask_arr[0]);
 
-    LD_UB3(src, src_stride, src0, src1, src2);
-    LD_UB2(dst, dst_stride, dst0, dst1);
+    LD_UB3(src, stride, src0, src1, src2);
+    LW2(dst, stride, tp0, tp1);
+    INSERT_W2_UB(tp0, tp1, dst_data);
     VSHF_B2_UB(src0, src1, src1, src2, mask, mask, src0, src1);
     DOTP_UB2_UH(src0, src1, coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1);
     MUL2(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_vt0, res_vt1);
@@ -1604,26 +1542,22 @@ static void avc_chroma_hv_and_aver_dst_4x2_msa(uint8_t *src, int32_t src_stride,
     res_vt0 += res_vt1;
     res_vt0 = (v8u16) __msa_srari_h((v8i16) res_vt0, 6);
     res_vt0 = __msa_sat_u_h(res_vt0, 7);
-    res = __msa_pckev_b((v16i8) res_vt0, (v16i8) res_vt0);
-    dst0 = (v16u8) __msa_insve_w((v4i32) dst0, 1, (v4i32) dst1);
-    dst0 = __msa_aver_u_b((v16u8) res, dst0);
+    dst0 = (v16u8) __msa_pckev_b((v16i8) res_vt0, (v16i8) res_vt0);
+    dst0 = __msa_aver_u_b(dst0, dst_data);
 
-    ST4x2_UB(dst0, dst, dst_stride);
+    ST4x2_UB(dst0, dst, stride);
 }
 
-static void avc_chroma_hv_and_aver_dst_4x4mul_msa(uint8_t *src,
-                                                  int32_t src_stride,
-                                                  uint8_t *dst,
-                                                  int32_t dst_stride,
-                                                  uint32_t coef_hor0,
-                                                  uint32_t coef_hor1,
-                                                  uint32_t coef_ver0,
-                                                  uint32_t coef_ver1,
-                                                  int32_t height)
+static void avc_chroma_hv_and_aver_dst_4x4_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
+                                               uint32_t coef_hor0,
+                                               uint32_t coef_hor1,
+                                               uint32_t coef_ver0,
+                                               uint32_t coef_ver1)
 {
-    uint32_t row;
+    uint32_t tp0, tp1, tp2, tp3;
     v16u8 src0, src1, src2, src3, src4;
-    v16u8 dst0, dst1, dst2, dst3;
+    v16u8 out, dst_data = { 0 };
     v8u16 res_hz0, res_hz1, res_hz2, res_hz3;
     v8u16 res_vt0, res_vt1, res_vt2, res_vt3;
     v16i8 mask;
@@ -1632,45 +1566,78 @@ static void avc_chroma_hv_and_aver_dst_4x4mul_msa(uint8_t *src,
     v16u8 coeff_hz_vec = (v16u8) __msa_ilvr_b(coeff_hz_vec0, coeff_hz_vec1);
     v8u16 coeff_vt_vec0 = (v8u16) __msa_fill_h(coef_ver0);
     v8u16 coeff_vt_vec1 = (v8u16) __msa_fill_h(coef_ver1);
-    v16u8 res0, res1;
 
     mask = LD_SB(&chroma_mask_arr[0]);
 
-    src0 = LD_UB(src);
-    src += src_stride;
-
-    for (row = (height >> 2); row--;) {
-        LD_UB4(src, src_stride, src1, src2, src3, src4);
-        src += (4 * src_stride);
-
-        LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
-        VSHF_B2_UB(src0, src1, src1, src2, mask, mask, src0, src1);
-        VSHF_B2_UB(src2, src3, src3, src4, mask, mask, src2, src3);
-        DOTP_UB4_UH(src0, src1, src2, src3, coeff_hz_vec, coeff_hz_vec,
-                    coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1, res_hz2,
-                    res_hz3);
-        MUL4(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_hz2,
-             coeff_vt_vec1, res_hz3, coeff_vt_vec0, res_vt0, res_vt1, res_vt2,
-             res_vt3);
-        ADD2(res_vt0, res_vt1, res_vt2, res_vt3, res_vt0, res_vt1);
-        SRARI_H2_UH(res_vt0, res_vt1, 6);
-        SAT_UH2_UH(res_vt0, res_vt1, 7);
-        PCKEV_B2_UB(res_vt0, res_vt0, res_vt1, res_vt1, res0, res1);
-
-        dst0 = (v16u8) __msa_insve_w((v4i32) dst0, 1, (v4i32) dst1);
-        dst1 = (v16u8) __msa_insve_w((v4i32) dst2, 1, (v4i32) dst3);
-
-        AVER_UB2_UB(res0, dst0, res1, dst1, dst0, dst1);
-
-        ST4x4_UB(dst0, dst1, 0, 1, 0, 1, dst, dst_stride);
-        dst += (4 * dst_stride);
-        src0 = src4;
-    }
+    LD_UB5(src, stride, src0, src1, src2, src3, src4);
+    LW4(dst, stride, tp0, tp1, tp2, tp3);
+    INSERT_W4_UB(tp0, tp1, tp2, tp3, dst_data);
+    VSHF_B2_UB(src0, src1, src1, src2, mask, mask, src0, src1);
+    VSHF_B2_UB(src2, src3, src3, src4, mask, mask, src2, src3);
+    DOTP_UB4_UH(src0, src1, src2, src3, coeff_hz_vec, coeff_hz_vec,
+                coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1, res_hz2,
+                res_hz3);
+    MUL4(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_hz2, coeff_vt_vec1,
+         res_hz3, coeff_vt_vec0, res_vt0, res_vt1, res_vt2, res_vt3);
+    ADD2(res_vt0, res_vt1, res_vt2, res_vt3, res_vt0, res_vt1);
+    SRARI_H2_UH(res_vt0, res_vt1, 6);
+    SAT_UH2_UH(res_vt0, res_vt1, 7);
+    out = (v16u8) __msa_pckev_b((v16i8) res_vt1, (v16i8) res_vt0);
+    out = __msa_aver_u_b(out, dst_data);
+    ST4x4_UB(out, out, 0, 1, 2, 3, dst, stride);
+}
+
+static void avc_chroma_hv_and_aver_dst_4x8_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
+                                               uint32_t coef_hor0,
+                                               uint32_t coef_hor1,
+                                               uint32_t coef_ver0,
+                                               uint32_t coef_ver1)
+{
+    uint32_t tp0, tp1, tp2, tp3;
+    v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, res0, res1;
+    v16u8 dst0 = { 0 }, dst1 = { 0 };
+    v8u16 res_hz0, res_hz1, res_hz2, res_hz3, res_hz4, res_hz5, res_hz6, res_hz7;
+    v8u16 res_vt0, res_vt1, res_vt2, res_vt3, res_vt4, res_vt5, res_vt6, res_vt7;
+    v16i8 mask;
+    v16i8 coeff_hz_vec0 = __msa_fill_b(coef_hor0);
+    v16i8 coeff_hz_vec1 = __msa_fill_b(coef_hor1);
+    v16u8 coeff_hz_vec = (v16u8) __msa_ilvr_b(coeff_hz_vec0, coeff_hz_vec1);
+    v8u16 coeff_vt_vec0 = (v8u16) __msa_fill_h(coef_ver0);
+    v8u16 coeff_vt_vec1 = (v8u16) __msa_fill_h(coef_ver1);
+
+    mask = LD_SB(&chroma_mask_arr[0]);
+
+    LD_UB5(src, stride, src0, src1, src2, src3, src4);
+    src += (5 * stride);
+    LD_UB4(src, stride, src5, src6, src7, src8);
+    LW4(dst, stride, tp0, tp1, tp2, tp3);
+    INSERT_W4_UB(tp0, tp1, tp2, tp3, dst0);
+    LW4(dst + 4 * stride, stride, tp0, tp1, tp2, tp3);
+    INSERT_W4_UB(tp0, tp1, tp2, tp3, dst1);
+    VSHF_B2_UB(src0, src1, src1, src2, mask, mask, src0, src1);
+    VSHF_B2_UB(src2, src3, src3, src4, mask, mask, src2, src3);
+    VSHF_B2_UB(src4, src5, src5, src6, mask, mask, src4, src5);
+    VSHF_B2_UB(src6, src7, src7, src8, mask, mask, src6, src7);
+    DOTP_UB4_UH(src0, src1, src2, src3, coeff_hz_vec, coeff_hz_vec,
+                coeff_hz_vec, coeff_hz_vec, res_hz0, res_hz1, res_hz2, res_hz3);
+    DOTP_UB4_UH(src4, src5, src6, src7, coeff_hz_vec, coeff_hz_vec,
+                coeff_hz_vec, coeff_hz_vec, res_hz4, res_hz5, res_hz6, res_hz7);
+    MUL4(res_hz0, coeff_vt_vec1, res_hz1, coeff_vt_vec0, res_hz2, coeff_vt_vec1,
+         res_hz3, coeff_vt_vec0, res_vt0, res_vt1, res_vt2, res_vt3);
+    MUL4(res_hz4, coeff_vt_vec1, res_hz5, coeff_vt_vec0, res_hz6, coeff_vt_vec1,
+         res_hz7, coeff_vt_vec0, res_vt4, res_vt5, res_vt6, res_vt7);
+    ADD2(res_vt0, res_vt1, res_vt2, res_vt3, res_vt0, res_vt1);
+    ADD2(res_vt4, res_vt5, res_vt6, res_vt7, res_vt2, res_vt3);
+    SRARI_H4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 6);
+    SAT_UH4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 7);
+    PCKEV_B2_UB(res_vt1, res_vt0, res_vt3, res_vt2, res0, res1);
+    AVER_UB2_UB(res0, dst0, res1, dst1, res0, res1);
+    ST4x8_UB(res0, res1, dst, stride);
 }
 
-static void avc_chroma_hv_and_aver_dst_4w_msa(uint8_t *src, int32_t src_stride,
-                                              uint8_t *dst, int32_t dst_stride,
+static void avc_chroma_hv_and_aver_dst_4w_msa(uint8_t *src, uint8_t *dst,
+                                              int32_t stride,
                                               uint32_t coef_hor0,
                                               uint32_t coef_hor1,
                                               uint32_t coef_ver0,
@@ -1678,30 +1645,30 @@ static void avc_chroma_hv_and_aver_dst_4w_msa(uint8_t *src, int32_t src_stride,
                                               int32_t height)
 {
     if (2 == height) {
-        avc_chroma_hv_and_aver_dst_4x2_msa(src, src_stride, dst, dst_stride,
-                                           coef_hor0, coef_hor1,
-                                           coef_ver0, coef_ver1);
-    } else {
-        avc_chroma_hv_and_aver_dst_4x4mul_msa(src, src_stride, dst, dst_stride,
-                                              coef_hor0, coef_hor1,
-                                              coef_ver0, coef_ver1, height);
+        avc_chroma_hv_and_aver_dst_4x2_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
+    } else if (4 == height) {
+        avc_chroma_hv_and_aver_dst_4x4_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
+    } else if (8 == height) {
+        avc_chroma_hv_and_aver_dst_4x8_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
     }
 }
 
-static void avc_chroma_hv_and_aver_dst_8w_msa(uint8_t *src, int32_t src_stride,
-                                              uint8_t *dst, int32_t dst_stride,
-                                              uint32_t coef_hor0,
-                                              uint32_t coef_hor1,
-                                              uint32_t coef_ver0,
-                                              uint32_t coef_ver1,
-                                              int32_t height)
+static void avc_chroma_hv_and_aver_dst_8x4_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
+                                               uint32_t coef_hor0,
+                                               uint32_t coef_hor1,
+                                               uint32_t coef_ver0,
+                                               uint32_t coef_ver1)
 {
-    uint32_t row;
+    uint64_t tp0, tp1, tp2, tp3;
     v16u8 src0, src1, src2, src3, src4, out0, out1;
     v8u16 res_hz0, res_hz1, res_hz2;
     v8u16 res_hz3, res_hz4;
     v8u16 res_vt0, res_vt1, res_vt2, res_vt3;
-    v16u8 dst0, dst1, dst2, dst3;
+    v16u8 dst0 = { 0 }, dst1 = { 0 };
     v16i8 mask;
     v16i8 coeff_hz_vec0 = __msa_fill_b(coef_hor0);
     v16i8 coeff_hz_vec1 = __msa_fill_b(coef_hor1);
@@ -1712,40 +1679,114 @@ static void avc_chroma_hv_and_aver_dst_8w_msa(uint8_t *src, int32_t src_stride,
     mask = LD_SB(&chroma_mask_arr[32]);
 
     src0 = LD_UB(src);
-    src += src_stride;
+    src += stride;
+    src0 = (v16u8) __msa_vshf_b(mask, (v16i8) src0, (v16i8) src0);
+    res_hz0 = __msa_dotp_u_h(src0, coeff_hz_vec);
+    LD_UB4(src, stride, src1, src2, src3, src4);
+    src += (4 * stride);
+    LD4(dst, stride, tp0, tp1, tp2, tp3);
+    INSERT_D2_UB(tp0, tp1, dst0);
+    INSERT_D2_UB(tp2, tp3, dst1);
+    VSHF_B2_UB(src1, src1, src2, src2, mask, mask, src1, src2);
+    VSHF_B2_UB(src3, src3, src4, src4, mask, mask, src3, src4);
+    DOTP_UB4_UH(src1, src2, src3, src4, coeff_hz_vec, coeff_hz_vec,
+                coeff_hz_vec, coeff_hz_vec, res_hz1, res_hz2, res_hz3, res_hz4);
+    MUL4(res_hz1, coeff_vt_vec0, res_hz2, coeff_vt_vec0, res_hz3, coeff_vt_vec0,
+         res_hz4, coeff_vt_vec0, res_vt0, res_vt1, res_vt2, res_vt3);
+    res_vt0 += (res_hz0 * coeff_vt_vec1);
+    res_vt1 += (res_hz1 * coeff_vt_vec1);
+    res_vt2 += (res_hz2 * coeff_vt_vec1);
+    res_vt3 += (res_hz3 * coeff_vt_vec1);
+    SRARI_H4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 6);
+    SAT_UH4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 7);
+    PCKEV_B2_UB(res_vt1, res_vt0, res_vt3, res_vt2, out0, out1);
+    AVER_UB2_UB(out0, dst0, out1, dst1, out0, out1);
+    ST8x4_UB(out0, out1, dst, stride);
+}
 
+static void avc_chroma_hv_and_aver_dst_8x8_msa(uint8_t *src, uint8_t *dst,
+                                               int32_t stride,
+                                               uint32_t coef_hor0,
+                                               uint32_t coef_hor1,
+                                               uint32_t coef_ver0,
+                                               uint32_t coef_ver1)
+{
+    uint64_t tp0, tp1, tp2, tp3;
+    v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+    v16u8 out0, out1, out2, out3;
+    v16u8 dst0 = { 0 }, dst1 = { 0 }, dst2 = { 0 }, dst3 = { 0 };
+    v8u16 res_hz0, res_hz1, res_hz2, res_hz3, res_hz4;
+    v8u16 res_hz5, res_hz6, res_hz7, res_hz8;
+    v8u16 res_vt0, res_vt1, res_vt2, res_vt3;
+    v8u16 res_vt4, res_vt5, res_vt6, res_vt7;
+    v16i8 mask;
+    v16i8 coeff_hz_vec0 = __msa_fill_b(coef_hor0);
+    v16i8 coeff_hz_vec1 = __msa_fill_b(coef_hor1);
+    v16u8 coeff_hz_vec = (v16u8) __msa_ilvr_b(coeff_hz_vec0, coeff_hz_vec1);
+    v8u16 coeff_vt_vec0 = (v8u16) __msa_fill_h(coef_ver0);
+    v8u16 coeff_vt_vec1 = (v8u16) __msa_fill_h(coef_ver1);
+
+    mask = LD_SB(&chroma_mask_arr[32]);
+
+    LD_UB5(src, stride, src0, src1, src2, src3, src4);
+    src += (5 * stride);
+    LD_UB4(src, stride, src5, src6, src7, src8);
     src0 = (v16u8) __msa_vshf_b(mask, (v16i8) src0, (v16i8) src0);
+    VSHF_B2_UB(src1, src1, src2, src2, mask, mask, src1, src2);
+    VSHF_B2_UB(src3, src3, src4, src4, mask, mask, src3, src4);
+    VSHF_B2_UB(src5, src5, src6, src6, mask, mask, src5, src6);
+    VSHF_B2_UB(src7, src7, src8, src8, mask, mask, src7, src8);
     res_hz0 = __msa_dotp_u_h(src0, coeff_hz_vec);
+    DOTP_UB4_UH(src1, src2, src3, src4, coeff_hz_vec, coeff_hz_vec,
+                coeff_hz_vec, coeff_hz_vec, res_hz1, res_hz2, res_hz3,
+                res_hz4);
+    DOTP_UB4_UH(src5, src6, src7, src8, coeff_hz_vec, coeff_hz_vec,
+                coeff_hz_vec, coeff_hz_vec, res_hz5, res_hz6, res_hz7, res_hz8);
+    MUL4(res_hz1, coeff_vt_vec0, res_hz2, coeff_vt_vec0, res_hz3,
+         coeff_vt_vec0, res_hz4, coeff_vt_vec0, res_vt0, res_vt1, res_vt2,
+         res_vt3);
+    MUL4(res_hz5, coeff_vt_vec0, res_hz6, coeff_vt_vec0, res_hz7,
+         coeff_vt_vec0, res_hz8, coeff_vt_vec0, res_vt4, res_vt5, res_vt6,
+         res_vt7);
+    LD4(dst, stride, tp0, tp1, tp2, tp3);
+    INSERT_D2_UB(tp0, tp1, dst0);
+    INSERT_D2_UB(tp2, tp3, dst1);
+    LD4(dst + 4 * stride, stride, tp0, tp1, tp2, tp3);
+    INSERT_D2_UB(tp0, tp1, dst2);
+    INSERT_D2_UB(tp2, tp3, dst3);
+    res_vt0 += (res_hz0 * coeff_vt_vec1);
+    res_vt1 += (res_hz1 * coeff_vt_vec1);
+    res_vt2 += (res_hz2 * coeff_vt_vec1);
+    res_vt3 += (res_hz3 * coeff_vt_vec1);
+    res_vt4 += (res_hz4 * coeff_vt_vec1);
+    res_vt5 += (res_hz5 * coeff_vt_vec1);
+    res_vt6 += (res_hz6 * coeff_vt_vec1);
+    res_vt7 += (res_hz7 * coeff_vt_vec1);
+    SRARI_H4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 6);
+    SRARI_H4_UH(res_vt4, res_vt5, res_vt6, res_vt7, 6);
+    SAT_UH4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 7);
+    SAT_UH4_UH(res_vt4, res_vt5, res_vt6, res_vt7, 7);
+    PCKEV_B2_UB(res_vt1, res_vt0, res_vt3, res_vt2, out0, out1);
+    PCKEV_B2_UB(res_vt5, res_vt4, res_vt7, res_vt6, out2, out3);
+    AVER_UB2_UB(out0, dst0, out1, dst1, out0, out1);
+    AVER_UB2_UB(out2, dst2, out3, dst3, out2, out3);
+    ST8x8_UB(out0, out1, out2, out3, dst, stride);
+}
 
-    for (row = (height >> 2); row--;) {
-        LD_UB4(src, src_stride, src1, src2, src3, src4);
-        src += (4 * src_stride);
-
-        LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-        VSHF_B2_UB(src1, src1, src2, src2, mask, mask, src1, src2);
-        VSHF_B2_UB(src3, src3, src4, src4, mask, mask, src3, src4);
-        DOTP_UB4_UH(src1, src2, src3, src4, coeff_hz_vec, coeff_hz_vec,
-                    coeff_hz_vec, coeff_hz_vec, res_hz1, res_hz2, res_hz3,
-                    res_hz4);
-        MUL4(res_hz1, coeff_vt_vec0, res_hz2, coeff_vt_vec0, res_hz3,
-             coeff_vt_vec0, res_hz4, coeff_vt_vec0, res_vt0, res_vt1, res_vt2,
-             res_vt3);
-
-        res_vt0 += (res_hz0 * coeff_vt_vec1);
-        res_vt1 += (res_hz1 * coeff_vt_vec1);
-        res_vt2 += (res_hz2 * coeff_vt_vec1);
-        res_vt3 += (res_hz3 * coeff_vt_vec1);
-
-        SRARI_H4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 6);
-        SAT_UH4_UH(res_vt0, res_vt1, res_vt2, res_vt3, 7);
-
-        PCKEV_B2_UB(res_vt1, res_vt0, res_vt3, res_vt2, out0, out1);
-        PCKEV_D2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
-        AVER_UB2_UB(out0, dst0, out1, dst1, out0, out1);
-        ST8x4_UB(out0, out1, dst, dst_stride);
-        dst += (4 * dst_stride);
-
-        res_hz0 = res_hz4;
+static void avc_chroma_hv_and_aver_dst_8w_msa(uint8_t *src, uint8_t *dst,
+                                              int32_t stride,
+                                              uint32_t coef_hor0,
+                                              uint32_t coef_hor1,
+                                              uint32_t coef_ver0,
+                                              uint32_t coef_ver1,
+                                              int32_t height)
+{
+    if (4 == height) {
+        avc_chroma_hv_and_aver_dst_8x4_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
+    } else if (8 == height) {
+        avc_chroma_hv_and_aver_dst_8x8_msa(src, dst, stride, coef_hor0,
+                                           coef_hor1, coef_ver0, coef_ver1);
     }
 }
 
@@ -1923,8 +1964,7 @@ void ff_avg_h264_chroma_mc8_msa(uint8_t *dst, uint8_t *src,
 
 
     if (x && y) {
-        avc_chroma_hv_and_aver_dst_8w_msa(src, stride, dst,
-                                          stride, x, (8 - x), y,
+        avc_chroma_hv_and_aver_dst_8w_msa(src, dst, stride, x, (8 - x), y,
                                           (8 - y), height);
     } else if (x) {
         avc_chroma_hz_and_aver_dst_8w_msa(src, dst, stride, x, (8 - x), height);
@@ -1941,8 +1981,7 @@ void ff_avg_h264_chroma_mc4_msa(uint8_t *dst, uint8_t *src,
     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
 
     if (x && y) {
-        avc_chroma_hv_and_aver_dst_4w_msa(src, stride, dst,
-                                          stride, x, (8 - x), y,
+        avc_chroma_hv_and_aver_dst_4w_msa(src, dst, stride, x, (8 - x), y,
                                           (8 - y), height);
     } else if (x) {
         avc_chroma_hz_and_aver_dst_4w_msa(src, dst, stride, x, (8 - x), height);
@@ -1961,8 +2000,7 @@ void ff_avg_h264_chroma_mc2_msa(uint8_t *dst, uint8_t *src,
     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
 
     if (x && y) {
-        avc_chroma_hv_and_aver_dst_2w_msa(src, stride, dst,
-                                          stride, x, (8 - x), y,
+        avc_chroma_hv_and_aver_dst_2w_msa(src, dst, stride, x, (8 - x), y,
                                           (8 - y), height);
     } else if (x) {
         avc_chroma_hz_and_aver_dst_2w_msa(src, dst, stride, x, (8 - x), height);