vp9: assert -> av_assert and fix associated compile error.
[ffmpeg.git] / libavcodec / rv40dsp.c
index 625ce9e..5579bd9 100644 (file)
  * RV40 decoder motion compensation functions
  */
 
+#include "libavutil/common.h"
+#include "libavutil/intreadwrite.h"
 #include "avcodec.h"
-#include "dsputil.h"
+#include "h264qpel.h"
+#include "mathops.h"
+#include "pixels.h"
+#include "rnd_avg.h"
 #include "rv34dsp.h"
+#include "libavutil/avassert.h"
 
 #define RV40_LOWPASS(OPNAME, OP) \
-static av_unused void OPNAME ## rv40_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride,\
+static void OPNAME ## rv40_qpel8_h_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\
                                                      const int h, const int C1, const int C2, const int SHIFT){\
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;\
     int i;\
     for(i = 0; i < h; i++)\
     {\
@@ -48,9 +54,9 @@ static av_unused void OPNAME ## rv40_qpel8_h_lowpass(uint8_t *dst, uint8_t *src,
     }\
 }\
 \
-static void OPNAME ## rv40_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride,\
+static void OPNAME ## rv40_qpel8_v_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\
                                            const int w, const int C1, const int C2, const int SHIFT){\
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;\
     int i;\
     for(i = 0; i < w; i++)\
     {\
@@ -80,7 +86,7 @@ static void OPNAME ## rv40_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstSt
     }\
 }\
 \
-static void OPNAME ## rv40_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride,\
+static void OPNAME ## rv40_qpel16_v_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\
                                             const int w, const int C1, const int C2, const int SHIFT){\
     OPNAME ## rv40_qpel8_v_lowpass(dst  , src  , dstStride, srcStride, 8, C1, C2, SHIFT);\
     OPNAME ## rv40_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride, 8, C1, C2, SHIFT);\
@@ -90,7 +96,7 @@ static void OPNAME ## rv40_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstS
     OPNAME ## rv40_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride, w-8, C1, C2, SHIFT);\
 }\
 \
-static void OPNAME ## rv40_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride,\
+static void OPNAME ## rv40_qpel16_h_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\
                                             const int h, const int C1, const int C2, const int SHIFT){\
     OPNAME ## rv40_qpel8_h_lowpass(dst  , src  , dstStride, srcStride, 8, C1, C2, SHIFT);\
     OPNAME ## rv40_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride, 8, C1, C2, SHIFT);\
@@ -102,72 +108,84 @@ static void OPNAME ## rv40_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstS
 \
 
 #define RV40_MC(OPNAME, SIZE) \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc10_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc30_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc01_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc11_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc21_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc31_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 52, 6);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc12_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc22_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc32_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 52, 6);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc03_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc13_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\
     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 52, 6);\
 }\
 \
-static void OPNAME ## rv40_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\
+static void OPNAME ## rv40_qpel ## SIZE ## _mc23_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
+{\
     uint8_t full[SIZE*(SIZE+5)];\
     uint8_t * const full_mid = full + SIZE*2;\
     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\
@@ -189,6 +207,83 @@ RV40_MC(put_, 16)
 RV40_MC(avg_, 8)
 RV40_MC(avg_, 16)
 
+#define PIXOP2(OPNAME, OP)                                              \
+static inline void OPNAME ## _pixels8_xy2_8_c(uint8_t *block,           \
+                                              const uint8_t *pixels,    \
+                                              ptrdiff_t line_size,      \
+                                              int h)                    \
+{                                                                       \
+    /* FIXME HIGH BIT DEPTH */                                          \
+    int j;                                                              \
+                                                                        \
+    for (j = 0; j < 2; j++) {                                           \
+        int i;                                                          \
+        const uint32_t a = AV_RN32(pixels);                             \
+        const uint32_t b = AV_RN32(pixels + 1);                         \
+        uint32_t l0 = (a & 0x03030303UL) +                              \
+                      (b & 0x03030303UL) +                              \
+                           0x02020202UL;                                \
+        uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) +                       \
+                      ((b & 0xFCFCFCFCUL) >> 2);                        \
+        uint32_t l1, h1;                                                \
+                                                                        \
+        pixels += line_size;                                            \
+        for (i = 0; i < h; i += 2) {                                    \
+            uint32_t a = AV_RN32(pixels);                               \
+            uint32_t b = AV_RN32(pixels + 1);                           \
+            l1 = (a & 0x03030303UL) +                                   \
+                 (b & 0x03030303UL);                                    \
+            h1 = ((a & 0xFCFCFCFCUL) >> 2) +                            \
+                 ((b & 0xFCFCFCFCUL) >> 2);                             \
+            OP(*((uint32_t *) block),                                   \
+               h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));            \
+            pixels += line_size;                                        \
+            block  += line_size;                                        \
+            a = AV_RN32(pixels);                                        \
+            b = AV_RN32(pixels + 1);                                    \
+            l0 = (a & 0x03030303UL) +                                   \
+                 (b & 0x03030303UL) +                                   \
+                      0x02020202UL;                                     \
+            h0 = ((a & 0xFCFCFCFCUL) >> 2) +                            \
+                 ((b & 0xFCFCFCFCUL) >> 2);                             \
+            OP(*((uint32_t *) block),                                   \
+               h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));            \
+            pixels += line_size;                                        \
+            block  += line_size;                                        \
+        }                                                               \
+        pixels += 4 - line_size * (h + 1);                              \
+        block  += 4 - line_size * h;                                    \
+    }                                                                   \
+}                                                                       \
+                                                                        \
+CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_8_c,                             \
+               OPNAME ## _pixels8_xy2_8_c,                              \
+               8)                                                       \
+
+#define op_avg(a, b) a = rnd_avg32(a, b)
+#define op_put(a, b) a = b
+PIXOP2(avg, op_avg)
+PIXOP2(put, op_put)
+#undef op_avg
+#undef op_put
+
+static void put_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
+{
+    put_pixels16_xy2_8_c(dst, src, stride, 16);
+}
+static void avg_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
+{
+    avg_pixels16_xy2_8_c(dst, src, stride, 16);
+}
+static void put_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
+{
+    put_pixels8_xy2_8_c(dst, src, stride, 8);
+}
+static void avg_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
+{
+    avg_pixels8_xy2_8_c(dst, src, stride, 8);
+}
+
 static const int rv40_bias[4][4] = {
     {  0, 16, 32, 16 },
     { 32, 28, 32, 28 },
@@ -197,7 +292,10 @@ static const int rv40_bias[4][4] = {
 };
 
 #define RV40_CHROMA_MC(OPNAME, OP)\
-static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\
+static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst /*align 8*/,\
+                                        uint8_t *src /*align 1*/,\
+                                        ptrdiff_t stride, int h, int x, int y)\
+{\
     const int A = (8-x) * (8-y);\
     const int B = (  x) * (8-y);\
     const int C = (8-x) * (  y);\
@@ -205,7 +303,7 @@ static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*a
     int i;\
     int bias = rv40_bias[y>>1][x>>1];\
     \
-    assert(x<8 && y<8 && x>=0 && y>=0);\
+    av_assert2(x<8 && y<8 && x>=0 && y>=0);\
 \
     if(D){\
         for(i = 0; i < h; i++){\
@@ -218,7 +316,7 @@ static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*a
         }\
     }else{\
         const int E = B + C;\
-        const int step = C ? stride : 1;\
+        const ptrdiff_t step = C ? stride : 1;\
         for(i = 0; i < h; i++){\
             OP(dst[0], (A*src[0] + E*src[step+0] + bias));\
             OP(dst[1], (A*src[1] + E*src[step+1] + bias));\
@@ -230,7 +328,10 @@ static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*a
     }\
 }\
 \
-static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\
+static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/,\
+                                        uint8_t *src/*align 1*/,\
+                                        ptrdiff_t stride, int h, int x, int y)\
+{\
     const int A = (8-x) * (8-y);\
     const int B = (  x) * (8-y);\
     const int C = (8-x) * (  y);\
@@ -238,7 +339,7 @@ static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*a
     int i;\
     int bias = rv40_bias[y>>1][x>>1];\
     \
-    assert(x<8 && y<8 && x>=0 && y>=0);\
+    av_assert2(x<8 && y<8 && x>=0 && y>=0);\
 \
     if(D){\
         for(i = 0; i < h; i++){\
@@ -255,7 +356,7 @@ static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*a
         }\
     }else{\
         const int E = B + C;\
-        const int step = C ? stride : 1;\
+        const ptrdiff_t step = C ? stride : 1;\
         for(i = 0; i < h; i++){\
             OP(dst[0], (A*src[0] + E*src[step+0] + bias));\
             OP(dst[1], (A*src[1] + E*src[step+1] + bias));\
@@ -278,7 +379,7 @@ RV40_CHROMA_MC(put_, op_put)
 RV40_CHROMA_MC(avg_, op_avg)
 
 #define RV40_WEIGHT_FUNC(size) \
-static void rv40_weight_func_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, int stride)\
+static void rv40_weight_func_rnd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\
 {\
     int i, j;\
 \
@@ -289,85 +390,326 @@ static void rv40_weight_func_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src
         src2 += stride;\
         dst  += stride;\
     }\
+}\
+static void rv40_weight_func_nornd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\
+{\
+    int i, j;\
+\
+    for (j = 0; j < size; j++) {\
+        for (i = 0; i < size; i++)\
+            dst[i] = (w2 * src1[i] + w1 * src2[i] + 0x10) >> 5;\
+        src1 += stride;\
+        src2 += stride;\
+        dst  += stride;\
+    }\
 }
 
 RV40_WEIGHT_FUNC(16)
 RV40_WEIGHT_FUNC(8)
 
-av_cold void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp) {
-    c->put_pixels_tab[0][ 0] = dsp->put_h264_qpel_pixels_tab[0][0];
+/**
+ * dither values for deblocking filter - left/top values
+ */
+static const uint8_t rv40_dither_l[16] = {
+    0x40, 0x50, 0x20, 0x60, 0x30, 0x50, 0x40, 0x30,
+    0x50, 0x40, 0x50, 0x30, 0x60, 0x20, 0x50, 0x40
+};
+
+/**
+ * dither values for deblocking filter - right/bottom values
+ */
+static const uint8_t rv40_dither_r[16] = {
+    0x40, 0x30, 0x60, 0x20, 0x50, 0x30, 0x30, 0x40,
+    0x40, 0x40, 0x50, 0x30, 0x20, 0x60, 0x30, 0x40
+};
+
+#define CLIP_SYMM(a, b) av_clip(a, -(b), b)
+/**
+ * weaker deblocking very similar to the one described in 4.4.2 of JVT-A003r1
+ */
+static av_always_inline void rv40_weak_loop_filter(uint8_t *src,
+                                                   const int step,
+                                                   const ptrdiff_t stride,
+                                                   const int filter_p1,
+                                                   const int filter_q1,
+                                                   const int alpha,
+                                                   const int beta,
+                                                   const int lim_p0q0,
+                                                   const int lim_q1,
+                                                   const int lim_p1)
+{
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
+    int i, t, u, diff;
+
+    for (i = 0; i < 4; i++, src += stride) {
+        int diff_p1p0 = src[-2*step] - src[-1*step];
+        int diff_q1q0 = src[ 1*step] - src[ 0*step];
+        int diff_p1p2 = src[-2*step] - src[-3*step];
+        int diff_q1q2 = src[ 1*step] - src[ 2*step];
+
+        t = src[0*step] - src[-1*step];
+        if (!t)
+            continue;
+
+        u = (alpha * FFABS(t)) >> 7;
+        if (u > 3 - (filter_p1 && filter_q1))
+            continue;
+
+        t *= 1 << 2;
+        if (filter_p1 && filter_q1)
+            t += src[-2*step] - src[1*step];
+
+        diff = CLIP_SYMM((t + 4) >> 3, lim_p0q0);
+        src[-1*step] = cm[src[-1*step] + diff];
+        src[ 0*step] = cm[src[ 0*step] - diff];
+
+        if (filter_p1 && FFABS(diff_p1p2) <= beta) {
+            t = (diff_p1p0 + diff_p1p2 - diff) >> 1;
+            src[-2*step] = cm[src[-2*step] - CLIP_SYMM(t, lim_p1)];
+        }
+
+        if (filter_q1 && FFABS(diff_q1q2) <= beta) {
+            t = (diff_q1q0 + diff_q1q2 + diff) >> 1;
+            src[ 1*step] = cm[src[ 1*step] - CLIP_SYMM(t, lim_q1)];
+        }
+    }
+}
+
+static void rv40_h_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,
+                                    const int filter_p1, const int filter_q1,
+                                    const int alpha, const int beta,
+                                    const int lim_p0q0, const int lim_q1,
+                                    const int lim_p1)
+{
+    rv40_weak_loop_filter(src, stride, 1, filter_p1, filter_q1,
+                          alpha, beta, lim_p0q0, lim_q1, lim_p1);
+}
+
+static void rv40_v_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,
+                                    const int filter_p1, const int filter_q1,
+                                    const int alpha, const int beta,
+                                    const int lim_p0q0, const int lim_q1,
+                                    const int lim_p1)
+{
+    rv40_weak_loop_filter(src, 1, stride, filter_p1, filter_q1,
+                          alpha, beta, lim_p0q0, lim_q1, lim_p1);
+}
+
+static av_always_inline void rv40_strong_loop_filter(uint8_t *src,
+                                                     const int step,
+                                                     const ptrdiff_t stride,
+                                                     const int alpha,
+                                                     const int lims,
+                                                     const int dmode,
+                                                     const int chroma)
+{
+    int i;
+
+    for(i = 0; i < 4; i++, src += stride){
+        int sflag, p0, q0, p1, q1;
+        int t = src[0*step] - src[-1*step];
+
+        if (!t)
+            continue;
+
+        sflag = (alpha * FFABS(t)) >> 7;
+        if (sflag > 1)
+            continue;
+
+        p0 = (25*src[-3*step] + 26*src[-2*step] + 26*src[-1*step] +
+              26*src[ 0*step] + 25*src[ 1*step] +
+              rv40_dither_l[dmode + i]) >> 7;
+
+        q0 = (25*src[-2*step] + 26*src[-1*step] + 26*src[ 0*step] +
+              26*src[ 1*step] + 25*src[ 2*step] +
+              rv40_dither_r[dmode + i]) >> 7;
+
+        if (sflag) {
+            p0 = av_clip(p0, src[-1*step] - lims, src[-1*step] + lims);
+            q0 = av_clip(q0, src[ 0*step] - lims, src[ 0*step] + lims);
+        }
+
+        p1 = (25*src[-4*step] + 26*src[-3*step] + 26*src[-2*step] + 26*p0 +
+              25*src[ 0*step] + rv40_dither_l[dmode + i]) >> 7;
+        q1 = (25*src[-1*step] + 26*q0 + 26*src[ 1*step] + 26*src[ 2*step] +
+              25*src[ 3*step] + rv40_dither_r[dmode + i]) >> 7;
+
+        if (sflag) {
+            p1 = av_clip(p1, src[-2*step] - lims, src[-2*step] + lims);
+            q1 = av_clip(q1, src[ 1*step] - lims, src[ 1*step] + lims);
+        }
+
+        src[-2*step] = p1;
+        src[-1*step] = p0;
+        src[ 0*step] = q0;
+        src[ 1*step] = q1;
+
+        if(!chroma){
+            src[-3*step] = (25*src[-1*step] + 26*src[-2*step] +
+                            51*src[-3*step] + 26*src[-4*step] + 64) >> 7;
+            src[ 2*step] = (25*src[ 0*step] + 26*src[ 1*step] +
+                            51*src[ 2*step] + 26*src[ 3*step] + 64) >> 7;
+        }
+    }
+}
+
+static void rv40_h_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,
+                                      const int alpha, const int lims,
+                                      const int dmode, const int chroma)
+{
+    rv40_strong_loop_filter(src, stride, 1, alpha, lims, dmode, chroma);
+}
+
+static void rv40_v_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,
+                                      const int alpha, const int lims,
+                                      const int dmode, const int chroma)
+{
+    rv40_strong_loop_filter(src, 1, stride, alpha, lims, dmode, chroma);
+}
+
+static av_always_inline int rv40_loop_filter_strength(uint8_t *src,
+                                                      int step, ptrdiff_t stride,
+                                                      int beta, int beta2,
+                                                      int edge,
+                                                      int *p1, int *q1)
+{
+    int sum_p1p0 = 0, sum_q1q0 = 0, sum_p1p2 = 0, sum_q1q2 = 0;
+    int strong0 = 0, strong1 = 0;
+    uint8_t *ptr;
+    int i;
+
+    for (i = 0, ptr = src; i < 4; i++, ptr += stride) {
+        sum_p1p0 += ptr[-2*step] - ptr[-1*step];
+        sum_q1q0 += ptr[ 1*step] - ptr[ 0*step];
+    }
+
+    *p1 = FFABS(sum_p1p0) < (beta << 2);
+    *q1 = FFABS(sum_q1q0) < (beta << 2);
+
+    if(!*p1 && !*q1)
+        return 0;
+
+    if (!edge)
+        return 0;
+
+    for (i = 0, ptr = src; i < 4; i++, ptr += stride) {
+        sum_p1p2 += ptr[-2*step] - ptr[-3*step];
+        sum_q1q2 += ptr[ 1*step] - ptr[ 2*step];
+    }
+
+    strong0 = *p1 && (FFABS(sum_p1p2) < beta2);
+    strong1 = *q1 && (FFABS(sum_q1q2) < beta2);
+
+    return strong0 && strong1;
+}
+
+static int rv40_h_loop_filter_strength(uint8_t *src, ptrdiff_t stride,
+                                       int beta, int beta2, int edge,
+                                       int *p1, int *q1)
+{
+    return rv40_loop_filter_strength(src, stride, 1, beta, beta2, edge, p1, q1);
+}
+
+static int rv40_v_loop_filter_strength(uint8_t *src, ptrdiff_t stride,
+                                       int beta, int beta2, int edge,
+                                       int *p1, int *q1)
+{
+    return rv40_loop_filter_strength(src, 1, stride, beta, beta2, edge, p1, q1);
+}
+
+av_cold void ff_rv40dsp_init(RV34DSPContext *c)
+{
+    H264QpelContext qpel;
+
+    ff_rv34dsp_init(c);
+    ff_h264qpel_init(&qpel, 8);
+
+    c->put_pixels_tab[0][ 0] = qpel.put_h264_qpel_pixels_tab[0][0];
     c->put_pixels_tab[0][ 1] = put_rv40_qpel16_mc10_c;
-    c->put_pixels_tab[0][ 2] = dsp->put_h264_qpel_pixels_tab[0][2];
+    c->put_pixels_tab[0][ 2] = qpel.put_h264_qpel_pixels_tab[0][2];
     c->put_pixels_tab[0][ 3] = put_rv40_qpel16_mc30_c;
     c->put_pixels_tab[0][ 4] = put_rv40_qpel16_mc01_c;
     c->put_pixels_tab[0][ 5] = put_rv40_qpel16_mc11_c;
     c->put_pixels_tab[0][ 6] = put_rv40_qpel16_mc21_c;
     c->put_pixels_tab[0][ 7] = put_rv40_qpel16_mc31_c;
-    c->put_pixels_tab[0][ 8] = dsp->put_h264_qpel_pixels_tab[0][8];
+    c->put_pixels_tab[0][ 8] = qpel.put_h264_qpel_pixels_tab[0][8];
     c->put_pixels_tab[0][ 9] = put_rv40_qpel16_mc12_c;
     c->put_pixels_tab[0][10] = put_rv40_qpel16_mc22_c;
     c->put_pixels_tab[0][11] = put_rv40_qpel16_mc32_c;
     c->put_pixels_tab[0][12] = put_rv40_qpel16_mc03_c;
     c->put_pixels_tab[0][13] = put_rv40_qpel16_mc13_c;
     c->put_pixels_tab[0][14] = put_rv40_qpel16_mc23_c;
-    c->put_pixels_tab[0][15] = ff_put_rv40_qpel16_mc33_c;
-    c->avg_pixels_tab[0][ 0] = dsp->avg_h264_qpel_pixels_tab[0][0];
+    c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_c;
+    c->avg_pixels_tab[0][ 0] = qpel.avg_h264_qpel_pixels_tab[0][0];
     c->avg_pixels_tab[0][ 1] = avg_rv40_qpel16_mc10_c;
-    c->avg_pixels_tab[0][ 2] = dsp->avg_h264_qpel_pixels_tab[0][2];
+    c->avg_pixels_tab[0][ 2] = qpel.avg_h264_qpel_pixels_tab[0][2];
     c->avg_pixels_tab[0][ 3] = avg_rv40_qpel16_mc30_c;
     c->avg_pixels_tab[0][ 4] = avg_rv40_qpel16_mc01_c;
     c->avg_pixels_tab[0][ 5] = avg_rv40_qpel16_mc11_c;
     c->avg_pixels_tab[0][ 6] = avg_rv40_qpel16_mc21_c;
     c->avg_pixels_tab[0][ 7] = avg_rv40_qpel16_mc31_c;
-    c->avg_pixels_tab[0][ 8] = dsp->avg_h264_qpel_pixels_tab[0][8];
+    c->avg_pixels_tab[0][ 8] = qpel.avg_h264_qpel_pixels_tab[0][8];
     c->avg_pixels_tab[0][ 9] = avg_rv40_qpel16_mc12_c;
     c->avg_pixels_tab[0][10] = avg_rv40_qpel16_mc22_c;
     c->avg_pixels_tab[0][11] = avg_rv40_qpel16_mc32_c;
     c->avg_pixels_tab[0][12] = avg_rv40_qpel16_mc03_c;
     c->avg_pixels_tab[0][13] = avg_rv40_qpel16_mc13_c;
     c->avg_pixels_tab[0][14] = avg_rv40_qpel16_mc23_c;
-    c->avg_pixels_tab[0][15] = ff_avg_rv40_qpel16_mc33_c;
-    c->put_pixels_tab[1][ 0] = dsp->put_h264_qpel_pixels_tab[1][0];
+    c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_c;
+    c->put_pixels_tab[1][ 0] = qpel.put_h264_qpel_pixels_tab[1][0];
     c->put_pixels_tab[1][ 1] = put_rv40_qpel8_mc10_c;
-    c->put_pixels_tab[1][ 2] = dsp->put_h264_qpel_pixels_tab[1][2];
+    c->put_pixels_tab[1][ 2] = qpel.put_h264_qpel_pixels_tab[1][2];
     c->put_pixels_tab[1][ 3] = put_rv40_qpel8_mc30_c;
     c->put_pixels_tab[1][ 4] = put_rv40_qpel8_mc01_c;
     c->put_pixels_tab[1][ 5] = put_rv40_qpel8_mc11_c;
     c->put_pixels_tab[1][ 6] = put_rv40_qpel8_mc21_c;
     c->put_pixels_tab[1][ 7] = put_rv40_qpel8_mc31_c;
-    c->put_pixels_tab[1][ 8] = dsp->put_h264_qpel_pixels_tab[1][8];
+    c->put_pixels_tab[1][ 8] = qpel.put_h264_qpel_pixels_tab[1][8];
     c->put_pixels_tab[1][ 9] = put_rv40_qpel8_mc12_c;
     c->put_pixels_tab[1][10] = put_rv40_qpel8_mc22_c;
     c->put_pixels_tab[1][11] = put_rv40_qpel8_mc32_c;
     c->put_pixels_tab[1][12] = put_rv40_qpel8_mc03_c;
     c->put_pixels_tab[1][13] = put_rv40_qpel8_mc13_c;
     c->put_pixels_tab[1][14] = put_rv40_qpel8_mc23_c;
-    c->put_pixels_tab[1][15] = ff_put_rv40_qpel8_mc33_c;
-    c->avg_pixels_tab[1][ 0] = dsp->avg_h264_qpel_pixels_tab[1][0];
+    c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_c;
+    c->avg_pixels_tab[1][ 0] = qpel.avg_h264_qpel_pixels_tab[1][0];
     c->avg_pixels_tab[1][ 1] = avg_rv40_qpel8_mc10_c;
-    c->avg_pixels_tab[1][ 2] = dsp->avg_h264_qpel_pixels_tab[1][2];
+    c->avg_pixels_tab[1][ 2] = qpel.avg_h264_qpel_pixels_tab[1][2];
     c->avg_pixels_tab[1][ 3] = avg_rv40_qpel8_mc30_c;
     c->avg_pixels_tab[1][ 4] = avg_rv40_qpel8_mc01_c;
     c->avg_pixels_tab[1][ 5] = avg_rv40_qpel8_mc11_c;
     c->avg_pixels_tab[1][ 6] = avg_rv40_qpel8_mc21_c;
     c->avg_pixels_tab[1][ 7] = avg_rv40_qpel8_mc31_c;
-    c->avg_pixels_tab[1][ 8] = dsp->avg_h264_qpel_pixels_tab[1][8];
+    c->avg_pixels_tab[1][ 8] = qpel.avg_h264_qpel_pixels_tab[1][8];
     c->avg_pixels_tab[1][ 9] = avg_rv40_qpel8_mc12_c;
     c->avg_pixels_tab[1][10] = avg_rv40_qpel8_mc22_c;
     c->avg_pixels_tab[1][11] = avg_rv40_qpel8_mc32_c;
     c->avg_pixels_tab[1][12] = avg_rv40_qpel8_mc03_c;
     c->avg_pixels_tab[1][13] = avg_rv40_qpel8_mc13_c;
     c->avg_pixels_tab[1][14] = avg_rv40_qpel8_mc23_c;
-    c->avg_pixels_tab[1][15] = ff_avg_rv40_qpel8_mc33_c;
+    c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_c;
 
     c->put_chroma_pixels_tab[0] = put_rv40_chroma_mc8_c;
     c->put_chroma_pixels_tab[1] = put_rv40_chroma_mc4_c;
     c->avg_chroma_pixels_tab[0] = avg_rv40_chroma_mc8_c;
     c->avg_chroma_pixels_tab[1] = avg_rv40_chroma_mc4_c;
 
-    c->rv40_weight_pixels_tab[0] = rv40_weight_func_16;
-    c->rv40_weight_pixels_tab[1] = rv40_weight_func_8;
+    c->rv40_weight_pixels_tab[0][0] = rv40_weight_func_rnd_16;
+    c->rv40_weight_pixels_tab[0][1] = rv40_weight_func_rnd_8;
+    c->rv40_weight_pixels_tab[1][0] = rv40_weight_func_nornd_16;
+    c->rv40_weight_pixels_tab[1][1] = rv40_weight_func_nornd_8;
+
+    c->rv40_weak_loop_filter[0]     = rv40_h_weak_loop_filter;
+    c->rv40_weak_loop_filter[1]     = rv40_v_weak_loop_filter;
+    c->rv40_strong_loop_filter[0]   = rv40_h_strong_loop_filter;
+    c->rv40_strong_loop_filter[1]   = rv40_v_strong_loop_filter;
+    c->rv40_loop_filter_strength[0] = rv40_h_loop_filter_strength;
+    c->rv40_loop_filter_strength[1] = rv40_v_loop_filter_strength;
 
-    if (HAVE_MMX)
-        ff_rv40dsp_init_x86(c, dsp);
+    if (ARCH_AARCH64)
+        ff_rv40dsp_init_aarch64(c);
+    if (ARCH_ARM)
+        ff_rv40dsp_init_arm(c);
+    if (ARCH_X86)
+        ff_rv40dsp_init_x86(c);
 }