vp9: invert order in l[] intra prediction array.

The directional intra predictors either don't care about order (dc, h,
dc_left, tm), or they prefer inverted order (vr, dr, hd). This allows
more efficient SIMD implementations.

2 files changed, 42 insertions, 42 deletions

diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c
index e79bddd1ba..ea2a17bf95 100644
--- a/libavcodec/vp9.c
+++ b/libavcodec/vp9.c
@@ -2379,11 +2379,11 @@ static av_always_inline int check_intra_mode(VP9Context *s, int mode, uint8_t **
 
             if (n_px_need <= n_px_have) {
                 for (i = 0; i < n_px_need; i++)
-                    l[i] = dst[i * stride - 1];
+                    l[n_px_need - 1 - i] = dst[i * stride - 1];
             } else {
                 for (i = 0; i < n_px_have; i++)
-                    l[i] = dst[i * stride - 1];
-                memset(&l[i], l[i - 1], n_px_need - n_px_have);
+                    l[n_px_need - 1 - i] = dst[i * stride - 1];
+                memset(l, l[n_px_need - n_px_have], n_px_need - n_px_have);
             }
         } else {
             memset(l, 129, 4 << tx);
@@ -2405,6 +2405,8 @@ static void intra_recon(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
     int tx = 4 * s->lossless + b->tx, uvtx = b->uvtx + 4 * s->lossless;
     int uvstep1d = 1 << b->uvtx, p;
     uint8_t *dst = s->dst[0], *dst_r = s->frames[CUR_FRAME].tf.f->data[0] + y_off;
+    LOCAL_ALIGNED_16(uint8_t, a_buf, [48]);
+    LOCAL_ALIGNED_16(uint8_t, l, [32]);
 
     for (n = 0, y = 0; y < end_y; y += step1d) {
         uint8_t *ptr = dst, *ptr_r = dst_r;
@@ -2412,8 +2414,7 @@ static void intra_recon(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
                                ptr_r += 4 * step1d, n += step) {
             int mode = b->mode[b->bs > BS_8x8 && b->tx == TX_4X4 ?
                                y * 2 + x : 0];
-            LOCAL_ALIGNED_16(uint8_t, a_buf, [48]);
-            uint8_t *a = &a_buf[16], l[32];
+            uint8_t *a = &a_buf[16];
             enum TxfmType txtp = vp9_intra_txfm_type[mode];
             int eob = b->skip ? 0 : b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
 
@@ -2444,8 +2445,7 @@ static void intra_recon(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
             for (x = 0; x < end_x; x += uvstep1d, ptr += 4 * uvstep1d,
                                    ptr_r += 4 * uvstep1d, n += step) {
                 int mode = b->uvmode;
-                LOCAL_ALIGNED_16(uint8_t, a_buf, [48]);
-                uint8_t *a = &a_buf[16], l[32];
+                uint8_t *a = &a_buf[16];
                 int eob = b->skip ? 0 : b->uvtx > TX_8X8 ? AV_RN16A(&s->uveob[p][n]) : s->uveob[p][n];
 
                 mode = check_intra_mode(s, mode, &a, ptr_r,
diff --git a/libavcodec/vp9dsp.c b/libavcodec/vp9dsp.c
index ec0b411c93..e2f99f72c8 100644
--- a/libavcodec/vp9dsp.c
+++ b/libavcodec/vp9dsp.c
@@ -84,10 +84,10 @@ static void vert_32x32_c(uint8_t *dst, ptrdiff_t stride,
 static void hor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                       const uint8_t *left, const uint8_t *top)
 {
-    AV_WN32A(dst + stride * 0, left[0] * 0x01010101U);
-    AV_WN32A(dst + stride * 1, left[1] * 0x01010101U);
-    AV_WN32A(dst + stride * 2, left[2] * 0x01010101U);
-    AV_WN32A(dst + stride * 3, left[3] * 0x01010101U);
+    AV_WN32A(dst + stride * 0, left[3] * 0x01010101U);
+    AV_WN32A(dst + stride * 1, left[2] * 0x01010101U);
+    AV_WN32A(dst + stride * 2, left[1] * 0x01010101U);
+    AV_WN32A(dst + stride * 3, left[0] * 0x01010101U);
 }
 
 static void hor_8x8_c(uint8_t *dst, ptrdiff_t stride,
@@ -96,7 +96,7 @@ static void hor_8x8_c(uint8_t *dst, ptrdiff_t stride,
     int y;
 
     for (y = 0; y < 8; y++) {
-        AV_WN64A(dst, left[y] * 0x0101010101010101ULL);
+        AV_WN64A(dst, left[7 - y] * 0x0101010101010101ULL);
         dst += stride;
     }
 }
@@ -107,7 +107,7 @@ static void hor_16x16_c(uint8_t *dst, ptrdiff_t stride,
     int y;
 
     for (y = 0; y < 16; y++) {
-        uint64_t p8 = left[y] * 0x0101010101010101ULL;
+        uint64_t p8 = left[15 - y] * 0x0101010101010101ULL;
 
         AV_WN64A(dst + 0, p8);
         AV_WN64A(dst + 8, p8);
@@ -121,7 +121,7 @@ static void hor_32x32_c(uint8_t *dst, ptrdiff_t stride,
     int y;
 
     for (y = 0; y < 32; y++) {
-        uint64_t p8 = left[y] * 0x0101010101010101ULL;
+        uint64_t p8 = left[31 - y] * 0x0101010101010101ULL;
 
         AV_WN64A(dst +  0, p8);
         AV_WN64A(dst +  8, p8);
@@ -137,7 +137,7 @@ static void tm_4x4_c(uint8_t *dst, ptrdiff_t stride,
     int y, tl = top[-1];
 
     for (y = 0; y < 4; y++) {
-        int l_m_tl = left[y] - tl;
+        int l_m_tl = left[3 - y] - tl;
 
         dst[0] = av_clip_uint8(top[0] + l_m_tl);
         dst[1] = av_clip_uint8(top[1] + l_m_tl);
@@ -153,7 +153,7 @@ static void tm_8x8_c(uint8_t *dst, ptrdiff_t stride,
     int y, tl = top[-1];
 
     for (y = 0; y < 8; y++) {
-        int l_m_tl = left[y] - tl;
+        int l_m_tl = left[7 - y] - tl;
 
         dst[0] = av_clip_uint8(top[0] + l_m_tl);
         dst[1] = av_clip_uint8(top[1] + l_m_tl);
@@ -173,7 +173,7 @@ static void tm_16x16_c(uint8_t *dst, ptrdiff_t stride,
     int y, tl = top[-1];
 
     for (y = 0; y < 16; y++) {
-        int l_m_tl = left[y] - tl;
+        int l_m_tl = left[15 - y] - tl;
 
         dst[ 0] = av_clip_uint8(top[ 0] + l_m_tl);
         dst[ 1] = av_clip_uint8(top[ 1] + l_m_tl);
@@ -201,7 +201,7 @@ static void tm_32x32_c(uint8_t *dst, ptrdiff_t stride,
     int y, tl = top[-1];
 
     for (y = 0; y < 32; y++) {
-        int l_m_tl = left[y] - tl;
+        int l_m_tl = left[31 - y] - tl;
 
         dst[ 0] = av_clip_uint8(top[ 0] + l_m_tl);
         dst[ 1] = av_clip_uint8(top[ 1] + l_m_tl);
@@ -613,7 +613,7 @@ static void diag_downright_4x4_c(uint8_t *dst, ptrdiff_t stride,
                                  const uint8_t *left, const uint8_t *top)
 {
     int tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],
-        l0 = left[0], l1 = left[1], l2 = left[2], l3 = left[3];
+        l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0];
 
     DST(0,3) = (l1 + l2 * 2 + l3 + 2) >> 2;
     DST(0,2) = DST(1,3) = (l0 + l1 * 2 + l2 + 2) >> 2;
@@ -632,11 +632,11 @@ static void diag_downright_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
     uint8_t v[size + size - 1]; \
 \
     for (i = 0; i < size - 2; i++) { \
-        v[i           ] = (left[size - 1 - i] + left[size - 2 - i] * 2 + left[size - 3 - i] + 2) >> 2; \
-        v[size + 1 + i] = (top[i]             + top[i + 1]         * 2 + top[i + 2]         + 2) >> 2; \
+        v[i           ] = (left[i] + left[i + 1] * 2 + left[i + 2] + 2) >> 2; \
+        v[size + 1 + i] = (top[i]  + top[i + 1]  * 2 + top[i + 2]  + 2) >> 2; \
     } \
-    v[size - 2] = (left[1] + left[0] * 2 + top[-1] + 2) >> 2; \
-    v[size - 1] = (left[0] + top[-1] * 2 + top[ 0] + 2) >> 2; \
+    v[size - 2] = (left[size - 2] + left[size - 1] * 2 + top[-1] + 2) >> 2; \
+    v[size - 1] = (left[size - 1] + top[-1] * 2 + top[ 0] + 2) >> 2; \
     v[size    ] = (top[-1] + top[0]  * 2 + top[ 1] + 2) >> 2; \
 \
     for (j = 0; j < size; j++) \
@@ -651,7 +651,7 @@ static void vert_right_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *left, const uint8_t *top)
 {
     int tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],
-        l0 = left[0], l1 = left[1], l2 = left[2];
+        l0 = left[3], l1 = left[2], l2 = left[1];
 
     DST(0,3) = (l0 + l1 * 2 + l2 + 2) >> 2;
     DST(0,2) = (tl + l0 * 2 + l1 + 2) >> 2;
@@ -673,14 +673,14 @@ static void vert_right_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
     uint8_t ve[size + size/2 - 1], vo[size + size/2 - 1]; \
 \
     for (i = 0; i < size/2 - 2; i++) { \
-        vo[i] = (left[size - 4 - i*2] + left[size - 3 - i*2] * 2 + left[size - 2 - i*2] + 2) >> 2; \
-        ve[i] = (left[size - 5 - i*2] + left[size - 4 - i*2] * 2 + left[size - 3 - i*2] + 2) >> 2; \
+        vo[i] = (left[i*2 + 3] + left[i*2 + 2] * 2 + left[i*2 + 1] + 2) >> 2; \
+        ve[i] = (left[i*2 + 4] + left[i*2 + 3] * 2 + left[i*2 + 2] + 2) >> 2; \
     } \
-    vo[size/2 - 2] = (left[0] + left[1] * 2 + left[2] + 2) >> 2; \
-    ve[size/2 - 2] = (top[-1] + left[0] * 2 + left[1] + 2) >> 2; \
+    vo[size/2 - 2] = (left[size - 1] + left[size - 2] * 2 + left[size - 3] + 2) >> 2; \
+    ve[size/2 - 2] = (top[-1] + left[size - 1] * 2 + left[size - 2] + 2) >> 2; \
 \
     ve[size/2 - 1] = (top[-1] + top[0] + 1) >> 1; \
-    vo[size/2 - 1] = (left[0] + top[-1] * 2 + top[0] + 2) >> 2; \
+    vo[size/2 - 1] = (left[size - 1] + top[-1] * 2 + top[0] + 2) >> 2; \
     for (i = 0; i < size - 1; i++) { \
         ve[size/2 + i] = (top[i] + top[i + 1] + 1) >> 1; \
         vo[size/2 + i] = (top[i - 1] + top[i] * 2 + top[i + 1] + 2) >> 2; \
@@ -699,7 +699,7 @@ def_vert_right(32)
 static void hor_down_4x4_c(uint8_t *dst, ptrdiff_t stride,
                            const uint8_t *left, const uint8_t *top)
 {
-    int l0 = left[0], l1 = left[1], l2 = left[2], l3 = left[3],
+    int l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0],
         tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2];
 
     DST(2,0) = (tl + a0 * 2 + a1 + 2) >> 2;
@@ -722,14 +722,14 @@ static void hor_down_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
     uint8_t v[size * 3 - 2]; \
 \
     for (i = 0; i < size - 2; i++) { \
-        v[i*2       ] = (left[size - 2 - i] + left[size - 1 - i] + 1) >> 1; \
-        v[i*2    + 1] = (left[size - 3 - i] + left[size - 2 - i] * 2 + left[size - 1 - i] + 2) >> 2; \
+        v[i*2       ] = (left[i + 1] + left[i + 0] + 1) >> 1; \
+        v[i*2    + 1] = (left[i + 2] + left[i + 1] * 2 + left[i + 0] + 2) >> 2; \
         v[size*2 + i] = (top[i - 1] + top[i] * 2 + top[i + 1] + 2) >> 2; \
     } \
-    v[size*2 - 2] = (top[-1] + left[0] + 1) >> 1; \
-    v[size*2 - 4] = (left[0] + left[1] + 1) >> 1; \
-    v[size*2 - 1] = (top[0]  + top[-1] * 2 + left[0] + 2) >> 2; \
-    v[size*2 - 3] = (top[-1] + left[0] * 2 + left[1] + 2) >> 2; \
+    v[size*2 - 2] = (top[-1] + left[size - 1] + 1) >> 1; \
+    v[size*2 - 4] = (left[size - 1] + left[size - 2] + 1) >> 1; \
+    v[size*2 - 1] = (top[0]  + top[-1] * 2 + left[size - 1] + 2) >> 2; \
+    v[size*2 - 3] = (top[-1] + left[size - 1] * 2 + left[size - 2] + 2) >> 2; \
 \
     for (j = 0; j < size; j++) \
         memcpy(dst + j*stride, v + size*2 - 2 - j*2, size); \
@@ -786,7 +786,7 @@ def_vert_left(32)
 static void hor_up_4x4_c(uint8_t *dst, ptrdiff_t stride,
                          const uint8_t *left, const uint8_t *top)
 {
-    int l0 = left[0], l1 = left[1], l2 = left[2], l3 = left[3];
+    int l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0];
 
     DST(0,0) = (l0 + l1 + 1) >> 1;
     DST(1,0) = (l0 + l1 * 2 + l2 + 2) >> 2;
@@ -805,17 +805,17 @@ static void hor_up_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
     uint8_t v[size*2 - 2]; \
 \
     for (i = 0; i < size - 2; i++) { \
-        v[i*2    ] = (left[i] + left[i + 1] + 1) >> 1; \
-        v[i*2 + 1] = (left[i] + left[i + 1] * 2 + left[i + 2] + 2) >> 2; \
+        v[i*2    ] = (left[size - i - 1] + left[size - i - 2] + 1) >> 1; \
+        v[i*2 + 1] = (left[size - i - 1] + left[size - i - 2] * 2 + left[size - i - 3] + 2) >> 2; \
     } \
-    v[size*2 - 4] = (left[size - 2] + left[size - 1] + 1) >> 1; \
-    v[size*2 - 3] = (left[size - 2] + left[size - 1] * 3 + 2) >> 2; \
+    v[size*2 - 4] = (left[1] + left[0] + 1) >> 1; \
+    v[size*2 - 3] = (left[1] + left[0] * 3 + 2) >> 2; \
 \
     for (j = 0; j < size / 2; j++) \
         memcpy(dst + j*stride, v + j*2, size); \
     for (j = size / 2; j < size; j++) { \
         memcpy(dst + j*stride, v + j*2, size*2 - 2 - j*2); \
-        memset(dst + j*stride + size*2 - 2 - j*2, left[size - 1], \
+        memset(dst + j*stride + size*2 - 2 - j*2, left[0], \
                2 + j*2 - size); \
     } \
 }