vp9/x86: invert hu_ipred left array ordering.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/vp9.c

@@ -2347,6 +2347,7 @@ static av_always_inline int check_intra_mode(VP9Context *s, int mode, uint8_t **
         uint8_t needs_top:1;
         uint8_t needs_topleft:1;
         uint8_t needs_topright:1;
+        uint8_t invert_left:1;
     } edges[N_INTRA_PRED_MODES] = {
         [VERT_PRED]            = { .needs_top  = 1 },
         [HOR_PRED]             = { .needs_left = 1 },
@@ -2356,7 +2357,7 @@ static av_always_inline int check_intra_mode(VP9Context *s, int mode, uint8_t **
         [VERT_RIGHT_PRED]      = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
         [HOR_DOWN_PRED]        = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
         [VERT_LEFT_PRED]       = { .needs_top  = 1, .needs_topright = 1 },
-        [HOR_UP_PRED]          = { .needs_left = 1 },
+        [HOR_UP_PRED]          = { .needs_left = 1, .invert_left = 1 },
         [TM_VP8_PRED]          = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
         [LEFT_DC_PRED]         = { .needs_left = 1 },
         [TOP_DC_PRED]          = { .needs_top  = 1 },
@@ -2429,13 +2430,24 @@ static av_always_inline int check_intra_mode(VP9Context *s, int mode, uint8_t **
             uint8_t *dst = x == 0 ? dst_edge : dst_inner;
             ptrdiff_t stride = x == 0 ? stride_edge : stride_inner;
 
-            if (n_px_need <= n_px_have) {
-                for (i = 0; i < n_px_need; i++)
-                    l[n_px_need - 1 - i] = dst[i * stride - 1];
+            if (edges[mode].invert_left) {
+                if (n_px_need <= n_px_have) {
+                    for (i = 0; i < n_px_need; i++)
+                        l[i] = dst[i * stride - 1];
+                } else {
+                    for (i = 0; i < n_px_have; i++)
+                        l[i] = dst[i * stride - 1];
+                    memset(&l[n_px_have], l[n_px_have - 1], n_px_need - n_px_have);
+                }
             } else {
-                for (i = 0; i < n_px_have; i++)
-                    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);
+                if (n_px_need <= n_px_have) {
+                    for (i = 0; i < n_px_need; i++)
+                        l[n_px_need - 1 - i] = dst[i * stride - 1];
+                } else {
+                    for (i = 0; i < n_px_have; i++)
+                        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);

libavcodec/vp9dsp.c

@@ -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[3], l1 = left[2], l2 = left[1], l3 = left[0];
+    int l0 = left[0], l1 = left[1], l2 = left[2], l3 = left[3];
 
     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[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[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[size*2 - 4] = (left[1] + left[0] + 1) >> 1; \
-    v[size*2 - 3] = (left[1] + left[0] * 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; \
 \
     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[0], \
+        memset(dst + j*stride + size*2 - 2 - j*2, left[size - 1], \
                2 + j*2 - size); \
     } \
 }

libavcodec/x86/vp9intrapred.asm

@@ -63,13 +63,11 @@ pb_6xm1_246_8toE: times 6 db -1
 pb_6xm1_BDF_0to6: times 6 db -1
                   db 11, 13, 15, 0, 1, 2, 3, 4, 5, 6
 pb_02468ACE_13579BDF: db 0, 2, 4, 6, 8, 10, 12, 14, 1, 3, 5, 7, 9, 11, 13, 15
-pb_7to1_9x0:  db 7, 6, 5, 4
-pb_3to1_5x0:  db 3, 2, 1
-              times 9 db 0
-pb_Fto0:      db 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
 
 pb_2:  times 32 db 2
 pb_15: times 16 db 15
+pb_0to2_5x3: db 0, 1, 2
+             times 5 db 3
 
 cextern pb_1
 cextern pb_3
@@ -1420,7 +1418,7 @@ HD_XMM_FUNCS avx
 INIT_MMX ssse3
 cglobal vp9_ipred_hu_4x4, 3, 3, 0, dst, stride, l
     movd                    m0, [lq]
-    pshufb                  m0, [pb_3to1_5x0]
+    pshufb                  m0, [pb_0to2_5x3]
     psrlq                   m1, m0, 8
     psrlq                   m2, m1, 8
     LOWPASS                  2,  1, 0, 3
@@ -1441,7 +1439,7 @@ cglobal vp9_ipred_hu_4x4, 3, 3, 0, dst, stride, l
 INIT_XMM %1
 cglobal vp9_ipred_hu_8x8, 3, 4, 4, dst, stride, l
     movq                    m0, [lq]
-    pshufb                  m0, [pb_7to1_9x0]
+    pshufb                  m0, [pb_0to6_9x7]
     psrldq                  m1, m0, 1
     psrldq                  m2, m1, 1
     LOWPASS                  2,  1, 0, 3
@@ -1466,7 +1464,6 @@ cglobal vp9_ipred_hu_8x8, 3, 4, 4, dst, stride, l
 INIT_XMM %1
 cglobal vp9_ipred_hu_16x16, 3, 4, 5, dst, stride, l
     mova                    m0, [lq]
-    pshufb                  m0, [pb_Fto0]
     mova                    m3, [pb_2toE_3xF]
     pshufb                  m1, m0, [pb_1toE_2xF]
     pshufb                  m2, m0, m3
@@ -1494,12 +1491,9 @@ cglobal vp9_ipred_hu_16x16, 3, 4, 5, dst, stride, l
 
 INIT_XMM %1
 cglobal vp9_ipred_hu_32x32, 3, 7, 7, dst, stride, l
-    mova                    m0, [lq]
-    mova                    m1, [lq+16]
-    mova                    m2, [pb_Fto0]
+    mova                    m1, [lq]
+    mova                    m0, [lq+16]
     mova                    m4, [pb_2toE_3xF]
-    pshufb                  m0, m2
-    pshufb                  m1, m2
     palignr                 m2, m0, m1,  1
     palignr                 m3, m0, m1,  2
     LOWPASS                  3,  2,  1,  5