summaryrefslogtreecommitdiff
path: root/vp9/common
diff options
context:
space:
mode:
Diffstat (limited to 'vp9/common')
-rw-r--r--vp9/common/arm/neon/vp9_reconintra_neon_asm.asm19
-rw-r--r--vp9/common/vp9_blockd.h10
-rw-r--r--vp9/common/vp9_onyxc_int.h8
-rw-r--r--vp9/common/vp9_reconintra.c118
4 files changed, 76 insertions, 79 deletions
diff --git a/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm b/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
index d4f6d9b48..14f574a50 100644
--- a/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
+++ b/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
@@ -298,8 +298,7 @@ loop_h
|vp9_tm_predictor_4x4_neon| PROC
; Load ytop_left = above[-1];
sub r12, r2, #1
- ldrb r12, [r12]
- vdup.u8 d0, r12
+ vld1.u8 {d0[]}, [r12]
; Load above 4 pixels
vld1.32 {d2[0]}, [r2]
@@ -309,10 +308,10 @@ loop_h
; Load left row by row and compute left + (above - ytop_left)
; 1st row and 2nd row
- ldrb r12, [r3], #1
- ldrb r2, [r3], #1
- vdup.u16 q1, r12
- vdup.u16 q2, r2
+ vld1.u8 {d2[]}, [r3]!
+ vld1.u8 {d4[]}, [r3]!
+ vmovl.u8 q1, d2
+ vmovl.u8 q2, d4
vadd.s16 q1, q1, q3
vadd.s16 q2, q2, q3
vqmovun.s16 d0, q1
@@ -321,10 +320,10 @@ loop_h
vst1.32 {d1[0]}, [r0], r1
; 3rd row and 4th row
- ldrb r12, [r3], #1
- ldrb r2, [r3], #1
- vdup.u16 q1, r12
- vdup.u16 q2, r2
+ vld1.u8 {d2[]}, [r3]!
+ vld1.u8 {d4[]}, [r3]
+ vmovl.u8 q1, d2
+ vmovl.u8 q2, d4
vadd.s16 q1, q1, q3
vadd.s16 q2, q2, q3
vqmovun.s16 d0, q1
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 319d34832..e53e15da9 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -124,9 +124,12 @@ struct macroblockd_plane {
int subsampling_y;
struct buf_2d dst;
struct buf_2d pre[2];
- const int16_t *dequant;
ENTROPY_CONTEXT *above_context;
ENTROPY_CONTEXT *left_context;
+ int16_t seg_dequant[MAX_SEGMENTS][2];
+
+ // encoder
+ const int16_t *dequant;
};
#define BLOCK_OFFSET(x, i) ((x) + (i) * 16)
@@ -141,7 +144,7 @@ typedef struct RefBuffer {
typedef struct macroblockd {
struct macroblockd_plane plane[MAX_MB_PLANE];
-
+ FRAME_COUNTS *counts;
int mi_stride;
MODE_INFO **mi;
@@ -159,6 +162,9 @@ typedef struct macroblockd {
int mb_to_top_edge;
int mb_to_bottom_edge;
+ FRAME_CONTEXT *fc;
+ int frame_parallel_decoding_mode;
+
/* pointers to reference frames */
RefBuffer *block_refs[2];
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index 188b03d41..045d35049 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -341,6 +341,14 @@ static INLINE void init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd) {
xd->plane[i].dqcoeff = xd->dqcoeff;
xd->above_context[i] = cm->above_context +
i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);
+
+ if (xd->plane[i].plane_type == PLANE_TYPE_Y) {
+ memcpy(xd->plane[i].seg_dequant, cm->y_dequant, sizeof(cm->y_dequant));
+ } else {
+ memcpy(xd->plane[i].seg_dequant, cm->uv_dequant, sizeof(cm->uv_dequant));
+ }
+ xd->fc = cm->fc;
+ xd->frame_parallel_decoding_mode = cm->frame_parallel_decoding_mode;
}
xd->above_seg_context = cm->above_seg_context;
diff --git a/vp9/common/vp9_reconintra.c b/vp9/common/vp9_reconintra.c
index f7c01fd1d..80034e4db 100644
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -102,6 +102,10 @@ static const uint8_t extend_modes[INTRA_MODES] = {
intra_pred_sized(type, 32)
#endif // CONFIG_VP9_HIGHBITDEPTH
+#define DST(x, y) dst[(x) + (y) * stride]
+#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
+#define AVG2(a, b) (((a) + (b) + 1) >> 1)
+
#if CONFIG_VP9_HIGHBITDEPTH
static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride,
int bs, const uint16_t *above,
@@ -112,18 +116,16 @@ static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride,
// First column.
for (r = 0; r < bs - 1; ++r) {
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+ dst[r * stride] = AVG2(left[r], left[r + 1]);
}
dst[(bs - 1) * stride] = left[bs - 1];
dst++;
// Second column.
for (r = 0; r < bs - 2; ++r) {
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
- left[r + 2], 2);
+ dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
}
- dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
- left[bs - 1] * 3, 2);
+ dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
dst[(bs - 1) * stride] = left[bs - 1];
dst++;
@@ -145,11 +147,9 @@ static INLINE void highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride,
(void) bd;
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
- dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
- above[r/2 + c + 1] * 2 +
- above[r/2 + c + 2], 2)
- : ROUND_POWER_OF_TWO(above[r/2 + c] +
- above[r/2 + c + 1], 1);
+ dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
+ above[(r >> 1) + c + 2])
+ : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
}
dst += stride;
}
@@ -163,9 +163,8 @@ static INLINE void highbd_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
(void) bd;
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
- dst[c] = r + c + 2 < bs * 2 ? ROUND_POWER_OF_TWO(above[r + c] +
- above[r + c + 1] * 2 +
- above[r + c + 2], 2)
+ dst[c] = r + c + 2 < bs * 2 ? AVG3(above[r + c], above[r + c + 1],
+ above[r + c + 2])
: above[bs * 2 - 1];
}
dst += stride;
@@ -180,20 +179,19 @@ static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
// first row
for (c = 0; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+ dst[c] = AVG2(above[c - 1], above[c]);
dst += stride;
// second row
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
dst += stride;
// the rest of first col
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[0] = AVG3(above[-1], left[0], left[1]);
for (r = 3; r < bs; ++r)
- dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
- left[r - 1], 2);
+ dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
// the rest of the block
for (r = 2; r < bs; ++r) {
@@ -208,14 +206,13 @@ static INLINE void highbd_d135_predictor(uint16_t *dst, ptrdiff_t stride,
const uint16_t *left, int bd) {
int r, c;
(void) bd;
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
- dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
dst += stride;
for (r = 1; r < bs; ++r) {
@@ -230,20 +227,19 @@ static INLINE void highbd_d153_predictor(uint16_t *dst, ptrdiff_t stride,
const uint16_t *left, int bd) {
int r, c;
(void) bd;
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+ dst[0] = AVG2(above[-1], left[0]);
for (r = 1; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+ dst[r * stride] = AVG2(left[r - 1], left[r]);
dst++;
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
- dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
dst++;
for (c = 0; c < bs - 2; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
+ dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
dst += stride;
for (r = 1; r < bs; ++r) {
@@ -359,10 +355,6 @@ static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride,
}
#endif // CONFIG_VP9_HIGHBITDEPTH
-#define DST(x, y) dst[(x) + (y) * stride]
-#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
-#define AVG2(a, b) (((a) + (b) + 1) >> 1)
-
void vp9_d207_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left) {
const int I = left[0];
@@ -386,16 +378,14 @@ static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
(void) above;
// first column
for (r = 0; r < bs - 1; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+ dst[r * stride] = AVG2(left[r], left[r + 1]);
dst[(bs - 1) * stride] = left[bs - 1];
dst++;
// second column
for (r = 0; r < bs - 2; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
- left[r + 2], 2);
- dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
- left[bs - 1] * 3, 2);
+ dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
+ dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
dst[(bs - 1) * stride] = left[bs - 1];
dst++;
@@ -418,19 +408,18 @@ void vp9_d63_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
const int E = above[4];
const int F = above[5];
const int G = above[6];
- const int H = above[7];
(void)left;
DST(0, 0) = AVG2(A, B);
DST(1, 0) = DST(0, 2) = AVG2(B, C);
DST(2, 0) = DST(1, 2) = AVG2(C, D);
DST(3, 0) = DST(2, 2) = AVG2(D, E);
+ DST(3, 2) = AVG2(E, F); // differs from vp8
DST(0, 1) = AVG3(A, B, C);
DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
- DST(3, 2) = AVG3(E, F, G);
- DST(3, 3) = AVG3(F, G, H);
+ DST(3, 3) = AVG3(E, F, G); // differs from vp8
}
static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
@@ -439,11 +428,9 @@ static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
(void) left;
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c)
- dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
- above[r/2 + c + 1] * 2 +
- above[r/2 + c + 2], 2)
- : ROUND_POWER_OF_TWO(above[r/2 + c] +
- above[r/2 + c + 1], 1);
+ dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
+ above[(r >> 1) + c + 2])
+ : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
dst += stride;
}
}
@@ -467,7 +454,7 @@ void vp9_d45_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
- DST(3, 3) = AVG3(G, H, H);
+ DST(3, 3) = H; // differs from vp8
}
static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
@@ -517,20 +504,19 @@ static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
// first row
for (c = 0; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+ dst[c] = AVG2(above[c - 1], above[c]);
dst += stride;
// second row
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
dst += stride;
// the rest of first col
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[0] = AVG3(above[-1], left[0], left[1]);
for (r = 3; r < bs; ++r)
- dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
- left[r - 1], 2);
+ dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
// the rest of the block
for (r = 2; r < bs; ++r) {
@@ -565,14 +551,13 @@ void vp9_d135_predictor_4x4(uint8_t *dst, ptrdiff_t stride,
static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
- dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
dst += stride;
for (r = 1; r < bs; ++r) {
@@ -610,20 +595,19 @@ void vp9_d153_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+ dst[0] = AVG2(above[-1], left[0]);
for (r = 1; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+ dst[r * stride] = AVG2(left[r - 1], left[r]);
dst++;
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
- dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
dst++;
for (c = 0; c < bs - 2; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
+ dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
dst += stride;
for (r = 1; r < bs; ++r) {
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-13 17:14:09 +0000