diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 715df98bd52bcd6a5e52fc01cc8363cda94a7afd..76f80d4d74dc07e80c21bbb39f79b21e8a6006df 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -340,7 +340,7 @@ struct buf_2d {
int stride;
};
-struct mb_plane {
+struct macroblockd_plane {
DECLARE_ALIGNED(16, int16_t, qcoeff[64 * 64]);
DECLARE_ALIGNED(16, int16_t, dqcoeff[64 * 64]);
DECLARE_ALIGNED(16, uint16_t, eobs[256]);
@@ -360,7 +360,7 @@ struct mb_plane {
BLOCK_OFFSET((x)->plane[2].field, ((i) - 20), 16))
typedef struct macroblockd {
- struct mb_plane plane[MAX_MB_PLANE];
+ struct macroblockd_plane plane[MAX_MB_PLANE];
/* 16 Y blocks, 4 U, 4 V, each with 16 entries. */
BLOCKD block[24];
@@ -924,6 +924,18 @@ static INLINE void foreach_predicted_block_uv(
foreach_predicted_block_in_plane(xd, bsize, plane, visit, arg);
}
}
+static int raster_block_offset(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize,
+ int plane, int block) {
+ const int bw = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+ const int stride = 4 << bw;
+ const int y = 4 * (block >> bw), x = 4 * (block & ((1 << bw) - 1));
+ return y * stride + x;
+}
+static int16_t* raster_block_offset_int16(MACROBLOCKD *xd,
+ BLOCK_SIZE_TYPE bsize,
+ int plane, int block, int16_t *base) {
+ return base + raster_block_offset(xd, bsize, plane, block);
+}
#if CONFIG_CODE_ZEROGROUP
static int get_zpc_used(TX_SIZE tx_size) {
diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh
index 1f838a574dee2bfcd5768a0204ade58c817f9990..020841c42e710bf5d9314fa54f6c492449600849 100644
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -617,12 +617,6 @@ prototype int vp9_block_error "int16_t *coeff, int16_t *dqcoeff, int block_size"
specialize vp9_block_error mmx sse2
vp9_block_error_sse2=vp9_block_error_xmm
-prototype void vp9_subtract_b "struct block *be, struct blockd *bd, int pitch"
-# TODO(jingning): The prototype function in c has been changed to remove
-# the use of predictor buffer in MACROBLOCKD. Need to modify the mmx and sse2
-# versions accordingly.
-specialize vp9_subtract_b
-
#
# Structured Similarity (SSIM)
#
diff --git a/vp9/decoder/vp9_decodframe.c b/vp9/decoder/vp9_decodframe.c
index fefa2af556de3ece9f0b4f752ef6f36d5ea067ee..91aca6f54861641160d2cf20ba35d7b361cc5a29 100644
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -305,7 +305,7 @@ static void decode_8x8(MACROBLOCKD *xd) {
static INLINE void dequant_add_y(MACROBLOCKD *xd, TX_TYPE tx_type, int idx) {
BLOCKD *const b = &xd->block[idx];
- struct mb_plane *const y = &xd->plane[0];
+ struct macroblockd_plane *const y = &xd->plane[0];
if (tx_type != DCT_DCT) {
vp9_dequant_iht_add_c(tx_type,
BLOCK_OFFSET(y->qcoeff, idx, 16),
diff --git a/vp9/encoder/vp9_asm_enc_offsets.c b/vp9/encoder/vp9_asm_enc_offsets.c
index 1a770dcf74df8310c3c8e78422fe8f7ed37480e1..db9a4fc4373a7ce170dd4119f8cf471938c00f7d 100644
--- a/vp9/encoder/vp9_asm_enc_offsets.c
+++ b/vp9/encoder/vp9_asm_enc_offsets.c
@@ -20,7 +20,6 @@
BEGIN
/* regular quantize */
-DEFINE(vp9_block_coeff, offsetof(BLOCK, coeff));
DEFINE(vp9_block_zbin, offsetof(BLOCK, zbin));
DEFINE(vp9_block_round, offsetof(BLOCK, round));
DEFINE(vp9_block_quant, offsetof(BLOCK, quant));
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index e511be8cd1c0f9c8bcd0bbbeccfd91ed99174316..a054269aac85a80cf00b6aa8c859b29669206dc0 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -25,7 +25,6 @@ typedef struct {
typedef struct block {
// 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
- int16_t *src_diff;
int16_t *coeff;
// 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
@@ -83,9 +82,13 @@ typedef struct {
int64_t txfm_rd_diff[NB_TXFM_MODES];
} PICK_MODE_CONTEXT;
+struct macroblock_plane {
+ DECLARE_ALIGNED(16, int16_t, src_diff[64*64]);
+};
+
typedef struct macroblock MACROBLOCK;
struct macroblock {
- DECLARE_ALIGNED(16, int16_t, src_diff[64*64+32*32*2]);
+ struct macroblock_plane plane[MAX_MB_PLANE];
DECLARE_ALIGNED(16, int16_t, coeff[64*64+32*32*2]);
// 16 Y blocks, 4 U blocks, 4 V blocks,
BLOCK block[24];
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 1f20d5e1c3861d87f4cf0e44fffd5e4659bb4da1..fd65dfa45e6936ca7c7b1c1e12946f37f23fef50 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1734,25 +1734,8 @@ void vp9_encode_frame(VP9_COMP *cpi) {
}
void vp9_setup_block_ptrs(MACROBLOCK *x) {
- int r, c;
int i;
- for (r = 0; r < 4; r++) {
- for (c = 0; c < 4; c++)
- x->block[r * 4 + c].src_diff = x->src_diff + r * 4 * 16 + c * 4;
- }
-
- for (r = 0; r < 2; r++) {
- for (c = 0; c < 2; c++)
- x->block[16 + r * 2 + c].src_diff = x->src_diff + 256 + r * 4 * 8 + c * 4;
- }
-
-
- for (r = 0; r < 2; r++) {
- for (c = 0; c < 2; c++)
- x->block[20 + r * 2 + c].src_diff = x->src_diff + 320 + r * 4 * 8 + c * 4;
- }
-
for (i = 0; i < 24; i++)
x->block[i].coeff = x->coeff + i * 16;
}
@@ -2100,14 +2083,6 @@ static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
VP9_COMMON *const cm = &cpi->common;
MACROBLOCK *const x = &cpi->mb;
MACROBLOCKD *const xd = &x->e_mbd;
- const uint8_t *src = x->src.y_buffer;
- uint8_t *dst = xd->plane[0].dst.buf;
- const uint8_t *usrc = x->src.u_buffer;
- uint8_t *udst = xd->plane[1].dst.buf;
- const uint8_t *vsrc = x->src.v_buffer;
- uint8_t *vdst = xd->plane[2].dst.buf;
- int src_y_stride = x->src.y_stride, dst_y_stride = xd->plane[0].dst.stride;
- int src_uv_stride = x->src.uv_stride, dst_uv_stride = xd->plane[1].dst.stride;
int n;
MODE_INFO *mi = x->e_mbd.mode_info_context;
unsigned int segment_id = mi->mbmi.segment_id;
@@ -2187,10 +2162,7 @@ static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
}
if (!x->skip) {
- vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride,
- bsize);
- vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
- udst, vdst, dst_uv_stride, bsize);
+ vp9_subtract_sb(x, bsize);
switch (xd->mode_info_context->mbmi.txfm_size) {
case TX_32X32:
diff --git a/vp9/encoder/vp9_encodeintra.c b/vp9/encoder/vp9_encodeintra.c
index 5c6559bdb5b846fae7e86942ac3bb42ac61b3c3a..32a92b8e7bd395a87328e9ac995ddb4f7194c3ee 100644
--- a/vp9/encoder/vp9_encodeintra.c
+++ b/vp9/encoder/vp9_encodeintra.c
@@ -37,7 +37,7 @@ int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred) {
}
}
- return vp9_get_mb_ss(x->src_diff);
+ return vp9_get_mb_ss(x->plane[0].src_diff);
}
static void encode_intra4x4block(MACROBLOCK *x, int ib) {
@@ -45,6 +45,9 @@ static void encode_intra4x4block(MACROBLOCK *x, int ib) {
BLOCK *be = &x->block[ib];
MACROBLOCKD * const xd = &x->e_mbd;
TX_TYPE tx_type;
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+ x->plane[0].src_diff);
assert(ib < 16);
@@ -54,16 +57,18 @@ static void encode_intra4x4block(MACROBLOCK *x, int ib) {
vp9_intra4x4_predict(&x->e_mbd, b, b->bmi.as_mode.first,
*(b->base_dst) + b->dst, b->dst_stride);
- vp9_subtract_b(be, b, 16);
+ vp9_subtract_block(4, 4, src_diff, 16,
+ *(be->base_src) + be->src, be->src_stride,
+ *(b->base_dst) + b->dst, b->dst_stride);
tx_type = get_tx_type_4x4(&x->e_mbd, ib);
if (tx_type != DCT_DCT) {
- vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+ vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
vp9_ht_quantize_b_4x4(x, ib, tx_type);
vp9_short_iht4x4(BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16),
b->diff, 16, tx_type);
} else {
- x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm4x4(src_diff, be->coeff, 32);
x->quantize_b_4x4(x, ib, 16);
vp9_inverse_transform_b_4x4(&x->e_mbd, xd->plane[0].eobs[ib],
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16),
@@ -86,10 +91,7 @@ void vp9_encode_intra16x16mby(VP9_COMMON *const cm, MACROBLOCK *x) {
TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
vp9_build_intra_predictors_sby_s(xd, BLOCK_SIZE_MB16X16);
- vp9_subtract_sby_s_c(x->src_diff,
- x->src.y_buffer, x->src.y_stride,
- xd->plane[0].dst.buf, xd->plane[0].dst.stride,
- BLOCK_SIZE_MB16X16);
+ vp9_subtract_sby(x, BLOCK_SIZE_MB16X16);
switch (tx_size) {
case TX_16X16:
@@ -123,11 +125,7 @@ void vp9_encode_intra16x16mbuv(VP9_COMMON *const cm, MACROBLOCK *x) {
TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
- vp9_subtract_sbuv_s_c(x->src_diff,
- x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
- xd->plane[1].dst.buf, xd->plane[2].dst.buf,
- xd->plane[1].dst.stride,
- BLOCK_SIZE_MB16X16);
+ vp9_subtract_sbuv(x, BLOCK_SIZE_MB16X16);
switch (tx_size) {
case TX_4X4:
@@ -153,6 +151,9 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
MACROBLOCKD *xd = &x->e_mbd;
BLOCKD *b = &xd->block[ib];
BLOCK *be = &x->block[ib];
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+ x->plane[0].src_diff);
const int iblock[4] = {0, 1, 4, 5};
int i;
TX_TYPE tx_type;
@@ -160,7 +161,9 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
vp9_intra8x8_predict(xd, b, b->bmi.as_mode.first,
*(b->base_dst) + b->dst, b->dst_stride);
// generate residual blocks
- vp9_subtract_4b_c(be, b, 16);
+ vp9_subtract_block(8, 8, src_diff, 16,
+ *(be->base_src) + be->src, be->src_stride,
+ *(b->base_dst) + b->dst, b->dst_stride);
if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
int idx = (ib & 0x02) ? (ib + 2) : ib;
@@ -169,12 +172,12 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
assert(idx < 16);
tx_type = get_tx_type_8x8(xd, ib);
if (tx_type != DCT_DCT) {
- vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type);
+ vp9_short_fht8x8(src_diff, (x->block + idx)->coeff, 16, tx_type);
x->quantize_b_8x8(x, idx, tx_type, 16);
vp9_short_iht8x8(dqcoeff, xd->block[ib].diff,
16, tx_type);
} else {
- x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32);
+ x->fwd_txm8x8(src_diff, (x->block + idx)->coeff, 32);
x->quantize_b_8x8(x, idx, DCT_DCT, 16);
vp9_short_idct8x8(dqcoeff, xd->block[ib].diff, 32);
}
@@ -182,18 +185,21 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
for (i = 0; i < 4; i++) {
int idx = ib + iblock[i];
int16_t * const dqcoeff = BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16);
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, idx,
+ x->plane[0].src_diff);
assert(idx < 16);
b = &xd->block[ib + iblock[i]];
be = &x->block[ib + iblock[i]];
tx_type = get_tx_type_4x4(xd, ib + iblock[i]);
if (tx_type != DCT_DCT) {
- vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+ vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
vp9_ht_quantize_b_4x4(x, ib + iblock[i], tx_type);
vp9_short_iht4x4(dqcoeff, b->diff, 16, tx_type);
} else if (!(i & 1) &&
get_tx_type_4x4(xd, ib + iblock[i] + 1) == DCT_DCT) {
- x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm8x4(src_diff, be->coeff, 32);
x->quantize_b_4x4_pair(x, ib + iblock[i], ib + iblock[i] + 1, 16);
vp9_inverse_transform_b_4x4(xd, xd->plane[0].eobs[ib + iblock[i]],
dqcoeff, b->diff, 32);
@@ -201,7 +207,7 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
dqcoeff + 16, (b + 1)->diff, 32);
i++;
} else {
- x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm4x4(src_diff, be->coeff, 32);
x->quantize_b_4x4(x, ib + iblock[i], 16);
vp9_inverse_transform_b_4x4(xd, xd->plane[0].eobs[ib + iblock[i]],
dqcoeff, b->diff, 32);
@@ -231,14 +237,20 @@ static void encode_intra_uv4x4(MACROBLOCK *x, int ib, int mode) {
int16_t * const dqcoeff = MB_SUBBLOCK_FIELD(xd, dqcoeff, ib);
const int plane = ib < 20 ? 1 : 2;
const int block = ib < 20 ? ib - 16 : ib - 20;
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, plane, block,
+ x->plane[plane].src_diff);
assert(ib >= 16 && ib < 24);
vp9_intra_uv4x4_predict(&x->e_mbd, b, mode,
*(b->base_dst) + b->dst, b->dst_stride);
- vp9_subtract_b(be, b, 8);
+ assert(xd->plane[1].subsampling_x == 1);
+ vp9_subtract_block(4, 4, src_diff, 8,
+ *(be->base_src) + be->src, be->src_stride,
+ *(b->base_dst) + b->dst, b->dst_stride);
- x->fwd_txm4x4(be->src_diff, be->coeff, 16);
+ x->fwd_txm4x4(src_diff, be->coeff, 16);
x->quantize_b_4x4(x, ib, 16);
vp9_inverse_transform_b_4x4(&x->e_mbd, xd->plane[plane].eobs[block],
dqcoeff, b->diff, 16);
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index c841c2823728d2190e17f29f8dad2e4334d56fe0..b6fe6dfa787d658e14cedf8999c48f7ac44374dd 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -20,102 +20,54 @@
#include "vp9/common/vp9_systemdependent.h"
#include "vp9_rtcd.h"
-void vp9_subtract_b_c(BLOCK *be, BLOCKD *bd, int pitch) {
- uint8_t *src_ptr = (*(be->base_src) + be->src);
- int16_t *diff_ptr = be->src_diff;
- uint8_t *pred_ptr = *(bd->base_dst) + bd->dst;
- int src_stride = be->src_stride;
- int dst_stride = bd->dst_stride;
-
+void vp9_subtract_block(int rows, int cols,
+ int16_t *diff_ptr, int diff_stride,
+ const uint8_t *src_ptr, int src_stride,
+ const uint8_t *pred_ptr, int pred_stride) {
int r, c;
- for (r = 0; r < 4; r++) {
- for (c = 0; c < 4; c++)
+ for (r = 0; r < rows; r++) {
+ for (c = 0; c < cols; c++)
diff_ptr[c] = src_ptr[c] - pred_ptr[c];
- diff_ptr += pitch;
- pred_ptr += dst_stride;
+ diff_ptr += diff_stride;
+ pred_ptr += pred_stride;
src_ptr += src_stride;
}
}
-void vp9_subtract_4b_c(BLOCK *be, BLOCKD *bd, int pitch) {
- uint8_t *src_ptr = (*(be->base_src) + be->src);
- int16_t *diff_ptr = be->src_diff;
- uint8_t *pred_ptr = *(bd->base_dst) + bd->dst;
- int src_stride = be->src_stride;
- int dst_stride = bd->dst_stride;
- int r, c;
- for (r = 0; r < 8; r++) {
- for (c = 0; c < 8; c++)
- diff_ptr[c] = src_ptr[c] - pred_ptr[c];
+static void subtract_plane(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize, int plane) {
+ const MACROBLOCKD * const xd = &x->e_mbd;
+ const int bw = 4 << (b_width_log2(bsize) - xd->plane[plane].subsampling_x);
+ const int bh = 4 << (b_height_log2(bsize) - xd->plane[plane].subsampling_y);
+ const uint8_t *src = plane == 0 ? x->src.y_buffer :
+ plane == 1 ? x->src.u_buffer : x->src.v_buffer;
+ const int src_stride = plane == 0 ? x->src.y_stride : x->src.uv_stride;
- diff_ptr += pitch;
- pred_ptr += dst_stride;
- src_ptr += src_stride;
- }
+ assert(plane < 3);
+ vp9_subtract_block(bh, bw,
+ x->plane[plane].src_diff, bw, src, src_stride,
+ xd->plane[plane].dst.buf, xd->plane[plane].dst.stride);
}
-void vp9_subtract_sby_s_c(int16_t *diff, const uint8_t *src, int src_stride,
- const uint8_t *pred, int dst_stride,
- BLOCK_SIZE_TYPE bsize) {
- const int bh = 16 << mb_height_log2(bsize), bw = 16 << mb_width_log2(bsize);
- int r, c;
-
- for (r = 0; r < bh; r++) {
- for (c = 0; c < bw; c++)
- diff[c] = src[c] - pred[c];
-
- diff += bw;
- pred += dst_stride;
- src += src_stride;
- }
+void vp9_subtract_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+ subtract_plane(x, bsize, 0);
}
-void vp9_subtract_sbuv_s_c(int16_t *diff, const uint8_t *usrc,
- const uint8_t *vsrc, int src_stride,
- const uint8_t *upred,
- const uint8_t *vpred, int dst_stride,
- BLOCK_SIZE_TYPE bsize) {
- const int bhl = mb_height_log2(bsize), bwl = mb_width_log2(bsize);
- const int uoff = (16 * 16) << (bhl + bwl), voff = (uoff * 5) >> 2;
- const int bw = 8 << bwl, bh = 8 << bhl;
- int16_t *udiff = diff + uoff;
- int16_t *vdiff = diff + voff;
- int r, c;
-
- for (r = 0; r < bh; r++) {
- for (c = 0; c < bw; c++)
- udiff[c] = usrc[c] - upred[c];
-
- udiff += bw;
- upred += dst_stride;
- usrc += src_stride;
- }
+void vp9_subtract_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+ int i;
- for (r = 0; r < bh; r++) {
- for (c = 0; c < bw; c++)
- vdiff[c] = vsrc[c] - vpred[c];
-
- vdiff += bw;
- vpred += dst_stride;
- vsrc += src_stride;
- }
+ for (i = 1; i < MAX_MB_PLANE; i++)
+ subtract_plane(x, bsize, i);
}
-static void subtract_mb(MACROBLOCK *x) {
- MACROBLOCKD *xd = &x->e_mbd;
- vp9_subtract_sby_s_c(x->src_diff, x->src.y_buffer, x->src.y_stride,
- xd->plane[0].dst.buf, xd->plane[0].dst.stride,
- BLOCK_SIZE_MB16X16);
- vp9_subtract_sbuv_s_c(x->src_diff, x->src.u_buffer, x->src.v_buffer,
- x->src.uv_stride,
- xd->plane[1].dst.buf, xd->plane[2].dst.buf,
- xd->plane[1].dst.stride,
- BLOCK_SIZE_MB16X16);
+void vp9_subtract_sb(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+ vp9_subtract_sby(x, bsize);
+ vp9_subtract_sbuv(x, bsize);
}
+
void vp9_transform_sby_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) - 1);
@@ -125,7 +77,7 @@ void vp9_transform_sby_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
for (n = 0; n < bw * bh; n++) {
const int x_idx = n & (bw - 1), y_idx = n >> bwl;
- vp9_short_fdct32x32(x->src_diff + y_idx * stride * 32 + x_idx * 32,
+ vp9_short_fdct32x32(x->plane[0].src_diff + y_idx * stride * 32 + x_idx * 32,
x->coeff + n * 1024, stride * 2);
}
}
@@ -143,10 +95,11 @@ void vp9_transform_sby_16x16(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
(y_idx * bstride + x_idx) * 4);
if (tx_type != DCT_DCT) {
- vp9_short_fht16x16(x->src_diff + y_idx * stride * 16 + x_idx * 16,
+ vp9_short_fht16x16(x->plane[0].src_diff +
+ y_idx * stride * 16 + x_idx * 16,
x->coeff + n * 256, stride, tx_type);
} else {
- x->fwd_txm16x16(x->src_diff + y_idx * stride * 16 + x_idx * 16,
+ x->fwd_txm16x16(x->plane[0].src_diff + y_idx * stride * 16 + x_idx * 16,
x->coeff + n * 256, stride * 2);
}
}
@@ -164,10 +117,10 @@ void vp9_transform_sby_8x8(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
const TX_TYPE tx_type = get_tx_type_8x8(xd, (y_idx * bstride + x_idx) * 2);
if (tx_type != DCT_DCT) {
- vp9_short_fht8x8(x->src_diff + y_idx * stride * 8 + x_idx * 8,
+ vp9_short_fht8x8(x->plane[0].src_diff + y_idx * stride * 8 + x_idx * 8,
x->coeff + n * 64, stride, tx_type);
} else {
- x->fwd_txm8x8(x->src_diff + y_idx * stride * 8 + x_idx * 8,
+ x->fwd_txm8x8(x->plane[0].src_diff + y_idx * stride * 8 + x_idx * 8,
x->coeff + n * 64, stride * 2);
}
}
@@ -185,10 +138,10 @@ void vp9_transform_sby_4x4(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
if (tx_type != DCT_DCT) {
- vp9_short_fht4x4(x->src_diff + y_idx * stride * 4 + x_idx * 4,
+ vp9_short_fht4x4(x->plane[0].src_diff + y_idx * stride * 4 + x_idx * 4,
x->coeff + n * 16, stride, tx_type);
} else {
- x->fwd_txm4x4(x->src_diff + y_idx * stride * 4 + x_idx * 4,
+ x->fwd_txm4x4(x->plane[0].src_diff + y_idx * stride * 4 + x_idx * 4,
x->coeff + n * 16, stride * 2);
}
}
@@ -197,9 +150,9 @@ void vp9_transform_sby_4x4(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
void vp9_transform_sbuv_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
assert(bsize == BLOCK_SIZE_SB64X64);
vp9_clear_system_state();
- vp9_short_fdct32x32(x->src_diff + 4096,
+ vp9_short_fdct32x32(x->plane[1].src_diff,
x->coeff + 4096, 64);
- vp9_short_fdct32x32(x->src_diff + 4096 + 1024,
+ vp9_short_fdct32x32(x->plane[2].src_diff,
x->coeff + 4096 + 1024, 64);
}
@@ -214,9 +167,9 @@ void vp9_transform_sbuv_16x16(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
for (n = 0; n < bw * bh; n++) {
const int x_idx = n & (bw - 1), y_idx = n >> (bwl - 1);
- x->fwd_txm16x16(x->src_diff + uoff + y_idx * stride * 16 + x_idx * 16,
+ x->fwd_txm16x16(x->plane[1].src_diff + y_idx * stride * 16 + x_idx * 16,
x->coeff + uoff + n * 256, stride * 2);
- x->fwd_txm16x16(x->src_diff + voff + y_idx * stride * 16 + x_idx * 16,
+ x->fwd_txm16x16(x->plane[2].src_diff + y_idx * stride * 16 + x_idx * 16,
x->coeff + voff + n * 256, stride * 2);
}
}
@@ -232,9 +185,9 @@ void vp9_transform_sbuv_8x8(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
for (n = 0; n < bw * bh; n++) {
const int x_idx = n & (bw - 1), y_idx = n >> (bwl - 1);
- x->fwd_txm8x8(x->src_diff + uoff + y_idx * stride * 8 + x_idx * 8,
+ x->fwd_txm8x8(x->plane[1].src_diff + y_idx * stride * 8 + x_idx * 8,
x->coeff + uoff + n * 64, stride * 2);
- x->fwd_txm8x8(x->src_diff + voff + y_idx * stride * 8 + x_idx * 8,
+ x->fwd_txm8x8(x->plane[2].src_diff + y_idx * stride * 8 + x_idx * 8,
x->coeff + voff + n * 64, stride * 2);
}
}
@@ -250,9 +203,9 @@ void vp9_transform_sbuv_4x4(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
for (n = 0; n < bw * bh; n++) {
const int x_idx = n & (bw - 1), y_idx = n >> (bwl - 1);
- x->fwd_txm4x4(x->src_diff + uoff + y_idx * stride * 4 + x_idx * 4,
+ x->fwd_txm4x4(x->plane[1].src_diff + y_idx * stride * 4 + x_idx * 4,
x->coeff + uoff + n * 16, stride * 2);
- x->fwd_txm4x4(x->src_diff + voff + y_idx * stride * 4 + x_idx * 4,
+ x->fwd_txm4x4(x->plane[2].src_diff + y_idx * stride * 4 + x_idx * 4,
x->coeff + voff + n * 16, stride * 2);
}
}
@@ -844,7 +797,7 @@ void vp9_encode_inter16x16(VP9_COMMON *const cm, MACROBLOCK *x,
MACROBLOCKD *const xd = &x->e_mbd;
vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
- subtract_mb(x);
+ vp9_subtract_sb(x, BLOCK_SIZE_MB16X16);
vp9_fidct_mb(cm, x);
vp9_recon_sb(xd, BLOCK_SIZE_MB16X16);
}
@@ -854,9 +807,7 @@ void vp9_encode_inter16x16y(MACROBLOCK *x, int mb_row, int mb_col) {
MACROBLOCKD *xd = &x->e_mbd;
vp9_build_inter_predictors_sby(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
- vp9_subtract_sby_s_c(x->src_diff, x->src.y_buffer, x->src.y_stride,
- xd->plane[0].dst.buf, xd->plane[0].dst.stride,
- BLOCK_SIZE_MB16X16);
+ vp9_subtract_sby(x, BLOCK_SIZE_MB16X16);
vp9_transform_sby_4x4(x, BLOCK_SIZE_MB16X16);
vp9_quantize_sby_4x4(x, BLOCK_SIZE_MB16X16);
diff --git a/vp9/encoder/vp9_encodemb.h b/vp9/encoder/vp9_encodemb.h
index 3c0d760a1933db794a41cffb0625301319dd4923..da134a86b847c5780f52a2042300dd17779e14bb 100644
--- a/vp9/encoder/vp9_encodemb.h
+++ b/vp9/encoder/vp9_encodemb.h
@@ -56,15 +56,12 @@ void vp9_optimize_sbuv_4x4(VP9_COMMON *const cm, MACROBLOCK *x,
void vp9_fidct_mb(VP9_COMMON *const cm, MACROBLOCK *x);
-void vp9_subtract_4b_c(BLOCK *be, BLOCKD *bd, int pitch);
-
-void vp9_subtract_sby_s_c(int16_t *diff, const uint8_t *src, int src_stride,
- const uint8_t *pred, int dst_stride,
- BLOCK_SIZE_TYPE bsize);
-void vp9_subtract_sbuv_s_c(int16_t *diff, const uint8_t *usrc,
- const uint8_t *vsrc, int src_stride,
- const uint8_t *upred,
- const uint8_t *vpred, int dst_stride,
- BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_block(int rows, int cols,
+ int16_t *diff_ptr, int diff_stride,
+ const uint8_t *src_ptr, int src_stride,
+ const uint8_t *pred_ptr, int pred_stride);
+void vp9_subtract_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_sb(MACROBLOCK *xd, BLOCK_SIZE_TYPE bsize);
#endif // VP9_ENCODER_VP9_ENCODEMB_H_
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 18ad10b8e9eefe2b2c006635defec8be81f2517e..5517b157402d54003f42fe013c1546ab3570c9ea 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -845,12 +845,9 @@ static void super_block_yrd(VP9_COMP *cpi,
int *skip, BLOCK_SIZE_TYPE bs,
int64_t txfm_cache[NB_TXFM_MODES]) {
VP9_COMMON *const cm = &cpi->common;
- MACROBLOCKD *const xd = &x->e_mbd;
int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB];
- uint8_t *src = x->src.y_buffer, *dst = xd->plane[0].dst.buf;
- int src_y_stride = x->src.y_stride, dst_y_stride = xd->plane[0].dst.stride;
- vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride, bs);
+ vp9_subtract_sby(x, bs);
if (bs >= BLOCK_SIZE_SB32X32)
super_block_yrd_32x32(cm, x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32],
@@ -877,7 +874,9 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
VP9_COMMON *const cm = &cpi->common;
BLOCK *be = x->block + ib;
BLOCKD *b = xd->block + ib;
-
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+ x->plane[0].src_diff);
ENTROPY_CONTEXT ta = *a, tempa = *a;
ENTROPY_CONTEXT tl = *l, templ = *l;
TX_TYPE tx_type = DCT_DCT;
@@ -917,15 +916,17 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
#endif
vp9_intra4x4_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
- vp9_subtract_b(be, b, 16);
+ vp9_subtract_block(4, 4, src_diff, 16,
+ *(be->base_src) + be->src, be->src_stride,
+ *(b->base_dst) + b->dst, b->dst_stride);
b->bmi.as_mode.first = mode;
tx_type = get_tx_type_4x4(xd, be - x->block);
if (tx_type != DCT_DCT) {
- vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+ vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
vp9_ht_quantize_b_4x4(x, be - x->block, tx_type);
} else {
- x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm4x4(src_diff, be->coeff, 32);
x->quantize_b_4x4(x, be - x->block, 16);
}
@@ -1107,10 +1108,12 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
ENTROPY_CONTEXT_PLANES ta, tl;
ENTROPY_CONTEXT *ta0, *ta1, besta0 = 0, besta1 = 0;
ENTROPY_CONTEXT *tl0, *tl1, bestl0 = 0, bestl1 = 0;
-
// perform transformation of dimension 8x8
// note the input and output index mapping
int idx = (ib & 0x02) ? (ib + 2) : ib;
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+ x->plane[0].src_diff);
assert(ib < 16);
for (mode = DC_PRED; mode <= TM_PRED; mode++) {
@@ -1123,14 +1126,16 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
vp9_intra8x8_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
- vp9_subtract_4b_c(be, b, 16);
+ vp9_subtract_block(8, 8, src_diff, 16,
+ *(be->base_src) + be->src, be->src_stride,
+ *(b->base_dst) + b->dst, b->dst_stride);
if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
TX_TYPE tx_type = get_tx_type_8x8(xd, ib);
if (tx_type != DCT_DCT)
- vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type);
+ vp9_short_fht8x8(src_diff, (x->block + idx)->coeff, 16, tx_type);
else
- x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32);
+ x->fwd_txm8x8(src_diff, (x->block + idx)->coeff, 32);
x->quantize_b_8x8(x, idx, tx_type, 16);
// compute quantization mse of 8x8 block
@@ -1162,20 +1167,24 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
distortion = 0;
rate_t = 0;
for (i = 0; i < 4; ++i) {
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
+ 0, ib + iblock[i],
+ x->plane[0].src_diff);
int do_two = 0;
b = &xd->block[ib + iblock[i]];
be = &x->block[ib + iblock[i]];
tx_type = get_tx_type_4x4(xd, ib + iblock[i]);
if (tx_type != DCT_DCT) {
- vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+ vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
vp9_ht_quantize_b_4x4(x, ib + iblock[i], tx_type);
} else if (!(i & 1) &&
get_tx_type_4x4(xd, ib + iblock[i] + 1) == DCT_DCT) {
- x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm8x4(src_diff, be->coeff, 32);
x->quantize_b_4x4_pair(x, ib + iblock[i], ib + iblock[i] + 1, 16);
do_two = 1;
} else {
- x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm4x4(src_diff, be->coeff, 32);
x->quantize_b_4x4(x, ib + iblock[i], 16);
}
distortion += vp9_block_error_c(be->coeff,
@@ -1531,12 +1540,8 @@ static void super_block_uvrd(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
- uint8_t *usrc = x->src.u_buffer, *udst = xd->plane[1].dst.buf;
- uint8_t *vsrc = x->src.v_buffer, *vdst = xd->plane[2].dst.buf;
- int src_uv_stride = x->src.uv_stride, dst_uv_stride = xd->plane[1].dst.stride;
- vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
- udst, vdst, dst_uv_stride, bsize);
+ vp9_subtract_sbuv(x, bsize);
if (mbmi->txfm_size >= TX_32X32 && bsize >= BLOCK_SIZE_SB64X64) {
super_block_uvrd_32x32(cm, x, rate, distortion, skippable, bsize);
@@ -1738,6 +1743,9 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
if (labels[i] == which_label) {
BLOCKD *bd = &x->e_mbd.block[i];
BLOCK *be = &x->block[i];
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, i,
+ x->plane[0].src_diff);
int thisdistortion;
vp9_build_inter_predictor(*(bd->base_pre) + bd->pre,
@@ -1760,8 +1768,10 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
&xd->subpix);
}
- vp9_subtract_b(be, bd, 16);
- x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+ vp9_subtract_block(4, 4, src_diff, 16,
+ *(be->base_src) + be->src, be->src_stride,
+ *(bd->base_dst) + bd->dst, bd->dst_stride);
+ x->fwd_txm4x4(src_diff, be->coeff, 32);
x->quantize_b_4x4(x, i, 16);
thisdistortion = vp9_block_error(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, i, 16), 16);
@@ -1809,6 +1819,9 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
const int idx = (ib & 8) + ((ib & 2) << 1);
BLOCKD *bd = &xd->block[ib];
BLOCK *be = &x->block[ib], *be2 = &x->block[idx];
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+ x->plane[0].src_diff);
int thisdistortion;
assert(idx < 16);
@@ -1826,11 +1839,13 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
&xd->subpix);
}
- vp9_subtract_4b_c(be, bd, 16);
+ vp9_subtract_block(8, 8, src_diff, 16,
+ *(be->base_src) + be->src, be->src_stride,
+ *(bd->base_dst) + bd->dst, bd->dst_stride);
if (xd->mode_info_context->mbmi.txfm_size == TX_4X4) {
if (otherrd) {
- x->fwd_txm8x8(be->src_diff, be2->coeff, 32);
+ x->fwd_txm8x8(src_diff, be2->coeff, 32);
x->quantize_b_8x8(x, idx, DCT_DCT, 16);
thisdistortion = vp9_block_error_c(be2->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
@@ -1843,9 +1858,13 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
xd->mode_info_context->mbmi.txfm_size = TX_4X4;
}
for (j = 0; j < 4; j += 2) {
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
+ 0, ib + iblock[j],
+ x->plane[0].src_diff);
bd = &xd->block[ib + iblock[j]];
be = &x->block[ib + iblock[j]];
- x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+ x->fwd_txm8x4(src_diff, be->coeff, 32);
x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1, 16);
thisdistortion = vp9_block_error_c(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32);
@@ -1866,7 +1885,11 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
if (otherrd) {
for (j = 0; j < 4; j += 2) {
BLOCK *be = &x->block[ib + iblock[j]];
- x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+ int16_t* const src_diff =
+ raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
+ 0, ib + iblock[j],
+ x->plane[0].src_diff);
+ x->fwd_txm8x4(src_diff, be->coeff, 32);
x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1, 16);
thisdistortion = vp9_block_error_c(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32);
@@ -1886,7 +1909,7 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
xd->mode_info_context->mbmi.txfm_size = TX_8X8;
}
}
- x->fwd_txm8x8(be->src_diff, be2->coeff, 32);
+ x->fwd_txm8x8(src_diff, be2->coeff, 32);
x->quantize_b_8x8(x, idx, DCT_DCT, 16);
thisdistortion = vp9_block_error_c(be2->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
@@ -3782,12 +3805,7 @@ static void rd_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
vp9_build_inter_predictors_sbuv(&x->e_mbd, mb_row, mb_col,
BLOCK_SIZE_MB16X16);
- vp9_subtract_sbuv_s_c(x->src_diff,
- x->src.u_buffer,
- x->src.v_buffer, x->src.uv_stride,
- xd->plane[1].dst.buf,
- xd->plane[2].dst.buf, xd->plane[1].dst.stride,
- BLOCK_SIZE_MB16X16);
+ vp9_subtract_sbuv(x, BLOCK_SIZE_MB16X16);
super_block_uvrd_4x4(cm, x, &rate_uv, &distortion_uv,
&uv_skippable, BLOCK_SIZE_MB16X16);
diff --git a/vp9/encoder/x86/vp9_x86_csystemdependent.c b/vp9/encoder/x86/vp9_x86_csystemdependent.c
index 04383fcb4a3b772e9b1b02c94f33100636e2a312..6016e14eb14ba167275a01aab3b6940781ac460a 100644
--- a/vp9/encoder/x86/vp9_x86_csystemdependent.c
+++ b/vp9/encoder/x86/vp9_x86_csystemdependent.c
@@ -17,7 +17,7 @@
// TODO(jimbankoski) Consider rewriting the c to take the same values rather
// than going through these pointer conversions
-#if HAVE_MMX
+#if 0 && HAVE_MMX
void vp9_short_fdct8x4_mmx(short *input, short *output, int pitch) {
vp9_short_fdct4x4_mmx(input, output, pitch);
vp9_short_fdct4x4_mmx(input + 4, output + 16, pitch);
@@ -38,7 +38,7 @@ void vp9_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch) {
#endif
-#if HAVE_SSE2
+#if 0 && HAVE_SSE2
void vp9_subtract_b_sse2_impl(unsigned char *z, int src_stride,
short *diff, unsigned char *predictor,
int pitch);