Merge branch 'master' into nextgenv2
diff --git a/test/vp9_intrapred_test.cc b/test/vp9_intrapred_test.cc
index bc85daf..2bebdcb 100644
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -141,7 +141,7 @@
&vpx_highbd_tm_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_tm_predictor_32x32_sse2,
&vpx_highbd_tm_predictor_32x32_c, 32, 8),
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
@@ -155,14 +155,14 @@
&vpx_highbd_v_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 8),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 8)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
::testing::Values(
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
@@ -176,7 +176,7 @@
&vpx_highbd_v_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 8),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 8)));
@@ -194,7 +194,7 @@
make_tuple(&vpx_highbd_tm_predictor_32x32_sse2,
&vpx_highbd_tm_predictor_32x32_c, 32,
10),
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 10),
@@ -211,14 +211,14 @@
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32,
10),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 10)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
::testing::Values(
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 10),
@@ -233,7 +233,7 @@
&vpx_highbd_v_predictor_16x16_c, 16, 10),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 10),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 10)));
@@ -251,7 +251,7 @@
make_tuple(&vpx_highbd_tm_predictor_32x32_sse2,
&vpx_highbd_tm_predictor_32x32_c, 32,
12),
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 12),
@@ -268,14 +268,14 @@
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32,
12),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 12)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
::testing::Values(
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 12),
@@ -290,7 +290,7 @@
&vpx_highbd_v_predictor_16x16_c, 16, 12),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 12),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 12)));
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index 2270a06..4295204 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -190,7 +190,8 @@
BLOCK_SIZE bsize,
int64_t rate,
int64_t dist,
- int skip) {
+ int skip,
+ struct macroblock_plane *const p) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
const int bw = num_8x8_blocks_wide_lookup[bsize];
@@ -198,12 +199,33 @@
const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
const int block_index = mi_row * cm->mi_cols + mi_col;
- const int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist,
- bsize);
+ int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist, bsize);
// Default is to not update the refresh map.
int new_map_value = cr->map[block_index];
int x = 0; int y = 0;
+ int is_skin = 0;
+ if (refresh_this_block == 0 &&
+ bsize <= BLOCK_16X16 &&
+ cpi->oxcf.content != VP9E_CONTENT_SCREEN) {
+ // Take center pixel in block to determine is_skin.
+ const int y_width_shift = (4 << b_width_log2_lookup[bsize]) >> 1;
+ const int y_height_shift = (4 << b_height_log2_lookup[bsize]) >> 1;
+ const int uv_width_shift = y_width_shift >> 1;
+ const int uv_height_shift = y_height_shift >> 1;
+ const int stride = p[0].src.stride;
+ const int strideuv = p[1].src.stride;
+ const uint8_t ysource =
+ p[0].src.buf[y_height_shift * stride + y_width_shift];
+ const uint8_t usource =
+ p[1].src.buf[uv_height_shift * strideuv + uv_width_shift];
+ const uint8_t vsource =
+ p[2].src.buf[uv_height_shift * strideuv + uv_width_shift];
+ is_skin = vp9_skin_pixel(ysource, usource, vsource);
+ if (is_skin)
+ refresh_this_block = 1;
+ }
+
// If this block is labeled for refresh, check if we should reset the
// segment_id.
if (cyclic_refresh_segment_id_boosted(mbmi->segment_id)) {
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.h b/vp9/encoder/vp9_aq_cyclicrefresh.h
index a5b3813..edf0a97 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.h
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.h
@@ -14,6 +14,8 @@
#include "vpx/vpx_integer.h"
#include "vp9/common/vp9_blockd.h"
+#include "vp9/encoder/vp9_block.h"
+#include "vp9/encoder/vp9_skin_detection.h"
#ifdef __cplusplus
extern "C" {
@@ -93,7 +95,8 @@
void vp9_cyclic_refresh_update_segment(struct VP9_COMP *const cpi,
MB_MODE_INFO *const mbmi,
int mi_row, int mi_col, BLOCK_SIZE bsize,
- int64_t rate, int64_t dist, int skip);
+ int64_t rate, int64_t dist, int skip,
+ struct macroblock_plane *const p);
void vp9_cyclic_refresh_update_sb_postencode(struct VP9_COMP *const cpi,
const MB_MODE_INFO *const mbmi,
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index f002788..7f94e19 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1045,7 +1045,7 @@
if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi, mi_row,
mi_col, bsize, ctx->rate, ctx->dist,
- x->skip);
+ x->skip, p);
}
}
@@ -1705,6 +1705,7 @@
MACROBLOCKD *const xd = &x->e_mbd;
MODE_INFO *const mi = xd->mi[0];
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+ struct macroblock_plane *const p = x->plane;
const struct segmentation *const seg = &cm->seg;
const int bw = num_8x8_blocks_wide_lookup[mi->mbmi.sb_type];
const int bh = num_8x8_blocks_high_lookup[mi->mbmi.sb_type];
@@ -1725,7 +1726,7 @@
} else {
// Setting segmentation map for cyclic_refresh.
vp9_cyclic_refresh_update_segment(cpi, mbmi, mi_row, mi_col, bsize,
- ctx->rate, ctx->dist, x->skip);
+ ctx->rate, ctx->dist, x->skip, p);
}
vp9_init_plane_quantizers(cpi, x);
}
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index cda38f6..ab647a4 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1349,11 +1349,25 @@
const InterpKernel *kernel = vp9_filter_kernels[mi->mbmi.interp_filter];
for (ref = 0; ref < 1 + is_compound; ++ref) {
- const uint8_t *pre = &pd->pre[ref].buf[vp9_raster_block_offset(BLOCK_8X8, i,
- pd->pre[ref].stride)];
+ const int bw = b_width_log2_lookup[BLOCK_8X8];
+ const int h = 4 * (i >> bw);
+ const int w = 4 * (i & ((1 << bw) - 1));
+ const struct scale_factors *sf = &xd->block_refs[ref]->sf;
+ int y_stride = pd->pre[ref].stride;
+ uint8_t *pre = pd->pre[ref].buf + (h * pd->pre[ref].stride + w);
+
+ if (vp9_is_scaled(sf)) {
+ const int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x));
+ const int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y));
+
+ y_stride = xd->block_refs[ref]->buf->y_stride;
+ pre = xd->block_refs[ref]->buf->y_buffer;
+ pre += scaled_buffer_offset(x_start + w, y_start + h,
+ y_stride, sf);
+ }
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- vp9_highbd_build_inter_predictor(pre, pd->pre[ref].stride,
+ vp9_highbd_build_inter_predictor(pre, y_stride,
dst, pd->dst.stride,
&mi->bmi[i].as_mv[ref].as_mv,
&xd->block_refs[ref]->sf, width, height,
@@ -1361,7 +1375,7 @@
mi_col * MI_SIZE + 4 * (i % 2),
mi_row * MI_SIZE + 4 * (i / 2), xd->bd);
} else {
- vp9_build_inter_predictor(pre, pd->pre[ref].stride,
+ vp9_build_inter_predictor(pre, y_stride,
dst, pd->dst.stride,
&mi->bmi[i].as_mv[ref].as_mv,
&xd->block_refs[ref]->sf, width, height, ref,
@@ -1370,7 +1384,7 @@
mi_row * MI_SIZE + 4 * (i / 2));
}
#else
- vp9_build_inter_predictor(pre, pd->pre[ref].stride,
+ vp9_build_inter_predictor(pre, y_stride,
dst, pd->dst.stride,
&mi->bmi[i].as_mv[ref].as_mv,
&xd->block_refs[ref]->sf, width, height, ref,
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 3f4fe11..a2a0674 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -291,10 +291,10 @@
specialize qw/vpx_highbd_v_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_tm_predictor_4x4/, "$sse_x86inc";
+ specialize qw/vpx_highbd_tm_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_dc_predictor_4x4/, "$sse_x86inc";
+ specialize qw/vpx_highbd_dc_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
specialize qw/vpx_highbd_dc_top_predictor_4x4/;
diff --git a/vpx_dsp/x86/highbd_intrapred_sse2.asm b/vpx_dsp/x86/highbd_intrapred_sse2.asm
index 9e6dcb5..233958a 100644
--- a/vpx_dsp/x86/highbd_intrapred_sse2.asm
+++ b/vpx_dsp/x86/highbd_intrapred_sse2.asm
@@ -17,24 +17,20 @@
pw_32: times 4 dd 32
SECTION .text
-INIT_MMX sse
+INIT_XMM sse2
cglobal highbd_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
GET_GOT goffsetq
movq m0, [aboveq]
movq m2, [leftq]
- DEFINE_ARGS dst, stride, one
- mov oned, 0x0001
- pxor m1, m1
- movd m3, oned
- pshufw m3, m3, 0x0
paddw m0, m2
- pmaddwd m0, m3
- packssdw m0, m1
- pmaddwd m0, m3
+ pshuflw m1, m0, 0xe
+ paddw m0, m1
+ pshuflw m1, m0, 0x1
+ paddw m0, m1
paddw m0, [GLOBAL(pw_4)]
psraw m0, 3
- pshufw m0, m0, 0x0
+ pshuflw m0, m0, 0x0
movq [dstq ], m0
movq [dstq+strideq*2], m0
lea dstq, [dstq+strideq*4]
@@ -261,43 +257,44 @@
jnz .loop
REP_RET
-INIT_MMX sse
-cglobal highbd_tm_predictor_4x4, 5, 6, 5, dst, stride, above, left, bps, one
+INIT_XMM sse2
+cglobal highbd_tm_predictor_4x4, 5, 5, 6, dst, stride, above, left, bps
movd m1, [aboveq-2]
movq m0, [aboveq]
- pshufw m1, m1, 0x0
+ pshuflw m1, m1, 0x0
+ movlhps m0, m0 ; t1 t2 t3 t4 t1 t2 t3 t4
+ movlhps m1, m1 ; tl tl tl tl tl tl tl tl
; Get the values to compute the maximum value at this bit depth
- mov oned, 1
- movd m3, oned
+ pcmpeqw m3, m3
movd m4, bpsd
- pshufw m3, m3, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -2
- mova m2, m3
+ psubw m0, m1 ; t1-tl t2-tl t3-tl t4-tl
psllw m3, m4
- add leftq, 8
- psubw m3, m2 ; max possible value
- pxor m4, m4 ; min possible value
- psubw m0, m1
-.loop:
- movq m1, [leftq+lineq*4]
- movq m2, [leftq+lineq*4+2]
- pshufw m1, m1, 0x0
- pshufw m2, m2, 0x0
- paddw m1, m0
+ pcmpeqw m2, m2
+ pxor m4, m4 ; min possible value
+ pxor m3, m2 ; max possible value
+ mova m1, [leftq]
+ pshuflw m2, m1, 0x0
+ pshuflw m5, m1, 0x55
+ movlhps m2, m5 ; l1 l1 l1 l1 l2 l2 l2 l2
paddw m2, m0
;Clamp to the bit-depth
- pminsw m1, m3
pminsw m2, m3
- pmaxsw m1, m4
pmaxsw m2, m4
;Store the values
- movq [dstq ], m1
- movq [dstq+strideq*2], m2
+ movq [dstq ], m2
+ movhpd [dstq+strideq*2], m2
lea dstq, [dstq+strideq*4]
- inc lineq
- jnz .loop
- REP_RET
+ pshuflw m2, m1, 0xaa
+ pshuflw m5, m1, 0xff
+ movlhps m2, m5
+ paddw m2, m0
+ ;Clamp to the bit-depth
+ pminsw m2, m3
+ pmaxsw m2, m4
+ ;Store the values
+ movq [dstq ], m2
+ movhpd [dstq+strideq*2], m2
+ RET
INIT_XMM sse2
cglobal highbd_tm_predictor_8x8, 5, 6, 5, dst, stride, above, left, bps, one
