Merge branch 'master' into nextgenv2
This commit is contained in:
Yaowu Xu
Коммит
7330108009
|
|
@ -702,18 +702,6 @@ INSTANTIATE_TEST_CASE_P(MMX, SADTest, ::testing::ValuesIn(mmx_tests));
|
|||
|
||||
#if HAVE_SSE
|
||||
#if CONFIG_USE_X86INC
|
||||
const SadMxNParam sse_tests[] = {
|
||||
make_tuple(4, 8, &vpx_sad4x8_sse, -1),
|
||||
make_tuple(4, 4, &vpx_sad4x4_sse, -1),
|
||||
};
|
||||
INSTANTIATE_TEST_CASE_P(SSE, SADTest, ::testing::ValuesIn(sse_tests));
|
||||
|
||||
const SadMxNAvgParam avg_sse_tests[] = {
|
||||
make_tuple(4, 8, &vpx_sad4x8_avg_sse, -1),
|
||||
make_tuple(4, 4, &vpx_sad4x4_avg_sse, -1),
|
||||
};
|
||||
INSTANTIATE_TEST_CASE_P(SSE, SADavgTest, ::testing::ValuesIn(avg_sse_tests));
|
||||
|
||||
const SadMxNx4Param x4d_sse_tests[] = {
|
||||
make_tuple(4, 8, &vpx_sad4x8x4d_sse, -1),
|
||||
make_tuple(4, 4, &vpx_sad4x4x4d_sse, -1),
|
||||
|
|
@ -736,6 +724,8 @@ const SadMxNParam sse2_tests[] = {
|
|||
make_tuple(8, 16, &vpx_sad8x16_sse2, -1),
|
||||
make_tuple(8, 8, &vpx_sad8x8_sse2, -1),
|
||||
make_tuple(8, 4, &vpx_sad8x4_sse2, -1),
|
||||
make_tuple(4, 8, &vpx_sad4x8_sse2, -1),
|
||||
make_tuple(4, 4, &vpx_sad4x4_sse2, -1),
|
||||
#if CONFIG_VP9_HIGHBITDEPTH
|
||||
make_tuple(64, 64, &vpx_highbd_sad64x64_sse2, 8),
|
||||
make_tuple(64, 32, &vpx_highbd_sad64x32_sse2, 8),
|
||||
|
|
@ -786,6 +776,8 @@ const SadMxNAvgParam avg_sse2_tests[] = {
|
|||
make_tuple(8, 16, &vpx_sad8x16_avg_sse2, -1),
|
||||
make_tuple(8, 8, &vpx_sad8x8_avg_sse2, -1),
|
||||
make_tuple(8, 4, &vpx_sad8x4_avg_sse2, -1),
|
||||
make_tuple(4, 8, &vpx_sad4x8_avg_sse2, -1),
|
||||
make_tuple(4, 4, &vpx_sad4x4_avg_sse2, -1),
|
||||
#if CONFIG_VP9_HIGHBITDEPTH
|
||||
make_tuple(64, 64, &vpx_highbd_sad64x64_avg_sse2, 8),
|
||||
make_tuple(64, 32, &vpx_highbd_sad64x32_avg_sse2, 8),
|
||||
|
|
|
|||
|
|
@ -1228,10 +1228,6 @@ static void setup_quantization(VP10_COMMON *const cm,
|
|||
cm->uv_dc_delta_q = read_delta_q(rb);
|
||||
cm->uv_ac_delta_q = read_delta_q(rb);
|
||||
cm->dequant_bit_depth = cm->bit_depth;
|
||||
|
||||
#if CONFIG_VP9_HIGHBITDEPTH
|
||||
xd->bd = (int)cm->bit_depth;
|
||||
#endif
|
||||
}
|
||||
|
||||
static void setup_segmentation_dequant(VP10_COMMON *const cm) {
|
||||
|
|
@ -2157,6 +2153,10 @@ static size_t read_uncompressed_header(VP10Decoder *pbi,
|
|||
|
||||
setup_loopfilter(&cm->lf, rb);
|
||||
setup_quantization(cm, rb);
|
||||
#if CONFIG_VP9_HIGHBITDEPTH
|
||||
xd->bd = (int)cm->bit_depth;
|
||||
#endif
|
||||
|
||||
setup_segmentation(cm, rb);
|
||||
|
||||
{
|
||||
|
|
|
|||
|
|
@ -680,9 +680,13 @@ static int choose_partitioning(VP9_COMP *cpi,
|
|||
int64_t thresholds[4] = {cpi->vbp_thresholds[0], cpi->vbp_thresholds[1],
|
||||
cpi->vbp_thresholds[2], cpi->vbp_thresholds[3]};
|
||||
|
||||
// For the variance computation under SVC mode, we treat the frame as key if
|
||||
// the reference (base layer frame) is key frame (i.e., is_key_frame == 1).
|
||||
const int is_key_frame = (cm->frame_type == KEY_FRAME ||
|
||||
(is_one_pass_cbr_svc(cpi) &&
|
||||
cpi->svc.layer_context[cpi->svc.temporal_layer_id].is_key_frame));
|
||||
// Always use 4x4 partition for key frame.
|
||||
const int is_key_frame = (cm->frame_type == KEY_FRAME);
|
||||
const int use_4x4_partition = is_key_frame;
|
||||
const int use_4x4_partition = cm->frame_type == KEY_FRAME;
|
||||
const int low_res = (cm->width <= 352 && cm->height <= 288);
|
||||
int variance4x4downsample[16];
|
||||
|
||||
|
|
@ -708,8 +712,7 @@ static int choose_partitioning(VP9_COMP *cpi,
|
|||
s = x->plane[0].src.buf;
|
||||
sp = x->plane[0].src.stride;
|
||||
|
||||
if (!is_key_frame && !(is_one_pass_cbr_svc(cpi) &&
|
||||
cpi->svc.layer_context[cpi->svc.temporal_layer_id].is_key_frame)) {
|
||||
if (!is_key_frame) {
|
||||
// In the case of spatial/temporal scalable coding, the assumption here is
|
||||
// that the temporal reference frame will always be of type LAST_FRAME.
|
||||
// TODO(marpan): If that assumption is broken, we need to revisit this code.
|
||||
|
|
@ -868,9 +871,7 @@ static int choose_partitioning(VP9_COMP *cpi,
|
|||
}
|
||||
}
|
||||
}
|
||||
// TODO(marpan): There is an issue with variance based on 4x4 average in
|
||||
// svc mode, don't allow it for now.
|
||||
if (is_key_frame || (low_res && !cpi->use_svc &&
|
||||
if (is_key_frame || (low_res &&
|
||||
vt.split[i].split[j].part_variances.none.variance >
|
||||
(thresholds[1] << 1))) {
|
||||
force_split[split_index] = 0;
|
||||
|
|
@ -892,7 +893,6 @@ static int choose_partitioning(VP9_COMP *cpi,
|
|||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Fill the rest of the variance tree by summing split partition values.
|
||||
avg_32x32 = 0;
|
||||
for (i = 0; i < 4; i++) {
|
||||
|
|
@ -937,7 +937,7 @@ static int choose_partitioning(VP9_COMP *cpi,
|
|||
// If variance of this 64x64 block is above (some threshold of) the average
|
||||
// variance over the sub-32x32 blocks, then force this block to split.
|
||||
if (!is_key_frame &&
|
||||
vt.part_variances.none.variance > (3 * avg_32x32) >> 3)
|
||||
vt.part_variances.none.variance > (5 * avg_32x32) >> 4)
|
||||
force_split[0] = 1;
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -990,10 +990,10 @@ add_proto qw/unsigned int vpx_sad8x4/, "const uint8_t *src_ptr, int src_stride,
|
|||
specialize qw/vpx_sad8x4 msa/, "$sse2_x86inc";
|
||||
|
||||
add_proto qw/unsigned int vpx_sad4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
|
||||
specialize qw/vpx_sad4x8 msa/, "$sse_x86inc";
|
||||
specialize qw/vpx_sad4x8 msa/, "$sse2_x86inc";
|
||||
|
||||
add_proto qw/unsigned int vpx_sad4x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
|
||||
specialize qw/vpx_sad4x4 mmx neon msa/, "$sse_x86inc";
|
||||
specialize qw/vpx_sad4x4 mmx neon msa/, "$sse2_x86inc";
|
||||
|
||||
#
|
||||
# Avg
|
||||
|
|
@ -1061,10 +1061,10 @@ add_proto qw/unsigned int vpx_sad8x4_avg/, "const uint8_t *src_ptr, int src_stri
|
|||
specialize qw/vpx_sad8x4_avg msa/, "$sse2_x86inc";
|
||||
|
||||
add_proto qw/unsigned int vpx_sad4x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
|
||||
specialize qw/vpx_sad4x8_avg msa/, "$sse_x86inc";
|
||||
specialize qw/vpx_sad4x8_avg msa/, "$sse2_x86inc";
|
||||
|
||||
add_proto qw/unsigned int vpx_sad4x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
|
||||
specialize qw/vpx_sad4x4_avg msa/, "$sse_x86inc";
|
||||
specialize qw/vpx_sad4x4_avg msa/, "$sse2_x86inc";
|
||||
|
||||
#
|
||||
# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
|
||||
|
|
|
|||
|
|
@ -222,8 +222,8 @@ SAD8XN 16, 1 ; sad8x16_avg_sse2
|
|||
SAD8XN 8, 1 ; sad8x8_avg_sse2
|
||||
SAD8XN 4, 1 ; sad8x4_avg_sse2
|
||||
|
||||
; unsigned int vpx_sad4x{4, 8}_sse(uint8_t *src, int src_stride,
|
||||
; uint8_t *ref, int ref_stride);
|
||||
; unsigned int vpx_sad4x{4, 8}_sse2(uint8_t *src, int src_stride,
|
||||
; uint8_t *ref, int ref_stride);
|
||||
%macro SAD4XN 1-2 0
|
||||
SAD_FN 4, %1, 7, %2
|
||||
mov n_rowsd, %1/4
|
||||
|
|
@ -236,10 +236,10 @@ SAD8XN 4, 1 ; sad8x4_avg_sse2
|
|||
movd m4, [refq+ref_stride3q]
|
||||
punpckldq m1, m2
|
||||
punpckldq m3, m4
|
||||
movlhps m1, m3
|
||||
%if %2 == 1
|
||||
pavgb m1, [second_predq+mmsize*0]
|
||||
pavgb m3, [second_predq+mmsize*1]
|
||||
lea second_predq, [second_predq+mmsize*2]
|
||||
lea second_predq, [second_predq+mmsize*1]
|
||||
%endif
|
||||
movd m2, [srcq]
|
||||
movd m5, [srcq+src_strideq]
|
||||
|
|
@ -247,20 +247,21 @@ SAD8XN 4, 1 ; sad8x4_avg_sse2
|
|||
movd m6, [srcq+src_stride3q]
|
||||
punpckldq m2, m5
|
||||
punpckldq m4, m6
|
||||
movlhps m2, m4
|
||||
psadbw m1, m2
|
||||
psadbw m3, m4
|
||||
lea refq, [refq+ref_strideq*4]
|
||||
paddd m0, m1
|
||||
lea srcq, [srcq+src_strideq*4]
|
||||
paddd m0, m3
|
||||
dec n_rowsd
|
||||
jg .loop
|
||||
|
||||
movhlps m1, m0
|
||||
paddd m0, m1
|
||||
movd eax, m0
|
||||
RET
|
||||
%endmacro
|
||||
|
||||
INIT_MMX sse
|
||||
INIT_XMM sse2
|
||||
SAD4XN 8 ; sad4x8_sse
|
||||
SAD4XN 4 ; sad4x4_sse
|
||||
SAD4XN 8, 1 ; sad4x8_avg_sse
|
||||
|
|
|
|||
8
vpxenc.c
8
vpxenc.c
|
|
@ -2060,9 +2060,11 @@ int main(int argc, const char **argv_) {
|
|||
|
||||
#if !CONFIG_WEBM_IO
|
||||
FOREACH_STREAM({
|
||||
stream->config.write_webm = 0;
|
||||
warn("vpxenc was compiled without WebM container support."
|
||||
"Producing IVF output");
|
||||
if (stream->config.write_webm) {
|
||||
stream->config.write_webm = 0;
|
||||
warn("vpxenc was compiled without WebM container support."
|
||||
"Producing IVF output");
|
||||
}
|
||||
});
|
||||
#endif
|
||||
|
||||
|
|
|
|||
Загрузка…
Ссылка в новой задаче