Merge "vp8-denoiser: Avoid doing the mcomp if we don't denoise."
diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index d6cc5e4..c4dfaf3 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -792,6 +792,67 @@
CompareInvReference(ref_txfm_, thresh_);
}
+class PartialTrans16x16Test
+ : public ::testing::TestWithParam<
+ std::tr1::tuple<FdctFunc, vpx_bit_depth_t> > {
+ public:
+ virtual ~PartialTrans16x16Test() {}
+ virtual void SetUp() {
+ fwd_txfm_ = GET_PARAM(0);
+ bit_depth_ = GET_PARAM(1);
+ }
+
+ virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+ vpx_bit_depth_t bit_depth_;
+ FdctFunc fwd_txfm_;
+};
+
+TEST_P(PartialTrans16x16Test, Extremes) {
+#if CONFIG_VP9_HIGHBITDEPTH
+ const int16_t maxval =
+ static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
+#else
+ const int16_t maxval = 255;
+#endif
+ const int minval = -maxval;
+ DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
+ DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
+
+ for (int i = 0; i < kNumCoeffs; ++i) input[i] = maxval;
+ output[0] = 0;
+ ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
+ EXPECT_EQ((maxval * kNumCoeffs) >> 1, output[0]);
+
+ for (int i = 0; i < kNumCoeffs; ++i) input[i] = minval;
+ output[0] = 0;
+ ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
+ EXPECT_EQ((minval * kNumCoeffs) >> 1, output[0]);
+}
+
+TEST_P(PartialTrans16x16Test, Random) {
+#if CONFIG_VP9_HIGHBITDEPTH
+ const int16_t maxval =
+ static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
+#else
+ const int16_t maxval = 255;
+#endif
+ DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
+ DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
+ ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+ int sum = 0;
+ for (int i = 0; i < kNumCoeffs; ++i) {
+ const int val = (i & 1) ? -rnd(maxval + 1) : rnd(maxval + 1);
+ input[i] = val;
+ sum += val;
+ }
+ output[0] = 0;
+ ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
+ EXPECT_EQ(sum >> 1, output[0]);
+}
+
using std::tr1::make_tuple;
#if CONFIG_VP9_HIGHBITDEPTH
@@ -824,6 +885,11 @@
make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
+INSTANTIATE_TEST_CASE_P(
+ C, PartialTrans16x16Test,
+ ::testing::Values(make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_8),
+ make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_10),
+ make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_12)));
#else
INSTANTIATE_TEST_CASE_P(
C, Trans16x16HT,
@@ -832,6 +898,9 @@
make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
+INSTANTIATE_TEST_CASE_P(C, PartialTrans16x16Test,
+ ::testing::Values(make_tuple(&vpx_fdct16x16_1_c,
+ VPX_BITS_8)));
#endif // CONFIG_VP9_HIGHBITDEPTH
#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -859,6 +928,9 @@
VPX_BITS_8),
make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3,
VPX_BITS_8)));
+INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans16x16Test,
+ ::testing::Values(make_tuple(&vpx_fdct16x16_1_sse2,
+ VPX_BITS_8)));
#endif // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
#if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -896,6 +968,9 @@
&idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
make_tuple(&idct16x16_12,
&idct16x16_256_add_12_sse2, 3167, VPX_BITS_12)));
+INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans16x16Test,
+ ::testing::Values(make_tuple(&vpx_fdct16x16_1_sse2,
+ VPX_BITS_8)));
#endif // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -912,5 +987,8 @@
make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 3,
VPX_BITS_8)));
+INSTANTIATE_TEST_CASE_P(MSA, PartialTrans16x16Test,
+ ::testing::Values(make_tuple(&vpx_fdct16x16_1_msa,
+ VPX_BITS_8)));
#endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
} // namespace
diff --git a/test/dct32x32_test.cc b/test/dct32x32_test.cc
index 87d8366..2a9942e 100644
--- a/test/dct32x32_test.cc
+++ b/test/dct32x32_test.cc
@@ -344,6 +344,28 @@
EXPECT_EQ((minval * kNumCoeffs) >> 3, output[0]);
}
+TEST_P(PartialTrans32x32Test, Random) {
+#if CONFIG_VP9_HIGHBITDEPTH
+ const int16_t maxval =
+ static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
+#else
+ const int16_t maxval = 255;
+#endif
+ DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
+ DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
+ ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+ int sum = 0;
+ for (int i = 0; i < kNumCoeffs; ++i) {
+ const int val = (i & 1) ? -rnd(maxval + 1) : rnd(maxval + 1);
+ input[i] = val;
+ sum += val;
+ }
+ output[0] = 0;
+ ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 32));
+ EXPECT_EQ(sum >> 3, output[0]);
+}
+
using std::tr1::make_tuple;
#if CONFIG_VP9_HIGHBITDEPTH
diff --git a/test/variance_test.cc b/test/variance_test.cc
index 6f50f78..a9ca07c 100644
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
@@ -74,6 +74,10 @@
return res;
}
+/* Note:
+ * Our codebase calculates the "diff" value in the variance algorithm by
+ * (src - ref).
+ */
static uint32_t variance_ref(const uint8_t *src, const uint8_t *ref,
int l2w, int l2h, int src_stride_coeff,
int ref_stride_coeff, uint32_t *sse_ptr,
@@ -87,14 +91,14 @@
for (int x = 0; x < w; x++) {
int diff;
if (!use_high_bit_depth_) {
- diff = ref[w * y * ref_stride_coeff + x] -
- src[w * y * src_stride_coeff + x];
+ diff = src[w * y * src_stride_coeff + x] -
+ ref[w * y * ref_stride_coeff + x];
se += diff;
sse += diff * diff;
#if CONFIG_VP9_HIGHBITDEPTH
} else {
- diff = CONVERT_TO_SHORTPTR(ref)[w * y * ref_stride_coeff + x] -
- CONVERT_TO_SHORTPTR(src)[w * y * src_stride_coeff + x];
+ diff = CONVERT_TO_SHORTPTR(src)[w * y * src_stride_coeff + x] -
+ CONVERT_TO_SHORTPTR(ref)[w * y * ref_stride_coeff + x];
se += diff;
sse += diff * diff;
#endif // CONFIG_VP9_HIGHBITDEPTH
@@ -309,15 +313,15 @@
void VarianceTest<VarianceFunctionType>::RefTest() {
for (int i = 0; i < 10; ++i) {
for (int j = 0; j < block_size_; j++) {
- if (!use_high_bit_depth_) {
- src_[j] = rnd_.Rand8();
- ref_[j] = rnd_.Rand8();
+ if (!use_high_bit_depth_) {
+ src_[j] = rnd_.Rand8();
+ ref_[j] = rnd_.Rand8();
#if CONFIG_VP9_HIGHBITDEPTH
- } else {
- CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() && mask_;
- CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() && mask_;
+ } else {
+ CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask_;
+ CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask_;
#endif // CONFIG_VP9_HIGHBITDEPTH
- }
+ }
}
unsigned int sse1, sse2;
unsigned int var1;
@@ -328,8 +332,10 @@
log2height_, stride_coeff,
stride_coeff, &sse2,
use_high_bit_depth_, bit_depth_);
- EXPECT_EQ(sse1, sse2);
- EXPECT_EQ(var1, var2);
+ EXPECT_EQ(sse1, sse2)
+ << "Error at test index: " << i;
+ EXPECT_EQ(var1, var2)
+ << "Error at test index: " << i;
}
}
@@ -346,8 +352,8 @@
ref_[ref_ind] = rnd_.Rand8();
#if CONFIG_VP9_HIGHBITDEPTH
} else {
- CONVERT_TO_SHORTPTR(src_)[src_ind] = rnd_.Rand16() && mask_;
- CONVERT_TO_SHORTPTR(ref_)[ref_ind] = rnd_.Rand16() && mask_;
+ CONVERT_TO_SHORTPTR(src_)[src_ind] = rnd_.Rand16() & mask_;
+ CONVERT_TO_SHORTPTR(ref_)[ref_ind] = rnd_.Rand16() & mask_;
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
@@ -361,8 +367,10 @@
log2height_, src_stride_coeff,
ref_stride_coeff, &sse2,
use_high_bit_depth_, bit_depth_);
- EXPECT_EQ(sse1, sse2);
- EXPECT_EQ(var1, var2);
+ EXPECT_EQ(sse1, sse2)
+ << "Error at test index: " << i;
+ EXPECT_EQ(var1, var2)
+ << "Error at test index: " << i;
}
}
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index 596427c..f304de3 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -353,11 +353,36 @@
}
}
+// TODO(slavarnway): Move this decoder version of
+// vp9_get_pred_context_switchable_interp() to vp9_pred_common.h and update the
+// encoder.
+//
+// Returns a context number for the given MB prediction signal
+static int dec_get_pred_context_switchable_interp(const MACROBLOCKD *xd) {
+ // Note:
+ // The mode info data structure has a one element border above and to the
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
+ const MODE_INFO *const left_mi = xd->left_mi;
+ const int left_type = left_mi ? left_mi->interp_filter : SWITCHABLE_FILTERS;
+ const MODE_INFO *const above_mi = xd->above_mi;
+ const int above_type = above_mi ? above_mi->interp_filter
+ : SWITCHABLE_FILTERS;
+
+ if (left_type == above_type)
+ return left_type;
+ else if (left_type == SWITCHABLE_FILTERS && above_type != SWITCHABLE_FILTERS)
+ return above_type;
+ else if (left_type != SWITCHABLE_FILTERS && above_type == SWITCHABLE_FILTERS)
+ return left_type;
+ else
+ return SWITCHABLE_FILTERS;
+}
static INLINE INTERP_FILTER read_switchable_interp_filter(
VP9_COMMON *const cm, MACROBLOCKD *const xd,
vpx_reader *r) {
- const int ctx = vp9_get_pred_context_switchable_interp(xd);
+ const int ctx = dec_get_pred_context_switchable_interp(xd);
const INTERP_FILTER type =
(INTERP_FILTER)vpx_read_tree(r, vp9_switchable_interp_tree,
cm->fc->switchable_interp_prob[ctx]);
@@ -373,9 +398,6 @@
const BLOCK_SIZE bsize = mi->sb_type;
int i;
- mi->ref_frame[0] = INTRA_FRAME;
- mi->ref_frame[1] = NONE;
-
switch (bsize) {
case BLOCK_4X4:
for (i = 0; i < 4; ++i)
@@ -399,6 +421,13 @@
}
mi->uv_mode = read_intra_mode_uv(cm, xd, r, mi->mode);
+
+ // Initialize interp_filter here so we do not have to check for inter block
+ // modes in dec_get_pred_context_switchable_interp()
+ mi->interp_filter = SWITCHABLE_FILTERS;
+
+ mi->ref_frame[0] = INTRA_FRAME;
+ mi->ref_frame[1] = NONE;
}
static INLINE int is_mv_valid(const MV *mv) {
diff --git a/vp9/encoder/vp9_noise_estimate.c b/vp9/encoder/vp9_noise_estimate.c
index abe6768..10ea010 100644
--- a/vp9/encoder/vp9_noise_estimate.c
+++ b/vp9/encoder/vp9_noise_estimate.c
@@ -91,7 +91,7 @@
} else {
if (ne->value > ne->thresh)
noise_level = kMedium;
- else if (ne->value > ((5 * ne->thresh) >> 3))
+ else if (ne->value > ((9 * ne->thresh) >> 4))
noise_level = kLow;
else
noise_level = kLowLow;
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index 9dd7394..545441b 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -621,15 +621,16 @@
if (cpi->common.frame_type == KEY_FRAME) {
active_worst_quality = curr_frame == 0 ? rc->worst_quality
- : rc->last_q[KEY_FRAME] * 2;
+ : rc->last_q[KEY_FRAME] << 1;
} else {
if (!rc->is_src_frame_alt_ref &&
(cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
- active_worst_quality = curr_frame == 1 ? rc->last_q[KEY_FRAME] * 5 / 4
+ active_worst_quality = curr_frame == 1 ? rc->last_q[KEY_FRAME] * 5 >> 2
: rc->last_q[INTER_FRAME];
} else {
- active_worst_quality = curr_frame == 1 ? rc->last_q[KEY_FRAME] * 2
- : rc->last_q[INTER_FRAME] * 2;
+ active_worst_quality = curr_frame == 1 ? rc->last_q[KEY_FRAME] << 1 :
+ VPXMIN(rc->last_q[INTER_FRAME] << 1,
+ (rc->avg_frame_qindex[INTER_FRAME] * 3 >> 1));
}
}
return VPXMIN(active_worst_quality, rc->worst_quality);
@@ -662,7 +663,7 @@
VPXMIN(rc->avg_frame_qindex[INTER_FRAME],
rc->avg_frame_qindex[KEY_FRAME]) :
rc->avg_frame_qindex[INTER_FRAME];
- active_worst_quality = VPXMIN(rc->worst_quality, ambient_qp * 5 / 4);
+ active_worst_quality = VPXMIN(rc->worst_quality, ambient_qp * 5 >> 2);
if (rc->buffer_level > rc->optimal_buffer_level) {
// Adjust down.
// Maximum limit for down adjustment, ~30%.
@@ -2047,7 +2048,7 @@
// for cases where there is small change from content that is completely
// static.
if (cpi->oxcf.rc_mode == VPX_VBR) {
- min_thresh = 30000;
+ min_thresh = 40000;
thresh = 2.0f;
}
if (avg_sad >
@@ -2061,14 +2062,15 @@
// For VBR, under scene change/high content change, force golden refresh.
if (cpi->oxcf.rc_mode == VPX_VBR &&
rc->high_source_sad &&
- cpi->refresh_golden_frame == 0 &&
cpi->ext_refresh_frame_flags_pending == 0) {
int target;
cpi->refresh_golden_frame = 1;
+ rc->gfu_boost = DEFAULT_GF_BOOST >> 1;
+ rc->baseline_gf_interval = VPXMIN(20,
+ VPXMAX(10, rc->baseline_gf_interval));
rc->frames_till_gf_update_due = rc->baseline_gf_interval;
if (rc->frames_till_gf_update_due > rc->frames_to_key)
rc->frames_till_gf_update_due = rc->frames_to_key;
- rc->gfu_boost = DEFAULT_GF_BOOST;
target = calc_pframe_target_size_one_pass_vbr(cpi);
vp9_rc_set_frame_target(cpi, target);
}
diff --git a/vpx_dsp/fwd_txfm.c b/vpx_dsp/fwd_txfm.c
index 39596b1..a5802e1 100644
--- a/vpx_dsp/fwd_txfm.c
+++ b/vpx_dsp/fwd_txfm.c
@@ -365,12 +365,12 @@
void vpx_fdct16x16_1_c(const int16_t *input, tran_low_t *output, int stride) {
int r, c;
- tran_low_t sum = 0;
+ int sum = 0;
for (r = 0; r < 16; ++r)
for (c = 0; c < 16; ++c)
sum += input[r * stride + c];
- output[0] = sum >> 1;
+ output[0] = (tran_low_t)(sum >> 1);
}
static INLINE tran_high_t dct_32_round(tran_high_t input) {
diff --git a/vpx_dsp/mips/fwd_txfm_msa.c b/vpx_dsp/mips/fwd_txfm_msa.c
index f66dd5f..0dd141f 100644
--- a/vpx_dsp/mips/fwd_txfm_msa.c
+++ b/vpx_dsp/mips/fwd_txfm_msa.c
@@ -237,11 +237,9 @@
}
void vpx_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
- out[1] = 0;
-
- out[0] = LD_HADD(input, stride);
- out[0] += LD_HADD(input + 8, stride);
- out[0] += LD_HADD(input + 16 * 8, stride);
- out[0] += LD_HADD(input + 16 * 8 + 8, stride);
- out[0] >>= 1;
+ int sum = LD_HADD(input, stride);
+ sum += LD_HADD(input + 8, stride);
+ sum += LD_HADD(input + 16 * 8, stride);
+ sum += LD_HADD(input + 16 * 8 + 8, stride);
+ out[0] = (int16_t)(sum >> 1);
}
diff --git a/vpx_dsp/x86/fwd_txfm_sse2.c b/vpx_dsp/x86/fwd_txfm_sse2.c
index bca72e8..e4deeec 100644
--- a/vpx_dsp/x86/fwd_txfm_sse2.c
+++ b/vpx_dsp/x86/fwd_txfm_sse2.c
@@ -40,7 +40,7 @@
in1 = _mm_add_epi32(tmp, in0);
in0 = _mm_slli_epi32(in1, 1);
- store_output(&in0, output);
+ output[0] = (tran_low_t)_mm_cvtsi128_si32(in0);
}
void vpx_fdct8x8_1_sse2(const int16_t *input, tran_low_t *output, int stride) {
@@ -80,7 +80,7 @@
in0 = _mm_srli_si128(sum, 8);
in1 = _mm_add_epi32(sum, in0);
- store_output(&in1, output);
+ output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
}
void vpx_fdct16x16_1_sse2(const int16_t *input, tran_low_t *output,
@@ -91,40 +91,39 @@
int i;
for (i = 0; i < 2; ++i) {
- input += 8 * i;
- in0 = _mm_load_si128((const __m128i *)(input + 0 * stride));
- in1 = _mm_load_si128((const __m128i *)(input + 1 * stride));
- in2 = _mm_load_si128((const __m128i *)(input + 2 * stride));
- in3 = _mm_load_si128((const __m128i *)(input + 3 * stride));
+ in0 = _mm_load_si128((const __m128i *)(input + 0 * stride + 0));
+ in1 = _mm_load_si128((const __m128i *)(input + 0 * stride + 8));
+ in2 = _mm_load_si128((const __m128i *)(input + 1 * stride + 0));
+ in3 = _mm_load_si128((const __m128i *)(input + 1 * stride + 8));
u0 = _mm_add_epi16(in0, in1);
u1 = _mm_add_epi16(in2, in3);
sum = _mm_add_epi16(sum, u0);
- in0 = _mm_load_si128((const __m128i *)(input + 4 * stride));
- in1 = _mm_load_si128((const __m128i *)(input + 5 * stride));
- in2 = _mm_load_si128((const __m128i *)(input + 6 * stride));
- in3 = _mm_load_si128((const __m128i *)(input + 7 * stride));
+ in0 = _mm_load_si128((const __m128i *)(input + 2 * stride + 0));
+ in1 = _mm_load_si128((const __m128i *)(input + 2 * stride + 8));
+ in2 = _mm_load_si128((const __m128i *)(input + 3 * stride + 0));
+ in3 = _mm_load_si128((const __m128i *)(input + 3 * stride + 8));
sum = _mm_add_epi16(sum, u1);
u0 = _mm_add_epi16(in0, in1);
u1 = _mm_add_epi16(in2, in3);
sum = _mm_add_epi16(sum, u0);
- in0 = _mm_load_si128((const __m128i *)(input + 8 * stride));
- in1 = _mm_load_si128((const __m128i *)(input + 9 * stride));
- in2 = _mm_load_si128((const __m128i *)(input + 10 * stride));
- in3 = _mm_load_si128((const __m128i *)(input + 11 * stride));
+ in0 = _mm_load_si128((const __m128i *)(input + 4 * stride + 0));
+ in1 = _mm_load_si128((const __m128i *)(input + 4 * stride + 8));
+ in2 = _mm_load_si128((const __m128i *)(input + 5 * stride + 0));
+ in3 = _mm_load_si128((const __m128i *)(input + 5 * stride + 8));
sum = _mm_add_epi16(sum, u1);
u0 = _mm_add_epi16(in0, in1);
u1 = _mm_add_epi16(in2, in3);
sum = _mm_add_epi16(sum, u0);
- in0 = _mm_load_si128((const __m128i *)(input + 12 * stride));
- in1 = _mm_load_si128((const __m128i *)(input + 13 * stride));
- in2 = _mm_load_si128((const __m128i *)(input + 14 * stride));
- in3 = _mm_load_si128((const __m128i *)(input + 15 * stride));
+ in0 = _mm_load_si128((const __m128i *)(input + 6 * stride + 0));
+ in1 = _mm_load_si128((const __m128i *)(input + 6 * stride + 8));
+ in2 = _mm_load_si128((const __m128i *)(input + 7 * stride + 0));
+ in3 = _mm_load_si128((const __m128i *)(input + 7 * stride + 8));
sum = _mm_add_epi16(sum, u1);
u0 = _mm_add_epi16(in0, in1);
@@ -132,6 +131,7 @@
sum = _mm_add_epi16(sum, u0);
sum = _mm_add_epi16(sum, u1);
+ input += 8 * stride;
}
u0 = _mm_setzero_si128();
@@ -149,7 +149,7 @@
in1 = _mm_add_epi32(sum, in0);
in1 = _mm_srai_epi32(in1, 1);
- store_output(&in1, output);
+ output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
}
void vpx_fdct32x32_1_sse2(const int16_t *input, tran_low_t *output,
@@ -221,7 +221,7 @@
in1 = _mm_add_epi32(sum, in0);
in1 = _mm_srai_epi32(in1, 3);
- store_output(&in1, output);
+ output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
}
#define DCT_HIGH_BIT_DEPTH 0
diff --git a/webmdec.cc b/webmdec.cc
index f541cfe..c4539a0 100644
--- a/webmdec.cc
+++ b/webmdec.cc
@@ -164,7 +164,7 @@
}
get_new_block = true;
}
- if (status) {
+ if (status || block_entry == NULL) {
return -1;
}
if (get_new_block) {
