Merge "Re-factor and simplify arnr filter."
diff --git a/test/aq_segment_test.cc b/test/aq_segment_test.cc
index 2f88b53..1b9c943 100644
--- a/test/aq_segment_test.cc
+++ b/test/aq_segment_test.cc
@@ -7,8 +7,6 @@
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
-#include <climits>
-#include <vector>
#include "third_party/googletest/src/include/gtest/gtest.h"
#include "test/codec_factory.h"
#include "test/encode_test_driver.h"
@@ -17,11 +15,12 @@
namespace {
-class AqSegmentTest : public ::libvpx_test::EncoderTest,
- public ::libvpx_test::CodecTestWith2Params<
- libvpx_test::TestMode, int> {
+class AqSegmentTest
+ : public ::libvpx_test::EncoderTest,
+ public ::libvpx_test::CodecTestWith2Params<libvpx_test::TestMode, int> {
protected:
AqSegmentTest() : EncoderTest(GET_PARAM(0)) {}
+ virtual ~AqSegmentTest() {}
virtual void SetUp() {
InitializeConfig();
@@ -39,10 +38,6 @@
}
}
- virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
- if (pkt->data.frame.flags & VPX_FRAME_IS_KEY) {
- }
- }
int set_cpu_used_;
int aq_mode_;
};
@@ -107,13 +102,8 @@
ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
}
-using std::tr1::make_tuple;
-
-#define VP9_FACTORY \
- static_cast<const libvpx_test::CodecFactory*> (&libvpx_test::kVP9)
-
VP9_INSTANTIATE_TEST_CASE(AqSegmentTest,
::testing::Values(::libvpx_test::kRealTime,
::libvpx_test::kOnePassGood),
- ::testing::Range(3, 9));
+ ::testing::Range(3, 9));
} // namespace
diff --git a/test/convolve_test.cc b/test/convolve_test.cc
index 6af2abb..2fcb2df 100644
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -264,7 +264,7 @@
uint8_t* const out = output();
DECLARE_ALIGNED(256, const int16_t, filter8[8]) = {0, 0, 0, 128, 0, 0, 0, 0};
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->h8_(in, kInputStride, out, kOutputStride, filter8, 16, filter8, 16,
Width(), Height()));
@@ -281,7 +281,7 @@
uint8_t* const out = output();
DECLARE_ALIGNED(256, const int16_t, filter8[8]) = {0, 0, 0, 128, 0, 0, 0, 0};
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->v8_(in, kInputStride, out, kOutputStride, filter8, 16, filter8, 16,
Width(), Height()));
@@ -298,7 +298,7 @@
uint8_t* const out = output();
DECLARE_ALIGNED(256, const int16_t, filter8[8]) = {0, 0, 0, 128, 0, 0, 0, 0};
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->hv8_(in, kInputStride, out, kOutputStride, filter8, 16, filter8, 16,
Width(), Height()));
@@ -356,17 +356,17 @@
Width(), Height());
if (filters == eighttap_smooth || (filter_x && filter_y))
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->hv8_(in, kInputStride, out, kOutputStride,
filters[filter_x], 16, filters[filter_y], 16,
Width(), Height()));
else if (filter_y)
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->v8_(in, kInputStride, out, kOutputStride,
kInvalidFilter, 16, filters[filter_y], 16,
Width(), Height()));
else
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->h8_(in, kInputStride, out, kOutputStride,
filters[filter_x], 16, kInvalidFilter, 16,
Width(), Height()));
@@ -414,17 +414,17 @@
Width(), Height());
if (filters == eighttap_smooth || (filter_x && filter_y))
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->hv8_avg_(in, kInputStride, out, kOutputStride,
filters[filter_x], 16, filters[filter_y], 16,
Width(), Height()));
else if (filter_y)
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->v8_avg_(in, kInputStride, out, kOutputStride,
filters[filter_x], 16, filters[filter_y], 16,
Width(), Height()));
else
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
UUT_->h8_avg_(in, kInputStride, out, kOutputStride,
filters[filter_x], 16, filters[filter_y], 16,
Width(), Height()));
@@ -494,9 +494,10 @@
*/
/* Test the horizontal filter. */
- REGISTER_STATE_CHECK(UUT_->h8_(in, kInputStride, out, kOutputStride,
- kChangeFilters[kInitialSubPelOffset],
- kInputPixelStep, NULL, 0, Width(), Height()));
+ ASM_REGISTER_STATE_CHECK(
+ UUT_->h8_(in, kInputStride, out, kOutputStride,
+ kChangeFilters[kInitialSubPelOffset],
+ kInputPixelStep, NULL, 0, Width(), Height()));
for (int x = 0; x < Width(); ++x) {
const int kFilterPeriodAdjust = (x >> 3) << 3;
@@ -508,9 +509,10 @@
}
/* Test the vertical filter. */
- REGISTER_STATE_CHECK(UUT_->v8_(in, kInputStride, out, kOutputStride,
- NULL, 0, kChangeFilters[kInitialSubPelOffset],
- kInputPixelStep, Width(), Height()));
+ ASM_REGISTER_STATE_CHECK(
+ UUT_->v8_(in, kInputStride, out, kOutputStride,
+ NULL, 0, kChangeFilters[kInitialSubPelOffset],
+ kInputPixelStep, Width(), Height()));
for (int y = 0; y < Height(); ++y) {
const int kFilterPeriodAdjust = (y >> 3) << 3;
@@ -522,12 +524,11 @@
}
/* Test the horizontal and vertical filters in combination. */
- REGISTER_STATE_CHECK(UUT_->hv8_(in, kInputStride, out, kOutputStride,
- kChangeFilters[kInitialSubPelOffset],
- kInputPixelStep,
- kChangeFilters[kInitialSubPelOffset],
- kInputPixelStep,
- Width(), Height()));
+ ASM_REGISTER_STATE_CHECK(
+ UUT_->hv8_(in, kInputStride, out, kOutputStride,
+ kChangeFilters[kInitialSubPelOffset], kInputPixelStep,
+ kChangeFilters[kInitialSubPelOffset], kInputPixelStep,
+ Width(), Height()));
for (int y = 0; y < Height(); ++y) {
const int kFilterPeriodAdjustY = (y >> 3) << 3;
@@ -560,10 +561,10 @@
for (int frac = 0; frac < 16; ++frac) {
for (int step = 1; step <= 32; ++step) {
/* Test the horizontal and vertical filters in combination. */
- REGISTER_STATE_CHECK(UUT_->hv8_(in, kInputStride, out, kOutputStride,
- eighttap[frac], step,
- eighttap[frac], step,
- Width(), Height()));
+ ASM_REGISTER_STATE_CHECK(UUT_->hv8_(in, kInputStride, out, kOutputStride,
+ eighttap[frac], step,
+ eighttap[frac], step,
+ Width(), Height()));
CheckGuardBlocks();
diff --git a/test/cpu_speed_test.cc b/test/cpu_speed_test.cc
index 961a0b8..9dca601 100644
--- a/test/cpu_speed_test.cc
+++ b/test/cpu_speed_test.cc
@@ -7,8 +7,6 @@
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
-#include <climits>
-#include <vector>
#include "third_party/googletest/src/include/gtest/gtest.h"
#include "test/codec_factory.h"
#include "test/encode_test_driver.h"
@@ -20,9 +18,9 @@
const int kMaxPSNR = 100;
-class CpuSpeedTest : public ::libvpx_test::EncoderTest,
- public ::libvpx_test::CodecTestWith2Params<
- libvpx_test::TestMode, int> {
+class CpuSpeedTest
+ : public ::libvpx_test::EncoderTest,
+ public ::libvpx_test::CodecTestWith2Params<libvpx_test::TestMode, int> {
protected:
CpuSpeedTest()
: EncoderTest(GET_PARAM(0)),
@@ -60,11 +58,6 @@
}
}
- virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
- if (pkt->data.frame.flags & VPX_FRAME_IS_KEY) {
- }
- }
-
virtual void PSNRPktHook(const vpx_codec_cx_pkt_t *pkt) {
if (pkt->data.psnr.psnr[0] < min_psnr_)
min_psnr_ = pkt->data.psnr.psnr[0];
@@ -126,11 +119,11 @@
ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
}
+
TEST_P(CpuSpeedTest, TestLowBitrate) {
// Validate that this clip encodes and decodes without a mismatch
// when passing in a very high min q. This pushes the encoder to producing
// lots of small partitions which might will test the other condition.
-
cfg_.rc_2pass_vbr_minsection_pct = 5;
cfg_.rc_2pass_vbr_minsection_pct = 2000;
cfg_.rc_target_bitrate = 200;
@@ -142,11 +135,6 @@
ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
}
-using std::tr1::make_tuple;
-
-#define VP9_FACTORY \
- static_cast<const libvpx_test::CodecFactory*> (&libvpx_test::kVP9)
-
VP9_INSTANTIATE_TEST_CASE(
CpuSpeedTest,
::testing::Values(::libvpx_test::kTwoPassGood, ::libvpx_test::kOnePassGood,
diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index e6a20fb..b7c2dcc 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -311,9 +311,9 @@
test_input_block[j] = src[j] - dst[j];
}
- REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
- test_temp_block, pitch_));
- REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
+ test_temp_block, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
for (int j = 0; j < kNumCoeffs; ++j) {
const uint32_t diff = dst[j] - src[j];
@@ -344,7 +344,7 @@
input_block[j] = rnd.Rand8() - rnd.Rand8();
fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
- REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
// The minimum quant value is 4.
for (int j = 0; j < kNumCoeffs; ++j)
@@ -375,8 +375,8 @@
}
fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
- REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
- output_block, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
+ output_block, pitch_));
// The minimum quant value is 4.
for (int j = 0; j < kNumCoeffs; ++j) {
@@ -421,7 +421,7 @@
for (int j = 1; j < kNumCoeffs; ++j)
output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
- REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
for (int j = 0; j < kNumCoeffs; ++j)
EXPECT_EQ(ref[j], dst[j]);
@@ -450,7 +450,7 @@
for (int j = 0; j < kNumCoeffs; ++j)
coeff[j] = round(out_r[j]);
- REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
+ ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
for (int j = 0; j < kNumCoeffs; ++j) {
const uint32_t diff = dst[j] - src[j];
diff --git a/test/dct32x32_test.cc b/test/dct32x32_test.cc
index 501c696..02250a9 100644
--- a/test/dct32x32_test.cc
+++ b/test/dct32x32_test.cc
@@ -112,8 +112,8 @@
test_input_block[j] = src[j] - dst[j];
}
- REGISTER_STATE_CHECK(fwd_txfm_(test_input_block, test_temp_block, 32));
- REGISTER_STATE_CHECK(inv_txfm_(test_temp_block, dst, 32));
+ ASM_REGISTER_STATE_CHECK(fwd_txfm_(test_input_block, test_temp_block, 32));
+ ASM_REGISTER_STATE_CHECK(inv_txfm_(test_temp_block, dst, 32));
for (int j = 0; j < kNumCoeffs; ++j) {
const uint32_t diff = dst[j] - src[j];
@@ -150,7 +150,7 @@
const int stride = 32;
vp9_fdct32x32_c(input_block, output_ref_block, stride);
- REGISTER_STATE_CHECK(fwd_txfm_(input_block, output_block, stride));
+ ASM_REGISTER_STATE_CHECK(fwd_txfm_(input_block, output_block, stride));
if (version_ == 0) {
for (int j = 0; j < kNumCoeffs; ++j)
@@ -189,7 +189,8 @@
const int stride = 32;
vp9_fdct32x32_c(input_extreme_block, output_ref_block, stride);
- REGISTER_STATE_CHECK(fwd_txfm_(input_extreme_block, output_block, stride));
+ ASM_REGISTER_STATE_CHECK(
+ fwd_txfm_(input_extreme_block, output_block, stride));
// The minimum quant value is 4.
for (int j = 0; j < kNumCoeffs; ++j) {
@@ -230,7 +231,7 @@
reference_32x32_dct_2d(in, out_r);
for (int j = 0; j < kNumCoeffs; ++j)
coeff[j] = round(out_r[j]);
- REGISTER_STATE_CHECK(inv_txfm_(coeff, dst, 32));
+ ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, dst, 32));
for (int j = 0; j < kNumCoeffs; ++j) {
const int diff = dst[j] - src[j];
const int error = diff * diff;
diff --git a/test/decode_test_driver.cc b/test/decode_test_driver.cc
index 8bea4cc..187a1c7 100644
--- a/test/decode_test_driver.cc
+++ b/test/decode_test_driver.cc
@@ -32,7 +32,7 @@
void *user_priv) {
vpx_codec_err_t res_dec;
InitOnce();
- REGISTER_STATE_CHECK(
+ API_REGISTER_STATE_CHECK(
res_dec = vpx_codec_decode(&decoder_,
cxdata, static_cast<unsigned int>(size),
user_priv, 0));
diff --git a/test/decode_to_md5.sh b/test/decode_to_md5.sh
index 28e29c6..6cb7d0e 100755
--- a/test/decode_to_md5.sh
+++ b/test/decode_to_md5.sh
@@ -39,7 +39,8 @@
return 1
fi
- eval "${decoder}" "${input_file}" "${output_file}" ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+ ${devnull}
[ -e "${output_file}" ] || return 1
diff --git a/test/decode_with_drops.sh b/test/decode_with_drops.sh
index 12e17de..9b2edb6 100755
--- a/test/decode_with_drops.sh
+++ b/test/decode_with_drops.sh
@@ -39,7 +39,8 @@
return 1
fi
- eval "${decoder}" "${input_file}" "${output_file}" "${drop_mode}" ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+ "${drop_mode}" ${devnull}
[ -e "${output_file}" ] || return 1
}
diff --git a/test/encode_test_driver.cc b/test/encode_test_driver.cc
index 709831e..b30c3e0 100644
--- a/test/encode_test_driver.cc
+++ b/test/encode_test_driver.cc
@@ -59,7 +59,7 @@
}
// Encode the frame
- REGISTER_STATE_CHECK(
+ API_REGISTER_STATE_CHECK(
res = vpx_codec_encode(&encoder_,
video.img(), video.pts(), video.duration(),
frame_flags, deadline_));
diff --git a/test/fdct4x4_test.cc b/test/fdct4x4_test.cc
index ec233d3..ab636bf 100644
--- a/test/fdct4x4_test.cc
+++ b/test/fdct4x4_test.cc
@@ -79,9 +79,9 @@
test_input_block[j] = src[j] - dst[j];
}
- REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
- test_temp_block, pitch_));
- REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
+ test_temp_block, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
for (int j = 0; j < kNumCoeffs; ++j) {
const uint32_t diff = dst[j] - src[j];
@@ -114,7 +114,7 @@
input_block[j] = rnd.Rand8() - rnd.Rand8();
fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
- REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
// The minimum quant value is 4.
for (int j = 0; j < kNumCoeffs; ++j)
@@ -145,8 +145,8 @@
}
fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
- REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
- output_block, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
+ output_block, pitch_));
// The minimum quant value is 4.
for (int j = 0; j < kNumCoeffs; ++j) {
@@ -175,7 +175,7 @@
fwd_txfm_ref(in, coeff, pitch_, tx_type_);
- REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
+ ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
for (int j = 0; j < kNumCoeffs; ++j) {
const uint32_t diff = dst[j] - src[j];
diff --git a/test/fdct8x8_test.cc b/test/fdct8x8_test.cc
index 146aa31..3e14c24 100644
--- a/test/fdct8x8_test.cc
+++ b/test/fdct8x8_test.cc
@@ -68,7 +68,7 @@
// Initialize a test block with input range [-255, 255].
for (int j = 0; j < 64; ++j)
test_input_block[j] = rnd.Rand8() - rnd.Rand8();
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
RunFwdTxfm(test_input_block, test_output_block, pitch_));
for (int j = 0; j < 64; ++j) {
@@ -97,7 +97,7 @@
// Initialize a test block with input range [-15, 15].
for (int j = 0; j < 64; ++j)
test_input_block[j] = (rnd.Rand8() >> 4) - (rnd.Rand8() >> 4);
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
RunFwdTxfm(test_input_block, test_output_block, pitch_));
for (int j = 0; j < 64; ++j) {
@@ -139,7 +139,7 @@
test_input_block[j] = src[j] - dst[j];
}
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
RunFwdTxfm(test_input_block, test_temp_block, pitch_));
for (int j = 0; j < 64; ++j) {
if (test_temp_block[j] > 0) {
@@ -152,7 +152,7 @@
test_temp_block[j] *= 4;
}
}
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
RunInvTxfm(test_temp_block, dst, pitch_));
for (int j = 0; j < 64; ++j) {
@@ -202,11 +202,11 @@
test_input_block[j] = src[j] - dst[j];
}
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
RunFwdTxfm(test_input_block, test_temp_block, pitch_));
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
fwd_txfm_ref(test_input_block, ref_temp_block, pitch_, tx_type_));
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
RunInvTxfm(test_temp_block, dst, pitch_));
for (int j = 0; j < 64; ++j) {
diff --git a/test/idct_test.cc b/test/idct_test.cc
index 1bbf80a..c7f609d 100644
--- a/test/idct_test.cc
+++ b/test/idct_test.cc
@@ -52,7 +52,7 @@
TEST_P(IDCTTest, TestAllZeros) {
int i;
- REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
+ ASM_REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
for (i = 0; i < 256; i++)
if ((i & 0xF) < 4 && i < 64)
@@ -65,7 +65,7 @@
int i;
input[0] = 4;
- REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
+ ASM_REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
for (i = 0; i < 256; i++)
if ((i & 0xF) < 4 && i < 64)
@@ -79,7 +79,7 @@
for (i = 0; i < 256; i++) predict[i] = i;
input[0] = 4;
- REGISTER_STATE_CHECK(UUT(input, predict, 16, output, 16));
+ ASM_REGISTER_STATE_CHECK(UUT(input, predict, 16, output, 16));
for (i = 0; i < 256; i++)
if ((i & 0xF) < 4 && i < 64)
@@ -93,7 +93,7 @@
for (i = 0; i < 16; i++) input[i] = i;
- REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
+ ASM_REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
for (i = 0; i < 256; i++)
if ((i & 0xF) > 3 || i > 63)
diff --git a/test/intrapred_test.cc b/test/intrapred_test.cc
index cefe192..d93251d 100644
--- a/test/intrapred_test.cc
+++ b/test/intrapred_test.cc
@@ -261,10 +261,10 @@
virtual void Predict(MB_PREDICTION_MODE mode) {
mbptr_->mode_info_context->mbmi.mode = mode;
- REGISTER_STATE_CHECK(pred_fn_(mbptr_,
- data_ptr_[0] - kStride,
- data_ptr_[0] - 1, kStride,
- data_ptr_[0], kStride));
+ ASM_REGISTER_STATE_CHECK(pred_fn_(mbptr_,
+ data_ptr_[0] - kStride,
+ data_ptr_[0] - 1, kStride,
+ data_ptr_[0], kStride));
}
intra_pred_y_fn_t pred_fn_;
diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 8c83034..63c7941 100644
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -55,7 +55,8 @@
// Check results match.
EXPECT_EQ(expected_res_dec, res_dec)
- << "Results don't match: frame number = " << video.frame_number();
+ << "Results don't match: frame number = " << video.frame_number()
+ << ". (" << decoder->DecodeError() << ")";
return !HasFailure();
}
diff --git a/test/partial_idct_test.cc b/test/partial_idct_test.cc
index f2171b2..fd24c75 100644
--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -118,8 +118,8 @@
= (output_ref_block[j] / 1828) * 1828;
}
- REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
- REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block1, dst2, size));
+ ASM_REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
+ ASM_REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block1, dst2, size));
for (int j = 0; j < block_size; ++j) {
const int diff = dst1[j] - dst2[j];
@@ -182,8 +182,8 @@
memcpy(test_coef_block2, test_coef_block1,
sizeof(*test_coef_block2) * block_size);
- REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
- REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block2, dst2, size));
+ ASM_REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
+ ASM_REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block2, dst2, size));
for (int j = 0; j < block_size; ++j) {
const int diff = dst1[j] - dst2[j];
diff --git a/test/postproc.sh b/test/postproc.sh
index c9c4e58..939a3e7 100755
--- a/test/postproc.sh
+++ b/test/postproc.sh
@@ -37,7 +37,8 @@
return 1
fi
- eval "${decoder}" "${input_file}" "${output_file}" ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+ ${devnull}
[ -e "${output_file}" ] || return 1
}
diff --git a/test/pp_filter_test.cc b/test/pp_filter_test.cc
index 86c2b0e..1144083 100644
--- a/test/pp_filter_test.cc
+++ b/test/pp_filter_test.cc
@@ -80,8 +80,9 @@
// Initialize pixels in the output to 99.
(void)vpx_memset(dst_image, 99, output_size);
- REGISTER_STATE_CHECK(GetParam()(src_image_ptr, dst_image_ptr, input_stride,
- output_stride, block_width, flimits, 16));
+ ASM_REGISTER_STATE_CHECK(
+ GetParam()(src_image_ptr, dst_image_ptr, input_stride,
+ output_stride, block_width, flimits, 16));
static const uint8_t expected_data[block_height] = {
4, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 4
diff --git a/test/register_state_check.h b/test/register_state_check.h
index 1ee149b..8d4beea 100644
--- a/test/register_state_check.h
+++ b/test/register_state_check.h
@@ -13,6 +13,20 @@
#include "third_party/googletest/src/include/gtest/gtest.h"
#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+
+// ASM_REGISTER_STATE_CHECK(asm_function)
+// Minimally validates the environment pre & post function execution. This
+// variant should be used with assembly functions which are not expected to
+// fully restore the system state. See platform implementations of
+// RegisterStateCheck for details.
+//
+// API_REGISTER_STATE_CHECK(api_function)
+// Performs all the checks done by ASM_REGISTER_STATE_CHECK() and any
+// additional checks to ensure the environment is in a consistent state pre &
+// post function execution. This variant should be used with API functions.
+// See platform implementations of RegisterStateCheckXXX for details.
+//
#if defined(_WIN64)
@@ -35,11 +49,6 @@
// Compares the state of xmm[6-15] at construction with their state at
// destruction. These registers should be preserved by the callee on
// Windows x64.
-// Usage:
-// {
-// RegisterStateCheck reg_check;
-// FunctionToVerify();
-// }
class RegisterStateCheck {
public:
RegisterStateCheck() { initialized_ = StoreRegisters(&pre_context_); }
@@ -75,9 +84,9 @@
CONTEXT pre_context_;
};
-#define REGISTER_STATE_CHECK(statement) do { \
- libvpx_test::RegisterStateCheck reg_check; \
- statement; \
+#define ASM_REGISTER_STATE_CHECK(statement) do { \
+ libvpx_test::RegisterStateCheck reg_check; \
+ statement; \
} while (false)
} // namespace libvpx_test
@@ -85,8 +94,6 @@
#elif defined(CONFIG_SHARED) && defined(HAVE_NEON_ASM) && defined(CONFIG_VP9) \
&& !CONFIG_SHARED && HAVE_NEON_ASM && CONFIG_VP9
-#include "vpx/vpx_integer.h"
-
extern "C" {
// Save the d8-d15 registers into store.
void vp9_push_neon(int64_t *store);
@@ -97,11 +104,6 @@
// Compares the state of d8-d15 at construction with their state at
// destruction. These registers should be preserved by the callee on
// arm platform.
-// Usage:
-// {
-// RegisterStateCheck reg_check;
-// FunctionToVerify();
-// }
class RegisterStateCheck {
public:
RegisterStateCheck() { initialized_ = StoreRegisters(pre_store_); }
@@ -129,9 +131,9 @@
int64_t pre_store_[8];
};
-#define REGISTER_STATE_CHECK(statement) do { \
- libvpx_test::RegisterStateCheck reg_check; \
- statement; \
+#define ASM_REGISTER_STATE_CHECK(statement) do { \
+ libvpx_test::RegisterStateCheck reg_check; \
+ statement; \
} while (false)
} // namespace libvpx_test
@@ -141,10 +143,54 @@
namespace libvpx_test {
class RegisterStateCheck {};
-#define REGISTER_STATE_CHECK(statement) statement
+#define ASM_REGISTER_STATE_CHECK(statement) statement
} // namespace libvpx_test
#endif // _WIN64
+#if ARCH_X86 || ARCH_X86_64
+#if defined(__GNUC__)
+
+namespace libvpx_test {
+
+// Checks the FPU tag word pre/post execution to ensure emms has been called.
+class RegisterStateCheckMMX {
+ public:
+ RegisterStateCheckMMX() {
+ __asm__ volatile("fstenv %0" : "=rm"(pre_fpu_env_));
+ }
+ ~RegisterStateCheckMMX() { EXPECT_TRUE(Check()); }
+
+ private:
+ // Checks the FPU tag word pre/post execution, returning false if not cleared
+ // to 0xffff.
+ bool Check() const {
+ EXPECT_EQ(0xffff, pre_fpu_env_[4])
+ << "FPU was in an inconsistent state prior to call";
+
+ uint16_t post_fpu_env[14];
+ __asm__ volatile("fstenv %0" : "=rm"(post_fpu_env));
+ EXPECT_EQ(0xffff, post_fpu_env[4])
+ << "FPU was left in an inconsistent state after call";
+ return !testing::Test::HasNonfatalFailure();
+ }
+
+ uint16_t pre_fpu_env_[14];
+};
+
+#define API_REGISTER_STATE_CHECK(statement) do { \
+ libvpx_test::RegisterStateCheckMMX reg_check; \
+ ASM_REGISTER_STATE_CHECK(statement); \
+} while (false)
+
+} // namespace libvpx_test
+
+#endif // __GNUC__
+#endif // ARCH_X86 || ARCH_X86_64
+
+#ifndef API_REGISTER_STATE_CHECK
+#define API_REGISTER_STATE_CHECK ASM_REGISTER_STATE_CHECK
+#endif
+
#endif // TEST_REGISTER_STATE_CHECK_H_
diff --git a/test/resize_util.sh b/test/resize_util.sh
index ab3dfd1..5e47271 100755
--- a/test/resize_util.sh
+++ b/test/resize_util.sh
@@ -38,7 +38,7 @@
return 1
fi
- eval "${resizer}" "${YUV_RAW_INPUT}" \
+ eval "${VPX_TEST_PREFIX}" "${resizer}" "${YUV_RAW_INPUT}" \
"${YUV_RAW_INPUT_WIDTH}x${YUV_RAW_INPUT_HEIGHT}" \
"${target_dimensions}" "${output_file}" ${frames_to_resize} \
${devnull}
diff --git a/test/sad_test.cc b/test/sad_test.cc
index 89d8c41..7e5bcef 100644
--- a/test/sad_test.cc
+++ b/test/sad_test.cc
@@ -149,9 +149,9 @@
const uint8_t* refs[] = {GetReference(0), GetReference(1),
GetReference(2), GetReference(3)};
- REGISTER_STATE_CHECK(GET_PARAM(2)(source_data_, source_stride_,
- refs, reference_stride_,
- results));
+ ASM_REGISTER_STATE_CHECK(GET_PARAM(2)(source_data_, source_stride_,
+ refs, reference_stride_,
+ results));
}
void CheckSADs() {
@@ -178,9 +178,9 @@
unsigned int ret;
const uint8_t* const reference = GetReference(block_idx);
- REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
- reference, reference_stride_,
- max_sad));
+ ASM_REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+ reference, reference_stride_,
+ max_sad));
return ret;
}
@@ -210,8 +210,8 @@
unsigned int ret;
const uint8_t* const reference = GetReference(block_idx);
- REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
- reference, reference_stride_));
+ ASM_REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+ reference, reference_stride_));
return ret;
}
diff --git a/test/simple_decoder.sh b/test/simple_decoder.sh
index 0be48e6..7eeaf71 100755
--- a/test/simple_decoder.sh
+++ b/test/simple_decoder.sh
@@ -37,7 +37,8 @@
return 1
fi
- eval "${decoder}" "${input_file}" "${output_file}" ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+ ${devnull}
[ -e "${output_file}" ] || return 1
}
diff --git a/test/simple_encoder.sh b/test/simple_encoder.sh
index a0b0e13..c4a6280 100755
--- a/test/simple_encoder.sh
+++ b/test/simple_encoder.sh
@@ -34,7 +34,7 @@
return 1
fi
- eval "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+ eval "${VPX_TEST_PREFIX}" "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
"${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" 9999 \
${devnull}
diff --git a/test/sixtap_predict_test.cc b/test/sixtap_predict_test.cc
index 0c600f4..ac7aa99 100644
--- a/test/sixtap_predict_test.cc
+++ b/test/sixtap_predict_test.cc
@@ -143,8 +143,9 @@
uint8_t *src = const_cast<uint8_t*>(test_data);
- REGISTER_STATE_CHECK(sixtap_predict_(&src[kSrcStride * 2 + 2 + 1], kSrcStride,
- 2, 2, dst_, kDstStride));
+ ASM_REGISTER_STATE_CHECK(
+ sixtap_predict_(&src[kSrcStride * 2 + 2 + 1], kSrcStride,
+ 2, 2, dst_, kDstStride));
for (int i = 0; i < height_; ++i)
for (int j = 0; j < width_; ++j)
@@ -169,7 +170,7 @@
xoffset, yoffset, dst_c_, kDstStride);
// Run test.
- REGISTER_STATE_CHECK(
+ ASM_REGISTER_STATE_CHECK(
sixtap_predict_(&src_[kSrcStride * 2 + 2 + 1], kSrcStride,
xoffset, yoffset, dst_, kDstStride));
diff --git a/test/subtract_test.cc b/test/subtract_test.cc
index 63e999d..2db3dd7 100644
--- a/test/subtract_test.cc
+++ b/test/subtract_test.cc
@@ -82,7 +82,7 @@
predictor += kDiffPredStride;
}
- REGISTER_STATE_CHECK(GetParam()(&be, &bd, kDiffPredStride));
+ ASM_REGISTER_STATE_CHECK(GetParam()(&be, &bd, kDiffPredStride));
base_src = *be.base_src;
src_diff = be.src_diff;
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index f9c09c6..8940027 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -653,6 +653,8 @@
e3ab35d4316c5e81325c50f5236ceca4bc0d35df vp90-2-15-segkey.webm.md5
9b7ca2cac09d34c4a5d296c1900f93b1e2f69d0d vp90-2-15-segkey_adpq.webm
8f46ba5f785d0c2170591a153e0d0d146a7c8090 vp90-2-15-segkey_adpq.webm.md5
+698a6910a97486b833073ef0c0b18d75dce57ee8 vp90-2-16-intra-only.webm
+5661b0168752969f055eec37b05fa9fa947dc7eb vp90-2-16-intra-only.webm.md5
0321d507ce62dedc8a51b4e9011f7a19aed9c3dc vp91-2-04-yuv444.webm
367e423dd41fdb49aa028574a2cfec5c2f325c5c vp91-2-04-yuv444.webm.md5
76024eb753cdac6a5e5703aaea189d35c3c30ac7 invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.ivf
diff --git a/test/test.mk b/test/test.mk
index 85212d9..ef81ab1 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -771,6 +771,8 @@
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey_adpq.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey_adpq.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-16-intra-only.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-16-intra-only.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp91-2-04-yuv444.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp91-2-04-yuv444.webm.md5
diff --git a/test/test_vectors.cc b/test/test_vectors.cc
index 41c9e26..4955887 100644
--- a/test/test_vectors.cc
+++ b/test/test_vectors.cc
@@ -180,6 +180,7 @@
"vp90-2-14-resize-fp-tiles-8-16.webm", "vp90-2-14-resize-fp-tiles-8-1.webm",
"vp90-2-14-resize-fp-tiles-8-2.webm", "vp90-2-14-resize-fp-tiles-8-4.webm",
"vp90-2-15-segkey.webm", "vp90-2-15-segkey_adpq.webm",
+ "vp90-2-16-intra-only.webm",
"vp91-2-04-yuv444.webm",
};
const int kNumVP9TestVectors = NELEMENTS(kVP9TestVectors);
diff --git a/test/tools_common.sh b/test/tools_common.sh
index 7f32905..3e69c36 100755
--- a/test/tools_common.sh
+++ b/test/tools_common.sh
@@ -16,6 +16,7 @@
set -e
devnull='> /dev/null 2>&1'
+VPX_TEST_PREFIX=""
elog() {
echo "$@" 1>&2
@@ -204,9 +205,12 @@
local decoder="${LIBVPX_BIN_PATH}/vpxdec${VPX_TEST_EXE_SUFFIX}"
if [ -z "${pipe_input}" ]; then
- eval "${decoder}" "$input" --summary --noblit "$@" ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${decoder}" "$input" --summary --noblit "$@" \
+ ${devnull}
else
- cat "${input}" | eval "${decoder}" - --summary --noblit "$@" ${devnull}
+ cat "${input}" \
+ | eval "${VPX_TEST_PREFIX}" "${decoder}" - --summary --noblit "$@" \
+ ${devnull}
fi
}
@@ -252,16 +256,14 @@
fi
if [ -z "${pipe_input}" ]; then
- eval "${encoder}" --codec=${codec} --width=${width} --height=${height} \
- --limit=${frames} ${use_ivf} ${extra_flags} --output="${output}" \
- "${input}" \
- ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${encoder}" --codec=${codec} --width=${width} \
+ --height=${height} --limit=${frames} ${use_ivf} ${extra_flags} \
+ --output="${output}" "${input}" ${devnull}
else
cat "${input}" \
- | eval "${encoder}" --codec=${codec} --width=${width} \
- --height=${height} --limit=${frames} ${use_ivf} ${extra_flags} \
- --output="${output}" - \
- ${devnull}
+ | eval "${VPX_TEST_PREFIX}" "${encoder}" --codec=${codec} \
+ --width=${width} --height=${height} --limit=${frames} ${use_ivf} \
+ ${extra_flags} --output="${output}" - ${devnull}
fi
if [ ! -e "${output}" ]; then
@@ -351,6 +353,9 @@
--help: Display this message and exit.
--test-data-path <path to libvpx test data directory>
--show-program-output: Shows output from all programs being tested.
+ --prefix: Allows for a user specified prefix to be inserted before all test
+ programs. Grants the ability, for example, to run test programs
+ within valgrind.
--verbose: Verbose output.
When the --bin-path option is not specified the script attempts to use
@@ -400,6 +405,10 @@
LIBVPX_TEST_DATA_PATH="$2"
shift
;;
+ --prefix)
+ VPX_TEST_PREFIX="$2"
+ shift
+ ;;
--verbose)
VPX_TEST_VERBOSE_OUTPUT=yes
;;
@@ -466,6 +475,7 @@
VPX_TEST_EXE_SUFFIX=${VPX_TEST_EXE_SUFFIX}
VPX_TEST_FILTER=${VPX_TEST_FILTER}
VPX_TEST_OUTPUT_DIR=${VPX_TEST_OUTPUT_DIR}
+ VPX_TEST_PREFIX=${VPX_TEST_PREFIX}
VPX_TEST_RAND=${VPX_TEST_RAND}
VPX_TEST_RUN_DISABLED_TESTS=${VPX_TEST_RUN_DISABLED_TESTS}
VPX_TEST_SHOW_PROGRAM_OUTPUT=${VPX_TEST_SHOW_PROGRAM_OUTPUT}
diff --git a/test/twopass_encoder.sh b/test/twopass_encoder.sh
index 95d49d6..1189e51 100755
--- a/test/twopass_encoder.sh
+++ b/test/twopass_encoder.sh
@@ -34,7 +34,7 @@
return 1
fi
- eval "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+ eval "${VPX_TEST_PREFIX}" "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
"${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" \
${devnull}
diff --git a/test/variance_test.cc b/test/variance_test.cc
index 9985695..5469770 100644
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
@@ -124,7 +124,8 @@
memset(ref_, j, block_size_);
unsigned int sse;
unsigned int var;
- REGISTER_STATE_CHECK(var = variance_(src_, width_, ref_, width_, &sse));
+ ASM_REGISTER_STATE_CHECK(
+ var = variance_(src_, width_, ref_, width_, &sse));
EXPECT_EQ(0u, var) << "src values: " << i << "ref values: " << j;
}
}
@@ -139,7 +140,8 @@
}
unsigned int sse1, sse2;
unsigned int var1;
- REGISTER_STATE_CHECK(var1 = variance_(src_, width_, ref_, width_, &sse1));
+ ASM_REGISTER_STATE_CHECK(
+ var1 = variance_(src_, width_, ref_, width_, &sse1));
const unsigned int var2 = variance_ref(src_, ref_, log2width_,
log2height_, &sse2);
EXPECT_EQ(sse1, sse2);
@@ -155,7 +157,7 @@
memset(ref_ + half, 0, half);
unsigned int sse;
unsigned int var;
- REGISTER_STATE_CHECK(var = variance_(src_, width_, ref_, width_, &sse));
+ ASM_REGISTER_STATE_CHECK(var = variance_(src_, width_, ref_, width_, &sse));
const unsigned int expected = block_size_ * 255 * 255 / 4;
EXPECT_EQ(expected, var);
}
@@ -246,8 +248,8 @@
}
unsigned int sse1, sse2;
unsigned int var1;
- REGISTER_STATE_CHECK(var1 = subpel_variance_(ref_, width_ + 1, x, y,
- src_, width_, &sse1));
+ ASM_REGISTER_STATE_CHECK(var1 = subpel_variance_(ref_, width_ + 1, x, y,
+ src_, width_, &sse1));
const unsigned int var2 = subpel_variance_ref(ref_, src_, log2width_,
log2height_, x, y, &sse2);
EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
@@ -269,8 +271,9 @@
}
unsigned int sse1, sse2;
unsigned int var1;
- REGISTER_STATE_CHECK(var1 = subpel_variance_(ref_, width_ + 1, x, y,
- src_, width_, &sse1, sec_));
+ ASM_REGISTER_STATE_CHECK(
+ var1 = subpel_variance_(ref_, width_ + 1, x, y,
+ src_, width_, &sse1, sec_));
const unsigned int var2 = subpel_avg_variance_ref(ref_, src_, sec_,
log2width_, log2height_,
x, y, &sse2);
diff --git a/test/vp8cx_set_ref.sh b/test/vp8cx_set_ref.sh
index ee10056..5d760bc 100755
--- a/test/vp8cx_set_ref.sh
+++ b/test/vp8cx_set_ref.sh
@@ -39,9 +39,9 @@
return 1
fi
- eval "${encoder}" "${YUV_RAW_INPUT_WIDTH}" "${YUV_RAW_INPUT_HEIGHT}" \
- "${YUV_RAW_INPUT}" "${output_file}" "${ref_frame_num}" \
- ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${encoder}" "${YUV_RAW_INPUT_WIDTH}" \
+ "${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" \
+ "${ref_frame_num}" ${devnull}
[ -e "${output_file}" ] || return 1
}
diff --git a/test/vp9_spatial_svc_encoder.sh b/test/vp9_spatial_svc_encoder.sh
index 8c9d130..a5728f6 100755
--- a/test/vp9_spatial_svc_encoder.sh
+++ b/test/vp9_spatial_svc_encoder.sh
@@ -39,10 +39,9 @@
return 1
fi
- eval "${encoder}" -w "${YUV_RAW_INPUT_WIDTH}" -h "${YUV_RAW_INPUT_HEIGHT}" \
- -k "${max_kf}" -f "${frames_to_encode}" "$@" "${YUV_RAW_INPUT}" \
- "${output_file}" \
- ${devnull}
+ eval "${VPX_TEST_PREFIX}" "${encoder}" -w "${YUV_RAW_INPUT_WIDTH}" \
+ -h "${YUV_RAW_INPUT_HEIGHT}" -k "${max_kf}" -f "${frames_to_encode}" \
+ "$@" "${YUV_RAW_INPUT}" "${output_file}" ${devnull}
[ -e "${output_file}" ] || return 1
}
diff --git a/test/vp9_thread_test.cc b/test/vp9_thread_test.cc
index 72719a6..d7fc4ee 100644
--- a/test/vp9_thread_test.cc
+++ b/test/vp9_thread_test.cc
@@ -18,7 +18,7 @@
#if CONFIG_WEBM_IO
#include "test/webm_video_source.h"
#endif
-#include "vp9/decoder/vp9_thread.h"
+#include "vp9/common/vp9_thread.h"
namespace {
@@ -35,6 +35,15 @@
vp9_get_worker_interface()->end(&worker_);
}
+ void Run(VP9Worker* worker) {
+ const bool synchronous = GetParam();
+ if (synchronous) {
+ vp9_get_worker_interface()->execute(worker);
+ } else {
+ vp9_get_worker_interface()->launch(worker);
+ }
+ }
+
VP9Worker worker_;
};
@@ -57,12 +66,7 @@
worker_.data1 = &hook_data;
worker_.data2 = &return_value;
- const bool synchronous = GetParam();
- if (synchronous) {
- vp9_get_worker_interface()->execute(&worker_);
- } else {
- vp9_get_worker_interface()->launch(&worker_);
- }
+ Run(&worker_);
EXPECT_NE(vp9_get_worker_interface()->sync(&worker_), 0);
EXPECT_FALSE(worker_.had_error);
EXPECT_EQ(5, hook_data);
@@ -81,12 +85,7 @@
worker_.data1 = &hook_data;
worker_.data2 = &return_value;
- const bool synchronous = GetParam();
- if (synchronous) {
- vp9_get_worker_interface()->execute(&worker_);
- } else {
- vp9_get_worker_interface()->launch(&worker_);
- }
+ Run(&worker_);
EXPECT_FALSE(vp9_get_worker_interface()->sync(&worker_));
EXPECT_EQ(1, worker_.had_error);
@@ -99,6 +98,39 @@
EXPECT_FALSE(worker_.had_error);
}
+TEST_P(VP9WorkerThreadTest, EndWithoutSync) {
+ // Create a large number of threads to increase the chances of detecting a
+ // race. Doing more work in the hook is no guarantee as any race would occur
+ // post hook execution in the main thread loop driver.
+ static const int kNumWorkers = 64;
+ VP9Worker workers[kNumWorkers];
+ int hook_data[kNumWorkers];
+ int return_value[kNumWorkers];
+
+ for (int n = 0; n < kNumWorkers; ++n) {
+ vp9_get_worker_interface()->init(&workers[n]);
+ return_value[n] = 1; // return successfully from the hook
+ workers[n].hook = ThreadHook;
+ workers[n].data1 = &hook_data[n];
+ workers[n].data2 = &return_value[n];
+ }
+
+ for (int i = 0; i < 2; ++i) {
+ for (int n = 0; n < kNumWorkers; ++n) {
+ EXPECT_NE(vp9_get_worker_interface()->reset(&workers[n]), 0);
+ hook_data[n] = 0;
+ }
+
+ for (int n = 0; n < kNumWorkers; ++n) {
+ Run(&workers[n]);
+ }
+
+ for (int n = kNumWorkers - 1; n >= 0; --n) {
+ vp9_get_worker_interface()->end(&workers[n]);
+ }
+ }
+}
+
TEST(VP9WorkerThreadTest, TestInterfaceAPI) {
EXPECT_EQ(0, vp9_set_worker_interface(NULL));
EXPECT_TRUE(vp9_get_worker_interface() != NULL);
diff --git a/test/vpx_temporal_svc_encoder.sh b/test/vpx_temporal_svc_encoder.sh
index b2e968f..fcc8cb4 100755
--- a/test/vpx_temporal_svc_encoder.sh
+++ b/test/vpx_temporal_svc_encoder.sh
@@ -21,6 +21,10 @@
echo "Libvpx test data must exist in LIBVPX_TEST_DATA_PATH."
return 1
fi
+ if [ "$(vpx_config_option_enabled CONFIG_TEMPORAL_DENOISING)" != "yes" ]; then
+ elog "Warning: Temporal denoising is disabled! Spatial denoising will be " \
+ "used instead, which is probably not what you want for this test."
+ fi
}
# Runs vpx_temporal_svc_encoder using the codec specified by $1 and output file
@@ -44,8 +48,8 @@
return 1
fi
- eval "${encoder}" "${YUV_RAW_INPUT}" "${output_file}" "${codec}" \
- "${YUV_RAW_INPUT_WIDTH}" "${YUV_RAW_INPUT_HEIGHT}" \
+ eval "${VPX_TEST_PREFIX}" "${encoder}" "${YUV_RAW_INPUT}" "${output_file}" \
+ "${codec}" "${YUV_RAW_INPUT_WIDTH}" "${YUV_RAW_INPUT_HEIGHT}" \
"${timebase_num}" "${timebase_den}" "${speed}" "${frame_drop_thresh}" \
"$@" \
${devnull}
diff --git a/vp8/decoder/onyxd_if.c b/vp8/decoder/onyxd_if.c
index 29fea61..1d763b6 100644
--- a/vp8/decoder/onyxd_if.c
+++ b/vp8/decoder/onyxd_if.c
@@ -386,6 +386,7 @@
decode_exit:
pbi->common.error.setjmp = 0;
+ vp8_clear_system_state();
return retcode;
}
int vp8dx_get_raw_frame(VP8D_COMP *pbi, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags)
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 09854a5..373dbeb 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -4886,6 +4886,7 @@
if (setjmp(cpi->common.error.jmp))
{
cpi->common.error.setjmp = 0;
+ vp8_clear_system_state();
return VPX_CODEC_CORRUPT_FRAME;
}
diff --git a/vp8/encoder/x86/denoising_sse2.c b/vp8/encoder/x86/denoising_sse2.c
index b84795c..3a4cf7e 100644
--- a/vp8/encoder/x86/denoising_sse2.c
+++ b/vp8/encoder/x86/denoising_sse2.c
@@ -30,9 +30,9 @@
_mm_srli_si128(hg_fe_dc_ba, 8));
const __m128i hgfedcba = _mm_add_epi32(hgfe_dcba,
_mm_srli_si128(hgfe_dcba, 4));
- unsigned int sum_diff = _mm_cvtsi128_si32(hgfedcba);
+ unsigned int sum_diff = abs(_mm_cvtsi128_si32(hgfedcba));
- return abs(sum_diff);
+ return sum_diff;
}
int vp8_denoiser_filter_sse2(unsigned char *mc_running_avg_y,
diff --git a/vp8/vp8_dx_iface.c b/vp8/vp8_dx_iface.c
index 56394fb..b695ddc 100644
--- a/vp8/vp8_dx_iface.c
+++ b/vp8/vp8_dx_iface.c
@@ -409,6 +409,7 @@
if (setjmp(pbi->common.error.jmp))
{
pbi->common.error.setjmp = 0;
+ vp8_clear_system_state();
/* same return value as used in vp8dx_receive_compressed_data */
return -1;
}
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index e1753a1..afe831a 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -257,10 +257,14 @@
xd->mi_stride = cm->mi_stride;
}
+static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
+ return cm->frame_type == KEY_FRAME || cm->intra_only;
+}
+
static INLINE const vp9_prob* get_partition_probs(const VP9_COMMON *cm,
int ctx) {
- return cm->frame_type == KEY_FRAME ? vp9_kf_partition_probs[ctx]
- : cm->fc.partition_prob[ctx];
+ return frame_is_intra_only(cm) ? vp9_kf_partition_probs[ctx]
+ : cm->fc.partition_prob[ctx];
}
static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) {
@@ -299,10 +303,6 @@
cm->prev_mip + cm->mi_stride + 1 : NULL;
}
-static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
- return cm->frame_type == KEY_FRAME || cm->intra_only;
-}
-
static INLINE void update_partition_context(MACROBLOCKD *xd,
int mi_row, int mi_col,
BLOCK_SIZE subsize,
diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index d4fcb62..86ae648 100644
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -151,7 +151,7 @@
return clamped_mv;
}
-static MV average_split_mvs(const struct macroblockd_plane *pd, int plane,
+static MV average_split_mvs(const struct macroblockd_plane *pd,
const MODE_INFO *mi, int ref, int block) {
const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
MV res = {0, 0};
@@ -190,7 +190,7 @@
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
const MV mv = mi->mbmi.sb_type < BLOCK_8X8
- ? average_split_mvs(pd, plane, mi, ref, block)
+ ? average_split_mvs(pd, mi, ref, block)
: mi->mbmi.mv[ref].as_mv;
// TODO(jkoleszar): This clamping is done in the incorrect place for the
@@ -288,7 +288,7 @@
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
const MV mv = mi->mbmi.sb_type < BLOCK_8X8
- ? average_split_mvs(pd, plane, mi, ref, block)
+ ? average_split_mvs(pd, mi, ref, block)
: mi->mbmi.mv[ref].as_mv;
@@ -389,7 +389,7 @@
}
// Skip border extension if block is inside the frame.
- if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width ||
+ if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width - 1 ||
y0 < 0 || y0 > frame_height - 1 || y1 < 0 || y1 > frame_height - 1) {
uint8_t *buf_ptr1 = ref_frame + y0 * pre_buf->stride + x0;
// Extend the border.
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index f52dccb..99f5a89 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -717,6 +717,9 @@
add_proto qw/void vp9_quantize_fp/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp9_quantize_fp/, "$ssse3_x86_64";
+add_proto qw/void vp9_quantize_fp_32x32/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+specialize qw/vp9_quantize_fp_32x32/, "$ssse3_x86_64";
+
add_proto qw/void vp9_quantize_b/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp9_quantize_b/, "$ssse3_x86_64";
@@ -783,12 +786,10 @@
$vp9_full_search_sad_sse4_1=vp9_full_search_sadx8;
add_proto qw/int vp9_refining_search_sad/, "const struct macroblock *x, struct mv *ref_mv, int sad_per_bit, int distance, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
-specialize qw/vp9_refining_search_sad sse3/;
-$vp9_refining_search_sad_sse3=vp9_refining_search_sadx4;
+specialize qw/vp9_refining_search_sad/;
add_proto qw/int vp9_diamond_search_sad/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
-specialize qw/vp9_diamond_search_sad sse3/;
-$vp9_diamond_search_sad_sse3=vp9_diamond_search_sadx4;
+specialize qw/vp9_diamond_search_sad/;
add_proto qw/int vp9_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
specialize qw/vp9_full_range_search/;
diff --git a/vp9/decoder/vp9_thread.c b/vp9/common/vp9_thread.c
similarity index 95%
rename from vp9/decoder/vp9_thread.c
rename to vp9/common/vp9_thread.c
index 348bdf6..1c6aec0 100644
--- a/vp9/decoder/vp9_thread.c
+++ b/vp9/common/vp9_thread.c
@@ -11,7 +11,7 @@
//
// Original source:
// http://git.chromium.org/webm/libwebp.git
-// 100644 blob 08ad4e1fecba302bf1247645e84a7d2779956bc3 src/utils/thread.c
+// 100644 blob 264210ba2807e4da47eb5d18c04cf869d89b9784 src/utils/thread.c
#include <assert.h>
#include <string.h> // for memset()
@@ -144,18 +144,19 @@
}
static void end(VP9Worker *const worker) {
- if (worker->status_ >= OK) {
#if CONFIG_MULTITHREAD
+ if (worker->impl_ != NULL) {
change_state(worker, NOT_OK);
pthread_join(worker->impl_->thread_, NULL);
pthread_mutex_destroy(&worker->impl_->mutex_);
pthread_cond_destroy(&worker->impl_->condition_);
-#else
- worker->status_ = NOT_OK;
-#endif
+ vpx_free(worker->impl_);
+ worker->impl_ = NULL;
}
- vpx_free(worker->impl_);
- worker->impl_ = NULL;
+#else
+ worker->status_ = NOT_OK;
+ assert(worker->impl_ == NULL);
+#endif
assert(worker->status_ == NOT_OK);
}
diff --git a/vp9/decoder/vp9_thread.h b/vp9/common/vp9_thread.h
similarity index 100%
rename from vp9/decoder/vp9_thread.h
rename to vp9/common/vp9_thread.h
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 8b96abb..55d5b4f 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -28,6 +28,7 @@
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_thread.h"
#include "vp9/common/vp9_tile_common.h"
#include "vp9/decoder/vp9_decodeframe.h"
@@ -38,7 +39,6 @@
#include "vp9/decoder/vp9_dthread.h"
#include "vp9/decoder/vp9_read_bit_buffer.h"
#include "vp9/decoder/vp9_reader.h"
-#include "vp9/decoder/vp9_thread.h"
#define MAX_VP9_HEADER_SIZE 80
@@ -605,8 +605,8 @@
: literal_to_filter[vp9_rb_read_literal(rb, 2)];
}
-static void read_frame_size(struct vp9_read_bit_buffer *rb,
- int *width, int *height) {
+void vp9_read_frame_size(struct vp9_read_bit_buffer *rb,
+ int *width, int *height) {
const int w = vp9_rb_read_literal(rb, 16) + 1;
const int h = vp9_rb_read_literal(rb, 16) + 1;
*width = w;
@@ -617,7 +617,7 @@
cm->display_width = cm->width;
cm->display_height = cm->height;
if (vp9_rb_read_bit(rb))
- read_frame_size(rb, &cm->display_width, &cm->display_height);
+ vp9_read_frame_size(rb, &cm->display_width, &cm->display_height);
}
static void apply_frame_size(VP9_COMMON *cm, int width, int height) {
@@ -649,7 +649,7 @@
static void setup_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
int width, height;
- read_frame_size(rb, &width, &height);
+ vp9_read_frame_size(rb, &width, &height);
apply_frame_size(cm, width, height);
setup_display_size(cm, rb);
}
@@ -669,7 +669,7 @@
}
if (!found)
- read_frame_size(rb, &width, &height);
+ vp9_read_frame_size(rb, &width, &height);
// Check that each of the frames that this frame references has valid
// dimensions.
@@ -1053,20 +1053,17 @@
return bit_reader_end;
}
-static void check_sync_code(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
- if (vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_0 ||
- vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_1 ||
- vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_2) {
- vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
- "Invalid frame sync code");
- }
-}
-
static void error_handler(void *data) {
VP9_COMMON *const cm = (VP9_COMMON *)data;
vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet");
}
+int vp9_read_sync_code(struct vp9_read_bit_buffer *const rb) {
+ return vp9_rb_read_literal(rb, 8) == VP9_SYNC_CODE_0 &&
+ vp9_rb_read_literal(rb, 8) == VP9_SYNC_CODE_1 &&
+ vp9_rb_read_literal(rb, 8) == VP9_SYNC_CODE_2;
+}
+
static BITSTREAM_PROFILE read_profile(struct vp9_read_bit_buffer *rb) {
int profile = vp9_rb_read_bit(rb);
profile |= vp9_rb_read_bit(rb) << 1;
@@ -1112,7 +1109,9 @@
cm->error_resilient_mode = vp9_rb_read_bit(rb);
if (cm->frame_type == KEY_FRAME) {
- check_sync_code(cm, rb);
+ if (!vp9_read_sync_code(rb))
+ vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
+ "Invalid frame sync code");
if (cm->profile > PROFILE_1)
cm->bit_depth = vp9_rb_read_bit(rb) ? BITS_12 : BITS_10;
cm->color_space = (COLOR_SPACE)vp9_rb_read_literal(rb, 3);
@@ -1150,9 +1149,18 @@
0 : vp9_rb_read_literal(rb, 2);
if (cm->intra_only) {
- check_sync_code(cm, rb);
+ if (!vp9_read_sync_code(rb))
+ vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
+ "Invalid frame sync code");
pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
+
+ // NOTE: The intra-only frame header does not include the specification of
+ // either the color format or color sub-sampling. VP9 specifies that the
+ // default color space should be YUV 4:2:0 in this case (normative).
+ cm->color_space = BT_601;
+ cm->subsampling_y = cm->subsampling_x = 1;
+
setup_frame_size(cm, rb);
} else {
pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
diff --git a/vp9/decoder/vp9_decodeframe.h b/vp9/decoder/vp9_decodeframe.h
index fb15645..e5d9d62 100644
--- a/vp9/decoder/vp9_decodeframe.h
+++ b/vp9/decoder/vp9_decodeframe.h
@@ -18,6 +18,7 @@
struct VP9Common;
struct VP9Decoder;
+struct vp9_read_bit_buffer;
void vp9_init_dequantizer(struct VP9Common *cm);
@@ -25,6 +26,10 @@
const uint8_t *data, const uint8_t *data_end,
const uint8_t **p_data_end);
+int vp9_read_sync_code(struct vp9_read_bit_buffer *const rb);
+void vp9_read_frame_size(struct vp9_read_bit_buffer *rb,
+ int *width, int *height);
+
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index d154e9d..e32637b 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -85,7 +85,6 @@
VP9_COMMON *const cm = &pbi->common;
int i;
- vp9_remove_common(cm);
vp9_get_worker_interface()->end(&pbi->lf_worker);
vpx_free(pbi->lf_worker.data1);
vpx_free(pbi->tile_data);
@@ -103,6 +102,7 @@
vp9_loop_filter_dealloc(&pbi->lf_row_sync, sb_rows);
}
+ vp9_remove_common(cm);
vpx_free(pbi);
}
@@ -246,6 +246,7 @@
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
+ vp9_clear_system_state();
// We do not know if the missing frame(s) was supposed to update
// any of the reference buffers, but we act conservative and
@@ -267,7 +268,10 @@
vp9_decode_frame(pbi, source, source + size, psource);
- swap_frame_buffers(pbi);
+ if (!cm->show_existing_frame)
+ swap_frame_buffers(pbi);
+ else
+ cm->frame_to_show = get_frame_new_buffer(cm);
vp9_clear_system_state();
@@ -291,6 +295,7 @@
int vp9_get_raw_frame(VP9Decoder *pbi, YV12_BUFFER_CONFIG *sd,
vp9_ppflags_t *flags) {
+ VP9_COMMON *const cm = &pbi->common;
int ret = -1;
#if !CONFIG_VP9_POSTPROC
(void)*flags;
@@ -300,15 +305,20 @@
return ret;
/* no raw frame to show!!! */
- if (pbi->common.show_frame == 0)
+ if (!cm->show_frame)
return ret;
pbi->ready_for_new_data = 1;
#if CONFIG_VP9_POSTPROC
- ret = vp9_post_proc_frame(&pbi->common, sd, flags);
+ if (!cm->show_existing_frame) {
+ ret = vp9_post_proc_frame(cm, sd, flags);
+ } else {
+ *sd = *cm->frame_to_show;
+ ret = 0;
+ }
#else
- *sd = *pbi->common.frame_to_show;
+ *sd = *cm->frame_to_show;
ret = 0;
#endif /*!CONFIG_POSTPROC*/
vp9_clear_system_state();
diff --git a/vp9/decoder/vp9_decoder.h b/vp9/decoder/vp9_decoder.h
index ab4f9a2..8e16e1c 100644
--- a/vp9/decoder/vp9_decoder.h
+++ b/vp9/decoder/vp9_decoder.h
@@ -18,10 +18,9 @@
#include "vp9/common/vp9_onyxc_int.h"
#include "vp9/common/vp9_ppflags.h"
+#include "vp9/common/vp9_thread.h"
-#include "vp9/decoder/vp9_decoder.h"
#include "vp9/decoder/vp9_dthread.h"
-#include "vp9/decoder/vp9_thread.h"
#ifdef __cplusplus
extern "C" {
diff --git a/vp9/decoder/vp9_dthread.h b/vp9/decoder/vp9_dthread.h
index a727e2a..423bd88 100644
--- a/vp9/decoder/vp9_dthread.h
+++ b/vp9/decoder/vp9_dthread.h
@@ -12,8 +12,8 @@
#define VP9_DECODER_VP9_DTHREAD_H_
#include "./vpx_config.h"
+#include "vp9/common/vp9_thread.h"
#include "vp9/decoder/vp9_reader.h"
-#include "vp9/decoder/vp9_thread.h"
struct VP9Common;
struct VP9Decoder;
diff --git a/vp9/encoder/vp9_denoiser.c b/vp9/encoder/vp9_denoiser.c
index f6393e0..dbf8cd7 100644
--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -190,12 +190,19 @@
MACROBLOCKD *filter_mbd = &mb->e_mbd;
MB_MODE_INFO *mbmi = &filter_mbd->mi[0]->mbmi;
+ MB_MODE_INFO saved_mbmi;
+ int i, j;
+ struct buf_2d saved_dst[MAX_MB_PLANE];
+ struct buf_2d saved_pre[MAX_MB_PLANE][2]; // 2 pre buffers
+
// We will restore these after motion compensation.
- MB_MODE_INFO saved_mbmi = *mbmi;
- struct buf_2d saved_dst = filter_mbd->plane[0].dst;
- struct buf_2d saved_pre[2];
- saved_pre[0] = filter_mbd->plane[0].pre[0];
- saved_pre[1] = filter_mbd->plane[0].pre[1];
+ saved_mbmi = *mbmi;
+ for (i = 0; i < MAX_MB_PLANE; ++i) {
+ for (j = 0; j < 2; ++j) {
+ saved_pre[i][j] = filter_mbd->plane[i].pre[j];
+ }
+ saved_dst[i] = filter_mbd->plane[i].dst;
+ }
mv_col = denoiser->best_sse_mv.as_mv.col;
mv_row = denoiser->best_sse_mv.as_mv.row;
@@ -224,67 +231,52 @@
// Set the pointers in the MACROBLOCKD to point to the buffers in the denoiser
// struct.
- filter_mbd->plane[0].pre[0].buf =
- block_start(denoiser->running_avg_y[frame].y_buffer,
- denoiser->running_avg_y[frame].y_stride,
- mi_row, mi_col);
- filter_mbd->plane[0].pre[0].stride = denoiser->running_avg_y[frame].y_stride;
-
- filter_mbd->plane[1].pre[0].buf =
- block_start(denoiser->running_avg_y[frame].u_buffer,
- denoiser->running_avg_y[frame].uv_stride,
- mi_row, mi_col);
- filter_mbd->plane[1].pre[0].stride = denoiser->running_avg_y[frame].uv_stride;
-
- filter_mbd->plane[2].pre[0].buf =
- block_start(denoiser->running_avg_y[frame].v_buffer,
- denoiser->running_avg_y[frame].uv_stride,
- mi_row, mi_col);
- filter_mbd->plane[2].pre[0].stride = denoiser->running_avg_y[frame].uv_stride;
-
- filter_mbd->plane[0].pre[1].buf =
- block_start(denoiser->running_avg_y[frame].y_buffer,
- denoiser->running_avg_y[frame].y_stride,
- mi_row, mi_col);
- filter_mbd->plane[0].pre[1].stride = denoiser->running_avg_y[frame].y_stride;
-
- filter_mbd->plane[1].pre[1].buf =
- block_start(denoiser->running_avg_y[frame].u_buffer,
- denoiser->running_avg_y[frame].uv_stride,
- mi_row, mi_col);
- filter_mbd->plane[1].pre[1].stride = denoiser->running_avg_y[frame].uv_stride;
-
- filter_mbd->plane[2].pre[1].buf =
- block_start(denoiser->running_avg_y[frame].v_buffer,
- denoiser->running_avg_y[frame].uv_stride,
- mi_row, mi_col);
- filter_mbd->plane[2].pre[1].stride = denoiser->running_avg_y[frame].uv_stride;
-
+ for (j = 0; j < 2; ++j) {
+ filter_mbd->plane[0].pre[j].buf =
+ block_start(denoiser->running_avg_y[frame].y_buffer,
+ denoiser->running_avg_y[frame].y_stride,
+ mi_row, mi_col);
+ filter_mbd->plane[0].pre[j].stride =
+ denoiser->running_avg_y[frame].y_stride;
+ filter_mbd->plane[1].pre[j].buf =
+ block_start(denoiser->running_avg_y[frame].u_buffer,
+ denoiser->running_avg_y[frame].uv_stride,
+ mi_row, mi_col);
+ filter_mbd->plane[1].pre[j].stride =
+ denoiser->running_avg_y[frame].uv_stride;
+ filter_mbd->plane[2].pre[j].buf =
+ block_start(denoiser->running_avg_y[frame].v_buffer,
+ denoiser->running_avg_y[frame].uv_stride,
+ mi_row, mi_col);
+ filter_mbd->plane[2].pre[j].stride =
+ denoiser->running_avg_y[frame].uv_stride;
+ }
filter_mbd->plane[0].dst.buf =
block_start(denoiser->mc_running_avg_y.y_buffer,
denoiser->mc_running_avg_y.y_stride,
mi_row, mi_col);
filter_mbd->plane[0].dst.stride = denoiser->mc_running_avg_y.y_stride;
-
filter_mbd->plane[1].dst.buf =
block_start(denoiser->mc_running_avg_y.u_buffer,
denoiser->mc_running_avg_y.uv_stride,
mi_row, mi_col);
- filter_mbd->plane[1].dst.stride = denoiser->mc_running_avg_y.y_stride;
-
+ filter_mbd->plane[1].dst.stride = denoiser->mc_running_avg_y.uv_stride;
filter_mbd->plane[2].dst.buf =
block_start(denoiser->mc_running_avg_y.v_buffer,
denoiser->mc_running_avg_y.uv_stride,
mi_row, mi_col);
- filter_mbd->plane[2].dst.stride = denoiser->mc_running_avg_y.y_stride;
+ filter_mbd->plane[2].dst.stride = denoiser->mc_running_avg_y.uv_stride;
vp9_build_inter_predictors_sby(filter_mbd, mv_row, mv_col, bs);
// Restore everything to its original state
- filter_mbd->plane[0].pre[0] = saved_pre[0];
- filter_mbd->plane[0].pre[1] = saved_pre[1];
- filter_mbd->plane[0].dst = saved_dst;
*mbmi = saved_mbmi;
+ for (i = 0; i < MAX_MB_PLANE; ++i) {
+ for (j = 0; j < 2; ++j) {
+ filter_mbd->plane[i].pre[j] = saved_pre[i][j];
+ }
+ filter_mbd->plane[i].dst = saved_dst[i];
+ }
mv_row = denoiser->best_sse_mv.as_mv.row;
mv_col = denoiser->best_sse_mv.as_mv.col;
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index eb9624d..cd0191e 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -320,10 +320,10 @@
switch (tx_size) {
case TX_32X32:
fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
- vp9_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
- p->quant, p->quant_shift, qcoeff, dqcoeff,
- pd->dequant, p->zbin_extra, eob, scan_order->scan,
- scan_order->iscan);
+ vp9_quantize_fp_32x32(coeff, 1024, x->skip_block, p->zbin, p->round_fp,
+ p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
+ pd->dequant, p->zbin_extra, eob, scan_order->scan,
+ scan_order->iscan);
break;
case TX_16X16:
vp9_fdct16x16(src_diff, coeff, diff_stride);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 1c514ba..beb2117 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -862,9 +862,7 @@
#if CONFIG_DENOISING
#ifdef OUTPUT_YUV_DENOISED
- if (cpi->oxcf.noise_sensitivity > 0) {
- yuv_denoised_file = fopen("denoised.yuv", "ab");
- }
+ yuv_denoised_file = fopen("denoised.yuv", "ab");
#endif
#endif
#ifdef OUTPUT_YUV_SRC
@@ -1122,9 +1120,7 @@
#if CONFIG_DENOISING
#ifdef OUTPUT_YUV_DENOISED
- if (cpi->oxcf.noise_sensitivity > 0) {
- fclose(yuv_denoised_file);
- }
+ fclose(yuv_denoised_file);
#endif
#endif
#ifdef OUTPUT_YUV_SRC
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index c66e003..9f8b37f 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -338,7 +338,6 @@
CYCLIC_REFRESH *cyclic_refresh;
fractional_mv_step_fp *find_fractional_mv_step;
- fractional_mv_step_comp_fp *find_fractional_mv_step_comp;
vp9_full_search_fn_t full_search_sad;
vp9_refining_search_fn_t refining_search_sad;
vp9_diamond_search_fn_t diamond_search_sad;
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index 9eb2fbc..6e04e2a 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -56,7 +56,7 @@
cpi->find_fractional_mv_step(
x, dst_mv, ref_mv, cpi->common.allow_high_precision_mv, x->errorperbit,
&v_fn_ptr, 0, mv_sf->subpel_iters_per_step, NULL, NULL, &distortion,
- &sse);
+ &sse, NULL, 0, 0);
}
xd->mi[0]->mbmi.mode = NEWMV;
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index c0edf45..4a3f895 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -172,15 +172,15 @@
return &buf[(r >> 3) * stride + (c >> 3)];
}
-/* returns subpixel variance error function */
-#define DIST(r, c) \
- vfp->svf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), z, \
- src_stride, &sse)
-
/* checks if (r, c) has better score than previous best */
#define CHECK_BETTER(v, r, c) \
if (c >= minc && c <= maxc && r >= minr && r <= maxr) { \
- thismse = (DIST(r, c)); \
+ if (second_pred == NULL) \
+ thismse = vfp->svf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), z, \
+ src_stride, &sse); \
+ else \
+ thismse = vfp->svaf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), \
+ z, src_stride, &sse, second_pred); \
if ((v = MVC(r, c) + thismse) < besterr) { \
besterr = v; \
br = r; \
@@ -266,105 +266,9 @@
int iters_per_step,
int *mvjcost, int *mvcost[2],
int *distortion,
- unsigned int *sse1) {
- const uint8_t *const z = x->plane[0].src.buf;
- const int src_stride = x->plane[0].src.stride;
- const MACROBLOCKD *xd = &x->e_mbd;
- unsigned int besterr = INT_MAX;
- unsigned int sse;
- unsigned int whichdir;
- int thismse;
- unsigned int halfiters = iters_per_step;
- unsigned int quarteriters = iters_per_step;
- unsigned int eighthiters = iters_per_step;
-
- const int y_stride = xd->plane[0].pre[0].stride;
- const int offset = bestmv->row * y_stride + bestmv->col;
- const uint8_t *const y = xd->plane[0].pre[0].buf;
-
- int rr = ref_mv->row;
- int rc = ref_mv->col;
- int br = bestmv->row * 8;
- int bc = bestmv->col * 8;
- int hstep = 4;
- const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
- const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
- const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
- const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
-
- int tr = br;
- int tc = bc;
-
- // central mv
- bestmv->row *= 8;
- bestmv->col *= 8;
-
- // calculate central point error
- besterr = vfp->vf(y + offset, y_stride, z, src_stride, sse1);
- *distortion = besterr;
- besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
-
- // 1/2 pel
- FIRST_LEVEL_CHECKS;
- if (halfiters > 1) {
- SECOND_LEVEL_CHECKS;
- }
- tr = br;
- tc = bc;
-
- // Note forced_stop: 0 - full, 1 - qtr only, 2 - half only
- if (forced_stop != 2) {
- hstep >>= 1;
- FIRST_LEVEL_CHECKS;
- if (quarteriters > 1) {
- SECOND_LEVEL_CHECKS;
- }
- tr = br;
- tc = bc;
- }
-
- if (allow_hp && vp9_use_mv_hp(ref_mv) && forced_stop == 0) {
- hstep >>= 1;
- FIRST_LEVEL_CHECKS;
- if (eighthiters > 1) {
- SECOND_LEVEL_CHECKS;
- }
- tr = br;
- tc = bc;
- }
- // These lines insure static analysis doesn't warn that
- // tr and tc aren't used after the above point.
- (void) tr;
- (void) tc;
-
- bestmv->row = br;
- bestmv->col = bc;
-
- if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
- (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
- return INT_MAX;
-
- return besterr;
-}
-
-#undef DIST
-/* returns subpixel variance error function */
-#define DIST(r, c) \
- vfp->svaf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), \
- z, src_stride, &sse, second_pred)
-
-int vp9_find_best_sub_pixel_comp_tree(const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
- int allow_hp,
- int error_per_bit,
- const vp9_variance_fn_ptr_t *vfp,
- int forced_stop,
- int iters_per_step,
- int *mvjcost, int *mvcost[2],
- int *distortion,
- unsigned int *sse1,
- const uint8_t *second_pred,
- int w, int h) {
+ unsigned int *sse1,
+ const uint8_t *second_pred,
+ int w, int h) {
const uint8_t *const z = x->plane[0].src.buf;
const int src_stride = x->plane[0].src.stride;
const MACROBLOCKD *xd = &x->e_mbd;
@@ -376,7 +280,6 @@
const unsigned int quarteriters = iters_per_step;
const unsigned int eighthiters = iters_per_step;
- DECLARE_ALIGNED_ARRAY(16, uint8_t, comp_pred, 64 * 64);
const int y_stride = xd->plane[0].pre[0].stride;
const int offset = bestmv->row * y_stride + bestmv->col;
const uint8_t *const y = xd->plane[0].pre[0].buf;
@@ -401,8 +304,13 @@
// calculate central point error
// TODO(yunqingwang): central pointer error was already calculated in full-
// pixel search, and can be passed in this function.
- vp9_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
- besterr = vfp->vf(comp_pred, w, z, src_stride, sse1);
+ if (second_pred != NULL) {
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, comp_pred, 64 * 64);
+ vp9_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
+ besterr = vfp->vf(comp_pred, w, z, src_stride, sse1);
+ } else {
+ besterr = vfp->vf(y + offset, y_stride, z, src_stride, sse1);
+ }
*distortion = besterr;
besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
@@ -456,7 +364,6 @@
#undef MVC
#undef PRE
-#undef DIST
#undef CHECK_BETTER
static INLINE int check_bounds(const MACROBLOCK *x, int row, int col,
@@ -944,94 +851,6 @@
int vp9_diamond_search_sad_c(const MACROBLOCK *x,
const search_site_config *cfg,
- MV *ref_mv, MV *best_mv,
- int search_param, int sad_per_bit, int *num00,
- const vp9_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv) {
- const MACROBLOCKD *const xd = &x->e_mbd;
- const struct buf_2d *const what = &x->plane[0].src;
- const struct buf_2d *const in_what = &xd->plane[0].pre[0];
- // search_param determines the length of the initial step and hence the number
- // of iterations
- // 0 = initial step (MAX_FIRST_STEP) pel : 1 = (MAX_FIRST_STEP/2) pel, 2 =
- // (MAX_FIRST_STEP/4) pel... etc.
- const search_site *const ss = &cfg->ss[search_param * cfg->searches_per_step];
- const int tot_steps = (cfg->ss_count / cfg->searches_per_step) - search_param;
- const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
- const uint8_t *best_address, *in_what_ref;
- int best_sad = INT_MAX;
- int best_site = 0;
- int last_site = 0;
- int i, j, step;
-
- clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
- in_what_ref = get_buf_from_mv(in_what, ref_mv);
- best_address = in_what_ref;
- *num00 = 0;
- *best_mv = *ref_mv;
-
- // Check the starting position
- best_sad = fn_ptr->sdf(what->buf, what->stride,
- best_address, in_what->stride) +
- mvsad_err_cost(x, best_mv, &fcenter_mv, sad_per_bit);
-
- i = 1;
-
- for (step = 0; step < tot_steps; step++) {
- for (j = 0; j < cfg->searches_per_step; j++) {
- const MV mv = {best_mv->row + ss[i].mv.row,
- best_mv->col + ss[i].mv.col};
- if (is_mv_in(x, &mv)) {
- int sad = fn_ptr->sdf(what->buf, what->stride,
- best_address + ss[i].offset, in_what->stride);
- if (sad < best_sad) {
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- best_site = i;
- }
- }
- }
-
- i++;
- }
-
- if (best_site != last_site) {
- best_mv->row += ss[best_site].mv.row;
- best_mv->col += ss[best_site].mv.col;
- best_address += ss[best_site].offset;
- last_site = best_site;
-#if defined(NEW_DIAMOND_SEARCH)
- while (1) {
- const MV this_mv = {best_mv->row + ss[best_site].mv.row,
- best_mv->col + ss[best_site].mv.col};
- if (is_mv_in(x, &this_mv)) {
- int sad = fn_ptr->sdf(what->buf, what->stride,
- best_address + ss[best_site].offset,
- in_what->stride);
- if (sad < best_sad) {
- sad += mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- best_mv->row += ss[best_site].mv.row;
- best_mv->col += ss[best_site].mv.col;
- best_address += ss[best_site].offset;
- continue;
- }
- }
- }
- break;
- };
-#endif
- } else if (best_address == in_what_ref) {
- (*num00)++;
- }
- }
- return best_sad;
-}
-
-int vp9_diamond_search_sadx4(const MACROBLOCK *x,
- const search_site_config *cfg,
MV *ref_mv, MV *best_mv, int search_param,
int sad_per_bit, int *num00,
const vp9_variance_fn_ptr_t *fn_ptr,
@@ -1152,8 +971,7 @@
unsigned int thissad = fn_ptr->sdf(what, what_stride, check_here,
in_what_stride);
if (thissad < bestsad) {
- thissad += mvsad_err_cost(&this_mv, &fcenter_mv,
- mvjsadcost, mvsadcost, sad_per_bit);
+ thissad += mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
if (thissad < bestsad) {
bestsad = thissad;
best_mv->row += ss[best_site].mv.row;
@@ -1427,51 +1245,6 @@
int search_range,
const vp9_variance_fn_ptr_t *fn_ptr,
const MV *center_mv) {
- const MV neighbors[4] = {{ -1, 0}, {0, -1}, {0, 1}, {1, 0}};
- const MACROBLOCKD *const xd = &x->e_mbd;
- const struct buf_2d *const what = &x->plane[0].src;
- const struct buf_2d *const in_what = &xd->plane[0].pre[0];
- const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
- unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
- get_buf_from_mv(in_what, ref_mv),
- in_what->stride) +
- mvsad_err_cost(x, ref_mv, &fcenter_mv, error_per_bit);
- int i, j;
-
- for (i = 0; i < search_range; i++) {
- int best_site = -1;
-
- for (j = 0; j < 4; j++) {
- const MV mv = {ref_mv->row + neighbors[j].row,
- ref_mv->col + neighbors[j].col};
- if (is_mv_in(x, &mv)) {
- unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
- get_buf_from_mv(in_what, &mv), in_what->stride);
- if (sad < best_sad) {
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, error_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- best_site = j;
- }
- }
- }
- }
-
- if (best_site == -1) {
- break;
- } else {
- ref_mv->row += neighbors[best_site].row;
- ref_mv->col += neighbors[best_site].col;
- }
- }
- return best_sad;
-}
-
-int vp9_refining_search_sadx4(const MACROBLOCK *x,
- MV *ref_mv, int error_per_bit,
- int search_range,
- const vp9_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv) {
const MACROBLOCKD *const xd = &x->e_mbd;
const MV neighbors[4] = {{ -1, 0}, {0, -1}, {0, 1}, {1, 0}};
const struct buf_2d *const what = &x->plane[0].src;
diff --git a/vp9/encoder/vp9_mcomp.h b/vp9/encoder/vp9_mcomp.h
index 07e410d..366f9af 100644
--- a/vp9/encoder/vp9_mcomp.h
+++ b/vp9/encoder/vp9_mcomp.h
@@ -98,27 +98,12 @@
const vp9_variance_fn_ptr_t *vfp,
int forced_stop, // 0 - full, 1 - qtr only, 2 - half only
int iters_per_step,
- int *mvjcost,
- int *mvcost[2],
- int *distortion,
- unsigned int *sse);
-
-extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree;
-
-typedef int (fractional_mv_step_comp_fp) (
- const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
- int allow_hp,
- int error_per_bit,
- const vp9_variance_fn_ptr_t *vfp,
- int forced_stop, // 0 - full, 1 - qtr only, 2 - half only
- int iters_per_step,
int *mvjcost, int *mvcost[2],
int *distortion, unsigned int *sse1,
const uint8_t *second_pred,
int w, int h);
-extern fractional_mv_step_comp_fp vp9_find_best_sub_pixel_comp_tree;
+extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree;
typedef int (*vp9_full_search_fn_t)(const MACROBLOCK *x,
const MV *ref_mv, int sad_per_bit,
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 7515f44..b2ab714 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -106,26 +106,31 @@
return const_motion;
}
-static void full_pixel_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
- BLOCK_SIZE bsize, int mi_row, int mi_col,
- int_mv *tmp_mv, int *rate_mv) {
+static int combined_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+ BLOCK_SIZE bsize, int mi_row, int mi_col,
+ int_mv *tmp_mv, int *rate_mv,
+ int64_t best_rd_sofar) {
MACROBLOCKD *xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0, 0}};
- int step_param;
- int sadpb = x->sadperbit16;
+ const int step_param = cpi->sf.mv.fullpel_search_step_param;
+ const int sadpb = x->sadperbit16;
MV mvp_full;
- int ref = mbmi->ref_frame[0];
+ const int ref = mbmi->ref_frame[0];
const MV ref_mv = mbmi->ref_mvs[ref][0].as_mv;
- int i;
-
- int tmp_col_min = x->mv_col_min;
- int tmp_col_max = x->mv_col_max;
- int tmp_row_min = x->mv_row_min;
- int tmp_row_max = x->mv_row_max;
-
+ int dis;
+ int rate_mode;
+ const int tmp_col_min = x->mv_col_min;
+ const int tmp_col_max = x->mv_col_max;
+ const int tmp_row_min = x->mv_row_min;
+ const int tmp_row_max = x->mv_row_max;
+ int rv = 0;
const YV12_BUFFER_CONFIG *scaled_ref_frame = vp9_get_scaled_ref_frame(cpi,
ref);
+ if (cpi->common.show_frame &&
+ (x->pred_mv_sad[ref] >> 3) > x->pred_mv_sad[LAST_FRAME])
+ return rv;
+
if (scaled_ref_frame) {
int i;
// Swap out the reference frame for a version that's been scaled to
@@ -133,28 +138,10 @@
// motion search code to be used without additional modifications.
for (i = 0; i < MAX_MB_PLANE; i++)
backup_yv12[i] = xd->plane[i].pre[0];
-
vp9_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
}
-
vp9_set_mv_search_range(x, &ref_mv);
- // TODO(jingning) exploiting adaptive motion search control in non-RD
- // mode decision too.
- step_param = cpi->sf.mv.fullpel_search_step_param;
-
- for (i = LAST_FRAME; i <= LAST_FRAME && cpi->common.show_frame; ++i) {
- if ((x->pred_mv_sad[ref] >> 3) > x->pred_mv_sad[i]) {
- tmp_mv->as_int = INVALID_MV;
-
- if (scaled_ref_frame) {
- int i;
- for (i = 0; i < MAX_MB_PLANE; i++)
- xd->plane[i].pre[0] = backup_yv12[i];
- }
- return;
- }
- }
assert(x->mv_best_ref_index[ref] <= 2);
if (x->mv_best_ref_index[ref] < 2)
mvp_full = mbmi->ref_mvs[ref][x->mv_best_ref_index[ref]].as_mv;
@@ -172,60 +159,39 @@
x->mv_row_min = tmp_row_min;
x->mv_row_max = tmp_row_max;
- if (scaled_ref_frame) {
- int i;
- for (i = 0; i < MAX_MB_PLANE; i++)
- xd->plane[i].pre[0] = backup_yv12[i];
- }
-
// calculate the bit cost on motion vector
mvp_full.row = tmp_mv->as_mv.row * 8;
mvp_full.col = tmp_mv->as_mv.col * 8;
+
*rate_mv = vp9_mv_bit_cost(&mvp_full, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
-}
-static void sub_pixel_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
- BLOCK_SIZE bsize, int mi_row, int mi_col,
- MV *tmp_mv) {
- MACROBLOCKD *xd = &x->e_mbd;
- MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
- struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0, 0}};
- int ref = mbmi->ref_frame[0];
- MV ref_mv = mbmi->ref_mvs[ref][0].as_mv;
- int dis;
+ rate_mode = cpi->inter_mode_cost[mbmi->mode_context[ref]]
+ [INTER_OFFSET(NEWMV)];
+ rv = !(RDCOST(x->rdmult, x->rddiv, (*rate_mv + rate_mode), 0) >
+ best_rd_sofar);
- const YV12_BUFFER_CONFIG *scaled_ref_frame = vp9_get_scaled_ref_frame(cpi,
- ref);
- if (scaled_ref_frame) {
- int i;
- // Swap out the reference frame for a version that's been scaled to
- // match the resolution of the current frame, allowing the existing
- // motion search code to be used without additional modifications.
- for (i = 0; i < MAX_MB_PLANE; i++)
- backup_yv12[i] = xd->plane[i].pre[0];
-
- vp9_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
+ if (rv) {
+ cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
+ cpi->common.allow_high_precision_mv,
+ x->errorperbit,
+ &cpi->fn_ptr[bsize],
+ cpi->sf.mv.subpel_force_stop,
+ cpi->sf.mv.subpel_iters_per_step,
+ x->nmvjointcost, x->mvcost,
+ &dis, &x->pred_sse[ref], NULL, 0, 0);
+ x->pred_mv[ref] = tmp_mv->as_mv;
}
- cpi->find_fractional_mv_step(x, tmp_mv, &ref_mv,
- cpi->common.allow_high_precision_mv,
- x->errorperbit,
- &cpi->fn_ptr[bsize],
- cpi->sf.mv.subpel_force_stop,
- cpi->sf.mv.subpel_iters_per_step,
- x->nmvjointcost, x->mvcost,
- &dis, &x->pred_sse[ref]);
-
if (scaled_ref_frame) {
int i;
for (i = 0; i < MAX_MB_PLANE; i++)
xd->plane[i].pre[0] = backup_yv12[i];
}
-
- x->pred_mv[ref] = *tmp_mv;
+ return rv;
}
+
static void model_rd_for_sb_y(VP9_COMP *cpi, BLOCK_SIZE bsize,
MACROBLOCK *x, MACROBLOCKD *xd,
int *out_rate_sum, int64_t *out_dist_sum,
@@ -544,28 +510,17 @@
continue;
if (this_mode == NEWMV) {
- int rate_mode = 0;
if (this_rd < (int64_t)(1 << num_pels_log2_lookup[bsize]))
continue;
-
- full_pixel_motion_search(cpi, x, bsize, mi_row, mi_col,
- &frame_mv[NEWMV][ref_frame], &rate_mv);
-
- if (frame_mv[NEWMV][ref_frame].as_int == INVALID_MV)
+ if (!combined_motion_search(cpi, x, bsize, mi_row, mi_col,
+ &frame_mv[NEWMV][ref_frame],
+ &rate_mv, best_rd))
continue;
-
- rate_mode = cpi->inter_mode_cost[mbmi->mode_context[ref_frame]]
- [INTER_OFFSET(this_mode)];
- if (RDCOST(x->rdmult, x->rddiv, rate_mv + rate_mode, 0) > best_rd)
- continue;
-
- sub_pixel_motion_search(cpi, x, bsize, mi_row, mi_col,
- &frame_mv[NEWMV][ref_frame].as_mv);
}
- if (this_mode != NEARESTMV)
- if (frame_mv[this_mode][ref_frame].as_int ==
- frame_mv[NEARESTMV][ref_frame].as_int)
+ if (this_mode != NEARESTMV &&
+ frame_mv[this_mode][ref_frame].as_int ==
+ frame_mv[NEARESTMV][ref_frame].as_int)
continue;
mbmi->mode = this_mode;
diff --git a/vp9/encoder/vp9_quantize.c b/vp9/encoder/vp9_quantize.c
index 4964e0f..370e1ce 100644
--- a/vp9/encoder/vp9_quantize.c
+++ b/vp9/encoder/vp9_quantize.c
@@ -104,6 +104,49 @@
*eob_ptr = eob + 1;
}
+// TODO(jingning) Refactor this file and combine functions with similar
+// operations.
+void vp9_quantize_fp_32x32_c(const int16_t *coeff_ptr, intptr_t n_coeffs,
+ int skip_block,
+ const int16_t *zbin_ptr, const int16_t *round_ptr,
+ const int16_t *quant_ptr,
+ const int16_t *quant_shift_ptr,
+ int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+ const int16_t *dequant_ptr,
+ int zbin_oq_value, uint16_t *eob_ptr,
+ const int16_t *scan, const int16_t *iscan) {
+ int i, eob = -1;
+ (void)zbin_ptr;
+ (void)quant_shift_ptr;
+ (void)zbin_oq_value;
+ (void)iscan;
+
+ vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(int16_t));
+ vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(int16_t));
+
+ if (!skip_block) {
+ for (i = 0; i < n_coeffs; i++) {
+ const int rc = scan[i];
+ const int coeff = coeff_ptr[rc];
+ const int coeff_sign = (coeff >> 31);
+ int tmp = 0;
+ int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+ if (abs_coeff >= (dequant_ptr[rc != 0] >> 2)) {
+ abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
+ abs_coeff = clamp(abs_coeff, INT16_MIN, INT16_MAX);
+ tmp = (abs_coeff * quant_ptr[rc != 0]) >> 15;
+ qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
+ dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
+ }
+
+ if (tmp)
+ eob = i;
+ }
+ }
+ *eob_ptr = eob + 1;
+}
+
void vp9_quantize_b_c(const int16_t *coeff_ptr, intptr_t count,
int skip_block,
const int16_t *zbin_ptr, const int16_t *round_ptr,
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index a8daa21..998fb3c 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1422,7 +1422,8 @@
cpi->sf.mv.subpel_iters_per_step,
x->nmvjointcost, x->mvcost,
&distortion,
- &x->pred_sse[mbmi->ref_frame[0]]);
+ &x->pred_sse[mbmi->ref_frame[0]],
+ NULL, 0, 0);
// save motion search result for use in compound prediction
seg_mvs[i][mbmi->ref_frame[0]].as_mv = *new_mv;
@@ -1838,7 +1839,7 @@
cpi->sf.mv.subpel_force_stop,
cpi->sf.mv.subpel_iters_per_step,
x->nmvjointcost, x->mvcost,
- &dis, &x->pred_sse[ref]);
+ &dis, &x->pred_sse[ref], NULL, 0, 0);
}
*rate_mv = vp9_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
@@ -1954,7 +1955,7 @@
if (bestsme < INT_MAX) {
int dis; /* TODO: use dis in distortion calculation later. */
unsigned int sse;
- bestsme = cpi->find_fractional_mv_step_comp(
+ bestsme = cpi->find_fractional_mv_step(
x, &tmp_mv,
&ref_mv[id].as_mv,
cpi->common.allow_high_precision_mv,
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 98d6825..f271182 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -253,6 +253,7 @@
}
if (speed >= 5) {
+ sf->use_quant_fp = cm->frame_type == KEY_FRAME ? 0 : 1;
sf->auto_min_max_partition_size = (cm->frame_type == KEY_FRAME) ?
RELAXED_NEIGHBORING_MIN_MAX : STRICT_NEIGHBORING_MIN_MAX;
sf->max_partition_size = BLOCK_32X32;
@@ -265,7 +266,6 @@
sf->max_delta_qindex = (cm->frame_type == KEY_FRAME) ? 20 : 15;
sf->partition_search_type = REFERENCE_PARTITION;
sf->use_nonrd_pick_mode = 1;
- sf->mv.search_method = FAST_DIAMOND;
sf->allow_skip_recode = 0;
}
@@ -287,7 +287,7 @@
sf->mv.reduce_first_step_size = 1;
}
if (speed >= 7) {
- sf->use_quant_fp = cm->frame_type == KEY_FRAME ? 0 : 1;
+ sf->mv.search_method = FAST_DIAMOND;
sf->mv.fullpel_search_step_param = 10;
sf->lpf_pick = LPF_PICK_MINIMAL_LPF;
sf->encode_breakout_thresh = (MIN(cm->width, cm->height) >= 720) ?
@@ -396,7 +396,6 @@
if (sf->mv.subpel_search_method == SUBPEL_TREE) {
cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree;
- cpi->find_fractional_mv_step_comp = vp9_find_best_sub_pixel_comp_tree;
}
cpi->mb.optimize = sf->optimize_coefficients == 1 && cpi->pass != 1;
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index f4b4e04..caa831c 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -178,7 +178,7 @@
&cpi->fn_ptr[BLOCK_16X16],
0, mv_sf->subpel_iters_per_step,
NULL, NULL,
- &distortion, &sse);
+ &distortion, &sse, NULL, 0, 0);
// Restore input state
x->plane[0].src = src;
diff --git a/vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm b/vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm
index 2d9f2b0..508e1d4 100644
--- a/vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm
+++ b/vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm
@@ -234,21 +234,18 @@
movifnidn quantq, quantmp
mova m1, [roundq] ; m1 = round
mova m2, [quantq] ; m2 = quant
-%ifidn %1, b_32x32
-; TODO(jingning) to be continued with 32x32 quantization process
+%ifidn %1, fp_32x32
pcmpeqw m5, m5
psrlw m5, 15
- paddw m0, m5
paddw m1, m5
- psrlw m0, 1 ; m0 = (m0 + 1) / 2
psrlw m1, 1 ; m1 = (m1 + 1) / 2
%endif
mova m3, [r2q] ; m3 = dequant
mov r3, qcoeffmp
mov r4, dqcoeffmp
mov r5, iscanmp
-%ifidn %1, b_32x32
- psllw m4, 1
+%ifidn %1, fp_32x32
+ psllw m2, 1
%endif
pxor m5, m5 ; m5 = dedicated zero
DEFINE_ARGS coeff, ncoeff, d1, qcoeff, dqcoeff, iscan, d2, d3, d4, d5, d6, eob
@@ -275,18 +272,19 @@
psignw m13, m10 ; m13 = reinsert sign
mova [qcoeffq+ncoeffq*2+ 0], m8
mova [qcoeffq+ncoeffq*2+16], m13
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
pabsw m8, m8
pabsw m13, m13
%endif
pmullw m8, m3 ; dqc[i] = qc[i] * q
punpckhqdq m3, m3
pmullw m13, m3 ; dqc[i] = qc[i] * q
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
psrlw m8, 1
psrlw m13, 1
psignw m8, m9
psignw m13, m10
+ psrlw m0, m3, 2
%endif
mova [dqcoeffq+ncoeffq*2+ 0], m8
mova [dqcoeffq+ncoeffq*2+16], m13
@@ -307,13 +305,17 @@
mova m10, [ coeffq+ncoeffq*2+16] ; m10 = c[i]
pabsw m6, m9 ; m6 = abs(m9)
pabsw m11, m10 ; m11 = abs(m10)
- pcmpeqw m7, m7
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
+ pcmpgtw m7, m6, m0
+ pcmpgtw m12, m11, m0
pmovmskb r6, m7
- pmovmskb r2, m7
+ pmovmskb r2, m12
+
or r6, r2
jz .skip_iter
%endif
+ pcmpeqw m7, m7
+
paddsw m6, m1 ; m6 += round
paddsw m11, m1 ; m11 += round
pmulhw m14, m6, m2 ; m14 = m6*q>>16
@@ -322,13 +324,13 @@
psignw m13, m10 ; m13 = reinsert sign
mova [qcoeffq+ncoeffq*2+ 0], m14
mova [qcoeffq+ncoeffq*2+16], m13
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
pabsw m14, m14
pabsw m13, m13
%endif
pmullw m14, m3 ; dqc[i] = qc[i] * q
pmullw m13, m3 ; dqc[i] = qc[i] * q
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
psrlw m14, 1
psrlw m13, 1
psignw m14, m9
@@ -349,7 +351,7 @@
add ncoeffq, mmsize
jl .ac_only_loop
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
jmp .accumulate_eob
.skip_iter:
mova [qcoeffq+ncoeffq*2+ 0], m5
@@ -397,3 +399,4 @@
INIT_XMM ssse3
QUANTIZE_FP fp, 7
+QUANTIZE_FP fp_32x32, 7
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 8c1f345..81fe6a6 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -50,6 +50,8 @@
VP9_COMMON_SRCS-yes += common/vp9_seg_common.c
VP9_COMMON_SRCS-yes += common/vp9_systemdependent.h
VP9_COMMON_SRCS-yes += common/vp9_textblit.h
+VP9_COMMON_SRCS-yes += common/vp9_thread.h
+VP9_COMMON_SRCS-yes += common/vp9_thread.c
VP9_COMMON_SRCS-yes += common/vp9_tile_common.h
VP9_COMMON_SRCS-yes += common/vp9_tile_common.c
VP9_COMMON_SRCS-yes += common/vp9_loopfilter.c
diff --git a/vp9/vp9_cx_iface.c b/vp9/vp9_cx_iface.c
index b150161..24dcbfa 100644
--- a/vp9/vp9_cx_iface.c
+++ b/vp9/vp9_cx_iface.c
@@ -353,7 +353,7 @@
oxcf->key_freq = cfg->kf_max_dist;
- oxcf->speed = clamp(abs(extra_cfg->cpu_used), 0, 7);
+ oxcf->speed = abs(extra_cfg->cpu_used);
oxcf->encode_breakout = extra_cfg->static_thresh;
oxcf->play_alternate = extra_cfg->enable_auto_alt_ref;
oxcf->noise_sensitivity = extra_cfg->noise_sensitivity;
diff --git a/vp9/vp9_dx_iface.c b/vp9/vp9_dx_iface.c
index c3ca7ee..2591852 100644
--- a/vp9/vp9_dx_iface.c
+++ b/vp9/vp9_dx_iface.c
@@ -20,6 +20,7 @@
#include "vp9/common/vp9_frame_buffers.h"
#include "vp9/decoder/vp9_decoder.h"
+#include "vp9/decoder/vp9_decodeframe.h"
#include "vp9/decoder/vp9_read_bit_buffer.h"
#include "vp9/vp9_iface_common.h"
@@ -98,8 +99,10 @@
static vpx_codec_err_t decoder_peek_si_internal(const uint8_t *data,
unsigned int data_sz,
vpx_codec_stream_info_t *si,
+ int *is_intra_only,
vpx_decrypt_cb decrypt_cb,
void *decrypt_state) {
+ int intra_only_flag = 0;
uint8_t clear_buffer[9];
if (data + data_sz <= data)
@@ -115,6 +118,8 @@
}
{
+ int show_frame;
+ int error_resilient;
struct vp9_read_bit_buffer rb = { data, data + data_sz, 0, NULL, NULL };
const int frame_marker = vp9_rb_read_literal(&rb, 2);
const int version = vp9_rb_read_bit(&rb);
@@ -126,6 +131,7 @@
if (version > 1) return VPX_CODEC_UNSUP_BITSTREAM;
if (vp9_rb_read_bit(&rb)) { // show an existing frame
+ vp9_rb_read_literal(&rb, 3); // Frame buffer to show.
return VPX_CODEC_OK;
}
@@ -133,18 +139,15 @@
return VPX_CODEC_UNSUP_BITSTREAM;
si->is_kf = !vp9_rb_read_bit(&rb);
+ show_frame = vp9_rb_read_bit(&rb);
+ error_resilient = vp9_rb_read_bit(&rb);
+
if (si->is_kf) {
const int sRGB = 7;
int colorspace;
- rb.bit_offset += 1; // show frame
- rb.bit_offset += 1; // error resilient
-
- if (vp9_rb_read_literal(&rb, 8) != VP9_SYNC_CODE_0 ||
- vp9_rb_read_literal(&rb, 8) != VP9_SYNC_CODE_1 ||
- vp9_rb_read_literal(&rb, 8) != VP9_SYNC_CODE_2) {
+ if (!vp9_read_sync_code(&rb))
return VPX_CODEC_UNSUP_BITSTREAM;
- }
colorspace = vp9_rb_read_literal(&rb, 3);
if (colorspace != sRGB) {
@@ -161,20 +164,28 @@
return VPX_CODEC_UNSUP_BITSTREAM;
}
}
+ vp9_read_frame_size(&rb, (int *)&si->w, (int *)&si->h);
+ } else {
+ intra_only_flag = show_frame ? 0 : vp9_rb_read_bit(&rb);
+ rb.bit_offset += error_resilient ? 0 : 2; // reset_frame_context
- // TODO(jzern): these are available on non-keyframes in intra only mode.
- si->w = vp9_rb_read_literal(&rb, 16) + 1;
- si->h = vp9_rb_read_literal(&rb, 16) + 1;
+ if (intra_only_flag) {
+ if (!vp9_read_sync_code(&rb))
+ return VPX_CODEC_UNSUP_BITSTREAM;
+ rb.bit_offset += REF_FRAMES; // refresh_frame_flags
+ vp9_read_frame_size(&rb, (int *)&si->w, (int *)&si->h);
+ }
}
}
-
+ if (is_intra_only != NULL)
+ *is_intra_only = intra_only_flag;
return VPX_CODEC_OK;
}
static vpx_codec_err_t decoder_peek_si(const uint8_t *data,
unsigned int data_sz,
vpx_codec_stream_info_t *si) {
- return decoder_peek_si_internal(data, data_sz, si, NULL, NULL);
+ return decoder_peek_si_internal(data, data_sz, si, NULL, NULL, NULL);
}
static vpx_codec_err_t decoder_get_si(vpx_codec_alg_priv_t *ctx,
@@ -266,13 +277,14 @@
// validate that we have a buffer that does not wrap around the top
// of the heap.
if (!ctx->si.h) {
+ int is_intra_only = 0;
const vpx_codec_err_t res =
- decoder_peek_si_internal(*data, data_sz, &ctx->si, ctx->decrypt_cb,
- ctx->decrypt_state);
+ decoder_peek_si_internal(*data, data_sz, &ctx->si, &is_intra_only,
+ ctx->decrypt_cb, ctx->decrypt_state);
if (res != VPX_CODEC_OK)
return res;
- if (!ctx->si.is_kf)
+ if (!ctx->si.is_kf && !is_intra_only)
return VPX_CODEC_ERROR;
}
diff --git a/vp9/vp9dx.mk b/vp9/vp9dx.mk
index 92ec6fd..1fcb36f 100644
--- a/vp9/vp9dx.mk
+++ b/vp9/vp9dx.mk
@@ -31,8 +31,6 @@
VP9_DX_SRCS-yes += decoder/vp9_detokenize.h
VP9_DX_SRCS-yes += decoder/vp9_decoder.c
VP9_DX_SRCS-yes += decoder/vp9_decoder.h
-VP9_DX_SRCS-yes += decoder/vp9_thread.c
-VP9_DX_SRCS-yes += decoder/vp9_thread.h
VP9_DX_SRCS-yes += decoder/vp9_dsubexp.c
VP9_DX_SRCS-yes += decoder/vp9_dsubexp.h
