diff options
author | Chih-hung Hsieh <chh@google.com> | 2016-01-20 17:50:13 +0000 |
---|---|---|
committer | android-build-merger <android-build-merger@google.com> | 2016-01-20 17:50:13 +0000 |
commit | b3cb8ab4ede8bb77f0bdef2715efc2c1e6267072 (patch) | |
tree | 28c4cf735dd5bd9cc8f1ccd06fff8a173b20d1cb /webrtc/modules/video_processing/main/source/content_analysis_sse2.cc | |
parent | a4acd9d6bc9b3b033d7d274316e75ee067df8d20 (diff) | |
parent | 9a337512d97e37afc142dee4fd50a41b741a87d2 (diff) | |
download | webrtc-nougat-dr1-release.tar.gz |
Merge "Merge upstream SHA 04cb763"android-cts_7.1_r1android-cts-7.1_r9android-cts-7.1_r8android-cts-7.1_r7android-cts-7.1_r6android-cts-7.1_r5android-cts-7.1_r4android-cts-7.1_r3android-cts-7.1_r29android-cts-7.1_r28android-cts-7.1_r27android-cts-7.1_r26android-cts-7.1_r25android-cts-7.1_r24android-cts-7.1_r23android-cts-7.1_r22android-cts-7.1_r21android-cts-7.1_r20android-cts-7.1_r2android-cts-7.1_r19android-cts-7.1_r18android-cts-7.1_r17android-cts-7.1_r16android-cts-7.1_r15android-cts-7.1_r14android-cts-7.1_r13android-cts-7.1_r12android-cts-7.1_r11android-cts-7.1_r10android-cts-7.1_r1android-cts-7.0_r9android-cts-7.0_r8android-cts-7.0_r7android-cts-7.0_r6android-cts-7.0_r5android-cts-7.0_r4android-cts-7.0_r33android-cts-7.0_r32android-cts-7.0_r31android-cts-7.0_r30android-cts-7.0_r3android-cts-7.0_r29android-cts-7.0_r28android-cts-7.0_r27android-cts-7.0_r26android-cts-7.0_r25android-cts-7.0_r24android-cts-7.0_r23android-cts-7.0_r22android-cts-7.0_r21android-cts-7.0_r20android-cts-7.0_r2android-cts-7.0_r19android-cts-7.0_r18android-cts-7.0_r17android-cts-7.0_r16android-cts-7.0_r15android-cts-7.0_r14android-cts-7.0_r13android-cts-7.0_r12android-cts-7.0_r11android-cts-7.0_r10android-cts-7.0_r1android-7.1.2_r9android-7.1.2_r8android-7.1.2_r6android-7.1.2_r5android-7.1.2_r4android-7.1.2_r39android-7.1.2_r38android-7.1.2_r37android-7.1.2_r36android-7.1.2_r33android-7.1.2_r32android-7.1.2_r30android-7.1.2_r3android-7.1.2_r29android-7.1.2_r28android-7.1.2_r27android-7.1.2_r25android-7.1.2_r24android-7.1.2_r23android-7.1.2_r2android-7.1.2_r19android-7.1.2_r18android-7.1.2_r17android-7.1.2_r16android-7.1.2_r15android-7.1.2_r14android-7.1.2_r13android-7.1.2_r12android-7.1.2_r11android-7.1.2_r10android-7.1.2_r1android-7.1.1_r9android-7.1.1_r8android-7.1.1_r7android-7.1.1_r61android-7.1.1_r60android-7.1.1_r6android-7.1.1_r59android-7.1.1_r58android-7.1.1_r57android-7.1.1_r56android-7.1.1_r55android-7.1.1_r54android-7.1.1_r53android-7.1.1_r52android-7.1.1_r51android-7.1.1_r50android-7.1.1_r49android-7.1.1_r48android-7.1.1_r47android-7.1.1_r46android-7.1.1_r45android-7.1.1_r44android-7.1.1_r43android-7.1.1_r42android-7.1.1_r41android-7.1.1_r40android-7.1.1_r4android-7.1.1_r39android-7.1.1_r38android-7.1.1_r35android-7.1.1_r33android-7.1.1_r32android-7.1.1_r31android-7.1.1_r3android-7.1.1_r28android-7.1.1_r27android-7.1.1_r26android-7.1.1_r25android-7.1.1_r24android-7.1.1_r23android-7.1.1_r22android-7.1.1_r21android-7.1.1_r20android-7.1.1_r2android-7.1.1_r17android-7.1.1_r16android-7.1.1_r15android-7.1.1_r14android-7.1.1_r13android-7.1.1_r12android-7.1.1_r11android-7.1.1_r10android-7.1.1_r1android-7.1.0_r7android-7.1.0_r6android-7.1.0_r5android-7.1.0_r4android-7.1.0_r3android-7.1.0_r2android-7.1.0_r1android-7.0.0_r9android-7.0.0_r8android-7.0.0_r7android-7.0.0_r6android-7.0.0_r5android-7.0.0_r4android-7.0.0_r36android-7.0.0_r35android-7.0.0_r34android-7.0.0_r33android-7.0.0_r32android-7.0.0_r31android-7.0.0_r30android-7.0.0_r3android-7.0.0_r29android-7.0.0_r28android-7.0.0_r27android-7.0.0_r24android-7.0.0_r21android-7.0.0_r19android-7.0.0_r17android-7.0.0_r15android-7.0.0_r14android-7.0.0_r13android-7.0.0_r12android-7.0.0_r11android-7.0.0_r10android-7.0.0_r1nougat-releasenougat-mr2.3-releasenougat-mr2.2-releasenougat-mr2.1-releasenougat-mr2-security-releasenougat-mr2-releasenougat-mr2-pixel-releasenougat-mr2-devnougat-mr1.8-releasenougat-mr1.7-releasenougat-mr1.6-releasenougat-mr1.5-releasenougat-mr1.4-releasenougat-mr1.3-releasenougat-mr1.2-releasenougat-mr1.1-releasenougat-mr1-volantis-releasenougat-mr1-security-releasenougat-mr1-releasenougat-mr1-flounder-releasenougat-mr1-devnougat-mr1-cts-releasenougat-mr0.5-releasenougat-dr1-releasenougat-devnougat-cts-releasenougat-bugfix-release
am: 9a337512d9
* commit '9a337512d97e37afc142dee4fd50a41b741a87d2': (797 commits)
Add tests for verifying transport feedback for audio and video.
Eliminate defines in talk/
Revert of Update with new default boringssl no-aes cipher suites. Re-enable tests. (patchset #3 id:40001 of https://codereview.webrtc.org/1550773002/ )
Remove assert which was incorrectly added to TcpPort::OnSentPacket.
Reland Connect TurnPort and TCPPort to AsyncPacketSocket::SignalSentPacket.
Update with new default boringssl no-aes cipher suites. Re-enable tests.
Revert of Connect TurnPort and TCPPort to AsyncPacketSocket::SignalSentPacket. (patchset #3 id:40001 of https://codereview.webrtc.org/1577873003/ )
Re-land: "Use an explicit identifier in Config"
Connect TurnPort and TCPPort to AsyncPacketSocket::SignalSentPacket.
Revert of Delete remnants of non-square pixel support from cricket::VideoFrame. (patchset #1 id:1 of https://codereview.webrtc.org/1586613002/ )
Remove libfuzzer trybot from default trybot set.
Add ramp-up tests for transport sequence number with and w/o audio.
Delete remnants of non-square pixel support from cricket::VideoFrame.
Fix IPAddress::ToSensitiveString() to avoid dependency on inet_ntop().
Revert of Storing raw audio sink for default audio track. (patchset #7 id:120001 of https://codereview.chromium.org/1551813002/ )
Re-enable tests that failed under Linux_Msan.
Revert of Use an explicit identifier in Config (patchset #4 id:60001 of https://codereview.webrtc.org/1538643004/ )
Roll chromium_revision 346fea9..099be58 (369082:369139)
Disable WebRtcVideoChannel2BaseTest.SendManyResizeOnce for TSan
Add build_protobuf variable.
...
Diffstat (limited to 'webrtc/modules/video_processing/main/source/content_analysis_sse2.cc')
-rw-r--r-- | webrtc/modules/video_processing/main/source/content_analysis_sse2.cc | 264 |
1 files changed, 0 insertions, 264 deletions
diff --git a/webrtc/modules/video_processing/main/source/content_analysis_sse2.cc b/webrtc/modules/video_processing/main/source/content_analysis_sse2.cc deleted file mode 100644 index 17b64ff280..0000000000 --- a/webrtc/modules/video_processing/main/source/content_analysis_sse2.cc +++ /dev/null @@ -1,264 +0,0 @@ -/* - * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. - * - * Use of this source code is governed by a BSD-style license - * that can be found in the LICENSE file in the root of the source - * tree. An additional intellectual property rights grant can be found - * in the file PATENTS. All contributing project authors may - * be found in the AUTHORS file in the root of the source tree. - */ - -#include "webrtc/modules/video_processing/main/source/content_analysis.h" - -#include <emmintrin.h> -#include <math.h> - -namespace webrtc { - -int32_t VPMContentAnalysis::TemporalDiffMetric_SSE2() { - uint32_t num_pixels = 0; // counter for # of pixels - const uint8_t* imgBufO = orig_frame_ + border_*width_ + border_; - const uint8_t* imgBufP = prev_frame_ + border_*width_ + border_; - - const int32_t width_end = ((width_ - 2*border_) & -16) + border_; - - __m128i sad_64 = _mm_setzero_si128(); - __m128i sum_64 = _mm_setzero_si128(); - __m128i sqsum_64 = _mm_setzero_si128(); - const __m128i z = _mm_setzero_si128(); - - for (uint16_t i = 0; i < (height_ - 2*border_); i += skip_num_) { - __m128i sqsum_32 = _mm_setzero_si128(); - - const uint8_t *lineO = imgBufO; - const uint8_t *lineP = imgBufP; - - // Work on 16 pixels at a time. For HD content with a width of 1920 - // this loop will run ~67 times (depending on border). Maximum for - // abs(o-p) and sum(o) will be 255. _mm_sad_epu8 produces 2 64 bit - // results which are then accumulated. There is no chance of - // rollover for these two accumulators. - // o*o will have a maximum of 255*255 = 65025. This will roll over - // a 16 bit accumulator as 67*65025 > 65535, but will fit in a - // 32 bit accumulator. - for (uint16_t j = 0; j < width_end - border_; j += 16) { - const __m128i o = _mm_loadu_si128((__m128i*)(lineO)); - const __m128i p = _mm_loadu_si128((__m128i*)(lineP)); - - lineO += 16; - lineP += 16; - - // Abs pixel difference between frames. - sad_64 = _mm_add_epi64 (sad_64, _mm_sad_epu8(o, p)); - - // sum of all pixels in frame - sum_64 = _mm_add_epi64 (sum_64, _mm_sad_epu8(o, z)); - - // Squared sum of all pixels in frame. - const __m128i olo = _mm_unpacklo_epi8(o,z); - const __m128i ohi = _mm_unpackhi_epi8(o,z); - - const __m128i sqsum_32_lo = _mm_madd_epi16(olo, olo); - const __m128i sqsum_32_hi = _mm_madd_epi16(ohi, ohi); - - sqsum_32 = _mm_add_epi32(sqsum_32, sqsum_32_lo); - sqsum_32 = _mm_add_epi32(sqsum_32, sqsum_32_hi); - } - - // Add to 64 bit running sum as to not roll over. - sqsum_64 = _mm_add_epi64(sqsum_64, - _mm_add_epi64(_mm_unpackhi_epi32(sqsum_32,z), - _mm_unpacklo_epi32(sqsum_32,z))); - - imgBufO += width_ * skip_num_; - imgBufP += width_ * skip_num_; - num_pixels += (width_end - border_); - } - - __m128i sad_final_128; - __m128i sum_final_128; - __m128i sqsum_final_128; - - // Bring sums out of vector registers and into integer register - // domain, summing them along the way. - _mm_store_si128 (&sad_final_128, sad_64); - _mm_store_si128 (&sum_final_128, sum_64); - _mm_store_si128 (&sqsum_final_128, sqsum_64); - - uint64_t *sad_final_64 = reinterpret_cast<uint64_t*>(&sad_final_128); - uint64_t *sum_final_64 = reinterpret_cast<uint64_t*>(&sum_final_128); - uint64_t *sqsum_final_64 = reinterpret_cast<uint64_t*>(&sqsum_final_128); - - const uint32_t pixelSum = sum_final_64[0] + sum_final_64[1]; - const uint64_t pixelSqSum = sqsum_final_64[0] + sqsum_final_64[1]; - const uint32_t tempDiffSum = sad_final_64[0] + sad_final_64[1]; - - // Default. - motion_magnitude_ = 0.0f; - - if (tempDiffSum == 0) return VPM_OK; - - // Normalize over all pixels. - const float tempDiffAvg = (float)tempDiffSum / (float)(num_pixels); - const float pixelSumAvg = (float)pixelSum / (float)(num_pixels); - const float pixelSqSumAvg = (float)pixelSqSum / (float)(num_pixels); - float contrast = pixelSqSumAvg - (pixelSumAvg * pixelSumAvg); - - if (contrast > 0.0) { - contrast = sqrt(contrast); - motion_magnitude_ = tempDiffAvg/contrast; - } - - return VPM_OK; -} - -int32_t VPMContentAnalysis::ComputeSpatialMetrics_SSE2() { - const uint8_t* imgBuf = orig_frame_ + border_*width_; - const int32_t width_end = ((width_ - 2 * border_) & -16) + border_; - - __m128i se_32 = _mm_setzero_si128(); - __m128i sev_32 = _mm_setzero_si128(); - __m128i seh_32 = _mm_setzero_si128(); - __m128i msa_32 = _mm_setzero_si128(); - const __m128i z = _mm_setzero_si128(); - - // Error is accumulated as a 32 bit value. Looking at HD content with a - // height of 1080 lines, or about 67 macro blocks. If the 16 bit row - // value is maxed out at 65529 for every row, 65529*1080 = 70777800, which - // will not roll over a 32 bit accumulator. - // skip_num_ is also used to reduce the number of rows - for (int32_t i = 0; i < (height_ - 2*border_); i += skip_num_) { - __m128i se_16 = _mm_setzero_si128(); - __m128i sev_16 = _mm_setzero_si128(); - __m128i seh_16 = _mm_setzero_si128(); - __m128i msa_16 = _mm_setzero_si128(); - - // Row error is accumulated as a 16 bit value. There are 8 - // accumulators. Max value of a 16 bit number is 65529. Looking - // at HD content, 1080p, has a width of 1920, 120 macro blocks. - // A mb at a time is processed at a time. Absolute max error at - // a point would be abs(0-255+255+255+255) which equals 1020. - // 120*1020 = 122400. The probability of hitting this is quite low - // on well behaved content. A specially crafted image could roll over. - // border_ could also be adjusted to concentrate on just the center of - // the images for an HD capture in order to reduce the possiblity of - // rollover. - const uint8_t *lineTop = imgBuf - width_ + border_; - const uint8_t *lineCen = imgBuf + border_; - const uint8_t *lineBot = imgBuf + width_ + border_; - - for (int32_t j = 0; j < width_end - border_; j += 16) { - const __m128i t = _mm_loadu_si128((__m128i*)(lineTop)); - const __m128i l = _mm_loadu_si128((__m128i*)(lineCen - 1)); - const __m128i c = _mm_loadu_si128((__m128i*)(lineCen)); - const __m128i r = _mm_loadu_si128((__m128i*)(lineCen + 1)); - const __m128i b = _mm_loadu_si128((__m128i*)(lineBot)); - - lineTop += 16; - lineCen += 16; - lineBot += 16; - - // center pixel unpacked - __m128i clo = _mm_unpacklo_epi8(c,z); - __m128i chi = _mm_unpackhi_epi8(c,z); - - // left right pixels unpacked and added together - const __m128i lrlo = _mm_add_epi16(_mm_unpacklo_epi8(l,z), - _mm_unpacklo_epi8(r,z)); - const __m128i lrhi = _mm_add_epi16(_mm_unpackhi_epi8(l,z), - _mm_unpackhi_epi8(r,z)); - - // top & bottom pixels unpacked and added together - const __m128i tblo = _mm_add_epi16(_mm_unpacklo_epi8(t,z), - _mm_unpacklo_epi8(b,z)); - const __m128i tbhi = _mm_add_epi16(_mm_unpackhi_epi8(t,z), - _mm_unpackhi_epi8(b,z)); - - // running sum of all pixels - msa_16 = _mm_add_epi16(msa_16, _mm_add_epi16(chi, clo)); - - clo = _mm_slli_epi16(clo, 1); - chi = _mm_slli_epi16(chi, 1); - const __m128i sevtlo = _mm_subs_epi16(clo, tblo); - const __m128i sevthi = _mm_subs_epi16(chi, tbhi); - const __m128i sehtlo = _mm_subs_epi16(clo, lrlo); - const __m128i sehthi = _mm_subs_epi16(chi, lrhi); - - clo = _mm_slli_epi16(clo, 1); - chi = _mm_slli_epi16(chi, 1); - const __m128i setlo = _mm_subs_epi16(clo, _mm_add_epi16(lrlo, tblo)); - const __m128i sethi = _mm_subs_epi16(chi, _mm_add_epi16(lrhi, tbhi)); - - // Add to 16 bit running sum - se_16 = _mm_add_epi16(se_16, _mm_max_epi16(setlo, - _mm_subs_epi16(z, setlo))); - se_16 = _mm_add_epi16(se_16, _mm_max_epi16(sethi, - _mm_subs_epi16(z, sethi))); - sev_16 = _mm_add_epi16(sev_16, _mm_max_epi16(sevtlo, - _mm_subs_epi16(z, sevtlo))); - sev_16 = _mm_add_epi16(sev_16, _mm_max_epi16(sevthi, - _mm_subs_epi16(z, sevthi))); - seh_16 = _mm_add_epi16(seh_16, _mm_max_epi16(sehtlo, - _mm_subs_epi16(z, sehtlo))); - seh_16 = _mm_add_epi16(seh_16, _mm_max_epi16(sehthi, - _mm_subs_epi16(z, sehthi))); - } - - // Add to 32 bit running sum as to not roll over. - se_32 = _mm_add_epi32(se_32, _mm_add_epi32(_mm_unpackhi_epi16(se_16,z), - _mm_unpacklo_epi16(se_16,z))); - sev_32 = _mm_add_epi32(sev_32, _mm_add_epi32(_mm_unpackhi_epi16(sev_16,z), - _mm_unpacklo_epi16(sev_16,z))); - seh_32 = _mm_add_epi32(seh_32, _mm_add_epi32(_mm_unpackhi_epi16(seh_16,z), - _mm_unpacklo_epi16(seh_16,z))); - msa_32 = _mm_add_epi32(msa_32, _mm_add_epi32(_mm_unpackhi_epi16(msa_16,z), - _mm_unpacklo_epi16(msa_16,z))); - - imgBuf += width_ * skip_num_; - } - - __m128i se_128; - __m128i sev_128; - __m128i seh_128; - __m128i msa_128; - - // Bring sums out of vector registers and into integer register - // domain, summing them along the way. - _mm_store_si128 (&se_128, _mm_add_epi64(_mm_unpackhi_epi32(se_32,z), - _mm_unpacklo_epi32(se_32,z))); - _mm_store_si128 (&sev_128, _mm_add_epi64(_mm_unpackhi_epi32(sev_32,z), - _mm_unpacklo_epi32(sev_32,z))); - _mm_store_si128 (&seh_128, _mm_add_epi64(_mm_unpackhi_epi32(seh_32,z), - _mm_unpacklo_epi32(seh_32,z))); - _mm_store_si128 (&msa_128, _mm_add_epi64(_mm_unpackhi_epi32(msa_32,z), - _mm_unpacklo_epi32(msa_32,z))); - - uint64_t *se_64 = reinterpret_cast<uint64_t*>(&se_128); - uint64_t *sev_64 = reinterpret_cast<uint64_t*>(&sev_128); - uint64_t *seh_64 = reinterpret_cast<uint64_t*>(&seh_128); - uint64_t *msa_64 = reinterpret_cast<uint64_t*>(&msa_128); - - const uint32_t spatialErrSum = se_64[0] + se_64[1]; - const uint32_t spatialErrVSum = sev_64[0] + sev_64[1]; - const uint32_t spatialErrHSum = seh_64[0] + seh_64[1]; - const uint32_t pixelMSA = msa_64[0] + msa_64[1]; - - // Normalize over all pixels. - const float spatialErr = (float)(spatialErrSum >> 2); - const float spatialErrH = (float)(spatialErrHSum >> 1); - const float spatialErrV = (float)(spatialErrVSum >> 1); - const float norm = (float)pixelMSA; - - // 2X2: - spatial_pred_err_ = spatialErr / norm; - - // 1X2: - spatial_pred_err_h_ = spatialErrH / norm; - - // 2X1: - spatial_pred_err_v_ = spatialErrV / norm; - - return VPM_OK; -} - -} // namespace webrtc |