- 27 Dec, 2017 9 commits
-
-
Yaowu Xu authored
Change-Id: I523d9a89493895eb6a7af1df30a39f36ca9f159f
-
Cheng Chen authored
Copy and x, y only convolve functions should be identical to the original 2d function. Change-Id: I2b84a1eed775e8d7cba9bc3baa6352227a0326c8
-
Frederic Barbier authored
This experiment has been adopted, we can simplify the code by dropping the associated preprocessor conditionals. Change-Id: Idf52f49d953b422f7789247df966d238fc34299b
-
Linfeng Zhang authored
Change-Id: I8e205e5b6310b345065200cfdac23f30badc3caa
-
Yaowu Xu authored
Change-Id: I69916bb6390dd9275341d8cd3fae2d8961e1cae3
-
Linfeng Zhang authored
Tested with 720p encoding and av1_cost_coeffs() saves 18% time. Change-Id: If6de7c539c4b01a3066bdc267fb375dfe77c2c50
-
Jingning Han authored
Sub8x8 blocks will not check the extended region in motion field. For regular block sizes, limit the extended region to be 3 points, down from 9 points check. Change-Id: I70f2631aa726ad01ee6bb83fffdf71ef82505888
-
Jingning Han authored
When either component of a motion vector is above 4096, drop this motion vector from the motion vector reference frame for later motion field projection use. The coding performance change is close to 0 for lowres and midres. This ensures that the motion vector and reference frame tuple can be efficiently stored within 32 bits. Change-Id: I9ae60a5caab2d3f49200abb5415532d82986839f
-
Tom Finegan authored
- Make the add_4bytes_obusize experiment part of the obu experiment. - Remove the add_4bytes_obusize experiment flags. - Update the encoder, decoder, and tooling sources. BUG=aomedia:1125 Change-Id: Ia5c443c855e52618257b39c44ca2632703bf83fd
-
- 26 Dec, 2017 2 commits
-
-
Zoe Liu authored
Change-Id: I40232b965ba28b4dcef7ebfe8697679f6b0ffc96
-
Zoe Liu authored
Use the frame ID that indicates the frame display order to identify whether two different reference frames exist for inter-coded frames. If there is only one unique reference valid in the reference buffer, there is no need to signal reference_mode. Instead, the decoder may identify such scenario and set reference_mode to SINGLE_REFERENCE. Change-Id: If7d374f5355f153c50b408be5a9956a833c976c3
-
- 25 Dec, 2017 3 commits
-
-
Debargha Mukherjee authored
Adjusst txfm stage ranges so that Cfg tests pass, however tests with --enable-coefficient-range-checking are still failing. Those are to be addressed subsequently. Change-Id: I1b15406a525a78322211e69a0d3a796823bb645e
-
Angie Chiang authored
Check if the global motion undergoes non-translation mode. If that is the case, do not force a newmv mode into globalmv mode. This will resolve a mismatch bug of txk_sel. Change-Id: I589f1619427f02045f8cf49bdea05f348f076236
-
Jingning Han authored
Make the translation global motion condition on block size consistent to support 4x16 and 16x4 coding block sizes. Change-Id: I23a0d04ca6cc08e8f29055c2c37f11b6ca204fec
-
- 24 Dec, 2017 2 commits
-
-
Yunqing Wang authored
Part2 of ext-warped-motion experiment was removed. Change-Id: I1b7aebad1fd7cb60d92c3ce959d5101228addae2
-
Debargha Mukherjee authored
Bakes in DISABLE_VARTX_FOR_CHROMA/DISABLE_SMLTX_FOR_CHROMA_INTRA macros. Cleans up code and removes the unwieldy uv_txsize_lookup array. Change-Id: I9aec8bfcfe218791517ab18e427bc645687916c3
-
- 23 Dec, 2017 8 commits
-
-
Yunqing Wang authored
Added optimized convolve functions for single reference case, so that no separate post rounding is needed and the result is written to the destination buffer directly. Duplicate code will be cleaned up later. Change-Id: Iffc0cc6e135b8b6f45a95c314d63368f5aa35f34
-
Yue Chen authored
Change-Id: I70ebb6ada7ec4a975a8984a2e1ea2fa51664a786
-
Zoe Liu authored
Encoder now writes the binary symbol for show_existing_frame to indicate the value of reset_decoder_state. The default value now is set as zero regardless. This enables the bitstream match with the decoder. With this patch when fwd-kf is enabled, one additional bit is signaled at the frame header for show_existing_frame, but without any coding performance impact. Change-Id: Ib0137e6a513011c0ac9506837d32e9ad405508af
-
Jingning Han authored
Reduce the reference motion vector search region over the spatial neighbor blocks for sub8x8 block sizes, in order to reduce the worst case context model parsing latency. Change-Id: I77a2a25483836cc02cf1784c93566fa7cff40fc8
-
Sarah Parker authored
The speed feature FINAL_PASS_TRELLIS_OPT is meant to disable optimize_b during the transform search but allow it for the final encode of blocks. There was a previously existing macro called DISABLE_TRELLISQ_SEARCH, which does the same thing. This patch merges the functionality so the macro serves only to enable the speed feature. Change-Id: Ieee70f97f817998b7ca275f6e4647cc89a330ad6
-
Urvang Joshi authored
Change-Id: I4658238c32ed419736cd5db2dd4fc26e9d8e50c5
-
Debargha Mukherjee authored
Only the col transforms are needed since the inverse transform is designed to do row first and then col. So the row transform can reuse the same configuration as the row transform of a square transform of the same size. Change-Id: I55e0bd6fca2765679be90364a65393e1787f42fe
-
Sarah Parker authored
0.05% drop in performance for 10 bit 0.03% drop in performance for 12 bit Updated relevant tests: - Use the fadst4 function from VP9 as the reference. - Update some max/avg error thresholds Change-Id: Ic8c5b591eea3309427d2bb42828d44e640f718a1
-
- 22 Dec, 2017 16 commits
-
-
Hui Su authored
Enable palette mode for 4x4, 4x8, 8x4, 4x16, 16x4, 8x32, 32x8, 16x64, 64x16 0.8% gain on screen_content keyframe coding. Change-Id: Ic3c089b74171ace9082a0d3ad9e27c8a27553789
-
Debargha Mukherjee authored
Only the row transforms are needed since the forward transform is designed to do col first and then row. So the col transform can reuse the same configuration as the col transform of a square block of the same size. Change-Id: I35d88146d8f8afeb685e958cb8df447f4d2b7aa1
-
Luc Trudeau authored
Small code cleaning and partially resolves issue 1171. I can't confirm that it's a complete fix because the code asserts further a long aomenc: ../av1/common/blockd.h:1064: av1_get_tx_type: Assertion `tx_type < TX_TYPES' failed. BUG=aomedia:1171 Change-Id: I3f1c91d751e937dd8e07f7e2f786a75eba4b6891
-
Sarah Parker authored
Currently, optimize_txb does not use the new-quant lookup table when it updates the coefficients. This fixes the function to avoid a mismatch. With this patch, new-quant adds 0.28% gains on top of lv-map-multi on 40 frames of lowres. Change-Id: Ic4e76c28c7e4e007d277fe92b9030e078768bab3
-
Zoe Liu authored
Change-Id: If84cd6da3f2ad9d989baac713e7bd1c50470acf1
-
Sebastien Alaiwan authored
Change-Id: I3faecc70004747a10cd51df9dbcb27b7c2d56703
-
Linfeng Zhang authored
10x - 50x faster than C code. av1_cost_coeffs_txb() is about 6% faster. av1_cost_coeffs() is about 3% faster. Change-Id: Ib9cbed02a65b9cb0c5deb7a5d99c95d0d8ba32c0
-
Linfeng Zhang authored
Change-Id: I9228a10fb0c50010a64aeb2761771522f378f33e
-
Linfeng Zhang authored
Tested with 720p encoding and av1_cost_coeffs() saves 12% time. Tried inline av1_cost_coeffs_txb() first but has no help. Change-Id: Idb59de8563010be6f39e9c803ee2d5137c5160f7
-
Debargha Mukherjee authored
Removes existing inconsistencies between chroma tx_sizes used for chroma loopfiltering. Includes various refactoring to remove the uv_txsize_lookup array eventually. BUG=aomedia:1090 Change-Id: Ib74299b41280ca3ebeaf9a9293242d531d68ad28
-
Debargha Mukherjee authored
This is essentially an implementation of Mozilla's big_chorma_tx proposal, and CFL is already using this. The option is turned on by default. Also includes some associated refactoring. AWCY Subset1 results: PSNR | PSNR Cb | PSNR Cr | PSNR HVS | SSIM | MS SSIM | CIEDE 2000 -0.0136 | -1.0317 | -1.3525 | -0.0140 | -0.0188 | -0.0156 | -0.4665 Link: https://beta.arewecompressedyet.com/?job=debargha-base-lvmap%402017-12-21T06%3A08%3A35.079Z&job=debargha-nosmltxi-lvmap%402017-12-21T06%3A10%3A57.767Z Also resolves the bug below: BUG=aomedia:1158 Change-Id: I9b806b57c008b7a9bb79357f0bc44dbb091e5278
-
Ralph Giles authored
INLINE is defined as __forceinline for vs* configs, but is the normal, compiler-discretion inline for gcc/clang configs. This makes many functions very large when building for windows targets, much larger than they are elsewhere. Use '__inline' as a consistent definition to get consistent function sizes. Thanks to David Major for the diagnosis. Change-Id: Ibb4f7e3a58d65b844d65cbd2e487d9f9f4cedc91
-
Yunqing Wang authored
Set function pointers for AV1 convolve functions in JNT_COMP. Change-Id: I9042b09c7c0222660b18b3a9ebb1379fd05b52c8
-
Yue Chen authored
Bug=aomedia:1170 Change-Id: I26a816c3d379ce0428967f2631c5c9282bf8f278
-
Jingning Han authored
Removing the drl dependency on the candidate list length appears to incur more than 0.3% compression performance loss. Hence remove this option from opt-ref-mv to allow better latency vs compression performance trade off. Change-Id: I6edaeb2d437996082b7bdd6cda7351426c5584b9
-