Cycle times:
inv_txfm_add_4x4_adst_adst_0_8bpc_c: 445.9
inv_txfm_add_4x4_adst_adst_0_8bpc_ssse3: 23.7
inv_txfm_add_4x4_adst_adst_1_8bpc_c: 443.7
inv_txfm_add_4x4_adst_adst_1_8bpc_ssse3: 52.6
inv_txfm_add_4x4_adst_dct_0_8bpc_c: 474.5
inv_txfm_add_4x4_adst_dct_0_8bpc_ssse3: 23.9
inv_txfm_add_4x4_adst_dct_1_8bpc_c: 482.0
inv_txfm_add_4x4_adst_dct_1_8bpc_ssse3: 51.1
inv_txfm_add_4x4_adst_flipadst_0_8bpc_c: 587.2
inv_txfm_add_4x4_adst_flipadst_0_8bpc_ssse3: 24.0
inv_txfm_add_4x4_adst_flipadst_1_8bpc_c: 457.2
inv_txfm_add_4x4_adst_flipadst_1_8bpc_ssse3: 52.8
inv_txfm_add_4x4_adst_identity_0_8bpc_c: 412.4
inv_txfm_add_4x4_adst_identity_0_8bpc_ssse3: 43.3
inv_txfm_add_4x4_adst_identity_1_8bpc_c: 412.0
inv_txfm_add_4x4_adst_identity_1_8bpc_ssse3: 43.3
inv_txfm_add_4x4_dct_adst_0_8bpc_c: 467.4
inv_txfm_add_4x4_dct_adst_0_8bpc_ssse3: 23.2
inv_txfm_add_4x4_dct_adst_1_8bpc_c: 588.3
inv_txfm_add_4x4_dct_adst_1_8bpc_ssse3: 48.6
inv_txfm_add_4x4_dct_dct_0_8bpc_c: 611.5
inv_txfm_add_4x4_dct_dct_0_8bpc_ssse3: 23.1
inv_txfm_add_4x4_dct_dct_1_8bpc_c: 576.2
inv_txfm_add_4x4_dct_dct_1_8bpc_ssse3: 47.6
inv_txfm_add_4x4_dct_flipadst_0_8bpc_c: 479.5
inv_txfm_add_4x4_dct_flipadst_0_8bpc_ssse3: 23.4
inv_txfm_add_4x4_dct_flipadst_1_8bpc_c: 549.3
inv_txfm_add_4x4_dct_flipadst_1_8bpc_ssse3: 48.3
inv_txfm_add_4x4_dct_identity_0_8bpc_c: 576.9
inv_txfm_add_4x4_dct_identity_0_8bpc_ssse3: 25.4
inv_txfm_add_4x4_dct_identity_1_8bpc_c: 610.7
inv_txfm_add_4x4_dct_identity_1_8bpc_ssse3: 25.1
inv_txfm_add_4x4_flipadst_adst_0_8bpc_c: 532.8
inv_txfm_add_4x4_flipadst_adst_0_8bpc_ssse3: 23.8
inv_txfm_add_4x4_flipadst_adst_1_8bpc_c: 666.7
inv_txfm_add_4x4_flipadst_adst_1_8bpc_ssse3: 61.0
inv_txfm_add_4x4_flipadst_dct_0_8bpc_c: 539.6
inv_txfm_add_4x4_flipadst_dct_0_8bpc_ssse3: 23.8
inv_txfm_add_4x4_flipadst_dct_1_8bpc_c: 484.6
inv_txfm_add_4x4_flipadst_dct_1_8bpc_ssse3: 51.1
inv_txfm_add_4x4_flipadst_flipadst_0_8bpc_c: 503.1
inv_txfm_add_4x4_flipadst_flipadst_0_8bpc_ssse3: 23.9
inv_txfm_add_4x4_flipadst_flipadst_1_8bpc_c: 463.0
inv_txfm_add_4x4_flipadst_flipadst_1_8bpc_ssse3: 54.0
inv_txfm_add_4x4_flipadst_identity_0_8bpc_c: 719.9
inv_txfm_add_4x4_flipadst_identity_0_8bpc_ssse3: 43.0
inv_txfm_add_4x4_flipadst_identity_1_8bpc_c: 456.8
inv_txfm_add_4x4_flipadst_identity_1_8bpc_ssse3: 44.1
inv_txfm_add_4x4_identity_adst_0_8bpc_c: 422.8
inv_txfm_add_4x4_identity_adst_0_8bpc_ssse3: 42.4
inv_txfm_add_4x4_identity_adst_1_8bpc_c: 417.1
inv_txfm_add_4x4_identity_adst_1_8bpc_ssse3: 42.3
inv_txfm_add_4x4_identity_dct_0_8bpc_c: 435.4
inv_txfm_add_4x4_identity_dct_0_8bpc_ssse3: 25.7
inv_txfm_add_4x4_identity_dct_1_8bpc_c: 434.1
inv_txfm_add_4x4_identity_dct_1_8bpc_ssse3: 25.3
inv_txfm_add_4x4_identity_flipadst_0_8bpc_c: 528.1
inv_txfm_add_4x4_identity_flipadst_0_8bpc_ssse3: 40.9
inv_txfm_add_4x4_identity_flipadst_1_8bpc_c: 720.0
inv_txfm_add_4x4_identity_flipadst_1_8bpc_ssse3: 41.8
inv_txfm_add_4x4_identity_identity_0_8bpc_c: 383.2
inv_txfm_add_4x4_identity_identity_0_8bpc_ssse3: 28.3
inv_txfm_add_4x4_identity_identity_1_8bpc_c: 378.9
inv_txfm_add_4x4_identity_identity_1_8bpc_ssse3: 28.2
inv_txfm_add_4x4_wht_wht_0_8bpc_c: 271.5
inv_txfm_add_4x4_wht_wht_0_8bpc_ssse3: 34.0
inv_txfm_add_4x4_wht_wht_1_8bpc_c: 266.0
inv_txfm_add_4x4_wht_wht_1_8bpc_ssse3: 33.9

Cycle times:
intra_pred_h_w4_8bpc_c: 146.6
intra_pred_h_w4_8bpc_ssse3: 30.6
intra_pred_h_w8_8bpc_c: 236.3
intra_pred_h_w8_8bpc_ssse3: 42.2
intra_pred_h_w16_8bpc_c: 446.6
intra_pred_h_w16_8bpc_ssse3: 55.8
intra_pred_h_w32_8bpc_c: 688.2
intra_pred_h_w32_8bpc_ssse3: 85.9
intra_pred_h_w64_8bpc_c: 634.2
intra_pred_h_w64_8bpc_ssse3: 169.2

Coef was originally chosen to accomodate 12 bits/component with
4 extra precision intermediates + some under/overflow range, but
it turns out that 12 bits/component only uses 2 extra precision
intermediates, so we don't need coef.

Fixes #213.

Fixes #211, dav1d_get_picture starts draining delayed pictures when it
gets called multiple times without dav1d_send_data call in between.

c->out can hold a valid picture on bitstream errors after 4903d87b
when frame multi-threading is used. Output this picture instead of
returning the return value of dav1d_parse_obus().

Fixes `assert(dst->data[0] == ((void*)0));` in dav1d_picture_ref with
clusterfuzz-testcase-minimized-dav1d_fuzzer_mt-5743306491822080. Also
fixes a memory leak of frames with
clusterfuzz-testcase-minimized-dav1d_fuzzer_mt-5655593017147392. Credits
to oss-fuzz.

Cycle times:
inv_txfm_add_4x4_dct_dct_0_8bpc_c: 492.6
inv_txfm_add_4x4_dct_dct_0_8bpc_ssse3: 22.6
inv_txfm_add_4x4_dct_dct_1_8bpc_c: 494.2
inv_txfm_add_4x4_dct_dct_1_8bpc_ssse3: 48.3

Fixes #210.

Luma output plane is used during chroma film grain. Fixes an use of
uninitialized value in iclip/apply_to_row_uv with
clusterfuzz-testcase-minimized-dav1d_fuzzer_mt-5636143299690496. Credits
to oss-fuzz.

Also use aligned reads and writes in sub_loop, and integrate sum_loop into
the main loop.

before:
cfl_ac_420_w4_8bpc_c: 367.4
cfl_ac_420_w4_8bpc_avx2: 72.8
cfl_ac_420_w8_8bpc_c: 621.6
cfl_ac_420_w8_8bpc_avx2: 85.1
cfl_ac_420_w16_8bpc_c: 983.4
cfl_ac_420_w16_8bpc_avx2: 141.0

after:
cfl_ac_420_w4_8bpc_c: 376.2
cfl_ac_420_w4_8bpc_avx2: 28.5
cfl_ac_420_w8_8bpc_c: 607.2
cfl_ac_420_w8_8bpc_avx2: 29.9
cfl_ac_420_w16_8bpc_c: 962.1
cfl_ac_420_w16_8bpc_avx2: 48.8

Fixes use of uninitialized vaue in apply_to_row_y() with
clusterfuzz-testcase-minimized-dav1d_fuzzer_mt-5753369222709248. Credits
to oss-fuzz.

This fixes a potential overflow when setting num_ticks_per_picture if
dav1d_get_vlc() returns (1 << 32) - 1.

Missed in the previous commits.

Fixes #200.

ftello isn't available on MSVC.

On MinGW, both ftello and _ftelli64 are available. MinGW provides
the _ftelli64 even on msvcrt versions that don't actually have
such a function by providing a statically linked wrapper. But
since the ftello function also is available (also provided as
a similar wrapper), keep using that.

off_t isn't available in all environments.

On Ubuntu 16.04, GCC 5.4, glibc 2.23, off_t is defined by stdio.h
normally, but if building with -std=c99 (which dav1d does), it no
longer is defined. If fcntl.h is included additionally (which isn't
done in dav1d), it still gets defined though.

off_t isn't available in MSVC at all.

Fixes an assert in dav1d_prepare_intra_edges_16bpc() with
clusterfuzz-testcase-minimized-dav1d_fuzzer_mt-5640098964570112. Credits
to oss-fuzz.

Adaption of the avx2 code