Valgrind error with tile threads

[GoogleCodeExporter]

What version are you using? On what operating system?
libvpx v1.3.0-5591-g6cf7b3b Linux 64

Please provide any additional information below.

==32349== Memcheck, a memory error detector
==32349== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==32349== Using Valgrind-3.10.0.SVN and LibVEX; rerun with -h for copyright info
==32349== Command: ./vpxenc --codec=vp9 --rt --cpu-used=-5 
--target-bitrate=1000 /home/aconverse/video/jimredvga_25fps.y4m -o /dev/null 
--tile-columns=1 -t 2 --limit=1
==32349== 
Pass 1/1 frame    1/0          0B       0 us 0.00 fpm [ETA  unknown] 
==32349== Source and destination overlap in memcpy(0x5cd2280, 0x5cd2280, 115952)
==32349==    at 0x4C2F71C: memcpy@@GLIBC_2.14 (in 
/usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==32349==    by 0x58BE2E: vp9_encode_tiles_mt (vp9_ethread.c:122)
==32349==    by 0x586926: encode_frame_internal (vp9_encodeframe.c:3937)
==32349==    by 0x4A0FB2: encode_frame_to_data_rate (vp9_encoder.c:2828)
==32349==    by 0x4A3339: vp9_get_compressed_data (vp9_encoder.c:3439)
==32349==    by 0x48B639: encoder_encode (vp9_cx_iface.c:1014)
==32349==    by 0x463248: vpx_codec_encode (vpx_encoder.c:223)
==32349==    by 0x40395E: main (vpxenc.c:1607)
==32349== 
==32349== Source and destination overlap in memcpy(0x5cee770, 0x5cee770, 27744)
==32349==    at 0x4C2F71C: memcpy@@GLIBC_2.14 (in 
/usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==32349==    by 0x58BE48: vp9_encode_tiles_mt (vp9_ethread.c:123)
==32349==    by 0x586926: encode_frame_internal (vp9_encodeframe.c:3937)
==32349==    by 0x4A0FB2: encode_frame_to_data_rate (vp9_encoder.c:2828)
==32349==    by 0x4A3339: vp9_get_compressed_data (vp9_encoder.c:3439)
==32349==    by 0x48B639: encoder_encode (vp9_cx_iface.c:1014)
==32349==    by 0x463248: vpx_codec_encode (vpx_encoder.c:223)
==32349==    by 0x40395E: main (vpxenc.c:1607)
==32349== 
==32349== Source and destination overlap in memcpy(0x5cf84bc, 0x5cf84bc, 13224)
==32349==    at 0x4C2F71C: memcpy@@GLIBC_2.14 (in 
/usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==32349==    by 0x58BE5E: vp9_encode_tiles_mt (vp9_ethread.c:124)
==32349==    by 0x586926: encode_frame_internal (vp9_encodeframe.c:3937)
==32349==    by 0x4A0FB2: encode_frame_to_data_rate (vp9_encoder.c:2828)
==32349==    by 0x4A3339: vp9_get_compressed_data (vp9_encoder.c:3439)
==32349==    by 0x48B639: encoder_encode (vp9_cx_iface.c:1014)
==32349==    by 0x463248: vpx_codec_encode (vpx_encoder.c:223)
==32349==    by 0x40395E: main (vpxenc.c:1607)
==32349== 
Pass 1/1 frame    1/1      42563B  340504b/f 8512600b/s   20674 ms (0.05 fps)
==32349== 
==32349== HEAP SUMMARY:
==32349==     in use at exit: 0 bytes in 0 blocks
==32349==   total heap usage: 36,316 allocs, 36,316 frees, 56,852,981 bytes 
allocated
==32349== 
==32349== All heap blocks were freed -- no leaks are possible

Original issue reported on code.google.com by aconve...@google.com on 5 Mar 2015 at 1:28

[GoogleCodeExporter]

Original comment by aconve...@google.com on 5 Mar 2015 at 7:17

• Changed state: Started

[GoogleCodeExporter]

Note there is a nightly which runs vpxenc + valgrind with vp9/rt [1]. My guess 
is this didn't get amended after multi-threaded encoding was added. Ideally 
that would pick this up.

[1] 
http://build.webmproject.org/jenkins/job/libvpx__example_test-multi-target/arch=
x86_64-linux-valgrind/

Original comment by jz...@google.com on 5 Mar 2015 at 8:38

[GoogleCodeExporter]

The ethread test doesn't actually use tiles, so the current test is valgrind 
clean because it doesn't actually test threading.   

https://gerrit.chromium.org/gerrit/#/c/73984/

Original comment by aconve...@google.com on 5 Mar 2015 at 8:46

[GoogleCodeExporter]

https://gerrit.chromium.org/gerrit/#/c/73983/1

Original comment by aconve...@google.com on 7 Mar 2015 at 8:44

• Changed state: Fixed