Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add support for multithreading in encoder #1248

Merged
merged 2 commits into from
Oct 9, 2020
Merged

Add support for multithreading in encoder #1248

merged 2 commits into from
Oct 9, 2020

Conversation

rouault
Copy link
Collaborator

@rouault rouault commented Apr 30, 2020 •
edited
Loading

Testing with a Sentinel2 10m band, T36JTT_20160914T074612_B02.jp2,
coming from S2A_MSIL1C_20160914T074612_N0204_R135_T36JTT_20160914T081456.SAFE

Decompress it to TIFF:

opj_decompress -i T36JTT_20160914T074612_B02.jp2 -o T36JTT_20160914T074612_B02.tif

Recompress it with similar parameters as original:

$ time bin/opj_compress -threads 0 -n 5 -c [256,256],[256,256],[256,256],[256,256],[256,256] -t 1024,1024 -PLT -i T36JTT_20160914T074612_B02.tif -o T36JTT_20160914T074612_B02_PLT.jp2 >/dev/null
real	0m20.386s
user	0m19.996s
sys	0m0.224s

$ time bin/opj_compress -threads 2 -n 5 -c [256,256],[256,256],[256,256],[256,256],[256,256] -t 1024,1024 -PLT -i T36JTT_20160914T074612_B02.tif -o T36JTT_20160914T074612_B02_PLT.jp2 >/dev/null
real	0m11.405s
user	0m21.772s
sys	0m0.320s

$ time bin/opj_compress -threads 4 -n 5 -c [256,256],[256,256],[256,256],[256,256],[256,256] -t 1024,1024 -PLT -i T36JTT_20160914T074612_B02.tif -o T36JTT_20160914T074612_B02_PLT.jp2 >/dev/null
real	0m6.339s
user	0m22.468s
sys	0m0.304s

$ time bin/opj_compress -threads 6 -n 5 -c [256,256],[256,256],[256,256],[256,256],[256,256] -t 1024,1024 -PLT -i T36JTT_20160914T074612_B02.tif -o T36JTT_20160914T074612_B02_PLT.jp2 >/dev/null
real	0m5.417s
user	0m27.548s
sys	0m0.332s

$ time bin/opj_compress -threads 8 -n 5 -c [256,256],[256,256],[256,256],[256,256],[256,256] -t 1024,1024 -PLT -i T36JTT_20160914T074612_B02.tif -o T36JTT_20160914T074612_B02_PLT.jp2 >/dev/null
real	0m4.800s
user	0m31.068s
sys	0m0.316s

Those tests are on a Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz (4 cores, hyper threaded). We get close to linear scaling until threads = 4 (x 1.78 speed-up for threads = 2, and x 3.21 speed-up for threads = 4) , and then more modest improvements due to hyper threading kicking.

Funded by ESA for S2-MPC project

@rouault
Copy link
Collaborator Author

rouault commented Apr 30, 2020

CC @savmickael This is ready for your review

@rouault
Add multithreading support in the T1 (entropy phase) encoder
97eb7e0 
- API wise, opj_codec_set_threads() can be used on the encoding side
- opj_compress has a -threads switch similar to opj_uncompress
@rouault
Add multithreaded support in the DWT encoder.
07d1f77 
Update the bench_dwt utility to have a -decode/-encode switch

Measured performance gains for DWT encoder on a
Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz (4 cores, hyper threaded)

Encoding time:
$ ./bin/bench_dwt -encode -num_threads 1
time for dwt_encode: total = 8.348 s, wallclock = 8.352 s

$ ./bin/bench_dwt -encode -num_threads 2
time for dwt_encode: total = 9.776 s, wallclock = 4.904 s

$ ./bin/bench_dwt -encode -num_threads 4
time for dwt_encode: total = 13.188 s, wallclock = 3.310 s

$ ./bin/bench_dwt -encode -num_threads 8
time for dwt_encode: total = 30.024 s, wallclock = 4.064 s

Scaling is probably limited by memory access patterns causing
memory access to be the bottleneck.
The slightly worse results with threads==8 than with thread==4
is due to hyperthreading being not appropriate here.
@rouault rouault merged commit 07d1f77 into uclouvain:master Oct 9, 2020
mtremer pushed a commit to ipfire/ipfire-2.x that referenced this pull request Apr 29, 2022
@pmu-ipf
openjpeg: Update to version 2.4.0
ca98d29 
- Update from version 2.3.1 to 2.4.0
- Update of rootfile
- Changelog
    2.4.0
	**Closed issues:**
		- OPENJPEG\_INSTALL\_DOC\_DIR does not control a destination directory where HTML docs would be installed. [\#1309](uclouvain/openjpeg#1309)
		- Heap-buffer-overflow in lib/openjp2/pi.c:312 [\#1302](uclouvain/openjpeg#1302)
		- Heap-buffer-overflow in lib/openjp2/t2.c:973 [\#1299](uclouvain/openjpeg#1299)
		- Heap-buffer-overflow in lib/openjp2/pi.c:623 [\#1293](uclouvain/openjpeg#1293)
		- Global-buffer-overflow in lib/openjp2/dwt.c:1980 [\#1286](uclouvain/openjpeg#1286)
		- Heap-buffer-overflow in lib/openjp2/tcd.c:2417 [\#1284](uclouvain/openjpeg#1284)
		- Heap-buffer-overflow in lib/openjp2/mqc.c:499 [\#1283](uclouvain/openjpeg#1283)
		- Openjpeg could not encode 32bit RGB float image [\#1281](uclouvain/openjpeg#1281)
		- Openjpeg could not encode 32bit RGB float image [\#1280](uclouvain/openjpeg#1280)
		- ISO/IEC 15444-1:2019 \(E\) compared with 'cio.h' [\#1277](uclouvain/openjpeg#1277)
		- Test-suite failure due to hash mismatch [\#1264](uclouvain/openjpeg#1264)
		- Heap use-after-free [\#1261](uclouvain/openjpeg#1261)
		- Memory leak when failing to allocate object... [\#1259](uclouvain/openjpeg#1259)
		- Memory leak of Tier 1 handle when OpenJPEG fails to set it as TLS... [\#1257](uclouvain/openjpeg#1257)
		- Any plan to build release for CVE-2020-8112/CVE-2020-6851 [\#1247](uclouvain/openjpeg#1247)
		- failing to convert 16-bit file: opj\_t2\_encode\_packet\(\): only 5251 bytes remaining in output buffer. 5621 needed. [\#1243](uclouvain/openjpeg#1243)
		- CMake+VS2017 Compile OK, thirdparty Compile OK, but thirdparty not install [\#1239](uclouvain/openjpeg#1239)
		- New release to solve CVE-2019-6988 ? [\#1238](uclouvain/openjpeg#1238)
		- Many tests fail to pass after the update of libtiff to version 4.1.0 [\#1233](uclouvain/openjpeg#1233)
		- Another heap buffer overflow in libopenjp2 [\#1231](uclouvain/openjpeg#1231)
		- Heap buffer overflow in libopenjp2 [\#1228](uclouvain/openjpeg#1228)
		- Endianness of binary volume \(JP3D\) [\#1224](uclouvain/openjpeg#1224)
		- New release to resolve CVE-2019-12973 [\#1222](uclouvain/openjpeg#1222)
		- how to set the block size,like 128,256 ? [\#1216](uclouvain/openjpeg#1216)
		- compress YUV files to motion jpeg2000 standard [\#1213](uclouvain/openjpeg#1213)
		- Repair/update Java wrapper, and include in release [\#1208](uclouvain/openjpeg#1208)
		- abc [\#1206](uclouvain/openjpeg#1206)
		- Slow decoding [\#1202](uclouvain/openjpeg#1202)
		- Installation question [\#1201](uclouvain/openjpeg#1201)
		- Typo in test\_decode\_area - \*ptilew is assigned instead of \*ptileh [\#1195](uclouvain/openjpeg#1195)
		- Creating a J2K file with one POC is broken [\#1191](uclouvain/openjpeg#1191)
		- Make fails on Arch Linux [\#1174](uclouvain/openjpeg#1174)
		- Heap buffer overflow in opj\_t1\_clbl\_decode\_processor\(\) triggered with Ghostscript [\#1158](uclouvain/openjpeg#1158)
		- opj\_stream\_get\_number\_byte\_left: Assertion `p\_stream-\>m\_byte\_offset \>= 0' failed. [\#1151](uclouvain/openjpeg#1151)
		- The fuzzer ignores too many inputs [\#1079](uclouvain/openjpeg#1079)
		- out of bounds read [\#1068](uclouvain/openjpeg#1068)
	**Merged pull requests:**
		- Change defined WIN32 [\#1310](uclouvain/openjpeg#1310) ([Jamaika1](https://github.com/Jamaika1))
		- docs: fix simple typo, producted -\> produced [\#1308](uclouvain/openjpeg#1308) ([timgates42](https://github.com/timgates42))
		- Set ${OPENJPEG\_INSTALL\_DOC\_DIR} to DESTINATION of HTMLs [\#1307](uclouvain/openjpeg#1307) ([lemniscati](https://github.com/lemniscati))
		- Use INC\_DIR for OPENJPEG\_INCLUDE\_DIRS \(fixes uclouvain\#1174\) [\#1306](uclouvain/openjpeg#1306) ([matthew-sharp](https://github.com/matthew-sharp))
		- pi.c: avoid out of bounds access with POC \(fixes \#1302\) [\#1304](uclouvain/openjpeg#1304) ([rouault](https://github.com/rouault))
		- Encoder: grow again buffer size [\#1303](uclouvain/openjpeg#1303) ([zodf0055980](https://github.com/zodf0055980))
		- opj\_j2k\_write\_sod\(\): avoid potential heap buffer overflow \(fixes \#1299\) \(probably master only\) [\#1301](uclouvain/openjpeg#1301) ([rouault](https://github.com/rouault))
		- pi.c: avoid out of bounds access with POC \(refs https://github.com/uclouvain/openjpeg/issues/1293\#issuecomment-737122836\) [\#1300](uclouvain/openjpeg#1300) ([rouault](https://github.com/rouault))
		- opj\_t2\_encode\_packet\(\): avoid out of bound access of \#1297, but likely not the proper fix [\#1298](uclouvain/openjpeg#1298) ([rouault](https://github.com/rouault))
		- opj\_t2\_encode\_packet\(\): avoid out of bound access of \#1294, but likely not the proper fix [\#1296](uclouvain/openjpeg#1296) ([rouault](https://github.com/rouault))
		- opj\_j2k\_setup\_encoder\(\): validate POC compno0 and compno1 \(fixes \#1293\) [\#1295](uclouvain/openjpeg#1295) ([rouault](https://github.com/rouault))
		- Encoder: avoid global buffer overflow on irreversible conversion when… [\#1292](uclouvain/openjpeg#1292) ([rouault](https://github.com/rouault))
		- Decoding: deal with some SPOT6 images that have tiles with a single tile-part with TPsot == 0 and TNsot == 0, and with missing EOC [\#1291](uclouvain/openjpeg#1291) ([rouault](https://github.com/rouault))
		- Free p\_tcd\_marker\_info to avoid memory leak [\#1288](uclouvain/openjpeg#1288) ([zodf0055980](https://github.com/zodf0055980))
		- Encoder: grow again buffer size [\#1287](uclouvain/openjpeg#1287) ([zodf0055980](https://github.com/zodf0055980))
		- Encoder: avoid uint32 overflow when allocating memory for codestream buffer \(fixes \#1243\) [\#1276](uclouvain/openjpeg#1276) ([rouault](https://github.com/rouault))
		- Java compatibility from 1.5 to 1.6 [\#1263](uclouvain/openjpeg#1263) ([jiapei100](https://github.com/jiapei100))
		- opj\_decompress: fix double-free on input directory with mix of valid and invalid images [\#1262](uclouvain/openjpeg#1262) ([rouault](https://github.com/rouault))
		- openjp2: Plug image leak when failing to allocate codestream index. [\#1260](uclouvain/openjpeg#1260) ([sebras](https://github.com/sebras))
		- openjp2: Plug memory leak when setting data as TLS fails. [\#1258](uclouvain/openjpeg#1258) ([sebras](https://github.com/sebras))
		- openjp2: Error out if failing to create Tier 1 handle. [\#1256](uclouvain/openjpeg#1256) ([sebras](https://github.com/sebras))
		- Testing for invalid values of width, height, numcomps [\#1254](uclouvain/openjpeg#1254) ([szukw000](https://github.com/szukw000))
		- Single-threaded performance improvements in forward DWT for 5-3 and 9-7 \(and other improvements\) [\#1253](uclouvain/openjpeg#1253) ([rouault](https://github.com/rouault))
		- Add support for multithreading in encoder [\#1248](uclouvain/openjpeg#1248) ([rouault](https://github.com/rouault))
		- Add support for generation of PLT markers in encoder [\#1246](uclouvain/openjpeg#1246) ([rouault](https://github.com/rouault))
		- Fix warnings about signed/unsigned casts in pi.c [\#1244](uclouvain/openjpeg#1244) ([rouault](https://github.com/rouault))
		- opj\_decompress: add sanity checks to avoid segfault in case of decoding error [\#1240](uclouvain/openjpeg#1240) ([rouault](https://github.com/rouault))
		- ignore wrong icc [\#1236](uclouvain/openjpeg#1236) ([szukw000](https://github.com/szukw000))
		- Implement writing of IMF profiles [\#1235](uclouvain/openjpeg#1235) ([rouault](https://github.com/rouault))
		- tests: add alternate checksums for libtiff 4.1 [\#1234](uclouvain/openjpeg#1234) ([rouault](https://github.com/rouault))
		- opj\_tcd\_init\_tile\(\): avoid integer overflow [\#1232](uclouvain/openjpeg#1232) ([rouault](https://github.com/rouault))
		- tests/fuzzers: link fuzz binaries using $LIB\_FUZZING\_ENGINE. [\#1230](uclouvain/openjpeg#1230) ([Dor1s](https://github.com/Dor1s))
		- opj\_j2k\_update\_image\_dimensions\(\): reject images whose coordinates are beyond INT\_MAX \(fixes \#1228\) [\#1229](uclouvain/openjpeg#1229) ([rouault](https://github.com/rouault))
		- Fix resource leaks [\#1226](uclouvain/openjpeg#1226) ([dodys](https://github.com/dodys))
		- abi-check.sh: fix false postive ABI error, and display output error log [\#1218](uclouvain/openjpeg#1218) ([rouault](https://github.com/rouault))
		- pi.c: avoid integer overflow, resulting in later invalid access to memory in opj\_t2\_decode\_packets\(\) [\#1217](uclouvain/openjpeg#1217) ([rouault](https://github.com/rouault))
		- Add check to validate SGcod/SPcoc/SPcod parameter values. [\#1211](uclouvain/openjpeg#1211) ([sebras](https://github.com/sebras))
		- Fix buffer overflow reading an image file less than four characters [\#1196](uclouvain/openjpeg#1196) ([robert-ancell](https://github.com/robert-ancell))
		- compression: emit POC marker when only one single POC is requested \(f… [\#1192](uclouvain/openjpeg#1192) ([rouault](https://github.com/rouault))
		- Fix several potential vulnerabilities  [\#1185](uclouvain/openjpeg#1185) ([Young-X](https://github.com/Young-X))
		- openjp2/j2k: Report error if all wanted components are not decoded. [\#1164](uclouvain/openjpeg#1164) ([sebras](https://github.com/sebras))

Signed-off-by: Adolf Belka <adolf.belka@ipfire.org>
Reviewed-by: Peter Müller <peter.mueller@ipfire.org>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@rouault