Add zlib-ng as an alternative zlib implementation by FantasqueX · Pull Request #24782 · opencv/opencv

FantasqueX · 2023-12-27T15:28:58Z

Zlib-ng is zlib replacement with optimizations for "next generation" systems. Its optimization may benifits image library decode and encode speed such as libpng. In our tests, if using zlib-ng and libpng combination on a x86_64 machine with AVX2, the time of imdecode amd imencode will drop 20% approximately. This patch enables zlib-ng's optimization if CV_DISABLE_OPTIMIZATION is OFF. Since Zlib-ng can dispatch intrinsics on the fly, port work is much easier.

Related discussion: #22573

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

I agree to contribute to the project under Apache 2 License.
To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
The PR is proposed to the proper branch
There is a reference to the original bug report and related work
There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
The feature is well documented and sample code can be built with the project CMake

在只开启 AVX2 intrinsics 时，png 格式 `imdecodemulti` 和 `imencodemulti` 耗时会减少约 20%。没有启用 name mangling，因为 openexr 会编译不过。经验证，在 Windows 和 Linux 下，zlib 的符号不会导出。上游 PR：opencv#24782

Zlib-ng is zlib replacement with optimizations for "next generation" systems. Its optimization may benifits image library decode and encode speed such as libpng. In our tests, if using zlib-ng and libpng combination on a x86_64 machine with AVX2, the time of `imdecode` amd `imencode` will drop 20% approximately. This patch enables zlib-ng's optimization if `CV_DISABLE_OPTIMIZATION` is OFF. Since Zlib-ng can dispatch intrinsics on the fly, port work is much easier. Related discussion: opencv#22573

FantasqueX · 2024-01-14T07:00:59Z

Hi, I've updated zlib-ng to 2.1.6. Could you please review this patch? @asmorkalov @opencv-alalek

FantasqueX · 2024-01-20T10:54:45Z

Gently ping~

asmorkalov · 2024-01-23T09:35:42Z

I apologize for the delay. Zlib is used py OpenCV itself, libpng, QT and other our 3rdparties. I'm looking on different options for OpenCV-Python package to re-use the same implementation and exclude conflicts.

asmorkalov · 2024-01-23T11:44:54Z

ARM v7 build is broken:

[ 11%] Building C object 3rdparty/zlib-ng/CMakeFiles/zlib.dir/arch/arm/slide_hash_armv6.c.o
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c: In function ‘slide_hash_chain’:
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:13:16: error: unknown type name ‘uint16x2_t’
     Z_REGISTER uint16x2_t v;
                ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:14:5: error: unknown type name ‘uint16x2_t’
     uint16x2_t p0, p1, p2, p3;
     ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:23:24: error: ‘uint16x2_t’ undeclared (first use in this function)
     n = size / (sizeof(uint16x2_t) * 4);
                        ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:23:24: note: each undeclared identifier is reported only once for each function it appears in
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:25:23: error: expected ‘)’ before ‘uint16x2_t’
         p0 = *((const uint16x2_t *)(table));
                       ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:25:9: warning: type defaults to ‘int’ in type name [-Wimplicit-int]
         p0 = *((const uint16x2_t *)(table));
         ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:25:14: error: invalid type argument of unary ‘*’ (have ‘int’)
         p0 = *((const uint16x2_t *)(table));
              ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:26:23: error: expected ‘)’ before ‘uint16x2_t’
         p1 = *((const uint16x2_t *)(table+2));
                       ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:26:9: warning: type defaults to ‘int’ in type name [-Wimplicit-int]
         p1 = *((const uint16x2_t *)(table+2));
         ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:26:14: error: invalid type argument of unary ‘*’ (have ‘int’)
         p1 = *((const uint16x2_t *)(table+2));
              ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:27:23: error: expected ‘)’ before ‘uint16x2_t’
         p2 = *((const uint16x2_t *)(table+4));
                       ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:27:9: warning: type defaults to ‘int’ in type name [-Wimplicit-int]
         p2 = *((const uint16x2_t *)(table+4));
         ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:27:14: error: invalid type argument of unary ‘*’ (have ‘int’)
         p2 = *((const uint16x2_t *)(table+4));
              ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:28:23: error: expected ‘)’ before ‘uint16x2_t’
         p3 = *((const uint16x2_t *)(table+6));
                       ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:28:9: warning: type defaults to ‘int’ in type name [-Wimplicit-int]
         p3 = *((const uint16x2_t *)(table+6));
         ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:28:14: error: invalid type argument of unary ‘*’ (have ‘int’)
         p3 = *((const uint16x2_t *)(table+6));
              ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:29:14: warning: implicit declaration of function ‘__uqsub16’ [-Wimplicit-function-declaration]
         p0 = __uqsub16(p0, v);
              ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:33:24: error: expected expression before ‘)’ token
         *((uint16x2_t *)(table)) = p0;
                        ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:34:24: error: expected expression before ‘)’ token
         *((uint16x2_t *)(table+2)) = p1;
                        ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:35:24: error: expected expression before ‘)’ token
         *((uint16x2_t *)(table+4)) = p2;
                        ^
/home/ubuntu/Projects/opencv/3rdparty/zlib-ng/arch/arm/slide_hash_armv6.c:36:24: error: expected expression before ‘)’ token
         *((uint16x2_t *)(table+6)) = p3;
                        ^
3rdparty/zlib-ng/CMakeFiles/zlib.dir/build.make:507: recipe for target '3rdparty/zlib-ng/CMakeFiles/zlib.dir/arch/arm/slide_hash_armv6.c.o' failed

OS: ubuntu 16.04, GCC 5.4.0, CMake: cmake -DCPU_BASELINE=NEON -DWITH_ZLIB_NG=ON -DBUILD_PNG=ON ../opencv

asmorkalov · 2024-01-23T11:56:40Z

Performance numbers for libpng on Intel i5-1135G7, Ubuntu 20.04:

Geometric mean (ms)

Name of Test 4.x-1  patched-1 patched-1 
                                  vs    
                                4.x-1   
                              (x-factor)
Decode::JPEG 43.926  44.043      1.00   
Encode::JPEG 38.879  38.880      1.00   
decode::PNG  23.325  12.710      1.84   
encode::PNG  49.257  30.369      1.62

asmorkalov · 2024-01-23T12:16:41Z

The same configuration with Spng:

Geometric mean (ms)

Name of Test 4.x-1  patched-1 patched-1 
                                  vs    
                                4.x-1   
                              (x-factor)
Decode::JPEG 43.528  43.789      0.99   
Encode::JPEG 39.319  38.999      1.01   
decode::PNG  19.943  13.462      1.48   
encode::PNG  64.591  46.950      1.38

FantasqueX · 2024-01-23T13:31:57Z

Thanks for testing. For Armv7 build failure, opencv's zlib-ng build option is -DZLIB_COMPAT=ON -DZLIB_ENABLE_TESTS=OFF -DWITH_GTEST=OFF . We should test whther zlib-ng 2.1.6 can be built with those options on Armv7. Since I dont't have an Armv7 environment, it'll take some time to investigate the build failure. I'll have a try with qemu-system.

asmorkalov · 2024-01-31T17:46:16Z

I made experiments with arm and found out that root CmakeLists.txt in zlib was migrated incorrectly. I'll return back to the PR soon.

FantasqueX · 2024-02-01T16:05:14Z

Sorry for late response. I encountered some difficulties using qemu-system. So, I tested my patch on raspberry pi 2b which is the latest armv7 device AFAIK. I cannot reproduce the build failure.

OS: raspberry pi OS based on debian 11, GCC 10.2.1-6+rpi1, CMake: cmake -GNinja -DCPU_BASELINE=NEON -DWITH_ZLIB_NG=ON -DBUILD_PNG=ON ..

FantasqueX · 2024-02-28T08:01:20Z

@asmorkalov Gently ping. It's been a month since I heard from you last time. Is there anything I can do to make this PR be merged sooner?

asmorkalov · 2024-03-04T14:46:38Z

I stuck a bit with the NEON issue in some configurations. I'll try to fix it and propose a solution this week. The PR is definitely useful and will be merged.

asmorkalov · 2024-03-28T13:04:52Z

The compilation issue is compiler dependent. I see the same errors with GCC 5.4 in Ubuntu 16.04, GCC 7.5 and 8.4, but not with GCC 10.2 on Raspberry Pi.

asmorkalov · 2024-05-03T07:55:16Z

Discussed the PR on the OpenCV Core team meeting. Decided to merge the PR as is. WITH_ZLIB_NG option is disabled by default. NEON issue with old compilers will be fixed later with another patch.

FantasqueX force-pushed the 4.x-zlib-ng branch from 279a3dc to e55a8cd Compare December 27, 2023 19:06

FantasqueX mentioned this pull request Dec 28, 2023

添加 zlib-ng 作为 zlib 的替代实现 aqrose-aidi-sdk-team/opencv#8

Merged

asmorkalov added the category: 3rdparty label Jan 9, 2024

FantasqueX force-pushed the 4.x-zlib-ng branch from e55a8cd to 0de26fd Compare January 14, 2024 06:59

asmorkalov self-requested a review January 23, 2024 09:35

asmorkalov added the optimization label Jan 23, 2024

asmorkalov added this to the 4.10.0 milestone Jan 23, 2024

asmorkalov approved these changes May 3, 2024

View reviewed changes

asmorkalov merged commit 6bf758e into opencv:4.x May 3, 2024

mshabunin mentioned this pull request Jun 14, 2024

Merge 4.x -> 5.x #25745

Merged

FantasqueX deleted the 4.x-zlib-ng branch June 28, 2024 13:09

Uh oh!

Conversation

FantasqueX commented Dec 27, 2023

Pull Request Readiness Checklist

Uh oh!

FantasqueX commented Jan 14, 2024

Uh oh!

FantasqueX commented Jan 20, 2024

Uh oh!

asmorkalov commented Jan 23, 2024

Uh oh!

asmorkalov commented Jan 23, 2024

Uh oh!

asmorkalov commented Jan 23, 2024

Uh oh!

asmorkalov commented Jan 23, 2024

Uh oh!

FantasqueX commented Jan 23, 2024

Uh oh!

asmorkalov commented Jan 31, 2024

Uh oh!

FantasqueX commented Feb 1, 2024

Uh oh!

FantasqueX commented Feb 28, 2024

Uh oh!

asmorkalov commented Mar 4, 2024

Uh oh!

asmorkalov commented Mar 28, 2024

Uh oh!

asmorkalov commented May 3, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants