Update IPP integration #26463

eplankin · 2024-11-14T16:49:55Z

Please merge together with opencv/opencv_3rdparty#88
Supported IPP version was updated to IPP 2022.0.0 for Linux and Windows. 32-bit binaries are dropped since this release.

Previous update: #25935

opencv-alalek · 2024-11-15T06:43:50Z

Looks like there is regression in norm.
Catched by perf_core test:

[ RUN      ] Size_MatType_NormType_norm2.norm2/58, where GetParam() = (127x61, 8UC1, NORM_L1|NORM_RELATIVE)
 Expected: 
[0.6575438084615143]
 Actual:
[0.6194119943083294]
/home/ubuntu/opencv/modules/ts/src/ts_perf.cpp:580: Failure
Failed
  Relative difference (0.038131814153184895 of 3.945262850769086e-05 allowed) between argument "n" and expected value is greater than 1.0000000000000001e-05 in 1 points
params    = (127x61, 8UC1, NORM_L1|NORM_RELATIVE)
termination reason:  reached maximum number of iterations
bytesIn   =      15494
bytesOut  =          0
samples   =          1
outliers  =          0
frequency = 1000000000
min       =       3339 = 0.00ms
median    =       3339 = 0.00ms
gmean     =       3339 = 0.00ms
gstddev   = 0.00000000 = 0.00ms for 97% dispersion interval
mean      =       3339 = 0.00ms
stddev    =          0 = 0.00ms
[  FAILED  ] Size_MatType_NormType_norm2.norm2/58, where GetParam() = (127x61, 8UC1, NORM_L1|NORM_RELATIVE) (0 ms)

(both Linux / Windows)

Could you please take a look?

eplankin · 2024-11-15T14:05:28Z

Looks like there is regression in norm. Catched by perf_core test:

[ RUN      ] Size_MatType_NormType_norm2.norm2/58, where GetParam() = (127x61, 8UC1, NORM_L1|NORM_RELATIVE)
 Expected: 
[0.6575438084615143]
 Actual:
[0.6194119943083294]
/home/ubuntu/opencv/modules/ts/src/ts_perf.cpp:580: Failure
Failed
  Relative difference (0.038131814153184895 of 3.945262850769086e-05 allowed) between argument "n" and expected value is greater than 1.0000000000000001e-05 in 1 points
params    = (127x61, 8UC1, NORM_L1|NORM_RELATIVE)
termination reason:  reached maximum number of iterations
bytesIn   =      15494
bytesOut  =          0
samples   =          1
outliers  =          0
frequency = 1000000000
min       =       3339 = 0.00ms
median    =       3339 = 0.00ms
gmean     =       3339 = 0.00ms
gstddev   = 0.00000000 = 0.00ms for 97% dispersion interval
mean      =       3339 = 0.00ms
stddev    =          0 = 0.00ms
[  FAILED  ] Size_MatType_NormType_norm2.norm2/58, where GetParam() = (127x61, 8UC1, NORM_L1|NORM_RELATIVE) (0 ms)

(both Linux / Windows)

Could you please take a look?

Yes, I'll take a look.

tizmajlo · 2024-11-15T16:07:55Z

3rdparty/ippicv/ippicv.cmake


  # Commit SHA in the opencv_3rdparty repo
-  set(IPPICV_COMMIT "7f55c0c26be418d494615afca15218566775c725")
+  set(IPPICV_COMMIT "d1cbea44d326eb0421fedcdd16de4630fd8c7ed0")


Shouldn't it be different for x86 and x86_64 as only x86_64 ICV packages are available since IPP 2022.0.0?

asmorkalov · 2024-12-16T15:49:00Z

@eplankin Friendly reminder. We plan OpenCV 4.11 release this month. It'll be great to finish the PR this week.

eplankin · 2024-12-18T12:02:25Z

@eplankin Friendly reminder. We plan OpenCV 4.11 release this month. It'll be great to finish the PR this week.

@asmorkalov It was confirmed that there is a bug in IPP, the fix will be available in the next release. Let's disable the call of IPP function until the next ICV update.

asmorkalov · 2024-12-18T19:20:40Z

@opencv-alalek could you take a look again?

opencv-alalek · 2024-12-23T08:14:23Z

Need to apply this patch to keep 32-bit archives usable as mentioned by @tizmajlo :

diff --git a/3rdparty/ippicv/ippicv.cmake b/3rdparty/ippicv/ippicv.cmake
index 9bf90af11c..7bfcf66993 100644
--- a/3rdparty/ippicv/ippicv.cmake
+++ b/3rdparty/ippicv/ippicv.cmake
@@ -17,6 +17,7 @@ function(download_ippicv root_var)
       set(OPENCV_ICV_NAME "ippicv_2022.0.0_lnx_intel64_20240904_general.tgz")
       set(OPENCV_ICV_HASH "63717ee0f918ad72fb5a737992a206d1")
     else()
+      set(IPPICV_COMMIT "7f55c0c26be418d494615afca15218566775c725")
       set(OPENCV_ICV_NAME "ippicv_2021.12.0_lnx_ia32_20240425_general.tgz")
       set(OPENCV_ICV_HASH "85ffa2b9ed7802b93c23fa27b0097d36")
     endif()
@@ -27,6 +28,7 @@ function(download_ippicv root_var)
       set(OPENCV_ICV_NAME "ippicv_2022.0.0_win_intel64_20240904_general.zip")
       set(OPENCV_ICV_HASH "3a6eca7cc3bce7159eb1443c6fca4e31")
     else()
+      set(IPPICV_COMMIT "7f55c0c26be418d494615afca15218566775c725")
       set(OPENCV_ICV_NAME "ippicv_2021.12.0_win_ia32_20240425_general.zip")
       set(OPENCV_ICV_HASH "8b1d2a23957d57624d0de8f2a5cae5f1")
     endif()

Otherwise it triggers error (remove .cache/ippicv or get clear source repository) due to wrong commit:

-- IPPICV: Downloading ippicv_2021.12.0_win_ia32_20240425_general.zip from https://raw.githubusercontent.com/opencv/opencv_3rdparty/d1cbea44d326eb0421fedcdd16de4630fd8c7ed0/ippicv/ippicv_2021.12.0_win_ia32_20240425_general.zip
-- Try 1 failed
CMake Warning at cmake/OpenCVDownload.cmake:251 (message):
  IPPICV: Download failed: 22;"HTTP response code said error"

opencv-alalek · 2024-12-25T12:30:17Z

@eplankin There are more cases with norm in OpenCL-related tests:

Linux OpenCL:

[ RUN      ] OCL_Arithm/Norm.NORM_INF_2args/4, where GetParam() = (CV_8U, Channels(3), false)
/build/precommit_opencl_linux/4.x/opencv/modules/core/test/ocl/test_arithm.cpp:1320: Failure
relativeError(cpuRes, gpuRes, 1e-6) evaluates to false, where
cpuRes evaluates to 0.98431372549019602
gpuRes evaluates to 0.98818897637795278
1e-6 evaluates to 9.9999999999999995e-07
Google Test trace:
/build/precommit_opencl_linux/4.x/opencv/modules/core/test/ocl/test_arithm.cpp:1311: NORM_RELATIVE
[ INFO:[email protected]] global ts.cpp:848 testTearDown Memory_usage (OpenCL): 5209 (base=400  current=4664)
[  FAILED  ] OCL_Arithm/Norm.NORM_INF_2args/4, where GetParam() = (CV_8U, Channels(3), false) (0 ms)
[ RUN      ] OCL_Arithm/Norm.NORM_INF_2args/5, where GetParam() = (CV_8U, Channels(3), true)
/build/precommit_opencl_linux/4.x/opencv/modules/core/test/ocl/test_arithm.cpp:1320: Failure
relativeError(cpuRes, gpuRes, 1e-6) evaluates to false, where
cpuRes evaluates to 0.98431372549019602
gpuRes evaluates to 0.98818897637795278
1e-6 evaluates to 9.9999999999999995e-07
Google Test trace:
/build/precommit_opencl_linux/4.x/opencv/modules/core/test/ocl/test_arithm.cpp:1311: NORM_RELATIVE
[ INFO:[email protected]] global ts.cpp:848 testTearDown Memory_usage (OpenCL): 618280 (base=400  current=559301)
[  FAILED  ] OCL_Arithm/Norm.NORM_INF_2args/5, where GetParam() = (CV_8U, Channels(3), true) (3 ms)

Windows OpenCL:

[ RUN      ] OCL_Arithm/Norm.NORM_INF_2args/4, where GetParam() = (CV_8U, Channels(3), false)
C:\build\precommit_opencl\4.x\opencv\modules\core\test\ocl\test_arithm.cpp(1320): error: relativeError(cpuRes, gpuRes, 1e-6) evaluates to false, where
cpuRes evaluates to 0.98431372549019602
gpuRes evaluates to 0.98818897637795278
1e-6 evaluates to 9.9999999999999995e-07
Google Test trace:
C:\build\precommit_opencl\4.x\opencv\modules\core\test\ocl\test_arithm.cpp(1311): NORM_RELATIVE
[ INFO:[email protected]] global ts.cpp:848 cvtest::testTearDown Memory_usage (OpenCL): 5209 (base=22056  current=26320)
[  FAILED  ] OCL_Arithm/Norm.NORM_INF_2args/4, where GetParam() = (CV_8U, Channels(3), false) (0 ms)
[ RUN      ] OCL_Arithm/Norm.NORM_INF_2args/5, where GetParam() = (CV_8U, Channels(3), true)
C:\build\precommit_opencl\4.x\opencv\modules\core\test\ocl\test_arithm.cpp(1320): error: relativeError(cpuRes, gpuRes, 1e-6) evaluates to false, where
cpuRes evaluates to 0.98431372549019602
gpuRes evaluates to 0.98818897637795278
1e-6 evaluates to 9.9999999999999995e-07
Google Test trace:
C:\build\precommit_opencl\4.x\opencv\modules\core\test\ocl\test_arithm.cpp(1311): NORM_RELATIVE
[ INFO:[email protected]] global ts.cpp:848 cvtest::testTearDown Memory_usage (OpenCL): 618280 (base=22056  current=580957)
[  FAILED  ] OCL_Arithm/Norm.NORM_INF_2args/5, where GetParam() = (CV_8U, Channels(3), true) (2 ms)

asmorkalov · 2025-01-27T08:29:16Z

Tests pass on Intel 1135g7 and AMD Ryzen 7 5700G + Nvidia JF 1080.

asmorkalov · 2025-01-27T08:29:47Z

@opencv-alalek Please take a look on the PR again.

opencv-alalek

👍

This PR could be merged.
I will merge opencv_3rdparty after that.

Update IPP integration opencv#26463 Please merge together with opencv/opencv_3rdparty#88 Supported IPP version was updated to IPP 2022.0.0 for Linux and Windows. 32-bit binaries are dropped since this release. Previous update: opencv#25935

Update IPP integration #27354 Please merge together with opencv/opencv_3rdparty#96 Supported IPP version was updated to IPP 2022.1.0 for Linux and Windows. Bugs in norm() function which caused failure of sanity check in performance tests were fixed, IPP calls were enabled. Previous update: #26463

Update IPP integration (v2022.0.0)

cf5e0c3

opencv-alalek self-requested a review November 14, 2024 18:09

tizmajlo reviewed Nov 15, 2024

View reviewed changes

opencv-alalek added this to the 4.11.0 milestone Nov 16, 2024

opencv-alalek added optimization category: 3rdparty labels Nov 16, 2024

IPP call of Norm_8u_L1() was disabled

9a34099

Updated commit hash for 32bit

5f0864e

opencv-alalek approved these changes Dec 24, 2024

View reviewed changes

asmorkalov modified the milestones: 4.11.0, 4.12.0 Dec 30, 2024

eplankin mentioned this pull request Jan 2, 2025

(.note.gnu.property+0x0): data is too short error during IPP linkage with NDK 27 for Android x86 #26072

Closed

Disabled other failed modes of Norm()

1cdf475

asmorkalov approved these changes Jan 27, 2025

View reviewed changes

opencv-alalek approved these changes Jan 27, 2025

View reviewed changes

asmorkalov merged commit ae57c54 into opencv:4.x Jan 27, 2025
28 of 31 checks passed

asmorkalov mentioned this pull request Feb 19, 2025

5.x merge 4.x #26939

Merged

eplankin mentioned this pull request May 23, 2025

Update IPP integration #27354

Merged

Uh oh!

Update IPP integration #26463

Update IPP integration #26463

Uh oh!

Conversation

eplankin commented Nov 14, 2024

Uh oh!

opencv-alalek commented Nov 15, 2024

Uh oh!

eplankin commented Nov 15, 2024

Uh oh!

tizmajlo Nov 15, 2024

Choose a reason for hiding this comment

Uh oh!

asmorkalov commented Dec 16, 2024

Uh oh!

eplankin commented Dec 18, 2024

Uh oh!

asmorkalov commented Dec 18, 2024

Uh oh!

opencv-alalek commented Dec 23, 2024

Uh oh!

opencv-alalek commented Dec 25, 2024

Uh oh!

asmorkalov commented Jan 27, 2025

Uh oh!

asmorkalov commented Jan 27, 2025

Uh oh!

opencv-alalek left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants