1
0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00

Merge pull request #20664 from anna-khakimova:ak/resize_simd

Fluid: SIMD for Resize Linear 8UC3

* SIMD for fluid::Resize 8U3C

* Rework horizontal pass + add 8U4C case

* Reproduce stackoverflow test

* StackOverflow test

* SSE42 impl

* SSE42 impl improvement

* GAPI:SSE42 simd opt for Resize 8UC3. Final version

* Fix tests

* Conditional compilation fix

* Applied comments

* Applied comments. Step2

* Applied comments. Step2
This commit is contained in:
Anna Khakimova
2021-11-10 16:19:52 +03:00
committed by GitHub
parent d934bb15b0
commit 3cfca01372
11 changed files with 1151 additions and 110 deletions
@@ -321,16 +321,28 @@ INSTANTIATE_TEST_CASE_P(TransposePerfTestCPU, TransposePerfTest,
INSTANTIATE_TEST_CASE_P(ResizePerfTestCPU, ResizePerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_16UC1, CV_16SC1),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
Values(cv::INTER_NEAREST, cv::INTER_LINEAR, cv::INTER_AREA),
Values(szSmall128, szVGA, sz720p, sz1080p),
Values(cv::Size(64, 64),
cv::Size(30, 30)),
cv::Size(32, 32)),
Values(cv::compile_args(CORE_CPU))));
INSTANTIATE_TEST_CASE_P(BottleneckKernelsPerfTestCPU, BottleneckKernelsConstInputPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values("cv/optflow/frames/1080p_00.png", "cv/optflow/frames/720p_00.png",
"cv/optflow/frames/VGA_00.png", "cv/dnn_face/recognition/Aaron_Tippin_0001.jpg"),
Values(cv::compile_args(CORE_CPU))));
INSTANTIATE_TEST_CASE_P(ResizeInSimpleGraphPerfTestCPU, ResizeInSimpleGraphPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC3),
Values(szSmall128, szVGA, sz720p, sz1080p),
Values(cv::compile_args(CORE_CPU))));
INSTANTIATE_TEST_CASE_P(ResizeFxFyPerfTestCPU, ResizeFxFyPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_16UC1, CV_16SC1),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
Values(cv::INTER_NEAREST, cv::INTER_LINEAR, cv::INTER_AREA),
Values(szSmall128, szVGA, sz720p, sz1080p),
Values(0.5, 0.1),
@@ -277,18 +277,31 @@ INSTANTIATE_TEST_CASE_P(ConvertToPerfTestFluid, ConvertToPerfTest,
Values(cv::compile_args(CORE_FLUID))));
INSTANTIATE_TEST_CASE_P(ResizePerfTestFluid, ResizePerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC3/*CV_8UC1, CV_16UC1, CV_16SC1*/),
Values(/*cv::INTER_NEAREST,*/ cv::INTER_LINEAR/*, cv::INTER_AREA*/),
Combine(Values(Tolerance_FloatRel_IntAbs(1e-5, 1).to_compare_f()),
Values(CV_8UC3),
Values(cv::INTER_LINEAR),
Values(szSmall128, szVGA, sz720p, sz1080p),
Values(cv::Size(64, 64),
cv::Size(30, 30)),
Values(cv::compile_args(CORE_FLUID))));
#define IMGPROC_FLUID cv::gapi::imgproc::fluid::kernels()
INSTANTIATE_TEST_CASE_P(BottleneckKernelsPerfTestFluid, BottleneckKernelsConstInputPerfTest,
Combine(Values(AbsSimilarPoints(0, 1).to_compare_f()),
Values("cv/optflow/frames/1080p_00.png", "cv/optflow/frames/720p_00.png",
"cv/optflow/frames/VGA_00.png", "cv/dnn_face/recognition/Aaron_Tippin_0001.jpg"),
Values(cv::compile_args(CORE_FLUID, IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(ResizeInSimpleGraphPerfTestFluid, ResizeInSimpleGraphPerfTest,
Combine(Values(Tolerance_FloatRel_IntAbs(1e-5, 1).to_compare_f()),
Values(CV_8UC3),
Values(szSmall128, szVGA, sz720p, sz1080p),
Values(cv::compile_args(CORE_FLUID, IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(ResizeFxFyPerfTestFluid, ResizeFxFyPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC3/*CV_8UC1, CV_16UC1, CV_16SC1*/),
Values(/*cv::INTER_NEAREST,*/ cv::INTER_LINEAR/*, cv::INTER_AREA*/),
Combine(Values(Tolerance_FloatRel_IntAbs(1e-5, 1).to_compare_f()),
Values(CV_8UC3),
Values(cv::INTER_LINEAR),
Values(szSmall128, szVGA, sz720p, sz1080p),
Values(0.5, 0.1),
Values(0.5, 0.1),