Skip to content

CUDA 12.2 fp16 dnn compilation error #23893

Closed
Closed
CUDA 12.2 fp16 dnn compilation error#23893
Labels
bugcategory: gpu/cuda (contrib)OpenCV 4.0+: moved to opencv_contrib
@cudawarped

Description

@cudawarped
cudawarped
opened on Jun 29, 2023

System Information

OpenCV version: 4.8.0
OS: Windows 11
Compiler: Visual Studio 2022
CUDA: 12.2

Detailed description

Switching from CUDA 12.1 to 12.2 results in several compilation like the one below when compiling the dnn module.

D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): error C2666: 'operator !=': overloaded functions have similar conversions 
FAILED: modules/dnn/CMakeFiles/opencv_dnn.dir/src/layers/normalize_bbox_layer.cpp.obj
C:\PROGRA~1\MICROS~2\2022\COMMUN~1\VC\Tools\MSVC\1434~1.319\bin\Hostx64\x64\cl.exe  /nologo /TP -DCVAPI_EXPORTS -DCV_CUDA4DNN=1 -DCV_OCL4DNN=1 -DENABLE_PLUGINS -DHAVE_FLATBUFFERS=1 -DHAVE_PROTOBUF=1 -D_CRT_SECURE_NO_WARNINGS=1 -D_USE_MATH_DEFINES -D_VARIADIC_MAX=10 -D_WIN32_WINNT=0x0601 -D__OPENCV_BUILD=1 -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -ID:\build\opencv\4_8_0\cuda_12_2_test\3rdparty\ippicv\ippicv_win\icv\include -ID:\build\opencv\4_8_0\cuda_12_2_test\3rdparty\ippicv\ippicv_win\iw\include -ID:\repos\opencv\opencv\modules\dnn\src -ID:\repos\opencv\opencv\modules\dnn\include -ID:\build\opencv\4_8_0\cuda_12_2_test\modules\dnn -ID:\repos\opencv\contrib\modules\cudev\include -ID:\repos\opencv\opencv\modules\core\include -ID:\repos\opencv\opencv\modules\imgproc\include -ID:\repos\opencv\opencv\modules\dnn\misc\caffe -ID:\repos\opencv\opencv\modules\dnn\misc\tensorflow -ID:\repos\opencv\opencv\modules\dnn\misc\onnx -ID:\repos\opencv\opencv\modules\dnn\misc\tflite -ID:\repos\opencv\opencv\3rdparty\include\opencl\1.2 -ID:\repos\opencv\opencv\modules\ts\include -ID:\repos\opencv\opencv\modules\imgcodecs\include -ID:\repos\opencv\opencv\modules\videoio\include -ID:\repos\opencv\opencv\modules\highgui\include -external:ID:\build\opencv\4_8_0\cuda_12_2_test -external:I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\include" -external:ID:\repos\opencv\opencv\3rdparty\flatbuffers\include -external:ID:\repos\opencv\opencv\3rdparty\protobuf\src -external:W0 /DWIN32 /D_WINDOWS /W4 /GR  /D _CRT_SECURE_NO_DEPRECATE /D _CRT_NONSTDC_NO_DEPRECATE /D _SCL_SECURE_NO_WARNINGS /Gy /bigobj /Oi  /fp:precise /FS     /EHa /wd4127 /wd4251 /wd4324 /wd4275 /wd4512 /wd4589 /wd4819 /wd4244 /wd4267 /wd4018 /wd4355 /wd4800 /wd4251 /wd4996 /wd4146 /wd4305 /wd4127 /wd4100 /wd4512 /wd4125 /wd4389 /wd4510 /wd4610 /wd4702 /wd4456 /wd4457 /wd4065 /wd4310 /wd4661 /wd4506 /wd4125 /wd4267 /wd4127 /wd4244 /wd4512 /wd4702 /wd4456 /wd4510 /wd4610 /wd4800 /wd4701 /wd4703 /wd4505 /wd4458  /MD /O2 /Ob2 /DNDEBUG /showIncludes /Fomodules\dnn\CMakeFiles\opencv_dnn.dir\src\layers\normalize_bbox_layer.cpp.obj /Fdlib\opencv_dnn480.pdb /FS -c D:\repos\opencv\opencv\modules\dnn\src\layers\normalize_bbox_layer.cpp
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): error C2666: 'operator !=': overloaded functions have similar conversions
D:\repos\opencv\opencv\modules\core\include\opencv2/core/persistence.hpp(1332): note: could be 'bool cv::operator !=(const cv::FileNodeIterator &,const cv::FileNodeIterator &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.inl.hpp(2807): note: or       'bool cv::operator !=(const cv::SparseMatConstIterator &,const cv::SparseMatConstIterator &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.inl.hpp(2398): note: or       'bool cv::operator !=(const cv::MatConstIterator &,const cv::MatConstIterator &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(3708): note: or       'cv::MatExpr cv::operator !=(double,const cv::Mat &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(3707): note: or       'cv::MatExpr cv::operator !=(const cv::Mat &,double)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(3706): note: or       'cv::MatExpr cv::operator !=(const cv::Mat &,const cv::Mat &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(583): note: or       'bool cv::operator !=(const cv::UMatData::MemoryFlag &,const int &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(266): note: or       'bool cv::operator !=(const cv::_InputArray::KindFlag &,const int &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(66): note: or       'bool cv::operator !=(const cv::AccessFlag &,const int &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(2086): note: or       'bool cv::operator !=(const cv::Range &,const cv::Range &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1279): note: or       'cv::hal_baseline::v_int64x2 cv::hal_baseline::operator !=(const cv::hal_baseline::v_int64x2 &,const cv::hal_baseline::v_int64x2 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1278): note: or       'cv::hal_baseline::v_uint64x2 cv::hal_baseline::operator !=(const cv::hal_baseline::v_uint64x2 &,const cv::hal_baseline::v_uint64x2 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1261): note: or       'cv::hal_baseline::v_float64x2 cv::hal_baseline::operator !=(const cv::hal_baseline::v_float64x2 &,const cv::hal_baseline::v_float64x2 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1260): note: or       'cv::hal_baseline::v_float32x4 cv::hal_baseline::operator !=(const cv::hal_baseline::v_float32x4 &,const cv::hal_baseline::v_float32x4 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1244): note: or       'cv::hal_baseline::v_int32x4 cv::hal_baseline::operator !=(const cv::hal_baseline::v_int32x4 &,const cv::hal_baseline::v_int32x4 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1244): note: or       'cv::hal_baseline::v_uint32x4 cv::hal_baseline::operator !=(const cv::hal_baseline::v_uint32x4 &,const cv::hal_baseline::v_uint32x4 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1243): note: or       'cv::hal_baseline::v_int16x8 cv::hal_baseline::operator !=(const cv::hal_baseline::v_int16x8 &,const cv::hal_baseline::v_int16x8 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1243): note: or       'cv::hal_baseline::v_uint16x8 cv::hal_baseline::operator !=(const cv::hal_baseline::v_uint16x8 &,const cv::hal_baseline::v_uint16x8 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1242): note: or       'cv::hal_baseline::v_int8x16 cv::hal_baseline::operator !=(const cv::hal_baseline::v_int8x16 &,const cv::hal_baseline::v_int8x16 &)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/hal/intrin_sse.hpp(1242): note: or       'cv::hal_baseline::v_uint8x16 cv::hal_baseline::operator !=(const cv::hal_baseline::v_uint8x16 &,const cv::hal_baseline::v_uint8x16 &)'
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\include\cuda_fp16.hpp(714): note: or       'bool operator !=(const __half &,const __half &)' [found using argument-dependent lookup]
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\include\cuda_fp16.hpp(917): note: or       'bool operator !=(const __half2 &,const __half2 &)' [found using argument-dependent lookup]
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\include\cuda_bf16.hpp(700): note: or       'bool operator !=(const __nv_bfloat16 &,const __nv_bfloat16 &)' [found using argument-dependent lookup]
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\include\cuda_bf16.hpp(903): note: or       'bool operator !=(const __nv_bfloat162 &,const __nv_bfloat162 &)' [found using argument-dependent lookup]
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(float, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(signed char, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(unsigned char, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(char, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(short, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(unsigned short, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(int, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(unsigned int, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(long, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(unsigned long, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(__int64, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(unsigned __int64, double)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: or       'built-in C++ operator!=(bool, double)'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.inl.hpp(2718): note: or       'bool cv::operator !=(const cv::MatIterator_<_Tp> &,const cv::MatIterator_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::MatIterator_<_Tp> &,const cv::MatIterator_<_Tp> &)': could not deduce template argument for 'const cv::MatIterator_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.inl.hpp(2581): note: or       'bool cv::operator !=(const cv::MatConstIterator_<_Tp> &,const cv::MatConstIterator_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::MatConstIterator_<_Tp> &,const cv::MatConstIterator_<_Tp> &)': could not deduce template argument for 'const cv::MatConstIterator_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(3712): note: or       'cv::MatExpr cv::operator !=(const cv::Matx<_Tp,m,n> &,const cv::Mat &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'cv::MatExpr cv::operator !=(const cv::Matx<_Tp,m,n> &,const cv::Mat &)': could not deduce template argument for 'const cv::Matx<_Tp,m,n> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/mat.hpp(3710): note: or       'cv::MatExpr cv::operator !=(const cv::Mat &,const cv::Matx<_Tp,m,n> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'cv::MatExpr cv::operator !=(const cv::Mat &,const cv::Matx<_Tp,m,n> &)': could not deduce template argument for 'const cv::Matx<_Tp,m,n> &' from 'double'
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(2276): note: or       'bool cv::operator !=(const cv::Scalar_<_Tp> &,const cv::Scalar_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Scalar_<_Tp> &,const cv::Scalar_<_Tp> &)': could not deduce template argument for 'const cv::Scalar_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(1964): note: or       'bool cv::operator !=(const cv::Rect_<_Tp> &,const cv::Rect_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Rect_<_Tp> &,const cv::Rect_<_Tp> &)': could not deduce template argument for 'const cv::Rect_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(1795): note: or       'bool cv::operator !=(const cv::Size_<_Tp> &,const cv::Size_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Size_<_Tp> &,const cv::Size_<_Tp> &)': could not deduce template argument for 'const cv::Size_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(1584): note: or       'bool cv::operator !=(const cv::Point3_<_Tp> &,const cv::Point3_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Point3_<_Tp> &,const cv::Point3_<_Tp> &)': could not deduce template argument for 'const cv::Point3_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(1337): note: or       'bool cv::operator !=(const cv::Point_<_Tp> &,const cv::Point_<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Point_<_Tp> &,const cv::Point_<_Tp> &)': could not deduce template argument for 'const cv::Point_<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/types.hpp(1052): note: or       'bool cv::operator !=(const cv::Complex<_Tp> &,const cv::Complex<_Tp> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Complex<_Tp> &,const cv::Complex<_Tp> &)': could not deduce template argument for 'const cv::Complex<_Tp> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\core\include\opencv2/core/matx.hpp(1365): note: or       'bool cv::operator !=(const cv::Matx<_Tp,m,n> &,const cv::Matx<_Tp,m,n> &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: 'bool cv::operator !=(const cv::Matx<_Tp,m,n> &,const cv::Matx<_Tp,m,n> &)': could not deduce template argument for 'const cv::Matx<_Tp,m,n> &' from 'T'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114): note: while trying to match the argument list '(T, double)'
        with
        [
            T=half
        ]
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(93): note: while compiling class template member function 'void cv::dnn::cuda4dnn::NormalizeOp<half>::forward(const std::vector<cv::Ptr<cv::dnn::dnn4_v20230620::BackendWrapper>,std::allocator<cv::Ptr<cv::dnn::dnn4_v20230620::BackendWrapper>>> &,const std::vector<cv::Ptr<cv::dnn::dnn4_v20230620::BackendWrapper>,std::allocator<cv::Ptr<cv::dnn::dnn4_v20230620::BackendWrapper>>> &,cv::dnn::cuda4dnn::csl::Workspace &)'
D:\repos\opencv\opencv\modules\dnn\src\layers\../op_cuda.hpp(194): note: see reference to class template instantiation 'cv::dnn::cuda4dnn::NormalizeOp<half>' being compiled
D:\repos\opencv\opencv\modules\dnn\src\layers\normalize_bbox_layer.cpp(332): note: see reference to function template instantiation 'cv::Ptr<cv::dnn::dnn4_v20230620::BackendNode> cv::dnn::make_cuda_node<cv::dnn::cuda4dnn::NormalizeOp,cv::dnn::cuda4dnn::csl::Stream,const cv::Mat&,cv::dnn::cuda4dnn::NormalizeConfiguration<float>&>(int,cv::dnn::cuda4dnn::csl::Stream &&,const cv::Mat &,cv::dnn::cuda4dnn::NormalizeConfiguration<float> &)' being compiled

it looks like a result of the CUDA Math API's being reworked in this version of CUDA, specifically changes to the C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\include\cuda_fp16.hpp file.

Steps to reproduce

cmake.exe -BBUILD_DIR -HOPENCV_DIR -GNinja -DOPENCV_EXTRA_MODULES_PATH=OPENCV_CONTRIB_MODULES -DBUILD_opencv_world=OFF
-DWITH_CUDA=ON -DCUDA_TOOLKIT_ROOT_DIR="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.2" -DCUDA_ARCH_BIN=8.6

cmake.exe --build BUILD_DIR --target opencv_dnn

Issue submission checklist

  • I report the issue, it's not a question
  • I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
  • I updated to the latest OpenCV version and the issue is still there
  • There is reproducer code and related data files (videos, images, onnx, etc)

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugcategory: gpu/cuda (contrib)OpenCV 4.0+: moved to opencv_contrib

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions