mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

Compile error on tensorrt8 branch #276

New Issue
Closed
opened 2022-01-11 08:38:59 +01:00 by mryau · 6 comments
mryau commented 2022-01-11 08:38:59 +01:00 (Migrated from github.com)
Copy Link
Copy Source

Ubuntu 20.04
Cuda 11.5
CudNN 8.3.1
TensorRT 8.2.2
OpenCV 4.52

Workflow as defined in README.md leads to the following error:

cmake .. 
make
[ 22%] Linking CXX shared library libkernels.so
/usr/bin/ld: CMakeFiles/kernels.dir/kernels_intermediate_link.o: relocation R_X86_64_PC32 against symbol `__fatbinwrap_dca1913b_17_activation_elu_cu_d406a46e' can not be used when making a shared object; recompile with -fPIC
/usr/bin/ld: final link failed: bad value
collect2: error: ld returned 1 exit status
make[2]: *** [CMakeFiles/kernels.dir/build.make:8225: libkernels.so] Error 1
make[1]: *** [CMakeFiles/Makefile2:872: CMakeFiles/kernels.dir/all] Error 2
make: *** [Makefile:130: all] Error 2

In case I explicitly say what build type I want all works perfectly:

cmake -DCMAKE_BUILD_TYPE=Release ..
make
...
[100%] Linking CXX executable test_yolo4_berkeley
[100%] Built target test_yolo4_berkeley
Ubuntu 20.04 Cuda 11.5 CudNN 8.3.1 TensorRT 8.2.2 OpenCV 4.52 Workflow as defined in README.md leads to the following error: ``` cmake .. make [ 22%] Linking CXX shared library libkernels.so /usr/bin/ld: CMakeFiles/kernels.dir/kernels_intermediate_link.o: relocation R_X86_64_PC32 against symbol `__fatbinwrap_dca1913b_17_activation_elu_cu_d406a46e' can not be used when making a shared object; recompile with -fPIC /usr/bin/ld: final link failed: bad value collect2: error: ld returned 1 exit status make[2]: *** [CMakeFiles/kernels.dir/build.make:8225: libkernels.so] Error 1 make[1]: *** [CMakeFiles/Makefile2:872: CMakeFiles/kernels.dir/all] Error 2 make: *** [Makefile:130: all] Error 2 ``` In case I explicitly say what build type I want all works perfectly: ``` cmake -DCMAKE_BUILD_TYPE=Release .. make ... [100%] Linking CXX executable test_yolo4_berkeley [100%] Built target test_yolo4_berkeley ```
perseusdg commented 2022-01-11 08:43:15 +01:00 (Migrated from github.com)
Copy Link
Copy Source

I believe its mentioned on the readme of the tensorrt8 branch to define build type https://github.com/ceccocats/tkDNN/tree/tensorrt8#how-to-compile-this-repo

I believe its mentioned on the readme of the tensorrt8 branch to define build type https://github.com/ceccocats/tkDNN/tree/tensorrt8#how-to-compile-this-repo
mochechan commented 2022-01-11 09:20:14 +01:00 (Migrated from github.com)
Copy Link
Copy Source

I also face the same problem with the tensorrt8 branch. Actually, the error messages are different if compiling with the master branch.

root@314654f59059:~/tkDNN# git checkout tensorrt8
Branch 'tensorrt8' set up to track remote branch 'tensorrt8' from 'origin'.
Switched to a new branch 'tensorrt8'
root@314654f59059:~/tkDNN# git status
On branch tensorrt8
Your branch is up to date with 'origin/tensorrt8'.

nothing to commit, working tree clean

root@314654f59059:~/tkDNN/build# cmake ..
-- CUDA Architecture: -arch=sm_86
-- CUDA Version: 11.4
-- CUDA Path: /usr/local/cuda
-- CUDA Libararies: /usr/local/cuda/lib64/libcudart_static.a;Threads::Threads;dl;/usr/lib/x86_64-linux-gnu/librt.so
-- CUDA Performance Primitives: /usr/local/cuda/lib64/libnppc.so;/usr/local/cuda/lib64/libnppial.so;/usr/local/cuda/lib64/libnppicc.so;;/usr/local/cuda/lib64/libnppidei.so;/usr/local/cuda/lib64/libnppif.so;/usr/local/cuda/lib64/libnppig.so;/usr/local/cuda/lib64/libnppim.so;/usr/local/cuda/lib64/libnppist.so;/usr/local/cuda/lib64/libnppisu.so;/usr/local/cuda/lib64/libnppitc.so;/usr/local/cuda/lib64/libnpps.so
-- Found CUDNN: /usr/lib/x86_64-linux-gnu/libcudnn.so
-- Found CUDNN include: /usr/include
-- Found NVINFER: /usr/lib/x86_64-linux-gnu/libnvinfer.so
-- Found NVINFER include: /usr/include/x86_64-linux-gnu
Eigen DIR: /usr/include/eigen3
install dir:/usr/local
-- Configuring done
-- Generating done
-- Build files have been written to: /root/tkDNN/build
root@314654f59059:~/tkDNN/build# make -j6
[  0%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_upsample.cu.o
[  1%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_elu.cu.o
[  1%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_leaky.cu.o
[  2%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_logistic.cu.o
[  3%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_mish.cu.o
[  3%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_relu_ceiling.cu.o
[  4%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_sigmoid.cu.o
[  4%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_convert.cu.o
[  5%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_deformable_conv.cu.o
[  6%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_fill.cu.o
[  6%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_normalize.cu.o
[  7%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_pooling.cu.o
[  7%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_postprocessing.cu.o
[  8%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_reorg.cu.o
[  9%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_resize.cu.o
[  9%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_scaladd.cu.o
[ 10%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_shortcut.cu.o
[ 11%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_softmax.cu.o
[ 12%] Building NVCC intermediate link file CMakeFiles/kernels.dir/kernels_intermediate_link.o
[ 13%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationLeakyRT.cpp.o
[ 14%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationLogisticRT.cpp.o
[ 14%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationMishRT.cpp.o
[ 14%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationReLUCeilingRT.cpp.o
[ 15%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/DeformableConvRT.cpp.o
[ 15%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/FlattenConcatRT.cpp.o
[ 16%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/MaxPoolingSizeRT.cpp.o
[ 17%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/RegionRT.cpp.o
[ 17%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ReorgRT.cpp.o
[ 18%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ReshapeRT.cpp.o
[ 18%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ResizeLayerRT.cpp.o
[ 19%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/RouteRT.cpp.o
[ 20%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ShortcutRT.cpp.o
[ 20%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/UpsampleRT.cpp.o
[ 21%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/YoloRT.cpp.o
[ 22%] Linking CXX shared library libkernels.so
/usr/bin/ld: CMakeFiles/kernels.dir/kernels_intermediate_link.o: relocation R_X86_64_PC32 against symbol `__fatbinwrap_49_tmpxft_000003f0_00000000_7_activation_elu_cpp1_ii_d406a46e' can not be used when making a shared object; recompile with -fPIC
/usr/bin/ld: final link failed: bad value
collect2: error: ld returned 1 exit status
make[2]: *** [CMakeFiles/kernels.dir/build.make:5477: libkernels.so] Error 1
make[1]: *** [CMakeFiles/Makefile2:179: CMakeFiles/kernels.dir/all] Error 2
make: *** [Makefile:136: all] Error 2
I also face the same problem with the tensorrt8 branch. Actually, the error messages are different if compiling with the master branch. ``` root@314654f59059:~/tkDNN# git checkout tensorrt8 Branch 'tensorrt8' set up to track remote branch 'tensorrt8' from 'origin'. Switched to a new branch 'tensorrt8' root@314654f59059:~/tkDNN# git status On branch tensorrt8 Your branch is up to date with 'origin/tensorrt8'. nothing to commit, working tree clean root@314654f59059:~/tkDNN/build# cmake .. -- CUDA Architecture: -arch=sm_86 -- CUDA Version: 11.4 -- CUDA Path: /usr/local/cuda -- CUDA Libararies: /usr/local/cuda/lib64/libcudart_static.a;Threads::Threads;dl;/usr/lib/x86_64-linux-gnu/librt.so -- CUDA Performance Primitives: /usr/local/cuda/lib64/libnppc.so;/usr/local/cuda/lib64/libnppial.so;/usr/local/cuda/lib64/libnppicc.so;;/usr/local/cuda/lib64/libnppidei.so;/usr/local/cuda/lib64/libnppif.so;/usr/local/cuda/lib64/libnppig.so;/usr/local/cuda/lib64/libnppim.so;/usr/local/cuda/lib64/libnppist.so;/usr/local/cuda/lib64/libnppisu.so;/usr/local/cuda/lib64/libnppitc.so;/usr/local/cuda/lib64/libnpps.so -- Found CUDNN: /usr/lib/x86_64-linux-gnu/libcudnn.so -- Found CUDNN include: /usr/include -- Found NVINFER: /usr/lib/x86_64-linux-gnu/libnvinfer.so -- Found NVINFER include: /usr/include/x86_64-linux-gnu Eigen DIR: /usr/include/eigen3 install dir:/usr/local -- Configuring done -- Generating done -- Build files have been written to: /root/tkDNN/build root@314654f59059:~/tkDNN/build# make -j6 [ 0%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_upsample.cu.o [ 1%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_elu.cu.o [ 1%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_leaky.cu.o [ 2%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_logistic.cu.o [ 3%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_mish.cu.o [ 3%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_relu_ceiling.cu.o [ 4%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_activation_sigmoid.cu.o [ 4%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_convert.cu.o [ 5%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_deformable_conv.cu.o [ 6%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_fill.cu.o [ 6%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_normalize.cu.o [ 7%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_pooling.cu.o [ 7%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_postprocessing.cu.o [ 8%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_reorg.cu.o [ 9%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_resize.cu.o [ 9%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_scaladd.cu.o [ 10%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_shortcut.cu.o [ 11%] Building NVCC (Device) object CMakeFiles/kernels.dir/src/kernels/kernels_generated_softmax.cu.o [ 12%] Building NVCC intermediate link file CMakeFiles/kernels.dir/kernels_intermediate_link.o [ 13%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationLeakyRT.cpp.o [ 14%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationLogisticRT.cpp.o [ 14%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationMishRT.cpp.o [ 14%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ActivationReLUCeilingRT.cpp.o [ 15%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/DeformableConvRT.cpp.o [ 15%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/FlattenConcatRT.cpp.o [ 16%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/MaxPoolingSizeRT.cpp.o [ 17%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/RegionRT.cpp.o [ 17%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ReorgRT.cpp.o [ 18%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ReshapeRT.cpp.o [ 18%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ResizeLayerRT.cpp.o [ 19%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/RouteRT.cpp.o [ 20%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/ShortcutRT.cpp.o [ 20%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/UpsampleRT.cpp.o [ 21%] Building CXX object CMakeFiles/kernels.dir/src/pluginsRT/YoloRT.cpp.o [ 22%] Linking CXX shared library libkernels.so /usr/bin/ld: CMakeFiles/kernels.dir/kernels_intermediate_link.o: relocation R_X86_64_PC32 against symbol `__fatbinwrap_49_tmpxft_000003f0_00000000_7_activation_elu_cpp1_ii_d406a46e' can not be used when making a shared object; recompile with -fPIC /usr/bin/ld: final link failed: bad value collect2: error: ld returned 1 exit status make[2]: *** [CMakeFiles/kernels.dir/build.make:5477: libkernels.so] Error 1 make[1]: *** [CMakeFiles/Makefile2:179: CMakeFiles/kernels.dir/all] Error 2 make: *** [Makefile:136: all] Error 2 ```
perseusdg commented 2022-01-11 09:25:40 +01:00 (Migrated from github.com)
Copy Link
Copy Source

If you define the build type it should resolve your issue,
tkDNN on tensorrt8 branches has its flags split into debug and release as shown below

if(UNIX)
    if(CMAKE_BUILD_TYPE MATCHES Release)
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3")
        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
    endif()

    if(CMAKE_BUILD_TYPE MATCHES Debug)
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g3")
        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -G -g)
    endif()
endif()

if you dont want to define the build type then append -fPIC to CMAKE_CXX_FLAGS similar to what is there on the master branch https://github.com/ceccocats/tkDNN/blob/04de9908a68588c32dcf94d84b59fdf71cf8d9a8/CMakeLists.txt#L6

If you define the build type it should resolve your issue, tkDNN on tensorrt8 branches has its flags split into debug and release as shown below ``` if(UNIX) if(CMAKE_BUILD_TYPE MATCHES Release) set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3") set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32) endif() if(CMAKE_BUILD_TYPE MATCHES Debug) set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g3") set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -G -g) endif() endif() ``` if you dont want to define the build type then append ```-fPIC``` to ```CMAKE_CXX_FLAGS``` similar to what is there on the master branch https://github.com/ceccocats/tkDNN/blob/04de9908a68588c32dcf94d84b59fdf71cf8d9a8/CMakeLists.txt#L6
mochechan commented 2022-01-11 09:46:31 +01:00 (Migrated from github.com)
Copy Link
Copy Source

Unfortunately, the "-fPIC" is already in my CMakeList.txt. Any other ideas? Thanks a lot.

    38	if(UNIX)
    39	    if(CMAKE_BUILD_TYPE MATCHES Release)
    40	        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3")
    41	        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
    42	    endif()
    43	
    44	    if(CMAKE_BUILD_TYPE MATCHES Debug)
    45	        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g3")
    46	        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -G -g)
    47	    endif()
    48	endif()
Unfortunately, the "-fPIC" is already in my CMakeList.txt. Any other ideas? Thanks a lot. ``` 38 if(UNIX) 39 if(CMAKE_BUILD_TYPE MATCHES Release) 40 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3") 41 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32) 42 endif() 43 44 if(CMAKE_BUILD_TYPE MATCHES Debug) 45 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g3") 46 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -G -g) 47 endif() 48 endif() ```
perseusdg commented 2022-01-11 09:52:32 +01:00 (Migrated from github.com)
Copy Link
Copy Source

I meant doing something like this

if(UNIX)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC")
    if(CMAKE_BUILD_TYPE MATCHES Release)
       ....
endif()

Since the CMakeList.txt you are using currently ,needs a build type to be defined for -fPIC , moving the -fPIC definition to just above the individual build type definitions should do the trick in case no build type is defined

I meant doing something like this ``` if(UNIX) set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC") if(CMAKE_BUILD_TYPE MATCHES Release) .... endif() ``` Since the CMakeList.txt you are using currently ,needs a build type to be defined for -fPIC , moving the -fPIC definition to just above the individual build type definitions should do the trick in case no build type is defined
mochechan commented 2022-01-11 10:24:45 +01:00 (Migrated from github.com)
Copy Link
Copy Source

It works. Thank you.

It works. Thank you.
This repo is archived. You cannot comment on issues.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
duplicate
enhancement
explanation
help wanted
invalid
question
wontfix
No Label
Milestone
No items
No Milestone
Assignees
Clear assignees
eeeck (Erick)
No Assignees
1 Participants
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: mmr/tkDNN#276
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?