lunar-walker opened a new issue #20256:
URL: https://github.com/apache/incubator-mxnet/issues/20256
## Description
When compiling version 1.8 with GPU & MKLDNN support from Windows 10 / x64
(Visual studio Community 2019), it fails with lots of errors of pattern calling
a __host__ function from a __global__ function is not allowed...
Initial investigation suggests some sort of confusion by CUDA Compiler for
mixing standard C/C++ libraries functions with CUDA Libraries functions i.e.
error is appearing when calling floor, ceil functions inside CUDA code (both
functions are available in standard C/C++ & CUDA libraries).
I'm using CUDA 11.2 (Pasted below initial out of build_windows.py for
environment)
### Error Message
../src/operator/contrib/adaptive_avg_pooling.cu(73): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::adaptiveaveragepool<float> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(73): error: identifier
"__floorf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(74): error: calling a
__host__ function("__ceilf") from a __global__
function("mxnet::op::adaptiveaveragepool<float> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(74): error: identifier
"__ceilf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(78): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::adaptiveaveragepool<float> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(78): error: identifier
"__floorf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(79): error: calling a
__host__ function("__ceilf") from a __global__
function("mxnet::op::adaptiveaveragepool<float> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(79): error: identifier
"__ceilf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(73): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::adaptiveaveragepool<double> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(73): error: identifier
"__floorf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(74): error: calling a
__host__ function("__ceilf") from a __global__
function("mxnet::op::adaptiveaveragepool<double> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(74): error: identifier
"__ceilf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(78): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::adaptiveaveragepool<double> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(78): error: identifier
"__floorf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(79): error: calling a
__host__ function("__ceilf") from a __global__
function("mxnet::op::adaptiveaveragepool<double> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(79): error: identifier
"__ceilf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(73): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::adaptiveaveragepool< ::mshadow::half::half_t> ") is not
allowed
../src/operator/contrib/adaptive_avg_pooling.cu(73): error: identifier
"__floorf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(74): error: calling a
__host__ function("__ceilf") from a __global__
function("mxnet::op::adaptiveaveragepool< ::mshadow::half::half_t> ") is not
allowed
../src/operator/contrib/adaptive_avg_pooling.cu(74): error: identifier
"__ceilf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(78): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::adaptiveaveragepool< ::mshadow::half::half_t> ") is not
allowed
../src/operator/contrib/adaptive_avg_pooling.cu(78): error: identifier
"__floorf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(79): error: calling a
__host__ function("__ceilf") from a __global__
function("mxnet::op::adaptiveaveragepool< ::mshadow::half::half_t> ") is not
allowed
../src/operator/contrib/adaptive_avg_pooling.cu(79): error: identifier
"__ceilf" is undefined in device code
../src/operator/contrib/adaptive_avg_pooling.cu(130): error: calling a
__host__ function("__floorf") from a __global__
function("mxnet::op::atomicadaptiveaveragegradinput<float> ") is not allowed
../src/operator/contrib/adaptive_avg_pooling.cu(130): error: identifier
"__floorf" is undefined in device code
......
......
......
## What have you tried to solve it?
1. Tried compiling using cmake / visual Studio solution
2. Tried compiling using ci/build_windows.py script
## Environment
<details>
<summary>Environment Information</summary>
----------Python Info----------
Version : 3.9.5
Compiler : MSC v.1928 64 bit (AMD64)
Build : ('tags/v3.9.5:0a7dcbd', 'May 3 2021 17:27:52')
Arch : ('64bit', 'WindowsPE')
------------Pip Info-----------
Version : 21.1.1
Directory : C:\Program
Files\WindowsApps\PythonSoftwareFoundation.Python.3.9_3.9.1520.0_x64__qbz5n2kfra8p0\lib\site-packages\pip
----------MXNet Info-----------
No MXNet installed.
----------System Info----------
Platform : Windows-10-10.0.19042-SP0
system : Windows
node : DESKTOP-6LE30FJ
release : 10
version : 10.0.19042
----------Hardware Info----------
machine : AMD64
processor : Intel64 Family 6 Model 94 Stepping 3, GenuineIntel
Name
Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz
</details>
<details>
<summary>Initial output of ci/build_windows.py</summary>
**********************************************************************
** Visual Studio 2019 Developer Command Prompt v16.9.4
** Copyright (c) 2021 Microsoft Corporation
**********************************************************************
[vcvarsall.bat] Environment initialized for: 'x64'
-- The C compiler identification is MSVC 19.28.29914.0
-- The CXX compiler identification is MSVC 19.28.29914.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files (x86)/Microsoft Visual
Studio/2019/Community/VC/Tools/MSVC/14.28.29910/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files (x86)/Microsoft Visual
Studio/2019/Community/VC/Tools/MSVC/14.28.29910/bin/Hostx64/x64/cl.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- CMAKE_CROSSCOMPILING FALSE
-- CMAKE_HOST_SYSTEM_PROCESSOR AMD64
-- CMAKE_SYSTEM_PROCESSOR AMD64
-- CMAKE_SYSTEM_NAME Windows
-- CMake version '3.20.1' using generator 'Ninja'
-- Looking for a CUDA compiler
-- Looking for a CUDA compiler - C:/Program Files/NVIDIA GPU Computing
Toolkit/CUDA/v11.2/bin/nvcc.exe
-- The CUDA compiler identification is NVIDIA 11.2.67
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - done
-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing
Toolkit/CUDA/v11.2/bin/nvcc.exe - skipped
-- Detecting CUDA compile features
-- Detecting CUDA compile features - done
-- The ASM_MASM compiler identification is MSVC
-- Found assembler: C:/Program Files (x86)/Microsoft Visual
Studio/2019/Community/VC/Tools/MSVC/14.28.29910/bin/Hostx64/x64/ml64.exe
CMake Deprecation Warning at 3rdparty/mkldnn/CMakeLists.txt:17
(cmake_minimum_required):
Compatibility with CMake < 2.8.12 will be removed from a future version of
CMake.
Update the VERSION argument <min> value or use a ...<max> suffix to tell
CMake that the project does not need compatibility with older versions.
-- Intel MKL-DNN compat: set DNNL_BUILD_EXAMPLES to MKLDNN_BUILD_EXAMPLES
with value `OFF`
-- Intel MKL-DNN compat: set DNNL_BUILD_TESTS to MKLDNN_BUILD_TESTS with
value `OFF`
-- Intel MKL-DNN compat: set DNNL_ENABLE_JIT_PROFILING to
MKLDNN_ENABLE_JIT_PROFILING with value `OFF`
-- Intel MKL-DNN compat: set DNNL_LIBRARY_TYPE to MKLDNN_LIBRARY_TYPE with
value `STATIC`
-- Intel MKL-DNN compat: set DNNL_ARCH_OPT_FLAGS to MKLDNN_ARCH_OPT_FLAGS
with value ``
-- Looking for pthread.h
-- Looking for pthread.h - not found
-- Found Threads: TRUE
-- Found OpenMP_C: -openmp (found version "2.0")
-- Found OpenMP_CXX: -openmp (found version "2.0")
-- Found OpenMP: TRUE (found version "2.0")
-- GPU support is disabled
-- Found Git: C:/Program Files/Git/cmd/git.exe (found version
"2.30.0.windows.2")
fatal: not a git repository (or any of the parent directories): .git
-- Primitive cache is enabled
-- Found MKL: D:/oneAPI/mkl/latest/include
-- Found MKL (include: D:/oneAPI/mkl/latest/include, lib:
D:/oneAPI/mkl/latest/lib/intel64/mkl_rt.lib
-- Found OpenCV: D:/code/opencvlib/build_opencv (found version "4.5.2")
found components: core highgui imgproc imgcodecs
-- OpenCV 4.5.2 found (D:/code/opencvlib/build_opencv)
-- OpenCV_LIBS=opencv_core;opencv_highgui;opencv_imgproc;opencv_imgcodecs
USE_LAPACK is ON
CMake Warning at 3rdparty/googletest/googletest/CMakeLists.txt:47 (project):
VERSION keyword not followed by a value or was followed by a value that
expanded to nothing.
CMake Deprecation Warning at
3rdparty/googletest/googletest/CMakeLists.txt:49 (cmake_minimum_required):
Compatibility with CMake < 2.8.12 will be removed from a future version of
CMake.
Update the VERSION argument <min> value or use a ...<max> suffix to tell
CMake that the project does not need compatibility with older versions.
-- Found PythonInterp: D:/miniconda3/python.exe (found version "3.7.9")
-- Found GTest: gtest
-- Found CUDNN: C:/cudnn/cuda/lib/x64/cudnn.lib
-- Looking for clock_gettime in rt
-- Looking for clock_gettime in rt - not found
-- Looking for fopen64
-- Looking for fopen64 - not found
-- Looking for C++ include cxxabi.h
-- Looking for C++ include cxxabi.h - not found
-- Looking for nanosleep
-- Looking for nanosleep - not found
-- Looking for backtrace
-- Looking for backtrace - not found
-- Could NOT find Backtrace (missing: Backtrace_LIBRARY
Backtrace_INCLUDE_DIR)
--
D:/code/apache-mxnet-src-1.8.0-incubating/3rdparty/dmlc-core/cmake/build_config.h.in
-> include/dmlc/build_config.h
-- CUDA: Using the following NVCC architecture flags
-gencode;arch=compute_75,code=sm_75
-- Found CUDAToolkit: C:/Program Files/NVIDIA GPU Computing
Toolkit/CUDA/v11.2/include (found version "11.2.67")
-- Performing Test SUPPORT_MSSE3
-- Performing Test SUPPORT_MSSE3 - Failed
-- Performing Test SUPPORT_MSSE2
-- Performing Test SUPPORT_MSSE2 - Failed
-- Determining F16C support
F16C instruction set is not yet supported for MSVC
-- CUDA: Adding NVCC options: --fatbin-options --compress-all
-- Google Test not found
-- Configuring done
CMake Warning (dev) in CMakeLists.txt:
Policy CMP0104 is not set: CMAKE_CUDA_ARCHITECTURES now detected for NVCC,
empty CUDA_ARCHITECTURES not allowed. Run "cmake --help-policy CMP0104"
for policy details. Use the cmake_policy command to set the policy and
suppress this warning.
CUDA_ARCHITECTURES is empty for target "mxnet_75".
This warning is for project developers. Use -Wno-dev to suppress it.
CMake Warning (dev) in CMakeLists.txt:
Policy CMP0104 is not set: CMAKE_CUDA_ARCHITECTURES now detected for NVCC,
empty CUDA_ARCHITECTURES not allowed. Run "cmake --help-policy CMP0104"
for policy details. Use the cmake_policy command to set the policy and
suppress this warning.
CUDA_ARCHITECTURES is empty for target "customop_gpu_lib".
This warning is for project developers. Use -Wno-dev to suppress it.
-- Generating done
-- Build files have been written to:
D:/code/apache-mxnet-src-1.8.0-incubating/build
</details>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
