macOS build failed

[lightingghost]

macOS 10.13
CUDA9.0 CUDNN7.0
bazel 0.7.0

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (4) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=float, IntType=tensorflow::int32, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=float, IntType=tensorflow::int32]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (4) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=float, IntType=tensorflow::int32, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=float, IntType=tensorflow::int32]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (4) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=float, IntType=tensorflow::int64, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=float, IntType=tensorflow::int64]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (4) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=float, IntType=tensorflow::int64, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=float, IntType=tensorflow::int64]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=double, IntType=tensorflow::int32, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=double, IntType=tensorflow::int32]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=double, IntType=tensorflow::int32, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=double, IntType=tensorflow::int32]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=double, IntType=tensorflow::int64, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=double, IntType=tensorflow::int64]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=double, IntType=tensorflow::int64, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=double, IntType=tensorflow::int64]" 
(251): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex64, IntType=tensorflow::int32, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex64, IntType=tensorflow::int32]" 
(252): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex64, IntType=tensorflow::int32, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex64, IntType=tensorflow::int32]" 
(252): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex64, IntType=tensorflow::int64, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex64, IntType=tensorflow::int64]" 
(252): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (8) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex64, IntType=tensorflow::int64, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex64, IntType=tensorflow::int64]" 
(252): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (16) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex128, IntType=tensorflow::int32, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex128, IntType=tensorflow::int32]" 
(253): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (16) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex128, IntType=tensorflow::int32, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex128, IntType=tensorflow::int32]" 
(253): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (16) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex128, IntType=tensorflow::int64, useSmem=true]" 
(231): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex128, IntType=tensorflow::int64]" 
(253): here

tensorflow/core/kernels/split_lib_gpu.cu.cc(122): error: specified alignment (16) is different from alignment (2) specified on a previous declaration
          detected during:
            instantiation of "void tensorflow::split_v_kernel<T,IntType,useSmem>(const T *, tensorflow::CudaDeviceArrayStruct<IntType, 8>, IntType, IntType, tensorflow::CudaDeviceArrayStruct<T *, 8>) [with T=tensorflow::complex128, IntType=tensorflow::int64, useSmem=false]" 
(236): here
            instantiation of "void tensorflow::SplitVOpGPULaunch<T, IntType>::Run(const Eigen::GpuDevice &, __nv_bool, const T *, int, int, const tensorflow::CudaDeviceArrayStruct<IntType, 8> &, const tensorflow::CudaDeviceArrayStruct<T *, 8> &) [with T=tensorflow::complex128, IntType=tensorflow::int64]" 
(253): here

16 errors detected in the compilation of "/var/folders/2f/891g9y691gzcy23blt644z6m0000gn/T//tmpxft_00002e92_00000000-6_split_lib_gpu.cu.cpp1.ii".
ERROR: /Users/odin/local/tensorflow/tensorflow/core/kernels/BUILD:387:1: output 'tensorflow/core/kernels/_objs/split_lib_gpu/tensorflow/core/kernels/split_lib_gpu.cu.pic.o' was not created.
ERROR: /Users/odin/local/tensorflow/tensorflow/core/kernels/BUILD:387:1: not all outputs were created or valid.
Target //tensorflow/tools/pip_package:build_pip_package failed to build
Use --verbose_failures to see the command lines of failed build steps.

[smitshilu]

Same issue here.

[chaoswith]

Same issue here too

[omerbenamram]

same here

[kanwei]

👍

[marmarko]

same here

[chaoswith]

finally build success!!

GTX 1080
10.13.1
CUDA9.0 CUDNN7.0
bazel 0.7.0

the point is delete text 'align(sizeof(T))'

concat_lib_gpu.impl.cu.cc
depthwise_conv_op_gpu.cu.cc
splite_lib_gpu.cu.cc

extern shared align(sizeof(T)) unsigned char smem[];
-> extern shared unsigned char smem[];

[smitshilu]

@chaoswith what Xcode command line tool version are you using?

[jart]

We only promise compatibility with XCode 7.2 at this time. See the source install docs. CC: @gunan

PTAL @ekelsen We're getting a compiler error about memory alignment, in recent versions of clang, on some GPU sources you worked on.

It also seems like PyTorch and a bunch of other projects have recently encountered this same issues. Maybe the compiler got stricter and has helped us identify a suboptimality.

@chaoswith While removing __align__ might be a successful workaround, it's probably not a fix we could use in the long run.

[smitshilu]

Build success! Thank you @chaoswith

[gunan]

The GPU build on macOS is completely community supported now, as there are no "Officially supported" systems released by apple that have NVIDIA GPUs.

But looks like this issue was resolved anyway?

[kanwei]

I finally got CUDA working on High Sierra, CUDA 9, and cuDNN 7, by downgrading to Xcode 8.3.3, removing the "align"s, doing the LD_LIBRARY_PATH modification, and removing "-lgomp" from the CUDA makefile. Pain in the butt, but it works!