RuntimeError: Error compiling objects for extension when install lightconv with setup.py provideded in examples

[Namco0816]

In the examples of lightweight and dynamic convolutions, to install the light/dynamic convolutions I ran the scripts:

python cuda_function_gen.py
python setup.py install

However I got the error "RuntimeError: Error compiling objects for extension"

some part of the error are as follows:

home/namco/fairseq/fairseq/modules/lightconv_layer/../cuda_utils.cu(99): error: namespace "std" has no member "cout"

/home/namco/fairseq/fairseq/modules/lightconv_layer/../cuda_utils.cu(100): error: namespace "std" has no member "endl"

/home/namco/fairseq/fairseq/modules/lightconv_layer/../cuda_utils.cu(101): error: namespace "std" has no member "cout"

/home/namco/fairseq/fairseq/modules/lightconv_layer/../cuda_utils.cu(101): error: namespace "std" has no member "endl"

/home/namco/fairseq/fairseq/modules/lightconv_layer/../cuda_utils.cu(49): error: identifier "SHFL_MASK" is undefined
          detected during:
            instantiation of "scalar_t blockReduce(scalar_t) [with scalar_t=float]" 
/home/namco/fairseq/fairseq/modules/lightconv_layer/lightconv_cuda_kernel.cu(238): here
            instantiation of "void lightconv_grad_wrt_weights_firstpass_short_kernel<FS,SB,padding_l,scalar_t>(const scalar_t *, const scalar_t *, int, int, int, int, int, float *) [with FS=3, SB=32, padding_l=1, scalar_t=double]" 
/home/namco/fairseq/fairseq/modules/lightconv_layer/lightconv_cuda_backward.cu(44): here

5 errors detected in the compilation of "/home/namco/fairseq/fairseq/modules/lightconv_layer/lightconv_cuda_kernel.cu".

Environment

• fairseq Version (e.g., 1.0 or main):main
• PyTorch Version (e.g., 1.0) 1.10
• OS (e.g., Linux):Ubuntu 20.04
• How you installed fairseq (pip, source):source
• Build command you used (if compiling from source):pip install --editable ./
• Python version:3.9.7
• CUDA/cuDNN version: Cuda compilation tools, release 11.3, V11.3.58 Build cuda_11.3.r11.3/compiler.29745058_0

[pixas]

Because you use a high version pytorch. You could change two places.

• in ../cuda_utils.cu, add two lines of code: #include <iostream> and #define SHFL_MASK 0xffffffff
• in lightconv_cuda.cpp, change AT_ASSERTM(x.type().is_cuda(), #x " must be a CUDA tensor") to AT_ASSERTM(x.device().is_cuda(), #x " must be a CUDA tensor")

After these modifications, I successfully build the kernel. I hope my successful experience can help you as well.

Environment

• fairseq Version: main
• PyTorch Version: 1.9
• OS: CentOS
• Python version: 3.7
• CUDA/cuDNN version: Cuda compilation tools, release 11.2, V11.2.152; Build cuda_11.2.r11.2/compiler.29618528_0

[Namco0816]

Thanks for your help! it works. Thank you!