warp reduction is disabled for complex numbers since __shfl_xor_sync() doesn't support complex number. The supported data types are:
_
T can be int, unsigned int, long, unsigned long, long long, unsigned long long, float or double. With the cuda_fp16.h header included, T can also be half or half2. Similarly, with the cuda_bf16.h header included, T can also be __nv_bfloat16 or __nv_bfloat162.
_
To add additional support may try: exchange the real and imag part of the complex separately. Need check performance if we are doing shfl twice.
🚀 The feature, motivation and pitch
warp reduction is disabled for complex numbers since __shfl_xor_sync() doesn't support complex number. The supported data types are:
_
_
To add additional support may try: exchange the real and imag part of the complex separately. Need check performance if we are doing shfl twice.
https://docs.nvidia.com/cuda/cuda-c-programming-guide/#:~:text=T%20can%20be%20int%2C%20unsigned%20int%2C%20long%2C%20unsigned%20long%2C%20long%20long%2C%20unsigned%20long%20long%2C%20float%20or%20double.%20With%20the%20cuda_fp16.h%20header%20included%2C%20T%20can%20also%20be%20__half%20or%20__half2.%20Similarly%2C%20with%20the%20cuda_bf16.h%20header%20included%2C%20T%20can%20also%20be%20__nv_bfloat16%20or%20__nv_bfloat162.
Alternatives
No response
Additional context
No response