[Issue]: AttributeError: module 'nvshmem.bindings' has no attribute 'fp16_sum_reduce_on_stream'

[foraxe]

How is this issue impacting you?

Application crash

Share Your Debug Logs

No response

Steps to Reproduce the Issue

fail when run nvshmem4py/perftest/reduction_on_stream.py

fp16

Command:

OMPI_ALLOW_RUN_AS_ROOT=1 \
 OMPI_ALLOW_RUN_AS_ROOT_CONFIRM=1 \
mpirun -np 4 -N 4 --bind-to none \
  python nvshmem4py/perftest/reduction_on_stream.py \
    -b 2M -e 2M -f 2 \
    -d fp16 -o sum \
    -w 5 -n 20

Former error:

AttributeError: module 'nvshmem.bindings' has no attribute 'fp16_sum_reduce_on_stream'. Did you mean: 'int16_sum_reduce_on_stream'?

bf16

Command:

OMPI_ALLOW_RUN_AS_ROOT=1 \
 OMPI_ALLOW_RUN_AS_ROOT_CONFIRM=1 \
mpirun -np 4 -N 4 --bind-to none \
  python nvshmem4py/perftest/reduction_on_stream.py \
    -b 2M -e 2M -f 2 \
    -d bf16 -o sum \
    -w 5 -n 20

Former error:

AttributeError: module 'nvshmem.bindings' has no attribute 'bf16_sum_reduce_on_stream'. Did you mean: 'int16_sum_reduce_on_stream'?

### Known Issue (Out of Scope)

The following dtypes are listed in the perftest argument parser choices but have no corresponding bindings:

* `ulonglong` — missing `ulonglong_*_reduce_on_stream` bindings
* `ptrdiff` — missing `ptrdiff_*_reduce_on_stream` bindings


### NVSHMEM Version

3.4.5+cuda 13.0

### Your platform details

_No response_

### Error Message & Behavior

_No response_

[foraxe]

fixed in #46

[benhg]

Thank you for the report. I will add better validation for the datatypes in nvshmem4py when I get a chance. Also, I think the type name for bfloat16 in nvshmem4py is bfloat16, not bf16. Could you try that?

[foraxe]

@benhg Thank you for the reply.

• The most type name in nvshmem4py is bfloat16. I have tested it and it is correct.
• But when run perftest, the utils.py defines the datatype with 'bf16':

        choices=[
            "int", "int32", "uint32", "int64", "uint64", "long", "longlong",
            "ulonglong", "size", "ptrdiff", "float", "double", "fp16", "bf16"
        ],

That's the reason why nvshmem4py/perftest/reduction_on_stream.py reports type error.

• Since the name bf16 is also widely used, it may be a good option to add dtype alias as the fix in [Bugfix][feat]: add support for fp16 and bf16 dtype aliases for collective operations. #46