[GitHub] [tvm] wrongtest opened a new pull request #10460: [CodeGen][CUDA] use hrint for cuda half rounding

[GitBox]

wrongtest opened a new pull request #10460:
URL: https://github.com/apache/tvm/pull/10460


   When cuda c codegen generate `tir.round` for fp16, there is not function 
named `hround`, but `hrint` for cuda half aithmetics. 
https://docs.nvidia.com/cuda/cuda-math-api/group__CUDA__MATH____HALF__FUNCTIONS.html#group__CUDA__MATH____HALF__FUNCTIONS_1gbbf7a989130edcbdbfbb4730f61c79b1
   
   Testcase to reproduce:
   ```python
   import tvm
   from tvm import relay
   from tvm.ir.module import IRModule
   x = relay.var("x", shape=[16], dtype="float16")
   y = relay.round(x)
   f = relay.Function([x], y)
   m = IRModule.from_expr(f)
   m = relay.transform.InferType()(m)
   relay.build(m, target="cuda")
   ```
   


-- 
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.

To unsubscribe, e-mail: commits-unsubscr...@tvm.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org