The following code failed at compilation after codegen, tested with CUDA 10.0:
import tvm
from tvm import relay
x = relay.var('x', shape=(32, 128), dtype='float16')
func = relay.Function([x], relay.abs(x))
mod = tvm.IRModule.from_expr(func)
with tvm.transform.PassContext(opt_level=3):
with tvm.target.Target('cuda'):
lib = relay.build(mod)
Is this a bug? It looks like tvm codegen misses generating abs function for half precision.