Does pytorch supports fp16xfp16->fp32 matmul?

1

At times, matrix multiplication between fp16 and fp16 can result in overflow. Does the Pytorch Matmul kernel support fp16 input and fp32 output?