Description
🚀 Feature
Provide a torch.minmax reduction operator that concurrently computes both the min and the max of the inputs.
Motivation
torch.min and torch.max often occur in pairs in pytorch code. It is desirable to have a concurrent min and max computation for efficiency.
Pitch
It is desirable to have a concurrent min and max computation for efficiency. This operator already exists internally as torch._aminmax and the proposal is to make this a public API
Alternatives
Build an FX optimizer (and/or a TorchScript JIT optimizer) that recognizes calls to torch.min and torch.max and replaces them with the optimized function.
Additional context
Computing min and max is often used together, e.g., in computing quantization parameters.