dbr quant overhead[6/x]: remove unneeded isinstance checks in `op_convert_before_hook` (#68371)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68371
`isinstance` has some overhead, changing the code in `op_convert_before_hook`
to use the information calculate during tracing instead which is cheaper.
Test Plan:
```
python test/test_quantization.py TestQuantizeDBR
```
function level benchmarking
```
// MobileNetV2, 1x3x224x224 input, % of time spent by function during DBR convert
// before
op_convert_before_hook = 3.55%
isinstance = 1.62%
// after
op_convert_before_hook = 2.89%
```
Reviewed By: jerryzh168
Differential Revision: D32463757
Pulled By: vkuzo
fbshipit-source-id: 129efe9c279a41f55b8bfd09132e21c0066298a6
pytorch
b7d58745 - dbr quant overhead[6/x]: remove unneeded isinstance checks in `op_convert_before_hook` (#68371)
