Interface changes to enable per channel quant. (#37618)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37618
This does not do any actual changes. Just introduces API changes and
some data struct changes to hold vector of data for zero point and
scale.
Test Plan:
Via unittests inside qnnpack, i.e., q8gemm-test, q8conv/dwconv test.
fully-conntected-test, convolution-test.
PT's quantization tests.
Imported from OSS
Differential Revision: D21339039
fbshipit-source-id: 4a20cff9795a105ddd31482d1f1fe2b1dbe18997
