[CPU] Support weight-compression dt s8

[usstq]

Details:

• FC with symmetrically quantized/compressed weight may have i8 (instead of u8) as weight data type (it saves the zero-point subtraction cost), this change added support to such weight dt.
• oneDNN fork PR: Support weight-compressed date type s8 oneDNN#249

Tickets:

• CVS-131042

[luweizhou2016]

so seems vnni2 doesn't support the compressed weight , or dynamic quantization?

[luweizhou2016]

@usstq , pls help to attach onednn side PR link.

[usstq]

@usstq , pls help to attach onednn side PR link.

It's openvinotoolkit/oneDNN#249

[luweizhou2016]

The code is okay for me. need @dmitry-gorokhov to take a look. Pls:

1. apply the comment changes.
2. change onednn commit message as https://wiki.ith.intel.com/pages/viewpage.action?pageId=2836774726. Refer the previous message.

[usstq]

@dmitry-gorokhov FYI, Here is the tricky logic added to fallback to u8/u4 with zp 128/8:

• we reorder layout of src into dst w/o type conversion
• after layout is made right, we shift i8/i4 value by adding zero-point to make them into u8/u4 range, as documented in https://oneapi-src.github.io/oneDNN/dev_guide_int8_computations.html?highlight=128#inputs-of-the-same-type-s8

[dmitry-gorokhov]

I think we have two follow-ups here:

1. Additional access to the weights (to perform 128/8 addition) adds some time to FIL metric. I suppose it is relatively not that high, but still worth to align with oneDNN team on possible merged solution with layout reorder
2. Signed to unsigned conversion will not be needed for LNL+ arch which incorporates VNNI with full precision combinations support.