add gemmx prameter support

[xiaoling-yi]

In this PR, we add the parameter support for gemmx. More specifically, we:

• set the gemmx pe array size, meshRow, tileSize, and meshCol as configurable parameters in the cfg file. Now we can set these parameters to different values, but please take care to modify the streamer parameters too! Just so you know, the transpose extension only works for 512 data width for readers.
• Note: the pipelined SIMD only works for meshRow * meshCol = 64 cases so if you set pe array size that doesn't have 64 output, please take care to set with_pipeline = false.
• modify software test and datagen.py to accommodate arbitrary pe array size. Now only fixed the matmul test. Conv test modification will be done later.
• relevant modification in snax-gen.py. The format tool did some format, so it looks like many changes but only very minimal changes actually.

[xiaoling-yi]

Would it be more useful though that there is a default setting for the GeMM config?

I don't know. Maybe it's better to ensure that people explicitly set the PE size? The streamer parameters should be matched with the PE array size

[rgantonio]

I don't know. Maybe it's better to ensure that people explicitly set the PE size? The streamer parameters should be matched with the PE array size

I'm not too picky so okay haha. Maybe we can add somekind of documentation on our documentation page for this 😄

[xiaoling-yi]

I don't know. Maybe it's better to ensure that people explicitly set the PE size? The streamer parameters should be matched with the PE array size

I'm not too picky so okay haha. Maybe we can add somekind of documentation on our documentation page for this 😄

We can!

[xiaoling-yi]

For this specific PR though:

Just so you know, the transpose extension only works for 512 data width for readers.

Note: the pipelined SIMD only works for meshRow * meshCol = 64 cases so if you set pe array size that doesn't have 64 output, please take care to set with_pipeline = false.

Can you please please please verify this somewhere in software instead of just this comment

I verified locally. Later relevant documentation will be added to push a warning on this.

Who sets TRANSPOSE_EXTENSION_ENABLE in the software makefile? Is there an automatic way this is set?

The streamer header file gen will set it if a transpose extension is instantiated.