Unsupported fusion blocks tensorcore pipeline

[yzhang93]

What happened?

When we add lowering configs on the matmul op to use tensorcore pipeline, we see very bad performance of the model because of the unsupported fusion with some ops (e.g. math.erf).

I've attached some example model IR that have bad performances. https://storage.googleapis.com/iree-shared-files/nod-perf/vivian/minilm.txt https://storage.googleapis.com/iree-shared-files/nod-perf/vivian/albert.txt https://storage.googleapis.com/iree-shared-files/nod-perf/vivian/funnel.txt

Steps to reproduce your issue

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

What component(s) does this issue relate to?

No response

Version information

No response

Additional context

No response

[MaheshRavishankar]

The title of the bug is misleading. Its not unsupported fusion, but unsupported vectorization?

[ThomasRaoux]

PR fix under review: https://github.com/iree-org/iree/pull/10460