Bump FPU to pulp-v0.1.1

[lucabertaccini]

Bump FPU to pulp-v0.1.1 which supports:

• Low and mixed-precision SDOTP operations (with stochastic rounding support)
• Compressed vector compare results
• FP8alt (1, 4, 3) format

CVA6 does not support the features listed above which are thus disabled in the CVA6 FPU instance.

[niwis]

Thanks a lot, LGTM. I have just two general questions:

1. Will these changes be upstreamed to openhwgroup:master in the near future, or is this a more permanent solution?
2. Can you confirm that adding OpenE906 is necessary for this PR? Maybe we need to double-check the license implications for this repo.

[lucabertaccini]

Hi @niwis:

1. The integration into openhwgroup:master has not been discussed yet
2. The DivSqrt unit from OpenE906 is not relevant for this PR (it is supported only by FP32-only CVFPU instances), but it is aligned with recent OpenHW developments (develop branch https://github.com/openhwgroup/cvfpu). In CVFPU, two license-related files have been added during the integration of this alternative DivSqrt module (https://github.com/openhwgroup/cvfpu/blob/develop/LICENSE.apache and https://github.com/openhwgroup/cvfpu/blob/develop/README.license.md)

[niwis]

@lucabertaccini

1. Okay, in that case I think we should make sure to regularly update the pulp-platform:pulp branch to include any relevant changes from openhwgroup:master so that they can be easily integrated into pulp-CVA6 (and other depandants)
2. would it be an option then to exclude the vendor directory in openhwgroup_cvfpu.vendor.hjson? To avoid unnecessary inflation of the code base (see this example)

[niwis]

I think we do not need to add vendor again... I will go ahead and remove the util dir