Conversation
tianrui-wei
left a comment
tianrui-wei
left a comment
There was a problem hiding this comment.
Very impressive work overall :)
abejgonzalez
left a comment
abejgonzalez
left a comment
There was a problem hiding this comment.
I need to look at spiketile.* still but here is a temp. review.
| val masterNode = visibilityNode | ||
| val slaveNode = TLIdentityNode() | ||
|
|
||
| override def isaDTS = "rv64gcv_Zfh" |
There was a problem hiding this comment.
Is this true with the current CY toolchain?
There was a problem hiding this comment.
This isn't passed to the toolchain.
There was a problem hiding this comment.
I know it's not passed but right now we can't compile code outside gc right? I'm not sure it would make sense to restrict this to rv64gc for now.
There was a problem hiding this comment.
rv64gcv_Zfh is a superset of rv64gc. The cpu is always allowed to report that it has more capability than the software needs.
abejgonzalez
left a comment
abejgonzalez
left a comment
There was a problem hiding this comment.
LGTM. Skimmed the C++/V side and it seems fine for now (esp. since this is an experimental feature)
|
Now that this exists, maybe it makes sense to boot Linux in CI with this. |
This needs a faster firemarshal linux image, |
3 participants
This PR integrates a spike processor model as a tile into chipyard.
The spike processor model supports any RISC-V ISA configuration Spike supports, and dynamically links with spike-as-a-library.
The spike processor interfaces with chipyard through a cache-coherent Tilelink model.
Of course, this shouldn't be used for any performance-modelling.
It is useful for stress-testing MMIO devices, or accelerators.
Adding RoCC support would be a cool extension.
Also, atomics aren't guaranteed to be atomic.
TODO:
Related PRs / Issues:
Type of change:
Impact:
Contributor Checklist:
mainas the base branch?changelog:<topic>label?changelog:label?.conda-lock.ymlfile if you updated the conda requirements file?Please Backport?