Error in vector stride? about hwpe-mac-engine HOT 5 CLOSED

I've constructed a small example that demonstrates the weird behavior well.
hwme.zip

Output is:

a:     2 b:     3 c:     0 d:    10
a:     2 b:     2 c:     0 d:    10
a:     2 b:     3 c:     0 d:    10
a:     2 b:     3 c:     0 d:    10
a:     2 b:     3 c:     0 d:     4
a:     2 b:     3 c:     0 d:    10
a:     2 b:     3 c:     0 d:    10

In the fifth row, the 4 should be a 10 instead.

from hwpe-mac-engine.

Hi @lstrz, I will gladly check your example when possible. It is possible that some parts of the accelerator, especially the datapath, contain a few bugs -- this specific accelerator is provided as an example so it's not thoroughly tested. I am more confident for what concerns the hwpe-stream and hwpe-ctrl IPs, although the versions used here could be (very slightly) outdated.

I'll let you know.

from hwpe-mac-engine.

Sounds good! I wasn't sure if I was doing something wrong or not, so while at it, I posted the example for narrowing down potential bugs in IPs you care about, as I have little experience in digital design.

from hwpe-mac-engine.

Ok I found and fixed the issue, which has always been there. One of the common cases (two simultaneous valid handshakes in different streams) was not properly checked in the datapath.
It was not previously exposed (some other change in the platform must have brought it out by slightly changing timings), but actually it impacted also my own test. It is now fixed in commit 087a7f3 of the HWME module, which you should get if you update the platform IPs with update-ips.

I also took the occasion to update hwpe-stream and hwpe-ctrl. They are not involved in this bug, but the new version should be anyways more robust, I've been recently working on a lot of testing on these versions.

from hwpe-mac-engine.

Perfect. Thank you very much!

from hwpe-mac-engine.