Notifications posted on pull requests now carry a marker naming their
source, as an HTML comment on their first line, which is invisible when
rendered. When posting, 'ci-pages-notify.bash' finds its own earlier
notifications with the same marker on that pull request, and deletes them.
A pull request therefore holds a single comment per source, rather than
accumulating one per run.
The superseded comments are not lost entirely. Each notification carries a
one line summary linking its report, stashed in an HTML comment, and the
notification superseding it inherits that summary, together with the
history the superseded one held itself, into a collapsed 'Workflow history'
section. That section therefore lists all the earlier reports, stays flat
however many runs a pull request sees, and holds no entry for the report
shown right above it.
The 'pr-notification' artifact is now uploaded via the new reusable
'upload-pr-notification' action, which validates the content the reporting
scripts produced, and records the source key and the pull request number
alongside it.
Fixes#8037
- Merge reusable-rtlmeter-build.yml into reusable-build.yml, selected by
an 'install' flag: archive the installed Verilator for RTLMeter, or
the whole repository tree for the test jobs.
- Align reusable-test.yml input arguments with reusable-build.yml
- Pass build/test options as booleans instead of 0/1.
- Inline the single-entry build matrices, and hoist matrix keys that are
constant across a job.
- If any simulated cycle counts mismatch, report them and fail the job
- Keep "Elapsed" only in cppbuild
- Ignore "Elapsed" and "CPU" samples less than 30 seconds - too noisy
- Widen acceptable ranges on "Elapsed" and "Memory"
- Add badge legend
The intention was to run scheduled with max 2 parallel jobs, however,
due to a bug they were run with 7 just like PR runs. This didn't seem to
cause problems, so change it all to 7, except for the hierarchical,
which runs on 6, for a total of a maximum 20 parallel RTLMeter run jobs
per workflow run.
Add a script that prints the job matrix of cases to run with RTLMeter.
This enables running different sets based on parameters. It will run
everything on scheduled nightly runs, but only a subset on PRs (to reduce
time and variance due to short cases).
Also prints the job matrix in descending duration order, which
effectively makes GitHub actions schedule longest jobs first to reduce
total workflow latency.
Chores:
- Remove ci-ccache-maint. This has never been useful and is just cruft.
- Remove then unused CI_COMMIT
- Change job names so they come out nicer in the web views
- Make os-name input to reusable-build always explicit
Improvements:
- Have at most build-test job in progress per branch
- Cancel in-progress build-test jobs on PRs
- In forks (that is, not on 'verilator/verilator'), cancel any
in-progress build-test jobs on push to the branch
Runs RTLMeter on PRs if the PR has the 'pr: rtlmeter' label applied to
it. This should make it easy to selectively require the RTLMeter
workflows to pass on a PR that is potentially invasive, per our own
judgement.
The manual for the BLKANDNBLK warning describes that it is safe to
disable that error if the updated ranges are non-overlapping. This
however was not true (see the added t_nba_mixed_update* tests).
In this patch we change V3Delayed to use a new ShadowVarMasked
scheme for variables that have mixed blocking and non-blocking
updates (or the FlagUnique scheme for unpacked variables), which
is in fact safe to use when the updated parts are non-overlapping.
Furthermore, mixed assignments are safe as far as scheduling is
concerned if either:
- They are to independent parts (bits/members/etc) (with this patch)
- Or if the blocking assignment is in clocked (or suspendable) logic.
The risk in scheduling is a race between the Post scheduled NBA
commit, and blocking assignments in combinational logic, which might
order incorrectly.
The second point highlights that we can handle stuff like this safely,
which is sometimes used in testbenches:
```systemverilog
always @(posedge clk) begin
if ($time == 0) a = 0;
end
always @(posedge clk) begin
if ($time > 0) a <= 2;
end
````
The only dangerous case is:
```systemverilog
always @(posedge clk) foo[idx] <= val;
assign foo[0] = bar;
```
Whit this patch, this will still resolve fine at run-time if 'idx' is
never zero, but might resolve incorrectly if 'idx' is zero.
With the above in mind, the BLKANDNBLK warning is now only issued if:
- We can't prove that the assignments are to non-overlapping bits
- And the blocking assignment is in combinational logic
These are the cases that genuinely require user attention to resolve.
With this patch, there are no more BLKANDNBLK warnings in the RTLMeter
designs.
Fixes#6122.
Geza Lore
dependabot[bot]
Wilson Snyder
Geza Lore