Make help target and VCS/Verilator Cleanup #650
Conversation
|
Ok the Chipyard simulation side of this should be good to review. I checked VCS simulation and Verilator simulations. Im trying to check Hammer sim right now. |
vcs.mk
Outdated
| # gcc configuration/optimization | ||
| #---------------------------------------------------------------------------------------- | ||
| # -flto slows down compilation on small-memory and breaks on firesim-manager | ||
| CMODE := -O3 -fbranch-probabilities -march=native |
There was a problem hiding this comment.
Where did this come from? Is it known to always be better than the default? Is it optimizing for compilation time or runtime? Is that what we want?
There was a problem hiding this comment.
This was copying what Verilator did (though to make Verilator not segfault I just kept the -O3). It is optimizing the runtime. The default is nothing (aka use -0 with nothing else).
There was a problem hiding this comment.
Does it actually improve runtime?
There was a problem hiding this comment.
I dunno. I can run the bmark-tests on Rocket with and without the CMODE flags. Im trying to get Hammer sim tested 1st.
There was a problem hiding this comment.
With CMODE flags:
Compile: 36s
Runtime for all bmark tests (serially): 7m 36s
Without CMODE flags:
Compile: 35s
Runtime for all bmark tests (serially): 7m 34s
There was a problem hiding this comment.
While this doesn't result in a large runtime difference, I think it should stay if we end up consolidating flags across VCS and Verilator.
There was a problem hiding this comment.
Fine. It is worth noting that adding code that doesn't affect things is confusing, but if this proves that we can merge the flags in the future I'm fine with leaving it in.
|
Ok. I just got around to checking Hammer Sim. I got the simulator to compile successfully but I am getting other RISCV toolchains problems (causing wrong TSI commands to be sent). I think that is a problem with my environment though not the PR. |
|
I filed issues for the things I suggested be updated in new PRs. I believe the HammerSim changes should work should I'm ok merging once CI passes. We're not close to a release now, so we have plenty of time to dog food the changes anyways. |
Related issue: #562
Type of change: other enhancement
Impact: other
Release Notes
This is a split of PR #562 with Makefile change minus the Multi-threading support. It includes changes to VCS so that it aligns with the same changes done to Verilator.