Skip to content

Update to latest Spike version - #87

Closed
jmgorius wants to merge 2 commits into
chipsalliance:mainfrom
jmgorius:main
Closed

jmgorius wants to merge 2 commits into
chipsalliance:mainfrom
jmgorius:main

Conversation

@jmgorius

@jmgorius jmgorius commented Mar 2, 2026

Copy link
Copy Markdown
Contributor

Disable DTB discovery, which was introduced in riscv-software-src/riscv-isa-sim@2c94ea4

Disable DTB discovery, which was introduced in riscv-software-src/riscv-isa-sim@2c94ea4

@BilalAli10x BilalAli10x left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I verified this fix locally. Adding the false argument to the sim_t constructor is required for compatibility with the latest Spike. My riscv-vector-tests build is now passing Stage 1 and Stage 2 with this change.

@nadime15

Copy link
Copy Markdown
Contributor

Could you update the hash and pull the latest? Since then, two new vector extensions, Zvabd and Zvzip, have been added (not yet ratified) to Spike, along with a fix to the RVV Crypto group.

@jmgorius

Copy link
Copy Markdown
Contributor Author

Sure! I checked locally, and no other API changes seem to have occured.

@ksco

ksco commented Apr 1, 2026

Copy link
Copy Markdown
Collaborator

There are failures in the RV64_Zvl64 combo.

@jmgorius

jmgorius commented Apr 1, 2026

Copy link
Copy Markdown
Contributor Author

There are, but I have no clue why this particular check would fail when the others all succeed. Maybe someone with more experience with the project may have some insights?

@nadime15

nadime15 commented Apr 1, 2026

Copy link
Copy Markdown
Contributor

@jmgorius Unfortunately, I don’t have a straightforward fix to propose. However, if you have the time, I would suggest gradually replacing the current spike hash with older ones and checking which version still passes. This should help narrow down where things start to break.

In the Makefile, you can provide a PATTERN. For now, generating cases only for vmulh_vv should significantly speed up testing and make it easier to isolate the issue.

@jmgorius

Copy link
Copy Markdown
Contributor Author

After some bisecting, it turns out that this Spike commit is the culprit: riscv-software-src/riscv-isa-sim@d1e34d0

I am trying to figure out how to fix the issue.

nadime15 added a commit to nadime15/riscv-vector-tests that referenced this pull request May 5, 2026
Per V spec, vmulh.{vv,vx}, vmulhu.{vv,vx}, vmulhsu.{vv,vx}
and vsmul.{vv,vx} are not available at EEW=64 in Zve64*. Spike
commit d1e34d0 enforces this, which broke the v64x64 build (no
full V at VLEN=64).

Add an Option.HasV flag derived from the march string and skip
SEW=64 generation for these eight instructions when V is absent.

Surfaced in chipsalliance#87 while bumping the Spike version.

Additionally update Spike to latest commit.

Co-authored-by: Jean-Michel Gorius <jmgorius@users.noreply.github.com>
ksco pushed a commit that referenced this pull request May 6, 2026
Per V spec, vmulh.{vv,vx}, vmulhu.{vv,vx}, vmulhsu.{vv,vx}
and vsmul.{vv,vx} are not available at EEW=64 in Zve64*. Spike
commit d1e34d0 enforces this, which broke the v64x64 build (no
full V at VLEN=64).

Add an Option.HasV flag derived from the march string and skip
SEW=64 generation for these eight instructions when V is absent.

Surfaced in #87 while bumping the Spike version.

Additionally update Spike to latest commit.

Co-authored-by: Jean-Michel Gorius <jmgorius@users.noreply.github.com>
@jmgorius

jmgorius commented May 6, 2026

Copy link
Copy Markdown
Contributor Author

Closing, since the update has been taken care of in b30515e.

@jmgorius jmgorius closed this May 6, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants