Stream: git-wasmtime

Topic: wasmtime / PR #13964 cranelift(x64): fold the u8×i8 dot-a...

Wasmtime GitHub notifications bot (Jul 24 2026 at 19:57):

darmie opened PR #13964 from darmie:x64-vpdpbusd-upstream to bytecodealliance:main:

Adds AVX-VNNI vpdpbusd (VEX.128.66.0F38.W0 50 /r) to the x64 backend plus a has_avx_vnni ISA flag with host detection, and folds the signed×unsigned i8 widening dot-accumulate tree into a single instruction. This is the x64 version of the aarch64 sdot fold.

The rule matches iadd(iadd_pairwise(... swiden(imul(swiden a, uwiden b)) ...), acc) and lowers it to x64_vpdpbusd acc b a: a sign-extended, b zero-extended, which is exactly what vpdpbusd computes. Only the mixed u8×i8 form is matched; signed×signed (the sdot pattern) would need vpdpbssd and is left alone. Unmatched patterns keep the widening chain.

Tested with a compile filetest (; check: vpdpbusd) and runtests (including a b > 127 case so the unsigned reading is observable), and executed on Alder Lake AVX-VNNI hardware. VEX.128 only; EVEX/256-bit are follow-ups.

Wasmtime GitHub notifications bot (Jul 24 2026 at 19:57):

darmie requested cfallin for a review on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 19:57):

darmie requested wasmtime-compiler-reviewers for a review on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 20:09):

darmie updated PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 20:09):

darmie requested wasmtime-core-reviewers for a review on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 20:27):

darmie updated PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:03):

github-actions[bot] added the label cranelift on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:03):

github-actions[bot] added the label cranelift:area:x64 on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:03):

github-actions[bot] added the label cranelift:meta on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:03):

github-actions[bot] added the label isle on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:03):

github-actions[bot] added the label wasmtime:api on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:03):

github-actions[bot] added the label wasmtime:config on PR #13964.

Wasmtime GitHub notifications bot (Jul 24 2026 at 22:04):

github-actions[bot] commented on PR #13964:

Subscribe to Label Action

cc @cfallin, @fitzgen

<details>
This issue or pull request has been labeled: "cranelift", "cranelift:area:x64", "cranelift:meta", "isle", "wasmtime:api", "wasmtime:config"

Thus the following users have been cc'd because of the following labels:

• cfallin: isle
• fitzgen: isle

To subscribe or unsubscribe from this label, edit the <code>.github/subscribe-to-label.json</code> configuration file.

Learn more.
</details>

Wasmtime GitHub notifications bot (Jul 24 2026 at 23:03):

github-actions[bot] commented on PR #13964:

Label Messager: wasmtime:config

It looks like you are changing Wasmtime's configuration options. Make sure to
complete this check list:

• [ ] If you added a new Config method, you wrote extensive documentation for
  it.

  <details>

  Our documentation should be of the following form:

  ```text
  Short, simple summary sentence.

  More details. These details can be multiple paragraphs. There should be
  information about not just the method, but its parameters and results as
  well.

  Is this method fallible? If so, when can it return an error?

  Can this method panic? If so, when does it panic?

  Example

  Optional example here.
  ```

  </details>

• [ ] If you added a new Config method, or modified an existing one, you
  ensured that this configuration is exercised by the fuzz targets.

  <details>

  For example, if you expose a new strategy for allocating the next instance
  slot inside the pooling allocator, you should ensure that at least one of our
  fuzz targets exercises that new strategy.

  Often, all that is required of you is to ensure that there is a knob for this
  configuration option in [wasmtime_fuzzing::Config][fuzzing-config] (or one
  of its nested structs).

  Rarely, this may require authoring a new fuzz target to specifically test this
  configuration. See [our docs on fuzzing][fuzzing-docs] for more details.

  </details>

• [ ] If you are enabling a configuration option by default, make sure that it
  has been fuzzed for at least two weeks before turning it on by default.

[fuzzing-config]: https://github.com/bytecodealliance/wasmtime/blob/ca0e8d0a1d8cefc0496dba2f77a670571d8fdcab/crates/fuzzing/src/generators.rs#L182-L194
[fuzzing-docs]: https://docs.wasmtime.dev/contributing-fuzzing.html

---

<details>

To modify this label's message, edit the <code>.github/label-messager/wasmtime-config.md</code> file.

To add new label messages or remove existing label messages, edit the
<code>.github/label-messager.json</code> configuration file.

Learn more.

</details>

Wasmtime GitHub notifications bot (Jul 25 2026 at 00:47):

darmie commented on PR #13964:

Thanks! A couple of clarifications on the auto-applied labels and checklist:

wasmtime:config / wasmtime:api: this PR doesn't add or change any public Config method or configuration option. The config.rs and engine.rs edits are internal host-feature-detection plumbing (detect_host_feature, and the flag→feature-name map in check_compatible_with_native_host), required whenever a new Cranelift x86 ISA setting is added (here has_avx_vnni) so a serialized module's ISA flags can be checked against the host at load time. Without it, loading any cached module errors with don't know how to test for target-specific flag "has_avx_vnni" at runtime. So the Config-documentation, Config-fuzzing, and default-enablement items don't apply.

Testing the new lowering: the vpdpbusd fold has a compile filetest (isa/x64/simd-vpdpbusd.clif) and an executed runtest (runtests/simd-vpdpbusd.clif), including a b > 127 case so the unsigned reading is observable; the runtest runs under the Intel-SDE CI job on emulated AVX-VNNI. One thing worth flagging directly: the assembler-x64 fuzz roundtrip skips vpdpbusd, because the bundled capstone (0.14) doesn't disassemble AVX-VNNI and so has no reference to compare against; the encoding is instead verified by that executed runtest. Happy to take a different approach to the disassembler gap if you'd prefer.

Wasmtime GitHub notifications bot (Jul 27 2026 at 21:06):

:thumbs_up: cfallin submitted PR review:

Thanks!

Wasmtime GitHub notifications bot (Jul 27 2026 at 21:06):

cfallin added PR #13964 cranelift(x64): fold the u8×i8 dot-accumulate tree to vpdpbusd (AVX-VNNI) to the merge queue

Wasmtime GitHub notifications bot (Jul 27 2026 at 22:11):

github-merge-queue[bot] removed PR #13964 cranelift(x64): fold the u8×i8 dot-accumulate tree to vpdpbusd (AVX-VNNI) from the merge queue

Wasmtime GitHub notifications bot (Jul 27 2026 at 22:15):

alexcrichton added PR #13964 cranelift(x64): fold the u8×i8 dot-accumulate tree to vpdpbusd (AVX-VNNI) to the merge queue

Wasmtime GitHub notifications bot (Jul 27 2026 at 23:02):

:check: alexcrichton merged PR #13964.

Wasmtime GitHub notifications bot (Jul 27 2026 at 23:02):

alexcrichton removed PR #13964 cranelift(x64): fold the u8×i8 dot-accumulate tree to vpdpbusd (AVX-VNNI) from the merge queue

Last updated: Jul 29 2026 at 05:03 UTC