Commit graph

9 commits

Author SHA1 Message Date
Vincent Ambo
d6c57eb957 refactor(tvix/eval): ensure VM operations fit in a single byte
This replaces the OpCode enum with a new Op enum which is guaranteed to fit in a
single byte. Instead of carrying enum variants with data, every variant that has
runtime data encodes it into the `Vec<u8>` that a `Chunk` now carries.

This has several advantages:

* Less stack space is required at runtime, and fewer allocations are required
  while compiling.
* The OpCode doesn't need to carry "weird" special-cased data variants anymore.
* It is faster (albeit, not by much). On my laptop, results consistently look
  approximately like this:

  Benchmark 1: ./before -E '(import <nixpkgs> {}).firefox.outPath' --log-level ERROR --no-warnings
  Time (mean ± σ):      8.224 s ±  0.272 s    [User: 7.149 s, System: 0.688 s]
  Range (min … max):    7.759 s …  8.583 s    10 runs

  Benchmark 2: ./after -E '(import <nixpkgs> {}).firefox.outPath' --log-level ERROR --no-warnings
  Time (mean ± σ):      8.000 s ±  0.198 s    [User: 7.036 s, System: 0.633 s]
  Range (min … max):    7.718 s …  8.334 s    10 runs

  See notes below for why the performance impact might be less than expected.
* It is faster while at the same time dropping some optimisations we previously
  performed.

This has several disadvantages:

* The code is closer to how one would write it in C or Go.
* Bit shifting!
* There is (for now) slightly more code than before.

On performance I have the following thoughts at the moment:

In order to prepare for adding GC, there's a couple of places in Tvix where I'd
like to fence off certain kinds of complexity (such as mutating bytecode, which,
for various reaons, also has to be part of data that is subject to GC). With
this change, we can drop optimisations like retroactively modifying existing
bytecode and *still* achieve better performance than before.

I believe that this is currently worth it to pave the way for changes that are
more significant for performance.

In general this also opens other avenues of optimisation: For example, we can
profile which argument sizes actually exist and remove the copy overhead of
varint decoding (which does show up in profiles) by using more adequately sized
types for, e.g., constant indices.

Known regressions:

* Op::Constant is no longer printing its values in disassembly (this can be
  fixed, I just didn't get around to it, will do separately).

Change-Id: Id9b3a4254623a45de03069dbdb70b8349e976743
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12191
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
2024-08-19 11:02:50 +00:00
Aspen Smith
d378111d77 feat(tvix/eval): Store hash in key of interner
Rather than storing the leaked allocation for the string as the key in
the interner, store the hash (using NoHashHashBuilder). I thought this
would improve performance, but it doesn't:

hello outpath           time:   [736.85 ms 748.42 ms 760.42 ms]
                        change: [-2.0754% +0.4798% +2.7096%] (p = 0.72 > 0.05)
                        No change in performance detected.

but it at least doesn't *hurt* performance, and it *does* avoid an
`unsafe`, so it's probably net good.

Change-Id: Ie413955bdb6f04b1f468f511e5ebce56e329fa37
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12049
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: aspen <root@gws.fyi>
2024-08-10 15:28:30 +00:00
Aspen Smith
b8f92a6d53 feat(tvix/eval): Forbid Hash{Map,Set}, use Fx instead
Per https://nnethercote.github.io/perf-book/hashing.html, we have
basically no reason to use the default hasher over a faster,
non-DoS-resistant hasher. This gives a nice perf boost basically for
free:

hello outpath           time:   [704.76 ms 714.91 ms 725.63 ms]
                        change: [-7.2391% -6.1018% -4.9189%] (p = 0.00 < 0.05)
                        Performance has improved.

Change-Id: If5587f444ed3af69f8af4eead6af3ea303b4ae68
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12046
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
Autosubmit: aspen <root@gws.fyi>
2024-08-07 12:38:40 +00:00
Ilan Joselevich
87d4b00ff5 chore(web/tvixbolt): Update to latest yew version
Change-Id: I8e89aea317f088142e8006b3a888ec6d28467b47
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12064
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Ilan Joselevich <personal@ilanjoselevich.com>
Tested-by: BuildkiteCI
2024-08-01 11:32:32 +00:00
Ilan Joselevich
7ca32d9f0b refactor(web/tvixbolt): buildRustPackage -> crate2nix
With the recent changes to crate2nix and buildRustCrate in nixpkgs it is
now possible to build tvixbolt via crate2nix like we do for other tvix
crates. We can reuse a lot of the customizations done in //tvix in
tvixbolt to avoid repeating ourselves.

A script for serving tvixbolt locally for testing purposes is also
available now through the .serve attribute of tvixbolt.

This change supersedes https://cl.tvl.fyi/c/depot/+/11821.

Change-Id: I4864df8b75aec73cf5fee2428924ed4cfbb32902
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11952
Tested-by: BuildkiteCI
Autosubmit: Ilan Joselevich <personal@ilanjoselevich.com>
Reviewed-by: flokli <flokli@flokli.de>
2024-07-05 20:19:39 +00:00
Florian Klink
afbe995aa9 chore(tvix): bump to data-encoding 2.6.0
Change-Id: I26af403bfa99e5d1cff24641a3dba908e1d06686
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11899
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
2024-07-01 06:57:01 +00:00
sterni
71435d9546 chore(3p/sources): Bump channels & overlays
* agenix has not been updated (https://github.com/ryantm/agenix/pull/241).
* wasm-bindgen bumped to 0.2.92 in Rust WASM projects
* 3p/lisp: port lispPackages from stable channel

  The Lisp package set we are using (`pkgs.lispPackages`) is the "old
  old" package set, whereas we were supposed to have been using
  `pkgs.lispPackages_new` (which is the "old new" package set).

  Either way we missed that train, and now there's a "new new" package
  set, but with a twist: Lisp packages in nixpkgs are now tied to
  their compilers, so the most generic way to access them seems to be
  from `pkgs.sbclPackages`.

  Switching to the packages from the "new new" package set doesn't
  work: Lots of stuff stops building if we just switch the sources
  over, and not everything is trivially fixable.

  For now we stay on the lispPackages from the stable channel. We need
  to look into the migration later.

  Or rewrite panettone.
* tvix: update generated protobuf files
* 3p/nixpkgs: pick trunk from stable channel; newer versions try to
  read files and do network I/O during build, but don't print enough
  details in error messages to figure out why.
* 3p/overlays: remove tdlib override (nixpkgs is currently new enough)
* 3p/overlays: override telega.el sources while updates are lagging in
  nixpkgs
* users/flokli/ipu6-softisp: update firmware paths, which NixOS now
  stores zstd-compressed.

Change-Id: I5a7a6c8b5d0688461bca92b9e6d654356d3a1cf1
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11711
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Reviewed-by: tazjin <tazjin@tvl.su>
2024-06-08 15:20:48 +00:00
Padraic-O-Mhuiris
5c3065b43a feat(tvix/eval): implement builtins.hashString
Implements md5, sha1, sha256 and sha512 using the related crates from
the RustCrypto hashes project (https://github.com/RustCrypto/hashes)

Change-Id: I00730dea44ec9ef85309edc27addab0ae88814b8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11005
Tested-by: BuildkiteCI
Reviewed-by: aspen <root@gws.fyi>
2024-02-23 16:04:35 +00:00
Vincent Ambo
782cfa9e33 chore(tvixbolt): move from //corp to //web
Assigning copyright to the TVL community (whatever that is), and
adding AGPL-3.0-or-later license.

I also cleaned up some of the stuff on the landing page.

Change-Id: I4dbca19406e00e5105fed50e8fb64e0fcca23e3a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11013
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
2024-02-23 08:08:07 +00:00
Renamed from corp/tvixbolt/Cargo.lock (Browse further)