NVPTX: Drop support for old architectures and old ISAs#152443

kjetilkjeka · 2026-02-10T16:29:16Z

This is the implementation of this MCP

I believe it was said that no FCP was needed, but if that is incorrect then the FCP is anyway scheduled to finish in 2 days so it can in any case be merged then.

rustbot · 2026-02-10T16:29:21Z

Some changes occurred in src/doc/rustc/src/platform-support

cc @Noratrieb

rustbot · 2026-02-10T16:29:24Z

r? @jackh726

rustbot has assigned @jackh726.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

Owners of files modified in this PR: compiler
compiler expanded to 66 candidates
Random selection from 13 candidates

kjetilkjeka · 2026-02-10T16:30:19Z

r? @ZuseZ4

kjetilkjeka · 2026-02-10T16:31:27Z

CC: @workingjubilee @kulst

kjetilkjeka · 2026-02-10T16:32:04Z

@rustbot label +O-NVPTX

kjetilkjeka · 2026-02-10T16:33:06Z

compiler/rustc_target/src/target_features.rs


 const NVPTX_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[
    // tidy-alphabetical-start
-    ("sm_20", Unstable(sym::nvptx_target_feature), &[]),


@kulst I assume it's correct to just remove the features for the things we do not support anymore?

ZuseZ4 · 2026-02-11T21:13:34Z

Changing a target baseline officially doesn't require a blog post: https://forge.rust-lang.org/compiler/proposals-and-stabilization.html#other-kind-of-target-changes

But since it's afaik the first time that we raise it for a GPU target and since nvptx is Tier 2, would you mind to still write one, so that we could merge it simultaneously?

https://blog.rust-lang.org/2023/09/25/Increasing-Apple-Version-Requirements/ and https://blog.rust-lang.org/2025/08/19/demoting-x86-64-apple-darwin-to-tier-2-with-host-tools/ are two related examples, but there are more if you want some inspiration.

Just a more user-friendly, short version of your MCP. Something motivating the change, mentioning the new behaviour, and saying that we will likely raise it further in the future, would be enough.

CodesInChaos · 2026-02-12T10:50:12Z

So this drops support for Pascal (10 series, P100), while keeping support for Volta (V100) and Turing (16 and 20 series)? Perhaps including that information in the docs would be helpful.

rustbot · 2026-03-04T16:56:24Z

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

kjetilkjeka · 2026-03-04T17:01:30Z

With regards to blogpost, is something like this acceptable? rust-lang/blog.rust-lang.org#1810

So this drops support for Pascal (10 series, P100), while keeping support for Volta (V100) and Turing (16 and 20 series)? Perhaps including that information in the docs would be helpful.

Yes. I linked to the document by Nvidia listing all their GPUs as it will quickly become very verbose to explicitly list a lot of GPUs. Let me know if you think it's easier to get hold of this information with these changes.

Since the branch date for 1.95 was approaching I decided to instead wait for 1.96. The version is stated to be 1.96 in the docs and blogpost right now and we should have plenty of time for that branch date.

ZuseZ4 · 2026-03-29T23:05:40Z

src/doc/rustc/src/platform-support/nvptx64-nvidia-cuda.md

 While the compiler accepts `#[target_feature(enable = "ptx80", enable = "sm_89")]`, it is not supported, may not behave as intended, and may become erroneous in the future.

+## Minimum SM and PTX support by Rust version
+Old hardware architectures and PTX ISA versions are periodically dropped support for. This table shows the minimum supported versions per Rust version.


Nit: Might just be because I'm not native, but it reads a bit awkward to me. Does

Support for old hardware architectures and PTX ISA versions is periodically dropped.

sound better?

ZuseZ4 · 2026-03-29T23:56:57Z

compiler/rustc_codegen_ssa/src/target_features.rs

        },
    );

+    // This is needed to ensure that we don't use PTX ISA versions that we do not support


I just looked up the ptx-release-history link from this PR to verify that we don't need to handle sm_80. Can you make it slightly more explicit for the next reader by saying that sm_{70,72,75} would default to ptx60, which is unsupported, whereas sm_80 would default to ptx70, which is supported?

ZuseZ4 · 2026-03-30T00:09:08Z

compiler/rustc_codegen_ssa/src/target_features.rs

+            None | Some("sm_70") | Some("sm_72") | Some("sm_75")
+        )
+    {
+        rust_features.push((true, "ptx70"));


This one is added after the sorting above, so it will come behind all other ptx versions.
Normally the largest would be ordered to come last and naturally be selected, but I think here you would overwrite the user selection for these sm, if the user selected a higher ptx, or?

rustbot assigned jackh726 Feb 10, 2026

rustbot assigned ZuseZ4 and unassigned jackh726 Feb 10, 2026

rustbot added the O-NVPTX Target: the NVPTX LLVM backend for running rust on GPUs, https://llvm.org/docs/NVPTXUsage.html label Feb 10, 2026

kjetilkjeka commented Feb 10, 2026

View reviewed changes

kjetilkjeka changed the title ~~NVPTX: Drop support for old hw and old ISAs~~ NVPTX: Drop support for old architectures and old ISAs Feb 10, 2026

This comment has been minimized.

Sign in to view

kjetilkjeka force-pushed the nvptx_drop_support_old_hw_and_isa branch from af90fab to 192c99f Compare February 10, 2026 16:53

This comment has been minimized.

Sign in to view

kjetilkjeka force-pushed the nvptx_drop_support_old_hw_and_isa branch from 192c99f to a516a55 Compare February 10, 2026 17:44

This comment has been minimized.

Sign in to view

kjetilkjeka force-pushed the nvptx_drop_support_old_hw_and_isa branch from a516a55 to 5f88388 Compare February 10, 2026 17:55

This comment has been minimized.

Sign in to view

kjetilkjeka force-pushed the nvptx_drop_support_old_hw_and_isa branch 3 times, most recently from 4cbf1bb to 9026a42 Compare February 10, 2026 21:12

kjetilkjeka force-pushed the nvptx_drop_support_old_hw_and_isa branch from 9026a42 to e04ea32 Compare March 4, 2026 16:56

kjetilkjeka force-pushed the nvptx_drop_support_old_hw_and_isa branch from e04ea32 to dc1bd5a Compare March 4, 2026 16:57

NVPTX: Drop support for old hw and old ISAs

dc1bd5a

ZuseZ4 reviewed Mar 30, 2026

View reviewed changes

Uh oh!

Conversation

kjetilkjeka commented Feb 10, 2026

Uh oh!

rustbot commented Feb 10, 2026

Uh oh!

rustbot commented Feb 10, 2026

Uh oh!

kjetilkjeka commented Feb 10, 2026

Uh oh!

kjetilkjeka commented Feb 10, 2026

Uh oh!

kjetilkjeka commented Feb 10, 2026

Uh oh!

kjetilkjeka Feb 10, 2026

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

ZuseZ4 commented Feb 11, 2026

Uh oh!

CodesInChaos commented Feb 12, 2026

Uh oh!

rustbot commented Mar 4, 2026

Uh oh!

kjetilkjeka commented Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ZuseZ4 Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

ZuseZ4 Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

ZuseZ4 Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

kjetilkjeka commented Mar 4, 2026 •

edited

Loading