Move shared offload globals and define per-kernel globals once #149788

Sa4dUs · 2025-12-08T21:57:26Z

This PR moves the shared LLVM global variables logic out of the offload intrinsic codegen and generates kernel-specific variables only ont he first call of the intrinsic.

r? @ZuseZ4

tracking:

Tracking Issue for GPU-offload #131513

Sa4dUs · 2025-12-08T22:18:33Z

compiler/rustc_codegen_llvm/src/base.rs

+            if cx.sess().opts.unstable_opts.offload.contains(&Offload::Enable)
+                && !cx.sess().target.is_like_gpu
+            {
+                cx.offload_globals.replace(Some(OffloadGlobals::declare(&cx)));


i'm a bit unsure about this location. we could also cache these globals and generate them on the first intrinsic call, but that felt like overloading intrinsic codegen a bit too much

i don't have a strong opinion though, so happy to go with whatever u think is best

I'd also prefer if the intrinsics stay simple and don't need global context about whether they're the first intrinsic or not.

The location seems fine, but can you leave a comment explaining that we have to leave a bunch of globals in the LLVM-IR host module? Similar to e.g. objective-c below. Otherwise lgtm.

ZuseZ4 · 2025-12-18T12:46:10Z

Thanks for the IR and code cleanup, the struct makes it much nicer.
Since this adds some entries to the fullCx, I'll run a

@bors try @rust-timer queue

I don't think it will have any impact, but then again, we thought the same when accidentally causing a regression with autodiff, so let's see.

Move shared offload globals and define per-kernel globals once

rust-bors · 2025-12-18T15:03:34Z

☀️ Try build successful (CI)
Build commit: b120fe9 (b120fe9e9f5cbf9dd3fe1c38c97b3b2d44dd94ee, parent: ed0006a7ba2dc8fabab8ea94d6f843886311b3c7)

rust-timer · 2025-12-18T15:44:49Z

Finished benchmarking commit (b120fe9): comparison URL.

Overall result: ❌ regressions - no action needed

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

@bors rollup=never
@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.1%	[1.1%, 1.1%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 481.444s -> 482.856s (0.29%)
Artifact size: 390.60 MiB -> 390.60 MiB (-0.00%)

ZuseZ4 · 2025-12-19T10:58:18Z

Benchmark seems noisy, and the Significance Factor 1.01 seems to agree, so let's ignore that.

@bors delegate+
r=me after adding a comment to base.rs (and squashing it into the first commit)

bors · 2025-12-19T10:58:22Z

✌️ @Sa4dUs, you can now approve this pull request!

If @ZuseZ4 told you to "r=me" after making some further change, please make that change, then do @bors r=@ZuseZ4

ZuseZ4 · 2025-12-19T14:29:06Z

@bors r+ rollup

Sa4dUs · 2025-12-19T14:29:09Z

@bors r=@ZuseZ4

bors · 2025-12-19T14:29:09Z

📌 Commit 04c2d2b has been approved by ZuseZ4

It is now in the queue for this repository.

bors · 2025-12-19T14:29:11Z

📌 Commit 04c2d2b has been approved by ZuseZ4

It is now in the queue for this repository.

Move shared offload globals and define per-kernel globals once This PR moves the shared LLVM global variables logic out of the `offload` intrinsic codegen and generates kernel-specific variables only ont he first call of the intrinsic. r? `@ZuseZ4`

Move shared offload globals and define per-kernel globals once This PR moves the shared LLVM global variables logic out of the `offload` intrinsic codegen and generates kernel-specific variables only ont he first call of the intrinsic. r? ``@ZuseZ4``

Move shared offload globals and define per-kernel globals once This PR moves the shared LLVM global variables logic out of the `offload` intrinsic codegen and generates kernel-specific variables only ont he first call of the intrinsic. r? ```@ZuseZ4```

Rollup of 7 pull requests Successful merges: - #149633 (Enable `outline-atomics` by default on AArch64 FreeBSD) - #149788 (Move shared offload globals and define per-kernel globals once) - #149989 (Improve filenames encoding and misc) - #150012 (rustc_target: Add `efiapi` ABI support for LoongArch) - #150116 (layout: Store inverse memory index in `FieldsShape::Arbitrary`) - #150151 (Destabilise `target-spec-json`) - #150159 (Split eii macro expansion code) r? `@ghost` `@rustbot` modify labels: rollup

Move shared offload globals and define per-kernel globals once This PR moves the shared LLVM global variables logic out of the `offload` intrinsic codegen and generates kernel-specific variables only ont he first call of the intrinsic. r? `@ZuseZ4` tracking: - rust-lang#131513

rustbot assigned ZuseZ4 Dec 8, 2025

Sa4dUs force-pushed the offload-cleanup branch from 9de91f2 to b357e0c Compare December 8, 2025 22:00

Sa4dUs commented Dec 8, 2025

View reviewed changes

ZuseZ4 added the F-gpu_offload `#![feature(gpu_offload)]` label Dec 10, 2025

ZuseZ4 mentioned this pull request Dec 10, 2025

Tracking Issue for GPU-offload #131513

Open

5 tasks

This comment has been minimized.

Sign in to view

rust-bors bot added a commit that referenced this pull request Dec 18, 2025

Auto merge of #149788 - Sa4dUs:offload-cleanup, r=<try>

b120fe9

Move shared offload globals and define per-kernel globals once

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Dec 18, 2025

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Dec 18, 2025

Sa4dUs added 3 commits December 19, 2025 13:27

Split runtime global logic and cache kernel specific one

3e4944d

Remove outdated comment

8bafb63

Remove region_id unnamed attr

04c2d2b

Sa4dUs force-pushed the offload-cleanup branch from 37df9e8 to 04c2d2b Compare December 19, 2025 12:37

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Dec 19, 2025

matthiaskrgr mentioned this pull request Dec 19, 2025

Rollup of 6 pull requests #150167

Closed

matthiaskrgr mentioned this pull request Dec 19, 2025

Rollup of 7 pull requests #150168

Closed

matthiaskrgr mentioned this pull request Dec 19, 2025

Rollup of 6 pull requests #150177

Open

Uh oh!

Move shared offload globals and define per-kernel globals once #149788

Are you sure you want to change the base?

Move shared offload globals and define per-kernel globals once #149788

Conversation

Sa4dUs commented Dec 8, 2025 • edited by ZuseZ4 Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Sa4dUs Dec 8, 2025

Choose a reason for hiding this comment

Uh oh!

ZuseZ4 Dec 19, 2025

Choose a reason for hiding this comment

Uh oh!

ZuseZ4 commented Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

rust-bors bot commented Dec 18, 2025

Uh oh!

This comment has been minimized.

rust-timer commented Dec 18, 2025

Overall result: ❌ regressions - no action needed

Instruction count

Max RSS (memory usage)

Cycles

Binary size

Uh oh!

ZuseZ4 commented Dec 19, 2025

Uh oh!

bors commented Dec 19, 2025

Uh oh!

ZuseZ4 commented Dec 19, 2025

Uh oh!

Sa4dUs commented Dec 19, 2025

Uh oh!

bors commented Dec 19, 2025

Uh oh!

bors commented Dec 19, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Sa4dUs commented Dec 8, 2025 •

edited by ZuseZ4

Loading

ZuseZ4 commented Dec 18, 2025 •

edited

Loading