offload: remove device.bin and pass device bitcode directly by sgasho · Pull Request #164455 · rust-lang/rust

sgasho · 2026-10-17T05:52:24Z

Since #162309 removed the clang-linker-wrapper step, nothing reads device.bin
anymore, and its triple/arch are already in the bitcode.

So, as the first step, I removed steps generating and extracting device.bin, instead we're going to use the device bitcode.

As a next step, the host pass could take the linked device output(.ptx for NVPTX, .elf for AMDGPU) instead, which
would remove the second codegen. I'd like to do that in a separate PR.

I'll update rustc-dev-guide in a follow-up PR in their repo.

I verified that this change works by running offload sample code on UofT gpunode.

r? ZuseZ4

ZuseZ4 · 2026-10-17T23:33:54Z

Iterating over all functions to get a target-cpu doesn't seem great, but should work for now and become unnecessary with the follow-up PR, so lgtm. Thanks!

@bors r+ rollup

rust-bors · 2026-10-17T23:33:57Z

📌 Commit 80d95cc has been approved by ZuseZ4

It is now in the queue for this repository.

offload: remove device.bin and pass device bitcode directly related: #164057 Since #162309 removed the `clang-linker-wrapper` step, nothing reads `device.bin` anymore, and its triple/arch are already in the bitcode. So, as the first step, I removed steps generating and extracting device.bin, instead we're going to use the device bitcode. As a next step, the host pass could take the linked device output(`.ptx` for NVPTX, `.elf` for AMDGPU) instead, which would remove the second codegen. I'd like to do that in a separate PR. I'll update rustc-dev-guide in a follow-up PR in their repo. I verified that this change works by running offload sample code on UofT gpunode. r? ZuseZ4

rust-log-analyzer · 2026-10-18T04:28:17Z

The job test-aarch64-msvc-1 failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

---- [run-make] tests\run-make\clear-error-blank-output stdout ----

error: rmake recipe failed to complete
status: exit code: 101
command: "C:\\a\\rust\\rust\\build\\aarch64-pc-windows-msvc\\test\\run-make\\clear-error-blank-output\\rmake.exe"
stdout: none
--- stderr -------------------------------

thread 'main' (4436) panicked at C:\a\rust\rust\tests\run-make\clear-error-blank-output\rmake.rs:11:64:
called `Result::unwrap()` on an `Err` value: Os { code: 232, kind: BrokenPipe, message: "The pipe is being closed." }
---
Currently active steps:
test::RunMake { test_compiler: Compiler { stage: 2, host: aarch64-pc-windows-msvc, forced_compiler: false }, target: aarch64-pc-windows-msvc } at src\bootstrap\src\core\build_steps\test.rs:2108
test::Compiletest { test_compiler: Compiler { stage: 2, host: aarch64-pc-windows-msvc, forced_compiler: false }, target: aarch64-pc-windows-msvc, mode: run-make, suite: "run-make", path: "tests/run-make", compare_mode: None } at src\bootstrap\src\core\build_steps\test.rs:2108
Build completed unsuccessfully in 1:59:49
make: *** [Makefile:115: ci-msvc-py] Error 1
  local time: Sat Oct 17 21:27:48 PDT 2026
  network time: Sun, 18 Oct 2026 04:27:48 GMT
##[error]Process completed with exit code 2.
##[group]Run echo "disk usage:"
echo "disk usage:"

rust-bors · 2026-10-18T04:29:14Z

💔 Test for 42d559d failed: CI. Failed job:

auto - test-aarch64-msvc-1 (web logs, enhanced plaintext logs)

ZuseZ4 · 2026-10-18T04:32:35Z

spurious

@bors r+ rollup

rust-bors · 2026-10-18T04:32:38Z

📌 Commit 80d95cc has been approved by ZuseZ4

It is now in the queue for this repository.

A failed build status on this PR was cleared due to the approval.

rust-bors · 2026-10-18T07:24:45Z

☀️ Test successful - CI
Approved by: ZuseZ4
Duration: 2h 51m 55s
Pushing 33295d7 to main...

github-actions · 2026-10-18T07:28:15Z

What is this?

This is an experimental post-merge analysis report that shows differences in test outcomes between the merged PR and its parent PR.

Comparing 02fb0a1 (parent) -> 33295d7 (this PR)

Test differences

Show 2 test diffs

2 doctest diffs were found. These are ignored, as they are noisy.

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard 33295d734447c7a2c22c44190f11bce6164a035a --output-dir test-dashboard

And then open test-dashboard/index.html in your browser to see an overview of all executed tests.

Job duration changes

dist-i686-linux: 1h 11m -> 1h 57m (+63.4%)
dist-arm-linux-gnueabi: 1h -> 1h 36m (+59.6%)
dist-various-2: 33m 38s -> 51m 24s (+52.8%)
test-x86_64-rust-for-linux: 34m 32s -> 52m 29s (+52.0%)
dist-aarch64-apple: 1h 38m -> 2h 25m (+48.4%)
dist-powerpc64-linux-gnu: 1h 6m -> 1h 36m (+45.6%)
dist-ohos-x86_64: 56m 46s -> 1h 20m (+41.7%)
test-x86_64-msvc-ext1: 2h 11m -> 1h 19m (-39.7%)
test-i686-gnu-nopt-1: 1h 22m -> 1h 53m (+38.9%)
test-x86_64-gnu-debug: 2h 2m -> 1h 15m (-38.4%)

How to interpret the job duration changes?

Job durations can vary a lot, based on the actual runner instance
that executed the job, system noise, invalidated caches, etc. The table above is provided
mostly for t-infra members, for simpler debugging of potential CI slow-downs.

rust-timer · 2026-10-18T08:10:45Z

Finished benchmarking commit (33295d7): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

This perf run didn't have relevant results for this metric.

Max RSS (memory usage)

Results (secondary -3.5%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-3.5%	[-4.4%, -2.0%]	4
All ❌✅ (primary)	-	-	0

Cycles

Results (secondary 5.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	5.0%	[5.0%, 5.0%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 489.75s -> 488.147s (-0.33%)
Artifact size: 409.00 MiB -> 408.98 MiB (-0.00%)

offload: pass the linked device image to the host pass Closes rust-lang#164057 Follow-up to: rust-lang#164455 The host pass now takes the linked device output (`<crate>.ptx` for NVPTX, `<crate>.elf` for AMDGPU) instead of the device bitcode. The device crate must now be built as a cdylib. rustc only passes --export-symbol to llvm-bitcode-linker for cdylibs, so a bin would lose its kernels to internalization. I had to add [lib] in Cargo.toml ```toml [lib] name = "sample_offload" path = "src/main.rs" crate-type = ["cdylib"] ``` Without it, the runtime fails at startup ``` PluginInterface error: Failure to init kernel: "requested object was not found in the binary image" error in cuModuleGetFunction('_RNvC14sample_offload6kernel'): named symbol not found omptarget error: Failed to load kernel _RNvC14sample_offload6kernel ``` I’m not sure whether bin device crates should export their kernels too, so I left the linker untouched for now. The second commit fixes another runtime error without `--emit=llvm-bc`/`llvm-ir`. (my understanding is that we should not have to set these after this PR, because no device.bc anymore) The “already wrapped” check in handle_offload relied on a value name, which LLVM drops unless --emit=llvm-bc/llvm-ir is given, so the kernel got wrapped twice. I verified on UofT nvidia gpunode (amdgpunodes was not available for my account). r? ZuseZ4

Rollup merge of #164896 - sgasho:offload-cleanup-2, r=ZuseZ4 offload: pass the linked device image to the host pass Closes #164057 Follow-up to: #164455 The host pass now takes the linked device output (`<crate>.ptx` for NVPTX, `<crate>.elf` for AMDGPU) instead of the device bitcode. The device crate must now be built as a cdylib. rustc only passes --export-symbol to llvm-bitcode-linker for cdylibs, so a bin would lose its kernels to internalization. I had to add [lib] in Cargo.toml ```toml [lib] name = "sample_offload" path = "src/main.rs" crate-type = ["cdylib"] ``` Without it, the runtime fails at startup ``` PluginInterface error: Failure to init kernel: "requested object was not found in the binary image" error in cuModuleGetFunction('_RNvC14sample_offload6kernel'): named symbol not found omptarget error: Failed to load kernel _RNvC14sample_offload6kernel ``` I’m not sure whether bin device crates should export their kernels too, so I left the linker untouched for now. The second commit fixes another runtime error without `--emit=llvm-bc`/`llvm-ir`. (my understanding is that we should not have to set these after this PR, because no device.bc anymore) The “already wrapped” check in handle_offload relied on a value name, which LLVM drops unless --emit=llvm-bc/llvm-ir is given, so the kernel got wrapped twice. I verified on UofT nvidia gpunode (amdgpunodes was not available for my account). r? ZuseZ4

rustbot assigned ZuseZ4 Oct 17, 2026

offload: remove device.bin to simplify OffloadWrapper

80d95cc

rust-cloud-vms Bot force-pushed the offload-cleanup-1 branch from 2c08c13 to 80d95cc Compare October 17, 2026 06:11

rust-bors Bot 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 Oct 17, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Oct 18, 2026

rust-bors Bot 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 Oct 18, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot added merged-by-bors This PR was explicitly merged by bors. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Oct 18, 2026

rust-bors Bot merged commit 33295d7 into rust-lang:main Oct 18, 2026
15 checks passed

rustbot added this to the 1.101.0 milestone Oct 18, 2026

This was referenced Oct 19, 2026

update offload usage rust-lang/rustc-dev-guide#3078

Merged

offload: pass the linked device image to the host pass #164896

Merged

Navigation Menu

Repository navigation

offload: remove device.bin and pass device bitcode directly - #164455

offload: remove device.bin and pass device bitcode directly#164455
rust-bors[bot] merged 1 commit into
rust-lang:mainfrom
sgasho:offload-cleanup-1

sgasho commented Oct 17, 2026 •

edited by rustbot

Loading

Uh oh!

ZuseZ4 commented Oct 17, 2026 •

edited

Loading

Uh oh!

rust-bors Bot commented Oct 17, 2026

Uh oh!

This comment has been minimized.

rust-log-analyzer commented Oct 18, 2026

Uh oh!

rust-bors Bot commented Oct 18, 2026

Uh oh!

ZuseZ4 commented Oct 18, 2026

Uh oh!

rust-bors Bot commented Oct 18, 2026

Uh oh!

This comment has been minimized.

rust-bors Bot commented Oct 18, 2026

Uh oh!

Uh oh!

github-actions Bot commented Oct 18, 2026

Uh oh!

rust-timer commented Oct 18, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

sgasho commented Oct 17, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ZuseZ4 commented Oct 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rust-bors Bot commented Oct 17, 2026

Uh oh!

This comment has been minimized.

rust-log-analyzer commented Oct 18, 2026

Uh oh!

rust-bors Bot commented Oct 18, 2026

Uh oh!

ZuseZ4 commented Oct 18, 2026

Uh oh!

rust-bors Bot commented Oct 18, 2026

Uh oh!

This comment has been minimized.

rust-bors Bot commented Oct 18, 2026

Uh oh!

Uh oh!

github-actions Bot commented Oct 18, 2026

Test differences

Job duration changes

Uh oh!

rust-timer commented Oct 18, 2026

Overall result: no relevant changes - no action needed

Instruction count

Max RSS (memory usage)

Cycles

Binary size

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

sgasho commented Oct 17, 2026 •

edited by rustbot

Loading

ZuseZ4 commented Oct 17, 2026 •

edited

Loading