[beetrees]

Typed GEPs are being removed from the rest of the compiler in #165146, and typed allocas were already removed, so change gpu_offload.rs to match. This PR also removes the constant zero GEPs (#t-compiler > GEP constant 0 in `gpu_offload.rs`).

r? @ZuseZ4

[ZuseZ4]

Thank you for the cleanups. I'm just returning from the llvm dev meeting, where I was promised by two vendors that they should provide us almost certainly with GPU CI. Until then, we'll still have to run tests by hand, unfortunately.

@sgasho since you should have an up-to-date build, can you test nvptx?

[ZuseZ4]

Can you please also update the now failing tests? Should be easy enough even without running them: control_flow.rs, scalar_host.rs, and slice_host.rs, test against the old typed alloca's. They're under ./tests/codegen-llvm/gpu_offload/

gpu_host.rs also tests for struct.__tgt_kernel_arguments (but fwiw, it failed on other things before).

[beetrees]

I've replaced all the allocas in the tests with the appropriate sizes for a 64-bit system with 8-aligned pointers. In addition to the files you mentioned, there was also one in region_host.rs.

[sgasho]

@sgasho since you should have an up-to-date build, can you test nvptx?

I tested the sample offload code on nvidia gpunode (RTX A2000 (sm_86)) and it worked 👍

[rustbot]

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.

[ZuseZ4]

@bors r+ rollup

[rust-bors]

📌 Commit cf4f14b has been approved by ZuseZ4

It is now in the queue for this repository.

[ZuseZ4]

@bors r-

Actually let me first try
@bors try jobs=test-various

[rust-bors]

This pull request was unapproved.

View changes since this unapproval

[rust-bors]

💔 Test for 31ff68f failed: CI. Failed job:

• try - test-various (web logs, enhanced plaintext logs)

[rust-log-analyzer]

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

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

[RUSTC-TIMING] jiff test:false 8.834
error[E0609]: no field `abi` on type `rustc_abi::Align`
   --> compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs:282:46
    |
282 |             let align = ty.default_align(cx).abi;
    |                                              ^^^ unknown field

For more information about this error, try `rustc --explain E0609`.
[RUSTC-TIMING] rustc_codegen_llvm test:false 4.477
error: could not compile `rustc_codegen_llvm` (lib) due to 1 previous error