Enable including atomics and friends in TUs that do not support them. #1736
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🟩 CI Results [ Failed: 0 | Passed: 302 | Total: 302 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 28 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust | |
| CUDA Experimental |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
miscco
requested changes
May 14, 2024
| (return __atomic_load_n_cuda(__a->get(), static_cast<__memory_order_underlying_t>(__order), _Sco{});), | ||
| NV_IS_DEVICE, | ||
| (__atomic_is_not_supported_pre_sm_60();), |
There was a problem hiding this comment.
We might need to make this a template that returns __atomic_underlying_t<_Sto> because otherwise we might have no return in that function
There was a problem hiding this comment.
Or:
Suggested change
| (__atomic_is_not_supported_pre_sm_60();), | |
| (__atomic_is_not_supported_pre_sm_60(); return {};), |
There was a problem hiding this comment.
Comment applies to a lot more places below.
There was a problem hiding this comment.
I think we just should mark the undefined function [[noinline]], that should kill any potential for stupid warnings.
I dont think so. With us moving towards linker errors we should be able to use NV_IF_TARGET to test this within the current framework |
|
Agreed with @miscco's review. |
Sure, you'd just need to rewrite all of the current |
It depends a bit on the tests, but we have moved towards defining a single test function anyway, which makes this rather easy to do |
miscco
reviewed
May 29, 2024
| @@ -64,11 +64,17 @@ struct __atomic_storage | |||
| } | |||
| }; | |||
|
|
|||
| #if defined(_CCCL_CUDA_COMPILER) | |||
| extern "C" _CCCL_DEVICE void __atomic_is_not_supported_pre_sm_60(); | |||
| #endif | |||
There was a problem hiding this comment.
Suggested change
| #endif | |
| #endif // _CCCL_CUDA_COMPILER |
bernhardmgruber
approved these changes
Jun 27, 2024
| (return __atomic_load_n_cuda(__a->get(), static_cast<__memory_order_underlying_t>(__order), _Sco{});), | ||
| NV_IS_DEVICE, | ||
| (__atomic_is_not_supported_pre_sm_60();), |
There was a problem hiding this comment.
Comment applies to a lot more places below.
griwes
reviewed
Aug 1, 2024
| (return __atomic_load_n_cuda(__a->get(), static_cast<__memory_order_underlying_t>(__order), _Sco{});), | ||
| NV_IS_DEVICE, | ||
| (__atomic_is_not_supported_pre_sm_60();), |
There was a problem hiding this comment.
I think we just should mark the undefined function [[noinline]], that should kill any potential for stupid warnings.
| @@ -45,7 +45,7 @@ function(libcudacxx_add_public_header_test header) | |||
| target_compile_options(headertest_${header_name} PRIVATE ${headertest_warning_levels_device}) | |||
|
|
|||
| # Ensure that if this is an atomic header, we only include the right architectures | |||
There was a problem hiding this comment.
This comment is stale with these changes.
miscco
approved these changes
Aug 5, 2024
🟨 CI finished in 4h 11m: Pass: 99%/417 | Total: 3d 05h | Avg: 11m 11s | Max: 1h 00m | Hits: 94%/524401
|
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust | |
| CUDA Experimental | |
| pycuda |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
| +/- | pycuda |
🏃 Runner counts (total jobs: 417)
| # | Runner |
|---|---|
| 305 | linux-amd64-cpu16 |
| 61 | linux-amd64-gpu-v100-latest-1 |
| 28 | linux-arm64-cpu16 |
| 23 | windows-amd64-cpu16 |
|
When can we expect this to be merged? |
|
/ok to test |
🟨 CI finished in 7h 05m: Pass: 99%/417 | Total: 2d 09h | Avg: 8m 14s | Max: 38m 15s | Hits: 97%/526130
|
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust | |
| CUDA Experimental | |
| pycuda |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
| +/- | pycuda |
🏃 Runner counts (total jobs: 417)
| # | Runner |
|---|---|
| 305 | linux-amd64-cpu16 |
| 61 | linux-amd64-gpu-v100-latest-1 |
| 28 | linux-arm64-cpu16 |
| 23 | windows-amd64-cpu16 |
🟩 CI finished in 3d 00h: Pass: 100%/417 | Total: 2d 09h | Avg: 8m 12s | Max: 38m 15s | Hits: 97%/526135
|
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust | |
| CUDA Experimental | |
| pycuda |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
| +/- | pycuda |
🏃 Runner counts (total jobs: 417)
| # | Runner |
|---|---|
| 305 | linux-amd64-cpu16 |
| 61 | linux-amd64-gpu-v100-latest-1 |
| 28 | linux-arm64-cpu16 |
| 23 | windows-amd64-cpu16 |
There are still issues with the solution we've come up with here. It's possible to defer the linker error to runtime when the TU is compiled from PTX to SASS during JIT or with |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This removes most
#errorprotections from atomics headers. This allows them to be included in TUs where they previously would force a failure if the__CUDA_ARCH__list included an architecture that was not supported.Link time failures are emitted when compiling for architectures where the API is not supported.
For example:
results in:
closes: #1083
Checklist