Add support for SimSYCL as a SYCL implementation #871

fknorr · 2024-02-17T17:59:28Z

This adds SimSYCL as a supported SYCL implementation, complete with CMake integration and an exclude-list for the "fast" conformance build.

This should be merged after #870 #872 and #873, because the CTS it will not actually compile for many of the non-excluded cases unless those bugs are addressed.

AlexeySachkov · 2024-02-19T15:50:08Z

cmake/AddSYCLExecutable.cmake

 if (NOT ${SYCL_IMPLEMENTATION} IN_LIST KNOWN_SYCL_IMPLEMENTATIONS)
    message(FATAL_ERROR
        "The SYCL CTS requires specifying a SYCL implementation with "
-        "-DSYCL_IMPLEMENTATION=[Intel_SYCL,DPCPP;hipSYCL]")
+        "-DSYCL_IMPLEMENTATION=[Intel_SYCL,DPCPP;hipSYCL;SimSYCL]")


I wonder if we should just put KNOWN_SYCL_IMPLEMENTATIONS directly into the message instead of modifying it every time we change list of known implementations?

Perhaps CTS_GRADE_SYCL_IMPLEMENTATION or something better, because we know quite many other implementations. :-)

AlexeySachkov · 2024-02-19T15:51:25Z

tests/event/event.cpp

+#if SYCL_CTS_COMPILING_WITH_SIMSYCL
+    SKIP("SimSYCL does not implement asynchronous execution.");
+#endif


I think that from the conformance point of view the better to use DISABLED_FOR_TEST_CASE macro so the test appears as failed.

I'm not sure this is even non-conformant. IIRC an implementation only needs to guarantee forward progress on an event when wait is called on it, which does not happen in any of these test cases.

I don't think this test requires a guarantee of forward progress, does it? It seems like it is just calling event::get_wait_list and making sure the list contains the direct dependencies of a pending command. There is no expectation that the command has completed when get_wait_list is called.

I agree, I don't understand why SimSYCL couldn't implement these features.

You're right, I read this code the wrong way! It needs the host task to launch asynchronously in the background at least once wait() is called, and then start running once the delayed_host_event is signalled. SimSYCL is fully synchronous and does not overlap command groups with the main thread, so the host task unconditionally deadlocks. This will probably never work. I changed the SKIP to FAIL to account for this.

tests/usm/CMakeLists.txt

psalz · 2024-02-21T13:18:54Z

Please also document SimSYCL in README.md and docs/README.adoc!

cmake/AdaptSimSYCL.cmake

psalz · 2024-02-21T14:08:05Z

Ideally we should also include SimSYCL in our CI setup. I think it should be relatively straightforward to add, analogously to hipSYCL/AdaptiveCpp.

cmake/AdaptSimSYCL.cmake

keryell · 2024-09-10T19:26:34Z

How do we move on this?

fknorr · 2024-12-18T13:17:41Z

Rebased onto #1011 and #874 and #872 , and now mentions SimSYCL in docs.

fknorr · 2025-01-08T11:55:36Z

I've added a CI step for a SimSYCL build, but creating the docker image needs some secrets. Not sure how to proceed on this.

psalz · 2025-01-08T12:24:20Z

I've added a CI step for a SimSYCL build, but creating the docker image needs some secrets. Not sure how to proceed on this.

I've launched a manual run to create the image here. It went through, but it looks like there's some issue while building the CTS itself!

fknorr · 2025-01-08T14:02:54Z

Thanks, it's building now!

keryell

That looks good after merge conflict resolution.

…ed_ptr<T[]>

atomicity_with_host_code requires a memory_scope argument (always `system`). atomic_ref<float> has no operator++ and must use fetch_add() instead.

tomdeakin · 2025-01-23T16:51:23Z

The WG discussed this today - can you resolve the CI failure please. Thanks!

bader · 2025-01-23T16:54:40Z

@fknorr, you need fix the compilation by updating SYCL_LANGUAGE_VERSION in SimSYCL implementation. We recently updated the test to verify the version value - #989.

fknorr · 2025-01-23T17:19:56Z

@bader I've turned these into runtime checks (2cf5cc4) so we don't need to re-build containers for now. I don't feel like the checks actually benefitted from being static assertions, please correct me if I'm wrong!

@AlexeySachkov I've drive-by fixed two inconsistencies (3dc17f6, 110c739) that caused a lot of warnings with GCC.

bader · 2025-01-23T17:25:07Z

@bader I've turned these into runtime checks (2cf5cc4) so we don't need to re-build containers for now. I don't feel like the checks actually benefitted from being static assertions, please correct me if I'm wrong!

Usually, the earlier you detect the problem the better. I.e. compile-time checks usually preferred over the run-time checks. I suggest we keep the check compile time but disable it for SimSYCL implementation. We should enable the check for SimSYCL with the implementation update. Does it sound okay to you?

fknorr · 2025-01-23T17:27:07Z

@tomdeakin Note that this PR is based on #1011, so we need to figure out what the correct behavior is supposed to be and implement it in SimSYCL before this PR can proceed (at least with the current filter list).

@bader Since we need to update SimSYCL ^ now anyway I'm going to implement the new behavior and drop that commit.

AlexeySachkov · 2025-01-24T12:41:49Z

@bader I've turned these into runtime checks (2cf5cc4) so we don't need to re-build containers for now. I don't feel like the checks actually benefitted from being static assertions, please correct me if I'm wrong!

See also discussion here. What we can do is switch catch2 static assertions to be runtime-only for the purpose of pre-commit in this repo, but leave them as static assertion in every other configuration (i.e. certification run and local development).

@AlexeySachkov I've drive-by fixed two inconsistencies (3dc17f6, 110c739) that caused a lot of warnings with GCC.

Thanks, warnings cleanup is always appreciated!

bader · 2025-01-24T16:24:09Z

What we can do is switch catch2 static assertions to be runtime-only for the purpose of pre-commit in this repo, but leave them as static assertion in every other configuration (i.e. certification run and local development).

I don't think this is a good idea. Technically, you suggest disabling all "static assertions" checks in pre-commit as CI do not run compiled tests (i.e. there is no run-time). Disabling either the compilation of the whole test via CMake filters or specific test cases via macro is preferable as it provides fine-grain control over tested functionality.

bader · 2025-01-24T22:48:06Z

tests/header/header_test.cpp

+  CHECK(std::is_same_v<decltype(SYCL_LANGUAGE_VERSION), long>);
+  CHECK(SYCL_LANGUAGE_VERSION == 202012L);


@fknorr, please, revert this change.
You can disable these checks for SimSYCL if needed.

psalz

LGTM! There are still quite a lot compiler warnings left, but it looks to me like most/all of those are legitimate issues with the CTS, and not SimSYCL's fault. We'll have to look into these separately I guess.

fknorr requested a review from a team as a code owner February 17, 2024 17:59

AlexeySachkov reviewed Feb 19, 2024

View reviewed changes

psalz reviewed Feb 21, 2024

View reviewed changes

tests/usm/CMakeLists.txt Outdated Show resolved Hide resolved

psalz reviewed Feb 21, 2024

View reviewed changes

cmake/AdaptSimSYCL.cmake Outdated Show resolved Hide resolved

psalz mentioned this pull request Feb 21, 2024

Rename hipSYCL to AdaptiveCpp, update CI and regenerate filter #874

Merged

keryell reviewed Feb 22, 2024

View reviewed changes

cmake/AdaptSimSYCL.cmake Show resolved Hide resolved

fknorr force-pushed the build-with-simsycl branch from ae3274e to 7d5f825 Compare December 17, 2024 17:22

fknorr mentioned this pull request Dec 18, 2024

Tooling: Track and verify CTS state celerity/SimSYCL#23

Open

fknorr force-pushed the build-with-simsycl branch from 7d5f825 to a330782 Compare December 18, 2024 13:15

fknorr force-pushed the build-with-simsycl branch from 3c601ad to 28402a6 Compare December 28, 2024 12:14

fknorr force-pushed the build-with-simsycl branch from 28402a6 to 47b7b26 Compare January 8, 2025 11:26

fknorr mentioned this pull request Jan 8, 2025

CI: Test Integration with Celerity / SYCLbench / SYCL CTS celerity/SimSYCL#7

Open

3 tasks

fknorr force-pushed the build-with-simsycl branch 2 times, most recently from 66159bf to 8518e9a Compare January 13, 2025 15:31

keryell approved these changes Jan 23, 2025

View reviewed changes

fknorr added 5 commits January 23, 2025 08:36

buffer::set_final_data accepts weak_ptr<T>, not weak_ptr<T[]> or shar…

16fa284

…ed_ptr<T[]>

Add required range -> id casts in handler test

3d78cbe

test_buffer: Avoid static initialization order fiasco

a75dacd

Do not increment/decrement invalid accessor iterators

4c050dc

Fix atomic_ref_stress / atomicity_with_host_code

60c6e96

atomicity_with_host_code requires a memory_scope argument (always `system`). atomic_ref<float> has no operator++ and must use fetch_add() instead.

fknorr force-pushed the build-with-simsycl branch 2 times, most recently from acfbda6 to 613b97d Compare January 23, 2025 07:53

Make SYCL_LANGUAGE_VERSION test a runtime check

2cf5cc4

Fix copy-paste error in legacy_input_iterator.h

3dc17f6

fknorr force-pushed the build-with-simsycl branch from 613b97d to 07edd6c Compare January 23, 2025 17:12

fknorr added 4 commits January 24, 2025 19:17

Fix copy-paste error in optional_kernel_features

76a9ce5

Remove unused variables in various tests

013fabb

Add support for SimSYCL as a SYCL implementation

0d4a60a

Add CI build step for SimSYCL

00c3a37

fknorr force-pushed the build-with-simsycl branch from 07edd6c to 00c3a37 Compare January 24, 2025 18:18

bader requested changes Jan 24, 2025

View reviewed changes

bader requested review from AlexeySachkov and psalz January 24, 2025 22:54

psalz approved these changes Jan 30, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for SimSYCL as a SYCL implementation #871

Add support for SimSYCL as a SYCL implementation #871

fknorr commented Feb 17, 2024 •

edited

Loading

AlexeySachkov Feb 19, 2024

keryell Apr 19, 2024

AlexeySachkov Feb 19, 2024

fknorr Dec 18, 2024

gmlueck Dec 18, 2024

Pennycook Jan 16, 2025

fknorr Jan 23, 2025

psalz commented Feb 21, 2024

psalz commented Feb 21, 2024

keryell commented Sep 10, 2024

fknorr commented Dec 18, 2024 •

edited

Loading

fknorr commented Jan 8, 2025

psalz commented Jan 8, 2025

fknorr commented Jan 8, 2025

keryell left a comment

tomdeakin commented Jan 23, 2025

bader commented Jan 23, 2025 •

edited

Loading

fknorr commented Jan 23, 2025 •

edited

Loading

bader commented Jan 23, 2025

fknorr commented Jan 23, 2025

AlexeySachkov commented Jan 24, 2025

bader commented Jan 24, 2025

bader Jan 24, 2025

psalz left a comment

		CHECK(std::is_same_v<decltype(SYCL_LANGUAGE_VERSION), long>);
		CHECK(SYCL_LANGUAGE_VERSION == 202012L);

Add support for SimSYCL as a SYCL implementation #871

Are you sure you want to change the base?

Add support for SimSYCL as a SYCL implementation #871

Conversation

fknorr commented Feb 17, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

psalz commented Feb 21, 2024

psalz commented Feb 21, 2024

keryell commented Sep 10, 2024

fknorr commented Dec 18, 2024 • edited Loading

fknorr commented Jan 8, 2025

psalz commented Jan 8, 2025

fknorr commented Jan 8, 2025

keryell left a comment

Choose a reason for hiding this comment

tomdeakin commented Jan 23, 2025

bader commented Jan 23, 2025 • edited Loading

fknorr commented Jan 23, 2025 • edited Loading

bader commented Jan 23, 2025

fknorr commented Jan 23, 2025

AlexeySachkov commented Jan 24, 2025

bader commented Jan 24, 2025

Choose a reason for hiding this comment

psalz left a comment

Choose a reason for hiding this comment

fknorr commented Feb 17, 2024 •

edited

Loading

fknorr commented Dec 18, 2024 •

edited

Loading

bader commented Jan 23, 2025 •

edited

Loading

fknorr commented Jan 23, 2025 •

edited

Loading