[SYCL] Adjust Driver and tests for addition of imf-bf16 library. #7917
Conversation
| // RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen "-device gen9" %s -### 2>&1 \ | ||
| // RUN: | FileCheck %s -check-prefix=BFLOAT16-FALLBACK-FALLBACK | ||
| // R UN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen "-device gen9" %s -### 2>&1 \ | ||
| // R UN: | FileCheck %s -check-prefix=BFLOAT16-FALLBACK-FALLBACK |
There was a problem hiding this comment.
What happened here? Looks like an unintentional space was added to the RUN:
There was a problem hiding this comment.
Somehow the full test passes locally but fails in CI.
I am trying to temporarily disable some tests to get to the root cause of the failure.
|
The one fail on AMD/HIP appears to be caused by use of the fp16 data type without checking that the GPU supports it. |
| // RUN: %clangxx -fsycl %s --sysroot=%S/Inputs/SYCL -### 2>&1 \ | ||
| // RUN: | FileCheck %s -check-prefix=BFLOAT16 | ||
|
|
||
| // test that fallback bfloat16 libraries are added in JIT mode with generic target | ||
| // RUN: %clangxx -fsycl -fsycl-targets=spir64 %s -### 2>&1 \ | ||
| // Test that fallback bfloat16 libraries are added in JIT mode with \ |
There was a problem hiding this comment.
The escape at the end of comment lines is not necessary
There was a problem hiding this comment.
Right. I'll remove them.
|
why we haven't merged this PR? |
|
I believe only @intel/llvm-gatekeepers can merge PRs. |
|
UPD: I'm taking it back, it's not unique for this PR failure
|
Changes have not yet been merged due to the failure. In accordance with the contribution guidelines unrelated failures should be mentioned and reasoning should be given for why it is unrelated, optimally with a link to a related issue and/or a PR addressing the unrelated failures. |
|
/verify with intel/llvm-test-suite#1495 |
|
@steffenlarsen I see the test failing in other PRs like #8033 , #8026 and #8031 My best guess, that this PR that was merged previously also required the test update, which was not merged . I have started 'verify with' to check this claim. |
|
An issue has been opened for the failure: #8040 |
|
@intel/llvm-gatekeepers guess we can merge this PR. The mentioned issue is known and reported and lucky guess with 'verify with' command makes the test to pass. |
This change adds the lib-imf-bf16 library into the libraries expected to be linked in by the Driver in various scenarios.