-
Notifications
You must be signed in to change notification settings - Fork 130
[SYCL] Add bfloat16 'hello world' host test. #1189
Conversation
Signed-off-by: Konstantin S Bobrovsky <[email protected]>
/verify with intel/llvm#6675 |
/verify with intel/llvm#6675 |
SYCL/BFloat16/bfloat_hw.cpp
Outdated
@@ -0,0 +1,50 @@ | |||
// RUN: %clangxx -fsycl %s -o %t.out | |||
// RUN: %HOST_RUN_PLACEHOLDER %t.out |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
// RUN: %HOST_RUN_PLACEHOLDER %t.out |
Host device is going to be removed soon. Moreover, you don't even use any devices in the test, so no need for any special placeholders
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe add a single_task
to verify it works on device as well then?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should I just say
// RUN: %t.out
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe add a single_task to verify it works on device as well then?
Yes, good idea. But this is E2E regression test for particular problem I found on host. Test coverage can be improved in separate patches, I believe.
Co-authored-by: aelovikov-intel <[email protected]>
/verify with intel/llvm#6675 |
Adding @rdeodhar who is working on moving bfloat16 support out of experimental. Rajiv, |
We're moving towards removing the to_float and from_float public interfaces. Also, raw is being removed. This will check the conversions:
|
I use neither to_float nor from_float - ? |
... but I use raw(). OK will try to get away only with constructors and bitcasts |
@rdeodhar, @aelovikov-intel, @AlexeySachkov - please review |
/verify with intel/llvm#6675 |
@@ -0,0 +1,60 @@ | |||
// RUN: %clangxx -fsycl %s -o %t.out | |||
// RUN: %t.out |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not sure if that is what we're supposed to do, but I honestly don't know.
Maybe someone else from @intel/llvm-reviewers-runtime know.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Seems fine to me.
Co-authored-by: aelovikov-intel <[email protected]>
Looks OK. |
/verify with intel/llvm#6675 |
Testing with intel/llvm#6675 has only known TGL-specific failures on Windows [2022-09-01T22:56:01.536Z] SYCL :: AtomicRef/assignment_atomic64.cpp |
bfloat_hw.cpp passed everywhere |
* [SYCL] Add bfloat16 'hello world' host test. Signed-off-by: Konstantin S Bobrovsky <[email protected]>
) * [SYCL] Add bfloat16 'hello world' host test. Signed-off-by: Konstantin S Bobrovsky <[email protected]>
Test for intel/llvm#6675.
Signed-off-by: Konstantin S Bobrovsky [email protected]