-
Notifications
You must be signed in to change notification settings - Fork 130
[SYCL][ESIMD] Add tests on simd load from accessors #921
[SYCL][ESIMD] Add tests on simd load from accessors #921
Conversation
To distinguish from the simd load constructors from accessor Signed-off-by: Kochetkov, Yuriy <[email protected]>
To use generic code for both USM and accessor-based tests Signed-off-by: Kochetkov, Yuriy <[email protected]>
To clarify exact value used for the load constructor tests Signed-off-by: Kochetkov, Yuriy <[email protected]>
Makes possible: - power-of-two smoke tests with overaligned<1> - byte order smoke tests Signed-off-by: Kochetkov, Yuriy <[email protected]>
Signed-off-by: Kochetkov, Yuriy <[email protected]>
Signed-off-by: Kochetkov, Yuriy <[email protected]>
Based on idea that offset should be a multiplier of alignment for simd load constructor to work as expected Signed-off-by: Kochetkov, Yuriy <[email protected]>
Rebased over the latest @vasilytric Would you mind to take a look? I've tried to split changes in |
Signed-off-by: Kochetkov, Yuriy <[email protected]>
@sndmitriev Would you mind to take a look? |
As there is a restriction on possible |
Signed-off-by: Kochetkov, Yuriy <[email protected]>
Co-authored-by: vasilytric <[email protected]>
@kbobrovs Would you mind to take a look? |
const auto single_type = | ||
named_type_pack<signed char>::generate("signed char"); | ||
|
||
passed &= for_all_combinations<ctors::run_test>( |
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.
This is really cool - the robustness of the infra to create new test combinations!
@vladimirlaz, the failure is unrelated - this is a clang++ crash on test compilation. Recently internal tickets were submitted on similar cases - e.g. by @dongkyunahn-intel. Please merge. Failed Tests (1): |
Based on idea that offset should be a multiplier of alignment flag, and alignment flag should be less strict than simd element alignment itself for simd load constructor to work as expected. Signed-off-by: Kochetkov, Yuriy [email protected]
Auto pulldown and update tc files for xmain branch on 20220512
…suite#921) Based on idea that offset should be a multiplier of alignment flag, and alignment flag should be less strict than simd element alignment itself for simd load constructor to work as expected. Signed-off-by: Kochetkov, Yuriy [email protected]
Based on idea that offset should be a multiplier of alignment flag, and
alignment flag should be less strict than simd element alignment itself
for simd load constructor to work as expected.
Signed-off-by: Kochetkov, Yuriy [email protected]