[SYCL] E2E test for interop_task with Level-Zero and OpenCL#252
[SYCL] E2E test for interop_task with Level-Zero and OpenCL#252vladimirlaz merged 12 commits intointel:intelfrom
Conversation
mikhail-nikolskiy
changed the title
bader
left a comment
bader
left a comment
There was a problem hiding this comment.
Please, apply the same comments to OpenCL test as well.
| // clang-format off | ||
| #include <level_zero/ze_api.h> | ||
| #include <CL/sycl/backend/level_zero.hpp> | ||
| // clang-format on |
There was a problem hiding this comment.
Could you clarify why clang-format comments are needed here, please?
There was a problem hiding this comment.
clang-format puts <level_zero/ze_api.h> after <CL/sycl/backend/level_zero.hpp> and compilation fails because backend/level_zero.hpp uses Level-zero types and expects ze_api.h included before
There was a problem hiding this comment.
I see. Alternative way to handle this is to put an empty line between ze_api.h and level_zero.hpp includes:
// Native API header
#include <level_zero/ze_api.h>
// SYCL API headers
#include <CL/sycl.hpp>
#include <CL/sycl/backend/level_zero.hpp>
| // clang-format off | ||
| #include <level_zero/ze_api.h> | ||
| #include <CL/sycl/backend/level_zero.hpp> | ||
| // clang-format on |
There was a problem hiding this comment.
I see. Alternative way to handle this is to put an empty line between ze_api.h and level_zero.hpp includes:
// Native API header
#include <level_zero/ze_api.h>
// SYCL API headers
#include <CL/sycl.hpp>
#include <CL/sycl/backend/level_zero.hpp>
bader
left a comment
bader
left a comment
There was a problem hiding this comment.
LGTM, except one minor comment regarding disabling clang-format.
Co-authored-by: vladimirlaz <vladimir.lazarev@intel.com>
|
@mikhail-nikolskiy, could you please resolve conflicts to let us follow up with the change? |
|
@mikhail-nikolskiy does the PR is still needed? |
Needed if we want tests for this small change merged into llvm repo https://github.com/intel/llvm/pull/3606/files, not needed otherwise |
|
@mikhail-nikolskiy could you please resolve the conflict? |
|
@vladimirlaz, BTW, I believe that test in |
nothing to be done. It is known regression and we are working on the fix. |
4 participants
Similar to #168 but extended for
to test llvm#3606
WIP because stable failure withconst void*exception observed (on Release build only) on interop-opencl-interop-task-mem.cpp line 46.The instability issue with
const void*exception fixed in test code by replacing '&' with '=' in interop_task lambda - copy accessor objects by value, not by reference.