-
Notifications
You must be signed in to change notification settings - Fork 744
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[SYCL][CUDA] Remove pi Event Callback implementation #1735
Conversation
Since introduction of host tasks in #1471, `piEventCallback` and related functionality is not required by the SYCL-RT. Removing the implementation of this behaviour from the CUDA backend simplifies the submission of operations to streams and overall increases performance. Signed-off-by: Ruyman Reyes <[email protected]>
@StuartDAdams / @steffenlarsen please take a look , this is basically reverting your original work on callbacks! |
It is glorious. Good riddance. |
sycl/plugins/cuda/pi_cuda.cpp
Outdated
PI_CHECK_ERROR(cuEventCreate(&evEnd_, CU_EVENT_DEFAULT)); | ||
|
||
if (queue_->properties_ & PI_QUEUE_PROFILING_ENABLE) { | ||
PI_CHECK_ERROR(cuEventCreate(&evQueued_, CU_EVENT_DEFAULT)); | ||
PI_CHECK_ERROR(cuEventCreate(&evStart_, CU_EVENT_DEFAULT)); | ||
} | ||
} else { | ||
cl::sycl::detail::pi::die("User-defined events not implemented"); |
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.
Would be good to give more granular message so that the user knows this error is coming from the CUDA backend. Same comment on line 2709.
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.
Fixed in adcf897
LGTM! |
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.
LGTM
@Ruyk some cuda tests are failing, please fix. |
The test failures don't seem to be caused by this PR. I have made a separate PR to fix the tests: #1763 |
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.
LGTM
Any ideas how this test failure might be introduced? I thought that buildbot check should had caught such regressions. |
* upstream/sycl: [SYCL] Handle KernelName templated using type with enum template argument (intel#1780) [SYCL] Fix KernelNameInfo generated for empty template parameter pack (intel#1775) [SYCL] Do not export utility methods from SYCLMemObjT (intel#1768) [Driver][SYCL] Fix processing when using -fsycl-link (intel#1765) [SYCL][NFC] Remove outdated confusing comment (intel#1779) [SYCL][NFC] Wrap classes in .cpp into a namespace to disable external linkage. (intel#1776) [SYCL][CUDA] Fixes CUDA unit tests that uses SYCL directly (intel#1763) [SYCL][Doc] Fix default device selection rules doc (intel#1769) [SYCL][CUDA] Remove pi Event Callback implementation (intel#1735) [SYCL] Throw exception if range/offset of kernel execution exceeds INT_MAX (intel#1713) [SYCL-PTX] Add intermediate layer to libclc to ease type management (intel#1712)
Since introduction of host tasks in #1471,
piEventCallback
and related functionality is not required by the SYCL-RT.Removing the implementation of this behaviour from the CUDA backend simplifies the submission of operations to streams and increases performance in multiple benchmarks.
Signed-off-by: Ruyman Reyes [email protected]