Skip to content

Comments

[UR][HIP][CUDA] Batched Changes#1774

Merged
kbenzie merged 21 commits intooneapi-src:mainfrom
hdelan:readytomerge-branch
Jun 20, 2024
Merged

[UR][HIP][CUDA] Batched Changes#1774
kbenzie merged 21 commits intooneapi-src:mainfrom
hdelan:readytomerge-branch

Conversation

JackAKirk and others added 18 commits June 5, 2024 07:07
When allocating new managed memory.

Signed-off-by: JackAKirk <jack.kirk@codeplay.com>
Covers case where command device outside the ur Context.

Signed-off-by: JackAKirk <jack.kirk@codeplay.com>
Signed-off-by: JackAKirk <jack.kirk@codeplay.com>
Signed-off-by: JackAKirk <jack.kirk@codeplay.com>
Use unique ptrs, and have the destructor call clear(), instead of
calling it manually. Also add a case for a std::bad_alloc.
If new fails then std::bad_alloc will be thrown. Also add the same
checking for urMemBufferCreate.
Make sure all calls of new in try catch which can catch a bad_alloc.
Make all calls to `new` safe in HIP adapter.
We can't make assumptions about interop events being on the same stream
as other interop events, since in UR we don't track the streams that
native events are recorded on.
The property should be of type size_t.

Note that we're removing a bunch of match lines for seemingly unrelated
properties, but in fact the CTS tests do not make use of the parameter
they claim, and in fact hard-code the UR_KERNEL_INFO_NUM_ARGS property.
Mirrors a similar change made to the CUDA adapater in 00761dc.
We were always returning the maximum log size (of 8192) even when the
actual log was much shorter. Strictly speaking, we should be returning
the log only up to the first null terminator. Doing otherwise could
cause strange behaviour, and may account for the flaky fails we've seen
in this test on both adapters.
The HIP and CUDA adapaters would like a more verbose logging level by
default. This patch allows customization per logger.

If the default log level is unsupplied, the old default continues to be
used. Users can also continue to override the logging level via the
various environment variables.
@hdelan hdelan requested review from a team as code owners June 19, 2024 14:08
@hdelan hdelan requested a review from npmiller June 19, 2024 14:08
@github-actions github-actions bot added common Changes or additions to common utilities conformance Conformance test suite issues. cuda CUDA adapter specific issues hip HIP adapter specific issues labels Jun 19, 2024
@hdelan hdelan force-pushed the readytomerge-branch branch from ae71aaf to 4929481 Compare June 19, 2024 15:01
@hdelan hdelan requested review from a team as code owners June 19, 2024 15:01
Hugh Delaney added 3 commits June 19, 2024 16:05
…aser/hip-context-native-handle', 'origin/fix-interop-event-bug', 'origin/make-imagecreate-safe' and 'Jack/cuda-malloc-managed-optimize' into readytomerge-branch
@hdelan hdelan force-pushed the readytomerge-branch branch from 4929481 to 836056e Compare June 19, 2024 15:07
@hdelan hdelan removed request for a team and npmiller June 19, 2024 15:16
@kbenzie kbenzie changed the title [UR][HIP][CUDA] Master Branch [UR][HIP][CUDA] Batched Changes Jun 20, 2024
@kbenzie kbenzie merged commit 975313c into oneapi-src:main Jun 20, 2024
martygrant pushed a commit to intel/llvm that referenced this pull request Jun 21, 2024
oneapi-src/unified-runtime#1774

---------

Co-authored-by: Kenneth Benzie (Benie) <k.benzie@codeplay.com>
@hdelan hdelan mentioned this pull request Jun 28, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

common Changes or additions to common utilities conformance Conformance test suite issues. cuda CUDA adapter specific issues hip HIP adapter specific issues

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants