Improve CUDA/HIP local argument handling #2298
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
cuda
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
3d64f76 to
977a240
Compare
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
99775d4 to
e9ecf06
Compare
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
e9ecf06 to
6c67530
Compare
EwanC
changed the title
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
6c67530 to
a13c0d1
Compare
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
2 times, most recently
from
f5a75b9 to
bed340f
Compare
hdelan
reviewed
Nov 18, 2024
aarongreig
approved these changes
Nov 18, 2024
test/conformance/exp_command_buffer/update/local_memory_update.cpp
Outdated
Show resolved
Hide resolved
test/conformance/exp_command_buffer/update/local_memory_update.cpp
Outdated
Show resolved
Hide resolved
test/conformance/exp_command_buffer/update/local_memory_update.cpp
Outdated
Show resolved
Hide resolved
test/conformance/exp_command_buffer/update/local_memory_update.cpp
Outdated
Show resolved
Hide resolved
test/conformance/exp_command_buffer/update/local_memory_update.cpp
Outdated
Show resolved
Hide resolved
Bensuo
reviewed
Nov 18, 2024
github-actions
bot
added
loader
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
2 times, most recently
from
23c24a4 to
0500f4c
Compare
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
0500f4c to
3e32dba
Compare
github-actions
bot
added
the
specification
hdelan
reviewed
Nov 19, 2024
hdelan
reviewed
Nov 19, 2024
frasercrmck
reviewed
Nov 19, 2024
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
3e32dba to
582f358
Compare
hdelan
approved these changes
Nov 20, 2024
frasercrmck
approved these changes
Nov 20, 2024
EwanC
added
ready to merge
After setting kernel arguments during update, we need to reset the amount of local memory used.
Iterate on previous solution so that the local argument offsets at following inidices are updated when an earlier local argument is updated
Co-authored-by: Ben Tracy <[email protected]> Co-authored-by: aarongreig <[email protected]>
EwanC
force-pushed
the
ewan/cuda_update_local_size
branch
from
582f358 to
e578228
Compare
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Dec 2, 2024
Tests UR PR oneapi-src/unified-runtime#2298 with additional SYCL-Graph local memory argument E2E tests. PR also sets the `pnext` and `snext` members of `ur_exp_command_buffer_update_kernel_launch_desc_t ` which were missing when calling into UR.
sarnex
pushed a commit
to intel/llvm
that referenced
this pull request
Dec 3, 2024
Tests UR PR oneapi-src/unified-runtime#2298 with additional SYCL-Graph local memory argument E2E tests. PR also sets the `pnext` and `snext` members of `ur_exp_command_buffer_update_kernel_launch_desc_t ` which were missing when calling into UR. --------- Co-authored-by: Kenneth Benzie (Benie) <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The current implementation of CUDA/HIP local memory argument handling does not update the offset of any following local memory arguments when a preceding local memory argument is set. Instead the expectation is that
clearLocalSize()is called after a kernel command has been appended/enqueued which clears the vector of local memory used by each argument. Then if the local memory arguments must be reset withurKernelSetArgLocal.This implementation causes problems for command-buffer kernel command update, where a user can pass a subset of local arguments to update, without a guarantee of passing all the local arguments in each update.
In this patch the local argument of CUDA/HIP is refactored so that when a local argument is set, any local arguments which follow are updated to account of changes in size & padding. This removes the need to have a
clearLocalSize()method and also for a user to have to seturKernelSetArgLocalbefore each kernel enqueue/append.DPC++ PR intel/llvm#16025