[core] Separate compilation, supporting C++ host side function references. #2216
Conversation
This file contains 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
amccaskey
reviewed
Sep 18, 2024
schweitzpgi
force-pushed
the
ch-first.class.kernels
branch
from
42c7ed2
to
8b13d0e
Compare
amccaskey
reviewed
Sep 18, 2024
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
1 similar comment
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
amccaskey
reviewed
Sep 24, 2024
schweitzpgi
force-pushed
the
ch-first.class.kernels
branch
2 times, most recently
from
d2924da
to
11a69e8
Compare
4 tasks
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
schweitzpgi
force-pushed
the
ch-first.class.kernels
branch
2 times, most recently
from
3177732
to
0df0f68
Compare
…nces. These changes add a new wrapper class to support interfacing between kernels. Kernels on the device side have their own names and calling conventions. Meanwhile, the C++ host code can capture references to these kernels. The C++ compiler assumes it can erase the functions completely and degenerate any reference to them as a pointer. Furthermore, it may inline or wrap this code in a thicket of template instantiations. But in order to be useful on the device side, the CUDA-Q runtime must be able to determine which kernel was wrapped in C++ code, lookup the device side code, and "link-time" optimize these calls, doing calling convention conversions, etc. Furthermore, what must be done is flavored by the execution environment. Update the call paths of hybridLaunchKernel so that we can use the new argument synthesis instead of falling back on and failing with quake synthesis. Add another deduction guide test (when using cudaq::sample). Mark all tests that use deduction guides as C++ 20 (since C++ 17 doesn't have guides). Add a hacky workaround for python. Python needs to use the new argument synthesis, but it is still using the old quake synthesis compiler pass. Attempt to add split-file to installed stuff. [Doesn't work! Why not?] Add guards to tests since some phases of the CI don't have the necessary test utilities properly installed. Mark failing state tests as UNSUPPORTED for now. These should be fixed when the state handling using new argument synthesis is fully implemented. Fix bug when there are multiple callables.
schweitzpgi
force-pushed
the
ch-first.class.kernels
branch
from
0df0f68
to
6f936d2
Compare
annagrin
reviewed
Sep 27, 2024
synthesis to be used to get the expected error. Reenable qvector_init_from_state.cpp. Hopefully this one was a red herring chasing its own CI tail.
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.
These changes add a new wrapper class to support interfacing between kernels. Kernels on the device side have their own names and calling conventions. Meanwhile, the C++ host code can capture references to these kernels. The C++ compiler assumes it can erase the functions completely and degenerate any reference to them as a pointer. Furthermore, it may inline or wrap this code in a thicket of template instantiations. But in order to be useful on the device side, the CUDA-Q runtime must be able to determine which kernel was wrapped in C++ code, lookup the device side code, and "link-time" optimize these calls, doing calling convention conversions, etc. Furthermore, what must be done is flavored by the execution environment.
Update the call paths of hybridLaunchKernel so that we can use the new argument synthesis instead of falling back on and failing with quake synthesis A couple of init state tests are marked XFAIL as they don't seem to behave with the new argument synthesis..