Skip to content
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.

Sign up for GitHub

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

Jump to bottom

[WIP] Device tasks with coroutines #253

Merged
merged 83 commits into from
Feb 27, 2024
Merged

[WIP] Device tasks with coroutines #253

merged 83 commits into from
Feb 27, 2024

Conversation

devreal
Copy link
Contributor

@devreal devreal commented Apr 12, 2023

An early implementation of device tasks with coroutines, still WIP. Single-process runs should work, distributed runs need a patch in PaRSEC. Implemented only in the PaRSEC backend. See included unit tests for usage examples.

@devreal devreal marked this pull request as draft April 12, 2023 15:58
@evaleev evaleev force-pushed the ttg-device-support-master-coro branch from cf9ed17 to 9bad08b Compare April 20, 2023 19:27
@devreal devreal force-pushed the ttg-device-support-master-coro branch from db6be87 to 91e13e8 Compare June 9, 2023 17:41
devreal and others added 27 commits July 6, 2023 16:18
@devreal
Mock of ttg::make_device_tt and views
835955e 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add new header to CMake files
7b6319b 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add device_mock example to compile the ttg device mock API
25d8c3c 
Signed-off-by: Joseph Schuchart <[email protected]>
@evaleev @devreal
[example/device_mock] added missing #include
831bc0f
@evaleev @devreal
[example/device_mock] declare as off-limits to clang-format
5618e86
@evaleev @devreal
[example/device_mock] cleanup
1e8fb69
@devreal
Partial implementation of ttg::make_device_tt
4d2b937 
Signed-off-by: Joseph Schuchart <[email protected]>
@evaleev @devreal
make device_mock compile again, and no need for MAWworld for no reason
11c4d4d
@devreal
Encapsulate callables passed to make_device_tt mock in a regular task
fa6359a 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Use the new multiindex key type in device mock example
0317069 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Remove superfluous todos
ffef761 
The views on the device will be handled by the backend, not in ttg::make_view

Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Provide a scope for views and replace ttg::span with ttg::ViewSpan
65668f4 
[ttg|std]::span is not a good fit because it carries a size template
argument that we don't care about but makes handling more difficult.
Plus, we need to encode the scope for each span.

Three scopes are available:
1) SyncIn: copy host data to device before invoking the kernel callable.
2) SyncOut: copy the device data back to the host before invoking the output callable.
3) Allocate: allocate but do not synchronize in or out.

Both SyncIn and SyncOut will allocate sufficient memory on the device.

Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add missing view include in madness backend
6f5e782 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Rename CUDA precompile guards:
7510b54 
- TTG_HAVE_CUDA enables CUDA support: synchronizing the stream between
  the kernel call and the output call.
- TTG_USE_CUDA_PREFETCH enables support for prefetching based on the view scope.

Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Remove noexcept from from default View ctor
de2ad29 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add missing terminal name in mock device example
3c2186b 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add a fence to the mock device test to make the termdet happy
a04a434 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add iovec.h to list of installed files.
de0d3c4 
Signed-off-by: Joseph Schuchart <[email protected]>
@evaleev @devreal
slight reorg of contrib dox + added device design notes
d26b43e
@evaleev @devreal
updated device design notes to point out the need for Views in both U…
816b6f2 
…M and DM models
@evaleev @devreal
make TTG compilable with C++20
ace8e17
@evaleev @devreal
avoid infinite recursion in ttg::{set_size,finalize}<i> that do not t…
b59677a 
…ake outterm as an argument
@evaleev @devreal
ttg_parsec::TT::static_op{,_noarg} return parsec_hook_return_t in pre…
557c423 
…paration for suspendable tasks
@evaleev @devreal
cleanup
aa7c1b7
@evaleev @devreal
find <coroutine>
302b745
@evaleev @devreal
added an example of a task body being a (trivial) coroutine
402e47d
@evaleev @devreal
[ci] try using gcc-10, first to provide support C++ corotuines, for u…
76b52bd 
…buntu runners
devreal and others added 27 commits July 6, 2023 16:18
@devreal
Initial implementation of coroutine-based device support with PaRSEC
9fd5b87 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Compiler fixes
3fe08a8 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Add device_coro.cc test file
78785a1 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Some more device coroutine examples
d7893df 
Signed-off-by: Joseph Schuchart <[email protected]>
@therault @devreal
Base of the beginning to log the appropriate info from the profiling
7a17baf
@therault @devreal
Update to appropriate version of parsec; use arrays of chars to store…
cbb30b6 
… the key in the profiling info under the parsec driver
@therault @devreal
Port deprecated types and functions to parsec master
cc12ba2
@therault @devreal
Port the PaRSEC backend over a candidate PaRSEC master
078ad46
@devreal
Make device coro tasks actually compile
a2caac6 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Many fixes to coro-based device tasks
8cc11a6 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Clean up code from ttg::meta::is_detected debugging
82c2a89
@devreal
Clean up code from ttg::meta::is_detected debugging
5df8350 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Implement ttg::buffer and ttg::devicescratch in the PaRSEC backend
e0c0b78 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Support for moving between device and host tasks, much more
93a7dc5 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
get_ptr is not an awaitable operation
542afc2 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Extend CUDA test and add support for sending objects containing a buffer
53fd5f7
@devreal
Fix msg header ctor and non-device tasks
c6b0416 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Draft for sending out of a device task
b9b7bdb 
Signed-off-by: Joseph Schuchart <[email protected]>
@evaleev @devreal
compile fixups for clang (N.B. clang is still unusable due to missing…
514acf8 
… std::atomic_ref in libc++)
@evaleev @devreal
Wrap various C/C++-only options using language-specific generator exp…
0c889ba 
…ressions to avoid pissing CUDA off
@evaleev @devreal
make ttg::buffer and device_coro's value_t serializable by boost
8094854
@devreal
CUDA tests: increment scratch buffer if provided
96e0d67 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Disable catching POSIX signals in Catch2 tests
71f123c 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Make CUDA tasks work in distributed, incl tests
4b6970a
@devreal
CUDA version of BSPMM
19adc69 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal
Call cublasDgemm with the right arguments
26341dc
@devreal
SPMM: make CUDA version work
8362742 
Signed-off-by: Joseph Schuchart <[email protected]>
@devreal devreal force-pushed the ttg-device-support-master-coro branch from e0735a7 to 8362742 Compare July 6, 2023 20:20
@devreal devreal mentioned this pull request Sep 29, 2023
Merged
@evaleev evaleev merged commit 8362742 into TESSEorg:master Feb 27, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@devreal @evaleev @therault