Extract CommandQueue interface #17352
Merged
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
ayerofieiev-tt
requested review from
cfjchu,
aliuTT,
tt-asaigal,
omilyutin-tt,
abhullar-tt,
pgkeller,
tt-aho,
tt-dma,
ubcheema and
davorchap
as code owners
ayerofieiev-tt
commented
Jan 30, 2025
ayerofieiev-tt
commented
Jan 30, 2025
| #include "trace_buffer.hpp" | ||
| #include "hardware_command_queue.hpp" | ||
| #include "memcpy.hpp" | ||
| #include "command_queue_interface.hpp" |
There was a problem hiding this comment.
not an interface really. need to find a better name, likely need to split. definitely need to move some things to cpp.
contains SystemMemoryManager, some "deprecated" singleton and a bunch of other things. Very big include.
There was a problem hiding this comment.
Yeah, definitely needs a split. Since now there is an actual "cq interface", maybe rename to "cq utils"?
ayerofieiev-tt
commented
Jan 30, 2025
tt-asaigal
reviewed
Jan 30, 2025
omilyutin-tt
approved these changes
Jan 30, 2025
| #include "trace_buffer.hpp" | ||
| #include "hardware_command_queue.hpp" | ||
| #include "memcpy.hpp" | ||
| #include "command_queue_interface.hpp" |
There was a problem hiding this comment.
Yeah, definitely needs a split. Since now there is an actual "cq interface", maybe rename to "cq utils"?
Comment on lines
+31
to
+32
| volatile bool is_dprint_server_hung() override; | ||
| volatile bool is_noc_hung() override; |
…nds. move things around
cfjchu
approved these changes
Jan 30, 2025
tt-asaigal
approved these changes
Jan 30, 2025
yieldthought
pushed a commit
that referenced
this pull request
Jan 31, 2025
### Ticket #17339 ### Problem description CommandQueue exposes all the details. CommandQueue has many things in the interface which I'd prefer us to cleanup. We need to allow MeshBuffer and MeshWorkload to go via CommandQueue. ### What's changed CommandQueue is now just an inteface command_queue.hpp is what you care about and it now includes way less. Moved command_queue_commands.hpp our of API folder. Yay! 😁 HWCommandQueue implements that interface. Device creates it inside. hardware_command_queue.hpp is relocated from api and can't be included. Yay! 😁 There is still some crap with 3 free functions which had Enqueue in them - which have nothing to do with CommandQueue - for now, located in command_queue.hpp 💩 There is some mess with command_queue_interface.hpp and command_queue_commands.hpp. Need to sort out separately. 💩 ### What's next I expect that next we work our an interface that works for both MeshBuffer and MeshWorkload, while still accommodating Buffer and Program 🔮 ```cpp // CommandQueue virtual void enqueue_program(Program& program, bool blocking) = 0; virtual void enqueue_write_buffer(Buffer& buffer, const void* src, const BufferRegion& region, bool blocking, tt::stl::Span<const SubDeviceId> sub_device_ids = {}) = 0; ``` vs ```cpp // MeshCommandQueue void enqueue_mesh_workload(MeshWorkload& mesh_workload, bool blocking); void enqueue_write_shard( std::shared_ptr<MeshBuffer>& mesh_buffer, const void* host_data, const Coordinate& coord, bool blocking); void enqueue_write_shard_to_sub_grid( const MeshBuffer& buffer, const void* host_data, const LogicalDeviceRange& device_range, bool blocking); void enqueue_write_mesh_buffer(const std::shared_ptr<MeshBuffer>& buffer, const void* host_data, bool blocking); ``` #### enqueue_program vs enqueue_mesh_workload I propose to update as `enqueue_workload(MeshWorkload)` and make sure that MeshWorkload can be (maybe even implicitly) constructed from Program #### enqueue_write_buffer vs enqueue_write* I don't know how to best handle. Need to align with @omilyutin-tt @tt-asaigal @cfjchu ### Checklist - [x] [Post commit CI](https://github.com/tenstorrent/tt-metal/actions/runs/13060406436)
nikileshx
pushed a commit
to nikileshx/tt-metal
that referenced
this pull request
Feb 3, 2025
### Ticket tenstorrent#17339 ### Problem description CommandQueue exposes all the details. CommandQueue has many things in the interface which I'd prefer us to cleanup. We need to allow MeshBuffer and MeshWorkload to go via CommandQueue. ### What's changed CommandQueue is now just an inteface command_queue.hpp is what you care about and it now includes way less. Moved command_queue_commands.hpp our of API folder. Yay! 😁 HWCommandQueue implements that interface. Device creates it inside. hardware_command_queue.hpp is relocated from api and can't be included. Yay! 😁 There is still some crap with 3 free functions which had Enqueue in them - which have nothing to do with CommandQueue - for now, located in command_queue.hpp 💩 There is some mess with command_queue_interface.hpp and command_queue_commands.hpp. Need to sort out separately. 💩 ### What's next I expect that next we work our an interface that works for both MeshBuffer and MeshWorkload, while still accommodating Buffer and Program 🔮 ```cpp // CommandQueue virtual void enqueue_program(Program& program, bool blocking) = 0; virtual void enqueue_write_buffer(Buffer& buffer, const void* src, const BufferRegion& region, bool blocking, tt::stl::Span<const SubDeviceId> sub_device_ids = {}) = 0; ``` vs ```cpp // MeshCommandQueue void enqueue_mesh_workload(MeshWorkload& mesh_workload, bool blocking); void enqueue_write_shard( std::shared_ptr<MeshBuffer>& mesh_buffer, const void* host_data, const Coordinate& coord, bool blocking); void enqueue_write_shard_to_sub_grid( const MeshBuffer& buffer, const void* host_data, const LogicalDeviceRange& device_range, bool blocking); void enqueue_write_mesh_buffer(const std::shared_ptr<MeshBuffer>& buffer, const void* host_data, bool blocking); ``` #### enqueue_program vs enqueue_mesh_workload I propose to update as `enqueue_workload(MeshWorkload)` and make sure that MeshWorkload can be (maybe even implicitly) constructed from Program #### enqueue_write_buffer vs enqueue_write* I don't know how to best handle. Need to align with @omilyutin-tt @tt-asaigal @cfjchu ### Checklist - [x] [Post commit CI](https://github.com/tenstorrent/tt-metal/actions/runs/13060406436)
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.
Ticket
#17339
Problem description
CommandQueue exposes all the details.
CommandQueue has many things in the interface which I'd prefer us to cleanup.
We need to allow MeshBuffer and MeshWorkload to go via CommandQueue.
What's changed
CommandQueue is now just an inteface
command_queue.hpp is what you care about and it now includes way less.
Moved command_queue_commands.hpp our of API folder. Yay! 😁
HWCommandQueue implements that interface.
Device creates it inside.
hardware_command_queue.hpp is relocated from api and can't be included. Yay! 😁
There is some mess with command_queue_interface.hpp. Need to sort out separately. 💩
What's next
I expect that next we work our an interface that works for both MeshBuffer and MeshWorkload, while still accommodating Buffer and Program 🔮
vs
enqueue_program vs enqueue_mesh_workload
I propose to update as
enqueue_workload(MeshWorkload)and make sure that MeshWorkload can be (maybe even implicitly) constructed from Programenqueue_write_buffer vs enqueue_write*
I don't know how to best handle. Need to align with @omilyutin-tt @tt-asaigal @cfjchu
Checklist