UCT/API: introduce sys_dev field for mem_alloc #10448
Open
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
Akshay-Venkatesh
requested review from
yosefe,
brminich,
rakhmets and
tvegas1
brminich
reviewed
Jan 24, 2025
src/uct/api/uct.h
Outdated
| UCT_MEM_ALLOC_PARAM_FIELD_NAME = UCS_BIT(4) | ||
| UCT_MEM_ALLOC_PARAM_FIELD_NAME = UCS_BIT(4), | ||
|
|
||
| /** Enables @ref uct_mem_alloc_params_t::sys_device */ |
There was a problem hiding this comment.
it is called sys_dev in uct_mem_alloc_params_t. Pls align one of them with the other.
also UCT_MEM_ALLOC_PARAM_FIELD_SYS_DEVICE
There was a problem hiding this comment.
UCT_MEM_ALLOC_PARAM_FIELD_SYS_DEV
yosefe
reviewed
Jan 25, 2025
src/uct/api/uct.h
Outdated
Comment on lines
2573
to
2575
| * Index of the system device on which memory is to be allocated. | ||
| * Eg: UCS_SYS_DEVICE_ID_UNKNOWN to allocate on host memory, or a specific | ||
| * index to allocate memory on GPU device. |
There was a problem hiding this comment.
Just
"System device on which memory is to be allocated, or UCS_SYS_DEVICE_ID_UNKNOWN to allow allocating on any device".
The host memory comment is misleading because mem_type field is used to select host memory
rakhmets
reviewed
Jan 28, 2025
rakhmets
reviewed
Feb 4, 2025
brminich
approved these changes
Mar 3, 2025
yosefe
reviewed
Mar 11, 2025
| @@ -539,10 +539,11 @@ ucs_status_t uct_mem_alloc_check_params(size_t length, | |||
|
|
|||
| ucs_status_t uct_md_mem_alloc(uct_md_h md, size_t *length_p, void **address_p, | |||
| ucs_memory_type_t mem_type, unsigned flags, | |||
| const char *alloc_name, uct_mem_h *memh_p) | |||
| const char *alloc_name, ucs_sys_device_t sys_dev, | |||
There was a problem hiding this comment.
move sys_dev after mem_type
| alloc_length = length; | ||
| log_level = (flags & UCT_MD_MEM_FLAG_HIDE_ERRORS) ? UCS_LOG_LEVEL_DEBUG : | ||
| UCS_LOG_LEVEL_ERROR; | ||
|
|
||
| ucs_trace("allocating %s: %s memory length %zu flags 0x%x", alloc_name, | ||
| ucs_memory_type_names[mem_type], alloc_length, flags); | ||
| ucs_trace("allocating %s: %s memory length %zu sys_dev %s flags 0x%x", |
| @@ -1160,7 +1160,7 @@ ucs_status_t | |||
| uct_ib_mlx5_devx_device_mem_alloc(uct_md_h uct_md, size_t *length_p, | |||
| void **address_p, ucs_memory_type_t mem_type, | |||
| unsigned flags, const char *alloc_name, | |||
| uct_mem_h *memh_p); | |||
| ucs_sys_device_t sys_dev, uct_mem_h *memh_p); | |||
There was a problem hiding this comment.
- fix indentation
- move sys_dev_after mem_type
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.
What? Why?
As we move towards supporting multi-device configurations and relaxing GPU context setting requirements, UCP and other UCT users need to be able to indicate on which system device the allocation is being requested on without having to rely on inferring the same from context set against calling thread. This patch extends allocation params struct to include system device. In this initial implementation all UCT allocators ignore the system device argument but eventually different transports can convert the given system device into a bus_id and allocate against that specific device.