gin/gdaki: Transcribe efa dp direct version bump - #1373
Open
anshumang wants to merge 5 commits into
Open
Conversation
anshumang
force-pushed
the
transcribe-bump-efa-dp-direct
branch
from
August 29, 2026 02:36
ffbfd23 to
8682367
Compare
anshumang
force-pushed
the
transcribe-bump-efa-dp-direct
branch
from
August 29, 2026 02:52
8682367 to
00b58c2
Compare
akkart-aws
reviewed
Sep 2, 2026
| attrs.rq_entry_size = rq_attr.entry_size; | ||
| attrs.sq_max_inline_data = gdaki_wide_wqe_inline_size; | ||
| attrs.sq_max_rdma_sges = gdaki_max_rdma_sges; | ||
| attrs.sq_wq_caps = EFA_CUDA_WQ_CAPS_64_BIT_REQ_ID_V2; |
Contributor
There was a problem hiding this comment.
Why dont we get attrs.sq_wq_caps from sq_attr.caps ?
I am assuming sq_attr.caps reports if the firmware actually supports 128bytes.
Right now we are always hardcoding that it supports req_id=64bit.
Sync the vendored CUDA datapath with efa-dp-direct Gerrit change 453817 patchset 1. This provides the version-dispatched host API, caller-owned descriptor storage, 64-bit request IDs, 128-byte inline RDMA writes, and runtime WQE geometry. Signed-off-by: Anshuman Goswami <anshumgo@amazon.com>
Build GDAKI QP and CQ descriptors through the frozen v1 operation table. Signed-off-by: Anshuman Goswami <anshumgo@amazon.com>
Thread backendVersion from createContext through QP/CQ construction and dispatch lifecycle operations through the matching efa-dp-direct operation table. Signed-off-by: Anshuman Goswami <anshumgo@amazon.com>
Build QP and CQ descriptors through efa_cuda_dp_v2 using the provider-reported SQ entry size and 64-bit request-ID capability. Keep backendVersion 1 on the v1 path. Update the GDAKI GPU tests to use EfaCudaWrBuilder and verify a request ID with nonzero high bits. Signed-off-by: Anshuman Goswami <anshumgo@amazon.com>
Adopt efa-dp-direct 1.0.0's version-dispatched, caller-owned host API. Map GIN backend versions 1 and 2 to API majors 0 and 1, respectively, copy initialized queue descriptors through the plugin accelerator abstraction, and use the validated single-RDMA-SGE limit for the major-1 WQE layout. The host library no longer calls CUDA directly, so build the unified host source and drop the libcuda dependency. Signed-off-by: Anshuman Goswami <anshumgo@amazon.com>
anshumang
force-pushed
the
transcribe-bump-efa-dp-direct
branch
from
September 4, 2026 01:12
5b0a26e to
8b2f0a9
Compare
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Adds backendVersion 2 to the GDAKI GIN path, matching efa-dp-direct's wide-WQE and 64-bit-request-ID support.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.