NEWS: Add v1.20.0 description to main branch

[roiedanino]

What?

Add v1.20.0 new features and fixes to NEWS

---

Features

UCP

NEWS Line	Commit	Details
Added new GPU device API for direct GPU-to-GPU communication	ed7b07b36	UCT/API: Introduce device API
Added host API for GPU device management	db2c287b3	UCP/API: Add GPU device and host API
Added device signaling API with cooperation levels and flags	3c3d48977	UCP/API/DEVICE: Add signaling API
Added API for working with offsets and channel id in device operations	1537aa946	UCP/API: Work with offsets and channel id
Added method to write to local counter in device operations	5d40ef70a	UCP/DEVICE: add method to write to local counter
Added local and remote address fields to memory list element in device API	3bfebd3a4	UCP/DEVICE: Add local and remote address fields to memory list element
Added device lane selection and allocated handle population	481340f03	UCP/DEVICE: Add lane selection and populate allocated handle
Added support for Direct NIC (DPU) data path with CUDA	e7f59b68f	UCP/IB/CUDA: Direct NIC data path support
Added rkey packing support for Direct NIC	e816f536d	UCP/CORE: Rkey pack featuring Direct NIC
Added sender flush mechanism when memory sys_dev differs from remote lane sys_dev	6c5b307bd	UCP/RKEY: Sender flush if memory sys_dev is not remote lane sys_dev
Added option to use single network device per protocol	c73ee1a08	UCP/PROTO: Added option to use single network device
Added MIN_RMA_CHUNK_SIZE configuration parameter	0f379b8b9	UCP: Added MIN_RMA_CHUNK_SIZE
Decreased default value for MIN_RMA_CHUNK_SIZE from 16k to 8k	4b81bef86	UCP: Decrease default value for MIN_RMA_CHUNK_SIZE from 16k to 8k
Improved protocol lane selection with find_lanes callback to minimize overhead	82eaa01e7	UCP/PROTO: Find lanes callback to minimize overhead
Improved send-zcopy latency factor for fast-completion cases	ef3c1109d	UCP/PROTO: increase the latency factor of send-zcopy only in cases of fast-completion
Improved multi-ppn performance estimation	179a0e728	UCP: Fix multi ppn perf estimation
Removed deprecated ucp_mem functions	9ccbdf0a1	UCP/MM: Remove deprecated function
Deprecated ucp_request_alloc API	e93727f0f	UCP/API: Deprecate ucp_request_alloc

UCT

NEWS Line	Commit	Details
Added new device API for GPU communication (rc_gda transport)	ed7b07b36 + 9ded802a0	UCT/API: Introduce device API + UCT/GDA: Rename transport to rc_gda
Added GDAKI transport with endpoint export to GPU	8649b0449 + 34317a76c	UCT/GDAKI: Add endpoint + Export EP to GPU
Added DEVX QP/CQ support on foreign memory	5f3c75da0	UCT/MLX5: Add DEVX QP/CQ on foreign memory support
Added device API implementation for CUDA_IPC transport	035793caa	UCT/CUDA_IPC: implement device api
Added device put multi, put partial, and atomic operations for CUDA_IPC	ccfd7487b	UCT/CUDA_IPC: implement device put multi, put partial, atomic
Added peer failure error handling capability for GDAKI	f6b87cb94	UCT/GDAKI: Add peer failure error handling cap
Added check for nvidia_peermem driver when using GDA transport	48e9e475d	UCT/GDA: Check that nvidia peermem driver is loaded
Enabled Direct NIC by default for IB transport	d215d23de	UCT/IB: Enable Direct NIC by default
Added XDR performance recognition	19615ab98	UCT: Add XDR perf recognition
Added support for mapping DMA_BUF handle via PCIe for Direct NIC	2da859c64	UCT/CUDA/CUDA_COPY: Fixed mapping of DMA_BUF handle
Improved GDR_COPY performance with fast-path cache lookup	21d6cd664	UCT/GDR_COPY: Use fast-path cache lookup

RDMA CORE (IB, ROCE, etc.)

NEWS Line	Commit	Details
Added ConnectX-9 device support	ee882f615	UCT/IB: add CX9 to spec list
Split dp_ordering flag for DV/DevX transports	5959ce74b	UCT/IB: Split dp_ordering flag for DV/DevX
Added VRF tables support for RoCE reachability check	fe9b01261	UCT/IB: Support VRF tables for RoCE reachability check
Added EFA-specific GPUDirect support detection	4c707c0b0	UCT/IB: Check EFA-specific GPUDirect support

TCP

NEWS Line	Commit	Details
Added routing table check during reachability verification	6bae8c5e8	UCT/TCP: check the routing table during reachability check in TCP

UCS

NEWS Line	Commit	Details
Introduced lightweight rwlock data structure	8bbe776d3	UCS: Introduce lightweight rwlock
Added built-in atomics for rcache rwlock	4a8fa51cf	UCS/RCACHE: Use built-in atomics for rwlock
Improved VFS symlink paths and duplicate object handling	c1fa6ec15	UCS/VFS: Improve symlink paths and handling of duplicate objects

CUDA

NEWS Line	Commit	Details
Added wrappers for NVML functions	0e1d78825	UCT/CUDA: Added wrappers for nvml functions
Added hook for cuLibraryGetGlobal	2fb4921cf	UCM/CUDA: Added hook for cuLibraryGetGlobal
Improved CUDA call logging	2bae0a55f	UCT/CUDA: CUDA call logging enhancment
Improved source/destination memory type detection for lane performance estimation	2b014468d	UCP/PROTO/CUDA: Detect src/dst memtype when estimating lane perf
Removed unsafe usage of cuCtxGetId	8d4f43ebf	UCT/CUDA: Removed unsafe usage of cuCtxGetId
Added support for cuCtxCreate_v4 for newer CUDA versions	99430174d	TEST: use cuCtxCreate_v4 by default for newer CUDA versions
Improved context management for CUDA_IPC operations	168414505	UCT/CUDA/CUDA_IPC: Set context associated with local buffer

UCM

NEWS Line	Commit	Details
Changed module info print to debug level by default	7363001e4	UCM: Print module info to debug by default

Tools

NEWS Line	Commit	Details
Added GDAKI kernel option to perftest	df3d8b949	PERF: Perftest GDAKI kernel option
Added UCP cuda device tests to perftest	176cec5e7	UCP/PERF: UCP cuda device real tests
Added MPI+CUDA example	4d06b0e75	TEST/MPI: Added MPI+CUDA example
Differentiated wakeup feature and extra info options in perftest	c76fa1521	TOOL/PERFTEST: Differentiate wakeup feature and extra info options

Build

NEWS Line	Commit	Details
Added ability to build CUDA device code for supported architectures	502a12b94 + 63be7441e	BUILD: Add ability to build CUDA code + build device code for supported CUDA arch
Added ucx.spec into tarball for Universal Build System support	de69ee57f	BUILD: Add ucx.spec into the tarball to support Universal Build System
Added CUDA 13 support	29831d319	AZP/RELEASE: Add CUDA 13 support
Added GDA build failure when gpunetio not found	401c1278b	BUILD/GDA: Fail if building with gpunetio but it cannot be found

Packaging

NEWS Line	Commit	Details
Moved driver level dependencies under Recommends section in Debian packages	a49b1b3ac	DEBIAN: Moved driver level dependencies under Recommends section
Added Provides field for upstream packages in Debian	8b312fffa	DEBIAN: Add Provides field for upstream packages
Migrated JUCX publish from OSSRH to Central Portal	d77431a87	AZP/RELEASE: Migrate JUCX publish from OSSRH to Central Portal
Added ib-mlx5-gda separate package	9ded802a0	UCT/GDA: Rename transport to rc_gda and packages to ib-mlx5-gda

CI/Testing

NEWS Line	Commit	Details
Added Rocky OS support to release pipeline	189c08a1b	AZP/RELEASE: Add Rocky OS support
Added RHEL 10 containers to build matrices	c8d815949	AZP: Add RHEL 10 containers to build matrices
Added Debian 13 to CI build stage	934a42b1d	AZP: Add Debian13 to CI Build stage
Added ARM build testing	22a08b751	AZP: Add build test on ARM
Switched to MOFED 25.07	37be3d687	AZP: Switch to MOFED 25.07
Switched GPU tests to Ubuntu 24.04 DOCA 3.1 (GPUNetIO) image	575512691	AZP: Switch GPU tests to Ubuntu 24.04 DOCA 3.1 (GPUNetIO) image
Added support for nvidia_peermem module in testing	cfec29bec	TEST/CI: Support nvidia_peermem module
Disabled Valgrind in CI Tests stage	ae53ac724	CONTRIB: Disable Valgrind in CI Tests stage
Disabled tag matching offload tests	2dd6c6553	TEST/GTEST: Disable tag matching offload tests

GO Bindings

NEWS Line	Commit	Details
Made go bindings thread safe	ba5cc7e06	BINDINGS/GO: Bug fix - make gobindings thread safe

Documentation

NEWS Line	Commit	Details
Added note about reachability check mode in README	def7672e6	README: add note about reachability check mode
Mentioned nvlink as supported transport	e80c579ba	DOC: Mention nvlink as supported tl
Documented return status for device APIs	5fe5dafc9	DEVICE: Document return status for device APIs

AWS EFA

NEWS Line	Commit	Details
Added RMA WRITE operations support	7fe7e4482	UCT/IB/EFA: Add RMA WRITE operations
Added flush and fence operations for SRD	f2f43b4d0	UCT/EFA/SRD: Add flush and fence, enable for UCT
Enabled EFA SRD support in tests	a0ecb0165	GTEST/UCP: Enable EFA SRD support

---

Bugfixes

UCP

NEWS Line	Commit	Details
Fixed fallback to blocking registration for network device only	b14d7f1ef	UCP/CORE: Fallback to blocking registration for network device only
Fixed flush_state validity check before using it	8a7afc1fd	UCP/FLUSH: Check flush_state validity before using it
Fixed single net dev filtering for single proto	e75960cf0	UCP: Remove single net dev filtering for single proto
Fixed rkey size estimation for rendezvous	af42831b8	UCP/RNDV: Fix rkey size estimation
Fixed memory invalidation without RNDV	e52c71541 + 5d7962a44	UCP/WIREUP: Don't request invalidate without RNDV
Fixed gather_pending_requests to execute only when reconfig occurs	c5654c44b	UCP/WIREUP: Moved gather_pending_requests to be executed only when reconfig occurs

UCT

NEWS Line	Commit	Details
Fixed CUDA_IPC protocol selection for cuda_ipc	7e03b7820	DEVICE/CUDA_IPC: Fix proto selection for cuda_ipc
Fixed GDA compilation issues	2c3b56765	UCT/GDA: Fix compilation
Fixed GDAKI wqe_idx overflow	9ed46592e	UCT/GDAKI: Fix wqe_idx overflow
Fixed MM FIFO room calculation for tail > head case	c4b647fb2	UCT/MM: Fix the FIFO room calculation for tail > head
Fixed CUDA_IPC indices handling in put partial	9455002de	UCT/CUDA_IPC/TEST: change indices handling in put partial
Removed DOCA runtime dependency from GDAKI	655feb5ba	UCT/GDAKI: Remove DOCA runtime dependency
Fixed GDA log spam by reducing DOCA log level	913cca674	UCT/GDA/MLX5: Reduce DOCA log spam by setting ERROR level
Fixed UAR support check when querying resources for GDA/MLX5	f60a22c2e	UCT/GDA/MLX5: Check UAR is supported when querying resources
Fixed crash in GGA transport when EXPORTED_MKEY flag is missing	a5297d6fb	GGA: Remove XGVMI assertion

CUDA

NEWS Line	Commit	Details
Fixed stack overflow bug when calling cuPointerGetAttribute	beab36c50	UCT/CUDA_COPY: Fix stack overflow bug when calling cuPointerGetAttribute
Fixed mapping of DMA_BUF handle for Direct NIC	2da859c64	UCT/CUDA/CUDA_COPY: Fixed mapping of DMA_BUF handle
Returned object to mpool in case of failure in CUDA_COPY	ab3d32fd3	UCT/CUDA/CUDA_COPY: Return object to the mpool in case of failure
Reduced log level of rkey unpacking failures	90b7edcd5	UCT/CUDA/CUDA_IPC: Reduced log level of rkey unpacking failures
Handled cuMemRelease error status properly	53c314243	UCT/CUDA/CUDA_IPC: Handle cuMemRelease error status
Fixed context setting for local buffer in CUDA_IPC	168414505	UCT/CUDA/CUDA_IPC: Set context associated with local buffer
Fixed host unregister error message (changed to diagnostic)	7114352c5	UCT/CUDA: Do not print host unregister error (use diag)
Fixed CUDA_IPC header installation	e163bf03a	UCT/CUDA_IPC: install missing headers

RDMA CORE (IB, ROCE, etc.)

NEWS Line	Commit	Details
Fixed RoCE network device name reading	1c743cdcf	UCT/IB/BASE: Fix roce ndev name read
Fixed Direct NIC related issues	b12909197	UCT/IB: Minor fixes related to Direct NIC
Reverted RC EP address size adaptation without flush_rkey	362bbc491	UCT/IB/RC: Revert adapt EP address size without flush_rkey

UCS

NEWS Line	Commit	Details
Fixed ARCH header inclusion when building with nvcc (arm_neon.h)	49172d8f9	UCS/ARCH: Don't include arm_neon.h when building with nvcc
Fixed VFS symlink path handling	c1fa6ec15	UCS/VFS: Improve symlink paths and handling of duplicate objects
Fixed netlink message receiving to continue until 'done' flag is set	6d96bec1a	UCS/SYS/NETLINK: Receive netlink messages continuously until 'done' flag is set

Build

NEWS Line	Commit	Details
Fixed NVCC search with explicit --with-cuda	79a25dcb5	BUILD/CUDA: Fix NVCC search when --with-cuda passed explicitly
Fixed ZE transport build failures	b4d395bd3	UCT/ZE: fix ZE transport build failures
Fixed ucs_arch_get_cpu_flag compilation	4f1039d87	UCS/ARCH: Fix ucs_arch_get_cpu_flag compilation
Fixed CUDA device code build for supported architectures	63be7441e	BUILD/CUDA: build device code for supported CUDA arch

Testing

NEWS Line	Commit	Details
Fixed test_jenkins CI issues	ee04ccdc3	CI: Fix test_jenkins
Decreased rwlock test duration	cf360d50e	TEST/RWLOCK: Decrease test duration
Fixed error counting in gtest	ac3773ac1	TEST/GTEST: Fix error counting in test
Enabled retries for test_arch.memcpy	b9b953567	TEST/ARCH: Enable retries for test_arch.memcpy
Fixed test_cuda_nvml condition relaxation	35459fe57	GTEST/UCT: Relaxed test_cuda_nvml.device_get_fabric_info condition
Skipped build when generating packages	a22a3e1e5	TEST/APPS: Skip build when generating packages
Fixed CUDA device restoration in tests	935ad58f1	TEST/CUDA: Restore original cuda device
Improved error detection in UCP device tests	f3366d629	TEST/UCP/DEVICE: Improve error detection
Fixed global topo state cleanup during gtest	61e00bb46	UCS/TOPO/TEST: Don't clean the global topo state during gtest

Tools

NEWS Line	Commit	Details
Fixed perftest CUDA kernel issues	d3ff21100	TOOLS/PERF: Perftest cuda kernel fixes

GO Bindings

NEWS Line	Commit	Details
Fixed go bindings compilation with CUDA	c9b7f6e53	BINDINGS/GO: Fixed go bindings compilation with CUDA

IB/EFA

NEWS Line	Commit	Details
Fixed error message when FLID is not available	3c7b2dd7c	UCP/WIREUP/IB: Fix error message when FLID is not available

Packaging

NEWS Line	Commit	Details
Fixed RPM SPEC debug_package macro execution on SLES16	de444fabe	RPM/SPEC: Do not execute %debug_package macro on SLES16

---

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

✨ Finishing touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

Tip

📝 Customizable high-level summaries are now available in beta!

You can now customize how CodeRabbit generates the high-level summary in your pull requests — including its content, structure, tone, and formatting.

• Provide your own instructions using the high_level_summary_instructions setting.
• Format the summary however you like (bullet lists, tables, multi-section layouts, contributor stats, etc.).
• Use high_level_summary_in_walkthrough to move the summary from the description to the walkthrough section.

Example instruction:

"Divide the high-level summary into five sections:

1. 📝 Description — Summarize the main change in 50–60 words, explaining what was done.
2. 📓 References — List relevant issues, discussions, documentation, or related PRs.
3. 📦 Dependencies & Requirements — Mention any new/updated dependencies, environment variable changes, or configuration updates.
4. 📊 Contributor Summary — Include a Markdown table showing contributions:
   | Contributor | Lines Added | Lines Removed | Files Changed |
5. ✔️ Additional Notes — Add any extra reviewer context.
   Keep each section concise (under 200 words) and use bullet or numbered lists for clarity."

Note: This feature is currently in beta for Pro-tier users, and pricing will be announced later.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}