| Index | index by Group | index by Distribution | index by Vendor | index by creation date | index by Name | Mirrors | Help | Search |
| Name: libfabric | Distribution: openSUSE:Factory:RISCV |
| Version: 1.22.0 | Vendor: openSUSE |
| Release: 6.1 | Build date: Mon Dec 2 09:47:15 2024 |
| Group: Development/Libraries/C and C++ | Build host: reproducible |
| Size: 259801 | Source RPM: libfabric-1.22.0-6.1.src.rpm |
| Packager: https://bugs.opensuse.org | |
| Url: http://www.github.com/ofiwg/libfabric | |
| Summary: User-space RDMA Fabric Interfaces | |
libfabric provides a user-space API to access high-performance fabric services, such as RDMA. This package only contains the fi_info binary.
BSD-2-Clause OR GPL-2.0-only
* Mon Dec 02 2024 Nicolas Morey <[email protected]>
- Completely remove building for AVX/AVX2 in PSM3 (bsc#1213538, bsc#1233356, bsc#1234014)
Runtime detection before initializing the provider is not enough as
PSM3 uses constructors which may include AVX insctruction.
Only requires SSE4.2 as it does make a large performance impact
in calculatin packet hashes.
- Remove psm3-fix-SIGILL-on-system-not-supporting-AVX.patch
- Add psm3-prevent-code-from-building-using-AVX-AVX2.patch
- Add _constraints to mark SSE4.2 as required
* Thu Nov 28 2024 Nicolas Morey <[email protected]>
- Add psm3-fix-SIGILL-on-system-not-supporting-AVX.patch to fix
SIGILL hapening during init on older CPU (bsc#1213538, bsc#1233356).
- Refresh libfabric-libtool.patch tu support patch -p0
* Mon Aug 05 2024 Filip Kastl <[email protected]>
- Add -Wno-incompatible-pointer-types to CFLAGS to enable building
for 32bit with GCC 14.
* Sun Aug 04 2024 Nicolas Morey <[email protected]>
- Update to 1.22.0
- Coll
- Fix Coverity issues
- Core
- General bug fixes
- hmem: change neuron get_dmabuf_fd error code
- Fix an error in the error handling path of fi_param_define()
- Makefile.am: Add Windows build files to distribution tarball
- hmem: disable ZE IPC
- Add profile variables for connections and memory allocated
- hmem: Fix `cuDeviceCanAccessPeer()` error reporting
- man: Update text for `len` parameter
- Add page size MR attr field
- man: Extend fi_mr_refresh support
- man: Improve FI_MR_ALLOCATED documentation
- man: Support optional MR desc
- man: Improve FI_MR_HMEM documentation
- Added ofi_get_realtime interfaces
- Add endpoint options for max message size and inject size
- Add Windows definition for `EREMOTEIO`
- EFA
- General improvement and bug fixes
- Handle recv cancel for zero copy recv
- Avoid iterating EP list in CQ read
- Add RDMA core errno for remote unknown peer
- Map EFA errnos to Libfabric codes
- Improve the zero-copy receive feature
- Improve the handshake enforcement procedure
- Support unsolicited rdma-write recv
- Support FI_MORE for eager send and rdma-write
- Improve the EFA_IO_COMP error code and explanation
- Improve the unit test for LL128 protocol
- Distinguish max RMA size from msg size
- Hooks
- dmabuf: Fix incompatible pointer warning
- OPX
- Add missing file needed for fabric direct build to release package
- Fix performance issue caused by not setting ACK bit in the single
SDMA packet case
- TID cache debug improvements
- Detection of driver lack of support for TID
- Multi-CTS support for TID
- Removal of statement that TID is not supported
- OPX Tracer improvements
- Improvements to OPX shared memory cleanup
- H to H performance improvements for build that supports HMEM
- Bug fix for a threshold check
- Bug fix for FI_SELECTIVE_COMPLETION
- CN5000 fixes
- Parameterization of various thresholds
- Further enhancements to support NVIDIA GPUs, included CUDA-allocated
bounce buffers and in-provider support for GDRCopy
- Enhancements to enable support for CN5000 hardware
- Better checking for TID support
- General TID enhancements
- Pkey error handling
- Send work queue splitting
- Support for OPX tracer for profiling purposes
- Coverity scan fixes
- Fixes and enhancements to logging and debug messages
- Intranode RMA read fixes
- Fix compile issues
- Fix shared memory segment index creation bug
- PSM3
- Update provider to sync with IEFS 11.7.0.0.110
- Improved auto-tuning features for PSM3, including dynamic Credit Flows
and detecting the presence of the rv kernel module
- Improved PSM3 intra-node performance for large message sizes
- SHM
- Added support for write() method to submit DSA work
- Touch all buffer pages after DSA page fault
- Add return and more descriptive error message
- Fix coverity about incorrect sign
- Fix memory leaks for srx
- Fix atomic read
- Sockets
- Fix Coverity issues
- USNIC
- Fix a few Coverity issues
- Util
- Discard outstanding operations in util_srx_close
- Enable profile on the size of bufpool allocated.
- Add more predefined profile variables.
- Fix issue while displaying addresses with fi_info -a <addr_format>
- fi_pingpong: Fix out of scope memory leak
- Add source address to fi_pingpong
- Verbs
- Flush CQ for SQ on no SQ credit
- Optimize search for device max inline size
- Enable profiling
- Fabtests
- pytest/shm: reduce the msg size in test_unexpected_msg
- Fix synapseai fabtests build
- Add pytests for EFA zero-copy receive
- Add benchmark option for `FI_OPT_MAX_MSG_SIZE`
- benchmarks: Add synapseai support
- Disable fi_rdm_tagged_peek test for ucx and psm3
- Add manual init sync to fi_rdm_multiclient and fi_rdm
- Refactor ft_sock_sync to take in a socket
- Add fi_rdm_bw test
- Skip rma_pingpong write tests
- Init rx_buf before sending data
- Add rma_pingpong tests to makefile
- pytest: use different message sizes for rma pingpong
- Fix missing fixture memory_type in test_rma_pingpong_range_no_inject
- pytest: account for process startup overhead in client-server tests
- pytest: save client process output to a file
- Support testing inject with cq data
- multinode: update arguments
- multi_ep: Fix memory leak
- rdm_tagged_peek: Align rx's msg_order with tx's
- Add backlog > 0 to listen call
* Wed Apr 03 2024 Nicolas Morey <[email protected]>
- Enable ucx and new efa provider on 64b architectures.
- Use a single changes file for libfabric and fabtests.
- Update to 1.21.0
- Core
- Various update and fixed in man pages
- Fix xpmem memory corruption
- Extend FI_PROVIDER_PATH to allow setting preferred DL provider
- Add a SECURITY.md file
- Document preferred threading model for scalable endpoints
- Move FI_PRIORITY to internal flag
- Remove FI_PROV_SPECIFIC
- Remove unimplemented or unused features
- Support cntr byte counting
- configure: Do not check for xpmem if disabled
- Add FI_PROGRESS_CONTROL_UNIFIED
- hmem/cuda: Get multiple attributes at once in cuda_is_addr_valid
- configure: Add -pipe by default to CFLAGS
- Selectively generate warnings on failed loading of DL providers
- hmem: introduce ofi_dev_reg_copy_*_iov ops
- Print provider path on fabric creation
- Introduce FI_OPT_SHARED_MEMORY_PERMITTED
- README.md: Add badge for openssf scorecard
- man: Regulate the fi_setopt call sequence.
- man: Clarify the usage of FI_RMOTE_CQ_DATA flag
- man: Add ucx provider to the fi_provider man page
- configure.ac: add extra check for 128 bit atomic support
- include/osd: align atomic complex definitions
- hmem/synapseai: Refine the error handling and warning
- Specify C11 standard for Visual Studio builds
- configure: Do not check for xpmem if disabled
- man page fixes
- EFA
- General improvement and bug fixes
- Propagate errnos from core functions untouched
- Create 1:1 relationship between libfabric CQs and IBV CQs
- Do not progress ep inside transmission call when hitting EAGAIN
- Remove unnecessary check in rdma write.
- Handle rx pkts error without ope
- Add a new rx pkt counter
- Enable runting for neuron with a different runt size
- Distinguish unresponsive receiver errors
- Remove unnecessary handshake in send path
- Don't fail the whole domain init if cudamalloc failed
- Introduce efa specific domain operations
- Implement FI_OPT_SHARED_MEMORY_PERMITTED
- Do not memset rxe to 0 on init
- Reduce # of error cases in happy path
- Add FI_EFA_USE_HUGE_PAGE to efa man page.
- Don't do handshake for local fi_write
- Add pingpong test after exhausting MRs
- Introduce utilities to exhaust MRs on EFA device
- Test EFA with a 1GiB message
- Do not abort on all deprecated env vars
- Onboard fi_mr_dmabuf API in mem reg ops.
- Try registering cuda memory via dmabuf when checking p2p
- Introduce HAVE_EFA_DMABUF_MR macro in configure
- Use long CTS protocol if long read and runting read protocols fail
because of memory registration limits
- Remove unnecessary check in rdma write.
- Enable runting for neuron with a different runt size
- Handle rx pkts error without ope
- Distinguish unresponsive receiver errors
- Add `efa_show_help()`
- Refactor error code definitions
- Remove error message assertions from CQ unit tests
- Refactor `efa_strerror()`
- Doxyfile: Configure tabs to 8 spaces
- Rename Doxyfile
- Hooks
- dmabuf_peer_mem: initialize fd to supress compiler warning
- NETDIR
- Removed. The functionality is intergrated into the verbs provider.
- OPX
- Fix compiler warnings and coverity issues
- General improvement and bug fixes
- Add GPU support to expected TID
- RZV RTS packet exclude empty immediate data
- Add more efficient check for cuda-resident user buffer
- Improve default HFI selection logic in multi rail environments
- Flush dead list opportunistically
- Add RISC-V support
- Make update HDRQ register frequency configurable at build time
- Removed all references to the reliability nack threshold env var
- Added missing tuneables, rearraged to match fi_info -e output
- Use BAR load/store macros
- Check HFI driver version to allow GPU-enabled build/run
- Added kernel and driver version check to allow/disallow expected receive TID
- Fix max SHM connections to allow up to 16 HFIs
- Use FI_HMEM_SYSTEM for Cuda-Managed (Unified) memory
- Handle FI_OPT_CUDA_API_PERMITTED
- Use contiguous send when only one iov present
- Always replay TID packets over SDMA
- Add Virtual Lane and Partition pkey (FI_OPX_SL and FI_OPX_PKEY)
- Forced AV type to be AV Map when requested AV is unsupported
- Reduce size of opx_shm_tx
- Add GPU support for RMA Atomic operations
- Add GPU support for RMA reads and writes
- Add HMEM debug counters
- Print debug counters upon receiving SIGUSR1
- Fix multi-receive to work with contiguous rzv payload
- Initial support for GPU / FI_HMEM
- Limit multipacket eager implementation to tagged sends
- Read, verify and store some hfi chip attributes
- PSM3
- Update provider to sync with IEFS 11.6.0.0.231
- Fix some conditional build errors
- RSTREAM
- Removed.
- RXM
- Add option to auto detect hmem iface of user buffers
- SHM
- Manually align 8 byte fields in memory region
- Close device_fds for connected peers when the EP is closed
- Print shm name and error code when failed to open
- Mark send as completed when a message is discarded
- Don't close dmabuf-fd when a request is done
- Revert the smr_region fields adjustment
- Fix various coverity issues
- Add ep to cq ep list once in cq bind
- Add ofi_buf_alloc error handling
- Revert the smr_region fields adjustment
- Don't close dmabuf-fd when a request is done
- Mark send as completed when a message is discarded
- Print shm name and error code when failed to open
- Close device_fds for connected peers when the EP is closed
- SOCKETS
- fix compiler warnings and coverity issues
- UCX
- Fix incorrect enum value in FI_DBG() and FI_WARN()
- USNIC
- Turn off compiler warnings of possible string truncation
- Util
- Make ep_list_lock noop for FI_PROGRESS_CONTROL_UNIFIED
- Save control progress model to util_domain
- Set import monitor state to idle upon close
- Add name field to memory monitors
- memhooks: Fix a bug when calculating mprotect region
- Modify domain_attr based on FI_AV_AUTH_KEY
- Verbs
- Non-blocking EP creation
- Address cm_id resource leak in rdma_reject path
- Redirected error handle logic for dmabuf failure in verbs
- Added rocr dmabuf support under verbs
- Windows: Check error code from GetPrivateData
- Add missing lock to protect SRX
- Fix compiler warnings about out of boundary access
- Fabtests
- Fix various coverity issues
- General improvement and bug fixes
- Add multi_ep test
- Serialize the run of fi_cq_test
- Utilize `junitparser` module directly
- Add progress models to SHM/EFA fabtests
- Add option to change progress model
- efa/rnr_cq_read_err: poll cq when hitting EAGAIN
- Allow testing multi_ep with shared/non-shared cq and av
- Print warning for HMEM iface init failure
- efa: Add small tx_rx size test
- pytest: Make ssh connection error pattern less stringent
- Add new exclude file for io_uring tests
- Add rma_pingpong benchmark
- efa: Make 1G tests run faster
- pytests: add command line argument for dmabuf reg
- Bump Libfabric API version.
- Add option to support dmabuf MR
- Add dmabuf ops for cuda.
- Replace strtok with strtok_r
- Add new exclude file for io_uring tests
* Mon Mar 25 2024 Nicolas Morey <[email protected]>
- Update to 1.20.1
- Core
- hmem/ze: Change the library name passed to dlopen
- hmem/ze: map device id to physical device
- hmem/ze: skip duplicate initialization
- hmem/ze: dynamically allocate device resources based on number of devices
- hmem/ze: fix hmem_ze_copy_engine variable look up
- hmem/ze: Increase ZE_MAX_DEVICES to 32
- man: Fix typo in fi_getinfo man page
- Fix compiler warning when compiling with ICX
- man: Fix fi_rxm.7 and fi_collective.3 man pages
- man: Update EFA docs for FI_EFA_INTER_MIN_READ_WRITE_SIZE
- EFA
- efa_rdm_ep_record_tx_op_submitted() rm peer lookup
- Remove peer lookup from efa_rdm_pke_sendv()
- Make handshake response use txe
- test: Only close SHM if SHM peer is Created
- Handshake code allocs txe via efa util
- Initialize txe.rma_iov_count to 0
- Switch fi_addr to efa_rdm_peer in trigger_handshake
- Downgrade EFA Endpoint Creation WARN to INFO
- Init srx_ctx before use
- Clean up generic_send path
- Pass in efa_rdm_ep to efa_rdm_msg_generic_recv()
- Make recv path slightly more efficient
- re-org rma write to avoid duplicate checks
- Add missing sync_memops call to writedata
- use peer pointer from txe in read, write and send
- Pass in peer pointer to txe
- Get rid of noop instruction from empty #define
- Remove noop memset
- Fix the ibv cq error handling.
- Don't do handshake for local read
- Fix a typo in configure.m4
- Make runt_size aligned
- OPX
- Initialize cq error data size
- RXM
- Fix data error with FI_OFI_RXM_USE_RNDV_WRITE=1
- SHM
- Fix coverity issue about resource leak
- Adjust the order of smr_region fields.
- Allocate peer device fds dynamically
- Util
- Fix coverity issue about missing lock
- Implement timeout in util_wait_yield_run()
- Fix bug in util_cq startup error case
- util_mem_hooks: add missing parantheses
- Verbs
- Windows: Resolve regression in user data retrieval
- Fabtests
- efa: Close ibv device after use
- efa: Get device MR limit from ibv_query_device
- efa: Add simple unexpected test to MR exhaustion test
- pytest: add a new ssh connection error pattern
* Thu Feb 29 2024 [email protected]
- Use %autosetup macro. Allows to eliminate the usage of deprecated
%patchN
* Sun Nov 19 2023 Nicolas Morey <[email protected]>
- Update to 1.20.0 (jsc#PED-5777, jsc#PED-5893, jsc#PED-5889)
- Core
- General bug fixes and code clean-up
- configure.ac: add extra check for 128 bit atomic support
- hmem/synapseai: Refine the error handling and warning
- Introduce FI_ENOMR
- hmem/cuda: fix a bug when calculating aligned size.
- Handle dmabuf for ofi_mr_cache* functions.
- Handle dmabuf flag in ofi_mr_attr_update
- Handle dmabuf for mr_map insert.
- man: Fix the description of virtual address when FI_MR_DMABUF is set
- man: Clarify the defition of FI_OPT_MIN_MULTI_RECV
- hmem/cuda: Add dmabuf fd ops functions
- include/ofi_atomic_queue: Properly align atomic values
- Define fi_av_set_user_id
- Support multiple auth keys per EP
- Simplify restricted-dl feature
- hmem: Only initalize synapseai if device exists
- Add "--enable-profile" option
- windows: Updated config.h
- Add environment variable for selective HMEM initialization
- Add restricted dlopen flag to configure options
- hmem: generalize the use of OFI_HMEM_DATA to non-cuda iface
- hmem: fail cuda_dev_register if gdrcopy is not enabled
- Add 1.7 ABI compat
- Define fi_domain_attr::max_ep_auth_key
- hmem: Add new op to hmem_ops for getting dmabuf fd
- hmem/cuda: Update cuda_gdrcopy_dev_register's signature
- mr_cache: Define ofi_mr_info::flags
- Add ABI compat for fi_cq_err_entry::src_addr
- Define fi_cq_err_entry::src_addr
- Add base_addr to fi_mr_dmabuf
- hmem: Set FI_HMEM_HOST_ALLOC for ze addr valid
- hmem: Support dev reg with FI_HMEM_ZE
- tostr: Added fi_tostr() for data type struct fi_cq_err_entry.
- hmem_ze: fix incorrect device id in copy function
- Introduce new profiling interface for low-level statistics
- hmem: Support dev reg with FI_HMEM_CUDA
- hmem: Support dev reg with FI_HMEM_ROCR
- hmem: Support dev reg with FI_HMEM_SYSTEM
- hmem: Define optimized HMEM memcpy APIs
- Implement memhooks atfork child handler
- hmem: Support ofi_hmem_get_base_addr with sys mem
- hmem: Add length field to ofi_hmem_get_base_addr
- mr_cache: Improve cache hit rate
- mr_cache: Purge dead regions in find
- mr_cache: Update find to remove invalid MR entries
- mr_cache: Update find with MM valid check
- Add direct support for dma-buf memory registration
- man/fi_tagged: Remove the peek for data ability
- indexer: Add byte idx abstraction
- Add missing FI_REMOTE_CQ_DATA for fi_inject_writedata
- Add configure flags for more sanitizers
- Fix fi_peer man page inconsistency
- include/fi_peer: Add cq_data to rx_entry, allow peer to modify on unexp
- Add XPMEM support
- EFA
- General bug fix and code clean-up
- Do not abort on all deprecated env vars
- Onboard fi_mr_dmabuf API in mem reg ops.
- Try registering cuda memory via dmabuf when checking p2p
- Introduce HAVE_EFA_DMABUF_MR macro in configure
- Add read nack protocol docs
- Receiver send NACK if runt read fails with ENOMR
- Sender switch to long CTS protocol if runt read fails with ENOMR
- Receiver send NACK if long read fails with ENOMR
- Update efa_rdm_rxe_map_remove to accept msg_id and addr
- Sender switch to long CTS protocol if long read fails with ENOMR
- Introduce new READ_NACK feature
- Use SHM's full inject size
- Add testing for small messages without inject
- Enable inject rdma write
- Use bounce buffer for 0 byte writes
- Onboard ofi_hmem_dev_register API
- Update cuda_gdrcopy_dev_register's signature
- Allocate pke_vec, recv_wr_vec, sge_vec from heap
- Close shm resource when it is disabled in ep
- Disable RUNTING for Neuron
- Move cuda-sync-memops from MR to EP
- Do not insert shm av inside efa progress engine
- Enable shm when FI_HMEM and FI_ATOMIC are requested
- Adjust posted receive size to pkt_size
- Do not create SHM peer when SHM is disabled
- Use correct threading model for shm
- Restrict RDMA read to compatible EFA devices
- Add EFA device version to handshake
- Add missing locks in efa_cntr_wait.
- Add writedata RNR fabtest
- Handle RNRs from RDMA writedata
- Check opt_len in efa_rdm_ep_getopt
- Use correct tx/rx op_flags for shm
- Hooks
- dmabuf: Initialize fd to supress compiler warning
- trace: Add log on FI_VAR_UNEXP_MSG_CNT when enabled.
- trace: Fixed trace log format on some attributes.
- OPX
- Fix compiler warnings
- PSM3
- Fix compiler warnings
- Update provider to sync with IEFS 11.5.1.1.1
- RXM
- Remove unused function
- Use gdrcopy in rma when emulating injection
- Use gdrcopy in eager send/recv
- Add hmem gdrcopy functions
- Remove unused dynamic rbuf support
- SHM
- General bug fixes and cleanup
- Add ofi_buf_alloc error handling
- Only copy header + msg on unexpected path
- Add FI_HMEM atomic support
- Add memory barrier before updating resp for atomic
- Add more error output
- Reduce atomic locking with ofi_mr_map_verify
- Only increment tx cntr when inject rma succeeded.
- Use peer cntr inc ops in smr_progress_cmd
- Allow for inject protocol to buffer more unexpected messages
- Change pending fs to bufpool to allow it to grow
- Add unexpected SAR buffering
- Use generic acronym for shm cap
- Move CMA to use the p2p infrastructure
- Add p2p abstraction
- Load DSA dependency dynamically
- Replace tx_lock with ep_lock
- Calculate comp vars when writing completion
- Move progress_sar above progress_cmd
- Rename SAR status enum to be more clear
- Make SAR protocol handle 0 byte transfer.
- Move selection logic to smr_select_proto()
- Sockets
- Fix compiler warnings
- Fix provider name and api version in returned fi_info struct
- TCP
- Add profiling interface support
- Pass through rdm_ep flags to msg eps
- Derive cq flags from op and msg flags
- Do not progress ep that is disconnected
- Set FI_MULTI_RECV for last completed RX slice
- Return an error if invalid sequence number received
- xnet_progress_rx() must only be called when connected
- Reset ep->rx_avail to 0 after RX queue is flushed
- Disable the EP if an error is detected for zero-copy
- Add debug tracking of transfer entries
- Negotiate support for rendezvous
- Add rendezvous protocol option
- Generalize xnet_send_ack
- Flatten protocol header definitions
- Remove unused dynamic rbuf support
- Define tcp specific protocol ops
- Remove unneeded and incorrect rx_entry init code
- UCX
- Add FI_HMEM support
- Initialize ep_flush to 1
- Util
- General bug fixes
- memhooks: Fix a bug when calculating mprotect region
- Check the return value of ofi_genlock_init()
- Update checks for FI_AV_AUTH_KEY
- Define domain primary and secondary caps
- Add profiling util functions
- Update util_cq to support err_data
- Update ofi_cq_readerr to use new memcpy
- Update ofi_cq_err_memcpy to handle err_data
- Zero util cancel err entry
- Move FI_REMOTE/LOCAL_COMM to secondary caps
- Alter domain max_ep_auth_key
- Add domain checks for max_ep_auth_key
- Revert util_cntr->ep_list_lock to ofi_mutex
- Add NIC FID functions to ofi.h
- Add EP and domain auth key checking
- Add bounds checks to ibuf get
- Define dlist_first_entry_or_null
- Update util_getinfo to dup auth_key
- Revert util_av, util_cq and util_cntr to mutex
- Add missing calls to (de)initialize monitor's mutexes
- Avoid attempting to cleanup an uninitialized MR cache
- Rename ofi_mr_info fields
- Add rv64g support to memory hooks
- Verbs
- Windows: Check error code from GetPrivateData
- Add missing lock to protect SRX
- Add synapseai dmabuf mr support
- Bug fix for matching domain name with device name
- Windows: Fetch rejected connection data
- Add support for DMA-buf memory registration
- Windows: Fix use-after-free in case of failure in fi_listen
- Windows: Map ND request type to ibverbs opcode
- Fix memory leak when creating EQ with unsupported wait object
- Track ep state to prevent duplicate shutdown events
- Fabtests
- Update man page
- pytests/efa: onboard dmabuf argument for test_mr
- pytest: make do_dmabuf_reg_for_hmem an cmdline argument
- Bump Libfabric API version.
- mr_test: Add dmabuf support
- Introduce ft_get_dmabuf_from_iov
- unexpected_msg: Use ft_reg_mr to register memory
- pytest: Allow registering mr with dmabuf
- Add dmabuf support to ft_reg_mr
- Add dmabuf ops for cuda.
- Test max inject size
- Add FI_HMEM support to fi_rdm_rma_event and fi_rdm tests
- memcopy-xe: Fix data verification error for device buffer
- dmabuf-rdma: Increase the number of NICs that can be tested
- dmabuf-rdma: Remove redundant libze_ops definition
- fi-mr-reg-xe: Skip native dmabuf reg test for system memory
- Check if fi_info is returned correctly in case of FI_CONNREQ
- cq_data: relax CQ data validation to cq_data_size
- Add ZE host alloc function
- Use common device host buffer for check_buf
- hmem_ze: allocate one cq and cl on init
- fi-mr-reg-xe: Add testing for dmabuf registration
- scripts: use yaml safe_load
- macos: Fix build error with clang
- multinode: Use FI_DELIVERY_COMPLETE for 'barrier'
- Handle partial read scenario for fi_xe_rdmabw test For cross node tests
- pytest/efa: add cuda memory marker
- pytest/efa: Skip some configuration for unexp msg test on neuron.
- runfabtests.py: ignore error due to no tests are collected.
- pytest/efa: extend unexpected msg test range
- pytest/shm: extend unexpected msg test range
- pytest: Allow running shm fabtests in parallel
- unexpected_msg.c: Allow running the test with FI_DELIVERY_COMPLETE
- runfabtests.sh: run fi_unexpected_msg with data validation
- pytest/shm: Extend test_unexpected_message
- unexpected_msg: Make tx/rx_size large enough
- pytest/shm: Extend shm's rma bw test
- Update shm.exclude
* Mon Sep 04 2023 Nicolas Morey <[email protected]>
- Update to 1.19.0
- Core
- General code cleanup and restructuring
- Add ofi_hmem_any_ipc_enabled()
- ofi_consume_iov allows 0-byte consume
- ofi_consume_iov consistency
- ofi_indexer: return error code when iterating
- getinfo: Add post filters for domain and fabric names
- Filter loopback device if iface is specified
- bsock: Fix error checking for -EAGAIN
- windows/osd: Remove unneeded check to silence coverity
- windows/osd: Move variable declaration to silence coverity
- Introduce gdrcopy awareness to hmem copy
- mr/cache: Fix fi_mr_info initialization
- hmem_cuda: remove gdrcopy from cuda hmem copy path
- iouring: Fix wrong indent in ofi_sockapi_accept_uring()
- Implement ofi_sockctx_uring_poll_add()
- hmem: introduce gdrcopy from/to cuda iov functions
- hmem: Deprecate `FI_HMEM_CUDA_ENABLE_XFER`
- hmem_cuda: Restrict CUDA IPC based on peer accessibility
- hmem_cuda: Log number of CUDA devices detected
- hmem_cuda: Refactor global variables
- tostr: Remove the extra dir "shared/" from "include/" and "src/" .
- hmem_ze: fix ZE is valid check
- hmem_rocr: fix offset calculation
- hmem_rocr: use ofi spinlock functions
- hmem_rocr: minor fixes
- hmem_neuron: convert warn to info for nrt_get_dmabuf_fd not found
- hmem_neuron: check existance of neuron devices during initialization
- tostr: Moved Windows functions in shared/ofi_str.c to windows/osd.h
- tostr: Add helper functions ofi_tostr_size() and ofi_tostr_count().
- EFA
- Onboard Peer API, use shm provider as a peer provider
- Uses util SRX framework in shared receive procedures.
- Register shm MR with hmem_data, allow shm to use gdrcopy for cuda data movement
- Finish the refactor for rxr squash.
- Use rdma-core WR API for send requests
- Check optlen in getopt call
- Fix the rdma-read support check in RMA and MSG operations
- Optimize ep lock usage
- Use an internal fi_mr_attr for memory registration
- Hooks
- Init field in mr_attr to silence coverity
- Add profiling hook provider
- Rename cq hooking functions' names
- Added trace for resource creation operations
- OPX
- Initialize ofi_mr_info
- Fix dput credit check
- Only allocate replay buffer if psn is valid
- Support SHM Intra-node communication between single server HFI devices
- Fix incorrect packet size in packet header when sending CTS packet
- Added check to address Coverity scan defect
- Add multi-entry caching to TID rendezvous
- Fall back to default domain name for TID fabric
- Properly handle multiple IOVs in fi_opx_tsendmsg
- Fix OPX Rzv RTS receive operation SHM error (DAOS-related)
- Fix non-tagged sends may incorrectly set FI_TAGGED in send completions
- Add more info to reliability IOV buffer validation check
- Move dput packet build functions to new inline include
- Use fi_mr_attr in fi_opx_mr
- Disable Pre-NAKing by default, throttle until all outstanding replays ACK'd
- Fix reliability bug when NAKing the last PSN
- Update HeaderQ Register more frequently
- No rbuf_wrap needed for expected receive (TID)
- Fixes for Coverity scan issues
- Enhanced tag matching
- Tune expected recv for unaligned buffers
- Observability: Add finer logging granularity
- Reduce RTS immediate data and fix packet estimate for odd TID lengths
- Add additional sources for FI_OPX_UUID
- Peer
- Add cq_data to rx_entry, allow peer to modify on unexp
- Introduce peer cntr API
- Add foreach_unspec_addr API
- Add size as an input of the get_tag
- PSM3
- Sync with IEFS 11.5.0.0.172
- SHM
- Only poll IPC list when ROCR IPC is enabled
- Allow for SAR and inject protocol to buffer more unexpected messages
- Remove unused sar fields
- Make SAR protocol handle 0 byte transfer
- Load DSA dependency dynamically
- Change recv entry freestack into bufpool
- Remove shm signal
- Use util peer cntr implementation
- Make SHM default to domain level threading level
- Replace internal shared receive implementation with util_srx
- Lock entire progress loop
- Fix ROCR data coherency
- Add FI_LOCAL_COMM to shm attrs
- Handle empty freestack
- Fix bug in configure.m4 in atomics_happy assignment happy
- Add memory barrier before update resp->status for SAR
- Do not use inline/inject for read op
- Allow shm to use gdrcopy
- Refactor protocol selection code
- Init map fi addrs to FI_ADDR_NOTAVAIL
- TCP
- General code cleanups
- Restrict which EPs can be opened per domain
- Increase CM error debug output
- Avoid calling close() on an invalid socket after accept error
- Mark the EP as disconnected before flushing the queues
- Add assertion failures for xnet_{monitor,halt}_sock
- Disable ofi_dynpoll_wait() for non-blocking progress
- Move PEP pollin operations to io_uring
- Move EP poll operations to io_uring
- Early exit if ofi_bsock_flush() has operation in progress
- Implement pollin sockctx in bsock
- Add missing call to xnet_submit_uring()
- Add return error to xnet_update_pollflag()
- Remove the cancel sockctx from the EP structure
- Move io_uring cqe from the stack to progress struct
- Reduce stack size for epoll event array
- handle NULL av in xnet_freeall_conns()
- UCX
- Publish FI_LOCAL_COMM and FI_REMOTE_COMM capabilities
- Fix configure error with newer MOFED
- Fix segfault in unsignalled completions
- Util
- Add FI_PEER support to util counter
- Refactor the usage of cntrs
- Change util_ep to be a genlock
- Add util shared receive implementation
- Update log message for invalid AV type message
- Fix fi_mr_info initialization
- Add peer ID to MR cache
- Store hmem_data in ofi_mr_map
- Split the cq progress and reading entries in ofi_cq_readfrom
- Verbs
- Add event lock to EQ to serialize closing ep
- Remove saved_wc_list and use CQ directly
- Consolidate peer_mem and dmabuf support check
- Fix vrb_add_credits signature
- Introduce new progress engine structure
- Simplify (and correct) locking around progress operations
- General code restructuring
- Fabtests
- Fix reading addressing options
- Allow to change only the OOB address
- Allow to use FI_ADDR_STR with -F
- Fix bw buffer utilization
- Separate RX and RMA counters
- Fix tx counter with RMA
- Add FI_CONTEXT mode to rdm_cntr_pingpong
- Add HMEM support to fi_unexpected_msg test
- Fix array OOB during fabtest list parsing
- Enable shm tagged_peek test
- Fix windows build warnings
- Make tx_buf and rx_buf aligned to 64 bytes by default
- Fix windows build warnings for sscanf
- Use dummy ft_pin_core on macOS
- Fix some header includes
- sock_test: Do not use epoll if not available
- recv_cancel: initialize error entry
- Fix wrong size used to allocate tx_msg_buf
- unexpected: change defaults to support tcp
- unexpected: add unknown unexpected peer test
- Enable a list of arbitrary message sizes
- Enabled data validation for rma read & write
- bw_rma operates on distinct buffer offsets
- ft_post_rma issues reads from remote's tx_buf
- General code cleanup and restructuring
- rdm_tagged_peek: fix race condition synchronization
- Add FI_LOCAL_COMM/FI_REMOTE_COMM presence check to fi_getinfo_test
- Correct ft_exchange_keys in prefix-mode
- Make rdm_tagged_peek test more general
- Add unit test for fi_setopt
* Mon Aug 07 2023 Nicolas Morey <[email protected]>
- Drop support for obsolete TrueScale (bsc#1212146)
* Mon Jul 03 2023 Nicolas Morey <[email protected]>
- Update to 1.18.1
- Core
- Fix build warning for ofi_dynpoll_get_fd
- EFA
- Handle 0-byte writes
- Apply byte_in_order_128_byte for all memory type
- Increase default shm_av_size to 256
- Force handshake before selecting rtm for non-system ifaces.
- Only select readbase_rtm when both sides support rdma-read
- Bugfix for initializing SHM offload
- Correct CPPFLAGS during configure
- Make setopt support sendrecv aligned 128 bytes
- Make data size to be 128 byte multiples for in-order aligned send/recv
- prepare local read pkt entry for in-order aligned send/recv.
- Disable gdrcopy and cudamemcpy for in-order aligned recv.
- Increase the pad size in rxr_pkt_entry
- Make readcopy pkt pool 128 byte aligned
- Introduce alignment to support in order aligned ops
- Fix a bug when calling ibv_query_qp_data_in_order
- RMA operations will ensure FI_ATOMIC cap
- RMA operations will ensure FI_RMA cap
- Unittest atomics without FI_ATOMIC cap.
- Unittest RMA without FI_RMA cap.
- Refactor pkt_entry assignment in poll_ibv loop
- Fixes for RDMA Write and Writedata
- RXM
- Revert rxm util peer CQ support
- Fix credit size parameter for flow ctrl
- SHM
- Fix DSA enable
- Assert read op and inject proto are mutually exclusive
- Fix ROCR data coherency
- Add FI_LOCAL_COMM to shm attrs
- Signal peer when peer is out of resources
- Handle empty freestack
- Fix bug in configure.m4 in atomics_happy assignment happy
- Add memory barrier before update resp->status for SAR
- Fix resource leak reported by coverity
- Switch cmd_ctx pool from freestack to bufpool
- Add iface parameter to smr_select_proto
- TCP
- Fix spinning on fi_trywait()
- Handle truncation of active message
- Handle prefetched data after reporting ETRUNC error
- Progress all ep's on unexp_msg_list when posting recv
- Removed unused saved_msg::ep field to fix assert
- Continue receiving after truncation error
- Create function to allocate internal msg buffer
- Add runtime setting for max saved message size
- Increase default max_saved value
- Dynamically allocate large saved Rx buffers
- Separate the max inject and recv buf size
- Remove 1-line xnet_cq_add_progress function
- Changed default wait object to epoll
- Handle case where epoll isn't natively supported
- Hold domain lock while deregistering memory
- Rename DL package from libnet to libtcp
- UCX
- Align the provider version with the libfabric version
- Verbs
- Delay device initialization to when fi_getinfo is called
- Consolidate peer_mem and dmabuf support check
- verbs_nd: Init len to 0 for WCSGetProviderPath call
- verbs_nd: Verify CQs are valid in rdma_create_qp
- verbs_nd: Initialize ibv_wc fields
- verbs_nd: Release lock in network direct error paths
- Fix vrb_add_credits signature
- Fix credit size parameter for flow ctrl
- Recover RXM connection from verbs QP in error state
- Fabtests
- Add ze-dlopen functions to component tests
- Call cudaSetDevice() for selected device
- pytest/efa: Adjust get_efa_devices()
- pytest/common: Support parallel neuron test
- pytest/common: Use different cuda device for parallel cuda set
- efa: Test_flood_peer.py increase timeout
- pytest/efa: Test to flood peer during startup
- fi-rdmabw-xe: Add option to set maximum message size
- fi-rdmabw-xe: Add option to set batch size
* Thu May 04 2023 Frederic Crozat <[email protected]>
- Add _multibuild to define additional spec files as additional
flavors.
Eliminates the need for source package links in OBS.
* Tue Apr 18 2023 Nicolas Morey <[email protected]>
- Update to 1.18.0
- Core
- rocr: fix offset calculation
- rocr: use ofi spinlock functions
- rocr: minor fixes
- neuron: convert warn to info for nrt_get_dmabuf_fd not found
- neuron: check existance of neuron devices during initialization
- neuron: Add support for neuron dma-buf
- ze: update ZE to support new driver index specification
- List variables read from config file
- Add switch to prefer system-config over environment
- Add basic system-config support for setting library variables
- Move peer provider defines into new header
- rocr: Support asynchronous memory copies
- rocr: Add support for ROCR IPC
- rocr: rename rocr data-structures
- synpaseai: return 0 for host_register and host_deregister
- fabric: Improve log level of provider mismatch
- cuda: Allow CUDA IPC when P2P disabled
- ze: add ZE command list pool to reuse command lists
- cuda: implement cuda_get_xfer_setting for non cuda build
- cuda: adjust FI_HMEM_CUDA_ENABLE_XFER behavior
- cuda.c: Add const to param to remove warning
- Add IFF_RUNNING check to indicate iface is up and running
- io_uring support enhancements
- EFA
- Implement CUDA support on instance types that do not support GPUDirect RDMA
- Implement fi_write using device's RDMA write capability
- Enrich error messages with debug and connection info
- Implement support for FI_OPT_EFA_USE_DEVICE_RDMA in fi_setopt
- Implement support for FI_OPT_CUDA_API_PERMITTED in fi_setopt
- Add support for neuron dma-buf
- Use gdrcopy to improve the intra-node CUDA communication performance for small messages
- Use shm provider's FI_AV_USER_ID support
- Fix bugs in efa provider’s shm info initialization procedure
- Hooks
- dmabuf_peer_mem: Handle IPC handle caching in L0
- trace: Add trace log for CM operation APIs
- trace: Change tag in trace log to hex format
- trace: Enhance trace log for data transfer API calls
- trace: Add trace log for API fi_cq_readerr()
- trace: Add trace log for CQ operation APIs
- Add tracing hook provider
- Net
- Net provider optimizations have been integrated into the tcp provider.
- Net provider has been removed as a reported provider.
- OPX
- Fixes for Coverity scan issues
- Enhanced tag matching
- Tune expected recv for unaligned buffers
- Add finer logging granularity
- Reduce RTS immediate data and fix packet estimate for odd TID lengths
- Add additional sources for FI_OPX_UUID
- Exclude opx from build if missing needed defines
- Move some logs to optimized builds
- Fix build warnings for unused return code from posix_memalign
- Add reliability sanity check to detect when send buffer is illegally altered
- SDMA Completion workaround for driver cache invalidation race condition
- Fix replay payload pointer increment
- Handle completion counter across multiple writes in SDMA
- Cleanup pointers after free()
- Modify domain creation to handle soft cache errors
- Two biband performance improvements
- Fixes based on Coverity Scan related to auto progress patch
- Changed poll many argument to rx_caps instead of caps
- Resync with server configured for Multi-Engines (DAOS CART Self Tests)
- Remove import_monitor as ENOSYS case
- Address memory leaks reported on OFIWG issues page
- General code cleanup
- Add replays over SDMA
- Implement basic TID Cache
- Revert work_pending check change
- Fix use_immediate_blocks
- Restore state after replay packet is NULL
- Fix memory leak from early arrival packets
- Fix segfault in SHM operations from uninitialized value in atomic path
- Prevent SDMA work entries from being reused with outstanding replays
- Set runtime as default for OPX_AV
- Fix RTS replay immediate data
- Fix errors caught by the upstream libfabric Coverity Scan
- fi_getInfo - Support multiple HFI devices
- Support OFI_PORT and Contiguous endpoint addresses for CART & Mercury
- Add fi_opx_tid.h to Makefile.include
- Fix progress checks and default domain
- Revert is_intranode simplification.
- Don't inline handle_ud_ping function
- Allow atomic fetch ops to use SDMA for sufficiently large counts
- Cleaned up FI_LOG_LEVEL=warn output
- Cleaned up unused macros for FI_REMOTE_COMM and FI_LOCAL_COMM
- Reset default progress to FI_PROGRESS_MANUAL
- Fixed GCC 10 build error with Auto Progress
- Add support for FI_PROGRESS_AUTO
- Use max allowed packet size in SDMA path when expected TID is off
- Expected receive (TID) rendezvous
- RMA Read/Write operations over SDMA
- Remove origin_rs from cts and dput packet header
- Fix for hang in DAOS CART tests
- Use single IOV for bounce buffer in SDMA requests.
- Check for FI_MULTI_RECV with bitwise OR instead of AND
- Fix for intermittent intra-node deadlock hang (DAOS CART tests)
- Fix to RPC transport error failure (DAOS CART tests)
- Fix for context->buf set to NULL
- Fix bad asserts
- Ensure atomicity of atomic ops
- fi_opx_cq_poll_inline count and head check fix
- Fix intermittent intra-node hang causing RPC timeouts (DAOS CART tests)
- PSM3
- Update provider to sync with IEFS 11.4.1.1.2
- Fix warnings from build
- Add oneapi ZE support to OFI configure
- RXD
- Ignore error path in av_close return
- RXM
- Handle NULL av in rxm_freeall_conns()
- Implement the FI_OPT_CUDA_API_PERMITTED option
- Write "len" field for remote write
- Ignore error path domain_close return
- Free coll_pool on ep close
- Update rxm to use util_cq FI_PEER support functions
- Fix incorrect CQ completion field
- Rename srx to msg_srx
- Disable FI_SOURCE if not requested
- Memory leaks removed
- Set offload_coll_mask based on actual configuration
- Report on coll offload capabilities with OFI_OFFLOAD_PROV_ONLY
- Fabric setups collective offload fabric
- Create eq for collective offload provider
- Close collective providers ep when rxm_ep is closed
- Fix incorrect use of OFI_UNUSED()
- Rework collective support to use collective provider(s)
- SHM
- Fix potential deadlock in smr_generic_rma()
- smr_generic_rma() wwrite error completion with positive errno
- Update SHM to use ROCR
- Fix incorrect discard call when cleaning up unexpected queues
- Separate smr_generic_msg into msg and tagged recv
- Fix start_msg call
- Implement the FI_OPT_CUDA_API_PERMITTED option
- Assert not valid atomic op
- Fix a bug in smr_av_insert
- Optimize locking on the SAR path
- Remove unneeded sar_cnt
- Optimize locking
- Enable multiple GPU/interface support
- Remove HMEM specific calls from atomic path
- Use util_cq FI_PEER support
- Import shm as device host memory
- Add HMEM flag to smr region
- Fix user_id support
- Write tx err comp to correct cq
- Fix index when setting FI_ADDR_USER_ID
- TCP
- Provider source has been replaced by net provider source
- Removed incorrect reporting of support for FI_ATOMIC
- Do not save unmatched messages until we have the peer's fi_addr
- Use internal flag for FI_CLAIM messages, versus a reserved tag bit
- Fix updating error counter when discarding saved messages
- Allow saved messages to be received after the underlying ep has been closed
- Enhanced debug logging in connection path
- Force CM progress on unconnected ep's when posting data transfers
- Support connect and accept calls with io_uring
- Fix segfault accessing an invalid fi_addr
- Add io_uring support for CM message exchange
- Move CM progress from fabric to EQ to improve multi-threaded performance
- Fix small memory leak destroying an EQ
- Fix race where same rx entry could be freed twice
- Handle NULL av in rdm ep cleanup
- Reduce stack use for epoll event array
- UCX
- New provider targeting Nvidia fabrics that layers over libucp
- Util
- Fix the behavior of cq_read for FI_PEER
- rocr: Fix compilation issue
- cuda: Use correct debug string calls
- Free cq->peer_cq on close
- Remove extra new line from av insert log
- Check for count = 0 in ofi_ip_av_insert
- rocr: Add support for ROCR IPC
- Add FI_PEER support to util_cq
- Disable FI_SOURCE if not requested
- Remove FID events from the EQ when closing endpoint
- Rework collective support to be a peer collective provider(s)
- Allow FI_PEER to pass CQ, EQ and AV attr checking
- Remove annoying WARNING message for FI_AFFINITY
- Add utility collective provider
- Verbs
- Implement the FI_OPT_CUDA_API_PERMITTED option
- Add support for ROCR IPC
- Fabtests
- Add fi_setopt_test unit test
- Update ze device registration calls
- fi-rdmabw-xe: Always use host buffer for synchronization
- Fix bug in posting RMA operation
- fi_cq_data: Extend test to fi_writedata
- fi_cq_data: Extend validation of completion data
- Rename fi_msg_inject tests to fi_inject_test to reflect its use
- fi_rdm_stress: Add count option to json key/pair options
- Add and fix OOB option handling in several tests
- fi_eq_test: Fix incorrect return value
- fi_rdm_multi_client: Increase the size of ep name buffer
- Add FI_MR_RAW to default mr_mode
- Support larger control messages needed by newer providers
- fi-rdmabw-xe: Update to work with the ucx provider
- fi_ubertest: Cleanup allocations in failure cases
- Change ft_reg_mr to not assume hmem iface & device
- fi_multinode: Bugfix multinode test for ze + verbs
- fi_multinode: Remove unused validation print
- fi_multinode: Skip tests for unsupported collective operations
- fi_ubertest: Fix data validation with device memory
- fi_peek_tagged: Restructure and expand test
* Mon Mar 20 2023 Nicolas Morey <[email protected]>
- Update to 1.17.1
- Core
- hmem_cuda Add const to param to remove warning
- Fix typos in fi_ext.h
- ofi_epoll: Remove unused hot_index struct member
- EFA
- Print local/peer addresses for RX write errors
- Unit test to verify no copy with shm for small host message
- Avoid unnecessary copy when sending data from shm
- Compare pci bus id in hints
- Fix double free in rxr endpoint init
- Hooks
- dmabuf_peer_mem: Handle IPC handle caching in L0
- OPX
- Exclude from build if missing needed defines
- Move some logs to optimized builds
- Fix build warnings for unused return code from posix_memalign
- Add reliability sanity check to detect when send buffer is illegally altered
- SDMA Completion workaround for driver cache invalidation race condition
- Fix replay payload pointer increment
- Handle completion counter across multiple writes in SDMA
- Cleanup pointers after free()
- Modify domain creation to handle soft cache errors
- Two biband performance improvements
- Fixes based on Coverity Scan related to auto progress patch
- Changed poll many argument to rx_caps instead of caps
- Resynch with server configured for Multi-Engines (DAOS CART Self Tests)
- Remove import_monitor as ENOSYS case
- Address memory leaks reported on OFIWG issues page
- Remove unused fields
- Fix unwanted print statement case
- Add replays over SDMA
- Implement basic TID Cache
- Revert work_pending check change
- Fix use_immediate_blocks
- Restore state after replay packet is NULL
- Fix memory leak from early arrival packets.
- Fix segfault in SHM operations from uninitialized value in atomic path.
- Prevent SDMA work entries from being reused with outstanding
replays pointing to bounce buf.
- Set runtime as default for OPX_AV
- Fix RTS replay immediate data
- Fix errors caught by the upstream libfabric Coverity Scan
- Support multiple HFI devices
- Support OFI_PORT and Contiguous endpoint addresses
- Update man pages
- Util
- util_cq: Remove annoying WARNING message for FI_AFFINITY
* Mon Dec 19 2022 Nicolas Morey <[email protected]>
- Update to 1.17.0
- Core
- Add IFF_RUNNING check to indicate iface is up and running
- General code cleanups
- Add abstraction for common io_uring operations
- Support ROCR get_base_addr
- Add a 'flags' parameter to fi_barrier()
- Introduce new calls for opening domain and endpoint with flags
- Add ability to re-sort the fi_info list
- Allowing layering of rxm over net provider
- General cleanup of provider filtering functions
- Add io_uring operations to be used by sockapi
- Modify internal handling of async socket operations
- Sockets operations are moved to a common sockapi abstraction
- Add support for Ze host register/unregister
- Add new offload provider type
- Rename fi_prov_context and simplify its use
- Convert interface prefix string checks to exact checks
- EFA
- Code cleanups and various bug fixes
- Improved debug logging and warnings and assertions
- Do not ignore hints->domain_attr->name
- Fix the calculation of REQ header size for a packet entry
- Fix default value for host memory's max_medium_msg_size
- Add tracepoints to send/recv/read ops
- Simplified emulated read protocol
- Set use_device_rdma according to efa device id
- Fix shm initialization path on error
- Fix Implementation of FI_EFA_INTER_MIN_READ_MESSAGE_SIZE
- Do not enable rdma_read if rxr_env.use_device_rdma is false
- Remove de-allocated CUDA memory region during registration
- Fix the error handling path of efa_mr_reg_impl()
- Fix rxr_ep unit tests involving ibv_cq_ex
- Add check of rdma-read capability for synapseai
- Report correct default for runt_size parameter
- Toggle cuda sync memops via environment variable.
- Net
- Continued fork of tcp provider, will eventually merge changes back
- Fix inject support
- Fix memory leak in peek/claim path
- General code cleanups and bug fixes from initial fork
- Allow looking ahead in tcp stream to handle out-of-order messages
- Add message tracing ability
- Fetch correct ep when posting to a loopback connection
- Release lock in case of error in rdm_close
- Fix error path in xnet_enable_rdm
- Add missing progress lock in srx cleanup
- Code restructuring and enhancements with longer term goal of supporting io_uring
- Disable the progress thread in most situations
- Rename DL from libxnet-fi to libnet-fi
- Add missing initialization calls for DL provider
- Add support for FI_PEEK, FI_CLAIM, and FI_DISCARD
- Include source address with CQ entry
- Fix support for FI_MULTI_RECV
- OPX
- Bug fixes and general code cleanup
- Fix progress checks and default domain
- Allow atomic fetch ops to use SDMA for sufficiently large counts
- Cleaned up FI_LOG_LEVEL=warn output
- Reset default progress to FI_PROGRESS_MANUAL
- Fixed GCC 10 build error with Auto Progress
- Add support for FI_PROGRESS_AUTO
- Use max allowed packet size in SDMA path when expected TID is turned off
- Expected receive (TID) rendezvous
- RMA Read/Write operations over SDMA
- Remove origin_rs from cts and dput packet header.
- Fix for hang - unable to match inbound packets with receive
context->src_addr (DAOS CART tests)
- Use single IOV for bounce buffer in SDMA requests.
- Check for FI_MULTI_RECV with bitwise OR instead of AND
- Fix for intermittent intra-node deadlock hang (DAOS CART tests)
- Fix to RPC transport error failure (DAOS CART tests)
- Fix for context->buf set to NULL
- Fix bad asserts
- Ensure atomicity of atomic ops
- fi_opx_cq_poll_inline count and head check fix
- Fix intermittent intra-node hang causing RPC timeouts (DAOS CART tests)
- Temporarily reduce SDMA queue ring size for possible driver bug workaround
- Fix alignment issue and asserts
- Enable more parallel SDMA operations
- PSM3
- Synced to IEFS 11.4.0.0.198
- Tech Preview Ubuntu 22.04 Support
- Tech Preview Intel DSA Support
- Improved Intel GPU Support
- Various performance improvements
- Various bug fixes
- RxM
- Always use rendezvous protocol for ZE device memory send
- Code cleanup
- Add option to free resources on AV removal
- SHM
- Fix user_id support
- Write tx err comp to correct cq
- Fix index when setting FI_ADDR_USER_ID
- Remove extraneous ofi_cirque_next() call
- Add support for FI_AV_USER_ID
- Fix multi_recv messaging
- General code restructuring for maintainability
- Implement shared completion queues
- Decouple error processing from cq completion path to avoid switch
- Fix incorrect op passed into recv cancel operation
- Enhanced SHM implementation with DSA offload
- Use multiple SAR buffers per copy operation
- Fix ZE IPC race condition on startup
- TCP
- Minor updates in preparation for io_uring support (via net provider)
- Util
- Add option to free resources on AV removal
- Add 'flags' parameter to new fi_barrier2() call
- Add debugging in ofi_mr_map_verify
- Rename internal bitmask struct to include ofi prefix
- Verbs
- Add option to disable dmabuf support
- FI_SOCKADDR includes support of FI_SOCKADDR_IB
- Fabtests
- shared: Expand hmem support
- fi_loopback: Add support for tagged messages
- fi_mr_test: add support of hmem
- fi_rdm_atomic: Fix hmem support
- fi_rdm_tagged_peek: Read messages in order, code cleanup and fixes
- fi_multinode: Add performance and runtime control options, cleanups
- benchmarks: Add data verification to some bw tests
- fi_multi_recv: Fix possible crash in cleanup
- Drop prov-net-fix-error-path-in-xnet_enable_rdm.patch which was merged upstream.
* Tue Nov 08 2022 Nicolas Morey-Chaisemartin <[email protected]>
- Add prov-net-fix-error-path-in-xnet_enable_rdm.patch to fix a deadlock
when no network interfaces are available (bsc#1205139)
* Mon Oct 10 2022 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.16.1
- Core
- Fix windows implementation to remove fd from poll set
- PSM3
- Add missing files to release tarball
- Util
- Handle NULL address insertion to fi_av_insert
- Drop prov-rxm-Disable-128-bit-atomics.patch which was merged upstream
* Thu Oct 06 2022 Nicolas Morey-Chaisemartin <[email protected]>
- Add prov-rxm-Disable-128-bit-atomics.patch to fix a potential
segfault on misaligned buffers.
* Fri Sep 30 2022 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.16.0 (jsc#PED-351, jsc#PED-190)
- Core
- Added HMEM IPC cache
- Use exact string comparison checks for network interfaces
- Restructuring of poll/epoll abstraction
- Add ability to disable locks completely in debug builds
- Serialize access to modifying the logging calls
- Minor fixes to fi_tostr text formatting
- Add hmem interface checks to memory registration
- EFA
- Added support of Synapse AI memory.
- Improved error message
- Net
- Temporarily forked, optimized version of tcp provider
- Focused on improved performance and scalability over tcp sockets
- Fork ensures tcp provider stability while net provider is developed
- Shares the tcp provider protocol and base implementation for msg endpoints
- Integrates direct support for rdm endpoints, using a derivative from rxm
- Implements own protocol for rdm endpoints, separate from rxm;tcp
- OPX
- Added initial support for SDMA
- General performance enhancements
- Performance improvements to reliability protocol
- Improved deferred work pending complete
- Added support for OPX_AV=runtime
- Support iov memory registration ops
- Added DAOS RPC support
- Atomic ops enhancements
- Improved documentation
- Debug build enhancements
- Fixed compiler warnings
- Reduced time to compile prov/opx code
- General bug fixes
- Fixed PSN wrapping scaling
- Added intranode fence
- Addressed bugs discovered by coverity scan
- PSM2
- Fix sending CQ data in some instances of fi_tsendmsg
- PSM3
- Updated to match Intel Ethernet Fabric Suite (IEFS) 11.3 release
- RxM
- Update to read multiple completions at once from msg provider
- Move RxM AV implementation to util code to share with net provider
- Minor code cleanups
- SHM
- Implement and use ipc_cache
- Add log messages for debugging and error tracking
- Fix check for FI_MR_HMEM mr_mode
- Move shm signal handlers initialization to EP
- Added log messages for errors detected
- TCP
- Fix incorrect signaling of the CQ
- Increase max number of poll events to retrieve
- Acquire ep lock prior to flushing socket in shutdown
- Verify ep state prior to progressing socket data
- Read cm error data when receiving connreq response
- Log error on connect failure
- Fix assertion failure in CQ progress function
- Util
- Fix text in log of UFFD ioctl failure
- Introduce cuda ipc monitor
- Fix CQ memory leak handling overflow
- Fix MR mode bit check for ver 1.5 and greater
- Add max_array_size to track/check array overflow
- Always progress transfers when reading from a CQ
- Handle NULL address insertion
- Try IPv4 before IPv6 addresses when starting name server
- Fix IP util av default address length
- Fix util IP getinfo path to read hints->addr_format
- Fix debug print mismatch
- Fix return code when memory allocation fails.
- Fix build sign warning in ofi_bufpool_region_alloc
- Minor code cleanups
- Print warning if an addr is inserted into an AV again
- Verbs
- Fix support of FI_SOCKADDR_IB when requested by the application
- Ensure all posted receives are flushed to the application
- Update ofi_mr_cache_search API for hmem IPC support
- Reduce logging verbosity for "no active ports"
- Fix incorrect length used in memory registration
- Various minor bug fixes for test failures
- Fix a memory leak getting IB address
- Implement verbs provider on Windows over NetworkDirect API
- Set and check address format correctly
- Only close qp if it was initialized
- Portable detection of loopback device
- Fabtests
- multi_ep: Separate EP resources and fix MR registration
- multi_recv: Fix possible crash and check for valid buffer
- unexpected_msg: Fix printf compiler warning
- dgram_pingpong.c: Use out-of-band sync
- multinode: Make multinode tests platform agnostic, fix formatting
- ubertest: Fix string comparison to include length, fix writedata completion check
- av_test: add support for -e <ep_type>
- New tests:
- dmabuf-rdma: Component level test for dma-buf RDMA
- sock_test: Component level performance test of poll, epoll, and select
- rdm_stress: Multi-threaded, multi-process stress test for RDM endpoints
- sighandler_test: Regression test for signal handler restoration
- Drop patches fixed upstream:
- prov-opx-Correctly-disable-OPX-if-unsupported.patch
- disable-flatten-attr.patch
* Mon Aug 01 2022 Martin Liška <[email protected]>
- Add disable-flatten-attr.patch that drops flatten attribute.
Note the flatten attribute results in huge compile time hog
in inliner (same the binary size would be huge).
- Use %make_build and enable LTO (boo#1133235).
- Synchronize used Patches.
* Thu Jun 23 2022 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.15.1
- Core
- Fix fi_info indentation error in fi_tostr
- hmem_ze: Add runtime option to choose specific copy engine
- Cleanup of configure HMEM checks
- Fixed stringop-truncation in ofi_ifaddr_get_speed
- Add utility provider log suffix to make logs easier to read
- Fix truncation of ipv6 addressing
- hmem: add support for AWS Trainium devices
- Fix potential sscanf overflows
- hmem: pass through device and flags when querying memory interface
- Rework locking in several areas to convert spinlocks to mutexes
- Add new locking abstractions to select lock types at runtime
- Add new FI_PROTO_RXM_TCP for optimized rxm over tcp path
- Fix windows implementation to remove fd from poll set
- EFA
- Added windows support through efawin (https://github.com/aws/efawin)
- Added support of AWS neuron.
- Added support of using gdrcopy to copy data from host to device.
- Fixed a bug that cause 0 byte read to fail.
- Fixed a memory corruption issue that can caused forked process to crash.
- Extended testing coverage through new pytest based testing framework.
- HOOKS
- Add new hooking provider dmabuf_peer_mem
- Enable DL build of hooking providers
- Add HMEM memory registration hook
- OPX
- New provider supporting Cornelis Networks Omni-path hardware
- PSM3
- Updated psm3 to match IEFS 11.2.0.0 release
- Added support for sockets (TCP/UDP) via a runtime selectable Hardware
Abstraction Layer (HAL)
- Added support for IPv6 addressing in RoCE and sockets
- Added various NIC selection filtering options (wildcarded NIC name,
address format, wildcarded IP subnet, link speed)
- Performance tuning in conjunction with OneAPI and OneCCL
- Improved PSM3_IDENTIFY output
- Rename most internal symbols to psm3_
- Corrected vulnerabilities found during Coverity scans
- configure options refined and help text improved
- PSM3_MULTI_EP has been deprecated (recommend always enabled, default
is enabled [same default as previous releases])
- Various bug fixes
- RxM
- Add check that atomic size is valid
- Add support to passthru calls to tcp provider in specific
- TCP
- Add assert to verify RMA source/target msg sizes match
- Wake-up threads blocked on CQ to update their poll events
- Fix use of incorrect events in progress handler
- Fixes for various compile warnings, mostly on Windows
- Add support for FI_RMA_EVENT capability
- Add support for completion counters
- Fix check for CQ data in tagged messages
- Add cancel support to shared rx context
- Add src_addr receive buffer matching
- Add provider control to assign a src_addr with an ep
- Handle trecv with FI_PEEK flag
- Allow binding a CQ with an SRX
- Restructuring of code in source files
- Handle EWOULDBLOCK returned by send call
- Add hot (active) pollfd
- SHM
- Properly chain the original signal handlers
- Avoid uninitialized variable with invalid atomic parameters
- Fix 0 byte SAR read
- Initialize len parameter to accept
- Refactor and simplify protocol code
- Remove broken support for 128-bit atomics
- Fix FI_INJECT flag support
- Add assert to verify RMA source/target msg sizes match
- Set domain threading to thread safe
- Fix possible use of uninitiated var in av_insert
- Util
- Fix sign warning in ofi_bufpool_region_alloc
- Remove unused variable from ofi_bufpool_destroy
- Fix check for valid datatype in ofi_atomic_valid
- Return with error if util_coll_sched_copy fails
- Fix use of uninitialized variable in ofi_ep_allreduce
- Fix memory access in ip_av_insertsym
- Track ep per collective operation not with multicast
- Restructure collective av set creation/destruction
- Change most locks from spin locks to mutexes
- Allow selection of spinlocks for CQ and domain objects
- Fix AV default addrlen
- Update fi_getinfo checks to include hints->addr_
- Handle NULL address insertion to fi_av_insert
- Verbs
- Initial changes for compiling on Windows (via NetworkDirect)
- Add a failover path to dma-buf based memory registration
- Replace use of spin locks with mutexes
- Check for valid qp prior to cleanup
- Set and check for address format correct in fi_getinfo
- Fabtests
- hmem_cuda: used device allocated host buff to fill device buf
- Add python scripts to control test execution
- test_configs: include util provider in core config file
- Add option "--pin-core"
- Only call nrt_init once
- Fix a bug in ft_neuron_cleanup
- Correct help for unit test programs
- Remove duplicate help prints from fi_mcast
- configure.ac: fix --enable-debug=no not properly detected
- msg_inject: handle the case ft_tsendmsg return -FI_EAGAIN
- Add AWS Trainium device support
- fi_inj_complete: Add FI_INJECT to fabtests
- inj_complete.c: Make arguments align with the other tests
- dgram_pingpong: handle the error return of fi_recv
- recv_cancel: Remove requirement for unexpected msg handling
- poll: Fix crash if unable to allocate pollset
- ubertest: Add GPU testing and validation support
- Add HMEM options parsing support
- Update and re-enable fi_multi_ep test
- Add prov-opx-Correctly-disable-OPX-if-unsupported.patch to disable
OPX compilation on non x86_64 systems
* Tue Apr 19 2022 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.14.1
- Core
- Use non-shared memory allocations to use MADV_DONTFORK safely
- Fix incorrect use of gdr_copy_from_mapping
- Ensure proper timeout time for pollfds to avoid early exit
- EFA
- Handle read completion properly for multi_recv
- Use shm's inject write when possible
- Support 0 byte read
- RxM
- Ensure signaling the CQ fd after writing completion
- Fix inject path for sending tagged messages with cq data
- Negotiate credit based flow control support over CM
- Add PID to CM messages to detect stale vs duplicate connections
- Fix race handling unexpected messages from unknown peers
- Fix possible leak of stack data in cm_accept
- Restrict reported caps based on core provider
- Delay starting listen until endpoint fully initialized
- Verify valid atomic size
- Sockets
- Fix coverity reports on uninitialized data
- Check for NULL pointers passed to memcpy
- Add missing error return code from sock_ep_enable
- TCP
- Fix performance regression resulting from sparse pollfd sets
- Fix assertion failure in CQ progress function
- Do not generate error completions for inject msgs
- Fix use of incorrect event names in progress handler
- Fix check for CQ data in tagged messages
- Make start_op array a static to reduce memory
- Wake-up threads blocked on CQ to update their poll events
- Verbs
- Generate error completions for all failed transmits
- Set all fields in the fi_fabric_attr for FI_CONNREQ events
- Set proper completion flags for all failed transfer
- Ensure that all attributes are provided when opening an endpoint
- Fix error handling in vrb_eq_read
- Fix memory leak in error case in vrb_get_sib
- Work-around bug in verbs HW not reported correct send opcodes
- Only call ibv_reg_dmabuf_mr when kernel support exists
- Add a failover path to dma-buf based memory registration
- Negotiate credit based flow control support over CM
* Mon Nov 22 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.14.0
- Add time stamps to log messages
- Fix gdrcopy calculation of memory region size when aligned
- Allow user to disable use of p2p transfers
- Update fi_tostr print FI_SHARED_CONTEXT text instead of value
- Update fi_tostr to output field names matching header file names
- Fix narrow race condition in ofi_init
- Add new fi_log_sparse API to rate limit repeated log output
- Define memory registration for buffers used for collective operations
- EFA, SHM, TCP, RXM, and verbs fixes
* Wed Nov 03 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Enable PSM3 provider (jsc#SLE-18754)
* Fri Oct 29 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.13.2
- Sort DL providers to ensure consistent load ordering
- Update hooking providers to handle fi_open_ops calls to avoid crashes
- Replace cassert with assert.h to avoid C++ headers in C code
- Enhance serialization for memory monitors to handle external monitors
- EFA, SHM, TCP, RxM and vers fixes
* Wed Aug 25 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.13.1
- Enable loading ZE library with dlopen()
- Add IPv6 support to fi_pingpong
- EFA, PSM3 and SHM fixes
* Wed Jul 07 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.13.0
- Fix behavior of fi_param_get parsing an invalid boolean value
- Add new APIs to open, export, and import specialized fid's
- Define ability to import a monitor into the registration cache
- Add API support for INT128/UINT128 atomics
- Fix incorrect check for provider name in getinfo filtering path
- Allow core providers to return default attributes which are lower then
maximum supported attributes in getinfo call
- Add option prefer external providers (in order discovered) over internal
providers, regardless of provider version
- Separate Ze (level-0) and DRM dependencies
- Always maintain a list of all discovered providers
- Fix incorrect CUDA warnings
- Fix bug in cuda init/cleanup checking for gdrcopy support
- Shift order providers are called from in fi_getinfo, move psm2 ahead of
psm3 and efa ahead of psmX
- See NEWS.md for changelog
* Fri Apr 02 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.12.1
- Fix initialization checks for CUDA HMEM support
- Fail if a memory monitor is requested but not available
- Adjust priority of psm3 provider to prefer HW specific providers,
such as efa and psm2
- EFA and PSM3 fixes
- See NEWS.md for changelog
* Tue Mar 09 2021 Nicolas Morey-Chaisemartin <[email protected]>
- Update to 1.12.0
- See NEWS.md for changelog
/usr/bin/fi_info /usr/bin/fi_pingpong /usr/bin/fi_strerror /usr/share/doc/packages/libfabric /usr/share/doc/packages/libfabric/NEWS.md /usr/share/licenses/libfabric /usr/share/licenses/libfabric/COPYING /usr/share/man/man1/fi_info.1.gz /usr/share/man/man1/fi_pingpong.1.gz /usr/share/man/man1/fi_strerror.1.gz
Generated by rpm2html 1.8.1
Fabrice Bellet, Sun Dec 8 23:52:46 2024