-
Notifications
You must be signed in to change notification settings - Fork 14
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Huawei 1230 #49
Huawei 1230 #49
Conversation
- added iface capability UCT_IFACE_FLAG_EP_KEEPALIVE to identify EP's which can provide keepalive functionality without UCX involving
…m_type GTEST/UCP: Expect SW RMA when managed memory is used
Signed-off-by: JonasZhou <JonasZhou@zhaoxin.com>
API/UCT: added UCT_IFACE_FLAG_EP_KEEPALIVE caps
UCP/AM: AM rndv fixes
- added support of keepalive feature for TCP transport
IB: Decrease log level for testing addr family gid index.
UCS/TIME-UNITS: added "inf" and "auto" values
UCP/API: Add recv_info to ucp_request_param_t
UCS/UCT/IB: Select SL not depending on AR support by default
UCS/ARCH: Add Zhaoxin cpu detection
UCP: SOCKADDR_CM_ENABLE=y by default
…ernal_failed UCP/EP/FLUSH: fix too early completion for failed EP
- If single fragment AM received without UCP_AM_SEND_REPLY flag it should be delivered to the user even if the corresponding rx ep is closed/failed. - If single fragment AM received with UCP_AM_SEND_REPLY flag it should be droppped if the corresponding rx ep is closed/failed, because reply ep can not be provided in the data callback. - If some fragment of multi-fragmented AM received and the corresponding rx ep is closed/failed, this fragment should be dropped regardless of send AM flags (no way to assemble a message without aux info storedd in ep extension) - If AM RTS is received and the corresponding rx ep is closed/failed, this RTS should be droppped and ATS with EP_TIMEOUT should be sent back to the sender.
* Renaming from wfe to wait_mem * Using average of ten performance runs as a reference point * Moving wait_mem to ucp_perf * Code styling fixes Signed-off-by: Pavel Shamis (Pasha) <pasharesearch@gmail.com>
…el_nb-10 UCP/EP/CLOSE: Make close EP discard lanes directly
…st-to-check TEST/TAG: Fix tag test to check inline data flag
…-test JUCX: catch exception on ep close.
…ad_test GTEST/UCP/TAG: Fix offload thresh check
UCP/AM: force eager protocol for old AM API - v1.10.x
…er_v.1.10 JUCX: Do not delete reference to listener connHandler (v.1.10.x)
AZP/RELEASE: Remove strict libibverbs dependency
AZP: Fix snapshot v1 10
UCT/TCP: add enable loopback flag. [v1.10]
* Adding missing items * Fixing news format to follow previous releases Signed-off-by: Pavel Shamis (Pasha) <pasharesearch@gmail.com>
NEWS: News update
UCT/DC/MLX5: Create DCI via DevX with full handshake option - v1.10.x
UCP/AM: Adjust max_short for UCP_AM_SEND_REPLY - v1.10.x
may happen if the list of components that support CM is longer than the available cms on the host (worker->cms).
…ss-null-cm-v1-10 UCP: handle a case of a null cm on the worker - v1.10.x
Signed-off-by: Pavel Shamis (Pasha) <pasharesearch@gmail.com>
NEWS: News update before release v1.10.1
Offering:hpc
Offering:hpc
Thanks for your pull request. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). The following commits have not yet signed CLA. 08b90b5 | IB: Decrease log level for testing addr family gid index.
GTEST/UCP: Expect SW RMA when managed memory is used Signed-off-by: JonasZhou JonasZhou@zhaoxin.com API/UCT: added UCT_IFACE_FLAG_EP_KEEPALIVE caps UCP/AM: AM rndv fixes
IB: Decrease log level for testing addr family gid index. UCS/TIME-UNITS: added "inf" and "auto" values UCP/API: Add recv_info to ucp_request_param_t UCS/UCT/IB: Select SL not depending on AR support by default UCS/ARCH: Add Zhaoxin cpu detection UCP: SOCKADDR_CM_ENABLE=y by default UCP/EP/FLUSH: fix too early completion for failed EP
Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com UCP/EP/CLOSE: Make close EP discard lanes directly TEST/TAG: Fix tag test to check inline data flag JUCX: catch exception on ep close. GTEST/UCP/TAG: Fix offload thresh check UCS/ARCH: Add SVE memcpy GTEST: Moving memory_wait test to perf IODEMO: add listen retry for server UCT: Flush TX CODE: Add style check support If a listener is destroyed while it has a pending connect request which UCP/AM: Drop AM data if rx ep is closed TCP/KEEPALIVE: enabled ep_check feature for TCP Jenkins: Add function for check state on device BUILD/JAVA: Fix build dependencies UCS/CONFIGURE: aarch64 default is to have HW_TIMER set, change config.h comment text to reflect that CONTRIB: Add valgrind suppression for rdma_create_event_channel UCP/RNDV/ROCM: add support for staging rndv protocol for ROCm
TOOLS/PERF: Fix test name printing format UCP: implement ucp_tag_msg_recv_nb as ucp_tag_msg_recv_nbx. UCP: copy user data only if flag is present. DOC: updated news. UCT/IB: get roce ndev name according to right gid but not fixed gid 0
JENKINS: update cuda module version to 11.1.1
UCP/WIREUP/GTEST: Fix dead code in CM disconnect
AZP: add warning for azure Fixes for iodemo and listener destroy flow UCS/CONFIG: fixed crash on incorrect value set UCS/DATASTRUCT: Add bitmap data struct UCP/AMO: Use remote_addr from AMO part of request instead of RMA UCP: Recv msg nbx routine. UCP/KEEPALIVE: removed incorrect assert API/UCT/IFACE: added keepalive_timeout value GTEST/UCP: Skip wireup 1sided disconnect test for TCP (workaround) UCP/WIREUP: use ucs topo to compare with mem_type md and adjust latency
TEST/IO_DEMO: Add a window on the client side, per every remote server IB/RDMACM: Add local and remote addresses to the reject error message. Ignore PFs without RDMA cap on BF2 according to its gid_tlb_len==0 UCP/HELLO-WORLD: added error simulations CONTRIB: Add valgrind suppression for rdma_bind_addr() UCP: Fix address unpacking error
UCP/CORE/GTEST: Rearrange fields in a UCP request to reduce its size JUCX: UCS Memory type constants. UCS/MPOOL: Make elem defined for Valgrind UCP/NBX: fixed external request free from CB UCP/GTEST: Fix and test ucp requests leak from the ptr_map UCT/CM/RDMACM: share dummy CQ per device JUCX: Prevent clang of formatting java files.
Since we may create RC QP from the progress thread in wireup_cm pack
UCP/TCP/KEEPALIVE: added processing of auto time UCT/RC: Protect rc_iface->ep_list and rc_iface->eps with a spinlock UCT/IB: Fixes for SL selection (v1.10.x)
(cherry picked from commit daa69c5) RC_MLX5/IFACE: fixed assert - v1.10 UCP/MEM_TYPE: Adjust mem type zcopy thresh if user sets UCX_ZCOPY_THRESH - v1.10.x UCT/IB/DEVX: Set modify-QP global address parameters only for GRH case - v1.10.x
(cherry picked from commit 4b4df04) UCT/IB/MLX5: DV UAR alloc type NC support - v1.10 UCP/TAG: Fix offload completion with inlined data - v.1.10.x DOC/RECV-NBX: removed incorrect note - v1.10 UCP/RNDV: Fix releasing of local request ID when switching to RNDV AM (v1.10.x) AZP: fix docker image UCP/PROTO: Handle AM short failure correctly [v1.10.x] UCT/IB/RC: Handle multiple flush cancel w/o completion [v1.10.x] use CM fallback on rdmacm route_resolve error. UCS/UCT/JENKINS: Enable the loopback IP as a tcp and testing resource
Introduce a minimal interval for sending packets with SOLICITED flag, UCT: Prevent segfault in uct_rdmacm_cm_ep_str when id is not initialized [v1.10.x] UCT/UD: Don't wake up remote peer for every ACKREQ packet -v1.10.x UCT/RDMACM: decrease log level from error to diag v1.10.x UCP/UCT/MD: add diag prints to mem_reg path, v1.10.x News format was changed slightly to include additional information Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com NEWS: News update before release AZP/IODEMO: fix get PID in corrupter - v1.10.x UCP/CORE/RNDV/GTEST: Drop packets with invalid ID and fix handling of status in RNDV RTS/RTR/data [v1.10.x]
DEB/PKG: add essential system dependencies. v1.10 CONFIG/TEST: Fix compilation on gcc11 - v1.10.x Set HAVE_HW_TIMER to zero for cross compile built With the rdma-core version 28 onwards the error code Fixing the error code check to look for the right Signed-off-by: Devesh Sharma devesh.sharma@broadcom.com UCT/IB: fix cq creation failure using old ibv api UCS: A fix for cross compilation support in configure - v1.10.x
UCP/AM: Fix releasing of deferred data - v1.10.x IB/ADDRESS: pack MTU value for non 4K value - v1.10 UCT/RC: Fix QP destroy for EXP flow - v1.10 Conflicts: UCP/RMA: Fix length check condition in RMA PUT short - v1.10.x UCP/RNDV: Set addr NULL in RTS if reg md_map is NULL. - v1.10.x UCP/CORE/WIREUP/GTEST: Fix check intersection with CM initial configuration [v1.10.x] Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com AUTHORS: Updating the list of authors - v1.10.0 NEWS: Update before RC3 - v1.10.0 AZP: add new OS for release CI UCP/CORE/WIREUP: Add missing async blocks (v1.10.x) UCS/TOPO: Use common prefix and char count for path distance estimation - v1.10.x UCT/CUDA_IPC: make cuda-ipc cache global - v1.10.x Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com NEWS: Updating news to reflect rc4 changes (cherry picked from commit cb69bcc) UCP/AM: Use correct ep for short send with reply - v1.10.x Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com NEWS: News update for v1.10.0 rc5 AZP: do not include libjucx into release packages. [v.1.10] NEWS: remove jucx from packages. [v.1.10] AZP/RELEASE: Add dockers with cuda11.2 for v1.10 NEWS: Update to v1.10 release date NEWS/BUILD: Update Build.Reason trigger and release date -v1.10.x
(cherry picked from commit e7894da) LIBPERF: fixed incorrect error handling - v1.10 UCT/IB: Fix port width check on HDR100 - v1.10.x AZP/RELEASE: launch jucx in lab UCS/SOCK: Fix RPM build with gcc11 on fedora34 - v1.10.x CONFIG/SPEC: Bump version to 1.10.1 UCP/RNDV: Fix in mem type pipeline - v1.10.x (cherry picked from commit bc408b3) UCP/AM: force eager protocol for old AM API - v1.10.x JUCX: Do not delete reference to listener connHandler (v.1.10.x) AZP/RELEASE: Remove strict libibverbs dependency AZP: Fix snapshot v1 10 UCT/TCP: add enable loopback flag. [v1.10]
Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com NEWS: News update UCT/DC/MLX5: Create DCI via DevX with full handshake option - v1.10.x UCP/AM: Adjust max_short for UCP_AM_SEND_REPLY - v1.10.x may happen if the list of components that support CM is longer than the UCP: handle a case of a null cm on the worker - v1.10.x Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com NEWS: News update before release v1.10.1 📝 Please access here to sign the CLA. It may take a couple minutes for the CLA signature to be fully registered; after that, please reply here with a new comment: /check-cla to verify. Thanks.
|
Alina Sklarevich seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
No description provided.