Functions — github.com/aws/aws-ofi-nccl

CodeHub MCPcopy Create free account

hub / github.com/aws/aws-ofi-nccl / functions

Functions1,193 in github.com/aws/aws-ofi-nccl

⨍ Functions 1,193◇ Types & classes 476

↓ 2 callersMethodrdma_device_get_recv_comm

* @brief Get endpoint recv communicator with given comm_id */

include/nccl_ofi_rdma.h:1814

↓ 2 callersMethodrdma_endpoint_abort

src/nccl_ofi_rdma.cpp:4109

↓ 2 callersMethodrecv

src/cm/nccl_ofi_cm_resources.cpp:292

↓ 2 callersMethodreg_mr_on_device

src/nccl_ofi_rdma.cpp:2826

↓ 2 callersMethodrelease_all_domain_and_ep

src/nccl_ofi_net.cpp:800

↓ 2 callersMethodrelease_all_ep

src/nccl_ofi_net.cpp:900

↓ 2 callersFunctionrma_write_impl

* @brief Write using DMA writemsg */

src/nccl_ofi_rdma.cpp:6423

↓ 2 callersFunctionrx_get_close_msg

* Get close message from rx buffer */

src/nccl_ofi_rdma.cpp:165

↓ 2 callersFunctionsend_comm_destroy

src/nccl_ofi_rdma.cpp:3936

↓ 2 callersMethodsend_receive_test

* Test send/receive with fresh buffer allocation per call (like master) */

tests/functional/functional_test.cpp:653

↓ 2 callersFunctionsendrecv_comm_mr_base_reg

src/nccl_ofi_sendrecv.cpp:750

↓ 2 callersMethodsendrecv_endpoint_abort

src/nccl_ofi_sendrecv.cpp:1001

↓ 2 callersFunctionsendrecv_get_local_address

* @brief Query local address for a libfabric endpoint * * @param Network device * * @return Local EP address, on success * NULL, others */

src/nccl_ofi_sendrecv.cpp:1582

↓ 2 callersFunctionsendrecv_mr_base_register

src/nccl_ofi_sendrecv.cpp:689

↓ 2 callersFunctionsendrecv_mr_buffers_register

* @brief Registers memory region (both HOST and CUDA) * * @return OFI memory handle for data transfer operations * @return 0 on success * non-zer

src/nccl_ofi_sendrecv.cpp:510

↓ 2 callersFunctionsendrecv_req_free

* @brief Prepares sendrecv request for reuse */

src/nccl_ofi_sendrecv.cpp:362

↓ 2 callersMethodset_conn_resp_msg_data

src/cm/nccl_ofi_cm.cpp:101

↓ 2 callersMethodset_gin_resources

include/nccl_ofi_rdma.h:1377

↓ 2 callersMethodset_info

* Set the trace information for LTTNG and NVTX * @param dev_arg: device ID * @param rank_arg: rank number * @param msg_seq_num_arg: message sequ

include/rdma/gin/nccl_ofi_gin_reqs.h:118

↓ 2 callersMethodsetup_connection

* Setup connection for a specific device using context's own fields */

tests/functional/functional_test.cpp:441

↓ 2 callersMethodteardown

tests/functional/functional_test.cpp:770

↓ 2 callersMethodunlock

* Manually unlock the mutex. After this function is called, the mutex * is no longer owned by this class. */

include/nccl_ofi_pthread.h:125

↓ 2 callersMethodwrite

src/nccl_ofi_rdma.cpp:6511

↓ 2 callersFunctionwrite_cpu_opening_tag

* @brief Write cpu opening tag to NCCL topology file * * @param node * Topology node of type `HWLOC_OBJ_NUMANODE` * @param file * Output file

src/nccl_ofi_topo.cpp:1400

↓ 2 callersFunctionzero_nccl_ofi_req

* @brief Zero out rdma request */

src/nccl_ofi_rdma.cpp:1909

↓ 1 callersFunctionNCCL_OFI_ROUND_UP_TO_POWER_OF_TWO

include/nccl_ofi_math.h:39

↓ 1 callersMethodaccept

src/cm/nccl_ofi_cm.cpp:66

↓ 1 callersFunctionaccept_wait_for_connection

src/nccl_ofi_rdma.cpp:4844

↓ 1 callersFunctionaddition_check

tests/unit/environ.cpp:35

↓ 1 callersMethodalloc_and_reg_flush_buff

* @brief Allocate and register buffer to flush RDMA operations. On * Success, receive communicator holds reference to flush buffer * and associa

src/nccl_ofi_sendrecv.cpp:1223

↓ 1 callersMethodalloc_and_reg_flush_buff

* @brief Allocate and register a buffer used to flush RDMA operations. * * See header for full documentation. */

src/nccl_ofi_rdma.cpp:3556

↓ 1 callersMethodalloc_comm_id

include/rdma/gin/nccl_ofi_gin_resources.h:281

↓ 1 callersFunctionalloc_rdma_read_req

src/nccl_ofi_rdma.cpp:4400

↓ 1 callersFunctionalloc_rdma_send_req

src/nccl_ofi_rdma.cpp:5175

↓ 1 callersFunctionalloc_rdma_write_req

src/nccl_ofi_rdma.cpp:5150

↓ 1 callersMethodallocate_recv_req

* @brief Allocate a new recv req from freelist */

src/nccl_ofi_rdma.cpp:3202

↓ 1 callersMethodanalyze_all

(self, min_size: int = 32, max_size: int = 32 * 1024 * 1024 * 1024)

contrib/python/ofi_nccl/tuner/cli/wrapper.py:221

↓ 1 callersMethodanalyze_message_range

(self, coll_type: NCCLFunc, min_size: int, max_size: int)

contrib/python/ofi_nccl/tuner/cli/wrapper.py:186

↓ 1 callersMethodawait_pending_requests

src/rdma/gin/nccl_ofi_gin.cpp:393

↓ 1 callersFunctionbuild_rdma_write_wqe

tests/functional/efa_direct_wqe.cpp:57

↓ 1 callersFunctioncalculateBestNChannelPat

* Calculate the best number of channel for PAT. Based on empirical data. */

src/tuner/nccl_ofi_regions.cpp:1932

↓ 1 callersFunctioncalculateBestNChannelTree

src/tuner/nccl_ofi_regions.cpp:1944

↓ 1 callersFunctioncalculateChunkSizeTreeLL128

src/tuner/nccl_ofi_regions.cpp:1874

↓ 1 callersFunctioncheck_custom

tests/unit/histogram_binner.cpp:39

↓ 1 callersFunctioncheck_histogram

tests/unit/histogram.cpp:36

↓ 1 callersFunctioncheck_invalid_variable

tests/unit/param.cpp:116

↓ 1 callersFunctioncheck_known_platforms

tests/unit/aws_platform_mapper.cpp:48

↓ 1 callersFunctioncheck_late_set

tests/unit/param.cpp:144

↓ 1 callersFunctioncheck_linear

tests/unit/histogram_binner.cpp:20

↓ 1 callersFunctioncheck_sources

tests/unit/param.cpp:126

↓ 1 callersFunctioncheck_string_to_value

tests/unit/param.cpp:14

↓ 1 callersFunctioncheck_timer_histogram

tests/unit/histogram.cpp:92

↓ 1 callersFunctioncheck_value_to_string

tests/unit/param.cpp:103

↓ 1 callersMethodcleanup_connection

* Cleanup connection at specific index using context's own vectors */

tests/functional/functional_test.cpp:561

↓ 1 callersFunctionclear_read_reqs_pending_back_pointers

Walk all MAX_NUM_RAILS slots — including any subreqs that returned -FI_EAGAIN on post() and got queued via add_pending_req — so none retain a pendi

src/rdma/gin/nccl_ofi_gin.cpp:678

↓ 1 callersFunctionclose_listen_recv_comm

* @brief Close receive communicator if listen request is not pending * Assumed to hold domain lock */

src/nccl_ofi_rdma.cpp:4829

↓ 1 callersMethodclose_ofi_ep

src/cm/nccl_ofi_cm_resources.cpp:309

↓ 1 callersMethodclose_ofi_eps

src/rdma/gin/nccl_ofi_gin_resources.cpp:166

↓ 1 callersMethodcomplete_init

src/nccl_ofi_rdma.cpp:7396

↓ 1 callersMethodconfig_endpoint

src/platform-aws.cpp:718

↓ 1 callersMethodconnect

src/rdma/gin/nccl_ofi_gin.cpp:123

↓ 1 callersFunctioncount_nodes_with_accel_or_nic_in_subtree

* @brief Count number of topology nodes that have a NIC or Nvidia GPU in its subtree * * * @param topo * Hwloc topology. Userdata pointers of top

src/nccl_ofi_topo.cpp:449

↓ 1 callersMethodcreate_device_rail_array

src/nccl_ofi_rdma.cpp:7176

↓ 1 callersMethodcreate_domain

src/nccl_ofi_rdma.cpp:7127

↓ 1 callersMethodcreate_endpoint

src/nccl_ofi_rdma.cpp:6995

↓ 1 callersFunctioncreate_groups_from_info_list

* @brief Split NIC info list into 'num_groups' and add each group to the * topology node corresponding to its leader (first NIC of the list). * *

src/nccl_ofi_topo.cpp:982

↓ 1 callersFunctioncreate_groups_from_info_lists

* @brief Split libfabric NIC info lists of topology nodes with 'num_groups' > 0 * into 'num_groups' lists (groups) and add these lists * to the co

src/nccl_ofi_topo.cpp:1074

↓ 1 callersFunctioncreate_ref_schedule

tests/unit/scheduler.cpp:60

↓ 1 callersMethodcreate_send_comm

src/nccl_ofi_rdma.cpp:6536

↓ 1 callersFunctionctrl_rx_buff_req_free

src/nccl_ofi_rdma.cpp:2168

↓ 1 callersMethoddecrease_rx_buff_cnt

src/nccl_ofi_rdma.cpp:836

↓ 1 callersMethoddecrement_outstanding_ack_counter

include/rdma/gin/nccl_ofi_gin.h:269

↓ 1 callersMethoddereg_mr_no_lock

src/nccl_ofi_rdma.cpp:2777

↓ 1 callersMethodderegmr_fn

C-style shim matching nccl_ofi_freelist_deregmr_fn; pass to freelist constructors. */

src/nccl_ofi_rdma.cpp:3120

↓ 1 callersMethoddevice_get_guid

src/platform-aws.cpp:939

↓ 1 callersMethoddevice_prepare_for_connection

src/nccl_ofi_rdma.cpp:7151

↓ 1 callersFunctiondigestHash

src/nccl_ofi_nccl_compat.cpp:80

↓ 1 callersFunctiondistance

* @brief Distance between x and nearest point on y0->y1 segment. * * @param x * point. * @param y0 * point at the beginning of the segment. *

src/tuner/nccl_ofi_regions.cpp:124

↓ 1 callersMethodeager_copy_to_sub_recv

* @brief Initiate eager copy for a specific sub-recv. * * For sub_idx == 0, uses the existing eager_copy_req path. * For sub_idx > 0, we complete v

src/nccl_ofi_rdma.cpp:910

↓ 1 callersFunctioneager_entry_less

include/nccl_ofi_rdma.h:835

↓ 1 callersFunctioneager_rx_buff_req_free

src/nccl_ofi_rdma.cpp:2122

↓ 1 callersFunctioneager_seq_before

Wrap-aware "a precedes b" for the 16-bit eager sequence. Correct as long as * the number of in-flight eager batches stays below 2^15, which is guaran

include/nccl_ofi_rdma.h:830

↓ 1 callersFunctioneatHash

The hash isn't just a function of the bytes but also where the bytes are split into different calls to eatHash().

src/nccl_ofi_nccl_compat.cpp:58

↓ 1 callersFunctionenable_hwloc_io_types

* brief Enable I/O discovery for hardware topology * * Hwloc I/O discovery is disabled by default. This function enables * topology discovery of I/

src/nccl_ofi_topo.cpp:246

↓ 1 callersMethodentry_set_defined

* Set memcheck guards of freelist entry's user data to accessible and defined */

include/nccl_ofi_freelist.h:234

↓ 1 callersMethodentry_set_undefined

* Set memcheck guards of freelist entry's user data to accessible but undefined */

include/nccl_ofi_freelist.h:220

↓ 1 callersMethodexecute

tests/functional/functional_test.cpp:779

↓ 1 callersFunctionfinish_connect

* @brief Execute second part of the connect functionality from listen/connect/accept * connection establishment * * Initialize communicator rails

src/nccl_ofi_rdma.cpp:2394

↓ 1 callersMethodflush

src/nccl_ofi_rdma.cpp:4206

↓ 1 callersFunctionfree_eager_copy_req

src/nccl_ofi_rdma.cpp:848

↓ 1 callersFunctionfree_flush_req

* @brief Free flush request */

src/nccl_ofi_rdma.cpp:2103

↓ 1 callersFunctionfree_read_req

* @brief Free read request */

src/nccl_ofi_rdma.cpp:1976

↓ 1 callersFunctionfree_recv_req

* @brief Free receive request */

src/nccl_ofi_rdma.cpp:2026

↓ 1 callersFunctionfree_recv_segms_req

* @brief Free receive segments request */

src/nccl_ofi_rdma.cpp:2061

↓ 1 callersFunctionfree_send_close_req

* @brief Free send close request */

src/nccl_ofi_rdma.cpp:2075

↓ 1 callersFunctionfree_send_req

* @brief Free send request */

src/nccl_ofi_rdma.cpp:1990

↓ 1 callersFunctionfree_write_req

* @brief Free write request */

src/nccl_ofi_rdma.cpp:1963

↓ 1 callersFunctiongenerate_topology_csv

(input_file, output_file)

contrib/scripts/topology_aware/hostfile-topologify.py:29

↓ 1 callersFunctiongetHash

src/nccl_ofi_nccl_compat.cpp:90

↓ 1 callersFunctiongetHostHash

src/nccl_ofi_nccl_compat.cpp:146

← previous next →301–400 of 1,193, ranked by callers