Functions — github.com/ai-dynamo/nixl

CodeHub MCPcopy Create free account

hub / github.com/ai-dynamo/nixl / functions

Functions3,108 in github.com/ai-dynamo/nixl

⨍ Functions 3,108◇ Types & classes 683↳ Endpoints 1

↓ 4 callersFunctionfindTomlNode

src/utils/common/configuration.cpp:113

↓ 4 callersFunctionfind_plugin

Helper function to find a plugin by name

src/bindings/rust/tests/tests.rs:143

↓ 4 callersFunctionformat_data_size

test/unit/plugins/hf3fs/nixl_hf3fs_mt_test.cpp:71

↓ 4 callersMethodget

src/plugins/ucx/mem_list.cpp:75

↓ 4 callersMethodgetAgent

test/gtest/unit/agent/agent.cpp:86

↓ 4 callersMethodgetAvgNicBandwidth

* @brief Retrieves the average NIC bandwidth in Gbps (decimal, as multiple of 10^9, not 2^30). * This is the speed as reported by fi_getinfo().

src/utils/libfabric/libfabric_topology.h:291

↓ 4 callersMethodgetAvgNicUpstreamBandwidth

* @brief Retrieves the average NIC upstream link bandwidth in Gbps (decimal, as multiple of * 10^9, not 2^30). This is the link speed of the PCIe

src/utils/libfabric/libfabric_topology.h:300

↓ 4 callersMethodgetBackendParams

test/gtest/test_transfer.cpp:118

↓ 4 callersMethodgetLocalMD

test/gtest/plugins/uccl/uccl_test.cpp:184

↓ 4 callersMethodgetMD

test/gtest/plugins/memory_handler.h:124

↓ 4 callersMethodget_chunk

(self, size, offset)

benchmark/kvbench/test/custom_traffic_perftest.py:86

↓ 4 callersMethodget_cq

src/plugins/gpunetio/verbs/verbs.h:59

↓ 4 callersMethodget_io_size

Calculates the IO size for one token per GPU for the Qwen3 model. Args: page_size (int): Size of memory pages for IO ope

benchmark/kvbench/models/qwen3.py:88

↓ 4 callersMethodget_lkey

src/plugins/gpunetio/verbs/verbs.h:151

↓ 4 callersMethodget_phase

Get current phase index.

examples/device/ep/tests/elastic/plan.py:101

↓ 4 callersMethodget_plugin_list

src/api/python/_api.py:301

↓ 4 callersFunctionget_precision_size

(precision: str)

benchmark/kvbench/models/utils.py:23

↓ 4 callersMethodget_serialized_descs

src/api/python/_api.py:1113

↓ 4 callersMethodget_telemetry

Gets telemetry data for this transfer request # Returns Transfer telemetry data containing timing and performance metrics # Errors `NoTelemetry` -

src/bindings/rust/src/xfer.rs:69

↓ 4 callersMethodget_throughput_mbps

test/unit/plugins/hf3fs/nixl_hf3fs_mt_test.cpp:132

↓ 4 callersFunctionhash_tensor

(t: torch.Tensor)

examples/device/ep/tests/utils.py:277

↓ 4 callersMethodht_dispatch

examples/device/ep/csrc/nixl_ep.cpp:615

↓ 4 callersMethodinit

test/gtest/error_handling.cpp:160

↓ 4 callersFunctioninitTestBlobEngine

test/gtest/unit/azure_blob/azure_blob.cpp:229

↓ 4 callersMethodinvalidate_local_metadata

( self, ip_addr: str = "", port: int = DEFAULT_COMM_PORT )

src/api/python/_api.py:912

↓ 4 callersMethodisInitiator

benchmark/nixlbench/src/worker/worker.cpp:187

↓ 4 callersMethodloadRemoteMD

test/gtest/plugins/uccl/uccl_test.cpp:191

↓ 4 callersMethodmem_type

src/bindings/rust/src/lib.rs:547

↓ 4 callersFunctionper_token_cast_back

(x_fp8: torch.Tensor, x_scales: torch.Tensor)

examples/device/ep/tests/utils.py:80

↓ 4 callersFunctionpopulateDescs

test/unit/plugins/mooncake/mooncake_backend_test.cpp:361

↓ 4 callersFunctionpopulateDescs

test/unit/plugins/ucx/ucx_backend_test.cpp:329

↓ 4 callersMethodprepMemView

src/core/nixl_agent.cpp:1727

↓ 4 callersMethodprep_xfer_dlist

( self, agent_name: str, xfer_list, mem_type: Optional[str] = None, ba

src/api/python/_api.py:495

↓ 4 callersFunctionprintParams

test/nixl/agent_example.cpp:528

↓ 4 callersFunctionprintParams

examples/cpp/nixl_example.cpp:47

↓ 4 callersFunctionprintParams

examples/cpp/nixl_etcd_example.cpp:60

↓ 4 callersFunctionprint_usage

test/unit/plugins/gds_mt/nixl_gds_mt_test.cpp:80

↓ 4 callersFunctionprint_usage

test/unit/plugins/cuda_gds/nixl_gds_test.cpp:67

↓ 4 callersMethodprogressActiveRails

src/utils/libfabric/libfabric_rail_manager.cpp:994

↓ 4 callersMethodregisterMemory

src/utils/libfabric/libfabric_rail.cpp:1283

↓ 4 callersMethodrelease

src/utils/libfabric/libfabric_rail.cpp:53

↓ 4 callersMethodresize

Resizes the list to the given size

src/bindings/rust/src/descriptors/reg.rs:181

↓ 4 callersFunctionsendCommMessage

src/core/nixl_listener.cpp:107

↓ 4 callersMethodserialize

Serializes the descriptor list to a byte vector using bincode

src/bindings/rust/src/descriptors/reg.rs:238

↓ 4 callersMethodsupportsNotif

test/gtest/mocks/mock_backend_engine.h:43

↓ 4 callersMethodsynchronize

benchmark/nixlbench/src/worker/worker.cpp:124

↓ 4 callersMethodunloadMD

src/plugins/ucx/ucx_backend.cpp:1022

↓ 4 callersMethodwarnAboutHardwareSupportMismatch

src/plugins/ucx/ucx_utils.cpp:607

↓ 4 callersFunctionwarn_cu_api

Log a non-fatal warning if a CUDA driver API call failed (e.g. during teardown). */

examples/device/ep/csrc/vmm.cpp:29

↓ 4 callersFunctionxferOpStr

src/core/nixl_enum_strings.cpp:41

↓ 3 callersMethodBuffer

examples/device/ep/csrc/nixl_ep.cpp:80

↓ 3 callersMethod_barrier_tp

Barrier for a traffic pattern

benchmark/kvbench/test/custom_traffic_perftest.py:147

↓ 3 callersMethod_prepare_tp

Timing everything in this function because it takes a lot of time

benchmark/kvbench/test/custom_traffic_perftest.py:212

↓ 3 callersFunction_run_xfer_telemetry_check

(agent1, agent2, expect_telemetry: bool = True)

test/python/test_nixl_api.py:250

↓ 3 callersMethodaddWorker

src/plugins/ucx/ucx_backend.cpp:225

↓ 3 callersMethodaddr

benchmark/nixlbench/src/worker/nixl/nixl_worker.cpp:388

↓ 3 callersMethodall_reduce

All-reduce operation with specified reduction operation

benchmark/kvbench/runtime/rt_base.py:55

↓ 3 callersMethodas_slice

Get a slice of the underlying data

src/bindings/rust/src/lib.rs:531

↓ 3 callersFunctionbench_kineto

( fn, kernel_names: Union[str, tuple], num_tests: int = 30, suppress_kineto_output: bool = Fal

examples/device/ep/tests/utils.py:189

↓ 3 callersMethodcapture

Capture a CUDA event on the current stream, i.e. `torch.cuda.current_stream()`. Returns: event: the captured event.

examples/device/ep/nixl_ep/buffer.py:129

↓ 3 callersMethodchange_unique

test/unit/plugins/gusli/nixl_gusli_test.cpp:56

↓ 3 callersFunctioncheckExistence

src/utils/common/configuration.h:151

↓ 3 callersMethodcheckXfer

src/plugins/ucx/ucx_backend.cpp:1303

↓ 3 callersFunctioncheck_data

(check_x, recv_gbl_rank_prefix_sum)

examples/device/ep/tests/test_ht.py:166

↓ 3 callersMethodclearActiveRails

src/utils/libfabric/libfabric_rail_manager.cpp:1257

↓ 3 callersMethodcode

src/plugins/posix/posix_backend.h:71

↓ 3 callersMethodcombine

(self, *args, **kwargs)

examples/device/ep/nixl_ep/buffer.py:519

↓ 3 callersMethodcomplete

src/plugins/ucx/ucx_backend.cpp:457

↓ 3 callersMethodconnect_ranks

examples/device/ep/csrc/nixl_ep.cpp:467

↓ 3 callersMethodcreateXferReq

test/gtest/plugins/uccl/uccl_test.cpp:197

↓ 3 callersMethoddestroy

benchmark/kvbench/test/custom_traffic_perftest.py:93

↓ 3 callersMethoddestroy

examples/device/ep/csrc/nixl_ep.cpp:280

↓ 3 callersFunctionenable_telemetry_with_temp_dir

Enables telemetry into a freshly created temporary directory for the lifetime of `env_guard`. The returned `TempDir` removes the directory when droppe

src/bindings/rust/tests/tests.rs:49

↓ 3 callersMethodexportEvent

src/core/telemetry/buffer_exporter.cpp:39

↓ 3 callersMethodfillData

test/gtest/error_handling.cpp:248

↓ 3 callersMethodfillData

test/gtest/plugins/uccl/uccl_test.cpp:241

↓ 3 callersFunctionfill_test_pattern

Helper function to fill buffer with repeating pattern

test/unit/plugins/hf3fs/nixl_hf3fs_test.cpp:76

↓ 3 callersFunctionfindAgentMetricSample

test/gtest/telemetry_prometheus_test.cpp:149

↓ 3 callersMethodflushEp

src/plugins/ucx/ucx_utils.cpp:338

↓ 3 callersFunctionformat_duration

Helper function to format duration

test/unit/plugins/hf3fs/nixl_hf3fs_test.cpp:94

↓ 3 callersFunctionformat_duration

Helper function to format duration

test/unit/plugins/gds_mt/nixl_gds_mt_test.cpp:216

↓ 3 callersFunctionformat_duration

Helper function to format duration

test/unit/plugins/posix/nixl_posix_test.cpp:128

↓ 3 callersFunctionformat_duration

Helper function to format duration

test/unit/plugins/cuda_gds/nixl_gds_test.cpp:194

↓ 3 callersMethodgenNotif

src/core/nixl_agent.cpp:1276

↓ 3 callersMethodgetAvailBackendPluginNames

src/core/nixl_plugin_manager.cpp:589

↓ 3 callersMethodgetAvgNumaNodeBandwidth

* @brief Retrieves the average bandwidth limit per NUMA node. This bandwidth limit of a single * NUMA node is the sum of the link speed of all to

src/utils/libfabric/libfabric_topology.h:282

↓ 3 callersMethodgetBackendOptions

src/core/nixl_plugin_manager.cpp:560

↓ 3 callersMethodgetCoveringIndex

src/infra/nixl_descriptors.cpp:484

↓ 3 callersFunctiongetNextXferId

src/utils/libfabric/libfabric_common.cpp:177

↓ 3 callersFunctiongetNonEmptyString

src/utils/common/configuration.h:141

↓ 3 callersMethodgetNotifs

Populate an empty received notif list. Elements are released within backend then.

src/api/cpp/backend/backend_engine.h:219

↓ 3 callersMethodgetNumChunks

src/plugins/ucx/ucx_backend.cpp:503

↓ 3 callersMethodgetNumNvidiaAccel

src/utils/libfabric/libfabric_topology.h:209

↓ 3 callersFunctiongetPcieAddressFromHwlocPcidev

hwloc helper methods

test/unit/utils/libfabric/libfabric_topology_test.cpp:366

↓ 3 callersMethodgetPublicData

test/gtest/mocks/mock_backend_engine.h:80

↓ 3 callersMethodgetSupportedMems

test/gtest/mocks/mock_backend_engine.h:47

↓ 3 callersMethodget_addr

src/plugins/gpunetio/verbs/verbs.h:166

↓ 3 callersMethodget_backend_mem_types

(self, backend: str)

src/api/python/_api.py:347

↓ 3 callersFunctionget_batch_size

(model: BaseModelArch, model_config: ModelConfig, io_size: int)

benchmark/kvbench/models/utils.py:36

↓ 3 callersMethodget_kv_size_per_token

Get the key-value cache size for the Qwen3 model (per token). The KV cache is based on the Grouped Query Attention mechanism:

benchmark/kvbench/models/qwen3.py:64

← previous next →301–400 of 3,108, ranked by callers