NCCL 2.28.3-1
Device API (Experimental) * Introduces device-side APIs to integrate NCCL communication directly into application kernels. * Supports LSA (Load/Store Access) for CUDA P2P communication over NVLink and some PCIe platforms. * Supports Multimem for hardware multicast using NVLink SHARP. * Adds initial framework for GIN (GPU-Initiated Networking), currently under development. * Introduces device communicators created using ncclDevCommCreate. * Enables device-side communication operations with synchronization (ncclLsaBarrierSession) and memory accessors (ncclGetLsaPointer, ncclGetLsaMultimemPointer). * Experimental APIs - signatures and functionality may evolve in future releases. * No ABI compatibility is guaranteed — applications must be recompiled with each new NCCL release. Symmetric memory improvements * Support for aggregating symmetric operations using ncclGroupStart/End APIs. * Reimplement symmetric kernels using device API. New Host APIs * Introduce new host collective APIs: ncclAlltoAll, ncclScatter, ncclGather. CE (Copy Engine) Collectives * Reduce SM utilization for alltoall, scatter, gather, and allgather within a single (MN)NVL domain. * Free up SM capacity for the application to do computation at the same time. * To enable the feature for ncclAllGather, ncclAlltoAll, ncclGather, ncclScatter, register buffers into symmetric windows and use the NCCL_CTA_POLICY_ZERO flag in the communicator config_t. NCCL Inspector Plugin * Introduces an Inspector plugin for always-on performance monitoring. * Produces structured JSON output with metadata, execution time, bandwidth, and optional event traces for each NCCL operation. * Enables integration with analysis tools such as Performance Exporter to visualize NCCL performance bottlenecks. * Lightweight to enable via environment variables NCCL_PROFILER_PLUGIN and NCCL_INSPECTOR_ENABLE. CMake support (Experiemental) * Adds a CMake build system as an alternative to existing Makefiles. * Known issues: pkg.build and Device API currently do not work with CMake. * The known issues will be addressed in a future release. Decreased max CTA count from 32 to 16 on Blackwell * SM overhead is decreased by 50% with this improvement. * This may cause some perf drop on Blackwell because of the reduced SM usage. * If the extra SM capacity is not desired, two options are available to restore to previous behavior: 1) Setting NCCL_MIN_CTAS=32 NCCL_MAX_CTAS=32 environment variables; 2) setting communicator config to over-write max CTA count to 32. * Based on community feedback, future versions may consider different trade-offs between performance and SM overhead. Plugins * Network * App-aware Network plugin. NCCL passes information about communication operations to be executed on the network end point. This allows for better tuning of network end points and their use in the plugins. * Improve handling of physical and virtual network devices and load/unload. * Network plugin version 11 - add explicit context and communication ID support for per communicator init/finalize. * Add Multi-Request Net API. Using this will help NCCL to anticipate multiple send/recv requests and optimize for it. See maxMultiRequestSize field in ncclNetProperties_v11_t. * Profiler * Add support for API events (group, collective, and p2p) and for tracking kernel launches in the profiler plugin. * Add Inspector Profiler Plugin (see section above). * Add a hook to Google’s CoMMA profiler on github. * Tuner * Expose NCCL tuning constants at tuner initialization via ncclTunerConstants_v5_t. * Add NVL Domain Information API. * Support multiple plugin types from a single shared object. New Parameterization and ncclConfig changes: * Add new option NCCL_MNNVL_CLIQUE_ID=-2 which will use rack serial number to partition the MNNVL clique. This will limit NVLink domains to GPUs within a single rack. * Add NCCL_NETDEVS_POLICY to control how NET devices are assigned to GPUs. The default (AUTO) is the policy used in previous versions. * Add NCCL_SINGLE_PROC_MEM_REG_ENABLE control variable to enable NVLS UB registration in the “one process, multiple ranks” case as opt in. * Move nChannelsPerNetPeer into ncclConfig. NCCL_NCHANNELS_PER_NET_PEER can override the value in ncclConfig. * Enable PxN over C2C by default * PxN over C2C will improve performance for Grace-Blackwell platforms by allowing NCCL to leverage the NIC attached to a peer GPU over NVLINK, C2C, and PCIe. * This behavior can be overridden by setting NCCL_PXN_C2C=0. Other Improvements: * Allow FP8 support for non-reductive operations on pre sm90 devices. (See https://github.com/pytorch/pytorch/pull/151594#discussion_r2135777776) * Fix NVLS+CollNet and temporarily disables COLLNET_CHAIN for >8 GPUs. * Only consider running interfaces for socket traffic. NCCL will not attempt to use interfaces that do not have the IFF_RUNNING bit. (https://github.com/NVIDIA/nccl/issues/1798) * Modernize mutex management. Convert to std::mutex and std::lock_guard. * Remove sm35 and sm50 GENCODE targets which have long been deprecated and were causing issues with the latest NCCL release builds. * Improved NVLS/NVLSTree tuning prediction to improve algorithm and protocol selection. * NVLSTree Tuning Fixes. Update tuning data for H100, GB200-NV72. * Respond better to RoCE link flaps. Instead of reporting an “unknown event” it will now report “GID table changed”. * Move libvirt bridge interface to the end of possible interfaces so that they are considered last. These interfaces are usually virtual bridges to relay traffic to containers running on the host and cannot be used for traffic to a remote node and are therefore unsuitable.
This commit is contained in:
Mark Santesson
+142
-13
@@ -10,10 +10,14 @@
|
||||
#include <sys/types.h>
|
||||
#include <stdint.h>
|
||||
#include <unistd.h>
|
||||
#include <cstring>
|
||||
#include "err.h"
|
||||
#include "profiler.h"
|
||||
#include "queue.h"
|
||||
#include <cuda_runtime.h>
|
||||
|
||||
#define MAX_CHANNELS 32
|
||||
#define MAX_STEPS 16
|
||||
#define MAX_STEPS 1024
|
||||
#define MAX_OPS 16 // Up to 64K ranks for PAT
|
||||
#define MAX_EVENTS_PER_REQ (8)
|
||||
|
||||
@@ -21,7 +25,7 @@ struct proxyOp;
|
||||
struct proxyStep;
|
||||
|
||||
struct netPlugin {
|
||||
uint8_t type;
|
||||
uint64_t type;
|
||||
int pluginType;
|
||||
int pluginVer;
|
||||
uint8_t pluginEvent;
|
||||
@@ -63,7 +67,7 @@ struct kernelCh {
|
||||
#define PROXY_STEP_MAX_STATES 3
|
||||
|
||||
struct proxyStep {
|
||||
uint8_t type; // type of event: network transfer
|
||||
uint64_t type; // type of event: network transfer
|
||||
int state;
|
||||
int step; // network transfer id in given channel
|
||||
int isSend; // send/recv channel operation
|
||||
@@ -76,7 +80,7 @@ struct proxyStep {
|
||||
};
|
||||
|
||||
struct proxyOp {
|
||||
uint8_t type; // type of event: proxy operation
|
||||
uint64_t type; // type of event: proxy operation
|
||||
uint8_t channelId; // channel id for this proxy operation
|
||||
pid_t pid;
|
||||
int rank;
|
||||
@@ -97,7 +101,7 @@ struct group;
|
||||
struct context;
|
||||
|
||||
struct proxyCtrl {
|
||||
uint8_t type;
|
||||
uint64_t type;
|
||||
struct context* ctx; // profiler context
|
||||
double startTs;
|
||||
double stopTs;
|
||||
@@ -107,12 +111,12 @@ struct proxyCtrl {
|
||||
|
||||
// task level event base structure
|
||||
struct taskEventBase {
|
||||
uint8_t type; // event type: collective/p2p
|
||||
uint64_t type; // event type: collective/p2p
|
||||
int rank; // rank of the operation in NCCL communicator
|
||||
const char* func; // ncclFunc*
|
||||
int refCount; // number of references for this operation
|
||||
struct group* parent; // parent event group
|
||||
struct taskEventBase* next; // next top level event in group
|
||||
void* parent; // parent API event
|
||||
struct taskEventBase* next; // next top level event
|
||||
double startTs;
|
||||
double stopTs;
|
||||
};
|
||||
@@ -147,7 +151,7 @@ struct p2p {
|
||||
};
|
||||
|
||||
struct group {
|
||||
uint8_t type;
|
||||
uint64_t type;
|
||||
struct context* ctx; // profiler context
|
||||
int groupId;
|
||||
int refCount;
|
||||
@@ -158,6 +162,70 @@ struct group {
|
||||
struct group* next; // next group event in queue
|
||||
};
|
||||
|
||||
struct collApi {
|
||||
uint64_t type;
|
||||
struct groupApi* parent;
|
||||
struct context* ctx; // profiler context
|
||||
int collApiId;
|
||||
int refCount;
|
||||
cudaStream_t stream;
|
||||
const char* func;
|
||||
size_t count;
|
||||
const char* datatype;
|
||||
int root;
|
||||
bool graphCaptured;
|
||||
struct taskEventBase* eventHead; // queue head for task events
|
||||
struct taskEventBase* eventTail; // queue tail for task events
|
||||
double startTs;
|
||||
double stopTs;
|
||||
struct collApi* next;
|
||||
};
|
||||
|
||||
struct p2pApi {
|
||||
uint64_t type;
|
||||
struct groupApi* parent;
|
||||
struct context* ctx; // profiler context
|
||||
int p2pApiId;
|
||||
int refCount;
|
||||
const char* func;
|
||||
cudaStream_t stream;
|
||||
size_t count;
|
||||
const char* datatype;
|
||||
bool graphCaptured;
|
||||
struct taskEventBase* eventHead; // queue head for task events
|
||||
struct taskEventBase* eventTail; // queue tail for task events
|
||||
double startTs;
|
||||
double stopTs;
|
||||
struct p2pApi* next;
|
||||
};
|
||||
|
||||
struct kernelLaunch {
|
||||
uint64_t type;
|
||||
struct groupApi* parent;
|
||||
cudaStream_t stream;
|
||||
int kernelLaunchId;
|
||||
double startTs;
|
||||
double stopTs;
|
||||
struct kernelLaunch* next;
|
||||
};
|
||||
|
||||
struct groupApi {
|
||||
uint64_t type;
|
||||
struct context* ctx;
|
||||
int groupApiId;
|
||||
int refCount;
|
||||
bool graphCaptured;
|
||||
int groupDepth;
|
||||
struct profilerQueue<struct p2pApi, &p2pApi::next> p2pApiEvents;
|
||||
struct profilerQueue<struct collApi, &collApi::next> collApiEvents;
|
||||
struct profilerQueue<struct kernelLaunch, &kernelLaunch::next> kernelLaunchEvents;
|
||||
double endOfncclGroupStartTs;
|
||||
double startOfncclGroupEndTs;
|
||||
double startTs;
|
||||
double stopTs;
|
||||
struct groupApi* next;
|
||||
};
|
||||
|
||||
// arrays for different event objects
|
||||
struct context {
|
||||
const char* commName;
|
||||
@@ -165,6 +233,26 @@ struct context {
|
||||
int nranks;
|
||||
int rank;
|
||||
|
||||
int groupApiPoolSize;
|
||||
int groupApiPoolBase;
|
||||
int groupApiPoolIndex;
|
||||
struct groupApi* groupApiPool;
|
||||
|
||||
int collApiPoolSize;
|
||||
int collApiPoolBase;
|
||||
int collApiPoolIndex;
|
||||
struct collApi* collApiPool;
|
||||
|
||||
int p2pApiPoolSize;
|
||||
int p2pApiPoolBase;
|
||||
int p2pApiPoolIndex;
|
||||
struct p2pApi* p2pApiPool;
|
||||
|
||||
int kernelLaunchPoolSize;
|
||||
int kernelLaunchPoolBase;
|
||||
int kernelLaunchPoolIndex;
|
||||
struct kernelLaunch* kernelLaunchPool;
|
||||
|
||||
int groupPoolSize;
|
||||
int groupPoolBase;
|
||||
int groupPoolIndex;
|
||||
@@ -186,9 +274,50 @@ struct context {
|
||||
struct proxyCtrl* proxyCtrlPool;
|
||||
};
|
||||
|
||||
int taskEventQueueEmpty(struct group* g);
|
||||
void taskEventQueueEnqueue(struct group* g, struct taskEventBase* event);
|
||||
struct taskEventBase* taskEventQueueHead(struct group* g);
|
||||
struct taskEventBase* taskEventQueueDequeue(struct group* g);
|
||||
template <typename T>
|
||||
inline int taskEventQueueEmpty(T *obj) {
|
||||
return obj->eventHead == NULL;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
inline void taskEventQueueEnqueue(T* obj, struct taskEventBase* event) {
|
||||
event->next = NULL;
|
||||
if (obj->eventHead) obj->eventTail->next = event;
|
||||
else obj->eventHead = event;
|
||||
obj->eventTail = event;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
inline struct taskEventBase* taskEventQueueHead(T *obj) {
|
||||
return obj->eventHead;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
inline struct taskEventBase* taskEventQueueDequeue(T* obj) {
|
||||
struct taskEventBase* tmp = obj->eventHead;
|
||||
obj->eventHead = obj->eventHead->next;
|
||||
if (obj->eventHead == NULL) obj->eventTail = NULL;
|
||||
return tmp;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
inline void resetTaskEvents(T *obj, struct context* ctx) {
|
||||
while (!taskEventQueueEmpty(obj)) {
|
||||
struct taskEventBase* base = taskEventQueueDequeue(obj);
|
||||
if (base->type == ncclProfileColl) {
|
||||
struct collective* c = (struct collective *)base;
|
||||
// reset event proxyOps & proxySteps
|
||||
memset(c->nProxyOps, 0, sizeof(int)*MAX_CHANNELS);
|
||||
// release collective events in the group and return them to the collective pool
|
||||
__atomic_fetch_add(&ctx->collPoolBase, 1, __ATOMIC_RELAXED);
|
||||
} else if (base->type == ncclProfileP2p) {
|
||||
struct p2p* p = (struct p2p *)base;
|
||||
// reset event proxyOp and proxySteps
|
||||
memset(&p->op, 0, sizeof(struct proxyOp)*MAX_CHANNELS);
|
||||
// release p2p events in the group and return them to the p2p pool
|
||||
__atomic_fetch_add(&ctx->p2pPoolBase, 1, __ATOMIC_RELAXED);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
Reference in New Issue
Block a user