Network user buffer support for collectives
 * Leverage user buffer registration to achieve zero-copy
   inter-node communications for Ring, NVLS and Collnet

Add RAS subsystem
 * Create a RAS thread keeping track of all NCCL communicators.
 * Add a ncclras tool contacting the RAS thread and getting a
   report.

Add fp8 support
 * Add support for e5m2 and e4m3 8-bit floating point operations.
 * Use Tree/PAT algorithms when possible for better numerical
   stability.

Add NIC fusion
 * Add a NET API to ask the network plugin to fuse a set of
   interfaces together.
 * Fuse multiple NICs under the same PCI switch as a single,
   larger NIC.

Socket connection failure retry
 * Retry in case of socket connection failure (unreachable host)
 * Avoid "Software caused connection abort" errors on retries

QP connection failure retry
 * Retry in case of IB QP connection failure during ibv_modify_qp.

NET API improvements
 * Allow plugins to force a flush in case data and completion
   ordering is not guaranteed.
 * Indicate when completion is not needed (e.g. for the LL128
   protocol), allowing plugins to skip generating a completion.
 * Allow for full offload of allgather operations when using one
   GPU per node.

NCCL_ALGO/NCCL_PROTO strict enforcement
 * Extend NCCL_ALGO/NCCL_PROTO syntax to be able to specify
   ALGO/PROTO filters for each collective operation.
 * Strictly enforce the ALGO/PROTO filters, no longer fall back
   on the ring algorithm when the filtering leaves no option and
   error out instead.

Enable CUMEM host allocations
 * Use cumem functions for host memory allocation by default.

Improved profiler plugin API
 * Avoid dependencies with NCCL includes.
 * Add information on whether the buffer is registered or not

Adjust PAT tuning
 * Improve transition between PAT and ring at scale.

Fix hangs when running with different CPU architectures
 * Detect when we use a mix of GPU architectures
 * Ensure Algo/Proto decisions are made based on that unified
   state.

Fix FD leak in UDS
 * Fix a leak when mapping buffers intra-node with cumem IPCs.

Fix crash when mixing buffer registration and graph buffer registration.
 * Separate local and graph registration to avoid crashes when we free
   buffers.

Fix user buffer registration with dmabuf
 * Make ncclSend/ncclRecv communication with buffer registration functional
   on network plugins relying on dmabuf for buffer registration.

Fix crash in IB code caused by uninitialized fields.

Fix non-blocking ncclSend/ncclRecv
 * Fix case where ncclSend/ncclRecv would return ncclSuccess in non-blocking
   mode even though the operation was not enqueued onto the stream.
 * Issue #1495

Various compiler tweaks and fixes
 * PR #758

Fix typo in ncclTopoPrintGraph
 * Issue #1468


[ROCm/rccl commit: 6aae379278]
This commit is contained in:
Sylvain Jeaugey
2024-12-18 08:26:06 -08:00
parent 758f28b359
commit db3bfd118f
97 changed files with 12588 additions and 3127 deletions
+18 -4
View File
@@ -197,12 +197,15 @@ struct ncclTaskColl {
int32_t algorithm:8, protocol:8;
uint32_t isCollnet:1, isNvls:1;
uint32_t devFuncId:30;
enum ncclRegBufferType regBufType;
int regBufType;
// number of elements in planner->ipcMemQueue associated with this collective
int nCleanupQueueElts;
void* sendMhandle;
void* recvMhandle;
void** sendNetHandles;
void** recvNetHandles;
void** srecvNetHandles;
// index for IPC record lookup
uintptr_t sendbuffOffset;
uintptr_t recvbuffOffset;
@@ -236,6 +239,7 @@ struct ncclKernelPlan {
struct ncclKernelPlan* next;
bool persistent; // aka captured in a graph
bool isHostCbEnq;
enum ncclDevWorkStorageType workStorageType;
bool kernelSpecialized;
void *kernelFn;
@@ -365,6 +369,7 @@ struct ncclKernelPlanner {
struct ncclIntruQueue<struct ncclTaskColl, &ncclTaskColl::next> collTaskQueue;
struct ncclIntruQueue<struct ncclWorkList, &ncclWorkList::next> collWorkQueue;
struct ncclIntruQueue<struct ncclWorkList, &ncclWorkList::next> tmpCollWorkQueue;
struct ncclIntruQueue<struct ncclCommCallback, &ncclCommCallback::next> collCleanupQueue;
//////////////////////////////////////////////////////////////////////////////
@@ -463,6 +468,8 @@ struct ncclComm {
// Counter for tracking CUDA launches (P2P and collectives included)
uint64_t opCount;
// Collective operation counter
uint64_t collOpCount;
// Channels for collectives
int nChannels; // connection nChannels
@@ -486,7 +493,6 @@ struct ncclComm {
ssize_t threadThresholds[NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS];
float latencies[NCCL_NUM_FUNCTIONS][NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS];
float bandwidths[NCCL_NUM_FUNCTIONS][NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS];
float ringbdw[NCCL_NUM_FUNCTIONS][NCCL_NUM_PROTOCOLS];
int maxThreads[NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS];
/* This attribute can indicate the states of communicators and return code of
@@ -532,7 +538,7 @@ struct ncclComm {
int proxyRefCountOld; /* store proxy post-atomic-sub refcount */
// Whether this communicator uses collNet
int collNetSupport;
bool collNetRegSupport;
bool isOneRPN;
uint8_t collNetSupportMatrix[4/*sum,prod,max,min*/][ncclNumTypes];
bool intraNodeP2pSupport;
int* collNetHeads;
@@ -560,6 +566,7 @@ struct ncclComm {
// Subset of those in groupNext list. Holds 0x1 if not needing preconnect.
struct ncclComm* preconnectNext;
int persistentRefs; // number of persistent plan-lists capturing this comm
int noncapturedRefs; // number of non-captured hostStreamPlanCallback on the stream
struct P2pSchedulePair { int sendRank; int recvRank; } *p2pSchedule;
struct ncclKernelPlanner planner;
@@ -599,9 +606,16 @@ struct ncclComm {
// buffer registration cache
struct ncclRegCache regCache;
int isAllNvlink;
bool useNetPXN;
bool useGdr;
int splitCount;
uint64_t endMagic;
};
static_assert(offsetof(struct ncclComm, startMagic) == 0, "startMagic must be the first field of ncclComm");
static_assert(offsetof(struct ncclComm, endMagic) == sizeof(struct ncclComm) - sizeof(uint64_t), "endMagic must be the last field of ncclComm");
enum ncclLaunchMode {
ncclLaunchModeInvalid=0,
ncclLaunchModeParallel,
@@ -644,7 +658,7 @@ inline ncclResult_t ncclCommPollEventCallbacks(struct ncclComm *comm) {
}
}
finish:
cudaThreadExchangeStreamCaptureMode(&mode);
CUDACHECK(cudaThreadExchangeStreamCaptureMode(&mode));
return ncclSuccess;
}