2.24.3-1

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]

projects/rccl/src/include/device.h

@@ -88,24 +88,18 @@ static_assert(NCCL_LL_CLEAN_MASK % NCCL_STEPS == 0, "Invalid NCCL_LL_CLEAN_MASK
 #define NCCL_LL128_SHMEM_ELEMS_PER_THREAD 8
 #define NCCL_LL128_SHMEM_SIZE (NCCL_LL128_SHMEM_ELEMS_PER_THREAD*NCCL_LL128_MAX_NTHREADS)
 
-#define NCCL_DIRECT_WRITE 0x01
-#define NCCL_DIRECT_READ  0x02
+#define NCCL_P2P_WRITE 0x01
+#define NCCL_P2P_READ  0x02
 #define NCCL_DIRECT_NIC   0x04
-#define NCCL_IPC_WRITE    0x08
-#define NCCL_IPC_READ     0x10
-#define NCCL_NVLS_MIN_POLL 0x20
+#define NCCL_NVLS_MIN_POLL 0x80
 
 // Number of named barriers supported by CUDA
 #define NCCL_MAX_GROUPS 16
 
-#define NCCL_MAX_COLLNET_SIZE (1L << 29)
-
-enum ncclRegBufferType {
-  NCCL_REGULAR_BUFFER = 0,
-  NCCL_IPC_REG_BUFFER = 1,
-  NCCL_NVLS_REG_BUFFER = 2,
-  NCCL_COLLNET_REG_BUFFER = 3
-};
+#define NCCL_REGULAR_BUFFER 0x00
+#define NCCL_IPC_REG_BUFFER 0x01
+#define NCCL_NVLS_REG_BUFFER 0x02
+#define NCCL_NET_REG_BUFFER 0x04
 
 struct ncclConnInfo {
   // Regular comm mechanism
@@ -143,8 +137,6 @@ struct ncclConnector {
   struct ncclTransportComm* transportComm;
   void* transportResources;
   struct ncclConnInfo conn;
-  int sendMemSameProcess;
-  int recvMemSameProcess;
 };
 
 struct ncclRing {
@@ -228,7 +220,7 @@ struct alignas(16) ncclDevWorkP2p {
   uint8_t sendChunkSize_u32fp8, recvChunkSize_u32fp8;
 
   uint8_t sendProtoLL:1, recvProtoLL:1;
-  uint8_t sendRegistered:1, recvRegistered:1;
+  uint8_t sendNetReg:1, recvNetReg:1;
   uint8_t sendIpcReg:1, recvIpcReg:1;
 };
 
@@ -267,7 +259,7 @@ struct alignas(16) ncclDevWorkColl {
   //   nChannels == (channelHi - channelLo) + 1
   uint32_t channelLo:8, channelHi:8;
   uint32_t nWarps:8;
-  uint32_t redOpArgIsPtr:1, regUsed:2, oneNode:1, direct:4;
+  uint32_t redOpArgIsPtr:1, regUsed:1, netRegUsed:1, oneNode:1, direct:2, isOneRPN:1;
   uint32_t root;
   void* recvbuff;
   void* sendbuff;
@@ -393,7 +385,7 @@ struct ncclDevComm {
   int nNodes;
   int buffSizes[NCCL_NUM_PROTOCOLS];
   int p2pChunkSize;
-  int isNvlink;
+  int isAllNvlink;
 
   // Work fifo return credits
   uint32_t* workConsumed/*[MAXCHANNELS]*/;
@@ -525,9 +517,7 @@ inline bool ncclNvlsSupported(int devRedOp, int type) {
   case ncclInt64:
   case ncclUint64:
   case ncclFloat16:
-  #if defined(__CUDA_BF16_TYPES_EXIST__)
   case ncclBfloat16:
-  #endif
     return devRedOp == ncclDevSum || devRedOp == ncclDevMinMax;
   case ncclFloat:
   case ncclDouble: