NCCL 2.26.3-1

Minimize the performance impact of the device kernel profiling support when
the profiler plugin is not loaded.

Reduce the overheads of CUDA graph capturing, which increased in NCCL
2.26.2 for large graphs.

Fix the exchange of enhanced connection establishment (ECE) options to
address potential slowdowns on networks utilizing RoCE.

Test if cuMem host allocations work and if not, disable them. Enabled by
default since NCCL 2.24 if the CUDA driver version is at least 12.6, such
allocations rely on NUMA support, which is by default not available under
Docker. We recommend invoking Docker with "--cap-add SYS_NICE" to enable
it.

Fix an initialization error when running with NCCL_NET_GDR_C2C=1 on
multiple MNNVL domains with non-uniform network configurations across
nodes.

Fix the printing of sub-seconds in the debug log when using a custom
NCCL_DEBUG_TIMESTAMP_FORMAT setting.

src/include/device.h

@@ -221,6 +221,7 @@ struct alignas(16) ncclDevWorkP2p {
   uint8_t sendProtoLL:1, recvProtoLL:1;
   uint8_t sendNetReg:1, recvNetReg:1;
   uint8_t sendIpcReg:1, recvIpcReg:1;
+  uint8_t profilerEnabled:1;
 };
 
 // Compute the subset of the data transfer corresponding to the given part index.
@@ -259,6 +260,7 @@ struct alignas(16) ncclDevWorkColl {
   uint32_t channelLo:8, channelHi:8;
   uint32_t nWarps:8;
   uint32_t redOpArgIsPtr:1, regUsed:1, netRegUsed:1, oneNode:1, direct:2, isOneRPN:1;
+  uint32_t profilerEnabled:1;
   uint32_t root;
   void* recvbuff;
   void* sendbuff;