AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

Revert "Port alltoall[v]" (#325)

Telusuri Sumber 
This reverts commit f4d5d3d620.
This commit is contained in:
Wenkai Du
2021-03-06 13:59:31 -08:00
committed by GitHub
orang tua c018edf0f2
melakukan 8e180cf087
25 mengubah file dengan 65 tambahan dan 451 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
tools/topo_expl/utils.cpp
+1 -35
Melihat File
@@ -30,7 +30,7 @@
#include "model.h"
#include "utils.h"
const char* ncclFuncStr[NCCL_NUM_FUNCTIONS+3] = { "Broadcast", "Reduce", "AllGather", "ReduceScatter", "AllReduce", "SendRecv", "AllToAll", "AllToAllv" };
const char* ncclFuncStr[NCCL_NUM_FUNCTIONS+1] = { "Broadcast", "Reduce", "AllGather", "ReduceScatter", "AllReduce", "SendRecv" };
const char* ncclAlgoStr[NCCL_NUM_ALGORITHMS] = { "Tree", "Ring", "CollNet" };
const char* ncclProtoStr[NCCL_NUM_PROTOCOLS] = { "LL", "LL128", "Simple" };
@@ -270,7 +270,6 @@ ncclResult_t initTransportsRank_1(struct ncclComm* comm, struct allGather1Data_t
int fullCudaCompCap;
int nChannels;
int gcn;
int alltoallDisable;
struct ncclGraphInfo tree;
struct ncclGraphInfo ring;
struct ncclGraphInfo collNet;
@@ -283,7 +282,6 @@ ncclResult_t initTransportsRank_1(struct ncclComm* comm, struct allGather1Data_t
NCCLCHECK(ncclTopoIdToIndex(comm->topo, GPU, myInfo->busId, &idx));
allGather3Data[rank].cudaCompCap = comm->topo->nodes[GPU].nodes[idx].gpu.cudaCompCap;
allGather3Data[rank].gcn = comm->topo->nodes[GPU].nodes[idx].gpu.gcn;
allGather3Data[rank].alltoallDisable = comm->topo->nodes[NET].count? 1 : comm->alltoallDisable;
allGather3Data[rank].nChannels = comm->nChannels = treeGraph.nChannels = ringGraph.nChannels =
std::min(treeGraph.nChannels, ringGraph.nChannels);
@@ -450,8 +448,6 @@ ncclResult_t ncclTransportP2pSetup(struct ncclComm* comm, struct ncclTopoGraph*
}
RCCL_PARAM(AllToAllDisable, "ALLTOALL_KERNEL_DISABLE", 0);
ncclResult_t initTransportsRank_3(struct ncclComm* comm, struct allGather3Data_t *allGather3Data,
struct ncclTopoGraph& treeGraph, struct ncclTopoGraph& ringGraph, struct ncclTopoGraph& collNetGraph) {
int rank = comm->rank;
@@ -481,11 +477,9 @@ ncclResult_t initTransportsRank_3(struct ncclComm* comm, struct allGather3Data_t
struct ncclTopoRanks** allTopoRanks;
NCCLCHECK(ncclCalloc(&allTopoRanks, comm->nRanks));
int gcn = allGather3Data[0].gcn;
int alltoallDisable = 0;
for (int i=0; i<nranks; i++) {
allTopoRanks[i] = &allGather3Data[i].topoRanks;
gcn = std::min(allGather3Data[i].gcn, gcn);
alltoallDisable = std::max(allGather3Data[i].alltoallDisable, alltoallDisable);
// Make sure we align all ranks so that the tuning is consistent across ranks
treeGraph.nChannels = ringGraph.nChannels = comm->nChannels = std::min(allGather3Data[i].nChannels, comm->nChannels);
treeGraph.sameChannels = std::min(allGather3Data[i].tree.sameChannels, treeGraph.sameChannels);
@@ -505,11 +499,6 @@ ncclResult_t initTransportsRank_3(struct ncclComm* comm, struct allGather3Data_t
collNetGraph.typeInter = std::min(allGather3Data[i].collNet.typeInter, collNetGraph.typeInter);
}
if (comm->alltoallDisable != alltoallDisable) {
comm->alltoallDisable = alltoallDisable;
}
INFO(NCCL_INIT, "RCCL AllToAll(v)/Scatter/Gather kernels %s", comm->alltoallDisable ? "disabled" : "enabled");
// count NETs used by ring
int nNets = 0;
int nets[MAXCHANNELS*2];
@@ -625,29 +614,6 @@ ncclResult_t initTransportsRank_3(struct ncclComm* comm, struct allGather3Data_t
// Compute nChannels per peer for p2p
NCCLCHECK(ncclTopoComputeP2pChannels(comm));
if (!alltoallDisable) {
int nc = comm->nChannels;
for (int c=0; c<nc; c++) {
const int peersPerChan = DIVUP(nranks, nc);
for (int p=0; p<peersPerChan; p++) {
// first channel is reserved for self copy
if ((c*peersPerChan+p)%nranks == 0)
continue;
int peerSend = (rank+c*peersPerChan+p)%nranks;
int peerRecv = (2*nranks+rank-(c*peersPerChan)%nranks-p)%nranks;
if (comm->channels[c].peers[peerSend].send.connected == 0) {
comm->connectSend[peerSend] |= (1<<c);
comm->connect = 1;
}
if (comm->channels[c].peers[peerRecv].recv.connected == 0) {
comm->connectRecv[peerRecv] |= (1<<c);
comm->connect = 1;
}
}
}
NCCLCHECK(ncclTransportP2pSetup(comm, NULL));
}
//NCCLCHECK(ncclCommSetIntra(comm, intraRank, intraRanks, intraRank0Comm));
//if (comm->nNodes) NCCLCHECK(ncclProxyCreate(comm));