AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'nccl/master' into develop

Browse Source
This commit is contained in:
Wenkai Du
2022-04-18 11:14:51 -07:00
parent fd2f1b3b88 9bfc1c6e35
commit d28e1cb44f
81 changed files with 7358 additions and 4468 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/topo.cc
+44 -13
View File
@@ -1,6 +1,6 @@
/*************************************************************************
* Copyright (c) 2016-2021, NVIDIA CORPORATION. All rights reserved.
* Modifications Copyright (c) 2019-2021 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2016-2022, NVIDIA CORPORATION. All rights reserved.
* Modifications Copyright (c) 2019-2022 Advanced Micro Devices, Inc. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -22,11 +22,11 @@
const char* topoNodeTypeStr[] = { "GPU", "PCI", "NVS", "CPU", "NIC", "NET" };
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
const char* topoLinkTypeStr[] = { "LOC", "XGMI", "", "PCI", "", "", "SYS", "NET" };
const char* topoPathTypeStr[] = { "LOC", "XGMI", "NVB", "PIX", "PXB", "PHB", "SYS" };
const char* topoLinkTypeStr[] = { "LOC", "XGMI", "", "PCI", "", "", "", "SYS", "NET" };
const char* topoPathTypeStr[] = { "LOC", "XGMI", "NVB", "PIX", "PXB", "PXN", "PHB", "SYS", "DIS" };
#else
const char* topoLinkTypeStr[] = { "LOC", "NVL", "", "PCI", "", "", "SYS", "NET" };
const char* topoPathTypeStr[] = { "LOC", "NVL", "NVB", "PIX", "PXB", "PHB", "SYS" };
const char* topoLinkTypeStr[] = { "LOC", "NVL", "", "PCI", "", "", "", "SYS", "NET" };
const char* topoPathTypeStr[] = { "LOC", "NVL", "NVB", "PIX", "PXB", "PXN", "PHB", "SYS", "DIS" };
#endif
/******************************************************************/
@@ -127,6 +127,7 @@ ncclResult_t ncclTopoCreateNode(struct ncclTopoSystem* system, struct ncclTopoNo
n->net.asic = 0ULL;
n->net.port = NCCL_TOPO_UNDEF;
n->net.width = 0.0;
n->net.latency = 0.0;
}
*node = n;
return ncclSuccess;
@@ -338,13 +339,14 @@ ncclResult_t ncclTopoAddNet(struct ncclXmlNode* xmlNet, struct ncclTopoSystem* s
ncclDebugNoWarn = NCCL_GRAPH;
int mbps;
if (xmlGetAttrInt(xmlNet, "speed", &mbps) != ncclSuccess) mbps = 0;
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "speed", &mbps, 0));
if (mbps <= 0) mbps = 10000; // Some NICs define speed = -1
net->net.width = mbps / 8000.0;
if (xmlGetAttrInt(xmlNet, "port", &net->net.port) != ncclSuccess) net->net.port = 0;
if (xmlGetAttrInt(xmlNet, "gdr", &net->net.gdrSupport) != ncclSuccess) net->net.gdrSupport = 0;
if (xmlGetAttrInt(xmlNet, "maxconn", &net->net.maxChannels) != ncclSuccess) net->net.maxChannels = MAXCHANNELS;
if (ncclCollNet && xmlGetAttrInt(xmlNet, "coll", &net->net.collSupport) != ncclSuccess) net->net.collSupport = 0;
if (xmlGetAttrFloat(xmlNet, "latency", &net->net.latency) != ncclSuccess) net->net.latency = 0;
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "port", &net->net.port, 0));
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "gdr", &net->net.gdrSupport, 0));
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "maxconn", &net->net.maxChannels, MAXCHANNELS));
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "coll", &net->net.collSupport, 0));
net->net.busId = busId;
ncclDebugNoWarn = 0;
@@ -653,6 +655,16 @@ static ncclResult_t xmlInitAttrUint64(struct ncclXmlNode* node, const char* attr
}
return ncclSuccess;
}
static ncclResult_t xmlInitAttrFloat(struct ncclXmlNode* node, const char* attrName, const float value) {
int index;
NCCLCHECK(xmlGetAttrIndex(node, attrName, &index));
if (index == -1) {
index = node->nAttrs++;
strncpy(node->attrs[index].key, attrName, MAX_STR_LEN);
snprintf(node->attrs[index].value, MAX_STR_LEN, "%f", value);
}
return ncclSuccess;
}
ncclResult_t ncclTopoGetSystem(struct ncclComm* comm, struct ncclTopoSystem** system) {
@@ -689,7 +701,7 @@ ncclResult_t ncclTopoGetSystem(struct ncclComm* comm, struct ncclTopoSystem** sy
// Auto-detect NICs if needed. net/collnet share the same xml/graph nodes,
// so we start with collnet so that it has precedence.
int netDevCount = 0;
if (ncclCollNet) {
if (collNetSupport()) {
NCCLCHECK(collNetDevices(&netDevCount));
for (int n=0; n<netDevCount; n++) {
ncclNetProperties_t props;
@@ -718,6 +730,7 @@ ncclResult_t ncclTopoGetSystem(struct ncclComm* comm, struct ncclTopoSystem** sy
NCCLCHECK(xmlSetAttrInt(netNode, "dev", n));
NCCLCHECK(xmlInitAttrInt(netNode, "speed", props.speed));
NCCLCHECK(xmlInitAttrInt(netNode, "port", props.port));
NCCLCHECK(xmlInitAttrFloat(netNode, "latency", props.latency));
NCCLCHECK(xmlInitAttrUint64(netNode, "guid", props.guid));
NCCLCHECK(xmlInitAttrInt(netNode, "maxconn", props.maxComms));
NCCLCHECK(xmlInitAttrInt(netNode, "gdr", props.ptrSupport & NCCL_PTR_CUDA ? 1 : 0));
@@ -737,7 +750,7 @@ ncclResult_t ncclTopoGetSystem(struct ncclComm* comm, struct ncclTopoSystem** sy
return ncclSuccess;
}
ncclResult_t ncclTopoGetLocalNet(struct ncclTopoSystem* system, int rank, int64_t* id, int rr) {
ncclResult_t ncclTopoGetLocalNet(struct ncclTopoSystem* system, int rank, int* id) {
int g;
NCCLCHECK(ncclTopoRankToIndex(system, rank, &g));
int minType = PATH_SYS;
@@ -754,6 +767,13 @@ ncclResult_t ncclTopoGetLocalNet(struct ncclTopoSystem* system, int rank, int64_
}
if (path->width == maxWidth && path->type == minType) nets[count++] = system->nodes[NET].nodes[n].id;
}
if (count == 0) {
*id = -1;
free(nets);
return ncclSuccess;
}
int rr = system->nodes[GPU].nodes[g].gpu.dev;
*id = nets[rr%count];
free(nets);
return ncclSuccess;
@@ -853,3 +873,14 @@ ncclResult_t ncclTopoGetCompCap(struct ncclTopoSystem* system, int* ccMin, int*
if (ccMax) *ccMax = max;
return ncclSuccess;
}
ncclResult_t ncclTopoGetLocalRank(struct ncclTopoSystem* system, int rank, int* localRank) {
for (int g=0; g<system->nodes[GPU].count; g++) {
if (system->nodes[GPU].nodes[g].gpu.rank == rank) {
*localRank = g;
return ncclSuccess;
}
}
WARN("Could not find local GPU with rank %d\n", rank);
return ncclInternalError;
}