yezhengmao/nccl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge tag 'v2.15.1-1'

Browse Source
This commit is contained in:
Sylvain Jeaugey 2022-10-07 11:00:26 -07:00
commit d128d62238
22 changed files with 303 additions and 76 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
makefiles/common.mk
View File

@ -31,13 +31,17 @@ CUDA8_GENCODE = -gencode=arch=compute_35,code=sm_35 \
-gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_61,code=sm_61
CUDA9_GENCODE = -gencode=arch=compute_70,code=sm_70 CUDA9_GENCODE = -gencode=arch=compute_70,code=sm_70
CUDA11_GENCODE = -gencode=arch=compute_80,code=sm_80 CUDA11_GENCODE = -gencode=arch=compute_80,code=sm_80
CUDA11_8_GENCODE = -gencode=arch=compute_90,code=sm_90
CUDA8_PTX = -gencode=arch=compute_61,code=compute_61 CUDA8_PTX = -gencode=arch=compute_61,code=compute_61
CUDA9_PTX = -gencode=arch=compute_70,code=compute_70 CUDA9_PTX = -gencode=arch=compute_70,code=compute_70
CUDA11_PTX = -gencode=arch=compute_80,code=compute_80 CUDA11_PTX = -gencode=arch=compute_80,code=compute_80
CUDA11_8_PTX = -gencode=arch=compute_90,code=compute_90
# Include Ampere support if we're using CUDA11 or above ifeq ($(shell test "0$(CUDA_MAJOR)" -eq 11 -a "0$(CUDA_MINOR)" -ge 8 -o "0$(CUDA_MAJOR)" -gt 11; echo $$?),0)
ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 11; echo $$?),0) # Include Hopper support if we're using CUDA11.8 or above
NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA9_GENCODE) $(CUDA11_GENCODE) $(CUDA11_8_GENCODE) $(CUDA11_8_PTX)
else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 11; echo $$?),0)
NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA9_GENCODE) $(CUDA11_GENCODE) $(CUDA11_PTX) NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA9_GENCODE) $(CUDA11_GENCODE) $(CUDA11_PTX)
# Include Volta support if we're using CUDA9 or above # Include Volta support if we're using CUDA9 or above
else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 9; echo $$?),0) else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 9; echo $$?),0)
@ -45,7 +49,7 @@ else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 9; echo $$?),0)
else else
NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA8_PTX) NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA8_PTX)
endif endif
#$(info NVCC_GENCODE is ${NVCC_GENCODE}) $(info NVCC_GENCODE is ${NVCC_GENCODE})
CXXFLAGS := -DCUDA_MAJOR=$(CUDA_MAJOR) -DCUDA_MINOR=$(CUDA_MINOR) -fPIC -fvisibility=hidden \ CXXFLAGS := -DCUDA_MAJOR=$(CUDA_MAJOR) -DCUDA_MINOR=$(CUDA_MINOR) -fPIC -fvisibility=hidden \
-Wall -Wno-unused-function -Wno-sign-compare -std=c++11 -Wvla \ -Wall -Wno-unused-function -Wno-sign-compare -std=c++11 -Wvla \

4
makefiles/version.mk
View File

@ -1,6 +1,6 @@
##### version ##### version
NCCL_MAJOR := 2 NCCL_MAJOR := 2
NCCL_MINOR := 14 NCCL_MINOR := 15
NCCL_PATCH := 3 NCCL_PATCH := 1
NCCL_SUFFIX := NCCL_SUFFIX :=
PKG_REVISION := 1 PKG_REVISION := 1

2
src/channel.cc
View File

@ -27,7 +27,7 @@ ncclResult_t initChannel(struct ncclComm* comm, int channelId) {
NCCLCHECK(ncclCudaCallocAsync(&channel->devRingUserRanks, nRanks, comm->deviceStream.stream)); NCCLCHECK(ncclCudaCallocAsync(&channel->devRingUserRanks, nRanks, comm->deviceStream.stream));
ncclCommPushCudaFree(comm, channel->devRingUserRanks); ncclCommPushCudaFree(comm, channel->devRingUserRanks);
NCCLCHECK(ncclStrongStreamRelease(ncclCudaGraphNull(), &comm->deviceStream)); NCCLCHECK(ncclStrongStreamRelease(ncclCudaGraphNone(), &comm->deviceStream));
for (int r=0; r < nRanks+1; ++r) { for (int r=0; r < nRanks+1; ++r) {
for (int b=0; b < NCCL_MAX_CONNS; b++) { for (int b=0; b < NCCL_MAX_CONNS; b++) {

103
src/enqueue.cc
View File

@ -940,14 +940,33 @@ ncclResult_t ncclLaunchPrepare(struct ncclComm* comm) {
struct ncclKernelPlan* planHead = ncclIntruQueueHead(&comm->planQueue); struct ncclKernelPlan* planHead = ncclIntruQueueHead(&comm->planQueue);
comm->unlaunchedPlansHead = planHead; comm->unlaunchedPlansHead = planHead;
// Semantically we want these dependencies for the kernels launched:
// 1. Launch host task on hostStream.
// 2. Launch kernel, depends on all of {deviceStream, hostStream, userStream[i]...}
// 3. {deviceStream, userStream[i]...} depend on kernel.
// We achieve this by:
// 1. userStream[0] waits on deviceStream
// 2. deviceStream waits on each of userStream[1...]
// 3. host task launch on hostStream
// 4. userStream[0] waits on hostStream
// 5. kernel launch on userStream[0]
// 6. deviceStream waits on userStream[0]
// 7. userStream[1...] each waits on deviceStream
// The two-level fan-in fan-out is because ncclStrongStreamWaitStream() requires
// at least one of the two streams to be strong-stream.
cudaStream_t launchStream = tasks->streams->stream;
NCCLCHECKGOTO(ncclStrongStreamAcquire(tasks->capturingGraph, &comm->deviceStream), result, failure); NCCLCHECKGOTO(ncclStrongStreamAcquire(tasks->capturingGraph, &comm->deviceStream), result, failure);
// Create dependency for nccl device work on user streams. // Create dependency for device stream on user streams. First from extra user
for (struct ncclCudaStreamList* l=tasks->streams; l != nullptr; l = l->next) { // streams to deviceStream. Then deviceStream to first user stream.
for (struct ncclCudaStreamList* l=tasks->streams->next; l != nullptr; l = l->next) {
NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, &comm->deviceStream, l->stream), result, failure); NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, &comm->deviceStream, l->stream), result, failure);
} }
NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, launchStream, &comm->deviceStream), result, failure);
if (persistent || comm->persistentRefs != 0) { if (persistent || comm->persistentRefs != 0) {
// We have to launch host tasks to push proxy args. We are careful to only
// do this if necessary since host tasks impose a high performance cost in CUDA.
bool acquired = false; bool acquired = false;
for (struct ncclKernelPlan* plan=planHead; plan != nullptr; plan = plan->next) { for (struct ncclKernelPlan* plan=planHead; plan != nullptr; plan = plan->next) {
if (plan->hasProxyOps) { if (plan->hasProxyOps) {
@ -959,6 +978,8 @@ ncclResult_t ncclLaunchPrepare(struct ncclComm* comm) {
} }
} }
if (acquired) { if (acquired) {
// Make to-be-launched kernels dependent on just-launched host stream tasks.
NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, launchStream, &comm->hostStream), result, failure);
NCCLCHECKGOTO(ncclStrongStreamRelease(tasks->capturingGraph, &comm->hostStream), result, failure); NCCLCHECKGOTO(ncclStrongStreamRelease(tasks->capturingGraph, &comm->hostStream), result, failure);
} }
} }
@ -984,14 +1005,67 @@ ncclResult_t ncclLaunchKernelBefore_NoUncapturedCuda(struct ncclComm* comm, stru
return ncclSuccess; return ncclSuccess;
} }
#if CUDART_VERSION >= 11080
#define NCCL_MAX_CGA_CLUSTER_SIZE 8
NCCL_PARAM(CGAClusterSize, "CGA_CLUSTER_SIZE", 0);
#endif
ncclResult_t ncclLaunchKernel(struct ncclComm* comm, struct ncclKernelPlan* plan) { ncclResult_t ncclLaunchKernel(struct ncclComm* comm, struct ncclKernelPlan* plan) {
struct ncclTasks* tasks = &comm->tasks; struct ncclTasks* tasks = &comm->tasks;
void *fn = plan->kernelFn;
cudaStream_t launchStream = tasks->streams->stream;
dim3 grid = {(unsigned)plan->channelCount, 1, 1}; dim3 grid = {(unsigned)plan->channelCount, 1, 1};
dim3 block = {(unsigned)plan->threadPerBlock, 1, 1}; dim3 block = {(unsigned)plan->threadPerBlock, 1, 1};
void *args[3] = {&comm->devComm, &plan->channelMask, &plan->workHead}; void *args[3] = {&comm->devComm, &plan->channelMask, &plan->workHead};
NCCLCHECK(ncclStrongStreamLaunchKernel(
tasks->capturingGraph, &comm->deviceStream, plan->kernelFn, grid, block, args, 0 #if CUDART_VERSION >= 11080
)); int driverVersion;
NCCLCHECK(ncclCudaDriverVersion(&driverVersion));
unsigned int clusterSize = 0;
clusterSize = ncclParamCGAClusterSize();
if (clusterSize > NCCL_MAX_CGA_CLUSTER_SIZE) {
static bool warned = false;
if (warned == false) {
WARN("NCCL_CGA_CLUSTER_SIZE value %d is too big. Limiting value to %d.",
clusterSize, NCCL_MAX_CGA_CLUSTER_SIZE);
warned = true;
}
clusterSize = NCCL_MAX_CGA_CLUSTER_SIZE;
}
if (clusterSize && driverVersion >= 11080) {
cudaLaunchConfig_t launchConfig = {0};
cudaLaunchAttribute launchAttrs[2];
/* Cooperative Group Array (CGA)
* On sm90 and later we have an extra level of hierarchy where we
* can group together several blocks within the Grid, called
* Thread Block Clusters.
* Clusters enable multiple thread blocks running concurrently
* across multiple SMs to synchronize and collaboratively fetch
* and exchange data. A cluster of blocks are guaranteed to be
* concurrently scheduled onto a group of SMs.
* The maximum value is 8 and it must be divisible into the grid dimensions
*/
// Grid dimension must be divisible by clusterSize
if (grid.x % clusterSize) clusterSize = 1;
launchAttrs[0].id = cudaLaunchAttributeClusterDimension;
launchAttrs[0].val.clusterDim = {clusterSize, 1, 1};
launchAttrs[1].id = cudaLaunchAttributeClusterSchedulingPolicyPreference;
launchAttrs[1].val.clusterSchedulingPolicyPreference = cudaClusterSchedulingPolicySpread;
launchConfig.gridDim = grid;
launchConfig.blockDim = block;
launchConfig.attrs = launchAttrs;
launchConfig.numAttrs = sizeof(launchAttrs)/sizeof(launchAttrs[0]);
launchConfig.stream = launchStream;
CUDACHECK(cudaLaunchKernelExC(&launchConfig, fn, args));
return ncclSuccess;
}
#endif
// Standard kernel launch
CUDACHECK(cudaLaunchKernel(fn, grid, block, args, 0, launchStream));
return ncclSuccess; return ncclSuccess;
} }
@ -1017,17 +1091,21 @@ ncclResult_t ncclLaunchFinish(struct ncclComm* comm) {
// Reset queue to empty without destroying plans since those will be sent // Reset queue to empty without destroying plans since those will be sent
// back to us for reclaiming via callbackQueue. // back to us for reclaiming via callbackQueue.
ncclIntruQueueConstruct(&comm->planQueue); ncclIntruQueueConstruct(&comm->planQueue);
// Close strong stream "transaction" encompassing cuda launches cudaStream_t launchStream = tasks->streams->stream; // First user stream gets launch
NCCLCHECKGOTO(ncclStrongStreamRelease(tasks->capturingGraph, &comm->deviceStream), result, resume1); // Create dependency for deviceStream on launchStream.
NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, &comm->deviceStream, launchStream), result, resume1);
resume1: resume1:
// Create dependency for user streams on nccl device work. // Create dependency for other user streams (skip launch stream).
struct ncclCudaStreamList* sl = tasks->streams; struct ncclCudaStreamList* sl = tasks->streams->next;
tasks->streams = nullptr; // reset streams to empty tasks->streams = nullptr; // Reset comm->tasks.streams to empty.
while (sl != nullptr) { while (sl != nullptr) {
NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, sl->stream, &comm->deviceStream), result, resume2); NCCLCHECKGOTO(ncclStrongStreamWaitStream(tasks->capturingGraph, sl->stream, &comm->deviceStream), result, resume2);
resume2: resume2:
sl = sl->next; sl = sl->next;
} }
// Release device stream as acquired in ncclLaunchPrepare()
NCCLCHECKGOTO(ncclStrongStreamRelease(tasks->capturingGraph, &comm->deviceStream), result, resume3);
resume3:;
} }
return result; return result;
} }
@ -1364,12 +1442,12 @@ static ncclResult_t taskAppend(struct ncclComm* comm, struct ncclInfo const* inf
NCCLCHECK(ncclChannelComputeFromBase(comm, channelBaseId, c, &channelId)); NCCLCHECK(ncclChannelComputeFromBase(comm, channelBaseId, c, &channelId));
if (isSendNotRecv) { if (isSendNotRecv) {
if (comm->channels[channelId].peers[peer].send[1].connected == 0) { // P2P uses only 1 connector if (comm->channels[channelId].peers[peer].send[1].connected == 0) { // P2P uses only 1 connector
comm->connectSend[peer] |= (1<<channelId); comm->connectSend[peer] |= (1UL<<channelId);
ncclGroupCommPreconnect(comm); ncclGroupCommPreconnect(comm);
} }
} else { } else {
if (comm->channels[channelId].peers[peer].recv[1].connected == 0) { // P2P uses only 1 connector if (comm->channels[channelId].peers[peer].recv[1].connected == 0) { // P2P uses only 1 connector
comm->connectRecv[peer] |= (1<<channelId); comm->connectRecv[peer] |= (1UL<<channelId);
ncclGroupCommPreconnect(comm); ncclGroupCommPreconnect(comm);
} }
} }
@ -1429,6 +1507,7 @@ static ncclResult_t taskAppend(struct ncclComm* comm, struct ncclInfo const* inf
} }
if (l->stream == info->stream) if (l->stream == info->stream)
break; // Already seen stream. break; // Already seen stream.
l = l->next;
} }
} }
return ncclSuccess; return ncclSuccess;

13
src/graph/paths.cc
View File

@ -399,6 +399,19 @@ ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* system, int64_t busId, int
return ncclSuccess; return ncclSuccess;
} }
// Set to 0 to disable the flush on Hopper when using GDR
NCCL_PARAM(NetForceFlush, "NET_FORCE_FLUSH", 1);
// Determine whether we need to flush the GDR recv buffers
ncclResult_t ncclTopoNeedFlush(struct ncclTopoSystem* system, int64_t busId, int* flush) {
int g;
NCCLCHECK(ncclTopoIdToIndex(system, GPU, busId, &g));
struct ncclTopoNode* gpu = system->nodes[GPU].nodes+g;
// Flush is required on Ampere and earlier
*flush = gpu->gpu.cudaCompCap < 90 ? 1 : ncclParamNetForceFlush();
return ncclSuccess;
}
NCCL_PARAM(NetDisableIntra, "NET_DISABLE_INTRA", 0); NCCL_PARAM(NetDisableIntra, "NET_DISABLE_INTRA", 0);
// Check whether going through the network would be faster than going through P2P/SHM. // Check whether going through the network would be faster than going through P2P/SHM.

2
src/graph/search.cc
View File

@ -727,7 +727,7 @@ ncclResult_t ncclTopoGetXmlFromGraphs(int ngraphs, struct ncclTopoGraph** graphs
} }
float speedArrayIntra[] = { 44.0, 30.0, 22.0, 18.0, 15.0, 12.0, 10.0, 9.0, 7.0, 6.0, 5.0, 4.0, 3.0 }; float speedArrayIntra[] = { 44.0, 30.0, 22.0, 18.0, 15.0, 12.0, 10.0, 9.0, 7.0, 6.0, 5.0, 4.0, 3.0 };
float speedArrayInter[] = { 48.0, 30.0, 24.0, 22.0, 18.0, 15.0, 12.0, 10.0, 9.0, 7.0, 6.0, 5.0, 4.0, 3.0, 2.4, 1.2, 0.24, 0.12 }; float speedArrayInter[] = { 48.0, 30.0, 28.0, 24.0, 22.0, 18.0, 15.0, 12.0, 10.0, 9.0, 7.0, 6.0, 5.0, 4.0, 3.0, 2.4, 1.2, 0.24, 0.12 };
#define NSPEEDSINTRA (sizeof(speedArrayIntra)/sizeof(float)) #define NSPEEDSINTRA (sizeof(speedArrayIntra)/sizeof(float))
#define NSPEEDSINTER (sizeof(speedArrayInter)/sizeof(float)) #define NSPEEDSINTER (sizeof(speedArrayInter)/sizeof(float))

30
src/graph/tuning.cc
View File

@ -72,10 +72,24 @@ static float hwLat [3][NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS] =
/* CollNetDirect (Simple)*/ { 0, 0, 10.7 }, /* CollNetChain (Simple)*/ { 0, 0, 10.7 } } /* CollNetDirect (Simple)*/ { 0, 0, 10.7 }, /* CollNetChain (Simple)*/ { 0, 0, 10.7 } }
}; };
/* Array indexes used below */
#define VOLTA_COMPCAP_IDX 0
#define AMPERE_COMPCAP_IDX 1
#define HOPPER_COMPCAP_IDX 2
// LL128 max BW per channel // LL128 max BW per channel
static const double ll128MaxBwPerCh = 20.0; static const double ll128MaxBwPerCh = 20.0;
static const double llMaxBws[2][3] = { /* Volta-N1/Intel-N2/Intel-N4) */ {39.0, 39.0, 20.4}, /* Ampere-N1/AMD-N2/AMD-N4) */ {87.7, 22.5 /*avg of ring & tree*/, 19.0} }; static const double llMaxBws[3][3] = {
static const double perChMaxTreeBws[2][3] = { /* Volta (N1/N2/N4) */ {26.5, 18.5, 10.0}, /* Ampere (N1/N2/N4) */ {24.0, 23.6, 17.8} }; /* Volta-N1/Intel-N2/Intel-N4) */ {39.0, 39.0, 20.4},
/* Ampere-N1/AMD-N2/AMD-N4) */ {87.7, 22.5 /*avg of ring & tree*/, 19.0},
/* Hopper-N1/AMD-N2/AMD-N4) */ {87.7, 22.5 /*avg of ring & tree*/, 19.0}
};
static const double perChMaxTreeBws[3][3] = {
/* Volta (N1/N2/N4) */ {26.5, 18.5, 10.0},
/* Ampere (N1/N2/N4) */ {24.0, 23.6, 17.8},
/* Hopper (N1/N2/N4) */ {24.0, 23.6, 17.8},
};
ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCompCap, struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph, struct ncclTopoGraph* collNetGraph) { ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCompCap, struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph, struct ncclTopoGraph* collNetGraph) {
int simpleDefaultThreads = (ringGraph->bwIntra*ringGraph->nChannels <= PCI_BW) ? 256 : NCCL_SIMPLE_MAX_NTHREADS; int simpleDefaultThreads = (ringGraph->bwIntra*ringGraph->nChannels <= PCI_BW) ? 256 : NCCL_SIMPLE_MAX_NTHREADS;
@ -94,14 +108,14 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
int nRanks = comm->nRanks; int nRanks = comm->nRanks;
if (nRanks <= 1) return ncclSuccess; if (nRanks <= 1) return ncclSuccess;
int compCap80 = minCompCap == 80 && maxCompCap == 80 ? 1 : 0; int compCapIndex = (minCompCap == 80 && maxCompCap == 80) ? AMPERE_COMPCAP_IDX : ((minCompCap == 90 && maxCompCap == 90) ? HOPPER_COMPCAP_IDX : VOLTA_COMPCAP_IDX);
int cpuArch, cpuVendor, cpuModel; int cpuArch, cpuVendor, cpuModel;
NCCLCHECK(ncclTopoCpuType(comm->topo, &cpuArch, &cpuVendor, &cpuModel)); NCCLCHECK(ncclTopoCpuType(comm->topo, &cpuArch, &cpuVendor, &cpuModel));
int index2 = nNodes <= 2 ? nNodes-1 : 2; int index2 = nNodes <= 2 ? nNodes-1 : 2;
// LL: for single node, we look at GPU type; for multi-node, we look at CPU type // LL: for single node, we look at GPU type; for multi-node, we look at CPU type
int index1 = nNodes == 1 ? compCap80 : cpuVendor == NCCL_TOPO_CPU_VENDOR_AMD ? 1 : 0; int index1 = nNodes == 1 ? compCapIndex : cpuVendor == NCCL_TOPO_CPU_VENDOR_AMD ? 1 : 0;
double llMaxBw = llMaxBws[index1][index2]; double llMaxBw = llMaxBws[index1][index2];
double perChMaxTreeBw = perChMaxTreeBws[compCap80][index2]; double perChMaxTreeBw = perChMaxTreeBws[compCapIndex][index2];
// De-penalize Tree/Simple latency on Power systems to favor Tree than Ring // De-penalize Tree/Simple latency on Power systems to favor Tree than Ring
if (cpuArch == NCCL_TOPO_CPU_ARCH_POWER) hwLat[NCCL_HW_PCI][NCCL_ALGO_TREE][NCCL_PROTO_SIMPLE] = hwLat[NCCL_HW_PCI][NCCL_ALGO_RING][NCCL_PROTO_SIMPLE]; if (cpuArch == NCCL_TOPO_CPU_ARCH_POWER) hwLat[NCCL_HW_PCI][NCCL_ALGO_TREE][NCCL_PROTO_SIMPLE] = hwLat[NCCL_HW_PCI][NCCL_ALGO_RING][NCCL_PROTO_SIMPLE];
float ppn = (float)nRanks / nNodes; // if ppn < 2, then we are sending/receiving at the same GPU through the NIC, apply some bw discount float ppn = (float)nRanks / nNodes; // if ppn < 2, then we are sending/receiving at the same GPU through the NIC, apply some bw discount
@ -128,7 +142,7 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
float busBw = graphs[a]->nChannels * bw; float busBw = graphs[a]->nChannels * bw;
// Various model refinements // Various model refinements
if (compCap80) busBw = std::min(busBw, 235.0f); if (compCapIndex == AMPERE_COMPCAP_IDX) busBw = std::min(busBw, 235.0f);
if (a == NCCL_ALGO_RING && p == NCCL_PROTO_LL) { busBw = std::min(llMaxBw, busBw * ((nNodes > 1 || coll == ncclFuncAllReduce || coll == ncclFuncReduce) ? 1.0/4.0 : 1.0/3.0)); } if (a == NCCL_ALGO_RING && p == NCCL_PROTO_LL) { busBw = std::min(llMaxBw, busBw * ((nNodes > 1 || coll == ncclFuncAllReduce || coll == ncclFuncReduce) ? 1.0/4.0 : 1.0/3.0)); }
if (a == NCCL_ALGO_RING && p == NCCL_PROTO_LL128) busBw = std::min(busBw * (ppn < 2 ? 0.7 : 0.92 /*120.0/128.0*/), ll128MaxBwPerCh*graphs[a]->nChannels); if (a == NCCL_ALGO_RING && p == NCCL_PROTO_LL128) busBw = std::min(busBw * (ppn < 2 ? 0.7 : 0.92 /*120.0/128.0*/), ll128MaxBwPerCh*graphs[a]->nChannels);
if (a == NCCL_ALGO_TREE) busBw = std::min(busBw*.92, graphs[a]->nChannels*perChMaxTreeBw); if (a == NCCL_ALGO_TREE) busBw = std::min(busBw*.92, graphs[a]->nChannels*perChMaxTreeBw);
@ -212,9 +226,9 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
for (int c=0; c<NCCL_NUM_FUNCTIONS; c++) for (int a=0; a<NCCL_NUM_ALGORITHMS; a++) for (int p=0; p<NCCL_NUM_PROTOCOLS; p++) { for (int c=0; c<NCCL_NUM_FUNCTIONS; c++) for (int a=0; a<NCCL_NUM_ALGORITHMS; a++) for (int p=0; p<NCCL_NUM_PROTOCOLS; p++) {
int pEnable = protoEnable[p]; int pEnable = protoEnable[p];
if (pEnable == 2 && p == NCCL_PROTO_LL128) { if (pEnable == 2 && p == NCCL_PROTO_LL128) {
// Enable LL128 by default only on Volta/Ampere+NVLink. Other cases are not tested and may cause silent data corruption. // Enable LL128 by default only on Volta/Ampere/Hopper+NVLink. Other cases are not tested and may cause silent data corruption.
pEnable = (graphs[a]->typeInter <= PATH_PXB) && graphs[a]->typeIntra <= PATH_NVL && pEnable = (graphs[a]->typeInter <= PATH_PXB) && graphs[a]->typeIntra <= PATH_NVL &&
((minCompCap == 70 && maxCompCap == 70) || (minCompCap == 80 && maxCompCap == 80)) ? 1 : 0; ((minCompCap == 70 && maxCompCap == 70) || (minCompCap == 80 && maxCompCap == 80) || (minCompCap == 90 && maxCompCap == 90)) ? 1 : 0;
} }
if (pEnable == 0) comm->bandwidths[c][a][p] = 0; if (pEnable == 0) comm->bandwidths[c][a][p] = 0;
// Only disable algo for Allreduce since others only have one // Only disable algo for Allreduce since others only have one

19
src/graph/xml.cc
View File

@ -628,7 +628,7 @@ ncclResult_t ncclTopoGetXmlFromGpu(struct ncclXmlNode* pciNode, nvmlDevice_t nvm
NCCLCHECK(xmlGetSub(gpuNode, "nvlink", &nvlNode)); NCCLCHECK(xmlGetSub(gpuNode, "nvlink", &nvlNode));
if (nvlNode == NULL) { if (nvlNode == NULL) {
// NVML NVLink detection // NVML NVLink detection
int maxNvLinks = (sm < 60) ? 0 : (sm < 70) ? 4 : (sm < 80) ? 6 : 12; int maxNvLinks = (sm < 60) ? 0 : (sm < 70) ? 4 : (sm < 80) ? 6 : (sm < 90) ? 12 : 18;
if (maxNvLinks > 0 && nvmlDev == NULL) { if (maxNvLinks > 0 && nvmlDev == NULL) {
WARN("No NVML device handle. Skipping nvlink detection."); WARN("No NVML device handle. Skipping nvlink detection.");
@ -641,8 +641,21 @@ ncclResult_t ncclTopoGetXmlFromGpu(struct ncclXmlNode* pciNode, nvmlDevice_t nvm
if ((ncclNvmlDeviceGetNvLinkCapability(nvmlDev, l, NVML_NVLINK_CAP_P2P_SUPPORTED, &canP2P) != ncclSuccess) || !canP2P) continue; if ((ncclNvmlDeviceGetNvLinkCapability(nvmlDev, l, NVML_NVLINK_CAP_P2P_SUPPORTED, &canP2P) != ncclSuccess) || !canP2P) continue;
// Make sure the Nvlink is up. The previous call should have trained the link. // Make sure the Nvlink is up. The previous call should have trained the link.
nvmlEnableState_t isActive; nvmlEnableState_t isActive = NVML_FEATURE_DISABLED;
if ((ncclNvmlDeviceGetNvLinkState(nvmlDev, l, &isActive) != ncclSuccess) || (isActive != NVML_FEATURE_ENABLED)) continue; #if CUDART_VERSION >= 11080
if (sm >= 90) {
nvmlFieldValue_t fv;
fv.fieldId = NVML_FI_DEV_NVLINK_GET_STATE;
fv.scopeId = l;
// fv.value will contain NV_FEATURE_ENABLED or NV_FEATURE_DISABLED
if ((ncclNvmlDeviceGetFieldValues(nvmlDev, 1, &fv) == ncclSuccess) && (fv.nvmlReturn == NVML_SUCCESS))
isActive = (nvmlEnableState_t) fv.value.uiVal;
} else /* FALLTHRU to GetNvLinkState if before SM90 */
#endif
{
(void) ncclNvmlDeviceGetNvLinkState(nvmlDev, l, &isActive);
}
if (isActive != NVML_FEATURE_ENABLED) continue;
// Try to figure out what's on the other side of the NVLink // Try to figure out what's on the other side of the NVLink
nvmlPciInfo_t remoteProc; nvmlPciInfo_t remoteProc;

4
src/group.cc
View File

@ -211,8 +211,8 @@ static void groupCleanup(struct ncclComm** groupCommHeadPtr, struct ncclComm** g
for (int i = 0; i < comm->nRanks; i++) { for (int i = 0; i < comm->nRanks; i++) {
comm->tasks.peers[i].sendSeen = false; comm->tasks.peers[i].sendSeen = false;
comm->tasks.peers[i].recvSeen = false; comm->tasks.peers[i].recvSeen = false;
comm->connectSend[i] = 0; comm->connectSend[i] = 0UL;
comm->connectRecv[i] = 0; comm->connectRecv[i] = 0UL;
} }
comm->unlaunchedPlansHead = nullptr; comm->unlaunchedPlansHead = nullptr;
// Reclaim abandoned kernel plan memory. Note ncclWork structs were already // Reclaim abandoned kernel plan memory. Note ncclWork structs were already

4
src/include/comm.h
View File

@ -168,8 +168,8 @@ struct ncclComm {
ncclCollNet_t* ncclCollNet; ncclCollNet_t* ncclCollNet;
void* bootstrap; void* bootstrap;
// Bitmasks for ncclTransportP2pSetup // Bitmasks for ncclTransportP2pSetup
uint32_t* connectSend; uint64_t* connectSend;
uint32_t* connectRecv; uint64_t* connectRecv;
int rank; // my rank in the communicator int rank; // my rank in the communicator
int nRanks; // number of GPUs in communicator int nRanks; // number of GPUs in communicator

16
src/include/cudawrap.h
View File

@ -8,6 +8,8 @@
#define NCCL_CUDAWRAP_H_ #define NCCL_CUDAWRAP_H_
#include <cuda.h> #include <cuda.h>
#include <cuda_runtime.h>
#include "checks.h"
#if CUDART_VERSION >= 11030 #if CUDART_VERSION >= 11030
#include <cudaTypedefs.h> #include <cudaTypedefs.h>
@ -83,6 +85,18 @@ DECLARE_CUDA_PFN_EXTERN(cuDriverGetVersion, 2020);
DECLARE_CUDA_PFN_EXTERN(cuGetProcAddress, 11030); DECLARE_CUDA_PFN_EXTERN(cuGetProcAddress, 11030);
ncclResult_t cudaLibraryInit(void); ncclResult_t ncclCudaLibraryInit(void);
extern int ncclCudaDriverVersionCache;
inline ncclResult_t ncclCudaDriverVersion(int* driver) {
int version = __atomic_load_n(&ncclCudaDriverVersionCache, __ATOMIC_RELAXED);
if (version == -1) {
CUDACHECK(cudaDriverGetVersion(&version));
__atomic_store_n(&ncclCudaDriverVersionCache, version, __ATOMIC_RELAXED);
}
*driver = version;
return ncclSuccess;
}
#endif #endif

1
src/include/graph.h
View File

@ -33,6 +33,7 @@ ncclResult_t ncclTopoGetNvbGpus(struct ncclTopoSystem* system, int rank, int* nr
ncclResult_t ncclTopoGetNetDev(struct ncclComm* comm, int rank, struct ncclTopoGraph* graph, int channelId, int peerRank, int* net, int* proxyRank); ncclResult_t ncclTopoGetNetDev(struct ncclComm* comm, int rank, struct ncclTopoGraph* graph, int channelId, int peerRank, int* net, int* proxyRank);
ncclResult_t ncclTopoCheckP2p(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* p2p, int *read, int* intermediateRank); ncclResult_t ncclTopoCheckP2p(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* p2p, int *read, int* intermediateRank);
ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* topo, int64_t busId, int netDev, int read, int* useGdr); ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* topo, int64_t busId, int netDev, int read, int* useGdr);
ncclResult_t ncclTopoNeedFlush(struct ncclTopoSystem* system, int64_t busId, int* flush);
ncclResult_t ncclTopoCheckNet(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* net); ncclResult_t ncclTopoCheckNet(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* net);
int ncclPxnDisable(struct ncclComm* comm); int ncclPxnDisable(struct ncclComm* comm);
ncclResult_t ncclTopoGetPxnRanks(struct ncclComm* comm, int** intermediateRanks, int* nranks); ncclResult_t ncclTopoGetPxnRanks(struct ncclComm* comm, int** intermediateRanks, int* nranks);

71
src/include/nvmlwrap.h
View File

@ -107,6 +107,75 @@ typedef enum nvmlGpuP2PCapsIndex_enum
NVML_P2P_CAPS_INDEX_UNKNOWN NVML_P2P_CAPS_INDEX_UNKNOWN
} nvmlGpuP2PCapsIndex_t; } nvmlGpuP2PCapsIndex_t;
/**
* Represents the type for sample value returned
*/
typedef enum nvmlValueType_enum
{
NVML_VALUE_TYPE_DOUBLE = 0,
NVML_VALUE_TYPE_UNSIGNED_INT = 1,
NVML_VALUE_TYPE_UNSIGNED_LONG = 2,
NVML_VALUE_TYPE_UNSIGNED_LONG_LONG = 3,
NVML_VALUE_TYPE_SIGNED_LONG_LONG = 4,
// Keep this last
NVML_VALUE_TYPE_COUNT
}nvmlValueType_t;
/**
* Union to represent different types of Value
*/
typedef union nvmlValue_st
{
double dVal; //!< If the value is double
unsigned int uiVal; //!< If the value is unsigned int
unsigned long ulVal; //!< If the value is unsigned long
unsigned long long ullVal; //!< If the value is unsigned long long
signed long long sllVal; //!< If the value is signed long long
}nvmlValue_t;
/**
* Field Identifiers.
*
* All Identifiers pertain to a device. Each ID is only used once and is guaranteed never to change.
*/
/* NVLink Speed */
#define NVML_FI_DEV_NVLINK_SPEED_MBPS_COMMON 90 //!< Common NVLink Speed in MBps for active links
#define NVML_FI_DEV_NVLINK_LINK_COUNT 91 //!< Number of NVLinks present on the device
/**
* Remote device NVLink ID
*
* Link ID needs to be specified in the scopeId field in nvmlFieldValue_t.
*/
#define NVML_FI_DEV_NVLINK_REMOTE_NVLINK_ID 146 //!< Remote device NVLink ID
/**
* NVSwitch: connected NVLink count
*/
#define NVML_FI_DEV_NVSWITCH_CONNECTED_LINK_COUNT 147 //!< Number of NVLinks connected to NVSwitch
#define NVML_FI_DEV_NVLINK_GET_SPEED 164
#define NVML_FI_DEV_NVLINK_GET_STATE 165
#define NVML_FI_DEV_NVLINK_GET_VERSION 166
#define NVML_FI_MAX 167 //!< One greater than the largest field ID defined above
/**
* Information for a Field Value Sample
*/
typedef struct nvmlFieldValue_st
{
unsigned int fieldId; //!< ID of the NVML field to retrieve. This must be set before any call that uses this struct. See the constants starting with NVML_FI_ above.
unsigned int scopeId; //!< Scope ID can represent data used by NVML depending on fieldId's context. For example, for NVLink throughput counter data, scopeId can represent linkId.
long long timestamp; //!< CPU Timestamp of this value in microseconds since 1970
long long latencyUsec; //!< How long this field value took to update (in usec) within NVML. This may be averaged across several fields that are serviced by the same driver call.
nvmlValueType_t valueType; //!< Type of the value stored in value
nvmlReturn_t nvmlReturn; //!< Return code for retrieving this value. This must be checked before looking at value, as value is undefined if nvmlReturn != NVML_SUCCESS
nvmlValue_t value; //!< Value for this field. This is only valid if nvmlReturn == NVML_SUCCESS
} nvmlFieldValue_t;
/* End of nvml.h */ /* End of nvml.h */
#endif // NCCL_NVML_DIRECT #endif // NCCL_NVML_DIRECT
@ -135,4 +204,6 @@ ncclResult_t ncclNvmlDeviceGetNvLinkRemotePciInfo(nvmlDevice_t device, unsigned
ncclResult_t ncclNvmlDeviceGetNvLinkCapability(nvmlDevice_t device, unsigned int link, nvmlNvLinkCapability_t capability, unsigned int *capResult); ncclResult_t ncclNvmlDeviceGetNvLinkCapability(nvmlDevice_t device, unsigned int link, nvmlNvLinkCapability_t capability, unsigned int *capResult);
ncclResult_t ncclNvmlDeviceGetCudaComputeCapability(nvmlDevice_t device, int* major, int* minor); ncclResult_t ncclNvmlDeviceGetCudaComputeCapability(nvmlDevice_t device, int* major, int* minor);
ncclResult_t ncclNvmlDeviceGetP2PStatus(nvmlDevice_t device1, nvmlDevice_t device2, nvmlGpuP2PCapsIndex_t p2pIndex, nvmlGpuP2PStatus_t* p2pStatus); ncclResult_t ncclNvmlDeviceGetP2PStatus(nvmlDevice_t device1, nvmlDevice_t device2, nvmlGpuP2PCapsIndex_t p2pIndex, nvmlGpuP2PStatus_t* p2pStatus);
ncclResult_t ncclNvmlDeviceGetFieldValues(nvmlDevice_t device, int valuesCount, nvmlFieldValue_t *values);
#endif // End include guard #endif // End include guard

5
src/include/strongstream.h
View File

@ -22,7 +22,7 @@ struct ncclCudaGraph {
#endif #endif
}; };
inline struct ncclCudaGraph ncclCudaGraphNull() { inline struct ncclCudaGraph ncclCudaGraphNone() {
struct ncclCudaGraph tmp; struct ncclCudaGraph tmp;
#if CUDART_VERSION >= 11030 #if CUDART_VERSION >= 11030
tmp.graph = nullptr; tmp.graph = nullptr;
@ -50,7 +50,6 @@ inline bool ncclCudaGraphSame(struct ncclCudaGraph a, struct ncclCudaGraph b) {
ncclResult_t ncclCudaGetCapturingGraph(struct ncclCudaGraph* graph, cudaStream_t stream); ncclResult_t ncclCudaGetCapturingGraph(struct ncclCudaGraph* graph, cudaStream_t stream);
ncclResult_t ncclCudaGraphAddDestructor(struct ncclCudaGraph graph, cudaHostFn_t fn, void* arg); ncclResult_t ncclCudaGraphAddDestructor(struct ncclCudaGraph graph, cudaHostFn_t fn, void* arg);
/* ncclStrongStream: An abstraction over CUDA streams that do not lose their /* ncclStrongStream: An abstraction over CUDA streams that do not lose their
* identity while being captured. Regular streams have the deficiency that the * identity while being captured. Regular streams have the deficiency that the
* captured form of a stream in one graph launch has no relation to the * captured form of a stream in one graph launch has no relation to the
@ -88,7 +87,7 @@ ncclResult_t ncclStrongStreamAcquire(
// Acquire-fence the strong stream assuming no graph is capturing. This permits // Acquire-fence the strong stream assuming no graph is capturing. This permits
// the caller to enqueue directly to the `ss->stream` member using native CUDA // the caller to enqueue directly to the `ss->stream` member using native CUDA
// calls. Strong stream must be released via: // calls. Strong stream must be released via:
// ncclStrongStreamRelease(ncclCudaGraphNull(), graphRefs, ss); // ncclStrongStreamRelease(ncclCudaGraphNone(), ss);
ncclResult_t ncclStrongStreamAcquireUncaptured(struct ncclStrongStream* ss); ncclResult_t ncclStrongStreamAcquireUncaptured(struct ncclStrongStream* ss);
// Release-fence of the strong stream. // Release-fence of the strong stream.

12
src/init.cc
View File

@ -416,7 +416,7 @@ static ncclResult_t devCommSetup(ncclComm_t comm) {
NCCLCHECK(ncclCudaMemcpyAsync(devCommAndChans, &tmpCommAndChans, 1, comm->deviceStream.stream)); NCCLCHECK(ncclCudaMemcpyAsync(devCommAndChans, &tmpCommAndChans, 1, comm->deviceStream.stream));
CUDACHECK(cudaStreamSynchronize(comm->deviceStream.stream)); CUDACHECK(cudaStreamSynchronize(comm->deviceStream.stream));
NCCLCHECK(ncclStrongStreamRelease(ncclCudaGraphNull(), &comm->deviceStream)); NCCLCHECK(ncclStrongStreamRelease(ncclCudaGraphNone(), &comm->deviceStream));
return ncclSuccess; return ncclSuccess;
} }
@ -955,13 +955,13 @@ collnet_cleanup:
for (int c=0; c<comm->p2pnChannelsPerPeer; c++) { for (int c=0; c<comm->p2pnChannelsPerPeer; c++) {
NCCLCHECK(ncclChannelCompute(comm, peer, c, ncclFuncSend, &channelId)); NCCLCHECK(ncclChannelCompute(comm, peer, c, ncclFuncSend, &channelId));
if (comm->channels[channelId].peers[peer].send[1].connected == 0) { if (comm->channels[channelId].peers[peer].send[1].connected == 0) {
comm->connectSend[peer] |= (1<<channelId); comm->connectSend[peer] |= (1UL<<channelId);
} }
} }
for (int c=0; c<comm->p2pnChannelsPerPeer; c++) { for (int c=0; c<comm->p2pnChannelsPerPeer; c++) {
NCCLCHECK(ncclChannelCompute(comm, peer, c, ncclFuncRecv, &channelId)); NCCLCHECK(ncclChannelCompute(comm, peer, c, ncclFuncRecv, &channelId));
if (comm->channels[channelId].peers[peer].recv[1].connected == 0) { if (comm->channels[channelId].peers[peer].recv[1].connected == 0) {
comm->connectRecv[peer] |= (1<<channelId); comm->connectRecv[peer] |= (1UL<<channelId);
} }
} }
} }
@ -1172,7 +1172,7 @@ ncclResult_t ncclCommInitRank(ncclComm_t* newcomm, int nranks, ncclUniqueId comm
NVTX3_FUNC_RANGE_IN(nccl_domain); NVTX3_FUNC_RANGE_IN(nccl_domain);
// Load the CUDA driver and dlsym hooks (can fail on old drivers) // Load the CUDA driver and dlsym hooks (can fail on old drivers)
(void) cudaLibraryInit(); (void)ncclCudaLibraryInit();
int cudaDev; int cudaDev;
CUDACHECK(cudaGetDevice(&cudaDev)); CUDACHECK(cudaGetDevice(&cudaDev));
@ -1187,7 +1187,7 @@ ncclResult_t ncclCommInitAll(ncclComm_t* comms, int ndev, const int* devlist) {
int totalnDev; int totalnDev;
int *gpuFlags = NULL; int *gpuFlags = NULL;
// Load the CUDA driver and dlsym hooks (can fail on old drivers) // Load the CUDA driver and dlsym hooks (can fail on old drivers)
(void) cudaLibraryInit(); (void)ncclCudaLibraryInit();
NCCLCHECKGOTO(PtrCheck(comms, "CommInitAll", "comms"), ret, fail); NCCLCHECKGOTO(PtrCheck(comms, "CommInitAll", "comms"), ret, fail);
if (ndev < 0) { if (ndev < 0) {
@ -1279,7 +1279,7 @@ ncclResult_t ncclCommInitRankConfig(ncclComm_t *newcomm, int nranks, ncclUniqueI
} }
if (blockingEnv == 1) internalConfigPtr->blocking = blockingEnv; if (blockingEnv == 1) internalConfigPtr->blocking = blockingEnv;
(void) cudaLibraryInit(); (void)ncclCudaLibraryInit();
CUDACHECKGOTO(cudaGetDevice(&cudaDev), ret, exit); CUDACHECKGOTO(cudaGetDevice(&cudaDev), ret, exit);
NCCLCHECKGOTO(ncclCommInitRankDev(newcomm, nranks, commId, myrank, cudaDev, internalConfigPtr), ret, fail); NCCLCHECKGOTO(ncclCommInitRankDev(newcomm, nranks, commId, myrank, cudaDev, internalConfigPtr), ret, fail);

13
src/misc/cudawrap.cc
View File

@ -38,7 +38,7 @@ DECLARE_CUDA_PFN(cuGetProcAddress, 11030);
#define CUDA_DRIVER_MIN_VERSION 11030 #define CUDA_DRIVER_MIN_VERSION 11030
static void *cudaLib; static void *cudaLib;
static int cudaDriverVersion; int ncclCudaDriverVersionCache = -1;
#if CUDART_VERSION >= 11030 #if CUDART_VERSION >= 11030
/* /*
@ -107,16 +107,17 @@ static void initOnceFunc() {
goto error; goto error;
} }
res = pfn_cuDriverGetVersion(&cudaDriverVersion); int driverVersion;
res = pfn_cuDriverGetVersion(&driverVersion);
if (res != 0) { if (res != 0) {
WARN("cuDriverGetVersion failed with %d", res); WARN("cuDriverGetVersion failed with %d", res);
goto error; goto error;
} }
INFO(NCCL_INIT, "cudaDriverVersion %d", cudaDriverVersion); INFO(NCCL_INIT, "cudaDriverVersion %d", driverVersion);
if (cudaDriverVersion < CUDA_DRIVER_MIN_VERSION) { if (driverVersion < CUDA_DRIVER_MIN_VERSION) {
// WARN("CUDA Driver version found is %d. Minimum requirement is %d", cudaDriverVersion, CUDA_DRIVER_MIN_VERSION); // WARN("CUDA Driver version found is %d. Minimum requirement is %d", driverVersion, CUDA_DRIVER_MIN_VERSION);
// Silently ignore version check mismatch for backwards compatibility // Silently ignore version check mismatch for backwards compatibility
goto error; goto error;
} }
@ -148,7 +149,7 @@ error:
return; return;
} }
ncclResult_t cudaLibraryInit() { ncclResult_t ncclCudaLibraryInit() {
pthread_once(&initOnceControl, initOnceFunc); pthread_once(&initOnceControl, initOnceFunc);
return initResult; return initResult;
} }

11
src/misc/nvmlwrap.cc
View File

@ -38,6 +38,7 @@ namespace {
NCCL_NVML_FN(nvmlDeviceGetNvLinkCapability, nvmlReturn_t, (nvmlDevice_t device, unsigned int link, nvmlNvLinkCapability_t capability, unsigned int *capResult)) NCCL_NVML_FN(nvmlDeviceGetNvLinkCapability, nvmlReturn_t, (nvmlDevice_t device, unsigned int link, nvmlNvLinkCapability_t capability, unsigned int *capResult))
NCCL_NVML_FN(nvmlDeviceGetCudaComputeCapability, nvmlReturn_t, (nvmlDevice_t device, int* major, int* minor)) NCCL_NVML_FN(nvmlDeviceGetCudaComputeCapability, nvmlReturn_t, (nvmlDevice_t device, int* major, int* minor))
NCCL_NVML_FN(nvmlDeviceGetP2PStatus, nvmlReturn_t, (nvmlDevice_t device1, nvmlDevice_t device2, nvmlGpuP2PCapsIndex_t p2pIndex, nvmlGpuP2PStatus_t* p2pStatus)) NCCL_NVML_FN(nvmlDeviceGetP2PStatus, nvmlReturn_t, (nvmlDevice_t device1, nvmlDevice_t device2, nvmlGpuP2PCapsIndex_t p2pIndex, nvmlGpuP2PStatus_t* p2pStatus))
NCCL_NVML_FN(nvmlDeviceGetFieldValues, nvmlReturn_t, (nvmlDevice_t device, int valuesCount, nvmlFieldValue_t *values))
std::mutex lock; // NVML has had some thread safety bugs std::mutex lock; // NVML has had some thread safety bugs
bool initialized = false; bool initialized = false;
@ -80,7 +81,8 @@ ncclResult_t ncclNvmlEnsureInitialized() {
{(void**)&pfn_nvmlDeviceGetNvLinkRemotePciInfo, "nvmlDeviceGetNvLinkRemotePciInfo"}, {(void**)&pfn_nvmlDeviceGetNvLinkRemotePciInfo, "nvmlDeviceGetNvLinkRemotePciInfo"},
{(void**)&pfn_nvmlDeviceGetNvLinkCapability, "nvmlDeviceGetNvLinkCapability"}, {(void**)&pfn_nvmlDeviceGetNvLinkCapability, "nvmlDeviceGetNvLinkCapability"},
{(void**)&pfn_nvmlDeviceGetCudaComputeCapability, "nvmlDeviceGetCudaComputeCapability"}, {(void**)&pfn_nvmlDeviceGetCudaComputeCapability, "nvmlDeviceGetCudaComputeCapability"},
{(void**)&pfn_nvmlDeviceGetP2PStatus, "nvmlDeviceGetP2PStatus"} {(void**)&pfn_nvmlDeviceGetP2PStatus, "nvmlDeviceGetP2PStatus"},
{(void**)&pfn_nvmlDeviceGetFieldValues, "nvmlDeviceGetFieldValues"}
}; };
for(Symbol sym: symbols) { for(Symbol sym: symbols) {
*sym.ppfn = dlsym(libhandle, sym.name); *sym.ppfn = dlsym(libhandle, sym.name);
@ -260,3 +262,10 @@ ncclResult_t ncclNvmlDeviceGetP2PStatus(
} }
return ncclSuccess; return ncclSuccess;
} }
ncclResult_t ncclNvmlDeviceGetFieldValues(nvmlDevice_t device, int valuesCount, nvmlFieldValue_t *values) {
NCCLCHECK(ncclNvmlEnsureInitialized());
std::lock_guard<std::mutex> locked(lock);
NVMLTRY(nvmlDeviceGetFieldValues, device, valuesCount, values);
return ncclSuccess;
}

12
src/misc/strongstream.cc
View File

@ -5,6 +5,7 @@
************************************************************************/ ************************************************************************/
#include "strongstream.h" #include "strongstream.h"
#include "cudawrap.h"
#include "checks.h" #include "checks.h"
#include "param.h" #include "param.h"
@ -14,10 +15,8 @@ ncclResult_t ncclCudaGetCapturingGraph(
struct ncclCudaGraph* graph, cudaStream_t stream struct ncclCudaGraph* graph, cudaStream_t stream
) { ) {
#if CUDART_VERSION >= 11030 #if CUDART_VERSION >= 11030
thread_local int driver = -1; int driver;
if (driver == -1) { NCCLCHECK(ncclCudaDriverVersion(&driver));
CUDACHECK(cudaDriverGetVersion(&driver));
}
if (driver < 11030) { if (driver < 11030) {
cudaStreamCaptureStatus status; cudaStreamCaptureStatus status;
unsigned long long gid; unsigned long long gid;
@ -192,11 +191,11 @@ ncclResult_t ncclStrongStreamWaitStream(
CUDACHECK(cudaEventRecord(b->event, b->stream)); CUDACHECK(cudaEventRecord(b->event, b->stream));
} }
CUDACHECK(cudaStreamWaitEvent(a->stream, b->event, 0)); CUDACHECK(cudaStreamWaitEvent(a->stream, b->event, 0));
a->eventIsLagging = 1;
} else { } else {
cudaGraphNode_t pair[2] = {a->node, b->node}; cudaGraphNode_t pair[2] = {a->node, b->node};
CUDACHECK(cudaGraphAddEmptyNode(&a->node, graph.graph, pair, 2)); CUDACHECK(cudaGraphAddEmptyNode(&a->node, graph.graph, pair, 2));
} }
a->eventIsLagging = 1;
#else #else
CUDACHECK(cudaEventRecord(b->event, b->stream)); CUDACHECK(cudaEventRecord(b->event, b->stream));
CUDACHECK(cudaStreamWaitEvent(a->stream, b->event, 0)); CUDACHECK(cudaStreamWaitEvent(a->stream, b->event, 0));
@ -232,9 +231,8 @@ ncclResult_t ncclStrongStreamWaitStream(
} }
cudaGraphNode_t pair[2] = {a->node, tie}; cudaGraphNode_t pair[2] = {a->node, tie};
CUDACHECK(cudaGraphAddEmptyNode(&a->node, graph.graph, pair, 2)); CUDACHECK(cudaGraphAddEmptyNode(&a->node, graph.graph, pair, 2));
a->eventIsLagging = 1;
} }
// a->eventIsLagging doesn't change since we are just updating the
// dependencies of a->node.
} }
#else #else
CUDACHECK(cudaEventRecord(a->event, b)); CUDACHECK(cudaEventRecord(a->event, b));

4
src/proxy.cc
View File

@ -1055,8 +1055,8 @@ void* ncclProxyService(void* _args) {
int asyncOpCount = 0; int asyncOpCount = 0;
while ((stop == 0 || (stop == 1 && npeers > 0)) && *comm->abortFlag == 0) { while ((stop == 0 || (stop == 1 && npeers > 0)) && *comm->abortFlag == 0) {
/* never let proxy service thread blocks in poll, or it cannot receive abortFlag. */ /* never let proxy service thread blocks in poll, or it cannot receive abortFlag. */
if (int error = poll(pollfds, NCCL_MAX_LOCAL_RANKS+1, asyncOpCount ? 0 : 500) < 0) { if (poll(pollfds, NCCL_MAX_LOCAL_RANKS+1, asyncOpCount ? 0 : 500) < 0) {
WARN("[Proxy Service] Poll failed with error %d", error); WARN("[Proxy Service] Poll failed: %s\n", strerror(errno));
return NULL; return NULL;
} }
if (pollfds[NCCL_MAX_LOCAL_RANKS].revents) { if (pollfds[NCCL_MAX_LOCAL_RANKS].revents) {

16
src/transport.cc
View File

@ -42,7 +42,7 @@ static ncclResult_t selectTransport(struct ncclComm* comm, struct ncclTopoGraph*
ncclResult_t ncclTransportP2pConnect(struct ncclComm* comm, int channelId, int nrecv, int* peerRecv, int nsend, int* peerSend, int connIndex) { ncclResult_t ncclTransportP2pConnect(struct ncclComm* comm, int channelId, int nrecv, int* peerRecv, int nsend, int* peerSend, int connIndex) {
TRACE(NCCL_INIT, "nsend %d nrecv %d", nsend, nrecv); TRACE(NCCL_INIT, "nsend %d nrecv %d", nsend, nrecv);
struct ncclChannel* channel = &comm->channels[channelId]; struct ncclChannel* channel = &comm->channels[channelId];
uint32_t mask = 1 << channelId; uint64_t mask = 1UL << channel->id;
for (int i=0; i<nrecv; i++) { for (int i=0; i<nrecv; i++) {
int peer = peerRecv[i]; int peer = peerRecv[i];
if (peer == -1 || peer >= comm->nRanks || peer == comm->rank || channel->peers[peer].recv[connIndex].connected) continue; if (peer == -1 || peer >= comm->nRanks || peer == comm->rank || channel->peers[peer].recv[connIndex].connected) continue;
@ -77,15 +77,15 @@ ncclResult_t ncclTransportP2pSetup(struct ncclComm* comm, struct ncclTopoGraph*
int bootstrapTag = (i<<8) + (graph ? graph->id+1 : 0); int bootstrapTag = (i<<8) + (graph ? graph->id+1 : 0);
int recvPeer = (comm->rank - i + comm->nRanks) % comm->nRanks; int recvPeer = (comm->rank - i + comm->nRanks) % comm->nRanks;
int sendPeer = (comm->rank + i) % comm->nRanks; int sendPeer = (comm->rank + i) % comm->nRanks;
uint32_t recvMask = comm->connectRecv[recvPeer]; uint64_t recvMask = comm->connectRecv[recvPeer];
uint32_t sendMask = comm->connectSend[sendPeer]; uint64_t sendMask = comm->connectSend[sendPeer];
struct ncclConnect* recvData = data; struct ncclConnect* recvData = data;
int sendChannels = 0, recvChannels = 0; int sendChannels = 0, recvChannels = 0;
int type; int type;
TIME_START(0); TIME_START(0);
for (int c=0; c<MAXCHANNELS; c++) { for (int c=0; c<MAXCHANNELS; c++) {
if (recvMask & (1<<c)) { if (recvMask & (1UL<<c)) {
NCCLCHECK(selectTransport<0>(comm, graph, recvData+recvChannels++, c, recvPeer, connIndex, &type)); NCCLCHECK(selectTransport<0>(comm, graph, recvData+recvChannels++, c, recvPeer, connIndex, &type));
if (type > highestType) highestType = type; if (type > highestType) highestType = type;
} }
@ -94,7 +94,7 @@ ncclResult_t ncclTransportP2pSetup(struct ncclComm* comm, struct ncclTopoGraph*
TIME_START(1); TIME_START(1);
struct ncclConnect* sendData = recvData+recvChannels; struct ncclConnect* sendData = recvData+recvChannels;
for (int c=0; c<MAXCHANNELS; c++) { for (int c=0; c<MAXCHANNELS; c++) {
if (sendMask & (1<<c)) { if (sendMask & (1UL<<c)) {
NCCLCHECK(selectTransport<1>(comm, graph, sendData+sendChannels++, c, sendPeer, connIndex, &type)); NCCLCHECK(selectTransport<1>(comm, graph, sendData+sendChannels++, c, sendPeer, connIndex, &type));
if (type > highestType) highestType = type; if (type > highestType) highestType = type;
} }
@ -119,7 +119,7 @@ ncclResult_t ncclTransportP2pSetup(struct ncclComm* comm, struct ncclTopoGraph*
TIME_START(3); TIME_START(3);
for (int c=0; c<MAXCHANNELS; c++) { for (int c=0; c<MAXCHANNELS; c++) {
if (sendMask & (1<<c)) { if (sendMask & (1UL<<c)) {
struct ncclConnector* conn = comm->channels[c].peers[sendPeer].send + connIndex; struct ncclConnector* conn = comm->channels[c].peers[sendPeer].send + connIndex;
NCCLCHECK(conn->transportComm->connect(comm, sendData++, 1, comm->rank, conn)); NCCLCHECK(conn->transportComm->connect(comm, sendData++, 1, comm->rank, conn));
conn->connected = 1; conn->connected = 1;
@ -129,7 +129,7 @@ ncclResult_t ncclTransportP2pSetup(struct ncclComm* comm, struct ncclTopoGraph*
TIME_STOP(3); TIME_STOP(3);
TIME_START(4); TIME_START(4);
for (int c=0; c<MAXCHANNELS; c++) { for (int c=0; c<MAXCHANNELS; c++) {
if (recvMask & (1<<c)) { if (recvMask & (1UL<<c)) {
struct ncclConnector* conn = comm->channels[c].peers[recvPeer].recv + connIndex; struct ncclConnector* conn = comm->channels[c].peers[recvPeer].recv + connIndex;
NCCLCHECK(conn->transportComm->connect(comm, recvData++, 1, comm->rank, conn)); NCCLCHECK(conn->transportComm->connect(comm, recvData++, 1, comm->rank, conn));
conn->connected = 1; conn->connected = 1;
@ -137,7 +137,7 @@ ncclResult_t ncclTransportP2pSetup(struct ncclComm* comm, struct ncclTopoGraph*
} }
} }
TIME_STOP(4); TIME_STOP(4);
comm->connectRecv[recvPeer] = comm->connectSend[sendPeer] = 0; comm->connectRecv[recvPeer] = comm->connectSend[sendPeer] = 0UL;
} }
CUDACHECK(cudaStreamSynchronize(transportSetupStream)); CUDACHECK(cudaStreamSynchronize(transportSetupStream));
CUDACHECK(cudaStreamDestroy(transportSetupStream)); CUDACHECK(cudaStreamDestroy(transportSetupStream));

7
src/transport/coll_net.cc
View File

@ -121,6 +121,7 @@ struct recvResources {
int netDev; int netDev;
int useGdr; int useGdr;
int useDmaBuf; int useDmaBuf;
int needFlush;
uint64_t* gdcSync; uint64_t* gdcSync;
uint64_t* gdcFlush; uint64_t* gdcFlush;
void* gdrDesc; void* gdrDesc;
@ -139,6 +140,7 @@ static ncclResult_t canConnect(int* ret, struct ncclTopoSystem* topo, struct ncc
struct setupReq { struct setupReq {
int netDev; int netDev;
int useGdr; int useGdr;
int needFlush;
}; };
@ -151,6 +153,8 @@ static ncclResult_t sendSetup(struct ncclComm* comm, struct ncclTopoGraph* graph
NCCLCHECK(ncclTopoGetNetDev(comm, myInfo->rank, graph, channelId, -1, &req.netDev, &proxyRank)); NCCLCHECK(ncclTopoGetNetDev(comm, myInfo->rank, graph, channelId, -1, &req.netDev, &proxyRank));
NCCLCHECK(ncclTopoCheckGdr(comm->topo, myInfo->busId, req.netDev, 1, &req.useGdr)); NCCLCHECK(ncclTopoCheckGdr(comm->topo, myInfo->busId, req.netDev, 1, &req.useGdr));
send->conn.direct |= req.useGdr ? NCCL_DIRECT_NIC : 0; send->conn.direct |= req.useGdr ? NCCL_DIRECT_NIC : 0;
// Determine whether we need to flush the GDR buffer on recv or not
if (req.useGdr) NCCLCHECK(ncclTopoNeedFlush(comm->topo, myInfo->busId, &req.needFlush));
NCCLCHECK(ncclTopoGetLocalRank(comm->topo, myInfo->rank, &send->proxyConn.localRank)); NCCLCHECK(ncclTopoGetLocalRank(comm->topo, myInfo->rank, &send->proxyConn.localRank));
NCCLCHECK(ncclProxyConnect(comm, TRANSPORT_COLLNET, 1, myInfo->rank, &send->proxyConn)); NCCLCHECK(ncclProxyConnect(comm, TRANSPORT_COLLNET, 1, myInfo->rank, &send->proxyConn));
@ -392,6 +396,7 @@ static ncclResult_t recvProxySetup(struct ncclProxyConnection* connection, struc
resources->netDev = req->netDev; resources->netDev = req->netDev;
resources->useGdr = req->useGdr; resources->useGdr = req->useGdr;
resources->needFlush = req->needFlush;
ncclNetProperties_t props; ncclNetProperties_t props;
NCCLCHECK(collNetGetProperties(comm, req->netDev, &props)); NCCLCHECK(collNetGetProperties(comm, req->netDev, &props));
/* DMA-BUF support */ /* DMA-BUF support */
@ -754,7 +759,7 @@ static ncclResult_t recvProxyProgress(struct ncclComm* comm, struct ncclProxyArg
TRACE(NCCL_NET, "recvProxy [%d/%d/%d] received, size %d", sub->received, group, buffSlot, totalSize); TRACE(NCCL_NET, "recvProxy [%d/%d/%d] received, size %d", sub->received, group, buffSlot, totalSize);
sub->received += args->sliceSteps; sub->received += args->sliceSteps;
sub->requests[buffSlot] = NULL; sub->requests[buffSlot] = NULL;
if (1 && reqFifo[group][buffSlot].size > 0 && resources->useGdr) { if (reqFifo[group][buffSlot].size > 0 && resources->useGdr && resources->needFlush) {
// GDRCOPY support // GDRCOPY support
if (resources->gdcFlush) { if (resources->gdcFlush) {
#if defined (__x86_64__) #if defined (__x86_64__)

12
src/transport/net.cc
View File

@ -118,6 +118,7 @@ struct recvResources {
int netDev; int netDev;
int useGdr; int useGdr;
int useDmaBuf; int useDmaBuf;
int needFlush;
int maxRecvs; int maxRecvs;
uint64_t* gdcSync; uint64_t* gdcSync;
uint64_t* gdcFlush; uint64_t* gdcFlush;
@ -152,6 +153,7 @@ struct setupReq {
int shared; int shared;
int netDev; int netDev;
int useGdr; int useGdr;
int needFlush;
int channelId; int channelId;
int connIndex; int connIndex;
}; };
@ -205,6 +207,9 @@ static ncclResult_t recvSetup(struct ncclComm* comm, struct ncclTopoGraph* graph
NCCLCHECK(ncclTopoGetNetDev(comm, myInfo->rank, graph, channelId, myInfo->rank, &req.netDev, &proxyRank)); NCCLCHECK(ncclTopoGetNetDev(comm, myInfo->rank, graph, channelId, myInfo->rank, &req.netDev, &proxyRank));
NCCLCHECK(ncclTopoCheckGdr(comm->topo, myInfo->busId, req.netDev, 0, &req.useGdr)); NCCLCHECK(ncclTopoCheckGdr(comm->topo, myInfo->busId, req.netDev, 0, &req.useGdr));
// Determine whether we need to flush the GDR buffer on recv or not
if (req.useGdr) NCCLCHECK(ncclTopoNeedFlush(comm->topo, myInfo->busId, &req.needFlush));
// We don't support PXN on receive yet // We don't support PXN on receive yet
NCCLCHECK(ncclProxyConnect(comm, TRANSPORT_NET, 0, myInfo->rank, &recv->proxyConn)); NCCLCHECK(ncclProxyConnect(comm, TRANSPORT_NET, 0, myInfo->rank, &recv->proxyConn));
@ -470,6 +475,7 @@ static ncclResult_t recvProxySetup(struct ncclProxyConnection* connection, struc
resources->netDev = req->netDev; resources->netDev = req->netDev;
resources->shared = connection->shared = req->shared; resources->shared = connection->shared = req->shared;
resources->useGdr = req->useGdr; resources->useGdr = req->useGdr;
resources->needFlush = req->needFlush;
resources->channelId = req->channelId; resources->channelId = req->channelId;
resources->connIndex = req->connIndex; resources->connIndex = req->connIndex;
ncclNetProperties_t props; ncclNetProperties_t props;
@ -1033,7 +1039,7 @@ static ncclResult_t recvProxyProgress(struct ncclComm* comm, struct ncclProxyArg
for (int i=0; i<NCCL_PROXY_MAX_SUBS; i++) sizes[i] = 0; for (int i=0; i<NCCL_PROXY_MAX_SUBS; i++) sizes[i] = 0;
NCCLCHECK(ncclNetTest(comm, subGroup->requests[step%NCCL_STEPS], &done, sizes)); NCCLCHECK(ncclNetTest(comm, subGroup->requests[step%NCCL_STEPS], &done, sizes));
if (done) { if (done) {
int useGdr = 0; int needFlush = 0;
int totalSize = 0; int totalSize = 0;
for (int i=0; i<NCCL_PROXY_MAX_SUBS; i++) totalSize += sizes[i]; for (int i=0; i<NCCL_PROXY_MAX_SUBS; i++) totalSize += sizes[i];
for (int i=0; i<subGroup->groupSize; i++) { for (int i=0; i<subGroup->groupSize; i++) {
@ -1042,11 +1048,11 @@ static ncclResult_t recvProxyProgress(struct ncclComm* comm, struct ncclProxyArg
for (uint64_t step=sub->received-args->sliceSteps; step<sub->received; step++) ncclProfilingRecord(args, s+i, step, ncclProxyProfileRecvFlushWait); for (uint64_t step=sub->received-args->sliceSteps; step<sub->received; step++) ncclProfilingRecord(args, s+i, step, ncclProxyProfileRecvFlushWait);
if (step < sub->nsteps) { if (step < sub->nsteps) {
struct recvResources* resources = (struct recvResources*) (sub->connection->transportResources); struct recvResources* resources = (struct recvResources*) (sub->connection->transportResources);
if (resources->useGdr) useGdr = 1; if (resources->useGdr) needFlush |= resources->needFlush;
} }
} }
subGroup->requests[step%NCCL_STEPS] = NULL; subGroup->requests[step%NCCL_STEPS] = NULL;
if (totalSize > 0 && p == NCCL_PROTO_SIMPLE && useGdr) { if (totalSize > 0 && p == NCCL_PROTO_SIMPLE && needFlush) {
// GDRCOPY support // GDRCOPY support
struct recvResources* resources = (struct recvResources*) (subGroup->connection->transportResources); struct recvResources* resources = (struct recvResources*) (subGroup->connection->transportResources);
if (resources->gdcFlush) { if (resources->gdcFlush) {