microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-12 01:10:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

filename changes

Browse Source
This commit is contained in:
Saeed Maleki
2023-04-06 23:55:11 +00:00
parent 8c6e361044
commit 1b2db68e93
5 changed files with 575 additions and 870 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Makefile
View File

@@ -129,13 +129,13 @@ LIBSONAME := $(LIBNAME).$(MSCCLPP_MAJOR)
LIBTARGET := $(BUILDDIR)/$(LIBDIR)/$(LIBNAME).$(MSCCLPP_MAJOR).$(MSCCLPP_MINOR).$(MSCCLPP_PATCH)
TESTSDIR := tests
TESTSSRCS := $(addprefix $(TESTSDIR)/,bootstrap_test.cc allgather_test.cu)
TESTSSRCS := $(addprefix $(TESTSDIR)/,bootstrap_test.cc allgather_test_standalone.cu)
TESTSOBJS := $(patsubst %.cc,%.o,$(TESTSSRCS)) $(patsubst %.cu,%.o,$(TESTSSRCS))
TESTSOBJTARGETS := $(TESTSOBJS:%=$(BUILDDIR)/$(OBJDIR)/%)
TESTSBINS := $(patsubst %.o,$(BUILDDIR)/$(BINDIR)/%,$(TESTSOBJS))
MSCLLPPTESTSOBJSDIR:= $(BUILDDIR)/$(OBJDIR)/$(TESTSDIR)
MSCLLPPTESTBINFILESLIST := allgather_test3
MSCLLPPTESTBINFILESLIST := allgather_test
MSCLLPPTESTBINS := $(MSCLLPPTESTBINFILESLIST:%=$(BUILDDIR)/$(BINDIR)/$(TESTSDIR)/%_perf)
INCLUDE := -Isrc -Isrc/include

420
tests/allgather_test.cu
View File

@@ -1,50 +1,16 @@
#include "mscclpp.h"
/*************************************************************************
* Copyright (c) 2016-2022, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
#include "mpi.h"
#endif // MSCCLPP_USE_MPI_FOR_TESTS
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include "comm.h"
#include "common.h"
#include <cuda_runtime.h>
#include <string>
#include <unistd.h>
#include <unordered_map>
static int nranksPerNode = 8;
// Propagate errors up
#define MSCCLPPCHECK(call) \
do { \
mscclppResult_t res = call; \
if (res != mscclppSuccess && res != mscclppInProgress) { \
/* Print the back trace*/ \
printf("Failure at %s:%d -> %s\n", __FILE__, __LINE__, mscclppGetErrorString(res)); \
return res; \
} \
} while (0)
// Check CUDA RT calls
#define CUDACHECK(cmd) \
do { \
cudaError_t err = cmd; \
if (err != cudaSuccess) { \
printf("%s:%d Cuda failure '%s'\n", __FILE__, __LINE__, cudaGetErrorString(err)); \
exit(EXIT_FAILURE); \
} \
} while (false)
// Measure current time in second.
static double getTime(void)
{
struct timespec tspec;
if (clock_gettime(CLOCK_MONOTONIC, &tspec) == -1) {
printf("clock_gettime failed\n");
exit(EXIT_FAILURE);
}
return (tspec.tv_nsec / 1.0e9) + tspec.tv_sec;
}
#define ALIGN 4
__constant__ mscclppDevConn_t constDevConns[16];
__device__ void allgather0(mscclppDevConn_t devConn, int rank, int world_size, int remoteRank, size_t nelemsPerGPU)
@@ -53,16 +19,16 @@ __device__ void allgather0(mscclppDevConn_t devConn, int rank, int world_size, i
// this thread's role is a sender role
// put your data asynchronously
if ((threadIdx.x % 32) == 0)
if (threadIdx.x % 32 == 0)
devConn.putWithSignal(rank * nelemsPerGPU * sizeof(int), nelemsPerGPU * sizeof(int));
// make sure everyone is put their data before some thread randomly blocks everyone else in signal
__syncthreads();
// push with flag and sync to make sure the data is received
if ((threadIdx.x % 32) == 0)
if (threadIdx.x % 32 == 0)
devConn.flush();
// this thread's role is a receiver role. wait on the semaphore to make sure the data is ready
if ((threadIdx.x % 32) == 0)
if (threadIdx.x % 32 == 0)
devConn.wait();
}
@@ -176,326 +142,84 @@ __global__ void kernel(int rank, int world_size, int nranksPerNode, size_t nelem
allgather2(devConn, rank, world_size, nranksPerNode, remoteRank, nelemsPerGPU);
}
int rankToLocalRank(int rank)
void AllGatherGetCollByteCount(size_t* sendcount, size_t* recvcount, size_t* paramcount, size_t* sendInplaceOffset,
size_t* recvInplaceOffset, size_t count, int nranks)
{
return rank % nranksPerNode;
size_t base = (count / (ALIGN * nranks)) * ALIGN;
*sendcount = base;
*recvcount = base * nranks;
*sendInplaceOffset = base;
*recvInplaceOffset = 0;
*paramcount = base;
}
int rankToNode(int rank)
testResult_t AllGatherInitData(struct threadArgs* args, int in_place)
{
return rank / nranksPerNode;
}
size_t sendcount = args->sendBytes / sizeof(int);
size_t recvcount = args->expectedBytes / sizeof(int);
// int nranks = args->totalProcs;
void print_usage(const char* prog)
{
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
printf("usage: %s IP:PORT [rank nranks]\n", prog);
#else
printf("usage: %s IP:PORT rank nranks\n", prog);
#endif
}
CUDACHECK(cudaSetDevice(args->gpus[0]));
int rank = args->proc;
CUDACHECK(cudaMemset(args->recvbuffs[0], 0, args->expectedBytes));
// void* data = in_place ? ((char*)args->recvbuffs[0]) + rank * args->sendBytes : args->sendbuffs[0];
void initializeAndAllocateAllGatherData(int rank, int world_size, size_t dataSize, size_t nelemsPerGPU, int** data_h,
int** data_d)
{
CUDACHECK(cudaMalloc(data_d, dataSize));
CUDACHECK(cudaMemset(*data_d, 0, dataSize));
*data_h = new int[nelemsPerGPU * world_size];
for (size_t i = 0; i < nelemsPerGPU * world_size; i++) {
int* dataHost = new int[recvcount];
for (size_t i = 0; i < recvcount; i++) {
int val = i + 1;
if (i / nelemsPerGPU == (size_t)rank) {
(*data_h)[i] = val;
if (i / sendcount == (size_t)rank) {
dataHost[i] = val;
} else {
(*data_h)[i] = 0;
dataHost[i] = 0;
}
}
CUDACHECK(cudaMemcpy(*data_d, *data_h, dataSize, cudaMemcpyHostToDevice));
CUDACHECK(cudaMemcpy(args->recvbuffs[0], dataHost, recvcount, cudaMemcpyHostToDevice));
for (int i = 0; i < static_cast<int>(recvcount); i++) {
dataHost[i] = i + 1;
}
CUDACHECK(cudaMemcpy(args->expected[0], dataHost, recvcount, cudaMemcpyHostToDevice));
delete dataHost;
CUDACHECK(cudaDeviceSynchronize());
return testSuccess;
}
mscclppResult_t setupMscclppConnections(int rank, int world_size, mscclppComm_t comm, int* data_d, size_t dataSize)
void AllGatherGetBw(size_t count, int typesize, double sec, double* algBw, double* busBw, int nranks)
{
int thisNode = rankToNode(rank);
int cudaNum = rankToLocalRank(rank);
std::string ibDevStr = "mlx5_ib" + std::to_string(cudaNum);
double baseBw = (double)(count * typesize * nranks) / 1.0E9 / sec;
for (int r = 0; r < world_size; ++r) {
if (r == rank)
continue;
mscclppTransport_t transportType;
const char* ibDev = ibDevStr.c_str();
if (rankToNode(r) == thisNode) {
ibDev = NULL;
transportType = mscclppTransportP2P;
} else {
transportType = mscclppTransportIB;
}
// Connect with all other ranks
MSCCLPPCHECK(mscclppConnect(comm, r, 0, data_d, dataSize, transportType, ibDev));
}
*algBw = baseBw;
double factor = ((double)(nranks - 1)) / ((double)nranks);
*busBw = baseBw * factor;
}
MSCCLPPCHECK(mscclppConnectionSetup(comm));
testResult_t AllGatherRunColl(void* sendbuff, void* recvbuff, int nranksPerNode, size_t count, mscclppComm_t comm,
cudaStream_t stream)
{
int worldSize = comm->nRanks;
kernel<<<1, 32 * (worldSize - 1), 0, stream>>>(comm->rank, worldSize, nranksPerNode, count / sizeof(int), 1);
return testSuccess;
}
struct testColl allGatherTest = {"AllGather", AllGatherGetCollByteCount, AllGatherInitData, AllGatherGetBw,
AllGatherRunColl};
void AllGatherGetBuffSize(size_t* sendcount, size_t* recvcount, size_t count, int nranks)
{
size_t paramcount, sendInplaceOffset, recvInplaceOffset;
AllGatherGetCollByteCount(sendcount, recvcount, &paramcount, &sendInplaceOffset, &recvInplaceOffset, count, nranks);
}
testResult_t AllGatherRunTest(struct threadArgs* args)
{
args->collTest = &allGatherTest;
mscclppDevConn_t* devConns;
int nCons;
MSCCLPPCHECK(mscclppGetAllDeviceConnections(comm, &devConns, &nCons));
MSCCLPPCHECK(mscclppGetAllDeviceConnections(args->comm, &devConns, &nCons));
CUDACHECK(cudaMemcpyToSymbol(constDevConns, devConns, sizeof(mscclppDevConn_t) * nCons));
return mscclppSuccess;
TESTCHECK(TimeTest(args));
return testSuccess;
}
void printUsage(const char* prog, bool isMpi)
{
if (isMpi) {
std::string st = "you are using MPI for this test\n";
st += "two possilbe usages are:\n";
st += "> " + std::string(prog) + "\n";
st += "or\n";
st += "> " + std::string(prog) + " -ip_port [ip:port]\n";
printf("%s", st.c_str());
} else {
std::string st = "you are NOT using MPI for this test\n";
st += "the only possible usage:\n";
st += "> " + std::string(prog) + " -ip_port [ip:port] -rank [rank] -nranks [nranks]\n";
printf("%s", st.c_str());
}
}
struct testEngine allGatherEngine = {AllGatherGetBuffSize, AllGatherRunTest};
std::unordered_map<std::string, std::string> parseArgs(int argc, const char* argv[], bool isMpi)
{
std::unordered_map<std::string, std::string> options;
for (int i = 1; i < argc; i++) {
std::string arg = argv[i];
if (arg == "-rankspernode") {
if (isMpi) {
fprintf(stderr, "Error: -rankspernode should not be specified with MPI.\n");
exit(-1);
}
if (i + 1 < argc) {
options["rankspernode"] = argv[++i];
} else {
fprintf(stderr, "Error: -rankspernode option requires an argument.\n");
;
exit(-1);
}
} else if (arg == "-kernel") {
if (i + 1 < argc) {
options["kernel"] = argv[++i];
} else {
fprintf(stderr, "Error: -kernel option requires an argument.\n");
exit(-1);
}
} else if (arg == "-ip_port") {
if (i + 1 < argc) {
options["ip_port"] = argv[++i];
} else {
fprintf(stderr, "Error: -ip_port option requires an argument.\n");
exit(-1);
}
} else if (arg == "-rank") {
if (isMpi) {
fprintf(stderr, "Error: -rank should not be specified with MPI.\n");
exit(-1);
}
if (i + 1 < argc) {
options["rank"] = argv[++i];
} else {
fprintf(stderr, "Error: -ip_port option requires an argument.\n");
exit(-1);
}
} else if (arg == "-nranks") {
if (isMpi) {
fprintf(stderr, "Error: -nranks should not be specified with MPI.\n");
exit(-1);
}
if (i + 1 < argc) {
options["nranks"] = argv[++i];
} else {
fprintf(stderr, "Error: -ip_port option requires an argument.\n");
exit(-1);
}
} else if (arg == "-datasize") {
if (i + 1 < argc) {
options["datasize"] = argv[++i];
} else {
fprintf(stderr, "Error: -datasize option requires an argument.\n");
exit(-1);
}
} else if (arg == "-help" || arg == "-h") {
printUsage(argv[0], isMpi);
exit(0);
} else {
fprintf(stderr, "Error: Unknown option %s\n", argv[i]);
exit(-1);
}
}
return options;
}
int main(int argc, const char* argv[])
{
bool isMpi = false;
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
isMpi = true;
#endif
auto parsedArgs = parseArgs(argc, argv, isMpi);
int rank;
int world_size;
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
MPI_Init(NULL, NULL);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &world_size);
// get the local number of nodes with MPI
MPI_Comm shmcomm;
MPI_Comm_split_type(MPI_COMM_WORLD, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL, &shmcomm);
int shmrank;
MPI_Comm_size(shmcomm, &shmrank);
nranksPerNode = shmrank;
MPI_Comm_free(&shmcomm);
#else
if (parsedArgs.find("rank") == parsedArgs.end() || parsedArgs.find("nranks") == parsedArgs.end()) {
printUsage(argv[0], isMpi);
exit(-1);
}
rank = std::stoi(parsedArgs["rank"]);
world_size = std::stoi(parsedArgs["nranks"]);
if (parsedArgs.find("rankspernode") == parsedArgs.end()) {
printUsage(argv[0], isMpi);
exit(-1);
}
nranksPerNode = std::stoi(parsedArgs["rankspernode"]);
#endif
int kernelNum = 0;
if (parsedArgs.find("kernel") != parsedArgs.end()) {
kernelNum = std::stoi(parsedArgs["kernel"]);
}
char* ip_port = NULL;
if (parsedArgs.find("ip_port") == parsedArgs.end()) {
printUsage(argv[0], isMpi);
exit(-1);
}
ip_port = (char*)parsedArgs["ip_port"].c_str();
int thisNode = rankToNode(rank);
int cudaNum = rankToLocalRank(rank);
CUDACHECK(cudaSetDevice(cudaNum));
if (rank == 0)
printf("Initializing MSCCL++\n");
mscclppComm_t comm;
MSCCLPPCHECK(mscclppCommInitRank(&comm, world_size, ip_port, rank));
int* data_d;
int* data_h;
size_t dataSize = 1024 * 1024 * 1024;
if (parsedArgs.find("datasize") != parsedArgs.end()) {
dataSize = std::stoul(parsedArgs["datasize"]);
}
size_t nelemsPerGPU = dataSize / sizeof(int) / world_size;
if (rank == 0)
printf("Initializing data for allgather test\n");
initializeAndAllocateAllGatherData(rank, world_size, dataSize, nelemsPerGPU, &data_h, &data_d);
if (rank == 0)
printf("Setting up the connection in MSCCL++\n");
MSCCLPPCHECK(setupMscclppConnections(rank, world_size, comm, data_d, dataSize));
if (rank == 0)
printf("Launching MSCCL++ proxy threads\n");
MSCCLPPCHECK(mscclppProxyLaunch(comm));
if (rank == 0)
printf("Testing the correctness of AllGather implementation\n");
cudaStream_t stream;
CUDACHECK(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
CUDACHECK(cudaDeviceSynchronize());
kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nranksPerNode, nelemsPerGPU, kernelNum);
CUDACHECK(cudaDeviceSynchronize());
CUDACHECK(cudaMemcpy(data_h, data_d, dataSize, cudaMemcpyDeviceToHost));
for (size_t i = 0; i < nelemsPerGPU * world_size; i++) {
int val = i + 1;
if (data_h[i] != val) {
printf("oh uh! data_h[%ld] (%d) != val (%d)\n", i, data_h[i], val);
break;
}
}
int tmp[16];
// A simple barrier
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
if (rank == 0)
printf("Successfully checked the correctness\n");
// Perf test
int iterwithoutcudagraph = 10;
if (rank == 0)
printf("Running %d iterations of the kernel without CUDA graph\n", iterwithoutcudagraph);
CUDACHECK(cudaStreamSynchronize(stream));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
for (int i = 0; i < iterwithoutcudagraph; ++i) {
kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nranksPerNode, nelemsPerGPU, kernelNum);
}
CUDACHECK(cudaStreamSynchronize(stream));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
// cudaGraph Capture
int cudagraphiter = 10;
if (rank == 0)
printf("Capturing %d iterations of the kernel in a CUDA graph\n", cudagraphiter);
cudaGraph_t graph;
cudaGraphExec_t instance;
cudaStreamBeginCapture(stream, cudaStreamCaptureModeGlobal);
for (int i = 0; i < cudagraphiter; ++i) {
kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nranksPerNode, nelemsPerGPU, kernelNum);
}
cudaStreamEndCapture(stream, &graph);
cudaGraphInstantiate(&instance, graph, NULL, NULL, 0);
int cudagraphwarmup = 10;
if (rank == 0)
printf("Warming up %d iterations of the CUDA graph with %d iterations of the kernel\n", cudagraphwarmup,
cudagraphiter);
for (int i = 0; i < cudagraphwarmup; ++i) {
cudaGraphLaunch(instance, stream);
}
CUDACHECK(cudaStreamSynchronize(stream));
// measure runtime
int cudagraphlaunch = 10;
if (rank == 0)
printf("Running %d iterations of the CUDA graph with %d iterations of the kernel\n", cudagraphlaunch,
cudagraphiter);
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
double t0, t1, ms, time_in_us;
t0 = getTime();
for (int i = 0; i < cudagraphlaunch; ++i) {
cudaGraphLaunch(instance, stream);
}
CUDACHECK(cudaStreamSynchronize(stream));
t1 = getTime();
ms = (t1 - t0) * 1000.0;
time_in_us = ms * 1000. / (float)cudagraphlaunch / (float)cudagraphiter;
printf("Rank %d report: size %lu time: %f us/iter algBW %f GBps\n", rank, dataSize, time_in_us,
(double)(dataSize) / 1e9 / (time_in_us / 1e6));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
if (rank == 0)
printf("Stopping MSCCL++ proxy threads\n");
MSCCLPPCHECK(mscclppProxyStop(comm));
if (rank == 0)
printf("Destroying MSCCL++ communicator\n");
MSCCLPPCHECK(mscclppCommDestroy(comm));
printf("Rank %d succeeded!\n", rank);
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
MPI_Finalize();
#endif
return 0;
}
#pragma weak mscclppTestEngine = allGatherEngine

295
tests/allgather_test2.cu
View File

@@ -1,295 +0,0 @@
#include "mscclpp.h"
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
#include "mpi.h"
#endif // MSCCLPP_USE_MPI_FOR_TESTS
#include <stdio.h>
#include <stdlib.h>
#include <string>
#include <unistd.h>
#define RANKS_PER_NODE 8
// Check CUDA RT calls
#define CUDACHECK(cmd) \
do { \
cudaError_t err = cmd; \
if (err != cudaSuccess) { \
printf("%s:%d Cuda failure '%s'\n", __FILE__, __LINE__, cudaGetErrorString(err)); \
exit(EXIT_FAILURE); \
} \
} while (false)
// Measure current time in second.
static double getTime(void)
{
struct timespec tspec;
if (clock_gettime(CLOCK_MONOTONIC, &tspec) == -1) {
printf("clock_gettime failed\n");
exit(EXIT_FAILURE);
}
return (tspec.tv_nsec / 1.0e9) + tspec.tv_sec;
}
__constant__ mscclppDevConn_t constDevConns[16];
__global__ void kernel(int rank, int world_size, size_t nelemsPerGPU)
{
if (threadIdx.x % 32 != 0)
return;
int warpId = threadIdx.x / 32;
bool isIB = false;
if (warpId >= world_size - 1)
isIB = true;
if (isIB)
warpId = warpId - (world_size - 1);
int remoteRank = (warpId < rank) ? warpId : warpId + 1;
mscclppDevConn_t devConn = constDevConns[remoteRank];
if (isIB)
devConn = constDevConns[remoteRank + world_size];
// Each warp receives data from different ranks
#if 1
// Trigger sending data, flag and synchronize after
devConn.putWithSignal(rank * nelemsPerGPU * sizeof(int), nelemsPerGPU * sizeof(int));
devConn.wait();
#else
for (int i = 1; i < world_size; i++) {
__syncthreads();
if (remoteRank != ((rank + i) % world_size))
continue;
// Trigger sending data, flag and synchronize after
size_t ibPortion = nelemsPerGPU / 12; // nelemsPerGPU/12;
if (isIB)
devConn.fifo.setTrigger(trig, mscclppFlag | mscclppData | mscclppSync,
rank * nelemsPerGPU * sizeof(int) + (nelemsPerGPU - ibPortion) * sizeof(int),
rank * nelemsPerGPU * sizeof(int) + (nelemsPerGPU - ibPortion) * sizeof(int),
ibPortion * sizeof(int));
else
devConn.fifo.setTrigger(trig, mscclppFlag | mscclppData | mscclppSync, rank * nelemsPerGPU * sizeof(int),
rank * nelemsPerGPU * sizeof(int), (nelemsPerGPU - ibPortion) * sizeof(int));
// Wait on the request to make sure it is safe to reuse buffer and flag
auto req = devConn.fifo.putWithSignal(dataOffset, dataSize);
devConn.fifo.sync(req);
}
// Wait for receiving data from remote rank
while (*proxyFlag == baseFlag)
;
#endif
}
int rankToLocalRank(int rank)
{
return rank % RANKS_PER_NODE;
}
int rankToNode(int rank)
{
return rank / RANKS_PER_NODE;
}
int cudaNumToIbNum(int cudaNum)
{
int ibNum;
if (cudaNum == 0) {
ibNum = 0;
} else if (cudaNum == 1) {
ibNum = 4;
} else if (cudaNum == 2) {
ibNum = 1;
} else if (cudaNum == 3) {
ibNum = 5;
} else if (cudaNum == 4) {
ibNum = 2;
} else if (cudaNum == 5) {
ibNum = 6;
} else if (cudaNum == 6) {
ibNum = 3;
} else if (cudaNum == 7) {
ibNum = 7;
} else {
printf("Invalid cudaNum: %d\n", cudaNum);
exit(EXIT_FAILURE);
}
return ibNum;
}
void print_usage(const char* prog)
{
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
printf("usage: %s IP:PORT [rank nranks]\n", prog);
#else
printf("usage: %s IP:PORT rank nranks\n", prog);
#endif
}
int main(int argc, const char* argv[])
{
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
if (argc != 2 && argc != 4) {
print_usage(argv[0]);
return -1;
}
const char* ip_port = argv[1];
int rank;
int world_size;
if (argc == 4) {
rank = atoi(argv[2]);
world_size = atoi(argv[3]);
} else {
MPI_Init(NULL, NULL);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &world_size);
}
#else
if (argc != 4) {
print_usage(argv[0]);
return -1;
}
const char* ip_port = argv[1];
int rank = atoi(argv[2]);
int world_size = atoi(argv[3]);
#endif
int localRank = rankToLocalRank(rank);
int thisNode = rankToNode(rank);
int cudaNum = localRank;
int ibNum = cudaNumToIbNum(cudaNum);
CUDACHECK(cudaSetDevice(cudaNum));
std::string ibDevStr = "mlx5_ib" + std::to_string(localRank);
mscclppComm_t comm;
MSCCLPPCHECK(mscclppCommInitRank(&comm, world_size, rank, ip_port));
int* data_d;
uint64_t* flag_d;
size_t data_size = 1536 * 1024 * 1024;
size_t nelemsPerGPU = data_size / sizeof(int) / world_size;
CUDACHECK(cudaMalloc(&data_d, data_size));
CUDACHECK(cudaMalloc(&flag_d, sizeof(uint64_t)));
CUDACHECK(cudaMemset(data_d, 0, data_size));
CUDACHECK(cudaMemset(flag_d, 0, sizeof(uint64_t)));
int* data_h = new int[nelemsPerGPU * world_size];
for (int i = 0; i < nelemsPerGPU * world_size; i++) {
size_t val = i + 1;
if (i / nelemsPerGPU == rank) {
data_h[i] = val;
} else {
data_h[i] = 0;
}
}
CUDACHECK(cudaMemcpy(data_d, data_h, data_size, cudaMemcpyHostToDevice));
mscclppDevConn_t devConns[16];
for (int r = 0; r < world_size; ++r) {
if (r == rank)
continue;
mscclppTransport_t transportType;
const char* ibDev = NULL;
transportType = mscclppTransportP2P;
// Connect with all other ranks
MSCCLPPCHECK(mscclppConnect(comm, &devConns[r], r, 0, data_d, data_size, flag_d, transportType, ibDev));
}
for (int r = 0; r < world_size; ++r) {
if (r == rank)
continue;
mscclppTransport_t transportType;
const char* ibDev = ibDevStr.c_str();
transportType = mscclppTransportIB;
// Connect with all other ranks
MSCCLPPCHECK(
mscclppConnect(comm, &devConns[r + world_size], r, 0, data_d, data_size, flag_d, transportType, ibDev));
}
MSCCLPPCHECK(mscclppConnectionSetup(comm));
MSCCLPPCHECK(mscclppProxyLaunch(comm));
CUDACHECK(cudaMemcpyToSymbol(constDevConns, devConns, sizeof(mscclppDevConn_t) * 2 * world_size));
cudaStream_t stream;
CUDACHECK(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
CUDACHECK(cudaDeviceSynchronize());
kernel<<<1, 32 * 2 * (world_size - 1), 0, stream>>>(rank, world_size, nelemsPerGPU);
CUDACHECK(cudaDeviceSynchronize());
CUDACHECK(cudaMemcpy(data_h, data_d, data_size, cudaMemcpyDeviceToHost));
CUDACHECK(cudaDeviceSynchronize());
for (size_t i = 0; i < nelemsPerGPU * world_size; i++) {
int val = i + 1;
if (data_h[i] != val) {
printf("oh uh things went wrong! data_h[%d] (%d) != val (%d)\n", i, data_h[i], val);
break;
}
}
int tmp[16];
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
// // Perf test
// cudaEvent_t ev_start;
// cudaEvent_t ev_end;
// CUDACHECK(cudaEventCreate(&ev_start));
// CUDACHECK(cudaEventCreate(&ev_end));
// warm up
// int warmupiter = 1000;
// for (int i = 0; i < warmupiter; ++i) {
// kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nelemsPerGPU);
// }
// CUDACHECK(cudaDeviceSynchronize());
// MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
// cudaGraph Capture
cudaGraph_t graph;
cudaGraphExec_t instance;
cudaStreamBeginCapture(stream, cudaStreamCaptureModeGlobal);
int cudagraphiter = 10;
for (int i = 0; i < cudagraphiter; ++i) {
kernel<<<1, 32 * 2 * (world_size - 1), 0, stream>>>(rank, world_size, nelemsPerGPU);
}
cudaStreamEndCapture(stream, &graph);
cudaGraphInstantiate(&instance, graph, NULL, NULL, 0);
int cudagraphwarmup = 10;
for (int i = 0; i < cudagraphwarmup; ++i) {
cudaGraphLaunch(instance, stream);
}
CUDACHECK(cudaStreamSynchronize(stream));
// measure runtime
// CUDACHECK(cudaEventRecord(ev_start, stream));
double t0 = getTime();
int cudagraphlaunch = 10;
for (int i = 0; i < cudagraphlaunch; ++i) {
// kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size);
cudaGraphLaunch(instance, stream);
}
// CUDACHECK(cudaEventRecord(ev_end, stream));
CUDACHECK(cudaStreamSynchronize(stream));
double t1 = getTime();
float ms = (t1 - t0) * 1000.0;
// CUDACHECK(cudaEventElapsedTime(&ms, ev_start, ev_end));
double time_in_us = ms * 1000. / (float)cudagraphlaunch / (float)cudagraphiter;
printf("rank: %d, time: %f us/iter algBW %f\n", rank, time_in_us,
(double)(data_size) / 1024. / 1024. / 1024. / (time_in_us / 1e6));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
MSCCLPPCHECK(mscclppProxyStop(comm));
MSCCLPPCHECK(mscclppCommDestroy(comm));
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
if (argc == 2) {
MPI_Finalize();
}
#endif
printf("Succeeded! %d\n", rank);
return 0;
}

225
tests/allgather_test3.cu
View File

@@ -1,225 +0,0 @@
/*************************************************************************
* Copyright (c) 2016-2022, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
#include "comm.h"
#include "common.h"
#include <cuda_runtime.h>
#include <string>
#define ALIGN 4
__constant__ mscclppDevConn_t constDevConns[16];
__device__ void allgather0(mscclppDevConn_t devConn, int rank, int world_size, int remoteRank, size_t nelemsPerGPU)
{
// this allgather is really simple and implemented as an alltoall
// this thread's role is a sender role
// put your data asynchronously
if (threadIdx.x % 32 == 0)
devConn.putWithSignal(rank * nelemsPerGPU * sizeof(int), nelemsPerGPU * sizeof(int));
// make sure everyone is put their data before some thread randomly blocks everyone else in signal
__syncthreads();
// push with flag and sync to make sure the data is received
if (threadIdx.x % 32 == 0)
devConn.flush();
// this thread's role is a receiver role. wait on the semaphore to make sure the data is ready
if (threadIdx.x % 32 == 0)
devConn.wait();
}
__device__ void localAllGather(mscclppDevConn_t devConn, int rank, int world_size, int nranksPerNode, int remoteRank,
uint64_t offset, uint64_t size)
{
// this allgather algorithm works as follows:
// Step 1: GPU rank i sends data to GPU rank (i+1) % nranksPerNode
// and waits for data from GPU rank (i-1) % nranksPerNode
// Step 2: GPU rank i sends data to GPU rank (i+2) % nranksPerNode
// ...
// This order is much better for DMA engine for NVLinks
for (int i = 1; i < nranksPerNode; i++) {
if ((remoteRank % nranksPerNode) == ((rank + i) % nranksPerNode)) {
// put your data to GPU (rank+i) % nranksPerNode and signal in one call
if ((threadIdx.x % 32) == 0)
devConn.putWithSignalAndFlush(offset, size);
}
// wait for the data from GPU (rank-i) % nranksPerNode to arrive
if ((remoteRank % nranksPerNode) == ((rank - i + nranksPerNode) % nranksPerNode)) {
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
asm volatile("bar.sync %0, %1;" ::"r"(11), "r"((nranksPerNode - 1) * 32) : "memory");
}
}
__device__ void allgather1(mscclppDevConn_t devConn, int rank, int world_size, int nranksPerNode, int remoteRank,
size_t nelemsPerGPU)
{
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank, rank * nelemsPerGPU * sizeof(int),
nelemsPerGPU * sizeof(int));
}
__device__ void allgather2(mscclppDevConn_t devConn, int rank, int world_size, int nranksPerNode, int remoteRank,
size_t nelemsPerGPU)
{
// this allgather is a pipelined and hierarchical one and only works for two nodes
// it is implemented as follows:
// Step 1: each node does a local allgather and concurrently,
// local GPU i exchange (piplineSize-1)/pipelineSize portion of their data with
// its cross-node neighbor (local GPU i on the other node) via IB
// Step 2: each node does a local allgather again with the data just received from its
// cross-node neighbor in step 1, and concurrently, exchange the rest of the data with
// its cross-node neighbor
// Step 3: each node does a local allgather for the last time with the rest of the data
int pipelineSize = 3;
// Step 1
// local allgather
if (remoteRank / nranksPerNode == rank / nranksPerNode) {
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank, rank * nelemsPerGPU * sizeof(int),
nelemsPerGPU * sizeof(int));
}
// cross-node exchange
if (remoteRank % nranksPerNode == rank % nranksPerNode) {
// opposite side
if ((threadIdx.x % 32) == 0)
devConn.putWithSignalAndFlush(rank * nelemsPerGPU * sizeof(int),
(nelemsPerGPU * (pipelineSize - 1)) / pipelineSize * sizeof(int));
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
__syncthreads();
// Step 2
// local allgather
int otherNghr = (rank + nranksPerNode) % world_size;
if (remoteRank / nranksPerNode == rank / nranksPerNode) {
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank, otherNghr * nelemsPerGPU * sizeof(int),
(nelemsPerGPU * (pipelineSize - 1)) / pipelineSize * sizeof(int));
}
// cross-node exchange
if (remoteRank % nranksPerNode == rank % nranksPerNode) {
// opposite side
if ((threadIdx.x % 32) == 0)
devConn.putWithSignalAndFlush((rank * nelemsPerGPU + (nelemsPerGPU * (pipelineSize - 1)) / pipelineSize) *
sizeof(int),
nelemsPerGPU / pipelineSize * sizeof(int));
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
__syncthreads();
// Step 3
// local allgather
if (remoteRank / nranksPerNode == rank / nranksPerNode) {
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank,
(otherNghr * nelemsPerGPU + (nelemsPerGPU * (pipelineSize - 1)) / pipelineSize) * sizeof(int),
nelemsPerGPU / pipelineSize * sizeof(int));
}
}
__global__ void kernel(int rank, int world_size, int nranksPerNode, size_t nelemsPerGPU, int kernel)
{
// find the mapping between remoteRank and devConns
int warpId = threadIdx.x / 32;
int remoteRank = (warpId < rank) ? warpId : warpId + 1;
// Each warp is responsible for one of the remote ranks
mscclppDevConn_t devConn = constDevConns[warpId];
if (kernel == 0)
allgather0(devConn, rank, world_size, remoteRank, nelemsPerGPU);
else if (kernel == 1)
allgather1(devConn, rank, world_size, nranksPerNode, remoteRank, nelemsPerGPU);
else if (kernel == 2)
allgather2(devConn, rank, world_size, nranksPerNode, remoteRank, nelemsPerGPU);
}
void AllGatherGetCollByteCount(size_t* sendcount, size_t* recvcount, size_t* paramcount, size_t* sendInplaceOffset,
size_t* recvInplaceOffset, size_t count, int nranks)
{
size_t base = (count / (ALIGN * nranks)) * ALIGN;
*sendcount = base;
*recvcount = base * nranks;
*sendInplaceOffset = base;
*recvInplaceOffset = 0;
*paramcount = base;
}
testResult_t AllGatherInitData(struct threadArgs* args, int in_place)
{
size_t sendcount = args->sendBytes / sizeof(int);
size_t recvcount = args->expectedBytes / sizeof(int);
// int nranks = args->totalProcs;
CUDACHECK(cudaSetDevice(args->gpus[0]));
int rank = args->proc;
CUDACHECK(cudaMemset(args->recvbuffs[0], 0, args->expectedBytes));
// void* data = in_place ? ((char*)args->recvbuffs[0]) + rank * args->sendBytes : args->sendbuffs[0];
int* dataHost = new int[recvcount];
for (size_t i = 0; i < recvcount; i++) {
int val = i + 1;
if (i / sendcount == (size_t)rank) {
dataHost[i] = val;
} else {
dataHost[i] = 0;
}
}
CUDACHECK(cudaMemcpy(args->recvbuffs[0], dataHost, recvcount, cudaMemcpyHostToDevice));
for (int i = 0; i < static_cast<int>(recvcount); i++) {
dataHost[i] = i + 1;
}
CUDACHECK(cudaMemcpy(args->expected[0], dataHost, recvcount, cudaMemcpyHostToDevice));
delete dataHost;
CUDACHECK(cudaDeviceSynchronize());
return testSuccess;
}
void AllGatherGetBw(size_t count, int typesize, double sec, double* algBw, double* busBw, int nranks)
{
double baseBw = (double)(count * typesize * nranks) / 1.0E9 / sec;
*algBw = baseBw;
double factor = ((double)(nranks - 1)) / ((double)nranks);
*busBw = baseBw * factor;
}
testResult_t AllGatherRunColl(void* sendbuff, void* recvbuff, int nranksPerNode, size_t count, mscclppComm_t comm,
cudaStream_t stream)
{
int worldSize = comm->nRanks;
kernel<<<1, 32 * (worldSize - 1), 0, stream>>>(comm->rank, worldSize, nranksPerNode, count / sizeof(int), 1);
return testSuccess;
}
struct testColl allGatherTest = {"AllGather", AllGatherGetCollByteCount, AllGatherInitData, AllGatherGetBw,
AllGatherRunColl};
void AllGatherGetBuffSize(size_t* sendcount, size_t* recvcount, size_t count, int nranks)
{
size_t paramcount, sendInplaceOffset, recvInplaceOffset;
AllGatherGetCollByteCount(sendcount, recvcount, &paramcount, &sendInplaceOffset, &recvInplaceOffset, count, nranks);
}
testResult_t AllGatherRunTest(struct threadArgs* args)
{
args->collTest = &allGatherTest;
mscclppDevConn_t* devConns;
int nCons;
MSCCLPPCHECK(mscclppGetAllDeviceConnections(args->comm, &devConns, &nCons));
CUDACHECK(cudaMemcpyToSymbol(constDevConns, devConns, sizeof(mscclppDevConn_t) * nCons));
TESTCHECK(TimeTest(args));
return testSuccess;
}
struct testEngine allGatherEngine = {AllGatherGetBuffSize, AllGatherRunTest};
#pragma weak mscclppTestEngine = allGatherEngine

501
tests/allgather_test_standalone.cu Normal file
View File

@@ -0,0 +1,501 @@
#include "mscclpp.h"
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
#include "mpi.h"
#endif // MSCCLPP_USE_MPI_FOR_TESTS
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <string>
#include <unistd.h>
#include <unordered_map>
static int nranksPerNode = 8;
// Propagate errors up
#define MSCCLPPCHECK(call) \
do { \
mscclppResult_t res = call; \
if (res != mscclppSuccess && res != mscclppInProgress) { \
/* Print the back trace*/ \
printf("Failure at %s:%d -> %s\n", __FILE__, __LINE__, mscclppGetErrorString(res)); \
return res; \
} \
} while (0)
// Check CUDA RT calls
#define CUDACHECK(cmd) \
do { \
cudaError_t err = cmd; \
if (err != cudaSuccess) { \
printf("%s:%d Cuda failure '%s'\n", __FILE__, __LINE__, cudaGetErrorString(err)); \
exit(EXIT_FAILURE); \
} \
} while (false)
// Measure current time in second.
static double getTime(void)
{
struct timespec tspec;
if (clock_gettime(CLOCK_MONOTONIC, &tspec) == -1) {
printf("clock_gettime failed\n");
exit(EXIT_FAILURE);
}
return (tspec.tv_nsec / 1.0e9) + tspec.tv_sec;
}
__constant__ mscclppDevConn_t constDevConns[16];
__device__ void allgather0(mscclppDevConn_t devConn, int rank, int world_size, int remoteRank, size_t nelemsPerGPU)
{
// this allgather is really simple and implemented as an alltoall
// this thread's role is a sender role
// put your data asynchronously
if ((threadIdx.x % 32) == 0)
devConn.putWithSignal(rank * nelemsPerGPU * sizeof(int), nelemsPerGPU * sizeof(int));
// make sure everyone is put their data before some thread randomly blocks everyone else in signal
__syncthreads();
// push with flag and sync to make sure the data is received
if ((threadIdx.x % 32) == 0)
devConn.flush();
// this thread's role is a receiver role. wait on the semaphore to make sure the data is ready
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
__device__ void localAllGather(mscclppDevConn_t devConn, int rank, int world_size, int nranksPerNode, int remoteRank,
uint64_t offset, uint64_t size)
{
// this allgather algorithm works as follows:
// Step 1: GPU rank i sends data to GPU rank (i+1) % nranksPerNode
// and waits for data from GPU rank (i-1) % nranksPerNode
// Step 2: GPU rank i sends data to GPU rank (i+2) % nranksPerNode
// ...
// This order is much better for DMA engine for NVLinks
for (int i = 1; i < nranksPerNode; i++) {
if ((remoteRank % nranksPerNode) == ((rank + i) % nranksPerNode)) {
// put your data to GPU (rank+i) % nranksPerNode and signal in one call
if ((threadIdx.x % 32) == 0)
devConn.putWithSignalAndFlush(offset, size);
}
// wait for the data from GPU (rank-i) % nranksPerNode to arrive
if ((remoteRank % nranksPerNode) == ((rank - i + nranksPerNode) % nranksPerNode)) {
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
asm volatile("bar.sync %0, %1;" ::"r"(11), "r"((nranksPerNode - 1) * 32) : "memory");
}
}
__device__ void allgather1(mscclppDevConn_t devConn, int rank, int world_size, int nranksPerNode, int remoteRank,
size_t nelemsPerGPU)
{
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank, rank * nelemsPerGPU * sizeof(int),
nelemsPerGPU * sizeof(int));
}
__device__ void allgather2(mscclppDevConn_t devConn, int rank, int world_size, int nranksPerNode, int remoteRank,
size_t nelemsPerGPU)
{
// this allgather is a pipelined and hierarchical one and only works for two nodes
// it is implemented as follows:
// Step 1: each node does a local allgather and concurrently,
// local GPU i exchange (piplineSize-1)/pipelineSize portion of their data with
// its cross-node neighbor (local GPU i on the other node) via IB
// Step 2: each node does a local allgather again with the data just received from its
// cross-node neighbor in step 1, and concurrently, exchange the rest of the data with
// its cross-node neighbor
// Step 3: each node does a local allgather for the last time with the rest of the data
int pipelineSize = 3;
// Step 1
// local allgather
if (remoteRank / nranksPerNode == rank / nranksPerNode) {
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank, rank * nelemsPerGPU * sizeof(int),
nelemsPerGPU * sizeof(int));
}
// cross-node exchange
if (remoteRank % nranksPerNode == rank % nranksPerNode) {
// opposite side
if ((threadIdx.x % 32) == 0)
devConn.putWithSignalAndFlush(rank * nelemsPerGPU * sizeof(int),
(nelemsPerGPU * (pipelineSize - 1)) / pipelineSize * sizeof(int));
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
__syncthreads();
// Step 2
// local allgather
int otherNghr = (rank + nranksPerNode) % world_size;
if (remoteRank / nranksPerNode == rank / nranksPerNode) {
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank, otherNghr * nelemsPerGPU * sizeof(int),
(nelemsPerGPU * (pipelineSize - 1)) / pipelineSize * sizeof(int));
}
// cross-node exchange
if (remoteRank % nranksPerNode == rank % nranksPerNode) {
// opposite side
if ((threadIdx.x % 32) == 0)
devConn.putWithSignalAndFlush((rank * nelemsPerGPU + (nelemsPerGPU * (pipelineSize - 1)) / pipelineSize) *
sizeof(int),
nelemsPerGPU / pipelineSize * sizeof(int));
if ((threadIdx.x % 32) == 0)
devConn.wait();
}
__syncthreads();
// Step 3
// local allgather
if (remoteRank / nranksPerNode == rank / nranksPerNode) {
localAllGather(devConn, rank, world_size, nranksPerNode, remoteRank,
(otherNghr * nelemsPerGPU + (nelemsPerGPU * (pipelineSize - 1)) / pipelineSize) * sizeof(int),
nelemsPerGPU / pipelineSize * sizeof(int));
}
}
__global__ void kernel(int rank, int world_size, int nranksPerNode, size_t nelemsPerGPU, int kernel)
{
// find the mapping between remoteRank and devConns
int warpId = threadIdx.x / 32;
int remoteRank = (warpId < rank) ? warpId : warpId + 1;
// Each warp is responsible for one of the remote ranks
mscclppDevConn_t devConn = constDevConns[warpId];
if (kernel == 0)
allgather0(devConn, rank, world_size, remoteRank, nelemsPerGPU);
else if (kernel == 1)
allgather1(devConn, rank, world_size, nranksPerNode, remoteRank, nelemsPerGPU);
else if (kernel == 2)
allgather2(devConn, rank, world_size, nranksPerNode, remoteRank, nelemsPerGPU);
}
int rankToLocalRank(int rank)
{
return rank % nranksPerNode;
}
int rankToNode(int rank)
{
return rank / nranksPerNode;
}
void print_usage(const char* prog)
{
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
printf("usage: %s IP:PORT [rank nranks]\n", prog);
#else
printf("usage: %s IP:PORT rank nranks\n", prog);
#endif
}
void initializeAndAllocateAllGatherData(int rank, int world_size, size_t dataSize, size_t nelemsPerGPU, int** data_h,
int** data_d)
{
CUDACHECK(cudaMalloc(data_d, dataSize));
CUDACHECK(cudaMemset(*data_d, 0, dataSize));
*data_h = new int[nelemsPerGPU * world_size];
for (size_t i = 0; i < nelemsPerGPU * world_size; i++) {
int val = i + 1;
if (i / nelemsPerGPU == (size_t)rank) {
(*data_h)[i] = val;
} else {
(*data_h)[i] = 0;
}
}
CUDACHECK(cudaMemcpy(*data_d, *data_h, dataSize, cudaMemcpyHostToDevice));
}
mscclppResult_t setupMscclppConnections(int rank, int world_size, mscclppComm_t comm, int* data_d, size_t dataSize)
{
int thisNode = rankToNode(rank);
int cudaNum = rankToLocalRank(rank);
std::string ibDevStr = "mlx5_ib" + std::to_string(cudaNum);
for (int r = 0; r < world_size; ++r) {
if (r == rank)
continue;
mscclppTransport_t transportType;
const char* ibDev = ibDevStr.c_str();
if (rankToNode(r) == thisNode) {
ibDev = NULL;
transportType = mscclppTransportP2P;
} else {
transportType = mscclppTransportIB;
}
// Connect with all other ranks
MSCCLPPCHECK(mscclppConnect(comm, r, 0, data_d, dataSize, transportType, ibDev));
}
MSCCLPPCHECK(mscclppConnectionSetup(comm));
mscclppDevConn_t* devConns;
int nCons;
MSCCLPPCHECK(mscclppGetAllDeviceConnections(comm, &devConns, &nCons));
CUDACHECK(cudaMemcpyToSymbol(constDevConns, devConns, sizeof(mscclppDevConn_t) * nCons));
return mscclppSuccess;
}
void printUsage(const char* prog, bool isMpi)
{
if (isMpi) {
std::string st = "you are using MPI for this test\n";
st += "two possilbe usages are:\n";
st += "> " + std::string(prog) + "\n";
st += "or\n";
st += "> " + std::string(prog) + " -ip_port [ip:port]\n";
printf("%s", st.c_str());
} else {
std::string st = "you are NOT using MPI for this test\n";
st += "the only possible usage:\n";
st += "> " + std::string(prog) + " -ip_port [ip:port] -rank [rank] -nranks [nranks]\n";
printf("%s", st.c_str());
}
}
std::unordered_map<std::string, std::string> parseArgs(int argc, const char* argv[], bool isMpi)
{
std::unordered_map<std::string, std::string> options;
for (int i = 1; i < argc; i++) {
std::string arg = argv[i];
if (arg == "-rankspernode") {
if (isMpi) {
fprintf(stderr, "Error: -rankspernode should not be specified with MPI.\n");
exit(-1);
}
if (i + 1 < argc) {
options["rankspernode"] = argv[++i];
} else {
fprintf(stderr, "Error: -rankspernode option requires an argument.\n");
;
exit(-1);
}
} else if (arg == "-kernel") {
if (i + 1 < argc) {
options["kernel"] = argv[++i];
} else {
fprintf(stderr, "Error: -kernel option requires an argument.\n");
exit(-1);
}
} else if (arg == "-ip_port") {
if (i + 1 < argc) {
options["ip_port"] = argv[++i];
} else {
fprintf(stderr, "Error: -ip_port option requires an argument.\n");
exit(-1);
}
} else if (arg == "-rank") {
if (isMpi) {
fprintf(stderr, "Error: -rank should not be specified with MPI.\n");
exit(-1);
}
if (i + 1 < argc) {
options["rank"] = argv[++i];
} else {
fprintf(stderr, "Error: -ip_port option requires an argument.\n");
exit(-1);
}
} else if (arg == "-nranks") {
if (isMpi) {
fprintf(stderr, "Error: -nranks should not be specified with MPI.\n");
exit(-1);
}
if (i + 1 < argc) {
options["nranks"] = argv[++i];
} else {
fprintf(stderr, "Error: -ip_port option requires an argument.\n");
exit(-1);
}
} else if (arg == "-datasize") {
if (i + 1 < argc) {
options["datasize"] = argv[++i];
} else {
fprintf(stderr, "Error: -datasize option requires an argument.\n");
exit(-1);
}
} else if (arg == "-help" || arg == "-h") {
printUsage(argv[0], isMpi);
exit(0);
} else {
fprintf(stderr, "Error: Unknown option %s\n", argv[i]);
exit(-1);
}
}
return options;
}
int main(int argc, const char* argv[])
{
bool isMpi = false;
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
isMpi = true;
#endif
auto parsedArgs = parseArgs(argc, argv, isMpi);
int rank;
int world_size;
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
MPI_Init(NULL, NULL);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &world_size);
// get the local number of nodes with MPI
MPI_Comm shmcomm;
MPI_Comm_split_type(MPI_COMM_WORLD, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL, &shmcomm);
int shmrank;
MPI_Comm_size(shmcomm, &shmrank);
nranksPerNode = shmrank;
MPI_Comm_free(&shmcomm);
#else
if (parsedArgs.find("rank") == parsedArgs.end() || parsedArgs.find("nranks") == parsedArgs.end()) {
printUsage(argv[0], isMpi);
exit(-1);
}
rank = std::stoi(parsedArgs["rank"]);
world_size = std::stoi(parsedArgs["nranks"]);
if (parsedArgs.find("rankspernode") == parsedArgs.end()) {
printUsage(argv[0], isMpi);
exit(-1);
}
nranksPerNode = std::stoi(parsedArgs["rankspernode"]);
#endif
int kernelNum = 0;
if (parsedArgs.find("kernel") != parsedArgs.end()) {
kernelNum = std::stoi(parsedArgs["kernel"]);
}
char* ip_port = NULL;
if (parsedArgs.find("ip_port") == parsedArgs.end()) {
printUsage(argv[0], isMpi);
exit(-1);
}
ip_port = (char*)parsedArgs["ip_port"].c_str();
int thisNode = rankToNode(rank);
int cudaNum = rankToLocalRank(rank);
CUDACHECK(cudaSetDevice(cudaNum));
if (rank == 0)
printf("Initializing MSCCL++\n");
mscclppComm_t comm;
MSCCLPPCHECK(mscclppCommInitRank(&comm, world_size, ip_port, rank));
int* data_d;
int* data_h;
size_t dataSize = 1024 * 1024 * 1024;
if (parsedArgs.find("datasize") != parsedArgs.end()) {
dataSize = std::stoul(parsedArgs["datasize"]);
}
size_t nelemsPerGPU = dataSize / sizeof(int) / world_size;
if (rank == 0)
printf("Initializing data for allgather test\n");
initializeAndAllocateAllGatherData(rank, world_size, dataSize, nelemsPerGPU, &data_h, &data_d);
if (rank == 0)
printf("Setting up the connection in MSCCL++\n");
MSCCLPPCHECK(setupMscclppConnections(rank, world_size, comm, data_d, dataSize));
if (rank == 0)
printf("Launching MSCCL++ proxy threads\n");
MSCCLPPCHECK(mscclppProxyLaunch(comm));
if (rank == 0)
printf("Testing the correctness of AllGather implementation\n");
cudaStream_t stream;
CUDACHECK(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
CUDACHECK(cudaDeviceSynchronize());
kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nranksPerNode, nelemsPerGPU, kernelNum);
CUDACHECK(cudaDeviceSynchronize());
CUDACHECK(cudaMemcpy(data_h, data_d, dataSize, cudaMemcpyDeviceToHost));
for (size_t i = 0; i < nelemsPerGPU * world_size; i++) {
int val = i + 1;
if (data_h[i] != val) {
printf("oh uh! data_h[%ld] (%d) != val (%d)\n", i, data_h[i], val);
break;
}
}
int tmp[16];
// A simple barrier
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
if (rank == 0)
printf("Successfully checked the correctness\n");
// Perf test
int iterwithoutcudagraph = 10;
if (rank == 0)
printf("Running %d iterations of the kernel without CUDA graph\n", iterwithoutcudagraph);
CUDACHECK(cudaStreamSynchronize(stream));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
for (int i = 0; i < iterwithoutcudagraph; ++i) {
kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nranksPerNode, nelemsPerGPU, kernelNum);
}
CUDACHECK(cudaStreamSynchronize(stream));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
// cudaGraph Capture
int cudagraphiter = 10;
if (rank == 0)
printf("Capturing %d iterations of the kernel in a CUDA graph\n", cudagraphiter);
cudaGraph_t graph;
cudaGraphExec_t instance;
cudaStreamBeginCapture(stream, cudaStreamCaptureModeGlobal);
for (int i = 0; i < cudagraphiter; ++i) {
kernel<<<1, 32 * (world_size - 1), 0, stream>>>(rank, world_size, nranksPerNode, nelemsPerGPU, kernelNum);
}
cudaStreamEndCapture(stream, &graph);
cudaGraphInstantiate(&instance, graph, NULL, NULL, 0);
int cudagraphwarmup = 10;
if (rank == 0)
printf("Warming up %d iterations of the CUDA graph with %d iterations of the kernel\n", cudagraphwarmup,
cudagraphiter);
for (int i = 0; i < cudagraphwarmup; ++i) {
cudaGraphLaunch(instance, stream);
}
CUDACHECK(cudaStreamSynchronize(stream));
// measure runtime
int cudagraphlaunch = 10;
if (rank == 0)
printf("Running %d iterations of the CUDA graph with %d iterations of the kernel\n", cudagraphlaunch,
cudagraphiter);
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
double t0, t1, ms, time_in_us;
t0 = getTime();
for (int i = 0; i < cudagraphlaunch; ++i) {
cudaGraphLaunch(instance, stream);
}
CUDACHECK(cudaStreamSynchronize(stream));
t1 = getTime();
ms = (t1 - t0) * 1000.0;
time_in_us = ms * 1000. / (float)cudagraphlaunch / (float)cudagraphiter;
printf("Rank %d report: size %lu time: %f us/iter algBW %f GBps\n", rank, dataSize, time_in_us,
(double)(dataSize) / 1e9 / (time_in_us / 1e6));
MSCCLPPCHECK(mscclppBootstrapAllGather(comm, tmp, sizeof(int)));
if (rank == 0)
printf("Stopping MSCCL++ proxy threads\n");
MSCCLPPCHECK(mscclppProxyStop(comm));
if (rank == 0)
printf("Destroying MSCCL++ communicator\n");
MSCCLPPCHECK(mscclppCommDestroy(comm));
printf("Rank %d succeeded!\n", rank);
#ifdef MSCCLPP_USE_MPI_FOR_TESTS
MPI_Finalize();
#endif
return 0;
}