Skip to content

GitLab

Commit a4ac3320 authored by lishen's avatar lishen
Browse files

通过线程池实现ipcsocket，满足节点内通信

parent d9d23f34
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 271 additions and 1645 deletions
examples/1_connection/3_sccl_ipc_socket/4_socket_mpi_data_sccl.cpp 0 → 100644
View file @ a4ac3320
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <vector> // 引入vector库
#include <thread> // 为了使用 std::this_thread::sleep_for
#include "mpi.h"
#include "net.h"
#include "ipc_socket.h"
#include "thread_pool.h"
using namespace sccl;
typedef class sccl::hardware::net::ipc_socket::scclIpcSocket scclIpcSocket_t;
template <typename T>
void send_data(T* ipcsocket, const void* data, size_t dataLen, int dst_rank, uint64_t dst_hash) {
if(ipcsocket->scclIpcSocketSendData(data, dataLen, dst_rank, dst_hash) != scclSuccess) {
perror("Failed to send data");
MPI_Abort(MPI_COMM_WORLD, 1);
}
}
template <typename T>
void recv_data(T* ipcsocket, void* buffer, size_t bufferLen, size_t* receivedLen) {
if(ipcsocket->scclIpcSocketRecvData(buffer, bufferLen, receivedLen) != scclSuccess) {
perror("Failed to receive data");
MPI_Abort(MPI_COMM_WORLD, 1);
}
}
int main(int argc, char* argv[]) {
MPI_Init(&argc, &argv);
int rank, size;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
int dst_hash = 12345;
scclIpcSocket_t ipcsocket(rank, dst_hash);
int sendDataLen = 256;
std::vector<char> sendData(sendDataLen);
std::vector<char> recvData(size * sendDataLen);
size_t receivedLen;
// 填充发送数据
snprintf(sendData.data(), sendData.size(), "Data from process %d", rank);
auto pthpool = ThreadPool(size * 2);
// 发送数据给所有其他进程
for(int i = 0; i < size; ++i) {
if(i != rank) {
auto task_send = std::bind(send_data<scclIpcSocket_t>, &ipcsocket, sendData.data(), sendData.size(), i, dst_hash);
pthpool.enqueue(task_send);
auto task_recv = std::bind(recv_data<scclIpcSocket_t>, &ipcsocket, recvData.data() + i * sendDataLen, sendDataLen, &receivedLen);
pthpool.enqueue(task_recv);
}
}
printf("sendData.size()=%d, receivedLen=%d\n", sendDataLen, int(receivedLen));
std::this_thread::sleep_for(std::chrono::seconds(2));
// 打印接收到的数据
for(int i = 0; i < size; ++i) {
printf("Process %d received from process %d: %s\n", rank, i, recvData.data() + i * 256);
}
MPI_Finalize();
return 0;
}
/*
单机执行
SCCL_DEBUG_LEVEL=ABORT SCCL_DEBUG_SUBSYS=BOOTSTRAP mpirun --allow-run-as-root -np 8 3_socket_mpi_data
*/
examples/1_connection/3_sccl_ipc_socket/compile1.sh 0 → 100644
View file @ a4ac3320
hipcc ./1_socket_mpi_fd.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/hardware_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvsymbols.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvwrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/net_ib.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/net_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ipc_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/archinfo.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/thread_pool.cpp \
-o 1_socket_mpi_fd \
-std=c++17 -g -O3 -fopenmp -DROC_SHMEM -D__HIP_PLATFORM_HCC__ -Wno-return-type \
-I ./ -I /usr/include -I /opt/dtk/include \
-I /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/ \
-L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \
-L /usr/lib/x86_64-linux-gnu -libverbs -lrdmacm \
-L /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/lib -lmpi \
-L /opt/dtk/lib -lamdhip64 -lrocm-core -lrocm_smi64 -pthread
examples/1_connection/3_sccl_ipc_socket/compile2.sh 0 → 100644
View file @ a4ac3320
hipcc ./2_socket_mpi_fd_pthpool.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/hardware_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvsymbols.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvwrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/net_ib.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/net_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ipc_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/archinfo.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/thread_pool.cpp \
-o 2_socket_mpi_fd_pthpool \
-std=c++17 -g -O3 -fopenmp -DROC_SHMEM -D__HIP_PLATFORM_HCC__ -Wno-return-type \
-I ./ -I /usr/include -I /opt/dtk/include \
-I /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/ \
-L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \
-L /usr/lib/x86_64-linux-gnu -libverbs -lrdmacm \
-L /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/lib -lmpi \
-L /opt/dtk/lib -lamdhip64 -lrocm-core -lrocm_smi64 -pthread
examples/1_connection/3_sccl_ipc_socket/compile3.sh 0 → 100644
View file @ a4ac3320
hipcc ./3_socket_mpi_data.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/hardware_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvsymbols.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvwrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/net_ib.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/net_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ipc_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/archinfo.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/thread_pool.cpp \
-o 3_socket_mpi_data \
-std=c++17 -g -O3 -fopenmp -DROC_SHMEM -D__HIP_PLATFORM_HCC__ -Wno-return-type \
-I ./ -I /usr/include -I /opt/dtk/include \
-I /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/ \
-L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \
-L /usr/lib/x86_64-linux-gnu -libverbs -lrdmacm \
-L /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/lib -lmpi \
-L /opt/dtk/lib -lamdhip64 -lrocm-core -lrocm_smi64 -pthread
examples/1_connection/3_socket_comm/client.cpp deleted 100644 → 0
View file @ d9d23f34
#include <iostream>
#include <string>
#include <cstring>
#include <unistd.h>
#include <arpa/inet.h>
void start_client(const std::string& server_ip, int server_port) {
int sock = 0;
struct sockaddr_in serv_addr;
char buffer[1024] = {0};
std::string message = "你好，服务器！";
// 创建 socket 文件描述符
if((sock = socket(AF_INET, SOCK_STREAM, 0)) < 0) {
std::cerr << "Socket creation error" << std::endl;
exit(EXIT_FAILURE);
}
serv_addr.sin_family = AF_INET;
serv_addr.sin_port = htons(server_port);
// 转换 IPv4 和 IPv6 地址
if(inet_pton(AF_INET, server_ip.c_str(), &serv_addr.sin_addr) <= 0) {
std::cerr << "Invalid address/ Address not supported" << std::endl;
close(sock);
exit(EXIT_FAILURE);
}
// 连接到服务器
if(connect(sock, (struct sockaddr*)&serv_addr, sizeof(serv_addr)) < 0) {
std::cerr << "Connection Failed" << std::endl;
close(sock);
exit(EXIT_FAILURE);
}
// 发送数据
send(sock, message.c_str(), message.length(), 0);
std::cout << "消息已发送" << std::endl;
// 接收响应
int valread = read(sock, buffer, 1024);
std::cout << "收到的响应: " << buffer << std::endl;
// 关闭连接
close(sock);
}
int main() {
std::string server_ip = "10.16.1.37";
int server_port = 6842;
start_client(server_ip, server_port);
return 0;
}
\ No newline at end of file
examples/1_connection/3_socket_comm/get_ip.cpp deleted 100644 → 0
View file @ d9d23f34
#include <iostream>
#include <ifaddrs.h>
#include <arpa/inet.h>
#include <net/if.h>
#include <stdlib.h>
#include <netdb.h>
#include <unistd.h>
#include <ifaddrs.h>
#include <net/if.h>
#include <vector>
#include <utility>
#include <unordered_set>
#include <unistd.h>
#include <sys/syscall.h>
#define NI_MAXHOST 1025
void get_ip_addresses() {
struct ifaddrs *ifaddr, *ifa;
char host[NI_MAXHOST];
if(getifaddrs(&ifaddr) == -1) {
perror("getifaddrs");
exit(EXIT_FAILURE);
}
for(ifa = ifaddr; ifa != NULL; ifa = ifa->ifa_next) {
if(ifa->ifa_addr == NULL)
continue;
if(ifa->ifa_addr->sa_family == AF_INET) { // 检查是否为 IPv4 地址
(void)getnameinfo(ifa->ifa_addr, sizeof(struct sockaddr_in), host, NI_MAXHOST, NULL, 0, NI_NUMERICHOST);
std::cout << "Interface: " << ifa->ifa_name << " Address: " << host << std::endl;
}
}
freeifaddrs(ifaddr);
}
int main() {
get_ip_addresses();
return 0;
}
\ No newline at end of file
examples/1_connection/3_socket_comm/server.cpp deleted 100644 → 0
View file @ d9d23f34
#include <iostream>
#include <string>
#include <cstring>
#include <unistd.h>
#include <arpa/inet.h>
#include <ifaddrs.h>
#include <net/if.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
void start_server() {
int server_fd, new_socket;
struct sockaddr_in address;
int addrlen = sizeof(address);
char buffer[1024] = {0};
std::string message = "消息已收到";
// 创建 socket 文件描述符
if((server_fd = socket(AF_INET, SOCK_STREAM, 0)) == 0) {
perror("socket failed");
exit(EXIT_FAILURE);
}
// 绑定地址和端口
address.sin_family = AF_INET;
address.sin_addr.s_addr = INADDR_ANY; // 自动获取所有 IP 地址
address.sin_port = htons(6842);
if(bind(server_fd, (struct sockaddr*)&address, sizeof(address)) < 0) {
perror("bind failed");
close(server_fd);
exit(EXIT_FAILURE);
}
// 获取绑定的端口号
socklen_t len = sizeof(address);
if(getsockname(server_fd, (struct sockaddr*)&address, &len) == -1) {
perror("getsockname failed");
close(server_fd);
exit(EXIT_FAILURE);
}
int port = ntohs(address.sin_port);
std::cout << "服务器已启动，端口: " << port << std::endl;
// 监听连接
if(listen(server_fd, 3) < 0) {
perror("listen");
close(server_fd);
exit(EXIT_FAILURE);
}
std::cout << "等待连接..." << std::endl;
// 接受客户端连接
if((new_socket = accept(server_fd, (struct sockaddr*)&address, (socklen_t*)&addrlen)) < 0) {
perror("accept");
close(server_fd);
exit(EXIT_FAILURE);
}
while(true) {
// 接收数据
int valread = read(new_socket, buffer, 1024);
if(valread == 0) {
break;
}
std::cout << "收到的消息: " << buffer << std::endl;
send(new_socket, message.c_str(), message.length(), 0);
memset(buffer, 0, sizeof(buffer));
}
// 关闭连接
close(new_socket);
close(server_fd);
}
int main() {
start_server();
return 0;
}
\ No newline at end of file
examples/1_connection/3_socket_comm/socket.cpp deleted 100644 → 0
View file @ d9d23f34
This diff is collapsed.
examples/1_connection/3_socket_comm/socket.h deleted 100644 → 0
View file @ d9d23f34
#pragma once
#include "debug.h"
#include "check.h"
#include <sys/socket.h>
#include <arpa/inet.h>
#include <netinet/tcp.h>
#include <netdb.h>
#include <fcntl.h>
#include <poll.h>
using namespace sccl;
struct netIf {
char prefix[64];
int port;
};
static thread_local int scclDebugNoWarn = 0;
#define SYSCHECK(call, name) \
do { \
int retval; \
SYSCHECKVAL(call, name, retval); \
} while(false)
#define SYSCHECKVAL(call, name, retval) \
do { \
SYSCHECKSYNC(call, name, retval); \
if(retval == -1) { \
WARN("Call to " name " failed : %s", strerror(errno)); \
return scclSystemError; \
} \
} while(false)
#define SYSCHECKSYNC(call, name, retval) \
do { \
retval = call; \
if(retval == -1 && (errno == EINTR || errno == EWOULDBLOCK || errno == EAGAIN)) { \
INFO(SCCL_LOG_CODEALL, "Call to " name " returned %s, retrying", strerror(errno)); \
} else { \
break; \
} \
} while(true)
#define EQCHECK(statement, value) \
do { \
if((statement) == value) { \
/* Print the back trace*/ \
INFO(SCCL_LOG_CODEALL, "%s:%d -> %d (%s)", __FILE__, __LINE__, scclSystemError, strerror(errno)); \
return scclSystemError; \
} \
} while(0);
#define NEQCHECKGOTO(statement, value, RES, label) \
do { \
if((statement) != value) { \
/* Print the back trace*/ \
RES = scclSystemError; \
INFO(SCCL_LOG_CODEALL, "%s:%d -> %d (%s)", __FILE__, __LINE__, RES, strerror(errno)); \
goto label; \
} \
} while(0);
#define SYSCHECKGOTO(statement, RES, label) \
do { \
if((statement) == -1) { \
/* Print the back trace*/ \
RES = scclSystemError; \
INFO(SCCL_LOG_CODEALL, "%s:%d -> %d (%s)", __FILE__, __LINE__, RES, strerror(errno)); \
goto label; \
} \
} while(0);
#define SCCLCHECKGOTO(call, RES, label) \
do { \
RES = call; \
if(RES != scclSuccess && RES != scclInProgress) { \
/* Print the back trace*/ \
if(scclDebugNoWarn == 0) \
INFO(SCCL_LOG_CODEALL, "%s:%d -> %d", __FILE__, __LINE__, RES); \
goto label; \
} \
INFO(SCCL_LOG_CODEALL, "check pass %s:%d -> %d", __FILE__, __LINE__, RES); \
} while(0);
#define EQCHECKGOTO(statement, value, RES, label) \
do { \
if((statement) == value) { \
/* Print the back trace*/ \
RES = scclSystemError; \
INFO(SCCL_LOG_CODEALL, "%s:%d -> %d (%s)", __FILE__, __LINE__, RES, strerror(errno)); \
goto label; \
} \
} while(0);
static int parseStringList(const char* string, struct netIf* ifList, int maxList) {
if(!string)
return 0;
const char* ptr = string;
int ifNum = 0;
int ifC = 0;
char c;
do {
c = *ptr;
if(c == ':') {
if(ifC > 0) {
ifList[ifNum].prefix[ifC] = '\0';
ifList[ifNum].port = atoi(ptr + 1);
ifNum++;
ifC = 0;
}
while(c != ',' && c != '\0')
c = *(++ptr);
} else if(c == ',' || c == '\0') {
if(ifC > 0) {
ifList[ifNum].prefix[ifC] = '\0';
ifList[ifNum].port = -1;
ifNum++;
ifC = 0;
}
} else {
ifList[ifNum].prefix[ifC] = c;
ifC++;
}
ptr++;
} while(ifNum < maxList && c);
return ifNum;
}
static bool matchIf(const char* string, const char* ref, bool matchExact) {
// Make sure to include '\0' in the exact case
int matchLen = matchExact ? strlen(string) + 1 : strlen(ref);
return strncmp(string, ref, matchLen) == 0;
}
static bool matchPort(const int port1, const int port2) {
if(port1 == -1)
return true;
if(port2 == -1)
return true;
if(port1 == port2)
return true;
return false;
}
static bool matchIfList(const char* string, int port, struct netIf* ifList, int listSize, bool matchExact) {
// Make an exception for the case where no user list is defined
if(listSize == 0)
return true;
for(int i = 0; i < listSize; i++) {
if(matchIf(string, ifList[i].prefix, matchExact) && matchPort(port, ifList[i].port)) {
return true;
}
}
return false;
}
#define MAX_IFS 16
#define MAX_IF_NAME_SIZE 16
#define SLEEP_INT 1000 // connection retry sleep interval in usec
#define RETRY_REFUSED_TIMES 2e4 // connection refused retry times before reporting a timeout (20 sec)
#define RETRY_TIMEDOUT_TIMES 3 // connection timed out retry times (each one can take 20s)
#define SOCKET_NAME_MAXLEN (NI_MAXHOST + NI_MAXSERV)
#define SCCL_SOCKET_MAGIC 0x564ab9f2fc4b9d6cULL
union scclSocketAddress {
struct sockaddr sa;
struct sockaddr_in sin;
struct sockaddr_in6 sin6;
};
enum scclSocketState {
scclSocketStateNone = 0,
scclSocketStateInitialized = 1,
scclSocketStateAccepting = 2,
scclSocketStateAccepted = 3,
scclSocketStateConnecting = 4,
scclSocketStateConnectPolling = 5,
scclSocketStateConnected = 6,
scclSocketStateReady = 7,
scclSocketStateClosed = 8,
scclSocketStateError = 9,
scclSocketStateNum = 10
};
enum scclSocketType {
scclSocketTypeUnknown = 0,
scclSocketTypeBootstrap = 1,
scclSocketTypeProxy = 2,
scclSocketTypeNetSocket = 3,
scclSocketTypeNetIb = 4
};
struct scclSocket {
int fd;
int acceptFd;
int timedOutRetries;
int refusedRetries;
union scclSocketAddress addr;
volatile uint32_t* abortFlag;
int asyncFlag;
enum scclSocketState state;
int salen;
uint64_t magic;
enum scclSocketType type;
};
const char* scclSocketToString(union scclSocketAddress* addr, char* buf, const int numericHostForm = 1);
scclResult_t scclSocketGetAddrFromString(union scclSocketAddress* ua, const char* ip_port_pair);
int scclFindInterfaceMatchSubnet(char* ifNames, union scclSocketAddress* localAddrs, union scclSocketAddress* remoteAddr, int ifNameMaxSize, int maxIfs);
int scclFindInterfaces(char* ifNames, union scclSocketAddress* ifAddrs, int ifNameMaxSize, int maxIfs);
// Initialize a socket
scclResult_t scclSocketInit(struct scclSocket* sock,
union scclSocketAddress* addr = NULL,
uint64_t magic = SCCL_SOCKET_MAGIC,
enum scclSocketType type = scclSocketTypeUnknown,
volatile uint32_t* abortFlag = NULL,
int asyncFlag = 0);
// Create a listening socket. sock->addr can be pre-filled with IP & port info. sock->fd is set after a successful call
scclResult_t scclSocketListen(struct scclSocket* sock);
scclResult_t scclSocketGetAddr(struct scclSocket* sock, union scclSocketAddress* addr);
// Connect to sock->addr. sock->fd is set after a successful call.
scclResult_t scclSocketConnect(struct scclSocket* sock, int portReuse = 0);
// Return socket connection state.
scclResult_t scclSocketReady(struct scclSocket* sock, int* running);
// Accept an incoming connection from listenSock->fd and keep the file descriptor in sock->fd, with the remote side IP/port in sock->addr.
scclResult_t scclSocketAccept(struct scclSocket* sock, struct scclSocket* ulistenSock);
scclResult_t scclSocketGetFd(struct scclSocket* sock, int* fd);
scclResult_t scclSocketSetFd(int fd, struct scclSocket* sock);
#define SCCL_SOCKET_SEND 0
#define SCCL_SOCKET_RECV 1
scclResult_t scclSocketProgress(int op, struct scclSocket* sock, void* ptr, int size, int* offset);
scclResult_t scclSocketWait(int op, struct scclSocket* sock, void* ptr, int size, int* offset);
scclResult_t scclSocketSend(struct scclSocket* sock, void* ptr, int size);
scclResult_t scclSocketRecv(struct scclSocket* sock, void* ptr, int size);
scclResult_t scclSocketTryRecv(struct scclSocket* sock, void* ptr, int size, int* closed, bool blocking);
scclResult_t scclSocketClose(struct scclSocket* sock);
examples/1_connection/3_socket_comm/test_socket_itf.cpp deleted 100644 → 0
View file @ d9d23f34
#include "socket.h"
#include "debug.h"
#include "check.h"
#include <sys/socket.h>
#include <arpa/inet.h>
#include <netinet/tcp.h>
#include <netdb.h>
#include <fcntl.h>
#include <poll.h>
using namespace sccl;
#define MAX_REQUESTS 8
#define MAX_THREADS 16
#define MAX_SOCKETS 64
struct scclNetSocketTask {
int op;
void* data;
int size;
struct scclSocket* sock;
int offset;
int used;
scclResult_t result;
};
struct scclNetSocketTaskQueue {
int next;
int len;
struct scclNetSocketTask* tasks;
};
struct scclNetSocketRequest {
int op;
void* data;
int size;
struct scclSocket* ctrlSock;
int offset;
int used;
struct scclNetSocketComm* comm;
struct scclNetSocketTask* tasks[MAX_SOCKETS];
int nSubs;
};
struct scclNetSocketThreadResources {
struct scclNetSocketTaskQueue threadTaskQueue;
int stop;
struct scclNetSocketComm* comm;
pthread_mutex_t threadLock;
pthread_cond_t threadCond;
};
struct scclNetSocketComm {
struct scclSocket ctrlSock;
struct scclSocket socks[MAX_SOCKETS];
int dev;
int hipDev;
int nSocks;
int nThreads;
int nextSock;
struct scclNetSocketRequest requests[MAX_REQUESTS];
pthread_t helperThread[MAX_THREADS];
struct scclNetSocketThreadResources threadResources[MAX_THREADS];
};
#define DIVUP(x, y) (((x) + (y) - 1) / (y))
#define MIN_CHUNKSIZE (64 * 1024)
template <typename T>
scclResult_t scclCallocDebug(T** ptr, size_t nelem, const char* filefunc, int line) {
void* p = malloc(nelem * sizeof(T));
if(p == NULL) {
WARN("Failed to malloc %ld bytes", nelem * sizeof(T));
return scclSystemError;
}
memset(p, 0, nelem * sizeof(T));
*ptr = (T*)p;
return scclSuccess;
}
#define scclCalloc(...) scclCallocDebug(__VA_ARGS__, __FILE__, __LINE__)
void scclSetThreadName(pthread_t thread, const char* fmt, ...) {
#ifdef _GNU_SOURCE
char threadName[16];
va_list vargs;
va_start(vargs, fmt);
vsnprintf(threadName, 16, fmt, vargs);
va_end(vargs);
pthread_setname_np(thread, threadName);
#endif
}
void* persistentSocketThread(void* args_) {
struct scclNetSocketThreadResources* resource = (struct scclNetSocketThreadResources*)args_;
struct scclNetSocketComm* comm = resource->comm;
struct scclNetSocketTaskQueue* myQueue = &resource->threadTaskQueue;
int nSocksPerThread = comm->nSocks / comm->nThreads;
while(1) {
int idle = 1;
int mark = myQueue->next; // mark newest task seen
for(int i = 0; i < myQueue->len; i += nSocksPerThread) {
int repeat;
do {
repeat = 0;
for(int j = 0; j < nSocksPerThread; j++) {
struct scclNetSocketTask* r = myQueue->tasks + i + j;
if(r != NULL && r->used == 1 && r->offset < r->size) {
r->result = scclSocketProgress(r->op, r->sock, r->data, r->size, &r->offset);
if(r->result != scclSuccess) {
WARN("NET/Socket : socket progress error");
return NULL;
}
idle = 0;
if(r->offset < r->size)
repeat = 1;
}
}
} while(repeat);
}
if(idle) {
pthread_mutex_lock(&resource->threadLock);
while(mark == myQueue->next && resource->stop == 0) { // no new tasks, wait
pthread_cond_wait(&resource->threadCond, &resource->threadLock);
}
pthread_mutex_unlock(&resource->threadLock);
}
if(resource->stop)
return NULL;
}
}
scclResult_t scclNetSocketGetTask(struct scclNetSocketComm* comm, int op, void* data, int size, struct scclNetSocketTask** req) {
int tid = comm->nextSock % comm->nThreads;
struct scclNetSocketThreadResources* res = comm->threadResources + tid;
struct scclNetSocketTaskQueue* queue = &res->threadTaskQueue;
// create helper threads and prepare per-thread task queue
if(queue->tasks == NULL) {
// each request can be divided up to nSocks tasks, and
// these tasks are distributed to nThreads threads,
// we need to make sure each thread queue has enough slots for MAX_REQUESTS
queue->len = MAX_REQUESTS * DIVUP(comm->nSocks, comm->nThreads);
SCCLCHECK(scclCalloc(&queue->tasks, queue->len));
queue->next = 0;
res->comm = comm;
pthread_mutex_init(&res->threadLock, NULL);
pthread_cond_init(&res->threadCond, NULL);
pthread_create(comm->helperThread + tid, NULL, persistentSocketThread, res);
scclSetThreadName(comm->helperThread[tid], "NCCL Sock%c%1u%2u%2u", op == SCCL_SOCKET_SEND ? 'S' : 'R', comm->dev, tid, comm->hipDev);
}
struct scclNetSocketTask* r = queue->tasks + queue->next;
if(r->used == 0) {
r->op = op;
r->data = data;
r->size = size;
r->sock = comm->socks + comm->nextSock;
r->offset = 0;
r->result = scclSuccess;
comm->nextSock = (comm->nextSock + 1) % comm->nSocks;
r->used = 1;
*req = r;
pthread_mutex_lock(&res->threadLock);
queue->next = (queue->next + 1) % queue->len;
pthread_cond_signal(&res->threadCond);
pthread_mutex_unlock(&res->threadLock);
return scclSuccess;
}
WARN("NET/Socket : unable to allocate subtasks");
return scclInternalError;
}
/**
* @brief 测试socket通信请求状态
*
* 该函数用于测试socket通信请求的完成状态，并处理数据传输过程。它会根据请求的不同状态（未开始、正在交换数据大小、已完成交换）执行相应的操作：
* - 如果请求未开始（used=0），则初始化状态
* - 如果正在交换数据大小（used=1），则处理数据大小交换逻辑
* - 如果已完成数据大小交换（used=2），则处理实际数据传输
*
* @param request 指向socket请求的指针
* @param done 输出参数，指示请求是否完成（1=完成，0=未完成）
* @param size 输出参数，返回传输的数据大小
* @return scclResult_t 返回操作结果状态码
*/
scclResult_t scclNetSocketTest(void* request, int* done, int* size) {
*done = 0;
struct scclNetSocketRequest* r = (struct scclNetSocketRequest*)request;
if(r == NULL) {
INFO(SCCL_LOG_CODEALL, "NET/Socket : test called with NULL request");
return scclInternalError;
}
INFO(SCCL_LOG_CODEALL, "NET/Socket : test called request used:%d\n", r->used);
if(r->used == 1) { /* try to send/recv size */
int data = r->size;
int offset = 0;
SCCLCHECK(scclSocketProgress(r->op, r->ctrlSock, &data, sizeof(int), &offset));
if(offset == 0)
return scclSuccess; /* Not ready -- retry later */
// Not sure we could ever receive less than 4 bytes, but just in case ...
if(offset < sizeof(int))
SCCLCHECK(scclSocketWait(r->op, r->ctrlSock, &data, sizeof(int), &offset));
// Check size is less or equal to the size provided by the user
if(r->op == SCCL_SOCKET_RECV && data > r->size) {
char line[SOCKET_NAME_MAXLEN + 1];
union scclSocketAddress addr;
scclSocketGetAddr(r->ctrlSock, &addr);
WARN("NET/Socket : peer %s message truncated : receiving %d bytes instead of %d. If you believe your socket network is in healthy state, \
there may be a mismatch in collective sizes or environment settings (e.g. SCCL_PROTO, SCCL_ALGO) between ranks",
scclSocketToString(&addr, line),
data,
r->size);
return scclInvalidUsage;
}
r->size = data;
r->offset = 0;
r->used = 2; // done exchanging size
// divide into subtasks
int chunkOffset = 0, i = 0;
if(r->comm->nSocks > 0) {
// each request can be divided up to nSocks tasks
int taskSize = std::max(MIN_CHUNKSIZE, DIVUP(r->size, r->comm->nSocks));
while(chunkOffset < r->size) {
int chunkSize = std::min(taskSize, r->size - chunkOffset);
SCCLCHECK(scclNetSocketGetTask(r->comm, r->op, (char*)(r->data) + chunkOffset, chunkSize, r->tasks + i++));
chunkOffset += chunkSize;
}
}
r->nSubs = i;
}
if(r->used == 2) { // already exchanged size
if(r->nSubs > 0) {
int nCompleted = 0;
for(int i = 0; i < r->nSubs; i++) {
struct scclNetSocketTask* sub = r->tasks[i];
if(sub->result != scclSuccess)
return sub->result;
if(sub->offset == sub->size)
nCompleted++;
}
if(nCompleted == r->nSubs) {
if(size)
*size = r->size;
*done = 1;
r->used = 0;
for(int i = 0; i < r->nSubs; i++) {
struct scclNetSocketTask* sub = r->tasks[i];
sub->used = 0;
}
}
} else { // progress request using main thread
if(r->offset < r->size) {
SCCLCHECK(scclSocketProgress(r->op, r->ctrlSock, r->data, r->size, &r->offset));
}
if(r->offset == r->size) {
if(size)
*size = r->size;
*done = 1;
r->used = 0;
}
}
}
return scclSuccess;
}
int main(int argc, char* argv[]) {
struct scclNetSocketRequest* request = (struct scclNetSocketRequest*)malloc(sizeof(struct scclNetSocketRequest));
request->op = SCCL_SOCKET_SEND;
request->used = 1;
request->size = 1024;
request->data = (char*)malloc(request->size);
request->ctrlSock = NULL;
request->comm = NULL;
request->nSubs = 0;
int done;
int sizes[32];
printf("test\n");
INFO(SCCL_LOG_CODEALL, "test INFO");
SCCLCHECK(scclSocketInit(request));
SCCLCHECK(scclNetSocketTest(request, &done, sizes));
if(done) {
printf("done\n");
}
}
\ No newline at end of file
examples/2_topo/0_demo_topo/compile_topo.sh
View file @ a4ac3320
...@@ -6,11 +6,11 @@ hipcc ./test_topo.cpp \ ...@@ -6,11 +6,11 @@ hipcc ./test_topo.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/utils.cc \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/utils.cc \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/archinfo.cc \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/archinfo.cc \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/nvmlwrap.cc \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/nvmlwrap.cc \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/ibvsymbols.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvsymbols.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/ibvwrap.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvwrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/net_ib.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/net_ib.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/host/socket.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/host/net_socket.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/net_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \
...@@ -25,8 +25,8 @@ hipcc ./test_topo.cpp \ ...@@ -25,8 +25,8 @@ hipcc ./test_topo.cpp \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/host/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \
-L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \ -L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \
-L /usr/lib/x86_64-linux-gnu -L /usr/lib/ \ -L /usr/lib/x86_64-linux-gnu -L /usr/lib/ \
......
examples/2_topo/1_demo_rocm/compile_rocm_smi.sh
View file @ a4ac3320
hipcc /public/home/lishen/Code/rocSHMEM/SCCL_v1/examples/2_topo/1_demo_rocm/test_rocm_smi.cpp \ hipcc /public/home/lishen/Code/rocSHMEM/SCCL_v1/examples/2_topo/1_demo_rocm/test_rocm_smi.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/rocm_smi_wrap.cc \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/rocm_smi_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/topo_utils.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/topo_utils.cpp \
-o test_topo \ -o test_topo \
-std=c++17 -g -O3 -fopenmp -D__HIP_PLATFORM_HCC__ \ -std=c++17 -g -O3 -fopenmp -D__HIP_PLATFORM_HCC__ \
...@@ -11,6 +11,7 @@ hipcc /public/home/lishen/Code/rocSHMEM/SCCL_v1/examples/2_topo/1_demo_rocm/test ...@@ -11,6 +11,7 @@ hipcc /public/home/lishen/Code/rocSHMEM/SCCL_v1/examples/2_topo/1_demo_rocm/test
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/topo \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/topo \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ \
-L /usr/lib/x86_64-linux-gnu \ -L /usr/lib/x86_64-linux-gnu \
-L /usr/lib/ \ -L /usr/lib/ \
-lamdhip64 -lrocm_smi64 -lamdhip64 -lrocm_smi64
\ No newline at end of file
examples/2_topo/1_demo_rocm/test_rocm_smi.cpp
View file @ a4ac3320
...@@ -11,21 +11,21 @@ using namespace sccl; ...@@ -11,21 +11,21 @@ using namespace sccl;
int main(int argc, char** argv) { int main(int argc, char** argv) {
printf("hello world\n"); printf("hello world\n");
(void)rocm_smi_init(); (void)sccl::hardware::topology::bootstrap::rocm_smi_init();
uint32_t num_devs; uint32_t num_devs;
(void)rocm_smi_getNumDevice(&num_devs); (void)sccl::hardware::topology::bootstrap::rocm_smi_getNumDevice(&num_devs);
printf("num_devs=%d\n", num_devs); printf("num_devs=%d\n", num_devs);
uint32_t deviceIndex = 0; uint32_t deviceIndex = 0;
char bus0[100] = "bus0"; char bus0[100] = "bus0";
(void)rocm_smi_getDevicePciBusIdString(deviceIndex, bus0, 100); (void)sccl::hardware::topology::bootstrap::rocm_smi_getDevicePciBusIdString(deviceIndex, bus0, 100);
printf("bus0=%s\n", bus0); printf("bus0=%s\n", bus0);
RSMI_IO_LINK_TYPE rsmi_type; RSMI_IO_LINK_TYPE rsmi_type;
int hops, count; int hops, count;
(void)rocm_smi_getLinkInfo(0, 8, &rsmi_type, &hops, &count); (void)sccl::hardware::topology::bootstrap::rocm_smi_getLinkInfo(0, 8, &rsmi_type, &hops, &count);
printf("rsmi_type=%d, hops=%d, count=%d\n", rsmi_type, hops, count); printf("rsmi_type=%d, hops=%d, count=%d\n", rsmi_type, hops, count);
// struct sccl::hardware::topology::topo::scclXml* xml; // struct sccl::hardware::topology::topo::scclXml* xml;
......
examples/2_topo/2_bootstrap/1_mpi_init.cpp
View file @ a4ac3320
...@@ -4,7 +4,8 @@ ...@@ -4,7 +4,8 @@
#include <stdint.h> #include <stdint.h>
#include "mpi.h" #include "mpi.h"
#include "net.h" #include "net.h"
#include "bootstrap_net.h" #include "bootstrap.h"
#include "hardware_utils.h"
using namespace sccl; using namespace sccl;
...@@ -23,17 +24,35 @@ int main(int argc, char* argv[]) { ...@@ -23,17 +24,35 @@ int main(int argc, char* argv[]) {
// ----------------------------------------------------------------------- // // ----------------------------------------------------------------------- //
INFO(SCCL_LOG_TOPO, "Bootstrap ...\n"); INFO(SCCL_LOG_TOPO, "Bootstrap ...\n");
struct scclRankInfo* rank_info;
struct sccl::hardware::topology::bootstrap::scclBootstrapComm* comm;
(void)sccl::hardware::topology::bootstrap::bootstrap_net::bootstrapNetInit(); SCCLCHECK(scclCalloc(&rank_info, 1));
SCCLCHECK(scclCalloc(&comm, 1));
rank_info->rank = rank;
rank_info->nRanks = nranks;
rank_info->localRanks = 2;
rank_info->hipDev = rank % rank_info->localRanks;
auto sccl_bootstrap = new sccl::hardware::topology::bootstrap::scclBootstrap(rank_info, comm);
SCCLCHECK(sccl_bootstrap->bootstrapInitCheck());
sccl::hardware::topology::bootstrap::printUniqueInfo(comm->unique_info);
int cuda_id;
HIPCHECK(hipGetDevice(&cuda_id));
printf("rank=%d, cuda_id=%d\n", rank, cuda_id);
MPI_Finalize(); MPI_Finalize();
} }
/* /*
单机执行 单机执行
SCCL_DEBUG_LEVEL=SCCL_LOG_ABORT mpirun --allow-run-as-root -np 2 1_mpi_init SCCL_DEBUG_LEVEL=ABORT mpirun --allow-run-as-root -np 4 1_mpi_init
SCCL_DEBUG_LEVEL=SCCL_LOG_INFO SCCL_DEBUG_POS=SCCL_LOG_CODEALL mpirun --allow-run-as-root -np 2 1_mpi_init SCCL_DEBUG_LEVEL=INFO SCCL_DEBUG_SUBSYS=ALL mpirun --allow-run-as-root -np 2 1_mpi_init
跨机执行 跨机执行
SCCL_DEBUG_LEVEL=SCCL_LOG_ABORT mpirun --allow-run-as-root --hostfile hostfile -np 16 ./1_mpi_init SCCL_DEBUG_LEVEL=ABORT mpirun --allow-run-as-root --hostfile hostfile -np 16 ./1_mpi_init
SCCL_DEBUG_LEVEL=ABORT SCCL_DEBUG_SUBSYS=BOOTSTRAP mpirun --allow-run-as-root --hostfile hostfile2 -np 4 ./1_mpi_init
*/ */
examples/2_topo/2_bootstrap/compile_mpi.sh
View file @ a4ac3320
hipcc ./1_mpi_init.cpp \ hipcc ./1_mpi_init.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/ibvsymbols.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/hardware_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/ibvwrap.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvsymbols.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/net_ib.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ibvwrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/host/socket.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/net_ib.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/host/net_socket.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/net_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ipc_socket.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/rocm_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/bootstrap_net.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/bootstrap_net.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ipcsocket.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/bootstrap_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/proxy.cpp \ /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/rocm_smi_wrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/topo_utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/archinfo.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/param.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/utils.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/bootstrap.cpp \
/public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/thread_pool.cpp \
-o 1_mpi_init \ -o 1_mpi_init \
-std=c++17 -g -O3 -fopenmp -DROC_SHMEM -D__HIP_PLATFORM_HCC__ \ -std=c++17 -g -O3 -fopenmp -DROC_SHMEM -D__HIP_PLATFORM_HCC__ -Wno-return-type \
-I ./ -I /usr/include -I /opt/dtk/include \ -I ./ -I /usr/include -I /opt/dtk/include \
-I /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/include/ \ -I /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/utils/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/include/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/include/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/device/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_ib/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/host/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/net_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ipc_socket/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/net/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ \ -I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/bootstrap/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/ \
-I /public/home/lishen/Code/rocSHMEM/SCCL_v1/src/hardware/topology/ \
-L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \ -L /public/home/lishen/Code/rocSHMEM/SCCL_v1 \
-L /usr/lib/x86_64-linux-gnu -libverbs -lrdmacm \ -L /usr/lib/x86_64-linux-gnu -libverbs -lrdmacm \
-L /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/lib -lmpi -L /public/home/lishen/Code/rocSHMEM/3rd_party/install/ompi/lib -lmpi \
-L /opt/dtk/lib -lamdhip64 -lrocm-core -lrocm_smi64 -pthread
export HSA_FORCE_FINE_GRAIN_PCIE="1"
export iommu=pt
examples/2_topo/2_bootstrap/hostfile 0 → 100644
View file @ a4ac3320
node037 slots=8
node038 slots=8
\ No newline at end of file
examples/2_topo/2_bootstrap/hostfile2 0 → 100644
View file @ a4ac3320
node037 slots=2
node038 slots=2
\ No newline at end of file
examples/get_cpp_files.py 0 → 100644
View file @ a4ac3320
import os
import glob
from pathlib import Path
def find_cpp_files(directory):
return [str(file) for file in Path(directory).rglob('*.cpp')]
def main():
src_path = "/public/home/lishen/Code/rocSHMEM/SCCL_v1/src"
cpp_files = find_cpp_files(src_path)
cpp_files.sort()
for cpp_file in cpp_files:
print(cpp_file+' \\')
if __name__ == "__main__":
main()
src/hardware/comm.h
View file @ a4ac3320
#pragma once #pragma once
#include <hip/hip_runtime.h>
#include <hip/hip_runtime_api.h>
#include <stdint.h> #include <stdint.h>
#include "base.h" #include "base.h"
#include "topo.h"
namespace sccl { namespace sccl {
namespace hardware { namespace hardware {
// 定义结构体 scclUniqueInfo，用于存储每个通信节点的信息
struct scclUniqueInfo {
int rank; // 当前节点的全局排名
int nRanks; // 总的节点数量
int localRank; // 当前节点在本地计算节点中的排名
int localRanks; // 本地计算节点中的节点总数
int cudaDev; // CUDA 设备 ID
int gdrSupport; // 是否支持 GPU 直接注册 (GDR)
uint64_t hostHash; // 主机哈希值
uint64_t pidHash; // 进程 ID 哈希值
int64_t busId; // 总线 ID
};
// // 定义结构体 scclCommBase，用于存储通信基础信息 // // 定义结构体 scclCommBase，用于存储通信基础信息
// struct scclCommBase { // struct scclCommBase {
// struct scclUniqueInfo* peerInfo; // 指向 peerInfo 结构体的指针，存储所有节点的信息 // struct scclUniqueInfo* peerInfo; // 指向 peerInfo 结构体的指针，存储所有节点的信息
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment