microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-03 05:01:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revised ProxyChannel interfaces (#400)

Browse Source 
* Renamed `ProxyChannel` -> `BaseProxyChannel` and `SimpleProxyChannel`
-> `ProxyChannel`. It makes the interface more consistent by defining
channels to be associated with a certain src/dst memory region:
`ProxyChannel` as "sema + src/dst + fifo" and `SmChannel` as "sema +
src/dst". BaseProxyChannel is not associated with any memory regions, as
"sema + fifo".
* `ProxyChannelDeviceHandle` now inherits from
`BaseProxyChannelDeviceHandle`, instead of having one as a member.
This commit is contained in:
Changho Hwang
2024-12-06 10:53:34 -08:00
committed by GitHub
parent f6305a3c1d
commit 756f24c697
25 changed files with 255 additions and 250 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
docs/getting-started/tutorials/initialization.md
View File

@@ -21,7 +21,7 @@ We will setup a mesh topology with eight GPUs. Each GPU will be connected to its
template <class T>
using DeviceHandle = mscclpp::DeviceHandle<T>;
__constant__ DeviceHandle<mscclpp::SimpleProxyChannel> constProxyChans[8];
__constant__ DeviceHandle<mscclpp::ProxyChannel> constProxyChans[8];
void setupMeshTopology(int rank, int worldsize, void* data, size_t dataSize) {
std::string ip_port = "10.0.0.4:50000";
@@ -55,17 +55,17 @@ void setupMeshTopology(int rank, int worldsize, void* data, size_t dataSize) {
comm.setup();
std::vector<DeviceHandle<mscclpp::SimpleProxyChannel>> proxyChannels;
std::vector<DeviceHandle<mscclpp::ProxyChannel>> proxyChannels;
for (size_t i = 0; i < semaphoreIds.size(); ++i) {
proxyChannels.push_back(mscclpp::deviceHandle(mscclpp::SimpleProxyChannel(
proxyChannels.push_back(mscclpp::deviceHandle(mscclpp::ProxyChannel(
proxyService.proxyChannel(semaphoreIds[i]), proxyService.addMemory(remoteMemories[i].get()),
proxyService.addMemory(localMemories[i]))));
}
if (proxyChannels.size() > sizeof(constProxyChans) / sizeof(DeviceHandle<mscclpp::SimpleProxyChannel>)) {
if (proxyChannels.size() > sizeof(constProxyChans) / sizeof(DeviceHandle<mscclpp::ProxyChannel>)) {
std::runtime_error("unexpected error");
}
CUDACHECK(cudaMemcpyToSymbol(constProxyChans, proxyChannels.data(),
sizeof(DeviceHandle<mscclpp::SimpleProxyChannel>) * proxyChannels.size()));
sizeof(DeviceHandle<mscclpp::ProxyChannel>) * proxyChannels.size()));
}
```