microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-24 14:54:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Refactor algo selection logic and introduce symmetric_memory env (#741)

Browse Source 
This PR refactors the algorithm selection logic in MSCCL++ and
introduces support for symmetric memory configuration through
environment variables.


1. Algorithm Selection Refactoring
Use separate class for algo selection. Could introduce more complex
logic for algo selection based on message size, arch, if cuda graph is
enabled and memory allocation method

2. Symmetric Memory Support
Introduced symmetricMemory parameter in algorithm context key
generation. Remove disableChannelCache env as is ambiguous

3. Add new args for build_default_algorithms 
Add flag_buffer, and flag_buffer_size args to build default algorithm.
Then we could use unified flag buffer for different algorithms, avoid
application hanging when switch algo for different message size.

---------

Co-authored-by: chhwang <8018170+chhwang@users.noreply.github.com>
Co-authored-by: Qinghua Zhou <qinghuazhou@microsoft.com>
Co-authored-by: Caio Rocha <caiorocha@microsoft.com>
This commit is contained in:
Binyang Li
2026-02-12 19:06:18 -08:00
committed by GitHub
parent dff3bc7bbb
commit bd68319e3e
43 changed files with 657 additions and 389 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
src/core/algorithm.cc
View File

@@ -3,6 +3,7 @@
#include <filesystem>
#include <mscclpp/algorithm.hpp>
#include <mscclpp/gpu_utils.hpp>
#include "logger.hpp"
@@ -40,12 +41,12 @@ NativeAlgorithm::NativeAlgorithm(std::string name, std::string collective, InitF
CommResult NativeAlgorithm::execute(std::shared_ptr<Communicator> comm, const void* input, void* output,
size_t inputSize, size_t outputSize, DataType dtype, ReduceOp op,
cudaStream_t stream, std::shared_ptr<Executor>, int nBlocks, int nThreadsPerBlock,
const std::unordered_map<std::string, uintptr_t>& extras) {
bool symmetricMemory, const std::unordered_map<std::string, uintptr_t>& extras) {
if (!initialized_) {
initFunc_(comm);
initialized_ = true;
}
AlgorithmCtxKey ctxKey = contextKeyGenFunc_(input, output, inputSize, outputSize, dtype);
AlgorithmCtxKey ctxKey = contextKeyGenFunc_(input, output, inputSize, outputSize, dtype, symmetricMemory);
auto it = contexts_.find(ctxKey);
if (it == contexts_.end()) {
auto ctx = contextInitFunc_(comm, input, output, inputSize, outputSize, dtype);
@@ -155,7 +156,7 @@ Algorithm::Constraint DslAlgorithm::constraint() const { return constraint_; }
CommResult DslAlgorithm::execute(std::shared_ptr<Communicator> comm, const void* input, void* output, size_t inputSize,
size_t outputSize, DataType dtype, ReduceOp, cudaStream_t stream,
std::shared_ptr<Executor> executor, int, int,
std::shared_ptr<Executor> executor, int, int, bool,
const std::unordered_map<std::string, uintptr_t>&) {
if (!executor) {
THROW(EXEC, Error, ErrorCode::InvalidUsage, "Executor is null in DslAlgorithm::execute");
@@ -198,4 +199,18 @@ std::shared_ptr<Algorithm> DslAlgorithm::build() { return shared_from_this(); }
// TODO: implement this
void DslAlgorithm::reset() {}
static std::weak_ptr<uint32_t> gDefaultFlagBuffer;
static size_t gDefaultFlagCount = 128;
std::pair<std::shared_ptr<void>, size_t> getDefaultFlagBuffer() {
std::shared_ptr<uint32_t> flagBuffer = gDefaultFlagBuffer.lock();
if (!flagBuffer) {
flagBuffer = mscclpp::detail::gpuCallocShared<uint32_t>(gDefaultFlagCount);
std::vector<uint32_t> initFlags(gDefaultFlagCount, 1);
mscclpp::gpuMemcpy(flagBuffer.get(), initFlags.data(), gDefaultFlagCount, cudaMemcpyHostToDevice);
gDefaultFlagBuffer = flagBuffer;
}
return {flagBuffer, gDefaultFlagCount * sizeof(uint32_t)};
}
} // namespace mscclpp