Merge pull request #9560 from chengduoZH/feature/fix_parallel_exe

Broadcast the gradient once it is generated
helinwang-patch-1
chengduo 7 years ago committed by GitHub
commit dd75fbde81
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

@ -55,6 +55,7 @@ std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
const ProgramDesc &program) const {
auto graph = new SSAGraph();
SSAGraph &result = *graph;
std::unordered_set<std::string> og_has_been_broadcast;
result.vars_.resize(places_.size());
bool is_forwarding = true;
@ -122,9 +123,15 @@ std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
if (!is_forwarding) {
auto var_names = op->OutputArgumentNames();
// Currently, we assume that once gradient is generated, it can be
// broadcast, and each gradient is only broadcast once. But there are no
// other cases, for example, we need to adjust the gradient according to
// the input when we get the gradient, which is not considered at present.
for (auto &og : var_names) {
if (grad_names_.count(og) != 0) { // is param grad
// Insert NCCL AllReduce Op
if (grad_names_.count(og) != 0 &&
og_has_been_broadcast.count(og) == 0) { // is param grad
// Insert NCCL AllReduce Op
og_has_been_broadcast.insert(og);
#ifdef PADDLE_WITH_CUDA
result.ops_.emplace_back(
new NCCLAllReduceOpHandle(local_scopes_, places_, *nccl_ctxs_));

Loading…
Cancel
Save