not process non task split when kernel by kernel
This commit is contained in:
parent
c1efb947ca
commit
cdb874a942
|
@ -194,6 +194,12 @@ void AscendSomas::IndependentNodeOutputProcess(const session::KernelGraph &graph
|
|||
}
|
||||
|
||||
void AscendSomas::NonTaskSplitProcess(const session::KernelGraph &graph) {
|
||||
// When not used task sink mode, should not process non task split mem-reuse.
|
||||
// Because the logic for processing non task split memory offset is in the function TaskGenerator:: GetTaskInput,
|
||||
// only run in task sink mode.
|
||||
if (!graph.is_graph_run_mode()) {
|
||||
return;
|
||||
}
|
||||
auto &kernel_cnodes = graph.execution_order();
|
||||
for (const auto &kernel : kernel_cnodes) {
|
||||
auto op_name = common::AnfAlgo::GetCNodeName(kernel);
|
||||
|
|
Loading…
Reference in New Issue