pytorch

8ea033a0 - [profiling] API For profiling backend memory events (#80350)

[profiling] API For profiling backend memory events (#80350) Summary: Adding new API call ``` mobile::getCurrentEdgeProfiler()->recordBackendMemoryEvent( ptr, alloc_size, total_allocated, total_reserved, device); ``` As well as another macro to use for recording backend memory events. These memory events will be captured in the trace file when we create the profiler: ``` { KinetoEdgeCPUProfiler profiler( module, trace_file_name, false, // record input_shapes true, // profile memory true, // record callstack false, // record flops true); // record module hierarchy module.forward(inputs); } ``` Test Plan: Testing to do in the next diff Differential Revision: D37116111 Pull Request resolved: https://github.com/pytorch/pytorch/pull/80350 Approved by: https://github.com/kimishpatel