RandySheriffH 657ab2f43c Sync between parent node and subgraph (#15757) ... By https://github.com/microsoft/onnxruntime/issues/14691, we found that there is a mis-reuse of GPU memory between NonZero(GPU) and Identity(GPU) which is a subgraph node in If(CPU). The NonZero gives a GPU output consumed by Transpose(GPU), after which that GPU output marks as free in BFCArena, and soon be reused by Identity(GPU) in a subgraph of If(CPU). However, NonZero(GPU) and Identity(GPU) run on separate cuda streams, there is no synchronization because the Identity node is in a subgraph of If(CPU). Meaning - Identity(GPU) can write to the memory when Transpose(GPU) is reading from it. --------- Co-authored-by: Randy Shuai <rashuai@microsoft.com>		2023-05-11 09:28:04 -07:00
..
3_gpu_streams.json	remove unnecessary waitOnEPStep when current node and the consumer node are in the same stream (#14173)	2023-01-20 07:35:15 -08:00
conv_add_relu.onnx
conv_add_relu_single_stream.json
conv_add_relu_single_stream_mismatch_device.json
conv_add_relu_single_stream_missing_node.json
cpu_if.onnx	Sync between parent node and subgraph (#15757)	2023-05-11 09:28:04 -07:00
memcpyToHost_same_stream_with_transpose.json	Make MemcpyToHost to a separate stream for performance gain (#14487)	2023-02-23 14:52:01 -08:00
multi_stream_double_stream.json
multi_stream_single_stream.json
simplified_ssd.onnx
simplified_ssd_cpu.json