[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-05 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-508660373 Good to see we have another option to implement operators with high performance! Here's a quick question: How to deal with TVM d

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-05 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-508819329 Oh, it's a bit different from what I thought. If we don't do jit, then tvm can't be used to implement custom op unless rebuilding M

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-07 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-509039433 @junrushao1994 The plugin idea looks too fantastic for me. Of course, If we can achieve that, then integrated TVM is able to provid

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-08 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-509452849 @yzhliu Actually this will be a performance problem. OMP threads will spin when parallel block is completed, this can help to avoid

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-09 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-509537185 @junrushao1994 No, it doesn't make sense. But currently we can specify OMP_NUM_THREADS = physical_core_num / 2 for two paralleled o

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-09 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-509582962 @junrushao1994 Let me clarify my concerns for TVM thread pool. 1> On default, threaded engine will only create single wor

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-09 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-509859385 @tqchen I agree with you that this is a technical issue and shouldn't block this PFC. As thread management should be considered fro

[GitHub] [incubator-mxnet] ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet

2019-07-09 Thread GitBox
ZhennanQin commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-509862734 @yidawang If we all agree that we should avoid using more than one thread pool implementation in the runtime, then I guess OMP is t