303 Commits (8fcbe8258d913e07db68a232dfed5568ac8ca5a0)

Author SHA1 Message Date
  Megvii Engine Team d2278f02d2 perf(imperative): speed up conv_transpose3d 3 years ago
  Megvii Engine Team 3a5347ed21 perf(imperative): speed up pooling 3 years ago
  Megvii Engine Team d9c4ef59fe perf(imperative): using simple hash key in heuristic cache 3 years ago
  Megvii Engine Team 26ea33c6a7 perf(imperative): improve convbwd performance 3 years ago
  Megvii Engine Team 2d72de8a87 fix(mge): fix infer output attrs fallible 3 years ago
  Megvii Engine Team 616352b009 fix(imperative): add dtype promote support for concat 3 years ago
  Megvii Engine Team 95a30eb6f6 perf(imperative): speed up stackmanager guard 3 years ago
  Megvii Engine Team f7a5fe17d3 feat(mge/config): add a config option for memory forwarding 3 years ago
  Megvii Engine Team 9ffc2c0a45 fix(mge): fix host performance loss caused by dtr 3 years ago
  Megvii Engine Team 4d5faa3f0a fix(imperative): using DnnOprCaller to avoid early destruction of dnn_opr 3 years ago
  Megvii Engine Team da620ca193 perf(imperative): specialize batchnorm implementation 3 years ago
  Megvii Engine Team 2a900a69cb perf(imperative): improve reduce op performance 3 years ago
  Megvii Engine Team c2293815b2 fix(autodiff): proxy_graph_detail::make_backward_graph support multiple opnodes 3 years ago
  Megvii Engine Team 273c0e8745 fix(autodiff): fix some bugs in relation to 2nd order grad 3 years ago
  Megvii Engine Team 9779bc7f6d fix(imperative): allow rng op infer shape fallible 3 years ago
  Megvii Engine Team 8fa9a8defa fix(imperative): fix dot-op implement 3 years ago
  Megvii Engine Team 6c413ba943 refactor(mge): refactor physical tensor 3 years ago
  Megvii Engine Team d56570d929 fix(megbrain): add rdnn to copybara 3 years ago
  Megvii Engine Team 7de1bb11ab fix(mge/utils): disable memory forwarding for subgraph 3 years ago
  Megvii Engine Team b7c9361f81 perf(mge/functional): add infer_output_attrs_fallible for some ops 3 years ago
  Megvii Engine Team a4327c4d25 perf(imperative): add dim_expansion transform for conv/bn1d 3 years ago
  Megvii Engine Team 72a70dd6a7 perf(imperative): specialize convolution implementation 3 years ago
  Megvii Engine Team 0a6f4a880e fix(mge/dtr): fix dtr problem 3 years ago
  Megvii Engine Team 529b394f9c fix(imperative): fix profiler problem 3 years ago
  Megvii Engine Team e64536a31e fix(imperative): fix the dtype promote problem when amp 3 years ago
  Megvii Engine Team 2b80806f21 perf(imperative/src): improve dot performance 3 years ago
  Megvii Engine Team 1709b3940b perf(mge/functional): speed up Broadcast and Reshape 3 years ago
  Megvii Engine Team 3e206d899b perf(mge/functional): speed up Split 3 years ago
  Megvii Engine Team 730ddc2d81 perf(interpreter): improve interpreter performance 3 years ago
  Megvii Engine Team 729242f9f8 refactor(imperative): move typecvt code of sereval ops to c++ 3 years ago
  Megvii Engine Team 3c3fc6f33c refactor(imperative): move python code of elemwise/reduce/conv2d/bn to c++ 3 years ago
  Megvii Engine Team 8446626193 perf(imperative/src): improve elemwise 3 years ago
  Megvii Engine Team e400b7ffe5 perf(imperative): enable memory forwarding for imperative 4 years ago
  Megvii Engine Team 84d1a440f0 fix(imperative): do not use output_desc in rng ops 3 years ago
  Megvii Engine Team 0cb60d646d feat(imperative): add output_descs for apply_on_physical_tensor 3 years ago
  Megvii Engine Team c7ded2fe2f refactor(imperative): remove unnecessary reverve in small vector 4 years ago
  Megvii Engine Team 8c2b916ef5 refactor(imperative): remove some methods in proxy graph 4 years ago
  Megvii Engine Team 2348a963f2 refactor(imperative): apply workspace limit hook to mini graph 4 years ago
  Megvii Engine Team fea46ea9a4 perf(imperative): add opr cache for apply_on_physical_tensor 4 years ago
  Megvii Engine Team 177001d5e5 refactor(dispatch): allow dynamic type creation 4 years ago
  Megvii Engine Team 150a6a6151 perf(dispatch/trace): remove unnecessary h2d for constant 4 years ago
  Megvii Engine Team 81d8c73a41 perf(dispatch/trace): serval tricks to speed up trace 4 years ago
  Megvii Engine Team 4fa6162027 perf(dispatch): improve performance of dispatch system 4 years ago
  Megvii Engine Team ca00177719 perf(dispatch): speed up dispatch system 4 years ago
  Megvii Engine Team 7bd848ce04 fix(subgraph): fix hand-written backward for serval jit-elemwise ops 4 years ago
  Megvii Engine Team 7be7656c9f fix(imperative): explicitly manage global structures 4 years ago
  Megvii Engine Team 62034fb262 fix(imperative): make CompNode finalize happens before global object destructor 4 years ago
  Megvii Engine Team b6ce02a152 fix(subgraph): fallback back to cg if jit unsupported 4 years ago
  Megvii Engine Team 225045236b perf(imperative): improve shape inference 4 years ago
  Megvii Engine Team 2775f4580c feat(subgraph): subgraph builder supports jit and custom grad 4 years ago