[GitHub] [incubator-mxnet] xidulu commented on issue #15928: [RFC] A faster version of Gamma sampling on GPU.
xidulu commented on issue #15928: [RFC] A faster version of Gamma sampling on GPU. URL: https://github.com/apache/incubator-mxnet/issues/15928#issuecomment-522209814 Hi @ptrendx , thanks for your reply, according to my discussion with @yzhliu , device-side API is much slower than host-side API. Also, could you please talk a little bit about the advantage of your approach compared with mine? thx :) This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on issue #15931: TBlob bug about dltensor
reminisce commented on issue #15931: TBlob bug about dltensor URL: https://github.com/apache/incubator-mxnet/issues/15931#issuecomment-522208880 I too encountered the illegal-memory-access error probably resulted from the root cause revealed here (I called `TBlob::reshape`). I think overriding assignment operator and copy constructor with `SetDLTensor()` explicitly called is reasonable. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet] branch master updated (5a4c01b -> a8b9728)
This is an automated email from the ASF dual-hosted git repository. patriczhao pushed a change to branch master in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git. from 5a4c01b [Opperf] Filter out deprecated ops (#15541) add a8b9728 Add disable attr to subgraph property (#15926) No new revisions were added by this update. Summary of changes: docs/tutorials/c++/subgraphAPI.md | 6 -- src/c_api/c_api_symbolic.cc | 8 src/executor/graph_executor.cc | 9 +++-- src/operator/subgraph/mkldnn/mkldnn_conv_property.h | 7 +++ src/operator/subgraph/mkldnn/mkldnn_fc_property.h | 7 +++ src/operator/subgraph/subgraph_property.h | 11 +++ 6 files changed, 32 insertions(+), 16 deletions(-)
[GitHub] [incubator-mxnet] pengzhao-intel merged pull request #15926: Add disable attr to subgraph property
pengzhao-intel merged pull request #15926: Add disable attr to subgraph property URL: https://github.com/apache/incubator-mxnet/pull/15926 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15926: Add disable attr to subgraph property
pengzhao-intel commented on issue #15926: Add disable attr to subgraph property URL: https://github.com/apache/incubator-mxnet/pull/15926#issuecomment-522208056 Thanks for the fixing and merging now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] marcoabreu edited a comment on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
marcoabreu edited a comment on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522203123 We had the same issue with clang unix if I recall correctly. So I'd say that this is not a bug but rather a symptom. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] marcoabreu commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
marcoabreu commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522203123 We had the same issue with clang if I recall correctly. So I'd say that this is not a bug but rather a symptom. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] Zha0q1 edited a comment on issue #15132: Profiler API Enhancements
Zha0q1 edited a comment on issue #15132: Profiler API Enhancements URL: https://github.com/apache/incubator-mxnet/pull/15132#issuecomment-522194858 > > The `ascending` argument is great, but why is it missing the most important sort criteria "total time"? That's usually THE first thing people look at - what operator takes most of the time in a network. This is definitely a must-have for the 'sort-by' argument .. > > Sure. I will create a pr and add "total_time" soon. will do this next Friday after I come back from my trip. Now I dont have access to my laptop This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] Zha0q1 commented on issue #15132: Profiler API Enhancements
Zha0q1 commented on issue #15132: Profiler API Enhancements URL: https://github.com/apache/incubator-mxnet/pull/15132#issuecomment-522194858 > > The `ascending` argument is great, but why is it missing the most important sort criteria "total time"? That's usually THE first thing people look at - what operator takes most of the time in a network. This is definitely a must-have for the 'sort-by' argument .. > > Sure. I will create a pr and add "total_time" soon. will do this on next Friday after I come back from my trip. Now I dont have access to my laptop This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API
mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314930220 ## File path: src/common/exec_utils.h ## @@ -621,6 +621,42 @@ inline nnvm::Graph AssignContext(nnvm::Graph g, return g; } +/*! + * \brief infers shapes, dtypes, stypes, contexts for the forward graph + */ +inline nnvm::Graph InferForwardAttrs(nnvm::Graph g, + mxnet::ShapeVector arg_shapes, + nnvm::DTypeVector arg_dtypes, + StorageTypeVector arg_stypes, + const Context& default_ctx, + const std::map& ctx_map, + const std::vector& in_arg_ctxes, + const std::vector& aux_state_ctxes, + bool partial_shape = false) { + const auto& indexed_graph = g.indexed_graph(); + const auto num_forward_inputs = indexed_graph.input_nodes().size(); + g = AssignContext(g, default_ctx, ctx_map, in_arg_ctxes, {}, + aux_state_ctxes, {}, num_forward_inputs, g.outputs.size()); Review comment: AssignContext sets 'context' attribute to graph https://github.com/apache/incubator-mxnet/blob/master/src/common/exec_utils.h#L512 which is used to set 'dev_mask' graph attribute later in InferStorageType https://github.com/apache/incubator-mxnet/blob/master/src/executor/infer_graph_attr_pass.cc#L713-L719 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. marcoabreu pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new 6856b46 Bump the publish timestamp. 6856b46 is described below commit 6856b467322a1b7f131d483d0c2b9f8d1df4e4f5 Author: mxnet-ci AuthorDate: Sat Aug 17 01:31:32 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..9f6e1a1 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Sat Aug 17 01:31:32 UTC 2019
[GitHub] [incubator-mxnet] larroy commented on a change in pull request #15922: Refactor for windows CI 'out of heap space' errors
larroy commented on a change in pull request #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#discussion_r314929315 ## File path: src/operator/tensor/broadcast_reduce_op.h ## @@ -1580,6 +1581,20 @@ void PickOpBackward(const nnvm::NodeAttrs& attrs, }); } +inline std::string get_reduce_axes_description(const std::string& op_name, int line) { Review comment: Of course I know that. The question is why isn't declared on the header and defined in the implementation file as usual. Seems like a big function to inline. Which adds bloat. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927157 ## File path: src/c_api/c_api_symbolic.cc ## @@ -1181,3 +1182,59 @@ int MXShallowCopySymbol(SymbolHandle src, SymbolHandle* out) { *out = out_sym; API_END_HANDLE_ERROR(delete out_sym); } + +int MXOptimizeForBackend(SymbolHandle sym_handle, + const char* backend_name, + SymbolHandle* ret_sym_handle, + const bool infer_shape_type, + const int dev_type, + const int dev_id, + const mx_uint len, + NDArrayHandle* in_args_handle, + const mx_uint num_options, + const char** keys, + const char** vals) { + nnvm::Symbol *s = new nnvm::Symbol(); + API_BEGIN(); + nnvm::Symbol *sym = static_cast(sym_handle); + *s = sym->Copy(); + nnvm::Graph orig_g = Symbol2Graph(*s); + if (infer_shape_type) { +Context default_ctx = Context::Create(static_cast(dev_type), dev_id); +std::map ctx_map; +NDArray **in_args_ptr = reinterpret_cast(in_args_handle); +mxnet::ShapeVector arg_shapes(len); +nnvm::DTypeVector arg_dtypes(len); +StorageTypeVector arg_stypes(len); +std::vector in_arg_ctxes(len); +std::vector aux_state_ctxes; +for (mx_uint i = 0; i < len; i++) { + const auto &in_arg = *(in_args_ptr[i]); + arg_shapes.push_back(in_arg.shape()); + arg_dtypes.push_back(in_arg.dtype()); + arg_stypes.push_back(in_arg.storage_type()); + in_arg_ctxes[i] = in_arg.ctx(); +} +orig_g = common::InferForwardAttrs(orig_g, arg_shapes, arg_dtypes, arg_stypes, default_ctx, + ctx_map, in_arg_ctxes, aux_state_ctxes, true); + } + std::vector> options_map; + for (mx_uint i = 0; i < num_options; ++i) { +options_map.emplace_back(keys[i], vals[i]); + } + const auto backend = mxnet::op::SubgraphBackendRegistry::Get()->GetSubgraphBackend(backend_name); + const auto& subgraph_prop_list = backend->GetSubgraphProperties(); + for (auto property : subgraph_prop_list) { +nnvm::Graph g = orig_g; +property->SetAttr("graph", g); Review comment: dont think we need this if we're going to pass it into the prePartition function please check intel's subgraph props to see if they're using the graph attribute, and if not please remove This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927384 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,37 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimize_for(self, backend, ctx=None, args=None, **kwargs): +"""Partition symbol and optimize it for a given backend""" +out = SymbolHandle() +if ctx is not None and args is not None): +infer_shape_type = False +else: +infer_shape_type = True +if not isinstance(ctx, Context): +raise TypeError("Context type error") Review comment: if you raise an exception here, that means ctx is required argument This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927455 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,37 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimize_for(self, backend, ctx=None, args=None, **kwargs): +"""Partition symbol and optimize it for a given backend""" +out = SymbolHandle() +if ctx is not None and args is not None): Review comment: should this set infer_shape_type = True? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314928296 ## File path: src/common/exec_utils.h ## @@ -621,6 +621,42 @@ inline nnvm::Graph AssignContext(nnvm::Graph g, return g; } +/*! + * \brief infers shapes, dtypes, stypes, contexts for the forward graph + */ +inline nnvm::Graph InferForwardAttrs(nnvm::Graph g, + mxnet::ShapeVector arg_shapes, + nnvm::DTypeVector arg_dtypes, + StorageTypeVector arg_stypes, + const Context& default_ctx, + const std::map& ctx_map, + const std::vector& in_arg_ctxes, + const std::vector& aux_state_ctxes, + bool partial_shape = false) { + const auto& indexed_graph = g.indexed_graph(); + const auto num_forward_inputs = indexed_graph.input_nodes().size(); + g = AssignContext(g, default_ctx, ctx_map, in_arg_ctxes, {}, + aux_state_ctxes, {}, num_forward_inputs, g.outputs.size()); Review comment: this is the only consumer of default_ctx, is it really necessary to just do the shape/type prop that we're looking for pre-partitioning? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927455 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,37 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimize_for(self, backend, ctx=None, args=None, **kwargs): +"""Partition symbol and optimize it for a given backend""" +out = SymbolHandle() +if ctx is not None and args is not None): Review comment: should this be OR? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927384 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,37 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimize_for(self, backend, ctx=None, args=None, **kwargs): +"""Partition symbol and optimize it for a given backend""" +out = SymbolHandle() +if ctx is not None and args is not None): +infer_shape_type = False +else: +infer_shape_type = True +if not isinstance(ctx, Context): +raise TypeError("Context type error") Review comment: if you raise an exception here, that means ctx is required argument This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927175 ## File path: src/c_api/c_api_symbolic.cc ## @@ -1181,3 +1182,59 @@ int MXShallowCopySymbol(SymbolHandle src, SymbolHandle* out) { *out = out_sym; API_END_HANDLE_ERROR(delete out_sym); } + +int MXOptimizeForBackend(SymbolHandle sym_handle, + const char* backend_name, + SymbolHandle* ret_sym_handle, + const bool infer_shape_type, + const int dev_type, + const int dev_id, + const mx_uint len, + NDArrayHandle* in_args_handle, + const mx_uint num_options, + const char** keys, + const char** vals) { + nnvm::Symbol *s = new nnvm::Symbol(); + API_BEGIN(); + nnvm::Symbol *sym = static_cast(sym_handle); + *s = sym->Copy(); + nnvm::Graph orig_g = Symbol2Graph(*s); + if (infer_shape_type) { +Context default_ctx = Context::Create(static_cast(dev_type), dev_id); +std::map ctx_map; +NDArray **in_args_ptr = reinterpret_cast(in_args_handle); +mxnet::ShapeVector arg_shapes(len); +nnvm::DTypeVector arg_dtypes(len); +StorageTypeVector arg_stypes(len); +std::vector in_arg_ctxes(len); +std::vector aux_state_ctxes; +for (mx_uint i = 0; i < len; i++) { + const auto &in_arg = *(in_args_ptr[i]); + arg_shapes.push_back(in_arg.shape()); + arg_dtypes.push_back(in_arg.dtype()); + arg_stypes.push_back(in_arg.storage_type()); + in_arg_ctxes[i] = in_arg.ctx(); +} +orig_g = common::InferForwardAttrs(orig_g, arg_shapes, arg_dtypes, arg_stypes, default_ctx, + ctx_map, in_arg_ctxes, aux_state_ctxes, true); + } + std::vector> options_map; + for (mx_uint i = 0; i < num_options; ++i) { +options_map.emplace_back(keys[i], vals[i]); + } + const auto backend = mxnet::op::SubgraphBackendRegistry::Get()->GetSubgraphBackend(backend_name); + const auto& subgraph_prop_list = backend->GetSubgraphProperties(); + for (auto property : subgraph_prop_list) { +nnvm::Graph g = orig_g; +property->SetAttr("graph", g); +property->PrePartition(g, options_map); +g.attrs["subgraph_property"] = std::make_shared(property); +g = ApplyPass(std::move(g), "BuildSubgraph"); +g.attrs.erase("subgraph_property"); +property->PostPartition(g, options_map); +property->RemoveAttr("graph"); Review comment: if we get rid of set, remove this one too This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API
samskalicky commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314927157 ## File path: src/c_api/c_api_symbolic.cc ## @@ -1181,3 +1182,59 @@ int MXShallowCopySymbol(SymbolHandle src, SymbolHandle* out) { *out = out_sym; API_END_HANDLE_ERROR(delete out_sym); } + +int MXOptimizeForBackend(SymbolHandle sym_handle, + const char* backend_name, + SymbolHandle* ret_sym_handle, + const bool infer_shape_type, + const int dev_type, + const int dev_id, + const mx_uint len, + NDArrayHandle* in_args_handle, + const mx_uint num_options, + const char** keys, + const char** vals) { + nnvm::Symbol *s = new nnvm::Symbol(); + API_BEGIN(); + nnvm::Symbol *sym = static_cast(sym_handle); + *s = sym->Copy(); + nnvm::Graph orig_g = Symbol2Graph(*s); + if (infer_shape_type) { +Context default_ctx = Context::Create(static_cast(dev_type), dev_id); +std::map ctx_map; +NDArray **in_args_ptr = reinterpret_cast(in_args_handle); +mxnet::ShapeVector arg_shapes(len); +nnvm::DTypeVector arg_dtypes(len); +StorageTypeVector arg_stypes(len); +std::vector in_arg_ctxes(len); +std::vector aux_state_ctxes; +for (mx_uint i = 0; i < len; i++) { + const auto &in_arg = *(in_args_ptr[i]); + arg_shapes.push_back(in_arg.shape()); + arg_dtypes.push_back(in_arg.dtype()); + arg_stypes.push_back(in_arg.storage_type()); + in_arg_ctxes[i] = in_arg.ctx(); +} +orig_g = common::InferForwardAttrs(orig_g, arg_shapes, arg_dtypes, arg_stypes, default_ctx, + ctx_map, in_arg_ctxes, aux_state_ctxes, true); + } + std::vector> options_map; + for (mx_uint i = 0; i < num_options; ++i) { +options_map.emplace_back(keys[i], vals[i]); + } + const auto backend = mxnet::op::SubgraphBackendRegistry::Get()->GetSubgraphBackend(backend_name); + const auto& subgraph_prop_list = backend->GetSubgraphProperties(); + for (auto property : subgraph_prop_list) { +nnvm::Graph g = orig_g; +property->SetAttr("graph", g); Review comment: dont think we need this if we're going to pass it into the prePartition function This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
larroy commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522189734 Marco. I understand this is a bug in the vs compiler version we use. I don't think this deprecates 32 bit support as the target arch doesn't change. We are compiling for 32 bit in arm7 right? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] DickJC123 commented on a change in pull request #15922: Refactor for windows CI 'out of heap space' errors
DickJC123 commented on a change in pull request #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#discussion_r314925116 ## File path: src/operator/tensor/broadcast_reduce_op.h ## @@ -1580,6 +1581,20 @@ void PickOpBackward(const nnvm::NodeAttrs& attrs, }); } +inline std::string get_reduce_axes_description(const std::string& op_name, int line) { Review comment: The short answer is that the function was formerly in broadcast_reduce_op_value.cc (with the inline keyword) and I copied it intact to broadcast_reduce_op.h. Also, removing `inline` breaks the compile with a 'multiple definitions' error. One shouldn't put the definition of a stand-alone function in a header file that's part of multiple translation units without the inline keyword. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API
mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314924282 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,41 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimizeFor(self, backend, ctx=None, args=None, **kwargs): Review comment: changed function name This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API
mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314924239 ## File path: src/operator/subgraph/subgraph_property.h ## @@ -221,6 +221,10 @@ class SubgraphProperty { return nullptr; } + virtual void PrePartition() {} Review comment: added graph and options_map as parameters This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API
mseth10 commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314924203 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,12 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimizeFor(self, backend): Review comment: added optional arguments This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ptrendx commented on a change in pull request #15545: Softmax fwd optimization for GPU
ptrendx commented on a change in pull request #15545: Softmax fwd optimization for GPU URL: https://github.com/apache/incubator-mxnet/pull/15545#discussion_r314919845 ## File path: src/common/cuda_utils.h ## @@ -615,6 +624,28 @@ __device__ inline DType ldg(const DType* address) { return *address; #endif } -#endif + Review comment: No, they don't. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
larroy commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522179936 Thanks for the clarification. I already approved. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on a change in pull request #15922: Refactor for windows CI 'out of heap space' errors
larroy commented on a change in pull request #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#discussion_r314917817 ## File path: src/operator/tensor/broadcast_reduce_op.h ## @@ -1580,6 +1581,20 @@ void PickOpBackward(const nnvm::NodeAttrs& attrs, }); } +inline std::string get_reduce_axes_description(const std::string& op_name, int line) { Review comment: why do you want to inline this? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #13736: gluon.utils.clip_global_norm/nd.dot with fp16 throws fatal message
ChaiBapchya commented on issue #13736: gluon.utils.clip_global_norm/nd.dot with fp16 throws fatal message URL: https://github.com/apache/incubator-mxnet/issues/13736#issuecomment-522175676 Fixed by https://github.com/apache/incubator-mxnet/pull/14102 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #15935: unused -mf16c compiler flag during android build
larroy commented on issue #15935: unused -mf16c compiler flag during android build URL: https://github.com/apache/incubator-mxnet/issues/15935#issuecomment-522175347 @mxnet-label-bot add [Call for Contribution] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #15935: unused -mf16c compiler flag during android build
larroy commented on issue #15935: unused -mf16c compiler flag during android build URL: https://github.com/apache/incubator-mxnet/issues/15935#issuecomment-522174966 @mxnet-label-bot add [Android, Build,Good First Issue] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15935: unused -mf16c compiler flag during android build
mxnet-label-bot commented on issue #15935: unused -mf16c compiler flag during android build URL: https://github.com/apache/incubator-mxnet/issues/15935#issuecomment-522174865 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: Build This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy opened a new issue #15935: unused -mf16c compiler flag during android build
larroy opened a new issue #15935: unused -mf16c compiler flag during android build URL: https://github.com/apache/incubator-mxnet/issues/15935 ``` ci/build.py -p android_armv7 [...] [98/290] ccache /usr/arm-linux-androideabi/bin/arm-linux-androideabi-clang++ -DDMLC_USE_CXX11 -DDMLC_USE_CXX11=1 -DMSHADOW_INT64_TENSOR_SIZE=0 -DMSHADOW_IN_CXX11 -DMSHADOW_USE_CBLAS=1 -DMSHADOW_USE_CUDA=0 -DMSHADOW_USE_MKL=0 -DMSHADOW_USE_SSE=0 -DMXNET_USE_BLAS_OPEN=1 -DMXNET_USE_NCCL=0 -DMXNET_USE_OPENCV=0 -DMXNET_USE_SIGNAL_HANDLER=1 -DNDEBUG=1 -I/work/mxnet/include -I/work/mxnet/src -isystem /usr/arm-linux-androideabi/include -I/work/mxnet/3rdparty/mshadow -I/work/mxnet/3rdparty/nvidia_cub -I/work/mxnet/3rdparty/tvm/nnvm/include -I/work/mxnet/3rdparty/tvm/include -I/work/mxnet/3rdparty/dmlc-core/include -I/work/mxnet/3rdparty/dlpack/include -D_GLIBCXX_ASSERTIONS -mf16c -Wall -Wno-unknown-pragmas -Wno-sign-compare -Wno-braced-scalar-init -O3 -g -std=c++11 -mf16c -std=c++0x -O2 -g -DNDEBUG -fPIC -MD -MT CMakeFiles/mxnet_static.dir/src/c_api/c_api_function.cc.o -MF CMakeFiles/mxnet_static.dir/src/c_api/c_api_function.cc.o.d -o CMakeFiles/mxnet_static.dir/src/c_api/c_api_function.cc.o -c /work/mxnet/src/c_api/c_api_function.cc clang60++: warning: argument unused during compilation: '-mf16c' [-Wunused-command-line-argument] clang60++: warning: argument unused during compilation: '-mf16c' [-Wunused-command-line-argument] ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] DickJC123 commented on issue #15922: Refactor for windows CI 'out of heap space' errors
DickJC123 commented on issue #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#issuecomment-522169267 Understandable. It's really only a lot of cut-and-paste, but I'll reach out to someone else. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] marcoabreu commented on issue #15922: Refactor for windows CI 'out of heap space' errors
marcoabreu commented on issue #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#issuecomment-522168628 I don't feel comfortable reviewing such an amount of c++ code and also a bit time constrained due to my business trip. So it would be great if somebody else could weigh in here This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] DickJC123 commented on issue #15922: Refactor for windows CI 'out of heap space' errors
DickJC123 commented on issue #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#issuecomment-52216 Hey, can you weigh in on this PR? With the bundled speed improvement of test_shuffle, it's a major step in stabilizing the CI. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] marcoabreu commented on issue #15922: Refactor for windows CI 'out of heap space' errors
marcoabreu commented on issue #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#issuecomment-522167346 Since October last year our CI system is unable to reliably communicate with GitHub. This results in lost status updates. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15925: [CI] illegal memory access
ChaiBapchya commented on issue #15925: [CI] illegal memory access URL: https://github.com/apache/incubator-mxnet/issues/15925#issuecomment-522167131 Another one same PR http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Funix-gpu/detail/PR-15736/13/pipeline/316 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] apeforest commented on issue #15703: Storage manager / memory usage regression in v1.5
apeforest commented on issue #15703: Storage manager / memory usage regression in v1.5 URL: https://github.com/apache/incubator-mxnet/issues/15703#issuecomment-522167234 Further narrowed it down to topk operator. There is some implementation of TopKImpl that did not allocate correct amount of GPU memory. Working on a PR now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy edited a comment on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
larroy edited a comment on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522162803 @DickJC123 I looked at laop tests, one root cause is the multiplication by a "random projection" in checking the numeric gradient done in the test, this initial multiplication creates an introduction of fp error. I had a look at this and tried to address, but I then had failures in other tests and run ouf of time allocated for this. https://github.com/apache/incubator-mxnet/blob/39bf4e062bac4fc23d7c8591005b99592e42ab25/python/mxnet/test_utils.py#L921 https://github.com/apache/incubator-mxnet/pull/15770 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
larroy commented on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522162803 @DickJC123 I looked at laop tests, one root cause is the multiplication by a "random projection" done in the test, this initial multiplication creates an introduction of fp error. I had a look at this and tried to address, but I then had failures in other tests and run ouf of time allocated for this. https://github.com/apache/incubator-mxnet/pull/15770 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy edited a comment on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'.
larroy edited a comment on issue #15882: Move Windows CI build to a 64-bit toolchain to fix 'out of heap space'. URL: https://github.com/apache/incubator-mxnet/pull/15882#issuecomment-522162803 @DickJC123 I looked at laop tests, one root cause is the multiplication by a "random projection" in checking the numeric gradient done in the test, this initial multiplication creates an introduction of fp error. I had a look at this and tried to address, but I then had failures in other tests and run ouf of time allocated for this. https://github.com/apache/incubator-mxnet/pull/15770 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15934: [CI] Validation Website build failure
ChaiBapchya commented on issue #15934: [CI] Validation Website build failure URL: https://github.com/apache/incubator-mxnet/issues/15934#issuecomment-522162130 @aaronmarkham any idea? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15934: [CI] Validation Website build failure
ChaiBapchya commented on issue #15934: [CI] Validation Website build failure URL: https://github.com/apache/incubator-mxnet/issues/15934#issuecomment-522162097 @mxnet-label-bot add [CI] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15934: [CI] Validation Website build failure
mxnet-label-bot commented on issue #15934: [CI] Validation Website build failure URL: https://github.com/apache/incubator-mxnet/issues/15934#issuecomment-522162019 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: CI, Doc This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya opened a new issue #15934: [CI] Validation Website build failure
ChaiBapchya opened a new issue #15934: [CI] Validation Website build failure URL: https://github.com/apache/incubator-mxnet/issues/15934 #15761 http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwebsite/detail/PR-15761/7/pipeline/ ``` make[1]: Entering directory '/work/mxnet/3rdparty/ps-lite' rm -rf zeromq-4.1.4.tar.gz zeromq-4.1.4 wget https://raw.githubusercontent.com/mli/deps/master/build/zeromq-4.1.4.tar.gz && tar --no-same-owner -zxf zeromq-4.1.4.tar.gz --2019-08-16 15:42:52-- https://raw.githubusercontent.com/mli/deps/master/build/zeromq-4.1.4.tar.gz Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.40.133 Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.40.133|:443... connected. rm -rf protobuf-2.5.0.tar.gz protobuf-2.5.0 wget https://raw.githubusercontent.com/mli/deps/master/build/protobuf-2.5.0.tar.gz && tar --no-same-owner -zxf protobuf-2.5.0.tar.gz --2019-08-16 15:42:52-- https://raw.githubusercontent.com/mli/deps/master/build/protobuf-2.5.0.tar.gz cd /work/mxnet/3rdparty/dmlc-core; make libdmlc.a USE_SSE=1 config=/work/mxnet/make/config.mk; cd /work/mxnet Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.40.133 make[1]: Entering directory '/work/mxnet/3rdparty/dmlc-core' Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.40.133|:443... HTTP request sent, awaiting response... connected. 200 OK Length: 1400012 (1.3M) [application/octet-stream] Saving to: 'zeromq-4.1.4.tar.gz' cd /work/mxnet/3rdparty/tvm/nnvm; make lib/libnnvm.a DMLC_CORE_PATH=/work/mxnet/3rdparty/dmlc-core; cd /work/mxnet make[1]: Entering directory '/work/mxnet/3rdparty/tvm/nnvm' g++ -std=c++11 -Wall -O2 -Iinclude -fPIC -I../include -I../3rdparty/dlpack/include -I../3rdparty/HalideIR/src -I../topi/include -I/work/mxnet/3rdparty/dmlc-core/include -MM -MT build/src/c_api/c_api_symbolic.o src/c_api/c_api_symbolic.cc >build/src/c_api/c_api_symbolic.d 0K .. .. .. .Unable to establish SSL connection. make/deps.mk:11: recipe for target '/work/mxnet/deps/include/google/protobuf/message.h' failed make[1]: *** [/work/mxnet/deps/include/google/protobuf/message.h] Error 4 make[1]: *** Waiting for unfinished jobs make[1]: Leaving directory '/work/mxnet/3rdparty/ps-lite' make: *** [PSLITE] Error 2 Makefile:594: recipe for target 'PSLITE' failed build.py: 2019-08-16 15:51:45,033Z INFO Waiting for status of container 31da00366c99 for 600 s. build.py: 2019-08-16 15:51:45,252Z INFO Container exit status: {'Error': None, 'StatusCode': 2} build.py: 2019-08-16 15:51:45,253Z ERROR Container exited with an error 😞 build.py: 2019-08-16 15:51:45,253Z INFO Executed command for reproduction: ci/build.py --docker-registry mxnetci --platform ubuntu_cpu --docker-build-retries 3 --shm-size 500m /work/runtime_functions.sh build_ubuntu_cpu_openblas ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #15285: Graph dumper
larroy commented on issue #15285: Graph dumper URL: https://github.com/apache/incubator-mxnet/pull/15285#issuecomment-522161736 Question, is there interest to get this merged if I address @ptrendx comments? The action points that I have in mind are just deduplicating node names. I have limited bandwidth right now to keep iterating in this pr. Could we summarize which changes are required to get this in or should I close the PR? Thanks. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on a change in pull request #14779: Fully connected, higher order grad
larroy commented on a change in pull request #14779: Fully connected, higher order grad URL: https://github.com/apache/incubator-mxnet/pull/14779#discussion_r314900617 ## File path: tests/python/unittest/test_higher_order_grad.py ## @@ -210,6 +217,168 @@ def check_second_order_unary(x, op, grad_grad_op, rtol=None, atol=None): x.grad.asnumpy(), rtol=rtol, atol=atol) +def arange_shape_like(y): +shape = y.shape +nelems = reduce(mul, shape) +x = nd.arange(nelems).reshape(shape) +return x + + +class NDArrayGenerator(object): +def __init__(self, dim, startdim=1): +self.dim = dim +self.curdim = startdim + +def __iter__(self): +return self + +@staticmethod +def gen(dimensions): +shape = rand_shape_nd(dimensions, 4) +nelems = reduce(mul, shape) +x = nd.arange(nelems).reshape(shape) +return x + +def next(self): +return self.__next__() + +def __next__(self): +if self.curdim > self.dim: +raise StopIteration +x = NDArrayGenerator.gen(self.curdim) +self.curdim += 1 +return x + + +def flatten2d_right(x): +s_0 = x.shape[0] +s_1 = reduce(mul, x.shape[1:]) +return x.reshape((s_0, s_1)) + + +def flatten2d_left(x): +s_0 = reduce(mul, x.shape[:-1]) +s_1 = x.shape[-1] +return x.reshape((s_0, s_1)) + + +@with_seed() +def test_dense_backward_flatten(): +print("2nd order gradient for Fully Connected, flatten=True") +for x in NDArrayGenerator(4,2): +hidden = random.randrange(1, 4) +net = gluon.nn.Sequential() +with net.name_scope(): +net.add(gluon.nn.Dense(hidden, flatten=True)) +net.initialize(mxnet.initializer.Constant(.5)) +x.attach_grad() +with autograd.record(): +y = net.forward(x) +o_y = arange_shape_like(y) # head gradient of y +params = [p.data() for p in net.collect_params().values()] +w = params[0] +b = params[1] +print("Checking y ({}) = x({}) * w^T({}) + b({})".format(y.shape, x.shape, w.shape, b.shape)) +x_grad = autograd.grad(heads=y, variables=x, head_grads=o_y, + create_graph=True, retain_graph=True)[0] +o_x_grad = arange_shape_like(x_grad) +w_grad_grad = autograd.grad(heads=x_grad, variables=w, +head_grads=o_x_grad, create_graph=False)[0] +w_grad = autograd.grad(heads=y, variables=w, head_grads=o_y, + create_graph=True, retain_graph=True)[0] +o_w_grad = arange_shape_like(w_grad) +x_grad_grad = autograd.grad(heads=w_grad, variables=x, +head_grads=o_w_grad, create_graph=False)[0] +# Expected results +w_grad_e = nd.dot(o_y, x, transpose_a=True) +w_grad_grad_e = nd.dot(o_y, o_x_grad, transpose_a=True) +x_grad_e = nd.dot(o_y, w) +x_grad_grad_e = nd.dot(o_y, o_w_grad) +ok_(w_grad.shape == w.shape) +ok_(w_grad_grad.shape == w.shape) +ok_(x_grad.shape == x.shape) +ok_(x_grad_grad.shape == x.shape) +#print("x_grad:\n{}".format(x_grad)); +#print("x_grad_e:\n{}".format(x_grad_e)); +#print("x_grad_grad:\n{}".format(x_grad_grad)); +#print("x_grad_grad_e:\n{}".format(x_grad_grad_e)); +#print("w_grad:\n{}".format(w_grad)); +#print("w_grad_e:\n{}".format(w_grad_e)); +#print("w_grad_grad:\n{}".format(w_grad_grad)); +#print("w_grad_grad_e:\n{}".format(w_grad_grad_e)); +w_grad_check = same(flatten2d_right(w_grad), flatten2d_right(w_grad_e)) +w_grad_grad_check = same(flatten2d_right(w_grad_grad), flatten2d_right(w_grad_grad_e)) +x_grad_check = same(flatten2d_right(x_grad), flatten2d_right(x_grad_e)) +x_grad_grad_check = same(flatten2d_right(x_grad_grad), flatten2d_right(x_grad_grad_e)) +ok_(x_grad_check) +ok_(w_grad_check) +ok_(x_grad_grad_check) +ok_(w_grad_grad_check) + +@with_seed() +def test_dense_backward_no_flatten(): +print("2nd order gradient for Fully Connected, flatten=False") +for x in NDArrayGenerator(5,3): +hidden = random.randrange(1, 4) +net = gluon.nn.Sequential() +with net.name_scope(): +net.add(gluon.nn.Dense(hidden, flatten=False)) +net.initialize(mxnet.initializer.Constant(.5)) +x.attach_grad() +with autograd.record(): +y = net.forward(x) +o_y = arange_shape_like(y) # head gradient of y +params = [p.data() for p in net.collect_params().values()] +w = params[0] +b = params[1] +print("Checking y ({}) = x({}) * w^T({}) + b({})".format(y.shape, x.shape, w.shape, b.shape)) +x_grad = autogra
[GitHub] [incubator-mxnet] apeforest commented on a change in pull request #15285: Graph dumper
apeforest commented on a change in pull request #15285: Graph dumper URL: https://github.com/apache/incubator-mxnet/pull/15285#discussion_r314895584 ## File path: include/mxnet/imperative.h ## @@ -160,8 +160,8 @@ class Imperative { private: friend class NDArray; - /*! \brief make constructor protected. */ - Imperative() { + Imperative() : +backward_graph_dump_enabled_(dmlc::GetEnv("MXNET_BACKWARD_GRAPH_DUMP", false)) { Review comment: If this is only used for debugging, I am not sure if adding a field to a production class is a good idea. Can we just pull the environment variable instead? It's not impacting production performance and not going to be used in multithreaded environment. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15933: indexing op does not report out of bound indices
mxnet-label-bot commented on issue #15933: indexing op does not report out of bound indices URL: https://github.com/apache/incubator-mxnet/issues/15933#issuecomment-522155910 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: Bug This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] eric-haibin-lin opened a new issue #15933: indexing op does not report out of bound indices
eric-haibin-lin opened a new issue #15933: indexing op does not report out of bound indices URL: https://github.com/apache/incubator-mxnet/issues/15933 out of bound error is reported in numpy: ``` >>> import numpy as np >>> np.ones((1,))[[10]] Traceback (most recent call last): File "", line 1, in IndexError: index 10 is out of bounds for axis 0 with size 1 >>> ``` but MXNet does not: ``` >>> import mxnet as mx >>> mx.nd.ones((1,))[[10]] [6.4e-44] ``` We can reuse the kernel in ` nd.take(mode='raise') ` to report such an error. @reminisce @haojin2 @hhexiy This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zachgk commented on issue #15932: Issues with SequenceReverse operator backward pass
zachgk commented on issue #15932: Issues with SequenceReverse operator backward pass URL: https://github.com/apache/incubator-mxnet/issues/15932#issuecomment-522154077 @apeforest This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zachgk commented on issue #15929: ./build.py -p ubuntu_tpu_tensorrt fails with error
zachgk commented on issue #15929: ./build.py -p ubuntu_tpu_tensorrt fails with error URL: https://github.com/apache/incubator-mxnet/issues/15929#issuecomment-522153342 ping @larroy This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. marcoabreu pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new d6ee198 Bump the publish timestamp. d6ee198 is described below commit d6ee1984be269fb30bb85871c2db1d4c3997054c Author: mxnet-ci AuthorDate: Fri Aug 16 21:04:58 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..a9fbd04 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Fri Aug 16 21:04:58 UTC 2019
[GitHub] [incubator-mxnet] aaronmarkham commented on issue #15884: [WIP] New Website: New Docs [1/3]
aaronmarkham commented on issue #15884: [WIP] New Website: New Docs [1/3] URL: https://github.com/apache/incubator-mxnet/pull/15884#issuecomment-522149243 I fixed one license issue, but now another popped up. I think that RAT checker just bails once it finds one issue instead of telling you the totality of files that need licenses. I'm going to ignore this check for now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] rondogency commented on a change in pull request #15886: [WIP] Graph Partition API
rondogency commented on a change in pull request #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#discussion_r314875614 ## File path: python/mxnet/symbol/symbol.py ## @@ -1437,6 +1437,41 @@ def _gen_atomic_symbol(self): return Symbol(handle) +def optimizeFor(self, backend, ctx=None, args=None, **kwargs): Review comment: should be optimize_for, since we don't do camelCase in api functions This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. marcoabreu pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new 72389c2 Bump the publish timestamp. 72389c2 is described below commit 72389c275c63d5f16611e937e049fc5f5f4c305c Author: mxnet-ci AuthorDate: Fri Aug 16 19:29:21 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..96712e3 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Fri Aug 16 19:29:21 UTC 2019
[GitHub] [incubator-mxnet] anirudh2290 opened a new issue #15932: Issues with SequenceReverse operator backward pass
anirudh2290 opened a new issue #15932: Issues with SequenceReverse operator backward pass URL: https://github.com/apache/incubator-mxnet/issues/15932 The behavior of SequenceReverse operator backward pass is different before and after #11742 https://github.com/apache/incubator-mxnet/pull/11742 The grad of sequence length has different values before and after this PR, causing issues for customers. My guess is this is because of an undefined behavior which is hit when casting from float to int or unsigned. The operator needs to be revisited to correct this issue. Below is a reproducible script. The dseq ndarrays will be different for mxnet version before and after #11742 ``` import mxnet as mx import numpy as np seq_len = mx.nd.load("seq_reverse_seq_len.params") inp = mx.nd.load("seq_reverse_data.params") ograd = mx.nd.load("seq_reverse_ograd.params") seq_len = seq_len[0] inp = inp[0] ograd = ograd[0] seq_len = seq_len.copyto(mx.gpu(0)) inp = inp.copyto(mx.gpu(0)) ograd = ograd.copyto(mx.gpu(0)) inp.attach_grad() seq_len.attach_grad() with mx.autograd.record(): out = mx.nd.SequenceReverse(inp, sequence_length=seq_len, use_sequence_length=True) out.backward(out_grad=ograd) dinp = inp.grad.asnumpy() dseq = seq_len.grad.asnumpy() mx.nd.waitall() ``` Below are the params files required: [seq_rev_params.zip](https://github.com/apache/incubator-mxnet/files/3510716/seq_rev_params.zip) This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] anirudh2290 commented on issue #15932: Issues with SequenceReverse operator backward pass
anirudh2290 commented on issue #15932: Issues with SequenceReverse operator backward pass URL: https://github.com/apache/incubator-mxnet/issues/15932#issuecomment-522110552 @access2rohit This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15932: Issues with SequenceReverse operator backward pass
mxnet-label-bot commented on issue #15932: Issues with SequenceReverse operator backward pass URL: https://github.com/apache/incubator-mxnet/issues/15932#issuecomment-522110457 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: Bug This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy edited a comment on issue #14940: Fix warnings
larroy edited a comment on issue #14940: Fix warnings URL: https://github.com/apache/incubator-mxnet/pull/14940#issuecomment-522106028 @marcoabreu This is a bit of a rant but why this doesn't apply to huge PRs like this? https://github.com/apache/incubator-mxnet/pull/15581/files instead is used as nitpicking for small PRs from external collaborators. If we want PRs which addresses isolated concerns, good, but let's be consistent and not have double standards. My point is that I think is not a good use of my time or the reviewers time to act as a linter. If some committers are so passionate about whitespace they should improve the linter and code analysis tools instead of alienating contributors. I find ok to be reminded of some whitespace issue here and there, but this is a shallow review which comes late. I think comitters should take into consideration volunteers time and effort a bit more instead of making things more difficult. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on issue #14940: Fix warnings
larroy commented on issue #14940: Fix warnings URL: https://github.com/apache/incubator-mxnet/pull/14940#issuecomment-522106028 @marcoabreu why this doesn't apply to huge PRs like this? https://github.com/apache/incubator-mxnet/pull/15581/files instead is used as nitpicking for small PRs from external collaborators. If we want PRs which addresses isolated concerns, good, but let's be consistent and not have double standards. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] Zha0q1 commented on issue #15132: Profiler API Enhancements
Zha0q1 commented on issue #15132: Profiler API Enhancements URL: https://github.com/apache/incubator-mxnet/pull/15132#issuecomment-522103033 > The `ascending` argument is great, but why is it missing the most important sort criteria "total time"? That's usually THE first thing people look at - what operator takes most of the time in a network. This is definitely a must-have for the 'sort-by' argument .. Sure. I will create a pr and add "total_time" soon. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] eric-haibin-lin closed issue #13736: gluon.utils.clip_global_norm/nd.dot with fp16 throws fatal message
eric-haibin-lin closed issue #13736: gluon.utils.clip_global_norm/nd.dot with fp16 throws fatal message URL: https://github.com/apache/incubator-mxnet/issues/13736 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #13736: gluon.utils.clip_global_norm/nd.dot with fp16 throws fatal message
ChaiBapchya commented on issue #13736: gluon.utils.clip_global_norm/nd.dot with fp16 throws fatal message URL: https://github.com/apache/incubator-mxnet/issues/13736#issuecomment-522102069 Unable to reproduce it With latest master ``` Python 3.6.5 |Anaconda, Inc.| (default, Apr 29 2018, 16:14:56) >>> import mxnet as mx >>> b = mx.nd.ones((10,), dtype='float16', ctx=mx.gpu()) [18:10:21] ../src/base.cc:84: Upgrade advisory: this mxnet has been built against cuDNN lib version 7401, which is older than the oldest version tested by CI (7600). Set MXNET_CUDNN_LIB_CHECKING=0 to quiet this warning. >>> mx.nd.dot(b,b) [10.] ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on issue #15931: TBlob bug about dltensor
hzfan commented on issue #15931: TBlob bug about dltensor URL: https://github.com/apache/incubator-mxnet/issues/15931#issuecomment-522101037 @mxnet-label-bot add [bug] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky edited a comment on issue #15886: [WIP] Graph Partition API
samskalicky edited a comment on issue #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#issuecomment-522096949 @mseth10 we also need to add the ability to reject creating a subgraph in the buildsubgraph.cc file: https://github.com/apache/incubator-mxnet/blob/e98fea3165670157090f2a2f644890452443803c/src/operator/subgraph/build_subgraph.cc#L574 We want to enable the subgraph property to return a null node to say reject creating a subgraph. The partitioning pass has a decycle feature that may removed nodes that were selected when calling the select function on the subgraph property. So the subgraph may be smaller than anticipated, and we want the ability to not create subgraphs based on some criteria in the subgraph property (ie. subgraph size too small). This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] eric-haibin-lin commented on issue #15132: Profiler API Enhancements
eric-haibin-lin commented on issue #15132: Profiler API Enhancements URL: https://github.com/apache/incubator-mxnet/pull/15132#issuecomment-522100984 The `ascending` argument is great, but why is it missing the most important sort criteria "total time"? That's usually THE first thing people look at - what operator takes most of the time in a network. This is definitely a must-have for the 'sort-by' argument .. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky edited a comment on issue #15886: [WIP] Graph Partition API
samskalicky edited a comment on issue #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#issuecomment-522067381 @ZhennanQin Thanks for your comments, we'll try and clarify our thought process. Apologies for the lack of clarity in this PR. You are correct in the current status of the code that @mseth10 has committed so far today. The PR is still WIP and more changes are coming, lets not make decisions on what is there today. Instead lets focus on the items i mentioned before (they will get into the PR description before we attempt to merge, I promise! :-D). We will add more arguments to the optimizeFor API in the coming commits. We need to add arguments to enable us to to do shape/type propagation prior to partitioning (so we can use shape/type info to select ops in subgraph properties), and we want to accept arbitrary options that we pass to the subgraph property for further configuration (ie. blacklisted ops). These new API changes will not be compatible with the current get_backend_symbol API. Since MXNet maintains semantic versioning for minor releases we cannot change the API yet. So instead we'll create another API (optimizeFor) along side get_backend_symbol for now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky edited a comment on issue #15886: [WIP] Graph Partition API
samskalicky edited a comment on issue #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#issuecomment-522067381 @ZhennanQin Thanks for your comments, we'll try and clarify our thought process. Apologies for the lack of clarity in this PR. You are correct in the current status of the code that @mseth10 has committed so far today. The PR is still WIP and more changes are coming, lets not make decisions on what is there today. Instead lets focus on the items i mentioned before (they will get into the PR description before we attempt to merge, I promise! :-D). We will add more arguments to the optimizeFor API in the coming commits. We need to add arguments to enable us to to do shape/type propagation prior to partitioning (so we can use shape/type info to select ops in subgraph properties), and we want to accept arbitrary options that we pass to the subgraph property for further configuration (ie. blacklisted ops). These new API changes will not be compatible with the current get_backend_symbol API. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15931: TBlob bug about dltensor
mxnet-label-bot commented on issue #15931: TBlob bug about dltensor URL: https://github.com/apache/incubator-mxnet/issues/15931#issuecomment-522099738 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: Bug This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] hzfan opened a new issue #15931: TBlob bug about dltensor
hzfan opened a new issue #15931: TBlob bug about dltensor URL: https://github.com/apache/incubator-mxnet/issues/15931 ## Description TBlob does not disable/overload the default copy constructor/assignment, so the default one is used. This results in shallow copy of dltensor_ (which is a field of type DLTensor in TBlob, see [here](https://github.com/apache/incubator-mxnet/blob/5a4c01bac9afd4e75227a0b4b1231bceffb204df/include/mxnet/tensor_blob.h#L415)) and memory leak. ## Environment info (Required) Python 3.7.3 Built from source (master at 5a4c01bac9afd4e75227a0b4b1231bceffb204df) ## Minimum reproducible example To reproduce this error, I made a minor change to the function [NumpyDotForward](https://github.com/apache/incubator-mxnet/blob/5a4c01bac9afd4e75227a0b4b1231bceffb204df/src/operator/numpy/np_dot-inl.h#L39) (in src/operator/numpy/np_dot-inl.h) for illustration. Here is the function after my modification. I modified one line, and added two lines (denoted by comments): ``` template inline void NumpyDotForward(const nnvm::NodeAttrs& attrs, const OpContext& ctx, const std::vector& inputs, const std::vector& req, const std::vector& outputs) { using namespace mshadow; using namespace mxnet_op; CHECK_EQ(inputs.size(), 2U); CHECK_EQ(outputs.size(), 1U); const TBlob& a = inputs[0]; const TBlob& b = inputs[1]; // const TBlob& out = outputs[0]; TBlob out = outputs[0]; // changed by me const mxnet::TShape a_shape = a.shape_; const mxnet::TShape b_shape = b.shape_; out = out.reshape(out.shape_); // added by me out = TBlob(out.dltensor()); // added MSHADOW_REAL_TYPE_SWITCH(out.type_flag_, DType, { if (b_shape.ndim() < 3) { // Case 1, 2, 3, 4, 5: a is N-D array (N >= 1) and b is vector or matrix, sum product //over the last axis of a and the first axis of b TensordotIntAxesImpl(1, ctx, a, b, out, req[0]); } else { // Case 3, 5.5: a is N-D array and b is M-D array (M > 2), sum product over the last axis // of a and the 2nd-to-last axis of b const Tuple a_axes_summed({a_shape.ndim() - 1}); const Tuple b_axes_summed({b_shape.ndim() - 2}); TensordotImpl(a_axes_summed, b_axes_summed, ctx, a, b, out, req); } }); } ``` ## Steps to reproduce 1. replace NumpyDotForward with the above one 2. build 3. run the following ``` from mxnet import np a = np.array([[1, 2, 3], [4, 5, 6]]) b = np.array([[1, 1], [1, 1], [1, 1]]) np.dot(a, b) ``` The expected result is ``` array([[ 6., 6.], [15., 15.]]) ``` But the real result is ``` array([[0., 0.], [0., 0.]]) ``` ## The cause of this problem TBlob.dltensor_.shape is a pointer. When TBlob b is assigned to TBlob a, the pointer gets shallow copied: ``` a.dltensor_.shape = b.dltensor_.shape ``` But b.dltensor_.shape points to b.shape_.data(). So when b is a temporary variable (like the return value of TBlob.reshape()), b.shape_.data() gets destroyed after the function returns. Now a.dltensor_.shape points to invalid memory. ## Possible solutions (IMO) - disable default assignment/copy constructor (declare them with private) - overload them and use SetDLTensor to avoid shallow copy Thank @yzhliu @reminisce @haojin2 for help. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on issue #15886: [WIP] Graph Partition API
samskalicky commented on issue #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#issuecomment-522096949 @mseth10 we also need to add the ability to reject creating a subgraph in the buildsubgraph.cc file: https://github.com/apache/incubator-mxnet/blob/e98fea3165670157090f2a2f644890452443803c/src/operator/subgraph/build_subgraph.cc#L574 We want to enable the subgraph property to return a null node to say reject This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] kohr-h commented on issue #13143: [MXNET-1206] Support NDArray indexing with None and Ellipsis
kohr-h commented on issue #13143: [MXNET-1206] Support NDArray indexing with None and Ellipsis URL: https://github.com/apache/incubator-mxnet/pull/13143#issuecomment-522081900 > @kohr-h Thanks for the response. Since returning views is a very frequent operation in many dataset loaders in MXNet, we have to keep the performance on a par with the previous version. @zoeygxy will keep working on improving the performance of the cases where views are returned. Although such special handling may lead code to look a little bit disorganized, the performance gain is still worth the effort because we don't want to introduce performance regression in training jobs. @reminisce Makes sense as a trade-off. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] kohr-h commented on issue #13143: [MXNET-1206] Support NDArray indexing with None and Ellipsis
kohr-h commented on issue #13143: [MXNET-1206] Support NDArray indexing with None and Ellipsis URL: https://github.com/apache/incubator-mxnet/pull/13143#issuecomment-522081223 @DickJC123 It's part of fe6336df4064f3086443d5abb7535fd549dd3a48 so I'll bounce the question to @reminisce. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] aaronmarkham commented on issue #15649: Installation Fails on Debian Stable Machine
aaronmarkham commented on issue #15649: Installation Fails on Debian Stable Machine URL: https://github.com/apache/incubator-mxnet/issues/15649#issuecomment-522067962 @larry77 Did changing the version work for you? Since we're tracking the version support in #14967, I'm going to close this. Please reopen if you're still having issues. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] aaronmarkham closed issue #15649: Installation Fails on Debian Stable Machine
aaronmarkham closed issue #15649: Installation Fails on Debian Stable Machine URL: https://github.com/apache/incubator-mxnet/issues/15649 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky edited a comment on issue #15886: [WIP] Graph Partition API
samskalicky edited a comment on issue #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#issuecomment-522067381 @ZhennanQin Thanks for your comments, we'll try and clarify our thought process. Apologies for the lack of clarity in this PR. You are correct in the current status of the code that @mseth10 has committed so far today. The PR is still WIP and more changes are coming, lets not make decisions on what is there today. Instead lets focus on the items i mentioned before (they will get into the PR description before we attempt to merge, I promise! :-D). We will add more arguments to the optimizeFor API in the coming commits. We need to add arguments to enable us to to do shape/type propagation, and we want to accept arbitrary options that we pass to the subgraph property. These new API changes will not be compatible with the current get_backend_symbol API. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] samskalicky commented on issue #15886: [WIP] Graph Partition API
samskalicky commented on issue #15886: [WIP] Graph Partition API URL: https://github.com/apache/incubator-mxnet/pull/15886#issuecomment-522067381 Basically, I think sym.optimizeFor('MKLDNN') is the same as current sym.get_backend_symbol('MKLDNN'), so I don't see the reason to duplicate it. Maybe we can directly rename get_backend_symbol to optimizeFor? @ZhennanQin Thanks for your comments, we'll try and clarify our thought process. Apologies for the lack of clarity in this PR. You are correct in the current status of the code that @mseth10 has committed so far today. The PR is still WIP and more changes are coming, lets not make decisions on what is there today. Instead lets focus on the items i mentioned before (they will get into the PR description before we attempt to merge, I promise! :-D). We will add more arguments to the optimizeFor API in the coming commits. We need to add arguments to enable us to to do shape/type propagation, and we want to accept arbitrary options that we pass to the subgraph property. These new API changes will not be compatible with the current get_backend_symbol API. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ptrendx commented on issue #15928: [RFC] A faster version of Gamma sampling on GPU.
ptrendx commented on issue #15928: [RFC] A faster version of Gamma sampling on GPU. URL: https://github.com/apache/incubator-mxnet/issues/15928#issuecomment-522055756 Hi @xidulu. I did not look at the differences in the implementation of host-side vs device-side API for RNG in MXNet, but if they are comparable in terms of performance, a possible better approach would be something like this: - launch only as many blocks and threads as necessary to fill the GPU, each having their own RNG - use following pseudocode ``` while(my_sample_id < N_samples) { float rng = generate_next_rng(); bool accepted = ... // compute whether this rng value is accepted if (accepted) { // write the result my_sample_id = next_sample(); } } ``` There are 2 ways of implementing `next_sample` here - either by `atomicInc` on some global counter or just by adding the total number of threads (so every thread processes the same number of samples). The atomic approach is potentially faster (as with the static assignment you could end up hitting a corner case where 1 thread would still do a lot more work than the other threads), but is nondeterministic, so I think static assignment is preferable here. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet] branch v1.5.x updated: fix license issues (#15806) (#15860)
This is an automated email from the ASF dual-hosted git repository. taolv pushed a commit to branch v1.5.x in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git The following commit(s) were added to refs/heads/v1.5.x by this push: new 6f7fbc8 fix license issues (#15806) (#15860) 6f7fbc8 is described below commit 6f7fbc8d03b4ecc3846a286e888d0cea1b41c1e9 Author: Tao Lv AuthorDate: Fri Aug 16 23:30:37 2019 +0800 fix license issues (#15806) (#15860) Conflicts: 3rdparty/mshadow --- LICENSE | 355 ++- src/operator/special_functions-inl.h | 19 ++ 2 files changed, 325 insertions(+), 49 deletions(-) diff --git a/LICENSE b/LICENSE index 0673e0e..4532449 100644 --- a/LICENSE +++ b/LICENSE @@ -458,6 +458,37 @@ 10. ONNX Export module For details, see, python/mxnet/contrib/onnx/mx2onnx/LICENSE +This module is based on https://github.com/NVIDIA/mxnet_to_onnx/blob/master/mx2onnx_converter/# +The original code has the following license: + +# Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# * Redistributions of source code must retain the above copyright +#notice, this list of conditions and the following disclaimer. +# * Redistributions in binary form must reproduce the above copyright +#notice, this list of conditions and the following disclaimer in the +#documentation and/or other materials provided with the distribution. +# * Neither the name of NVIDIA CORPORATION nor the names of its +#contributors may be used to endorse or promote products derived +#from this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY +# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR +# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY +# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +Now it has been changed by Apache MXNet project. Modifications have the following license: + # Licensed to the Apache Software Foundation (ASF) under one # or more contributor license agreements. See the NOTICE file # distributed with this work for additional information @@ -657,56 +688,282 @@ === 14. FindJeMalloc.cmake -For details, see, cmake/Modules/FindJeMalloc.cmake - -Licensed to the Apache Software Foundation (ASF) under one -or more contributor license agreements. See the NOTICE file -distributed with this work for additional information -regarding copyright ownership. The ASF licenses this file -to you under the Apache License, Version 2.0 (the -"License"); you may not use this file except in compliance -with the License. You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, -software distributed under the License is distributed on an -"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -KIND, either express or implied. See the License for the -specific language governing permissions and limitations -under the License. - - -Copyright (c) 2014 Thomas Heller -Copyright (c) 2007-2012 Hartmut Kaiser -Copyright (c) 2010-2011 Matt Anderson -Copyright (c) 2011 Bryce Lelbach - -Distributed under the Boost Software License, Version 1.0. -Boost Software License - Version 1.0 - August 17th, 2003 - -Permission is hereby granted, free of charge, to any person or organization -obtaining a copy of the software and accompanying documentation covered by -this license (the "Software") to use, reproduce, display, distribute, -execute, and transmit the Software, and to prepare derivative works of the -Software, and to permit third-parties to whom the Software is furnished to -do so, all subject to the following: - -The copyright notices in the Software and this entire statement, including -the above license grant, this restriction and the following disclaimer, -must be included i
[GitHub] [incubator-mxnet] TaoLv merged pull request #15860: [v1.5.x] fix license issues (#15806)
TaoLv merged pull request #15860: [v1.5.x] fix license issues (#15806) URL: https://github.com/apache/incubator-mxnet/pull/15860 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] TaoLv commented on issue #15930: Fix dtype inference in arange_like operator
TaoLv commented on issue #15930: Fix dtype inference in arange_like operator URL: https://github.com/apache/incubator-mxnet/pull/15930#issuecomment-522049857 @eric-haibin-lin @fhieber This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] TaoLv opened a new pull request #15930: Fix dtype inference in arange_like operator
TaoLv opened a new pull request #15930: Fix dtype inference in arange_like operator URL: https://github.com/apache/incubator-mxnet/pull/15930 ## Description ## Remove the dtype argument from parameter structure and use ElemwiseType instead. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [ ] Feature1, tests, (and when applicable, API doc) - [ ] Feature2, tests, (and when applicable, API doc) ## Comments ## - If this change is a backward incompatible change, why must this change be made. - Interesting edge cases to note here This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on a change in pull request #15769: Benchmark doc fix
ChaiBapchya commented on a change in pull request #15769: Benchmark doc fix URL: https://github.com/apache/incubator-mxnet/pull/15769#discussion_r314770899 ## File path: benchmark/opperf/README.md ## @@ -47,7 +47,10 @@ Hence, in this utility, we will build the functionality to allow users and devel ## Prerequisites -Make sure to build the flavor of MXNet, for example - with/without MKL, with CUDA 9 or 10.1 etc., on which you would like to measure operator performance. Finally, you need to add path to your cloned MXNet repository to the PYTHONPATH. +Provided you have MXNet installed (any version >= 1.5.1), all you need to use opperf utility is to add path to your cloned MXNet repository to the PYTHONPATH. Review comment: In the note I have added link to how to install MXNet (page) instead of giving just 1 of the 10 different ways to install mxnet This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ElaineBao commented on issue #15870: update support MKLDNN BN conditions
ElaineBao commented on issue #15870: update support MKLDNN BN conditions URL: https://github.com/apache/incubator-mxnet/pull/15870#issuecomment-522048331 Seems something wrong on CI, retrigger it again. If it's still failed, will look into it. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15770: [WIP] Reduce numerical error on numerical gradient calculations
ChaiBapchya commented on issue #15770: [WIP] Reduce numerical error on numerical gradient calculations URL: https://github.com/apache/incubator-mxnet/pull/15770#issuecomment-522048180 @larroy Thanks for diving deep on this issue! If this solves the problem (adding instead of multiplying random matrix) would be great! Can you address merge conflicts and retrigger the CI? Also I skimmed through a few CI pipelines. Error seem to be related to this change.. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet] branch v1.5.x updated: prevent TRT_Logger to be destroyed before TRT engine (#14898) (#15877)
This is an automated email from the ASF dual-hosted git repository. taolv pushed a commit to branch v1.5.x in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git The following commit(s) were added to refs/heads/v1.5.x by this push: new bd2b5a2 prevent TRT_Logger to be destroyed before TRT engine (#14898) (#15877) bd2b5a2 is described below commit bd2b5a28e775004c504c8154a90884bf9c06cd9d Author: Kellen Sunderland AuthorDate: Fri Aug 16 08:20:32 2019 -0700 prevent TRT_Logger to be destroyed before TRT engine (#14898) (#15877) * prevent TRT_Logger to be destroyed before TRT engine * use unique_ptr for trt_logger/parser/engine/executor ownership * reduce line length for lint --- src/operator/subgraph/tensorrt/onnx_to_tensorrt.cc | 35 src/operator/subgraph/tensorrt/onnx_to_tensorrt.h | 66 ++ src/operator/subgraph/tensorrt/tensorrt-inl.h | 25 src/operator/subgraph/tensorrt/tensorrt.cc | 4 +- 4 files changed, 69 insertions(+), 61 deletions(-) diff --git a/src/operator/subgraph/tensorrt/onnx_to_tensorrt.cc b/src/operator/subgraph/tensorrt/onnx_to_tensorrt.cc index 7dbc54b..27f6da4 100644 --- a/src/operator/subgraph/tensorrt/onnx_to_tensorrt.cc +++ b/src/operator/subgraph/tensorrt/onnx_to_tensorrt.cc @@ -48,23 +48,6 @@ using std::endl; namespace onnx_to_tensorrt { -struct InferDeleter { - template -void operator()(T* obj) const { - if ( obj ) { -obj->destroy(); - } -} -}; - -template -inline std::shared_ptr InferObject(T* obj) { - if ( !obj ) { -throw std::runtime_error("Failed to create object"); - } - return std::shared_ptr(obj, InferDeleter()); -} - std::string onnx_ir_version_string(int64_t ir_version = onnx::IR_VERSION) { int onnx_ir_major = ir_version / 100; int onnx_ir_minor = ir_version % 100 / 1; @@ -83,7 +66,9 @@ void PrintVersion() { << NV_TENSORRT_PATCH << endl; } -std::tuple onnxToTrtCtx( +std::tuple, + unique_ptr, + std::unique_ptr > onnxToTrtCtx( const std::string& onnx_model, int32_t max_batch_size, size_t max_workspace_size, @@ -91,10 +76,10 @@ std::tuple onnxToTrtCtx( bool debug_builder) { GOOGLE_PROTOBUF_VERIFY_VERSION; - TRT_Logger trt_logger(verbosity); - auto trt_builder = InferObject(nvinfer1::createInferBuilder(trt_logger)); - auto trt_network = InferObject(trt_builder->createNetwork()); - auto trt_parser = nvonnxparser::createParser(trt_network.get(), trt_logger); + auto trt_logger = std::unique_ptr(new TRT_Logger(verbosity)); + auto trt_builder = nvinfer1::createInferBuilder(*trt_logger); + auto trt_network = trt_builder->createNetwork(); + auto trt_parser = InferObject(nvonnxparser::createParser(trt_network, *trt_logger)); ::ONNX_NAMESPACE::ModelProto parsed_model; // We check for a valid parse, but the main effect is the side effect // of populating parsed_model @@ -139,8 +124,10 @@ std::tuple onnxToTrtCtx( trt_builder->setMaxBatchSize(max_batch_size); trt_builder->setMaxWorkspaceSize(max_workspace_size); trt_builder->setDebugSync(debug_builder); - nvinfer1::ICudaEngine* trt_engine = trt_builder->buildCudaEngine(*trt_network.get()); - return std::make_tuple(trt_engine, trt_parser); + auto trt_engine = InferObject(trt_builder->buildCudaEngine(*trt_network)); + trt_builder->destroy(); + trt_network->destroy(); + return std::make_tuple(std::move(trt_engine), std::move(trt_parser), std::move(trt_logger)); } } // namespace onnx_to_tensorrt diff --git a/src/operator/subgraph/tensorrt/onnx_to_tensorrt.h b/src/operator/subgraph/tensorrt/onnx_to_tensorrt.h index 3e8ea1b..b89422f 100644 --- a/src/operator/subgraph/tensorrt/onnx_to_tensorrt.h +++ b/src/operator/subgraph/tensorrt/onnx_to_tensorrt.h @@ -32,6 +32,7 @@ #include #include +#include #include #include #include @@ -40,33 +41,51 @@ namespace onnx_to_tensorrt { +struct InferDeleter { + template +void operator()(T* obj) const { + if ( obj ) { +obj->destroy(); + } +} +}; + +template +using unique_ptr = std::unique_ptr; + +template +inline unique_ptr InferObject(T* obj) { + if ( !obj ) { +throw std::runtime_error("Failed to create object"); + } + return unique_ptr(obj, InferDeleter()); +} + class TRT_Logger : public nvinfer1::ILogger { -nvinfer1::ILogger::Severity _verbosity; -std::ostream* _ostream; + nvinfer1::ILogger::Severity _verbosity; + std::ostream* _ostream; public: -TRT_Logger(Severity verbosity = Severity::kWARNING, - std::ostream& ostream = std::cout) -: _verbosity(verbosity), _ostream(&ostream) {} -void log(Severity severity, const char* msg) override { -if ( severity <= _verbosity ) { -time_t rawtime = std::time(0); -char buf[256]; -st
[incubator-mxnet] branch v1.5.x updated: [MXNET-1086] added sub and mul to ONNX->TensorRT conversion (#15344) (#15875)
This is an automated email from the ASF dual-hosted git repository. taolv pushed a commit to branch v1.5.x in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git The following commit(s) were added to refs/heads/v1.5.x by this push: new 964f288 [MXNET-1086] added sub and mul to ONNX->TensorRT conversion (#15344) (#15875) 964f288 is described below commit 964f2880c8c312a0e02579823e7c69875b604619 Author: Kellen Sunderland AuthorDate: Fri Aug 16 08:19:55 2019 -0700 [MXNET-1086] added sub and mul to ONNX->TensorRT conversion (#15344) (#15875) * added sub and mul to ONNX->TensorRT conversion * add test for elementwise ops in TRT --- CMakeLists.txt| 2 +- src/operator/subgraph/tensorrt/nnvm_to_onnx-inl.h | 12 src/operator/subgraph/tensorrt/nnvm_to_onnx.cc| 12 tests/python/tensorrt/test_ops.py | 68 +++ 4 files changed, 93 insertions(+), 1 deletion(-) diff --git a/CMakeLists.txt b/CMakeLists.txt index 2142a09..0148ac3 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -47,7 +47,7 @@ mxnet_option(ENABLE_CUDA_RTC "Build with CUDA runtime compilation support" mxnet_option(BUILD_CPP_EXAMPLES "Build cpp examples" ON) mxnet_option(INSTALL_EXAMPLES "Install the example source files." OFF) mxnet_option(USE_SIGNAL_HANDLER "Print stack traces on segfaults." ON) -mxnet_option(USE_TENSORRT "Enable infeference optimization with TensorRT." OFF) +mxnet_option(USE_TENSORRT "Enable inference optimization with TensorRT." OFF) mxnet_option(USE_ASAN "Enable Clang/GCC ASAN sanitizers." OFF) mxnet_option(ENABLE_TESTCOVERAGE "Enable compilation with test coverage metric output" OFF) mxnet_option(USE_INT64_TENSOR_SIZE "Use int64_t to represent the total number of elements in a tensor" OFF) diff --git a/src/operator/subgraph/tensorrt/nnvm_to_onnx-inl.h b/src/operator/subgraph/tensorrt/nnvm_to_onnx-inl.h index f5bf8b7..55b3d93 100644 --- a/src/operator/subgraph/tensorrt/nnvm_to_onnx-inl.h +++ b/src/operator/subgraph/tensorrt/nnvm_to_onnx-inl.h @@ -125,6 +125,16 @@ void ConvertElementwiseAdd(NodeProto *node_proto, const nnvm::IndexedGraph &ig, const array_view &inputs); +void ConvertElementwiseSub(NodeProto *node_proto, +const NodeAttrs &attrs, +const nnvm::IndexedGraph &ig, +const array_view &inputs); + +void ConvertElementwiseMul(NodeProto *node_proto, +const NodeAttrs &attrs, +const nnvm::IndexedGraph &ig, +const array_view &inputs); + void ConvertConcatenate(NodeProto *node_proto, const NodeAttrs &attrs, const nnvm::IndexedGraph &ig, @@ -151,6 +161,8 @@ static const std::unordered_map converter_map = {"Concat", ConvertConcatenate}, {"Dropout", ConvertDropout}, {"elemwise_add", ConvertElementwiseAdd}, + {"elemwise_sub", ConvertElementwiseSub}, + {"elemwise_mul", ConvertElementwiseMul}, {"Flatten", ConvertFlatten}, {"FullyConnected", ConvertFullyConnected}, {"Pad", ConvertPad}, diff --git a/src/operator/subgraph/tensorrt/nnvm_to_onnx.cc b/src/operator/subgraph/tensorrt/nnvm_to_onnx.cc index 111995d..6116f29 100644 --- a/src/operator/subgraph/tensorrt/nnvm_to_onnx.cc +++ b/src/operator/subgraph/tensorrt/nnvm_to_onnx.cc @@ -406,6 +406,18 @@ void ConvertElementwiseAdd(NodeProto* node_proto, const NodeAttrs& /*attrs*/, node_proto->set_op_type("Add"); } +void ConvertElementwiseSub(NodeProto* node_proto, const NodeAttrs& /*attrs*/, + const nnvm::IndexedGraph& /*ig*/, + const array_view& /*inputs*/) { + node_proto->set_op_type("Sub"); +} + +void ConvertElementwiseMul(NodeProto* node_proto, const NodeAttrs& /*attrs*/, + const nnvm::IndexedGraph& /*ig*/, + const array_view& /*inputs*/) { + node_proto->set_op_type("Mul"); +} + void ConvertConcatenate(NodeProto* node_proto, const NodeAttrs& attrs, const nnvm::IndexedGraph& /*ig*/, const array_view& /*inputs*/) { diff --git a/tests/python/tensorrt/test_ops.py b/tests/python/tensorrt/test_ops.py new file mode 100644 index 000..2df9104 --- /dev/null +++ b/tests/python/tensorrt/test_ops.py @@ -0,0 +1,68 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agree
[GitHub] [incubator-mxnet] TaoLv merged pull request #15877: [v1.5.x] prevent TRT_Logger to be destroyed before TRT engine (#14898)
TaoLv merged pull request #15877: [v1.5.x] prevent TRT_Logger to be destroyed before TRT engine (#14898) URL: https://github.com/apache/incubator-mxnet/pull/15877 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] TaoLv merged pull request #15875: [v1.5.x] Added sub and mul to ONNX->TensorRT conversion (#15344)
TaoLv merged pull request #15875: [v1.5.x] Added sub and mul to ONNX->TensorRT conversion (#15344) URL: https://github.com/apache/incubator-mxnet/pull/15875 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya edited a comment on issue #11720: test_operator.test_laop_3 has fixed seed that can mask flakiness
ChaiBapchya edited a comment on issue #11720: test_operator.test_laop_3 has fixed seed that can mask flakiness URL: https://github.com/apache/incubator-mxnet/issues/11720#issuecomment-521450313 Again #15736 http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwindows-gpu/detail/PR-15736/6/pipeline/ http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwindows-gpu/detail/PR-15736/11/pipeline/ This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya edited a comment on issue #15880: [CI] unix cpu validation Timeout
ChaiBapchya edited a comment on issue #15880: [CI] unix cpu validation Timeout URL: https://github.com/apache/incubator-mxnet/issues/15880#issuecomment-521894929 Another one #15736 http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Funix-cpu/detail/PR-15736/10/ http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Funix-cpu/detail/PR-15736/11/pipeline/291 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] DickJC123 commented on issue #15922: Refactor for windows CI 'out of heap space' errors
DickJC123 commented on issue #15922: Refactor for windows CI 'out of heap space' errors URL: https://github.com/apache/incubator-mxnet/pull/15922#issuecomment-522040068 @larroy @marcoabreu Do you understand why the CI doesn't show here as completely passing? The 'miscellaneous' part also passed if you look at the details. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel edited a comment on issue #15884: [WIP] New Website: New Docs [1/3]
pengzhao-intel edited a comment on issue #15884: [WIP] New Website: New Docs [1/3] URL: https://github.com/apache/incubator-mxnet/pull/15884#issuecomment-522021352 @sandeep-krishnamurthy do we have the plan to put the unified dashboard #15757 into the new website? If so, the user can find the latest performance on the website. Maybe includes GluonCV/NLP/TS, DGL in the future as well. Current [perf page](https://mxnet.incubator.apache.org/versions/master/faq/perf.html#intel-cpu) is really out-of-date and we are refreshing it now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel edited a comment on issue #15884: [WIP] New Website: New Docs [1/3]
pengzhao-intel edited a comment on issue #15884: [WIP] New Website: New Docs [1/3] URL: https://github.com/apache/incubator-mxnet/pull/15884#issuecomment-522021352 @sandeep-krishnamurthy do we have the plan to put the unified dashboard #15757 into the new website? If so, the user can find the latest performance on the website. Maybe includes GluonCV/NLP/TS, DGL in the future as well. Current [perf page](https://mxnet.incubator.apache.org/versions/master/faq/perf.html) is really out-of-date and we are refreshing it now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15884: [WIP] New Website: New Docs [1/3]
pengzhao-intel commented on issue #15884: [WIP] New Website: New Docs [1/3] URL: https://github.com/apache/incubator-mxnet/pull/15884#issuecomment-522021352 @sandeep-krishnamurthy do we have the plan to put the unified dashboard #15757 into the new website? If so, the user will find the latest performance on the website. Maybe includes GluonCV/NLP/TS, DGL in the future as well. Current [perf page](https://mxnet.incubator.apache.org/versions/master/faq/perf.html) is really out-of-date and we are refreshing it now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15884: [WIP] New Website: New Docs [1/3]
pengzhao-intel commented on issue #15884: [WIP] New Website: New Docs [1/3] URL: https://github.com/apache/incubator-mxnet/pull/15884#issuecomment-522019476 Thanks, @ThomasDelteil @xinyu-intel @TaoLv please take a look at the new website and we need to move our documents in new home :) This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zixuanweeei commented on issue #15880: [CI] unix cpu validation Timeout
zixuanweeei commented on issue #15880: [CI] unix cpu validation Timeout URL: https://github.com/apache/incubator-mxnet/issues/15880#issuecomment-522014868 From the last comment by @ChaiBapchya, we also found that `test_operator.test_convolution_independent_gradients` costed too much. And that test was conducted on a library compiled with MKL-DNN. So it will cost more on CPU context when MXNet is compiled without MKL-DNN. Should PR #15922 work for test_shuffle, we would reduce the cost from `test_operator.test_convolution_independent_gradients`. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. marcoabreu pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new f3b2a08 Bump the publish timestamp. f3b2a08 is described below commit f3b2a088f8e7955e0617c1b6c609e8a383726bfe Author: mxnet-ci AuthorDate: Fri Aug 16 13:29:11 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..952201d --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Fri Aug 16 13:29:11 UTC 2019