FM / opt

/usr/local/lib64/python3.6/site-packages/torch/include/caffe2/opt/

/usr/local/lib64/python3.6/site-packages/torch/include/caffe2/opt

Name	Size	Mode	Actions
annotations.h	2141	0644	edit dl rm
backend_cutting.h	586	0644	edit dl rm
backend_transformer_base.h	2773	0644	edit dl rm
bound_shape_inferencer.h	5884	0644	edit dl rm
converter.h	3013	0644	edit dl rm
device.h	442	0644	edit dl rm
distributed.h	1126	0644	edit dl rm
fakefp16_transform.h	584	0644	edit dl rm
fusion.h	4141	0644	edit dl rm
glow_net_transform.h	1625	0644	edit dl rm
mobile.h	388	0644	edit dl rm
onnxifi_op.h	19497	0644	edit dl rm
onnxifi_transformer.h	6859	0644	edit dl rm
onnx_convert.h	1279	0644	edit dl rm
optimizer.h	402	0644	edit dl rm
optimize_ideep.h	351	0644	edit dl rm
passes.h	2642	0644	edit dl rm
shape_info.h	4784	0644	edit dl rm
tvm_transformer.h	2992	0644	edit dl rm

Edit: /usr/local/lib64/python3.6/site-packages/torch/include/caffe2/opt/glow_net_transform.h (1625B)

#pragma once

#include <string>
#include <unordered_set>
#include <vector>

#include <caffe2/core/net.h>
#include <caffe2/core/workspace.h>
#include <caffe2/opt/shape_info.h>
#include <caffe2/proto/caffe2_pb.h>

C10_DECLARE_string(onnxifi_blacklist);
C10_DECLARE_string(onnxifi_blacklist_ops);

namespace caffe2 {
namespace glow {
/// Onnxifi transformation on the net and workspace.  We also
/// needed the input data/shape to populate the shape. In addition, we take a \p
/// blocklist to control and mask what ops we want to consider in onnxifi
/// process. We can also set whether to use ONNX proto or C2 proto through
/// ONNXIFI interface.
void onnxifi(
    NetDef* net,
    Workspace* ws,
    const std::vector<std::string>& input_names,
    const std::vector<std::string>& output_names,
    const std::vector<std::string>& weight_names,
    const std::unordered_set<int>& blocklist,
    const ShapeInfoMap& shape_hints_max_bs,
    bool use_onnx,
    size_t max_batch_size = 0,
    size_t max_seq_size = 0,
    bool load_model_by_blob = false,
    bool predictor_net_ssa_rewritten = false,
    const std::unordered_map<int, ShapeInfoMap> &shape_hints_per_bs = {},
    const c10::optional<std::string> &blacklist_ops = c10::nullopt,
    const c10::optional<size_t> &min_ops = c10::nullopt,
    const std::unordered_set<std::string> &blocklist_blobs = {},
    const c10::optional<bool> & verify_only_single_subnet = c10::nullopt);

std::unordered_set<int> ParseNetPositionList(const std::string& str);
std::unordered_set<std::string> ParseBlockListOps(const std::string& str);

} // namespace glow
} // namespace caffe2

#pragma once #include #include #include #include #include #include #include C10_DECLARE_string(onnxifi_blacklist); C10_DECLARE_string(onnxifi_blacklist_ops); namespace caffe2 { namespace glow { /// Onnxifi transformation on the net and workspace. We also /// needed the input data/shape to populate the shape. In addition, we take a \p /// blocklist to control and mask what ops we want to consider in onnxifi /// process. We can also set whether to use ONNX proto or C2 proto through /// ONNXIFI interface. void onnxifi( NetDef* net, Workspace* ws, const std::vector& input_names, const std::vector& output_names, const std::vector& weight_names, const std::unordered_set& blocklist, const ShapeInfoMap& shape_hints_max_bs, bool use_onnx, size_t max_batch_size = 0, size_t max_seq_size = 0, bool load_model_by_blob = false, bool predictor_net_ssa_rewritten = false, const std::unordered_map &shape_hints_per_bs = {}, const c10::optional &blacklist_ops = c10::nullopt, const c10::optional &min_ops = c10::nullopt, const std::unordered_set &blocklist_blobs = {}, const c10::optional & verify_only_single_subnet = c10::nullopt); std::unordered_set ParseNetPositionList(const std::string& str); std::unordered_set ParseBlockListOps(const std::string& str); } // namespace glow } // namespace caffe2