From 9e7875437f2f547095b06f123a0a9ad07ec02475 Mon Sep 17 00:00:00 2001
From: "A. Unique TensorFlower" <gardener@tensorflow.org>
Date: Mon, 31 Jul 2017 16:06:05 -0700
Subject: [PATCH] Add the option of including Shape, ShapeN, Size and Rank in
the standard TensorFlow constant propagation pass, when the inputs to those
Ops have sufficiently known static shape.
PiperOrigin-RevId: 163762750
---
tensorflow/compiler/tf2xla/xla_compiler.cc | 2 +-
.../core/common_runtime/constant_folding.cc | 415 ++++++++++++++----
.../core/common_runtime/constant_folding.h | 5 +
.../common_runtime/constant_folding_test.cc | 191 ++++++++
.../core/common_runtime/direct_session.cc | 3 +-
tensorflow/core/common_runtime/function.cc | 4 +-
.../core/common_runtime/graph_optimizer.cc | 8 +-
.../core/common_runtime/graph_optimizer.h | 17 +-
.../core/distributed_runtime/graph_mgr.cc | 3 +-
.../core/grappler/grappler_item_builder.cc | 3 +-
10 files changed, 562 insertions(+), 89 deletions(-)
diff --git a/tensorflow/compiler/tf2xla/xla_compiler.cc b/tensorflow/compiler/tf2xla/xla_compiler.cc
index 6c81d4e9f82..89d92173c37 100644
--- a/tensorflow/compiler/tf2xla/xla_compiler.cc
+++ b/tensorflow/compiler/tf2xla/xla_compiler.cc
@@ -161,7 +161,7 @@ Status XlaCompiler::CompileFunction(
opts.set_do_constant_folding(true);
GraphOptimizer optimizer(opts);
optimizer.Optimize(flib_runtime_.get(), flib_runtime_->env(),
- /*device=*/nullptr, &graph);
+ /*device=*/nullptr, &graph, /*shape_map=*/nullptr);
VLOG(1) << "====================================================";
TF_RETURN_IF_ERROR(
diff --git a/tensorflow/core/common_runtime/constant_folding.cc b/tensorflow/core/common_runtime/constant_folding.cc
index 914683d9fa3..8dfb8b45de2 100644
--- a/tensorflow/core/common_runtime/constant_folding.cc
+++ b/tensorflow/core/common_runtime/constant_folding.cc
@@ -43,11 +43,182 @@ namespace tensorflow {
namespace {
-bool IsConstantFoldable(const Node* n,
- const std::function<bool(const Node*)>& consider) {
+// Test to see if the Op is one that turns into a constant when its
+// inputs' shapes are known.
+bool IsShapeOp(const Node* n) {
+ const auto& ts = n->type_string();
+ return ts == "Shape" || ts == "ShapeN" || ts == "Rank" || ts == "Size";
+}
+
+// Reads the partially-known shape of each of n's inputs from shape_map, and
+// stores it to input_shapes. Returns false if any input does not have a shape
+// in shape_map.
+bool ReadPartialShapesFromShapeMap(
+ const Node* n,
+ const std::unordered_map<const Node*, std::vector<PartialTensorShape>>*
+ shape_map,
+ std::vector<PartialTensorShape>* input_shapes) {
+ CHECK(shape_map != nullptr);
+ for (const Edge* in : n->in_edges()) {
+ // Don't need to check if incoming control edges have known shapes.
+ if (in->IsControlEdge()) continue;
+ if (shape_map->count(in->src()) == 0) {
+ // One of n's inputs doesn't have known shapes, so don't replace n.
+ return false;
+ }
+ const auto& known_shape = shape_map->at(in->src());
+ CHECK_GT(known_shape.size(), in->src_output());
+ input_shapes->push_back(known_shape[in->src_output()]);
+ }
+ return true;
+}
+
+// If all of n's inputs have fully-defined shapes, inserts those shapes as a
+// vector of Tensors in the shape_replacement_map.
+bool MaybeReplaceShapeOrShapeNOp(
+ const Node* n, const std::vector<PartialTensorShape>& input_shapes,
+ std::unordered_map<const Node*, std::vector<Tensor>>*
+ shape_replacement_map) {
+ std::vector<Tensor> defined_shape;
+ for (const auto& shape : input_shapes) {
+ if (!shape.IsFullyDefined()) {
+ return false;
+ }
+ const int rank = shape.dims();
+ DataType op_type = n->output_type(0);
+ Tensor t(op_type, TensorShape({rank}));
+ if (op_type == DT_INT64) {
+ auto vec = t.vec<int64>();
+ for (int i = 0; i < rank; ++i) {
+ vec(i) = shape.dim_size(i);
+ }
+ } else {
+ CHECK(op_type == DT_INT32);
+ auto vec = t.vec<int32>();
+ for (int i = 0; i < rank; ++i) {
+ if (shape.dim_size(i) > INT_MAX) {
+ VLOG(1) << "Node " << n->name() << " has input shape dimension " << i
+ << " of " << shape.dim_size(i) << " but type INT32 "
+ << " so not replacing as constant: this will trigger a "
+ "runtime error later.";
+ return false;
+ }
+ vec(i) = static_cast<int32>(shape.dim_size(i));
+ }
+ }
+ defined_shape.push_back(t);
+ }
+ // All the inputs had known shapes so we can replace the node by constants
+ // later in the rewrite.
+ shape_replacement_map->insert({n, defined_shape});
+ return true;
+}
+
+// If n's input has defined rank, inserts that rank as a Tensor in the
+// shape_replacement_map.
+bool MaybeReplaceRankOp(const Node* n,
+ const std::vector<PartialTensorShape>& input_shapes,
+ std::unordered_map<const Node*, std::vector<Tensor>>*
+ shape_replacement_map) {
+ CHECK_EQ(input_shapes.size(), 1);
+ if (input_shapes[0].unknown_rank()) {
+ return false;
+ }
+ Tensor t(DT_INT32, TensorShape({}));
+ t.scalar<int32>()() = input_shapes[0].dims();
+ shape_replacement_map->insert({n, {t}});
+ return true;
+}
+
+// If n's input has defined size, inserts that size as a Tensor in the
+// shape_replacement_map.
+bool MaybeReplaceSizeOp(const Node* n,
+ const std::vector<PartialTensorShape>& input_shapes,
+ std::unordered_map<const Node*, std::vector<Tensor>>*
+ shape_replacement_map) {
+ CHECK_EQ(input_shapes.size(), 1);
+ if (!input_shapes[0].IsFullyDefined()) {
+ return false;
+ }
+ DataType op_type = n->output_type(0);
+ Tensor t(op_type, TensorShape({}));
+ int64 size = input_shapes[0].num_elements();
+ if (op_type == DT_INT64) {
+ t.scalar<int64>()() = size;
+ } else {
+ CHECK(op_type == DT_INT32);
+ if (size > INT_MAX) {
+ VLOG(1) << "Node " << n->name() << " has input shape size " << size
+ << " but type INT32 "
+ << " so not replacing as constant: this will trigger a runtime "
+ "error later.";
+ return false;
+ }
+ t.scalar<int32>()() = static_cast<int32>(size);
+ }
+ shape_replacement_map->insert({n, {t}});
+ return true;
+}
+
+// If n is a shape Op (Shape, ShapeN, Rank, or Size) and its inputs have their
+// shapes specified in shape_map, then adds to shape_replacement_map a mapping
+// from n to a vector of Tensors, where Tensor k is the (statically known) value
+// on n's kth output edge. shape_replacement_map has an entry for n iff
+// MaybeReplaceShapeOp returns true, so it's valid to use
+// shape_replacement_map->count(n) as a test to see if n is a shape op that can
+// be replaced.
+bool MaybeReplaceShapeOp(
+ const Node* n,
+ const std::unordered_map<const Node*, std::vector<PartialTensorShape>>*
+ shape_map,
+ std::unordered_map<const Node*, std::vector<Tensor>>*
+ shape_replacement_map) {
+ if (shape_map == nullptr || !IsShapeOp(n)) {
+ return false;
+ }
+ // input_shapes will contain the shapes of each of n's inputs.
+ std::vector<PartialTensorShape> input_shapes;
+ if (!ReadPartialShapesFromShapeMap(n, shape_map, &input_shapes)) {
+ return false;
+ }
+ const auto& ts = n->type_string();
+ if (ts == "Shape" || ts == "ShapeN") {
+ if (!MaybeReplaceShapeOrShapeNOp(n, input_shapes, shape_replacement_map)) {
+ return false;
+ }
+ } else if (ts == "Rank") {
+ if (!MaybeReplaceRankOp(n, input_shapes, shape_replacement_map)) {
+ return false;
+ }
+ } else {
+ CHECK_EQ(ts, "Size");
+ if (!MaybeReplaceSizeOp(n, input_shapes, shape_replacement_map)) {
+ return false;
+ }
+ }
+ return true;
+}
+
+// Returns true if n can be evaluated as constant. shape_map maps from
+// nodes to the partially-known shapes of their outputs. consider if
+// non-null returns a bool indicating whether a given (non-Const,
+// non-Shape) node is eligible to be
+// constant-propagated. shape_replacement_map is filled in with a
+// vector of constant output tensors for constant-foldable shape nodes
+// (Shape, ShapeN, Size, or Rank).
+bool IsConstantFoldable(
+ const Node* n,
+ const std::unordered_map<const Node*, std::vector<PartialTensorShape>>*
+ shape_map,
+ const std::function<bool(const Node*)>& consider,
+ std::unordered_map<const Node*, std::vector<Tensor>>*
+ shape_replacement_map) {
if (n->IsConstant()) {
return true;
}
+ if (MaybeReplaceShapeOp(n, shape_map, shape_replacement_map)) {
+ return true;
+ }
if (n->op_def().is_stateful()) {
return false;
}
@@ -82,56 +253,81 @@ bool IsConstantFoldable(const Node* n,
return true;
}
+// If n is eligible for constant-folding, adds it to nodes, and places its
+// control dependencies and those transitively of its constant-foldable inputs
+// into constant_control_deps. If n is a constant-foldable shape node (Shape,
+// ShapeN, Rank, or Size), also puts its outputs into shape_replacement_map.
+void ConsiderConstantFoldableNode(
+ Node* n, const ConstantFoldingOptions& opts, std::vector<Node*>* nodes,
+ std::unordered_map<const Node*, gtl::FlatSet<Node*>>* constant_control_deps,
+ std::unordered_map<const Node*, std::vector<Tensor>>* shape_replacement_map,
+ bool* internal_node_inserted) {
+ if (IsConstantFoldable(n, opts.shape_map, opts.consider,
+ shape_replacement_map)) {
+ // A node is constant provided all of its non-control incoming Tensors come
+ // from constant nodes, or it's a shape Op with statically known inputs in
+ // which case it is placed in shape_replacement_map.
+ //
+ // We allow control dependencies from non-constant nodes to constant nodes,
+ // but to preserve the graph structure we must transfer the control
+ // dependency onto any constant replacement.
+ bool all_parents_constant = true;
+ for (const Edge* in : n->in_edges()) {
+ // Allows non-constant -> constant control edges.
+ if (!in->IsControlEdge() &&
+ constant_control_deps->count(in->src()) == 0) {
+ all_parents_constant = false;
+ break;
+ }
+ }
+ if (all_parents_constant || shape_replacement_map->count(n) != 0) {
+ gtl::FlatSet<Node*>& control_deps = (*constant_control_deps)[n];
+ for (const Edge* e : n->in_edges()) {
+ if (constant_control_deps->count(e->src()) == 0) {
+ // This branch is taken if the incoming edge is a control dependency,
+ // in which case we want to add it to the dependencies being
+ // accumulated for this node, or the incoming edge is not
+ // constant. The latter may happen when n is a shape node and the
+ // source has known shape. In that case add a control dependency from
+ // the source node, since there was previously a data dependency and
+ // we want to preserve sequencing constraints.
+ if (!e->src()->IsSource()) {
+ control_deps.insert(e->src());
+ }
+ } else {
+ // If the parent has been accumulating control dependencies, add all
+ // of its transitive control deps.
+ const gtl::FlatSet<Node*>& parent_deps =
+ (*constant_control_deps)[e->src()];
+ control_deps.insert(parent_deps.begin(), parent_deps.end());
+ }
+ }
+ nodes->push_back(n);
+ if (!n->IsConstant()) {
+ *internal_node_inserted = true;
+ }
+ }
+ }
+}
+
// Returns the constant foldable nodes in `nodes` in topological order.
// Populates `constant_control_deps` with the non-constant control dependencies
// of each constant node.
void FindConstantFoldableNodes(
- const Graph* graph, ConstantFoldingOptions opts, std::vector<Node*>* nodes,
- std::unordered_map<const Node*, gtl::FlatSet<Node*>>*
- constant_control_deps) {
+ const Graph* graph, const ConstantFoldingOptions& opts,
+ std::vector<Node*>* nodes,
+ std::unordered_map<const Node*, gtl::FlatSet<Node*>>* constant_control_deps,
+ std::unordered_map<const Node*, std::vector<Tensor>>*
+ shape_replacement_map) {
bool internal_node_inserted = false;
- // Walk the nodes in data flow order
- ReverseDFS(
- *graph, nullptr,
- [nodes, constant_control_deps, &internal_node_inserted, opts](Node* n) {
- if (IsConstantFoldable(n, opts.consider)) {
- // A node is constant provided all of its non-control
- // incoming Tensors come from constant nodes.
- //
- // We allow control dependencies from non-constant nodes to constant
- // nodes, but to preserve the graph structure we must transfer the
- // control dependency onto any constant replacement.
- bool all_parents_constant = true;
- for (const Edge* in : n->in_edges()) {
- // Allows non-constant -> constant control edges.
- if (!in->IsControlEdge() &&
- constant_control_deps->count(in->src()) == 0) {
- all_parents_constant = false;
- break;
- }
- }
- if (all_parents_constant) {
- gtl::FlatSet<Node*>& control_deps = (*constant_control_deps)[n];
- for (const Edge* e : n->in_edges()) {
- if (constant_control_deps->count(e->src()) == 0) {
- if (!e->src()->IsSource()) {
- control_deps.insert(e->src());
- }
- } else {
- // If the parent is constant, add all of its transitive control
- // deps.
- const gtl::FlatSet<Node*>& parent_deps =
- (*constant_control_deps)[e->src()];
- control_deps.insert(parent_deps.begin(), parent_deps.end());
- }
- }
- nodes->push_back(n);
- if (!n->IsConstant()) {
- internal_node_inserted = true;
- }
- }
- }
- });
+ // Walk the nodes in data flow order.
+ ReverseDFS(*graph, nullptr,
+ [nodes, constant_control_deps, shape_replacement_map,
+ &internal_node_inserted, &opts](Node* n) {
+ ConsiderConstantFoldableNode(
+ n, opts, nodes, constant_control_deps, shape_replacement_map,
+ &internal_node_inserted);
+ });
// If we have inserted just leaf level nodes, then there is nothing to fold.
if (!internal_node_inserted) {
nodes->clear();
@@ -141,31 +337,93 @@ void FindConstantFoldableNodes(
typedef std::pair<Node*, int> NodeAndOutput;
+int64 UniqueConstantId() {
+ static std::atomic_int_fast64_t id;
+ return id.fetch_add(1);
+}
+
+// Adds n to constant_graph which is being built up for subsequent evaluation of
+// constant propagation. node_map is the mapping of nodes in the original graph
+// to nodes in the constant graph. The value of an entry in node_map is a vector
+// of nodes because a ShapeN node in the original graph is replaced by a vector
+// of Constant nodes in the constant graph.
+void AddNodeToConstantGraph(
+ Node* n, std::unordered_map<Node*, std::vector<Node*>>* node_map,
+ Graph* constant_graph) {
+ std::vector<Node*>& added = (*node_map)[n];
+ added.push_back(constant_graph->CopyNode(n));
+ for (const Edge* in_edge : n->in_edges()) {
+ // Don't copy control edges to the constant graph.
+ if (!in_edge->IsControlEdge()) {
+ Node* in = in_edge->src();
+ auto it = node_map->find(in);
+ CHECK(it != node_map->end())
+ << n->DebugString() << " <-" << in->DebugString();
+ if (it->second.size() == 1) {
+ constant_graph->AddEdge(it->second[0], in_edge->src_output(), added[0],
+ in_edge->dst_input());
+ } else {
+ // The original source node had multiple outputs and was replaced by a
+ // vector of constants, so the edge comes from the 0th output of the kth
+ // added constant, rather than the kth output of the added node as in
+ // the standard case above.
+ constant_graph->AddEdge(it->second[in_edge->src_output()], 0, added[0],
+ in_edge->dst_input());
+ }
+ }
+ }
+}
+
+// Replaces constant-foldable shape node n by a vector of constants in
+// constant_graph, which is being built up for subsequent evaluation of constant
+// propagation. node_map is the mapping of nodes in the original graph to nodes
+// in the constant graph. The value of an entry in node_map is a vector of nodes
+// because a ShapeN node in the original graph is replaced by a vector of
+// Constant nodes in the constant graph.
+void AddShapeNodeToConstantGraph(
+ Node* n,
+ const std::unordered_map<const Node*, std::vector<Tensor>>&
+ shape_replacement_map,
+ std::unordered_map<Node*, std::vector<Node*>>* node_map,
+ Graph* constant_graph) {
+ std::vector<Node*>& added = (*node_map)[n];
+ const string& node_name = n->name();
+ for (const Tensor& t : shape_replacement_map.at(n)) {
+ auto builder =
+ NodeDefBuilder(strings::StrCat(constant_graph->NewName(node_name),
+ "__cf__", UniqueConstantId()),
+ "Const")
+ .Attr("dtype", t.dtype())
+ .Attr("value", t);
+ NodeDef def;
+ CHECK(builder.Finalize(&def).ok());
+ Node* constant_node;
+ CHECK(NodeBuilder(builder).Finalize(constant_graph, &constant_node).ok());
+ added.push_back(constant_node);
+ }
+ // Don't copy incoming edges to shape nodes that are being replaced.
+}
+
// Given the constant foldable nodes in 'nodes', returns a new graph 'g'. 'g'
// will contain copies of the nodes in 'nodes'. In addition, if there is an edge
// going from a node 'n' in 'nodes' to another node in 'orig_graph' but not in
// 'nodes', then 'tensors_to_fetch' will contain the mapping from the
// corresponding copy of 'n' and the edge number in 'g' to 'n'.
-Graph* GetConstantGraph(const Graph* orig_graph,
- const std::vector<Node*>& nodes,
- std::map<NodeAndOutput, Node*>* tensors_to_fetch) {
+Graph* GetConstantGraph(
+ const Graph* orig_graph, const std::vector<Node*>& nodes,
+ const std::unordered_map<const Node*, std::vector<Tensor>>&
+ shape_replacement_map,
+ std::map<NodeAndOutput, Node*>* tensors_to_fetch) {
Graph* constant_graph = new Graph(orig_graph->op_registry());
- std::unordered_map<Node*, Node*> node_map;
- node_map[orig_graph->source_node()] = constant_graph->source_node();
- node_map[orig_graph->sink_node()] = constant_graph->sink_node();
+ std::unordered_map<Node*, std::vector<Node*>> node_map;
+ node_map[orig_graph->source_node()] = {constant_graph->source_node()};
+ node_map[orig_graph->sink_node()] = {constant_graph->sink_node()};
for (Node* n : nodes) {
- Node* added = constant_graph->CopyNode(n);
- node_map[n] = added;
- for (const Edge* in_edge : n->in_edges()) {
- // Don't copy control edges to the constant graph.
- if (!in_edge->IsControlEdge()) {
- Node* in = in_edge->src();
- auto it = node_map.find(in);
- CHECK(it != node_map.end())
- << n->DebugString() << " <-" << in->DebugString();
- constant_graph->AddEdge(it->second, in_edge->src_output(), added,
- in_edge->dst_input());
- }
+ if (shape_replacement_map.count(n) == 0) {
+ AddNodeToConstantGraph(n, &node_map, constant_graph);
+ } else {
+ AddShapeNodeToConstantGraph(n, shape_replacement_map, &node_map,
+ constant_graph);
}
}
@@ -173,8 +431,19 @@ Graph* GetConstantGraph(const Graph* orig_graph,
for (const Edge* out_edge : added_nodes.first->out_edges()) {
if (node_map.count(out_edge->dst()) == 0) {
if (out_edge->IsControlEdge()) continue;
- tensors_to_fetch->insert(
- {{added_nodes.second, out_edge->src_output()}, added_nodes.first});
+ if (added_nodes.second.size() == 1) {
+ tensors_to_fetch->insert(
+ {{added_nodes.second[0], out_edge->src_output()},
+ added_nodes.first});
+ } else {
+ // The node had multiple outputs and was replaced by a
+ // vector of constants, so the NodeAndOutput is the 0th
+ // output of the kth added constant, rather than the kth
+ // output of the added node as in the standard case above.
+ tensors_to_fetch->insert(
+ {{added_nodes.second[out_edge->src_output()], 0},
+ added_nodes.first});
+ }
}
}
}
@@ -182,11 +451,6 @@ Graph* GetConstantGraph(const Graph* orig_graph,
return constant_graph;
}
-int64 UniqueConstantId() {
- static std::atomic_int_fast64_t id;
- return id.fetch_add(1);
-}
-
// Replaces the identified Tensor in 'graph' by a 'Const' node with
// the value supplied in 'constant'. 'partition_device', if non-null
// is the device where the graph executes. Returns true if the
@@ -291,8 +555,9 @@ Status ConstantFold(const ConstantFoldingOptions& opts,
std::vector<Node*> constant_foldable_nodes;
std::unordered_map<const Node*, gtl::FlatSet<Node*>> constant_control_deps;
+ std::unordered_map<const Node*, std::vector<Tensor>> shape_replacement_map;
FindConstantFoldableNodes(graph, opts, &constant_foldable_nodes,
- &constant_control_deps);
+ &constant_control_deps, &shape_replacement_map);
if (constant_foldable_nodes.empty()) {
VLOG(1) << "No constant foldable nodes found";
*was_mutated = false;
@@ -302,7 +567,8 @@ Status ConstantFold(const ConstantFoldingOptions& opts,
std::map<NodeAndOutput, Node*> tensors_to_fetch;
std::unique_ptr<Graph> constant_graph(
- GetConstantGraph(graph, constant_foldable_nodes, &tensors_to_fetch));
+ GetConstantGraph(graph, constant_foldable_nodes, shape_replacement_map,
+ &tensors_to_fetch));
DumpGraph("Constant graph", constant_graph.get());
if (tensors_to_fetch.empty()) {
@@ -337,7 +603,6 @@ Status ConstantFold(const ConstantFoldingOptions& opts,
if (!s.ok()) {
VLOG(1) << "Could not fetch constants: " << s;
*was_mutated = false;
- // This is not an error, so return the status as OK.
return s;
}
diff --git a/tensorflow/core/common_runtime/constant_folding.h b/tensorflow/core/common_runtime/constant_folding.h
index 93289b875f5..07cb5def8b4 100644
--- a/tensorflow/core/common_runtime/constant_folding.h
+++ b/tensorflow/core/common_runtime/constant_folding.h
@@ -29,6 +29,11 @@ struct ConstantFoldingOptions {
// If "consider" is not a nullptr, then only constant fold a node "n" if
// consider(n) returns true.
std::function<bool(const Node*)> consider = nullptr;
+ // If shape_map is not a nullptr, it is a map from node n to a
+ // vector of the (potentially partially-known) shapes of its
+ // outputs.
+ const std::unordered_map<const Node*, std::vector<PartialTensorShape>>*
+ shape_map; // not owned
};
// Perform constant folding optimization on "graph".
diff --git a/tensorflow/core/common_runtime/constant_folding_test.cc b/tensorflow/core/common_runtime/constant_folding_test.cc
index 4a8560960ed..c76ad647a0a 100644
--- a/tensorflow/core/common_runtime/constant_folding_test.cc
+++ b/tensorflow/core/common_runtime/constant_folding_test.cc
@@ -363,6 +363,197 @@ TEST_F(ConstantFoldingTest, ControlDependencies) {
}
}
+TEST_F(ConstantFoldingTest, SimpleShapeKnown) {
+ Graph g(OpRegistry::Global());
+ {
+ Scope s = Scope::NewRootScope();
+ Output recv0 = ops::_Recv(s.WithOpName("recv0"), DT_FLOAT, "recv0",
+ "sender", 0, "receiver");
+ auto shape = ops::Shape(s.WithOpName("shape"), recv0);
+ Output recv1 = ops::_Recv(s.WithOpName("recv1"), DT_FLOAT, "recv1",
+ "sender", 0, "receiver");
+ auto shape_n = ops::ShapeN(s.WithOpName("shape_n"), {recv0, recv1});
+ auto rank = ops::Rank(s.WithOpName("rank"), recv0);
+ auto size = ops::Size(s.WithOpName("size"), recv1);
+ auto recv2 = ops::_Recv(s.WithOpName("recv2"), DT_FLOAT, "recv2", "sender",
+ 0, "receiver");
+ auto c = ops::Const<int>(s.WithControlDependencies(recv2), 3);
+ auto add0 = ops::Add(s.WithControlDependencies(c), rank, size);
+ auto add1 = ops::Add(s, shape, shape_n[0]);
+ auto add2 = ops::Add(s, shape_n[1], shape_n[1]);
+ auto send0 = ops::_Send(s.WithOpName("send0"), add0, "send0", "sender", 0,
+ "receiver");
+ auto send1 = ops::_Send(s.WithOpName("send1"), add1, "send1", "sender", 0,
+ "receiver");
+ auto send2 = ops::_Send(s.WithOpName("send2"), add2, "send2", "sender", 0,
+ "receiver");
+ TF_ASSERT_OK(s.ToGraph(&g));
+ }
+ std::unordered_map<string, Node*> orig_index = NodeNameIndex(g);
+ Node* recv0 = orig_index.at("recv0");
+ Node* recv1 = orig_index.at("recv1");
+ PartialTensorShape ps0;
+ int r0_dims[] = {1, 2};
+ TF_EXPECT_OK(PartialTensorShape::MakePartialShape(r0_dims, 2, &ps0));
+ PartialTensorShape ps1;
+ int r1_dims[] = {2, 3, 4};
+ TF_EXPECT_OK(PartialTensorShape::MakePartialShape<int>(r1_dims, 3, &ps1));
+ std::unordered_map<const Node*, std::vector<PartialTensorShape>> map;
+ map[recv0].push_back(ps0);
+ map[recv1].push_back(ps1);
+ ConstantFoldingOptions opts;
+ opts.shape_map = ↦
+ bool was_mutated;
+ TF_EXPECT_OK(
+ ConstantFold(opts, nullptr, Env::Default(), nullptr, &g, &was_mutated));
+ EXPECT_TRUE(was_mutated);
+
+ std::unordered_map<string, Node*> index = NodeNameIndex(g);
+ Node* recv2 = index.at("recv2");
+ Node* send0 = index.at("send0");
+ Node* send1 = index.at("send1");
+ Node* send2 = index.at("send2");
+
+ ASSERT_EQ(1, send0->num_inputs());
+ Node* cf0 = *(send0->in_nodes().begin());
+ ExpectNodeEqual<int>(cf0, {26}, {});
+
+ ASSERT_EQ(1, send1->num_inputs());
+ Node* cf1 = *(send1->in_nodes().begin());
+ ExpectNodeEqual<int>(cf1, {2, 4}, {2});
+
+ ASSERT_EQ(1, send2->num_inputs());
+ Node* cf2 = *(send2->in_nodes().begin());
+ ExpectNodeEqual<int>(cf2, {4, 6, 8}, {3});
+
+ ASSERT_EQ(3, cf0->in_edges().size());
+ for (const Edge* e : cf0->in_edges()) {
+ EXPECT_TRUE(e->IsControlEdge());
+ EXPECT_TRUE(e->src() == recv0 || e->src() == recv1 || e->src() == recv2)
+ << e->src()->name();
+ }
+
+ ASSERT_EQ(2, cf1->in_edges().size());
+ for (const Edge* e : cf1->in_edges()) {
+ EXPECT_TRUE(e->IsControlEdge());
+ EXPECT_TRUE(e->src() == recv0 || e->src() == recv1) << e->src()->name();
+ }
+
+ ASSERT_EQ(2, cf2->in_edges().size());
+ for (const Edge* e : cf2->in_edges()) {
+ EXPECT_TRUE(e->IsControlEdge());
+ EXPECT_TRUE(e->src() == recv0 || e->src() == recv1) << e->src()->name();
+ }
+}
+
+TEST_F(ConstantFoldingTest, PartialShape) {
+ Graph g(OpRegistry::Global());
+ {
+ Scope s = Scope::NewRootScope();
+ Output recv0 = ops::_Recv(s.WithOpName("recv0"), DT_FLOAT, "recv0",
+ "sender", 0, "receiver");
+ Output recv1 = ops::_Recv(s.WithOpName("recv1"), DT_FLOAT, "recv1",
+ "sender", 0, "receiver");
+ auto shape = ops::Shape(s.WithOpName("shape"), recv0);
+ auto rank0 = ops::Rank(s.WithOpName("rank0"), recv0);
+ auto rank1 = ops::Rank(s.WithOpName("rank1"), recv1);
+ auto size = ops::Size(s.WithOpName("size"), recv0);
+ auto send0 = ops::_Send(s.WithOpName("send0"), rank0, "send0", "sender", 0,
+ "receiver");
+ auto send1 = ops::_Send(s.WithOpName("send1"), shape, "send1", "sender", 0,
+ "receiver");
+ auto send2 = ops::_Send(s.WithOpName("send2"), size, "send2", "sender", 0,
+ "receiver");
+ auto send3 = ops::_Send(s.WithOpName("send3"), rank1, "send3", "sender", 0,
+ "receiver");
+ TF_ASSERT_OK(s.ToGraph(&g));
+ }
+ std::unordered_map<string, Node*> orig_index = NodeNameIndex(g);
+ Node* recv0 = orig_index.at("recv0");
+ Node* recv1 = orig_index.at("recv1");
+ PartialTensorShape ps0;
+ int r0_dims[] = {-1, -1};
+ TF_EXPECT_OK(PartialTensorShape::MakePartialShape(r0_dims, 2, &ps0));
+ PartialTensorShape ps1;
+ std::unordered_map<const Node*, std::vector<PartialTensorShape>> map;
+ map[recv0].push_back(ps0);
+ map[recv1].push_back(ps1);
+ ConstantFoldingOptions opts;
+ opts.shape_map = ↦
+ bool was_mutated;
+ TF_EXPECT_OK(
+ ConstantFold(opts, nullptr, Env::Default(), nullptr, &g, &was_mutated));
+ EXPECT_TRUE(was_mutated);
+
+ std::unordered_map<string, Node*> index = NodeNameIndex(g);
+ Node* shape = index.at("shape");
+ Node* size = index.at("size");
+ Node* rank1 = index.at("rank1");
+ Node* send0 = index.at("send0");
+ Node* send1 = index.at("send1");
+ Node* send2 = index.at("send2");
+ Node* send3 = index.at("send3");
+
+ ASSERT_EQ(1, send0->num_inputs());
+ Node* cf0 = *(send0->in_nodes().begin());
+ ExpectNodeEqual<int>(cf0, {2}, {});
+
+ ASSERT_EQ(1, send1->num_inputs());
+ Node* ncf1 = *(send1->in_nodes().begin());
+ EXPECT_EQ(ncf1, shape);
+
+ ASSERT_EQ(1, send2->num_inputs());
+ Node* ncf2 = *(send2->in_nodes().begin());
+ EXPECT_EQ(ncf2, size);
+
+ ASSERT_EQ(1, send3->num_inputs());
+ Node* ncf3 = *(send3->in_nodes().begin());
+ EXPECT_EQ(ncf3, rank1);
+}
+
+TEST_F(ConstantFoldingTest, ConstShapeKnown) {
+ Graph g(OpRegistry::Global());
+ {
+ Scope s = Scope::NewRootScope();
+ auto recv0 = ops::_Recv(s.WithOpName("recv0"), DT_FLOAT, "recv0", "sender",
+ 0, "receiver");
+ auto c0 =
+ ops::Const<int>(s.WithOpName("c0").WithControlDependencies(recv0), 1);
+ auto rank = ops::Rank(s.WithOpName("rank"), c0);
+ auto add0 = ops::Add(s, rank, rank);
+ auto send0 = ops::_Send(s.WithOpName("send0"), add0, "send0", "sender", 0,
+ "receiver");
+ TF_ASSERT_OK(s.ToGraph(&g));
+ }
+ std::unordered_map<string, Node*> orig_index = NodeNameIndex(g);
+ Node* c0 = orig_index.at("c0");
+ PartialTensorShape ps0;
+ int c0_dims[] = {};
+ TF_EXPECT_OK(PartialTensorShape::MakePartialShape(c0_dims, 0, &ps0));
+ std::unordered_map<const Node*, std::vector<PartialTensorShape>> map;
+ map[c0].push_back(ps0);
+ ConstantFoldingOptions opts;
+ opts.shape_map = ↦
+ bool was_mutated;
+ TF_EXPECT_OK(
+ ConstantFold(opts, nullptr, Env::Default(), nullptr, &g, &was_mutated));
+ EXPECT_TRUE(was_mutated);
+
+ std::unordered_map<string, Node*> index = NodeNameIndex(g);
+ Node* recv0 = index.at("recv0");
+ Node* send0 = index.at("send0");
+
+ ASSERT_EQ(1, send0->num_inputs());
+ Node* cf0 = *(send0->in_nodes().begin());
+ ExpectNodeEqual<int>(cf0, {0}, {});
+
+ ASSERT_EQ(1, cf0->in_edges().size());
+ for (const Edge* e : cf0->in_edges()) {
+ EXPECT_TRUE(e->IsControlEdge());
+ EXPECT_TRUE(e->src() == recv0) << e->src()->name();
+ }
+}
+
namespace {
const char kTestMemRegionName[] = "test://test";
diff --git a/tensorflow/core/common_runtime/direct_session.cc b/tensorflow/core/common_runtime/direct_session.cc
index 294ad13a0a4..aaba2aa7875 100644
--- a/tensorflow/core/common_runtime/direct_session.cc
+++ b/tensorflow/core/common_runtime/direct_session.cc
@@ -1194,7 +1194,8 @@ Status DirectSession::GetOrCreateExecutors(
};
params.node_outputs_cb = node_outputs_callback_;
- optimizer.Optimize(lib, options_.env, device, &iter->second);
+ optimizer.Optimize(lib, options_.env, device, &iter->second,
+ /*shape_map=*/nullptr);
// EXPERIMENTAL: tfdbg inserts debug nodes in the graph.
if (!options.debug_options.debug_tensor_watch_opts().empty()) {
diff --git a/tensorflow/core/common_runtime/function.cc b/tensorflow/core/common_runtime/function.cc
index 64c3747ce1c..b7fad68de76 100644
--- a/tensorflow/core/common_runtime/function.cc
+++ b/tensorflow/core/common_runtime/function.cc
@@ -461,7 +461,7 @@ void OptimizeGraph(FunctionLibraryRuntime* lib, std::unique_ptr<Graph>* g) {
opts.set_do_function_inlining(true);
opts.set_do_constant_folding(true);
GraphOptimizer optimizer(opts);
- optimizer.Optimize(lib, lib->env(), lib->device(), g);
+ optimizer.Optimize(lib, lib->env(), lib->device(), g, /*shape_map=*/nullptr);
}
Status FunctionLibraryRuntimeImpl::CreateItem(Handle handle, Item** item) {
@@ -470,7 +470,7 @@ Status FunctionLibraryRuntimeImpl::CreateItem(Handle handle, Item** item) {
std::unique_ptr<Graph> g(new Graph(lib_def_));
CopyGraph(*fbody->graph, g.get());
- optimizer_.Optimize(this, env(), device(), &g);
+ optimizer_.Optimize(this, env(), device(), &g, /*shape_map=*/nullptr);
TF_RETURN_IF_ERROR(EnsureMemoryTypes(DeviceType(device()->device_type()),
device()->name(), g.get()));
diff --git a/tensorflow/core/common_runtime/graph_optimizer.cc b/tensorflow/core/common_runtime/graph_optimizer.cc
index edfecfae06e..c32d9f8a2e9 100644
--- a/tensorflow/core/common_runtime/graph_optimizer.cc
+++ b/tensorflow/core/common_runtime/graph_optimizer.cc
@@ -33,8 +33,11 @@ GraphOptimizer::GraphOptimizer(const OptimizerOptions& opts) : opts_(opts) {
GraphOptimizer::~GraphOptimizer() {}
-void GraphOptimizer::Optimize(FunctionLibraryRuntime* runtime, Env* env,
- Device* device, std::unique_ptr<Graph>* graph) {
+void GraphOptimizer::Optimize(
+ FunctionLibraryRuntime* runtime, Env* env, Device* device,
+ std::unique_ptr<Graph>* graph,
+ const std::unordered_map<const Node*, std::vector<PartialTensorShape>>*
+ shape_map) {
Graph* g = graph->get();
DumpGraph("Initial", g);
@@ -57,6 +60,7 @@ void GraphOptimizer::Optimize(FunctionLibraryRuntime* runtime, Env* env,
if (opts_.do_constant_folding()) {
ConstantFoldingOptions cf_opts;
+ cf_opts.shape_map = shape_map;
bool was_mutated;
ConstantFold(cf_opts, runtime, env, device, g, &was_mutated)
.IgnoreError();
diff --git a/tensorflow/core/common_runtime/graph_optimizer.h b/tensorflow/core/common_runtime/graph_optimizer.h
index a6b10356ce1..c145adde829 100644
--- a/tensorflow/core/common_runtime/graph_optimizer.h
+++ b/tensorflow/core/common_runtime/graph_optimizer.h
@@ -30,12 +30,17 @@ class GraphOptimizer {
~GraphOptimizer();
// Applies optimization passes specified in 'opts' to 'graph'.
- // Maybe replace *graph with a new graph object.
- // 'device' is device on which the 'graph' will execute. It's passed to the
- // optimizers so that they can respect constraints if any, that should be
- // respected.
- void Optimize(FunctionLibraryRuntime* runtime, Env* env, Device* device,
- std::unique_ptr<Graph>* graph);
+ // Maybe replace *graph with a new graph object. 'device' is device
+ // on which the 'graph' will execute. It's passed to the optimizers
+ // so that they can respect constraints if any, that should be
+ // respected. If shape_map is not null it maps from nodes in graph
+ // to partially-known shapes of their outputs, and may be used,
+ // e.g., in the constant folding pass.
+ void Optimize(
+ FunctionLibraryRuntime* runtime, Env* env, Device* device,
+ std::unique_ptr<Graph>* graph,
+ const std::unordered_map<const Node*, std::vector<PartialTensorShape>>*
+ shape_map);
private:
OptimizerOptions opts_;
diff --git a/tensorflow/core/distributed_runtime/graph_mgr.cc b/tensorflow/core/distributed_runtime/graph_mgr.cc
index 8d4b1be0261..ce186700b33 100644
--- a/tensorflow/core/distributed_runtime/graph_mgr.cc
+++ b/tensorflow/core/distributed_runtime/graph_mgr.cc
@@ -244,7 +244,8 @@ Status GraphMgr::InitItem(const string& session, const GraphDef& gdef,
}
};
- optimizer.Optimize(lib, worker_env_->env, params.device, &subgraph);
+ optimizer.Optimize(lib, worker_env_->env, params.device, &subgraph,
+ /*shape_map=*/nullptr);
// EXPERIMENTAL: tfdbg inserts debug nodes (i.e., probes) to the graph.
if (!debug_options.debug_tensor_watch_opts().empty()) {
diff --git a/tensorflow/core/grappler/grappler_item_builder.cc b/tensorflow/core/grappler/grappler_item_builder.cc
index 0c2801e8bc3..ed8bbe58c46 100644
--- a/tensorflow/core/grappler/grappler_item_builder.cc
+++ b/tensorflow/core/grappler/grappler_item_builder.cc
@@ -127,7 +127,8 @@ Status OptimizeGraph(const GraphDef& graph_def, GraphDef* output_graph_def,
// Optimize the graph.
GraphOptimizer optimizer(*optimizer_opts);
- optimizer.Optimize(flib.get(), env, devices[0], &graphptr);
+ optimizer.Optimize(flib.get(), env, devices[0], &graphptr,
+ /*shape_map=*/nullptr);
graphptr->ToGraphDef(output_graph_def);
return Status::OK();