[NFC/cleanup] Remove more entries of variables.

Author: Roman Dzhabarov

examples/char-rnn.cpp

@@ -275,7 +275,7 @@ int main(int argc, char **argv) {
     // Generate a sentence by running inference over and over again.
     for (unsigned i = 0; i < generateChars; i++) {
       // Generate a char:
-      updateVariables(ctx, {X}, {&currCharInfer});
+      updateInputPlaceholders(ctx, {X}, {&currCharInfer});
       EE.run();
 
       // Pick a char at random from the softmax distribution.

examples/cifar10.cpp

@@ -159,7 +159,7 @@ void testCIFAR10() {
     for (unsigned int i = 0; i < 100 / minibatchSize; i++) {
       Tensor sample(ElemKind::FloatTy, {minibatchSize, 32, 32, 3});
       sample.copyConsecutiveSlices(&images, minibatchSize * i);
-      updateVariables(ctx, {A}, {&sample});
+      updateInputPlaceholders(ctx, {A}, {&sample});
       EE.run();
 
       for (unsigned int iter = 0; iter < minibatchSize; iter++) {

examples/fr2en.cpp

@@ -391,7 +391,7 @@ void Model::translate(const std::vector<std::string> &batch) {
         (words.size() - 1) + j * MAX_LENGTH;
   }
 
-  updateVariables(ctx, {input_, seqLength_}, {&input, &seqLength});
+  updateInputPlaceholders(ctx, {input_, seqLength_}, {&input, &seqLength});
   EE_.run();
 
   auto OH = ctx.get(output_)->getHandle<int64_t>();

include/glow/ExecutionEngine/ExecutionEngine.h

@@ -91,15 +91,15 @@ class ExecutionEngine final {
 
 /// This method updates the placeholders in \p ph with the tensor content
 /// values \p inputs, in \p ctx.
-void updateVariables(Context &ctx, llvm::ArrayRef<Placeholder *> ph,
-                     llvm::ArrayRef<Tensor *> inputs);
+void updateInputPlaceholders(Context &ctx, llvm::ArrayRef<Placeholder *> ph,
+                             llvm::ArrayRef<Tensor *> inputs);
 
 /// This method updates the placeholders in the module. The placeholders are
 /// found by name
 ///  in \p ph with the tensor content values \p inputs.
-void updateInputsByName(Context &ctx, Module *mod,
-                        llvm::ArrayRef<llvm::StringRef> ph,
-                        llvm::ArrayRef<Tensor *> inputs);
+void updateInputPlaceholdersByName(Context &ctx, Module *mod,
+                                   llvm::ArrayRef<llvm::StringRef> ph,
+                                   llvm::ArrayRef<Tensor *> inputs);
 
 /// Runs \p iterations iterations of the compiled function. The method updates a
 /// global counter and future invocations of this method continue running

lib/Converter/FunctionConverter.cpp

@@ -17,7 +17,7 @@
 
 #include "glow/Graph/Graph.h" // For Function.
 #include "glow/Graph/Node.h"  // For Node.
-#include "glow/Graph/Nodes.h" // For Placeholder and Variable.
+#include "glow/Graph/Nodes.h" // For Placeholder and Constant.
 
 using namespace glow;
 

lib/ExecutionEngine/ExecutionEngine.cpp

@@ -41,27 +41,27 @@ void ExecutionEngine::setBackend(Backend *backend) {
 
 ExecutionEngine::~ExecutionEngine() = default;
 
-void glow::updateVariables(Context &ctx, llvm::ArrayRef<Placeholder *> ph,
-                           llvm::ArrayRef<Tensor *> inputs) {
+void glow::updateInputPlaceholders(Context &ctx,
+                                   llvm::ArrayRef<Placeholder *> ph,
+                                   llvm::ArrayRef<Tensor *> inputs) {
   assert(inputs.size() == ph.size() &&
          "The number of inputs does not match the number of Placeholders");
 
-  // Update the input variables.
   for (int i = 0, e = ph.size(); i < e; i++) {
     assert(ph[i] && "Invalid value");
     auto *backingTensor = ctx.get(ph[i]);
     assert(backingTensor && "Can't find the placeholder");
     auto dim = inputs[i]->dims();
     (void)dim;
     assert(backingTensor->getType().isEqual(inputs[i]->getType()) &&
-           "Mismatch on Variable and Tensor types.");
+           "Mismatch on Placeholder and Tensor types.");
     backingTensor->assign(inputs[i]);
   }
 }
 
-void glow::updateInputsByName(Context &ctx, Module *mod,
-                              llvm::ArrayRef<llvm::StringRef> ph,
-                              llvm::ArrayRef<Tensor *> inputs) {
+void glow::updateInputPlaceholdersByName(Context &ctx, Module *mod,
+                                         llvm::ArrayRef<llvm::StringRef> ph,
+                                         llvm::ArrayRef<Tensor *> inputs) {
   assert(inputs.size() == ph.size() &&
          "The number of inputs does not match the number of Placeholders");
 
@@ -70,7 +70,7 @@ void glow::updateInputsByName(Context &ctx, Module *mod,
     Tensor *t = inputs[i];
     assert(t && "Invalid tensor.");
     assert(p && "Invalid placeholder.");
-    updateVariables(ctx, {p}, {t});
+    updateInputPlaceholders(ctx, {p}, {t});
   }
 }
 
@@ -87,7 +87,7 @@ void glow::runBatch(ExecutionEngine &EE, Context &ctx, size_t iterations,
 
   assert(!inputs.empty() && "No inputs");
   assert(inputs.size() == ph.size() &&
-         "The number of inputs does not match the number of variables");
+         "The number of inputs does not match the number of placeholders");
 
   // For each iteration in the batch:
   for (size_t j = 0; j < iterations; j++) {

lib/Graph/Graph.cpp

@@ -2361,7 +2361,7 @@ void Function::verify() const {
     }
   }
 
-  // Now check that the variables that are written to are either:
+  // Now check that the placeholders that are written to are either:
   // - Written by a save node, or
   // - Are only used by the node that writes them
   // If this check fails, that means we have implicit memory

lib/IR/GraphScheduler.cpp

@@ -103,7 +103,8 @@ void ChildMemSizeBasedScheduler::orderChildNodesAndSchedule(Node *N) {
 
   // SaveNode hack:
   // We don't model memory dependencies, but we still need to honor them.
-  // Make sure the SaveNode happens after the last use of the output variable.
+  // Make sure the SaveNode happens after the last use of the output
+  // placeholder.
   if (auto *save = dyn_cast<SaveNode>(N)) {
     auto *destination = save->getOutput().getNode();
     for (NodeUse &use : destination->getUsers()) {

lib/Onnxifi/Base.cpp

@@ -79,7 +79,7 @@ onnxStatus Graph::run() {
 
   // Run inference.
   auto &EE = backendPtr_->getEE();
-  updateVariables(ctx_, phs, tensors);
+  updateInputPlaceholders(ctx_, phs, tensors);
   EE.run();
 
   // Copy outputs to the addresses specified in the outputNodeToBuffer_.

lib/Optimizer/IROptimizer.cpp

@@ -481,12 +481,12 @@ void makeWeightsConst(IRFunction &M) {
       }
     }
 
-    // Mark the variable as read only.
+    // Mark the constant as read only.
     if (readOnly) {
       W->setMutability(WeightVar::MutabilityKind::Constant);
     } else {
       assert(W->getMutability() != WeightVar::MutabilityKind::Constant &&
-             "Variables defined as Const cannot be written into.");
+             "Const cannot be written into.");
     }
   }
 }

lib/Optimizer/Lower.cpp

@@ -293,7 +293,7 @@ static void lowerSGDNode(Function *F, const SGDNode &SGD) {
 
   float momentum = SGD.getMomentum();
 
-  assert(W.dims() == G.dims() && "Invalid variables sizes for SGDNode");
+  assert(W.dims() == G.dims() && "Invalid weight/gradient sizes for SGDNode");
 
   float L1Decay = SGD.getL1Decay();
   float L2Decay = SGD.getL2Decay();

lib/Optimizer/Partition.cpp

@@ -64,9 +64,9 @@ class NodeFunctionMap {
   Function *operator[](Node *n) { return nodeToFunction_[n]; }
 };
 
-/// If \p node has a single input that is not a variable, return it.  Otherwise
+/// If \p node has a single input that is not a storage, return it.  Otherwise
 /// return nullptr.
-Node *singleNonVariableInput(Node *node) {
+Node *singleNonStorageInput(Node *node) {
   Node *nonVarInput = nullptr;
 
   for (unsigned i = 0, e = node->getNumInputs(); i < e; i++) {
@@ -96,7 +96,7 @@ NodeFunctionMap selectBasicBlockPartitions(Function *F) {
 
     // If node has only one input, and that input has only one output, place it
     // in the same partition.
-    auto *in = singleNonVariableInput(node);
+    auto *in = singleNonStorageInput(node);
     if (in && in->getNumUsers() == 1) {
       auto it = mapping.find(in);
       assert(it != mapping.end());
@@ -126,9 +126,9 @@ FunctionDAG doPartitioning(Function *F, NodeFunctionMap &mapping) {
     mapping[&N]->addNode(clone);
   }
 
-  // For any dependency that crosses a partition, add a variable and save
+  // For any dependency that crosses a partition, add a placeholder and save
   // node. Record the dependence in the function graph.
-  llvm::DenseMap<Node *, Placeholder *> variables;
+  llvm::DenseMap<Node *, Placeholder *> placeholders;
   for (auto *F : mapping.getFunctions()) {
     for (auto &N : F->getNodes()) {
       for (unsigned inp = 0, e = N.getNumInputs(); inp < e; inp++) {
@@ -143,23 +143,23 @@ FunctionDAG doPartitioning(Function *F, NodeFunctionMap &mapping) {
         // Add this dependence to the FunctionDAG.
         G.add(F, inputF);
 
-        // If we've already created a variable for this dependence, use it.
-        auto it = variables.find(input.getNode());
-        if (it != variables.end()) {
+        // If we've already created a placeholder for this dependence, use it.
+        auto it = placeholders.find(input.getNode());
+        if (it != placeholders.end()) {
           N.setNthInput(inp, it->second);
           continue;
         }
 
-        // Create a new variable to represent this dependence.
+        // Create a new placeholder to represent this dependence.
         auto *save = inputF->createSave("tmp", input);
         auto *tmp = save->getPlaceholder();
-        variables[input.getNode()] = tmp;
+        placeholders[input.getNode()] = tmp;
         N.setNthInput(inp, tmp);
       }
     }
   }
 
-  // Update links between nodes in the cloned functions.  Add variables (and
+  // Update links between nodes in the cloned functions. Add placeholders (and
   // save nodes) where a link crosses a partition boundary.
   for (auto *F : mapping.getFunctions()) {
     for (auto &N : F->getNodes()) {

tests/unittests/BackendTest.cpp

@@ -68,7 +68,7 @@ TEST(Interpreter, profileQuantizationForANetwork) {
 
   // TODO: Verify histogram itself, for now just verify min and max.
   // Run inference first time and capture tensor stats.
-  updateVariables(ctx, {A}, {&inputs});
+  updateInputPlaceholders(ctx, {A}, {&inputs});
   EE.run();
 
   QuantizationProfileNode *profile{nullptr};
@@ -95,7 +95,7 @@ TEST(Interpreter, profileQuantizationForANetwork) {
 
   // Run inference for the second time with new min and max.
   inputs.getHandle() = {0.2f, 1.6f, 0.5f, 1.3f};
-  updateVariables(ctx, {A}, {&inputs});
+  updateInputPlaceholders(ctx, {A}, {&inputs});
   EE.run();
   min = CI.raw(0);
   max = CI.raw(1);
@@ -137,7 +137,7 @@ TEST_P(BackendTest, simpleInference) {
   ctx.allocate(S->getPlaceholder());
   EE_.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {input}, {&inputs});
+  updateInputPlaceholders(ctx, {input}, {&inputs});
   EE_.run();
 }
 

tests/unittests/BackendTestUtils.cpp

@@ -66,7 +66,7 @@ void inferIntLookupTableNet(Tensor *input, Tensor *out,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var}, {input});
+  updateInputPlaceholders(ctx, {var}, {input});
   EE.run();
   out->assign(resultTensor);
 }
@@ -107,7 +107,8 @@ void inferConvNet(Tensor *inputs, Tensor *filter, Tensor *bias, Tensor *out,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {inputP, filterP, biasP}, {inputs, filter, bias});
+  updateInputPlaceholders(ctx, {inputP, filterP, biasP},
+                          {inputs, filter, bias});
   EE.run();
   out->assign(resultTensor);
 }
@@ -149,7 +150,7 @@ void trainConvNet(Tensor *inputs, Tensor *kernel1, Tensor *bias1,
 
   runBatch(EE, ctx, 8, sampleCounter, {var1, var2}, {inputs, selected});
   EE.compile(CompilationMode::Infer, F, ctx);
-  updateVariables(ctx, {var1, var2}, {inputs, selected});
+  updateInputPlaceholders(ctx, {var1, var2}, {inputs, selected});
   EE.run();
   out->assign(resultTensor);
 }
@@ -167,7 +168,7 @@ void inferLocalResponseNormalizationNet(Tensor *inputs, Tensor *out,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var}, {inputs});
+  updateInputPlaceholders(ctx, {var}, {inputs});
   EE.run();
   out->assign(resultTensor);
 }
@@ -248,7 +249,7 @@ void trainAvgPoolNet(Tensor *inputs, Tensor *weights, Tensor *bias,
   runBatch(EE, ctx, 10, sampleCounter, {var1, var2}, {inputs, selected});
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var1, var2}, {inputs, selected});
+  updateInputPlaceholders(ctx, {var1, var2}, {inputs, selected});
   EE.run();
   out->assign(resultTensor);
 }
@@ -306,7 +307,7 @@ void inferSmallConv(Tensor *inputs, Tensor *out, BackendKind kind) {
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {in}, {inputs});
+  updateInputPlaceholders(ctx, {in}, {inputs});
   EE.run();
 
   out->assign(resultTensor);
@@ -543,7 +544,7 @@ void trainSoftMaxNet(Tensor *inputs, Tensor *weights, Tensor *bias,
   runBatch(EE, ctx, 30, sampleCounter, {var1, var2}, {inputs, selected});
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var1, var2}, {inputs, selected});
+  updateInputPlaceholders(ctx, {var1, var2}, {inputs, selected});
   EE.run();
   out->assign(resultTensor);
 }
@@ -567,7 +568,7 @@ void inferTanhConcatNet(Tensor *input1, Tensor *input2, Tensor *input3,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var1, var2, var3}, {input1, input2, input3});
+  updateInputPlaceholders(ctx, {var1, var2, var3}, {input1, input2, input3});
   EE.run();
   out->assign(resultTensor);
 }
@@ -590,7 +591,7 @@ void inferBasicConvNet(Tensor *inputs, Tensor *out, BackendKind kind,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var}, {inputs});
+  updateInputPlaceholders(ctx, {var}, {inputs});
   EE.run();
   out->assign(resultTensor);
 }
@@ -643,7 +644,7 @@ void inferMixedNet(Tensor *inputs, Tensor *out, BackendKind kind) {
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var}, {inputs});
+  updateInputPlaceholders(ctx, {var}, {inputs});
   EE.run();
   out->assign(resultTensor);
 }
@@ -687,8 +688,8 @@ void inferComplexNet1(Tensor *inputs1, Tensor *inputs2, Tensor *inputs3,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var1, var2, var3, var4},
-                  {inputs1, inputs2, inputs3, inputs4});
+  updateInputPlaceholders(ctx, {var1, var2, var3, var4},
+                          {inputs1, inputs2, inputs3, inputs4});
   EE.run();
   out->assign(resultTensor);
 }
@@ -728,7 +729,7 @@ void inferTinyResnet(Tensor *input, Tensor *out, std::vector<Tensor> &weights,
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {in}, {input});
+  updateInputPlaceholders(ctx, {in}, {input});
   EE.run();
   out->assign(resultTensor);
 }
@@ -761,7 +762,7 @@ void inferExtract3D(Tensor *input, Tensor *out, BackendKind kind) {
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {inputs}, {input});
+  updateInputPlaceholders(ctx, {inputs}, {input});
   EE.run();
   out->assign(resultTensor);
 }
@@ -790,7 +791,7 @@ void inferMaxSplat(Tensor *input, Tensor *out, BackendKind kind) {
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {var}, {input});
+  updateInputPlaceholders(ctx, {var}, {input});
   EE.run();
   out->assign(resultTensor);
 }

tests/unittests/GemmTest.cpp

@@ -57,7 +57,7 @@ void infer(Tensor *out, Tensor *lhs, Tensor *rhs) {
 
   EE.compile(CompilationMode::Infer, F, ctx);
 
-  updateVariables(ctx, {lhsVar, rhsVar}, {lhs, rhs});
+  updateInputPlaceholders(ctx, {lhsVar, rhsVar}, {lhs, rhs});
   EE.run();
 
   out->assign(res);

tests/unittests/HyphenTest.cpp

@@ -304,7 +304,7 @@ struct HyphenNetwork {
         bi = numSamples - batchSize;
       }
       auto batchInputs = inputs.getUnowned({batchSize, 6, 27}, {bi, 0, 0});
-      updateVariables(ctx_, {input_}, {&batchInputs});
+      updateInputPlaceholders(ctx_, {input_}, {&batchInputs});
       EE.run();
 
       // Check each output in the batch.