Remove the old dnn network implementation. This commit removes the ol…

examples/cifar10.cpp

@@ -80,7 +80,7 @@ void testCIFAR10() {
     A = bb.createWeightVar(ElemKind::FloatTy, {minibatchSize, 32, 32, 3});
     E = bb.createWeightVar(ElemKind::IndexTy, {minibatchSize, 1});
 
-    // Create the rest of the network.  NodeBase *SM = createSimpleNet(N, A, E);
+    // Create the rest of the network.
     auto *CV0 = bb.createConvOp(A, 16, 5, 1, 2);
     auto *RL0 = bb.createRELUOp(*CV0);
     auto *MP0 = bb.createPoolOp(*RL0, PoolInst::OpKind::kMax, 2, 2, 0);

include/glow/IR/IR.h

@@ -4,8 +4,8 @@
 #include "glow/IR/Type.h"
 #include "glow/IR/UseDef.h"
 
-#include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/StringRef.h"
 
 #include <list>
 #include <vector>

include/glow/IR/Instrs.h

@@ -88,6 +88,17 @@ class ConvolutionInst : public Instruction {
   size_t getDepth() const { return depth_; }
 
   void verify() const;
+
+  /// Calculate the size of the output tensor based on the convolution
+  /// parameters.
+  static std::pair<size_t, size_t> calculateOutputDims(size_t sx, size_t sy,
+                                                       size_t pad,
+                                                       size_t filterSize,
+                                                       size_t stride) {
+    size_t outsx = ((sx + pad * 2 - filterSize) / stride + 1);
+    size_t outsy = ((sy + pad * 2 - filterSize) / stride + 1);
+    return {outsx, outsy};
+  }
 };
 
 class PoolInst : public Instruction {

include/glow/Importer/Caffe2.h

@@ -68,8 +68,7 @@ class caffe2ModelLoader {
   caffe2ModelLoader(const std::string &netDescFilename,
                     const std::string &netWeightFilename,
                     llvm::ArrayRef<const char *> names,
-                    llvm::ArrayRef<Tensor *> tensors,
-                    Interpreter &IP);
+                    llvm::ArrayRef<Tensor *> tensors, Interpreter &IP);
 
   /// \returns the output of the network. This is usually the result of the last
   /// softmax or regression layer.