TOPI C++ bugfixes (#864)

* TOPI C++ bugfixes

* Fix lint

topi/include/topi/cuda/dense.h

@@ -66,7 +66,7 @@ inline tvm::Tensor dense_cuda(const Target& target,
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_dense(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_dense(const Target &target, const Array<Tensor>& outs) {
   if (target.target_name == "cuda" &&
     target.libs.count("cublas") > 0) {
     return topi::generic::schedule_extern(target, outs);

topi/include/topi/cuda/extern.h

@@ -25,7 +25,7 @@ namespace cuda {
  *
  * \return The schedule given by sch
  */
-Schedule ScheduleOutputForExtern(Target target, Operation op, Schedule sch) {
+inline Schedule ScheduleOutputForExtern(Target target, Operation op, Schedule sch) {
   auto x = op.output(0);
   auto fused = Fuse(sch[x], sch[x]->op.as<ComputeOpNode>()->axis);
   auto num_thread = target.max_num_threads;
@@ -45,7 +45,7 @@ Schedule ScheduleOutputForExtern(Target target, Operation op, Schedule sch) {
 *
 * \return A schedule for the op.
 */
-Schedule schedule_extern(const Target& target, Array<Tensor> outs) {
+inline Schedule schedule_extern(const Target& target, Array<Tensor> outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/cuda/injective.h

@@ -22,7 +22,7 @@ namespace cuda {
 * \param op The operation representing the injective operation.
 * \param s The schedule to apply this scheduling to
 */
-void ScheduleInjectiveOp(const Target &target, Operation op, Schedule s) {
+inline void ScheduleInjectiveOp(const Target &target, Operation op, Schedule s) {
   auto x = op.output(0);
   auto fused = Fuse(s[x], s[x]->op.as<ComputeOpNode>()->axis);
   auto num_thread = target.max_num_threads;
@@ -40,7 +40,7 @@ void ScheduleInjectiveOp(const Target &target, Operation op, Schedule s) {
  *
  * \return A schedule for the given ops.
  */
-Schedule schedule_injective(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_injective(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/cuda/pooling.h

@@ -25,7 +25,7 @@ namespace cuda {
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_pool(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_pool(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);
@@ -90,7 +90,7 @@ Schedule schedule_pool(const Target &target, const Array<Tensor>& outs) {
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_global_pool(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_global_pool(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/cuda/softmax.h

@@ -24,7 +24,7 @@ namespace cuda {
  *
  * \return A schedule for the given ops.
  */
-Schedule schedule_softmax(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_softmax(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/detail/array_utils.h

@@ -21,7 +21,7 @@ using namespace tvm;
  * \return True iff the given array contains the given item.
  */
 template<typename T>
-bool contains(Array<T> array, T item) {
+inline bool contains(Array<T> array, T item) {
   for (auto& i : array) {
     if (i == item) {
       return true;

topi/include/topi/detail/constant_utils.h

@@ -22,7 +22,7 @@ using namespace tvm;
  *
  * \return true if the given expr is a constant int or uint, false otherwise.
  */
-bool IsConstInt(Expr expr) {
+inline bool IsConstInt(Expr expr) {
   return
     expr->derived_from<tvm::ir::IntImm>() ||
     expr->derived_from<tvm::ir::UIntImm>();
@@ -36,7 +36,7 @@ bool IsConstInt(Expr expr) {
  *
  * \return The integer value.
  */
-int64_t GetConstInt(Expr expr) {
+inline int64_t GetConstInt(Expr expr) {
   if (expr->derived_from<tvm::ir::IntImm>()) {
     return expr.as<tvm::ir::IntImm>()->value;
   }
@@ -56,7 +56,7 @@ int64_t GetConstInt(Expr expr) {
  *
  * \return A vector of the integer values
  */
-std::vector<int> GetConstIntValues(Array<Expr> exprs, const std::string& var_name) {
+inline std::vector<int> GetConstIntValues(Array<Expr> exprs, const std::string& var_name) {
   std::vector<int> result;
   for (auto expr : exprs) {
     CHECK(IsConstInt(expr)) << "All elements of " << var_name << " must be constant integers";

topi/include/topi/detail/extern.h

@@ -24,7 +24,7 @@ using namespace tvm;
  *
  * \return The Buffer object
  */
-Buffer DeclExternBuffer(Array<Expr> shape,
+inline Buffer DeclExternBuffer(Array<Expr> shape,
                         Type dtype,
                         std::string name) {
   auto data = var(name, Handle());
@@ -56,7 +56,7 @@ using FExtern = std::function<Expr(Array<Buffer>, Array<Buffer>)>;
  * be one output Tensor for each element of out_shapes, with dtype equal to the corresponding
  * element of out_types.
  */
-Array<Tensor> make_extern(const Array< Array<Expr> >& out_shapes,
+inline Array<Tensor> make_extern(const Array< Array<Expr> >& out_shapes,
                           const std::vector<Type>& out_types,
                           const Array<Tensor>& inputs,
                           FExtern fextern,
@@ -95,7 +95,7 @@ Array<Tensor> make_extern(const Array< Array<Expr> >& out_shapes,
  *
  * \return An expression representing the pack operation
  */
-Expr pack_buffer(Buffer buf) {
+inline Expr pack_buffer(Buffer buf) {
   CHECK_GT(buf->shape.size(), 0) << "buf shape must have at least one element";
   auto shape = tvm::ir::Call::make(Handle(), tvm::ir::intrinsic::tvm_stack_make_shape,
                                    buf->shape, tvm::ir::Call::CallType::Intrinsic);
@@ -127,7 +127,7 @@ Expr pack_buffer(Buffer buf) {
  *
  * \return An expression representing the invocation
  */
-Expr call_packed(Array<Expr> args) {
+inline Expr call_packed(Array<Expr> args) {
   return tvm::ir::Call::make(Int(32), tvm::ir::intrinsic::tvm_call_packed,
                              args, tvm::ir::Call::CallType::Intrinsic);
 }

topi/include/topi/detail/fuse.h

@@ -20,7 +20,7 @@ using namespace tvm;
  *
  * \return The fused iteration variable
  */
-IterVar Fuse(Stage stage, const Array<IterVar>& args) {
+inline IterVar Fuse(Stage stage, const Array<IterVar>& args) {
   CHECK_GE(args.size(), 1) << "Fuse requires at least 1 arg";
 
   auto fused = args[0];

topi/include/topi/detail/pad_utils.h

@@ -23,7 +23,7 @@ using namespace tvm;
  * \return An array of 4 elements, representing padding sizes for
  * each individual side. The array is in the order { top, left, bottom, right }
  */
-Array<Expr> GetPadTuple(Expr pad_h, Expr pad_w) {
+inline Array<Expr> GetPadTuple(Expr pad_h, Expr pad_w) {
   pad_h *= 2;
   pad_w *= 2;
 

topi/include/topi/generic/default.h

@@ -24,7 +24,7 @@ namespace generic {
  *
  * \return A schedule for the given ops.
  */
-Schedule default_schedule(const Target& target, Array<Tensor> outs, bool auto_inline) {
+inline Schedule default_schedule(const Target& target, Array<Tensor> outs, bool auto_inline) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/generic/extern.h

@@ -23,7 +23,7 @@ namespace generic {
 *
 * \return A schedule for the op.
 */
-Schedule schedule_extern(const Target& target, Array<Tensor> outs) {
+inline Schedule schedule_extern(const Target& target, Array<Tensor> outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/generic/injective.h

@@ -24,7 +24,7 @@ namespace generic {
  *
  * \return A schedule for the given ops.
  */
-Schedule schedule_injective(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_injective(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/nn/flatten.h

@@ -33,7 +33,7 @@ inline Tensor flatten(const Tensor& x,
   auto ishape = x->shape;
   int dim = 1;
   for (size_t i = 1; i < ishape.size(); ++i) {
-    dim = dim * static_cast<int>(GetConstInt(ishape[i]));
+    dim = dim * static_cast<int>(topi::detail::GetConstInt(ishape[i]));
   }
 
   Array<Expr> oshape({ ishape[0], dim });

topi/include/topi/rocm/dense.h

@@ -67,7 +67,7 @@ inline tvm::Tensor dense_rocm(const Target& target,
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_dense(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_dense(const Target &target, const Array<Tensor>& outs) {
   if (target.target_name == "rocm" &&
     target.libs.count("rocblas") > 0) {
     return topi::generic::schedule_extern(target, outs);

topi/include/topi/x86/bnn.h

@@ -23,7 +23,7 @@ namespace x86 {
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_binarize_pack(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_binarize_pack(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);
@@ -55,7 +55,7 @@ Schedule schedule_binarize_pack(const Target &target, const Array<Tensor>& outs)
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_binary_dense(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_binary_dense(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/x86/default.h

@@ -24,7 +24,9 @@ namespace x86 {
 *
 * \return A schedule for the given ops.
 */
-Schedule default_schedule(const Target &target, const Array<Tensor>& outs, bool auto_inline) {
+inline Schedule default_schedule(const Target &target,
+                                 const Array<Tensor>& outs,
+                                 bool auto_inline) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);

topi/include/topi/x86/injective.h

@@ -23,7 +23,7 @@ namespace x86 {
 *
 * \return A schedule for the given ops.
 */
-Schedule schedule_injective(const Target &target, const Array<Tensor>& outs) {
+inline Schedule schedule_injective(const Target &target, const Array<Tensor>& outs) {
   Array<Operation> out_ops;
   for (auto t : outs) {
     out_ops.push_back(t->op);