Merge pull request tensorflow#8206 from conqer/fix_xla_compiler_warnings

Fix compile warnings.

Author: caisq

tensorflow/compiler/xla/layout_util.cc

@@ -287,7 +287,8 @@ Layout CreateDefaultLayoutForRank(int64 rank) {
 /* static */ std::vector<int64> LayoutUtil::MakeLogicalToPhysical(
     const Layout& layout) {
   std::vector<int64> logical_to_physical(layout.minor_to_major_size());
-  for (int64 physical = 0; physical < logical_to_physical.size(); ++physical) {
+  for (std::vector<int64>::size_type physical = 0;
+       physical < logical_to_physical.size(); ++physical) {
     const int64 logical = Major(layout, physical);
     logical_to_physical[logical] = physical;
   }

tensorflow/compiler/xla/metric_table_report.cc

@@ -76,10 +76,10 @@ void MetricTableReport::WriteReportToInfoLog(double expected_metric_sum) {
   // Write something to the log normally to get the date-time and file prefix.
   LOG(INFO) << "Writing report to log.";
 
-  int64 pos = 0;
+  string::size_type pos = 0;
   const string report = MakeReport(expected_metric_sum);
   while (pos < report.size()) {
-    int64 end_of_line = report.find('\n', pos);
+    size_t end_of_line = report.find('\n', pos);
     if (end_of_line == string::npos) {
       end_of_line = report.size();
     }
@@ -247,7 +247,7 @@ string MetricTableReport::MetricString(double metric) {
     sp1.remove_prefix(1);
   }
   // Copy rest of input characters.
-  for (int64 i = 0; i < sp1.size(); ++i) {
+  for (tensorflow::StringPiece::size_type i = 0; i < sp1.size(); ++i) {
     if (i > 0 && (sp1.size() - i) % 3 == 0) {
       output.push_back(',');
     }

tensorflow/compiler/xla/service/cpu/dot_op_emitter.cc

@@ -191,12 +191,12 @@ tensorflow::Status DotOpEmitter::Emit() {
   // from the rhs index are the lower dimensions in the index so we add them
   // first.
   llvm_ir::IrArray::Index target_index;
-  for (int dimension = 0; dimension < lhs_index.size(); ++dimension) {
+  for (size_t dimension = 0; dimension < lhs_index.size(); ++dimension) {
     if (dimension != lhs_reduction_dimension) {
       target_index.push_back(lhs_index[dimension]);
     }
   }
-  for (int dimension = 0; dimension < rhs_index.size(); ++dimension) {
+  for (size_t dimension = 0; dimension < rhs_index.size(); ++dimension) {
     if (dimension != rhs_reduction_dimension) {
       target_index.push_back(rhs_index[dimension]);
     }
@@ -332,7 +332,7 @@ llvm_ir::IrArray::Index DotOpEmitter::EmitOperandArrayLoopNest(
   llvm_ir::IrArray::Index index =
       loop_nest->AddLoopsForShapeOnDimensions(shape, dimensions, name_suffix);
   // Verify every dimension except the reduction dimension was set in the index.
-  for (int dimension = 0; dimension < index.size(); ++dimension) {
+  for (size_t dimension = 0; dimension < index.size(); ++dimension) {
     if (dimension == reduction_dimension) {
       DCHECK_EQ(nullptr, index[dimension]);
     } else {

tensorflow/compiler/xla/service/generic_transfer_manager.cc

@@ -68,7 +68,8 @@ Status GenericTransferManager::TransferLiteralFromDevice(
         ShallowCopyTupleFromDevice(executor, source, device_shape));
     TF_RET_CHECK(element_buffers.size() ==
                  ShapeUtil::TupleElementCount(device_shape));
-    for (int64 i = 0; i < element_buffers.size(); ++i) {
+    for (std::vector<se::DeviceMemoryBase>::size_type i = 0;
+         i < element_buffers.size(); ++i) {
       const Shape& element_device_shape = device_shape.tuple_shapes(i);
       const Shape& element_literal_shape = literal_shape.tuple_shapes(i);
       Literal* element_literal = literal->add_tuple_literals();

tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/dump_ir_pass.cc

@@ -80,7 +80,7 @@ class DumpIrPass : public llvm::FunctionPass {
 char DumpIrPass::id_ = 0;
 
 void IrDumpingPassManager::run(llvm::Module &module) {
-  for (int i = 0; i < passes_.size(); ++i) {
+  for (std::vector<llvm::Pass*>::size_type i = 0; i < passes_.size(); ++i) {
     llvm::Pass *P = passes_[i];
     if (dump_ir_) {
       const llvm::PassInfo *PI =

tensorflow/compiler/xla/service/hlo_instruction.cc

@@ -548,7 +548,8 @@ HloInstruction* HloInstruction::CloneAndFuseInternal(
     // See if this operand is already an operand of the fusion node.
     CHECK_EQ(operands_.size(), fused_parameters_.size());
     HloInstruction* fused_param = nullptr;
-    for (int64 i = 0; i < operands_.size(); ++i) {
+    for (std::vector<HloInstruction*>::size_type i = 0;
+         i < operands_.size(); ++i) {
       if (operands_[i] == operand) {
         fused_param = fused_parameters_[i];
         break;
@@ -639,7 +640,8 @@ void HloInstruction::CheckFusionInstruction() const {
       CHECK(!root_owned);
       root_owned = true;
     }
-    for (int i = 0; i < fused_parameters_.size(); ++i) {
+    for (std::vector<HloInstruction*>::size_type i = 0;
+         i < fused_parameters_.size(); ++i) {
       if (fused_parameters_[i] == instruction.get()) {
         CHECK(!parameter_owned[i]);
         parameter_owned[i] = true;
@@ -648,7 +650,7 @@ void HloInstruction::CheckFusionInstruction() const {
   }
   CHECK(root_owned);
   // Make sure all the parameter_owned entries are set
-  for (int i = 0; i < parameter_owned.size(); i++) {
+  for (std::vector<bool>::size_type i = 0; i < parameter_owned.size(); i++) {
     CHECK(parameter_owned[i]);
   }
 
@@ -681,7 +683,7 @@ void HloInstruction::CheckFusionInstruction() const {
                                 operands_[param_no]->shape()));
   }
   // Make sure all the parameter_numbers entries were seen
-  for (int i = 0; i < parameter_numbers.size(); i++) {
+  for (std::vector<bool>::size_type i = 0; i < parameter_numbers.size(); i++) {
     CHECK(parameter_numbers[i]);
   }
 
@@ -1384,7 +1386,7 @@ string HloInstruction::ToString(bool compact_operands) const {
   }
   if (!slice_starts_.empty() && !slice_limits_.empty()) {
     std::vector<string> bounds;
-    for (int i = 0; i < slice_starts_.size(); ++i) {
+    for (std::vector<int64>::size_type i = 0; i < slice_starts_.size(); ++i) {
       bounds.push_back(tensorflow::strings::StrCat("[", slice_starts_[i], ":",
                                                    slice_limits_[i], "]"));
     }
@@ -1399,7 +1401,8 @@ string HloInstruction::ToString(bool compact_operands) const {
     const auto append_dims = [&](const std::vector<string>& dims,
                                  const Shape& shape) {
       CHECK_EQ(dims.size(), ShapeUtil::Rank(shape));
-      for (int64 logical = 0; logical < dims.size(); ++logical) {
+      for (std::vector<string>::size_type logical = 0; logical < dims.size();
+           ++logical) {
         int64 physical = logical;
         if (!shape.layout().minor_to_major().empty()) {
           physical = LayoutUtil::Major(shape.layout(), logical);
@@ -1812,7 +1815,8 @@ namespace {
 bool OrderIsTopologicalSort(const std::vector<const HloInstruction*>& order) {
   // Create a map from instruction to its position in 'order'.
   std::unordered_map<const HloInstruction*, int> order_position;
-  for (int i = 0; i < order.size(); i++) {
+  for (std::vector<const HloInstruction*>::size_type i = 0; i < order.size();
+       i++) {
     if (!order_position.insert({order[i], i}).second) {
       // Instruction order[i] is duplicated in the order.
       return false;
@@ -2059,7 +2063,8 @@ HloInstruction::UseKind HloInstruction::OperandElementUse(int64 i) const {
               return UseKind::kUse;
             }
             if (cache.count(&hlo) == 0) {
-              for (int64 j = 0; j < hlo.operands_.size(); ++j) {
+              for (std::vector<HloInstruction*>::size_type j = 0;
+                   j < hlo.operands_.size(); ++j) {
                 UseKind old = cache[&hlo];
                 UseKind updated = plus(
                     old, std::min(hlo.OperandElementUse(j),

tensorflow/compiler/xla/service/hlo_ordering.cc

@@ -90,7 +90,8 @@ SequentialHloOrdering::SequentialHloOrdering(
   // Create a map from instruction to its order position.
   for (auto computation_order : module_sequence_) {
     const std::vector<const HloInstruction*>& order = computation_order.second;
-    for (int i = 0; i < order.size(); ++i) {
+    for (std::vector<const HloInstruction*>::size_type i = 0; i < order.size();
+         ++i) {
       DCHECK_EQ(0, order_position_.count(order[i]));
       order_position_.emplace(order[i], i);
     }

tensorflow/compiler/xla/service/llvm_ir/ir_array.cc

@@ -200,7 +200,7 @@ llvm::Value* IrArray::EmitArrayElementAddress(
   // We perform broadcasting when the operand shape has dimension(s) of size
   // 1. In this case we fix the index value for that dimension to zero. This
   // effectively broadcasts along this dimension.
-  for (int64 i = 0; i < index.size(); ++i) {
+  for (size_t i = 0; i < index.size(); ++i) {
     auto dim = shape_->dimensions(i);
     actual_index.push_back(dim == 1 ? ir_builder->getInt64(0) : index[i]);
     is_implicit_broadcast |= dim == 1;

tensorflow/compiler/xla/service/service.cc

@@ -348,7 +348,8 @@ StatusOr<std::vector<std::unique_ptr<Executable>>> Service::BuildExecutables(
   const string& directory_path = flags->xla_dump_computations_to;
   const string& other_directory_path = flags->xla_dump_executions_to;
   if ((!directory_path.empty() || !other_directory_path.empty())) {
-    for (int64 i = 0; i < versioned_handles.size(); ++i) {
+    for (std::vector<VersionedComputationHandle>::size_type i = 0;
+         i < versioned_handles.size(); ++i) {
       TF_ASSIGN_OR_RETURN(std::unique_ptr<SessionModule> session_module,
                           computation_tracker_.SnapshotComputation(
                               versioned_handles[i].handle));
@@ -537,7 +538,8 @@ Service::ExecuteParallelAndRegisterResult(
   }
 
   // Wait for all executions to complete.
-  for (int64 i = 0; i < result_handles.size(); ++i) {
+  for (std::vector<GlobalDataHandle>::size_type i = 0;
+       i < result_handles.size(); ++i) {
     if (!streams[i]->BlockHostUntilDone()) {
       return InternalError("failed to complete execution for stream %lld", i);
     }

tensorflow/compiler/xla/service/shape_inference.cc

@@ -1037,7 +1037,8 @@ ShapeInference::InferDegenerateDimensionBroadcastShape(
   }
 
   std::vector<int64> sizes;
-  for (int64 dimension = 0; dimension < starts.size(); ++dimension) {
+  for (tensorflow::gtl::ArraySlice<int64>::size_type dimension = 0;
+       dimension < starts.size(); ++dimension) {
     int64 start_index = starts[dimension];
     int64 limit_index = limits[dimension];
     if (start_index < 0) {
@@ -1110,7 +1111,8 @@ ShapeInference::InferDegenerateDimensionBroadcastShape(
         slice_sizes.size(), ShapeUtil::Rank(operand_shape));
   }
 
-  for (int64 dim = 0; dim < slice_sizes.size(); ++dim) {
+  for (tensorflow::gtl::ArraySlice<int64>::size_type dim = 0;
+       dim < slice_sizes.size(); ++dim) {
     const int64 input_dim_size = operand_shape.dimensions(dim);
     const int64 slice_dim_size = slice_sizes[dim];
     if (slice_dim_size <= 0) {
@@ -1370,7 +1372,8 @@ ShapeInference::InferDegenerateDimensionBroadcastShape(
   }
 
   // All arguments must be compatible with the program shape.
-  for (int i = 0; i < arg_shapes.size(); ++i) {
+  for (tensorflow::gtl::ArraySlice<const Shape*>::size_type i = 0;
+       i < arg_shapes.size(); ++i) {
     const Shape& arg_shape = *arg_shapes[i];
     const Shape& param_shape = to_apply.parameters(i);
     if (!ShapeUtil::Compatible(arg_shape, param_shape)) {

tensorflow/compiler/xla/service/tuple_points_to_analysis.cc

@@ -280,7 +280,8 @@ Status TuplePointsToAnalysis::HandleTuple(
 
   // A tuple contains references to all input operands and transitively any
   // references in those operands.
-  for (int64 i = 0; i < operands.size(); ++i) {
+  for (tensorflow::gtl::ArraySlice<HloInstruction*>::size_type i = 0;
+       i < operands.size(); ++i) {
     const PointsToSet& operand_points_to_set =
         *FindOrDie(points_to_, operands[i]);
 

tensorflow/compiler/xla/shape_util.cc

@@ -643,7 +643,7 @@ bool CompareShapes(const Shape& lhs, const Shape& rhs, bool compare_layouts) {
   // The dimensions in minor_to_major need to be renumbered to account for the
   // degenerate dimensions which have removed. Decrement each dimension number
   // once for each degenerate dimension which has a smaller number.
-  for (int i = 0; i < minor_to_major.size(); ++i) {
+  for (std::vector<int64>::size_type i = 0; i < minor_to_major.size(); ++i) {
     int adjustment = 0;
     for (int64 dim : degenerate_dimensions) {
       if (minor_to_major[i] > dim) {