AddN for variants adds in a tree structure (pairwise summation)

Improves numerical precision (if applicable) using pairwise summation:
https://en.wikipedia.org/wiki/Pairwise_summation

Thanks to Rasmus Larsen for the succinct binary tree aggregation pseudocode.

Also adds AsString for Variant types: this emits the Variant as a string via
its DebugString().

PiperOrigin-RevId: 340246073
Change-Id: I009281f46cbea30d6e33ecf79a1723d62e96cc6d

Author: ebrevdo

tensorflow/compiler/mlir/tensorflow/ir/tf_generated_ops.td

@@ -509,7 +509,7 @@ array([b'3.14', b'2.72'], dtype=object)
   }];
 
   let arguments = (ins
-    TensorOf<[TF_Bool, TF_Complex128, TF_Complex64, TF_Float32, TF_Float64, TF_Int16, TF_Int32, TF_Int64, TF_Int8]>:$input,
+    TensorOf<[TF_Bool, TF_Complex128, TF_Complex64, TF_Float32, TF_Float64, TF_Int16, TF_Int32, TF_Int64, TF_Int8, TF_Variant]>:$input,
 
     DefaultValuedAttr<I64Attr, "-1">:$precision,
     DefaultValuedAttr<BoolAttr, "false">:$scientific,
@@ -15226,4 +15226,4 @@ execution the transfer corresponds to.}]>:$dynamic_key,
   let results = (outs);
 
   TF_DerivedOperandTypeListAttr Tinputs = TF_DerivedOperandTypeListAttr<0>;
-}
+}

tensorflow/core/kernels/aggregate_ops.h

@@ -370,24 +370,77 @@ class AddNOp<Device, Variant, OpKernelT, OpKernelConstructionT,
               i, " has shape: ", ctx->input(i).shape().DebugString(), "."));
     }
 
-    // Step 2: attempt to add using
+    // Step 2: Sum input variants in a tree-like structure using
     //   BinaryOpVariants(ADD_VARIANT_BINARY_OP, ...)
     //   For the output create a default-constructed variant object.
-    // TODO(ebrevdo): Perform summation in a tree-structure.
-    Tensor out(cpu_allocator(), DT_VARIANT, TensorShape({}));
-    Variant* v_out = &(out.scalar<Variant>()());
-    OP_REQUIRES_OK(ctx, BinaryOpVariants<Device>(
-                            ctx, ADD_VARIANT_BINARY_OP,
-                            ctx->input(0).template scalar<Variant>()(),
-                            ctx->input(1).template scalar<Variant>()(), v_out));
-    for (int i = 2; i < num; ++i) {
-      const Variant tmp = std::move(*v_out);
-      const Variant& inp = ctx->input(i).template scalar<Variant>()();
-      OP_REQUIRES_OK(ctx, BinaryOpVariants<Device>(ctx, ADD_VARIANT_BINARY_OP,
-                                                   inp, tmp, v_out));
+    //
+    // Pairwise summation provides better numerical precision by
+    // reducing round-off error:
+    //
+    //   https://en.wikipedia.org/wiki/Pairwise_summation
+    //
+    // These two vectors are used to store and mark intermediate sums.
+    gtl::InlinedVector<bool, 4> temp_filled(num, false);
+    gtl::InlinedVector<Variant, 4> temp(num);
+
+    // Tree-based summation.
+    int skip = 1;
+    int n = num;
+    while (skip < n) {
+      int i = skip;
+      while (i < n) {
+        // TODO(ebrevdo, rmlarsen): Parallelize the pairwise summations in the
+        // inner loop if the variants are "large".
+
+        // x[i - skip] += x[i]
+        OP_REQUIRES_OK(ctx,
+                       AddVariantTo(ctx, i - skip, i, &temp, &temp_filled));
+        // We won't use this index again, recover its memory.
+        temp[i].clear();
+        i += 2 * skip;
+      }
+      if (i == n) {
+        // x[0] += x[i - skip]
+        OP_REQUIRES_OK(ctx,
+                       AddVariantTo(ctx, 0, i - skip, &temp, &temp_filled));
+        // We won't use this index again, recover its memory.
+        temp[i - skip].clear();
+        n -= skip;
+      }
+      skip *= 2;
     }
+
+    Tensor out(cpu_allocator(), DT_VARIANT, TensorShape({}));
+    out.scalar<Variant>()() = std::move(temp[0]);
     ctx->set_output(0, out);
   }
+
+ private:
+  // AddVariantTo efficiently performs:
+  //    temp[lhs_ix] <- array(lhs_ix) + array(rhs_ix)
+  // where array(ix) := (temp_filled[ix]
+  //                     ? temp[ix]
+  //                     : ctx->input(ix).scalar<Variant>()())
+  // This reduces (possibly expensive) copying of Variants from
+  // the inputs into temp at the lowest levels of the summation tree.
+  static inline Status AddVariantTo(OpKernelContextT* ctx, const int lhs_ix,
+                                    const int rhs_ix,
+                                    gtl::InlinedVector<Variant, 4>* temp,
+                                    gtl::InlinedVector<bool, 4>* temp_filled) {
+    Variant tmp;
+    if (temp_filled->at(lhs_ix)) tmp = std::move(temp->at(lhs_ix));
+    const Variant& a = temp_filled->at(lhs_ix)
+                           ? tmp
+                           : ctx->input(lhs_ix).template scalar<Variant>()();
+    const Variant& b = temp_filled->at(rhs_ix)
+                           ? temp->at(rhs_ix)
+                           : ctx->input(rhs_ix).template scalar<Variant>()();
+    Variant* c = &temp->at(lhs_ix);
+    TF_RETURN_IF_ERROR(
+        BinaryOpVariants<Device>(ctx, ADD_VARIANT_BINARY_OP, a, b, c));
+    temp_filled->at(lhs_ix) = true;
+    return Status::OK();
+  }
 };
 
 }  // namespace tensorflow

tensorflow/core/kernels/as_string_op.cc

@@ -20,6 +20,9 @@ limitations under the License.
 #include "tensorflow/core/framework/kernel_def_builder.h"
 #include "tensorflow/core/framework/op_kernel.h"
 #include "tensorflow/core/framework/tensor.h"
+#include "tensorflow/core/framework/variant.h"
+#include "tensorflow/core/framework/variant_encode_decode.h"
+#include "tensorflow/core/framework/variant_tensor_data.h"
 #include "tensorflow/core/lib/core/errors.h"
 #include "tensorflow/core/lib/core/status.h"
 #include "tensorflow/core/lib/strings/stringprintf.h"
@@ -112,6 +115,8 @@ class AsStringOp : public OpKernel {
         break;
       case DT_BOOL:
         break;
+      case DT_VARIANT:
+        break;
       default:
         bool type_not_supported = true;
         OP_REQUIRES(ctx, !type_not_supported,
@@ -156,6 +161,12 @@ class AsStringOp : public OpKernel {
           output_flat(i) = (input_flat(i)) ? "true" : "false";
         }
       } break;
+      case (DT_VARIANT): {
+        const auto& input_flat = input_tensor->flat<Variant>();
+        for (int i = 0; i < input_flat.size(); ++i) {
+          output_flat(i) = input_flat(i).DebugString();
+        }
+      } break;
       case (DT_COMPLEX64): {
         const auto& input_flat = input_tensor->flat<complex64>();
         for (int i = 0; i < input_flat.size(); ++i) {

tensorflow/core/kernels/as_string_op_test.cc

@@ -18,6 +18,9 @@ limitations under the License.
 #include "tensorflow/core/framework/tensor.h"
 #include "tensorflow/core/framework/tensor_testutil.h"
 #include "tensorflow/core/framework/types.h"
+#include "tensorflow/core/framework/variant.h"
+#include "tensorflow/core/framework/variant_encode_decode.h"
+#include "tensorflow/core/framework/variant_tensor_data.h"
 #include "tensorflow/core/kernels/ops_testutil.h"
 #include "tensorflow/core/kernels/ops_util.h"
 #include "tensorflow/core/lib/core/status_test_util.h"
@@ -148,6 +151,25 @@ TEST_F(AsStringGraphTest, Bool) {
   test::ExpectTensorEqual<tstring>(expected, *GetOutput(0));
 }
 
+TEST_F(AsStringGraphTest, Variant) {
+  TF_ASSERT_OK(Init(DT_VARIANT));
+
+  AddInput(DT_VARIANT, TensorShape({4}));
+  auto inputs = mutable_input(0)->flat<Variant>();
+  inputs(0) = 2;
+  inputs(1) = 3;
+  inputs(2) = true;
+  inputs(3) = Tensor("hi");
+  TF_ASSERT_OK(RunOpKernel());
+  Tensor expected(allocator(), DT_STRING, TensorShape({4}));
+  test::FillValues<tstring>(
+      &expected, {"Variant<type: int value: 2>", "Variant<type: int value: 3>",
+                  "Variant<type: bool value: 1>",
+                  ("Variant<type: tensorflow::Tensor value: Tensor<type: string"
+                   " shape: [] values: hi>>")});
+  test::ExpectTensorEqual<tstring>(expected, *GetOutput(0));
+}
+
 TEST_F(AsStringGraphTest, String) {
   Status s = Init(DT_STRING);
   ASSERT_EQ(error::INVALID_ARGUMENT, s.code());

tensorflow/core/ops/string_ops.cc

@@ -116,7 +116,7 @@ REGISTER_OP("AsString")
     .Output("output: string")
     .Attr(
         "T: {int8, int16, int32, int64, complex64, complex128, float, double, "
-        "bool}")
+        "bool, variant}")
     .Attr("precision: int = -1")
     .Attr("scientific: bool = false")
     .Attr("shortest: bool = false")

tensorflow/python/kernel_tests/BUILD

@@ -1606,6 +1606,7 @@ cuda_py_test(
         "//tensorflow/python:client_testlib",
         "//tensorflow/python:framework_for_generated_wrappers",
         "//tensorflow/python:math_ops",
+        "//tensorflow/python:string_ops",
         "//third_party/py/numpy",
     ],
 )

tensorflow/python/kernel_tests/aggregate_ops_test.py

@@ -26,8 +26,8 @@
 from tensorflow.python.framework import tensor_shape
 from tensorflow.python.framework import test_util
 from tensorflow.python.ops import array_ops
-from tensorflow.python.ops import logging_ops
 from tensorflow.python.ops import math_ops
+from tensorflow.python.ops import string_ops
 from tensorflow.python.platform import test
 
 
@@ -100,24 +100,28 @@ def create_constant_variant(value):
     # TODO(ebrevdo): Re-enable use_gpu=True once non-DMA Variant
     # copying between CPU and GPU is supported.
     with self.session(use_gpu=False):
-      variant_const_3 = create_constant_variant(3)
-      variant_const_4 = create_constant_variant(4)
-      variant_const_5 = create_constant_variant(5)
-      # 3 + 3 + 5 + 4 = 15.
-      result = math_ops.add_n((variant_const_3, variant_const_3,
-                               variant_const_5, variant_const_4))
+      num_tests = 127
+      values = list(range(100))
+      variant_consts = [create_constant_variant(x) for x in values]
+      sum_count_indices = np.random.randint(1, 29, size=num_tests)
+      sum_indices = [
+          np.random.randint(100, size=count) for count in sum_count_indices]
+      expected_sums = [np.sum(x) for x in sum_indices]
+      variant_sums = [math_ops.add_n([variant_consts[i] for i in x])
+                      for x in sum_indices]
 
-      # Smoke test -- ensure this executes without trouble.
+      # We use as_string() to get the Variant DebugString for the
+      # variant_sums; we know its value so we can check via string equality
+      # here.
+      #
       # Right now, non-numpy-compatible objects cannot be returned from a
       # session.run call; similarly, objects that can't be converted to
       # native numpy types cannot be passed to ops.convert_to_tensor.
-      # For now, run the test and examine the output to see that the result is
-      # equal to 15.
-      result_op = logging_ops.Print(
-          result, [variant_const_3, variant_const_4, variant_const_5, result],
-          message=("Variants stored an int: c(3), c(4), c(5), "
-                   "add_n(c(3), c(3), c(5), c(4)): ")).op
-      result_op.run()
+      variant_sums_string = string_ops.as_string(variant_sums)
+      self.assertAllEqual(
+          variant_sums_string,
+          ["Variant<type: int value: {}>".format(s).encode("utf-8")
+           for s in expected_sums])
 
 
 if __name__ == "__main__":