Method RunR2ToR1Test

tensorflow/compiler/xla/tests/reduce_test.cc:230–256

Runs an R2 => R1 reduction test with the given number of (rows, cols).

  // Runs an R2 => R1 reduction test with the given number of (rows, cols).
  void RunR2ToR1Test(int64 rows, int64 cols, int64 minor = 1, int64 major = 0) {
    XlaBuilder builder(TestName());
    XlaComputation add_f32 = CreateScalarAddComputation(F32, &builder);
    const Shape input_shape = ShapeUtil::MakeShape(F32, {rows, cols});
    auto input = Parameter(&builder, 0, input_shape, "input");
    auto zero = ConstantR0<float>(&builder, 0.0);
    Reduce(input, zero, add_f32, /*dimensions_to_reduce=*/{0});

    Array2D<float> input_data(rows, cols);
    input_data.FillRandom(3.14f, 0.04);
    Literal input_literal = LiteralUtil::CreateR2FromArray2D(input_data);
    input_literal =
        input_literal.Relayout(LayoutUtil::MakeLayout({minor, major}));
    std::unique_ptr<GlobalData> input_global_data =
        client_->TransferToServer(input_literal).ConsumeValueOrDie();

    std::vector<float> expected;
    for (int64 colno = 0; colno < cols; ++colno) {
      float column_sum = 0;
      for (int64 rowno = 0; rowno < rows; ++rowno) {
        column_sum += input_data(rowno, colno);
      }
      expected.push_back(column_sum);
    }
    ComputeAndCompareR1<float>(&builder, expected, {input_global_data.get()},
                               ErrorSpec(0.01, 1e-4));
  }

  template <typename NativeT>
  void ComputeAndCompareGeneric(