From f6c8cc0e99c48cd8ebe3c5a94fad07273c0e91ea Mon Sep 17 00:00:00 2001
From: Antonio Sanchez <cantonios@google.com>
Date: Fri, 29 Oct 2021 22:03:44 +0000
Subject: [PATCH] Fix TensorReduction warnings and error bound for sum accuracy
 test.

The sum accuracy test currently uses the default test precision for
the given scalar type.  However, scalars are generated via a normal
distribution, and given a large enough count and strong enough random
generator, the expected sum is zero.  This causes the test to
periodically fail.

Here we estimate an upper-bound for the error as `sqrt(N) * prec` for
summing N values, with each having an approximate epsilon of `prec`.

Also fixed a few warnings generated by MSVC when compiling the
reduction test.
---
 unsupported/Eigen/CXX11/src/Tensor/Tensor.h          | 2 +-
 unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h | 4 ++--
 unsupported/test/cxx11_tensor_reduction.cpp          | 6 +++++-
 3 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/unsupported/Eigen/CXX11/src/Tensor/Tensor.h b/unsupported/Eigen/CXX11/src/Tensor/Tensor.h
index 76e97cd1a..26358d545 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/Tensor.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/Tensor.h
@@ -76,7 +76,7 @@ class Tensor : public TensorBase<Tensor<Scalar_, NumIndices_, Options_, IndexTyp
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
     enum {
-      IsAligned = bool(EIGEN_MAX_ALIGN_BYTES>0) & !(Options_&DontAlign),
+      IsAligned = (EIGEN_MAX_ALIGN_BYTES>0) && !(Options_&DontAlign),
       Layout = Options_ & RowMajor ? RowMajor : ColMajor,
       CoordAccess = true,
       RawAccess = true
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
index b5922477a..2939b9862 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
@@ -633,7 +633,7 @@ static const bool RunningOnGPU = false;
           m_fastOutputStrides[i] = internal::TensorIntDivisor<Index>(m_outputStrides[i]);
         }
       } else {
-        m_outputStrides[NumOutputDims - 1] = 1;
+        m_outputStrides[static_cast<size_t>(NumOutputDims - 1)] = 1;
         for (int i = NumOutputDims - 2; i >= 0; --i) {
           m_outputStrides[i] = m_outputStrides[i + 1] * m_dimensions[i + 1];
           m_fastOutputStrides[i] = internal::TensorIntDivisor<Index>(m_outputStrides[i]);
@@ -680,7 +680,7 @@ static const bool RunningOnGPU = false;
             ? internal::array_prod(input_dims)
             : (static_cast<int>(Layout) == static_cast<int>(ColMajor))
                   ? m_preservedStrides[0]
-                  : m_preservedStrides[NumOutputDims - 1];
+                  : m_preservedStrides[static_cast<size_t>(NumOutputDims - 1)];
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
diff --git a/unsupported/test/cxx11_tensor_reduction.cpp b/unsupported/test/cxx11_tensor_reduction.cpp
index b7611d7b0..afcb79570 100644
--- a/unsupported/test/cxx11_tensor_reduction.cpp
+++ b/unsupported/test/cxx11_tensor_reduction.cpp
@@ -504,7 +504,11 @@ void test_sum_accuracy() {
     for (int i = 0; i < num_elements; ++i) {
       expected_sum += static_cast<double>(tensor(i));
     }
-    VERIFY_IS_APPROX(sum(), static_cast<ScalarType>(expected_sum));
+    // Scale tolerance to account for # elements.  Otherwise, we periodically fail, since
+    // E[sum] == prescribed_mean == 0 for the first iteration.
+    double err = Eigen::numext::abs(static_cast<double>(sum()) - expected_sum);
+    double tol = Eigen::numext::sqrt(num_elements) * static_cast<double>(test_precision<ScalarType>()) * numext::maxi(1.0, prescribed_mean);
+    VERIFY(err < tol);
   }
 }