Author: Shengqi Chen <harry-chen@outlook.com>
Description: Fix compilation with Eigen 3.4.0
 onnxruntime >= 1.19.0 invokes eigen3 with PropagateNaN, which is not supported in eigen 3.4.0.
 This patch removes PropagateNaN from code.
 See: https://github.com/microsoft/onnxruntime/commit/7543dd040b2d32109a2718d7276d3aca1edadaae
 See: https://gitlab.com/libeigen/eigen/-/commit/5d918b82a80118ebb19572770a0c8e1f5fe06b91
Forwarded: not-needed

--- a/onnxruntime/core/providers/cpu/math/element_wise_ops.cc
+++ b/onnxruntime/core/providers/cpu/math/element_wise_ops.cc
@@ -813,7 +813,7 @@
   for (int index = 1; index < inputCount; index++) {
     auto& data_n = *ctx->Input<Tensor>(index);
     ORT_ENFORCE(data_n.Shape() == shape, "All inputs must have the same shape");
-    min = min.array().template min<Eigen::PropagateNaN>(EigenMap<float>(data_n).array());
+    min = min.array().min(EigenMap<float>(data_n).array());
   }
 
   return Status::OK();
@@ -829,15 +829,15 @@
     ProcessBroadcastSpanFuncs funcs{
         [](BroadcastHelper& per_iter_bh) {
           per_iter_bh.OutputEigen<T>() =
-              per_iter_bh.EigenInput1<T>().array().template min<Eigen::PropagateNaN>(per_iter_bh.ScalarInput0<T>());
+              per_iter_bh.EigenInput1<T>().array().min(per_iter_bh.ScalarInput0<T>());
         },
         [](BroadcastHelper& per_iter_bh) {
           per_iter_bh.OutputEigen<T>() =
-              per_iter_bh.EigenInput0<T>().array().template min<Eigen::PropagateNaN>(per_iter_bh.ScalarInput1<T>());
+              per_iter_bh.EigenInput0<T>().array().min(per_iter_bh.ScalarInput1<T>());
         },
         [](BroadcastHelper& per_iter_bh) {
           per_iter_bh.OutputEigen<T>() =
-              per_iter_bh.EigenInput0<T>().array().template min<Eigen::PropagateNaN>(
+              per_iter_bh.EigenInput0<T>().array().min(
                   per_iter_bh.EigenInput1<T>().array());
         }};
 
@@ -865,10 +865,10 @@
         EigenVectorArrayMap<Eigen::half> output_vec_map(output, num_elements);
 
         if (is_min) {
-          output_vec_map = input_1_vec_map.template min<Eigen::PropagateNaN>(
+          output_vec_map = input_1_vec_map.min(
               static_cast<Eigen::half>(per_iter_bh.ScalarInput0<MLFloat16>()));
         } else {
-          output_vec_map = input_1_vec_map.template max<Eigen::PropagateNaN>(
+          output_vec_map = input_1_vec_map.max(
               static_cast<Eigen::half>(per_iter_bh.ScalarInput0<MLFloat16>()));
         }
       },
@@ -882,10 +882,10 @@
         EigenVectorArrayMap<Eigen::half> output_vec_map(output, num_elements);
 
         if (is_min) {
-          output_vec_map = input_0_vec_map.template min<Eigen::PropagateNaN>(
+          output_vec_map = input_0_vec_map.min(
               static_cast<Eigen::half>(per_iter_bh.ScalarInput1<MLFloat16>()));
         } else {
-          output_vec_map = input_0_vec_map.template max<Eigen::PropagateNaN>(
+          output_vec_map = input_0_vec_map.max(
               static_cast<Eigen::half>(per_iter_bh.ScalarInput1<MLFloat16>()));
         }
       },
@@ -902,9 +902,9 @@
         EigenVectorArrayMap<Eigen::half> output_vec_map(output, num_elements);
 
         if (is_min) {
-          output_vec_map = input_0_vec_map.template min<Eigen::PropagateNaN>(input_1_vec_map);
+          output_vec_map = input_0_vec_map.min(input_1_vec_map);
         } else {
-          output_vec_map = input_0_vec_map.template max<Eigen::PropagateNaN>(input_1_vec_map);
+          output_vec_map = input_0_vec_map.max(input_1_vec_map);
         }
       }};
 
@@ -940,7 +940,7 @@
   for (int index = 1; index < inputCount; index++) {
     auto& data_n = *ctx->Input<Tensor>(index);
     ORT_ENFORCE(data_n.Shape() == shape, "All inputs must have the same shape");
-    max = max.array().template max<Eigen::PropagateNaN>(EigenMap<float>(data_n).array());
+    max = max.array().max(EigenMap<float>(data_n).array());
   }
 
   return Status::OK();
@@ -956,15 +956,15 @@
     ProcessBroadcastSpanFuncs funcs{
         [](BroadcastHelper& per_iter_bh) {
           per_iter_bh.OutputEigen<T>() =
-              per_iter_bh.EigenInput1<T>().array().template max<Eigen::PropagateNaN>(per_iter_bh.ScalarInput0<T>());
+              per_iter_bh.EigenInput1<T>().array().max(per_iter_bh.ScalarInput0<T>());
         },
         [](BroadcastHelper& per_iter_bh) {
           per_iter_bh.OutputEigen<T>() =
-              per_iter_bh.EigenInput0<T>().array().template max<Eigen::PropagateNaN>(per_iter_bh.ScalarInput1<T>());
+              per_iter_bh.EigenInput0<T>().array().max(per_iter_bh.ScalarInput1<T>());
         },
         [](BroadcastHelper& per_iter_bh) {
           per_iter_bh.OutputEigen<T>() =
-              per_iter_bh.EigenInput0<T>().array().template max<Eigen::PropagateNaN>(
+              per_iter_bh.EigenInput0<T>().array().max(
                   per_iter_bh.EigenInput1<T>().array());
         }};