From dc26459b9910d8c1fda964917635ee8277dd2614 Mon Sep 17 00:00:00 2001
From: Benoit Steiner <benoit.steiner.goog@gmail.com>
Date: Fri, 19 Feb 2016 15:16:54 +0000
Subject: Implemented protate() for CUDA

---
 Eigen/src/Core/arch/CUDA/PacketMath.h | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/Eigen/src/Core/arch/CUDA/PacketMath.h b/Eigen/src/Core/arch/CUDA/PacketMath.h
index d5dcc7fa3..a32b41e18 100644
--- a/Eigen/src/Core/arch/CUDA/PacketMath.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMath.h
@@ -272,6 +272,35 @@ template<> EIGEN_DEVICE_FUNC inline double predux_mul<double2>(const double2& a)
   return a.x * a.y;
 }
 
+template<size_t offset>
+struct protate_impl<offset, float4>
+{
+  static float4 run(const float4& a) {
+    if (offset == 0) {
+      return make_float4(a.x, a.y, a.z, a.w);
+    }
+    if (offset == 1) {
+      return make_float4(a.w, a.x, a.y, a.z);
+    }
+    if (offset == 2) {
+      return make_float4(a.z, a.w, a.x, a.y);
+    }
+    return make_float4(a.y, a.z, a.w, a.x);
+  }
+};
+
+template<size_t offset>
+struct protate_impl<offset, double2>
+{
+  static double2 run(const double2& a) {
+    if (offset == 0) {
+      return make_double2(a.x, a.y);
+    }
+    return make_double2(a.y, a.x);
+  }
+};
+
+
 template<> EIGEN_DEVICE_FUNC inline float4  pabs<float4>(const float4& a) {
   return make_float4(fabsf(a.x), fabsf(a.y), fabsf(a.z), fabsf(a.w));
 }
-- 
cgit v1.2.3