Implement pbroadcast4 on altivec

2 files changed, 25 insertions, 3 deletions

diff --git a/Eigen/src/Core/arch/AltiVec/Complex.h b/Eigen/src/Core/arch/AltiVec/Complex.h
index 8fdffad5e..5409ddedd 100644
--- a/Eigen/src/Core/arch/AltiVec/Complex.h
+++ b/Eigen/src/Core/arch/AltiVec/Complex.h
@@ -229,7 +229,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pcplxflip<Packet2cf>(const Packet2cf& x
   return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX_REV));
 }
 
-template<> EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
 {
   Packet4f tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_COMPLEX_TRANSPOSE_0);
   kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_COMPLEX_TRANSPOSE_1);
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
index 8a67354e4..0e9adf450 100755
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -168,6 +168,28 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   {
   return vc;
 }
 
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = vec_ld(0,a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4i>(const int *a,
+                      Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3)
+{
+  a3 = vec_ld(0,a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
+
 template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, int stride)
 {
   float EIGEN_ALIGN16 af[4];
@@ -538,7 +560,7 @@ struct palign_impl<Offset,Packet4i>
   }
 };
 
-template<> EIGEN_DEVICE_FUNC inline void
+EIGEN_DEVICE_FUNC inline void
 ptranspose(PacketBlock<Packet4f,4>& kernel) {
   Packet4f t0, t1, t2, t3;
   t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
@@ -551,7 +573,7 @@ ptranspose(PacketBlock<Packet4f,4>& kernel) {
   kernel.packet[3] = vec_mergel(t1, t3);
 }
 
-template<> EIGEN_DEVICE_FUNC inline void
+EIGEN_DEVICE_FUNC inline void
 ptranspose(PacketBlock<Packet4i,4>& kernel) {
   Packet4i t0, t1, t2, t3;
   t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);