diff --git a/Eigen/src/Core/products/GeneralBlockPanelKernel.h b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
--- a/Eigen/src/Core/products/GeneralBlockPanelKernel.h
+++ b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
@@ -1740,24 +1740,24 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Sca
       const Scalar* b3 = &rhs[(j2+3)*rhsStride];
       
       Index k=0;
       if((PacketSize%4)==0) // TODO enbale vectorized transposition for PacketSize==2 ??
       {
         for(; k<peeled_k; k+=PacketSize) {
           PacketBlock<Packet,(PacketSize%4)==0?4:PacketSize> kernel;
           kernel.packet[0] = ploadu<Packet>(&b0[k]);
-          kernel.packet[1] = ploadu<Packet>(&b1[k]);
-          kernel.packet[2] = ploadu<Packet>(&b2[k]);
-          kernel.packet[3] = ploadu<Packet>(&b3[k]);
+          kernel.packet[1%PacketSize] = ploadu<Packet>(&b1[k]);
+          kernel.packet[2%PacketSize] = ploadu<Packet>(&b2[k]);
+          kernel.packet[3%PacketSize] = ploadu<Packet>(&b3[k]);
           ptranspose(kernel);
           pstoreu(blockB+count+0*PacketSize, cj.pconj(kernel.packet[0]));
-          pstoreu(blockB+count+1*PacketSize, cj.pconj(kernel.packet[1]));
-          pstoreu(blockB+count+2*PacketSize, cj.pconj(kernel.packet[2]));
-          pstoreu(blockB+count+3*PacketSize, cj.pconj(kernel.packet[3]));
+          pstoreu(blockB+count+1*PacketSize, cj.pconj(kernel.packet[1%PacketSize]));
+          pstoreu(blockB+count+2*PacketSize, cj.pconj(kernel.packet[2%PacketSize]));
+          pstoreu(blockB+count+3*PacketSize, cj.pconj(kernel.packet[3%PacketSize]));
           count+=4*PacketSize;
         }
       }
       for(; k<depth; k++)
       {
         blockB[count+0] = cj(b0[k]);
         blockB[count+1] = cj(b1[k]);
         blockB[count+2] = cj(b2[k]);