diff --git a/RTCP/Cobalt/GPUProc/share/gpu/kernels/FIR_Filter.cu b/RTCP/Cobalt/GPUProc/share/gpu/kernels/FIR_Filter.cu
index 875a7befcaf536c243d69b325f3cf0d56b564168..654616eaaac7bca82e783322fc5c5c3b9dd6a3a2 100644
--- a/RTCP/Cobalt/GPUProc/share/gpu/kernels/FIR_Filter.cu
+++ b/RTCP/Cobalt/GPUProc/share/gpu/kernels/FIR_Filter.cu
@@ -106,7 +106,7 @@ typedef const float (*WeightsType)[NR_CHANNELS][NR_TAPS];
 // this is faster than doing "pol ? sin(phi) : cos(phi)"
 // because that statement forces CUDA to still compute both
 // as GPUs always compute both branches.
-inline __device__ float sincos_d2f_select(float phi, int ri)
+inline __device__ float sincos_d2f_select(double phi, int ri)
 {
   double r[2];