Okay, I could reduce it somewhat, it seems the compiler crashes on the constant float3 parameter.
// compiles __kernel void select_dot_float3( const __global float3 * first, const float4 second, // a waste of precious parameter memory! 😉 __global float * result, const uint count ) { uint tid = get_global_id(0); if(tid >= count) return; result[tid] = dot(first[tid], second.xyz); } // crashes __kernel void select_dot_float3( const __global float3 * first, const float3 second, // look here for error! __global float * result, const uint count ) { uint tid = get_global_id(0); if(tid >= count) return; result[tid] = dot(first[tid], second); }
Before Christmas most likely.
