The previous attempt (dsl_feature_completeness_2019-08-23) to enable arbitrary kernel functions was a failure: we get significant performance loss (25-100%) if step_number is not passed as a template parameter to the integration kernel. Apparently the CUDA compiler cannot perform some optimizations if there is a if/else construct in a performance-critical part which cannot be evaluated at compile time. This branch keeps step_number as a template parameter but takes rest of the user parameters as uniforms (dt is no longer passed as a function parameter but as an uniform with the DSL instead).

acc/mhd_solver/stencil_definition.sdh

@@ -15,6 +15,7 @@ uniform int AC_bin_steps;
 uniform int AC_bc_type;
 
 // Real params
+uniform Scalar AC_dt;
 // Spacing
 uniform Scalar AC_dsx;
 uniform Scalar AC_dsy;

acc/mhd_solver/stencil_process.sps

@@ -294,8 +294,9 @@ out ScalarField out_tt(VTXBUF_TEMPERATURE);
 #endif
 
 Kernel void
-solve(Scalar dt)
+solve()
 {
+    Scalar dt = AC_dt;
     out_lnrho = rk3(out_lnrho, lnrho, continuity(uu, lnrho), dt);
 
 #if LMAGNETIC