Fix for inexact division sizes

Author: utkarsh530

src/kernels.jl

@@ -66,7 +66,11 @@ end
 
     group_particles = @localmem SPSOGBest{T1, T2} (gs)
 
-    i > length(gpu_particles) && return
+    if tidx == 1
+        fill!(group_particles, SPSOGBest(gbest.position, convert(typeof(gbest.cost), Inf)))
+    end
+
+    @synchronize
 
     @inbounds particle = gpu_particles[i]
 

src/lowerlevel_solve.jl

@@ -8,7 +8,7 @@ function vectorized_solve!(prob,
     backend = get_backend(gpu_particles)
 
     ## TODO: Get dynamic workgroupsize
-    workgroupsize = (64,)
+    workgroupsize = (min(length(gpu_particles), 1024),)
 
     update_particle_kernel = update_particle_states!(backend, workgroupsize)