It may be better to find a way to share this code between the smooth brush and here. So, maybe:

• move this function to gpencil_utils.c as "bool gp_smooth_stroke(bGDstroke *gps, int i, float inf)"
• add a "bool gp_smooth_stroke(bGDstroke *gps, int i, float inf);" to gpencil_intern.h
• use "gp_smooth_stroke" in this file
• simplify the gp_brush_smooth_apply code to:

static bool gp_brush_smooth_apply(tGP_BrushEditData *gso, bGPDstroke *gps, int i,
                                  const int radius, const int co[2])
{
	GP_EditBrush_Data *brush = gso->brush;
	bGPDspoint *pt = &gps->points[i];
	float inf = gp_brush_influence_calc(gso, radius, co);

        /* perform smoothing */
        return gp_smooth_stroke(gps, i, inf);
}

"subpoints" is not really that descriptive. It was only upon re-reading the code that I realised that this is the new total number of points in the subdivided stroke. Maybe something like "new_totpoints" or something like that instead?

Is there any reason why you need to have these on the stack, and thus end up copying the data back and forth? It would probably be more efficient to just do:

bGPDspoint *pt_a;
bGPDspoint *pt_b;
bGPDspoint *pt_new;

then:

 pt_a = &gps->points[i];
 pt_b = &gps->points[i];
 pt_n = &gps->points[i + 1];

 /* Interpolate all values */
interp_v3_v3(&pt_n->x, &pt_a->x, &pt_b->x, 0.5f);
interpf(&pt_n->pressure, pt_a->pressure, pt_b->pressure);
// ...

Also note, the use of the interp functions from BLI_math instead of inlining the math here :)

AFAICT, there isn't really any reason why we need to make copies of everything first. It doesn't look like the ranges will overlap...