Please see this example where the key computation, the sine of a radian, was moved to an external library, cordic_sine: Simple summation 8x slower than in Julia - #44 by FortranFan.
Once that was done, note in all the subsequent scenarios, the ratio of compute times with Fortran vs Julia were around unity, meaning no real difference i.e., the specific advantages with an enthusiast-driven macro implementation toward vectorization and specific “fast” trig functions leading to the original 8x difference became inapplicable.