The code produces the correct output on the smaller benchmarks. However on larger benchmarks such as benchmark 5 the code produces the wrong output.
Looking in depth at the code, the error value (which is what determines how many iterations are run for each timestep) decreases at a slower rate than it does on other platforms. This means the program takes a very long time to run, hitting the iteration limit every time step and producing the wrong output. This same code works to produce the correct output on all other tested platforms (Radeon VII, GTX 2080TI, Intel UHD P630 Graphics) for all benchmark sizes.
The code produces the correct output on the smaller benchmarks. However on larger benchmarks such as benchmark 5 the code produces the wrong output.
Looking in depth at the code, the error value (which is what determines how many iterations are run for each timestep) decreases at a slower rate than it does on other platforms. This means the program takes a very long time to run, hitting the iteration limit every time step and producing the wrong output. This same code works to produce the correct output on all other tested platforms (Radeon VII, GTX 2080TI, Intel UHD P630 Graphics) for all benchmark sizes.