Merge remote-tracking branch 'thomas/GS_WP' into neuralpint

Author: tlunet

pySDC/implementations/convergence_controller_classes/step_size_limiter.py

@@ -192,12 +192,7 @@ def _round_step_size(dt, fac, digits):
 
     def get_new_step_size(self, controller, S, **kwargs):
         """
-        Enforce an upper and lower limit to the step size here.
-        Be aware that this is only tested when a new step size has been determined. That means if you set an initial
-        value for the step size outside of the limits, and you don't do any further step size control, that value will
-        go through.
-        Also, the final step is adjusted such that we reach Tend as best as possible, which might give step sizes below
-        the lower limit set here.
+        Round step size here
 
         Args:
             controller (pySDC.Controller): The controller

pySDC/implementations/problem_classes/RayleighBenard.py

@@ -131,8 +131,10 @@ def __init__(
         self.Dz = S1 @ Dz
         self.Dzz = S2 @ Dzz
 
-        kappa = (Rayleigh * Prandtl) ** (-1 / 2.0)
-        nu = (Rayleigh / Prandtl) ** (-1 / 2.0)
+        # compute rescaled Rayleigh number to extract viscosity and thermal diffusivity
+        Ra = Rayleigh / (abs(BCs['T_top'] - BCs['T_bottom']) * self.axes[1].L ** 3)
+        kappa = (Ra * Prandtl) ** (-1 / 2.0)
+        nu = (Ra / Prandtl) ** (-1 / 2.0)
 
         # construct operators
         L_lhs = {

pySDC/projects/GPU/analysis_scripts/parallel_scaling.py

@@ -134,6 +134,7 @@ def plot_scaling_test(self, ax, quantity='time', **plotting_params):  # pragma:
                         timing_step = get_sorted(stats, type='timing_step')
 
                     t_mean = np.mean([me[1] for me in timing_step])
+                    t_min = np.min([me[1] for me in timing_step][1:])
 
                     if quantity == 'throughput':
                         timings[np.prod(procs) / self.tasks_per_node] = experiment.res**self.ndim / t_mean
@@ -147,6 +148,8 @@ def plot_scaling_test(self, ax, quantity='time', **plotting_params):  # pragma:
                         timings[np.prod(procs) / self.tasks_per_node] = t_mean
                     elif quantity == 'time_per_task':
                         timings[np.prod(procs)] = t_mean
+                    elif quantity == 'min_time_per_task':
+                        timings[np.prod(procs)] = t_min
                     else:
                         raise NotImplementedError
                 except (FileNotFoundError, ValueError):
@@ -167,6 +170,7 @@ def plot_scaling_test(self, ax, quantity='time', **plotting_params):  # pragma:
             'throughput_per_task': 'throughput / DoF/s',
             'time': r'$t_\mathrm{step}$ / s',
             'time_per_task': r'$t_\mathrm{step}$ / s',
+            'min_time_per_task': r'minimal $t_\mathrm{step}$ / s',
             'efficiency': 'efficiency / DoF/s/task',
         }
         ax.set_ylabel(labels[quantity])
@@ -358,6 +362,7 @@ def plot_scalings(problem, **kwargs):  # pragma: no cover
         ('RBC', 'throughput'): {'x': [1 / 10, 64], 'y': [2e4, 2e4 * 640]},
         ('RBC', 'time'): {'x': [1 / 10, 64], 'y': [60, 60 / 640]},
         ('RBC', 'time_per_task'): {'x': [1, 640], 'y': [60, 60 / 640]},
+        ('RBC', 'min_time_per_task'): {'x': [1, 640], 'y': [60, 60 / 640]},
         ('RBC', 'throughput_per_task'): {'x': [1 / 1, 640], 'y': [2e4, 2e4 * 640]},
     }
 
@@ -373,7 +378,7 @@ def plot_scalings(problem, **kwargs):  # pragma: no cover
     fig.savefig(path, bbox_inches='tight')
     print(f'Saved {path!r}', flush=True)
 
-    for quantity in ['time', 'throughput', 'time_per_task', 'throughput_per_task'][::-1]:
+    for quantity in ['time', 'throughput', 'time_per_task', 'throughput_per_task', 'min_time_per_task'][::-1]:
         fig, ax = plt.subplots(figsize=figsize_by_journal('TUHH_thesis', 1, 0.6))
         for config in configs:
             config.plot_scaling_test(ax=ax, quantity=quantity)

pySDC/projects/GPU/analysis_scripts/plot_RBC_matrix.py

@@ -61,6 +61,36 @@ def plot_ultraspherical():
     plt.show()
 
 
+def plot_DCT():
+    fig, axs = plt.subplots(1, 3, figsize=figsize_by_journal('TUHH_thesis', 1, 0.28), sharey=True)
+
+    N = 8
+    color = 'black'
+
+    x = np.linspace(0, 3, N)
+    y = x**3 - 4 * x**2
+    axs[0].plot(y, marker='o', color=color)
+
+    y_m = np.append(y, y[::-1])
+    axs[1].scatter(np.arange(2 * N)[::2], y_m[::2], marker='<', color=color)
+    axs[1].scatter(np.arange(2 * N)[1::2], y_m[1::2], marker='>', color=color)
+    axs[1].plot(np.arange(2 * N), y_m, color=color)
+
+    v = y_m[::2]
+    axs[2].plot(np.arange(N), v, color=color, marker='x')
+
+    axs[0].set_title('original')
+    axs[1].set_title('mirrored')
+    axs[2].set_title('periodically reordered')
+
+    for ax in axs:
+        # ax.set_xlabel(r'$n$')
+        ax.set_yticks([])
+    fig.savefig('plots/DCT_via_FFT.pdf', bbox_inches='tight', dpi=300)
+
+
 if __name__ == '__main__':
     setup_mpl()
-    plot_ultraspherical()
+    plot_DCT()
+    # plot_ultraspherical()
+    plt.show()

pySDC/projects/GPU/configs/RBC_configs.py

@@ -385,7 +385,7 @@ def get_controller_params(self, *args, **kwargs):
 class RayleighBenard_large(RayleighBenardRegular):
     # res_per_plume = 256
     # vertical_res = 1024
-    Ra = 2e7
+    Ra = 3.2e8
     relaxation_steps = 5
 
     def get_description(self, *args, **kwargs):

pySDC/projects/Resilience/RBC.py

@@ -14,7 +14,7 @@
 
 import numpy as np
 
-PROBLEM_PARAMS = {'Rayleigh': 2e4, 'nx': 256, 'nz': 128, 'max_cached_factorizations': 30}
+PROBLEM_PARAMS = {'Rayleigh': 3.2e5, 'nx': 256, 'nz': 128, 'max_cached_factorizations': 30}
 
 
 def u_exact(self, t, u_init=None, t_init=None, recompute=False, _t0=None):