Fixed incorrect condition to grow time step in CPU branch

Author: Johannes Lang

RG-Evo

@@ -16,6 +16,7 @@ struct SimulationDataSnapshot {
     std::vector<double> debug_step_memory;
     double energy;
     double t_current;
+    int method;  // 1=RK54, 2=SSPRK104, 3+=SERK2(2*(method-2))
     int current_len, current_loop;
     SimulationConfig config_snapshot;
 

RG-Evo-shared

@@ -18,8 +18,8 @@ module purge
 
 # Load a current GPU toolchain. The cluster module tree moved away from
 # accel/nvhpc/24.9, so try the CUDA modules that are available now.
-if ! module load cuda/12.5 2>/dev/null; then
-  if ! module load cuda/12.2 2>/dev/null; then
+if ! module load accel/cuda/12.5 2>/dev/null; then
+  if ! module load accel/cuda/12.2 2>/dev/null; then
     module load accel/nvhpc/24.9
   fi
 fi

include/io/io_utils.hpp

@@ -9,6 +9,7 @@
 #include "convolution/convolution.cuh"
 #include "core/device_utils.cuh"
 #include "EOMs/time_steps.hpp"
+#include "EOMs/rk_data.hpp"
 #include "version/version_info.hpp"
 #include <fstream>
 #include <iostream>
@@ -485,8 +486,7 @@ void saveParametersToFileAsync(const std::string& dirPath, double delta, double
     params << "current_loop = " << snapshot.current_loop << std::endl;
     params << "current_delta = " << delta << std::endl;
     params << "current_delta_t = " << delta_t << std::endl;
-    // Note: We can't access rk->host->init in async context, so we'll use a default
-    params << "current_method = N/A (async)" << std::endl;
+    params << "current_method = " << (snapshot.method == 1 ? "RK54" : snapshot.method == 2 ? "SSPRK104" : "SERK2(" + std::to_string(2 * (snapshot.method - 2)) + ")") << std::endl;
     params << "current_t1grid_size = " << snapshot.t1grid.size() << std::endl;
     params << "current_QK0 = " << snapshot.QKv[(snapshot.t1grid.size() - 1) * snapshot.current_len] << std::endl;
     params << "current_QR0 = " << snapshot.QRv[(snapshot.t1grid.size() - 1) * snapshot.current_len] << std::endl;

run.sh

@@ -11,6 +11,8 @@
 #include "convolution/convolution.cuh"
 #include "core/device_utils.cuh"
 #include "EOMs/time_steps.hpp"
+#include "EOMs/rk_data.hpp"
+#include "EOMs/device_rk_data.hpp"
 #include "version/version_info.hpp"
 #include <fstream>
 #include <iostream>
@@ -148,7 +150,7 @@ SimulationDataSnapshot createDataSnapshot()
         copyDeviceToSnapshot(snapshot.QKB1int, sim->device->QKB1int);
         copyDeviceToSnapshot(snapshot.QRB1int, sim->device->QRB1int);
         copyDeviceToSnapshot(snapshot.theta, sim->device->theta);
-        snapshot.t_current = snapshot.t1grid.back();
+        cudaMemcpy(&snapshot.method, &rk->device->init, sizeof(int), cudaMemcpyDeviceToHost); // Store the RK method used on GPU
     } else {
         // For CPU runs, copy from host vectors as before
         snapshot.QKv = sim->host->QKv;
@@ -164,6 +166,7 @@ SimulationDataSnapshot createDataSnapshot()
         snapshot.QRB1int = sim->host->QRB1int;
         snapshot.theta = sim->host->theta;
         snapshot.t_current = sim->host->t1grid.back();
+        snapshot.method = rk->host->init;  // Store the RK method used on CPU
     }
 
     // Calculate energy

src/io/io_output.cpp

@@ -396,7 +396,7 @@ void saveParametersToFileAsync(const std::string& dirPath, double delta, double
     params << "current_loop = " << snapshot.current_loop << std::endl;
     params << "current_delta = " << delta << std::endl;
     params << "current_delta_t = " << delta_t << std::endl;
-    params << "current_method = N/A (async)" << std::endl;
+    params << "current_method = " << (snapshot.method == 1 ? "RK54" : snapshot.method == 2 ? "SSPRK104" : "SERK2(" + std::to_string(2 * (snapshot.method - 2)) + ")") << std::endl;
     params << "current_t1grid_size = " << snapshot.t1grid.size() << std::endl;
     params << "current_QK0 = " << snapshot.QKv[(snapshot.t1grid.size() - 1) * snapshot.current_len] << std::endl;
     params << "current_QR0 = " << snapshot.QRv[(snapshot.t1grid.size() - 1) * snapshot.current_len] << std::endl;

src/io/io_output.cu

@@ -31,6 +31,7 @@ SimulationDataSnapshot createDataSnapshot() {
     snapshot.QRB1int = sim->host->QRB1int;
     snapshot.theta = sim->host->theta;
     snapshot.t_current = sim->host->t1grid.empty() ? 0.0 : sim->host->t1grid.back();
+    snapshot.method = rk->host->init;  // Store the RK method used on CPU
 
     std::vector<double> temp(config.len, 0.0);
     std::vector<double> lastQKv = getLastLenEntries(sim->host->QKv, config.len);

src/io/io_save_params.cpp

@@ -137,7 +137,7 @@ int runSimulation() {
 #endif
         config.delta_old = config.delta;
         config.delta = update(pool);
-        
+
         if (config.tail_fit_enabled) {
 #if DMFE_WITH_CUDA
     // Optional tail fit/blend near theta->1 (toggle with --tail-fit)
@@ -250,11 +250,14 @@ int runSimulation() {
             }
         }
 
-        // primitive time-step adaptation
+        // primitive time-step adaptation        
 #if DMFE_WITH_CUDA
-        if (config.delta < config.delta_max && config.loop > 5 &&
+        if ((config.gpu && (config.delta < config.delta_max && config.loop > 5 &&
+            (config.delta < 1.1 * config.delta_old || config.delta_old < config.delta_max/1000) &&
+            config.rmax[rk->device->init-1] / config.specRad > config.delta_t && (config.gpu ? sim->device->delta_t_ratio.back() : sim->host->delta_t_ratio.back()) == 1.0)) ||
+            (!config.gpu && (config.delta < config.delta_max && config.loop > 5 &&
             (config.delta < 1.1 * config.delta_old || config.delta_old < config.delta_max/1000) &&
-            config.rmax[rk->device->init-1] / config.specRad > config.delta_t && (config.gpu ? sim->device->delta_t_ratio.back() : sim->host->delta_t_ratio.back()) == 1.0)
+            config.rmax[rk->host->init-1] / config.specRad > config.delta_t && sim->host->delta_t_ratio.back() == 1.0)))
 #else
         if (config.delta < config.delta_max && config.loop > 5 &&
             (config.delta < 1.1 * config.delta_old || config.delta_old < config.delta_max/1000) &&
@@ -316,7 +319,7 @@ int runSimulation() {
 #endif
 
         // Status: continuous multi-line TUI for TTY when debug is off; otherwise line-per-iteration
-        std::string method = (rk->host->init == 1 ? "RK54" : rk->host->init == 2 ? "SSPRK104" : "SERK2(" + std::to_string(2 * (rk->host->init - 2)) + ")");
+        std::string method = config.gpu ? (rk->device->init == 1 ? "RK54" : rk->device->init == 2 ? "SSPRK104" : "SERK2(" + std::to_string(2 * (rk->device->init - 2)) + ")") : (rk->host->init == 1 ? "RK54" : rk->host->init == 2 ? "SSPRK104" : "SERK2(" + std::to_string(2 * (rk->host->init - 2)) + ")");
         if (continuous_status) {
             ui.update_status(t, config.tmax, config.loop, config.delta_t, method);
         } else if (config.debug) {