[Bf-blender-cvs] [17bff65c6bb] functions: Use multithreading when moving particles forward

[Jacques Lucke]

Commit: 17bff65c6bb70599c7506229e66e397aed560b96
Author: Jacques Lucke
Date:   Tue Jun 18 12:35:22 2019 +0200
Branches: functions
https://developer.blender.org/rB17bff65c6bb70599c7506229e66e397aed560b96

Use multithreading when moving particles forward

Timings in my current test file (nanoseconds per particle per time step):
* Debug Build: 150ns
* Optimized Build: 47ns
* With threading: 9ns

===================================================================

M	source/blender/simulations/bparticles/c_wrapper.cpp
M	source/blender/simulations/bparticles/playground_solver.cpp

===================================================================

diff --git a/source/blender/simulations/bparticles/c_wrapper.cpp b/source/blender/simulations/bparticles/c_wrapper.cpp
index 226e2471449..9d6d24ac625 100644
--- a/source/blender/simulations/bparticles/c_wrapper.cpp
+++ b/source/blender/simulations/bparticles/c_wrapper.cpp
@@ -4,6 +4,7 @@
 #include "playground_solver.hpp"
 #include "emitter.hpp"
 #include "BLI_noise.h"
+#include "BLI_timeit.hpp"
 
 #define WRAPPERS(T1, T2) \
   inline T1 unwrap(T2 value) \
@@ -139,6 +140,7 @@ void BParticles_state_step(BParticlesSolver solver_c,
                            BParticlesState state_c,
                            float elapsed_seconds)
 {
+  SCOPED_TIMER("step");
   Solver *solver = unwrap(solver_c);
   WrappedState *wrapped_state = unwrap(state_c);
 
diff --git a/source/blender/simulations/bparticles/playground_solver.cpp b/source/blender/simulations/bparticles/playground_solver.cpp
index 4d99d179e75..74925c801dc 100644
--- a/source/blender/simulations/bparticles/playground_solver.cpp
+++ b/source/blender/simulations/bparticles/playground_solver.cpp
@@ -1,4 +1,5 @@
 #include "BLI_small_vector.hpp"
+#include "BLI_task.h"
 
 #include "particles_container.hpp"
 #include "playground_solver.hpp"
@@ -82,14 +83,33 @@ class SimpleSolver : public Solver {
     std::cout << "Block amount: " << particles.active_blocks().size() << "\n";
   }
 
+  struct StepBlocksParallelData {
+    SimpleSolver *solver;
+    MyState &state;
+    ArrayRef<ParticlesBlock *> blocks;
+    float elapsed_seconds;
+  };
+
   BLI_NOINLINE void step_blocks(MyState &state,
                                 ArrayRef<ParticlesBlock *> blocks,
                                 float elapsed_seconds)
   {
-    for (ParticlesBlock *block : blocks) {
-      AttributeArrays attributes = block->slice_active();
-      this->step_slice(state, attributes, elapsed_seconds);
-    }
+    ParallelRangeSettings settings;
+    BLI_parallel_range_settings_defaults(&settings);
+
+    StepBlocksParallelData data = {this, state, blocks, elapsed_seconds};
+
+    BLI_task_parallel_range(0, blocks.size(), (void *)&data, step_block_cb, &settings);
+  }
+
+  static void step_block_cb(void *__restrict userdata,
+                            const int index,
+                            const ParallelRangeTLS *__restrict UNUSED(tls))
+  {
+    StepBlocksParallelData *data = (StepBlocksParallelData *)userdata;
+    ParticlesBlock *block = data->blocks[index];
+    AttributeArrays attributes = block->slice_active();
+    data->solver->step_slice(data->state, attributes, data->elapsed_seconds);
   }
 
   BLI_NOINLINE void step_slice(MyState &state, AttributeArrays buffers, float elapsed_seconds)