From 5f565c457dbcb93cfcb33e89c3958502c8ca74d2 Mon Sep 17 00:00:00 2001
From: austin <austin@austinhoover.net>
Date: Wed, 11 Dec 2024 11:49:17 -0500
Subject: [PATCH] swap diffuse to use generic solver func

---
 src/main/c/src/fluid/sim/grid2/density.c  | 29 +--------
 src/main/c/src/fluid/sim/grid2/velocity.c | 75 +----------------------
 2 files changed, 5 insertions(+), 99 deletions(-)

diff --git a/src/main/c/src/fluid/sim/grid2/density.c b/src/main/c/src/fluid/sim/grid2/density.c
index 699cb793..17ef8f63 100644
--- a/src/main/c/src/fluid/sim/grid2/density.c
+++ b/src/main/c/src/fluid/sim/grid2/density.c
@@ -9,6 +9,7 @@
 #include "fluid/sim/grid2/solver_consts.h"
 #include "fluid/sim/grid2/utilities.h"
 #include "math/ode/multigrid.h"
+#include "math/ode/gauss_seidel.h"
 
 
 /**
@@ -49,35 +50,9 @@ LIBRARY_API void fluid_grid2_solveDiffuseDensity(
     float * x = GET_ARR_RAW(d,CENTER_LOC);
     float * x0 = GET_ARR_RAW(d0,CENTER_LOC);
     
-    __m256 aScalar = _mm256_set1_ps(a);
-    __m256 cScalar = _mm256_set1_ps(c);
-
     for(int l = 0; l < FLUID_GRID2_LINEARSOLVERTIMES; l++){
         //iterate
-        for(k=1; k<DIM-1; k++){
-            for(j=1; j<DIM-1; j++){
-                int n = 0;
-                //solve as much as possible vectorized
-                for(i = 1; i < DIM-1; i=i+8){
-                    __m256 vector = _mm256_loadu_ps(&x[IX(i-1,j,k)]);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&x[IX(i+1,j,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&x[IX(i,j-1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&x[IX(i,j+1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&x[IX(i,j,k-1)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&x[IX(i,j,k+1)]));
-                    vector = _mm256_mul_ps(vector,aScalar);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&x0[IX(i,j,k)]));
-                    vector = _mm256_div_ps(vector,cScalar);
-                    _mm256_storeu_ps(&x[IX(i,j,k)],vector);
-                }
-                //If there is any leftover, perform manual solving
-                if(i>DIM-1){
-                    for(i=i-8; i < DIM-1; i++){
-                        x[IX(i,j,k)] = (x0[IX(i,j,k)] + a*(x[IX(i-1,j,k)]+x[IX(i+1,j,k)]+x[IX(i,j-1,k)]+x[IX(i,j+1,k)]+x[IX(i,j,k-1)]+x[IX(i,j,k+1)]))/c;
-                    }
-                }
-            }
-        }
+        solver_gauss_seidel_iterate_parallel(x,x0,a,c,DIM);
 
         //set bounds
         fluid_grid2_set_bounds(0,x);
diff --git a/src/main/c/src/fluid/sim/grid2/velocity.c b/src/main/c/src/fluid/sim/grid2/velocity.c
index 58c6807b..2f977e91 100644
--- a/src/main/c/src/fluid/sim/grid2/velocity.c
+++ b/src/main/c/src/fluid/sim/grid2/velocity.c
@@ -58,84 +58,15 @@ LIBRARY_API void fluid_grid2_solveVectorDiffuse (
     float * v0 = GET_ARR_RAW(jrv0,CENTER_LOC);
     float * w0 = GET_ARR_RAW(jrw0,CENTER_LOC);
     
-    __m256 aScalar = _mm256_set1_ps(a);
-    __m256 cScalar = _mm256_set1_ps(c);
-
     for(int l = 0; l < FLUID_GRID2_LINEARSOLVERTIMES; l++){
         //transform u direction
-        for(k=1; k<DIM-1; k++){
-            for(j=1; j<DIM-1; j++){
-                //solve as much as possible vectorized
-                for(i = 1; i < DIM-1; i=i+8){
-                    __m256 vector = _mm256_loadu_ps(&u[IX(i-1,j,k)]);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&u[IX(i+1,j,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&u[IX(i,j-1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&u[IX(i,j+1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&u[IX(i,j,k-1)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&u[IX(i,j,k+1)]));
-                    vector = _mm256_mul_ps(vector,aScalar);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&u0[IX(i,j,k)]));
-                    vector = _mm256_div_ps(vector,cScalar);
-                    _mm256_storeu_ps(&u[IX(i,j,k)],vector);
-                }
-                //If there is any leftover, perform manual solving
-                if(i>DIM-1){
-                    for(i=i-8; i < DIM-1; i++){
-                        u[IX(i,j,k)] = (u0[IX(i,j,k)] + a*(u[IX(i-1,j,k)]+u[IX(i+1,j,k)]+u[IX(i,j-1,k)]+u[IX(i,j+1,k)]+u[IX(i,j,k-1)]+u[IX(i,j,k+1)]))/c;
-                    }
-                }
-            }
-        }
+        solver_gauss_seidel_iterate_parallel(u,u0,a,c,DIM);
 
         //transform v direction
-        for(k=1; k<DIM-1; k++){
-            for(j=1; j<DIM-1; j++){
-                //solve as much as possible vectorized
-                for(i = 1; i < DIM-1; i=i+8){
-                    __m256 vector = _mm256_loadu_ps(&v[IX(i-1,j,k)]);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&v[IX(i+1,j,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&v[IX(i,j-1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&v[IX(i,j+1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&v[IX(i,j,k-1)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&v[IX(i,j,k+1)]));
-                    vector = _mm256_mul_ps(vector,aScalar);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&v0[IX(i,j,k)]));
-                    vector = _mm256_div_ps(vector,cScalar);
-                    _mm256_storeu_ps(&v[IX(i,j,k)],vector);
-                }
-                //If there is any leftover, perform manual solving
-                if(i>DIM-1){
-                    for(i=i-8; i < DIM-1; i++){
-                        v[IX(i,j,k)] = (v0[IX(i,j,k)] + a*(v[IX(i-1,j,k)]+v[IX(i+1,j,k)]+v[IX(i,j-1,k)]+v[IX(i,j+1,k)]+v[IX(i,j,k-1)]+v[IX(i,j,k+1)]))/c;
-                    }
-                }
-            }
-        }
+        solver_gauss_seidel_iterate_parallel(v,v0,a,c,DIM);
 
         //transform w direction
-        for(k=1; k<DIM-1; k++){
-            for(j=1; j<DIM-1; j++){
-                //solve as much as possible vectorized
-                for(i = 1; i < DIM-1; i=i+8){
-                    __m256 vector = _mm256_loadu_ps(&w[IX(i-1,j,k)]);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&w[IX(i+1,j,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&w[IX(i,j-1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&w[IX(i,j+1,k)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&w[IX(i,j,k-1)]));
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&w[IX(i,j,k+1)]));
-                    vector = _mm256_mul_ps(vector,aScalar);
-                    vector = _mm256_add_ps(vector,_mm256_loadu_ps(&w0[IX(i,j,k)]));
-                    vector = _mm256_div_ps(vector,cScalar);
-                    _mm256_storeu_ps(&w[IX(i,j,k)],vector);
-                }
-                //If there is any leftover, perform manual solving
-                if(i>DIM-1){
-                    for(i=i-8; i < DIM-1; i++){
-                        w[IX(i,j,k)] = (w0[IX(i,j,k)] + a*(w[IX(i-1,j,k)]+w[IX(i+1,j,k)]+w[IX(i,j-1,k)]+w[IX(i,j+1,k)]+w[IX(i,j,k-1)]+w[IX(i,j,k+1)]))/c;
-                    }
-                }
-            }
-        }
+        solver_gauss_seidel_iterate_parallel(w,w0,a,c,DIM);
 
         //set bounds
         fluid_grid2_set_bounds(FLUID_GRID2_BOUND_DIR_U,u);