studiorailgun/Renderer
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

small optimizations for performance
Some checks failed
studiorailgun/Renderer/pipeline/head There was a failure building this commit
Details

Browse Source
This commit is contained in:
austin 2024-12-12 00:31:47 -05:00
parent c96113133d
commit 4010b2beb7
2 changed files with 52 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/main/c/src/fluid/sim/grid2/density.c
View File

@ -25,15 +25,33 @@ void fluid_grid2_addDensity(
){ ){
int i; int i;
int size=DIM*DIM*DIM; int size=DIM*DIM*DIM;
__m256 minVec = _mm256_set1_ps(MIN_FLUID_VALUE);
__m256 maxVec = _mm256_set1_ps(MAX_FLUID_VALUE);
__m256 existing;
__m256 delta;
float * x = GET_ARR_RAW(d,CENTER_LOC); float * x = GET_ARR_RAW(d,CENTER_LOC);
float * s = GET_ARR_RAW(d0,CENTER_LOC); float * s = GET_ARR_RAW(d0,CENTER_LOC);
for(i=0; i<size; i++){ for(i=0; i<size; i=i+8){
x[i] += dt*s[i]; existing = _mm256_loadu_ps(&x[i]);
if(x[i] < MIN_FLUID_VALUE){ delta = _mm256_loadu_ps(&s[i]);
x[i] = MIN_FLUID_VALUE; _mm256_storeu_ps(&x[i],
} else if(x[i] > MAX_FLUID_VALUE){ _mm256_max_ps(
x[i] = MAX_FLUID_VALUE; _mm256_min_ps(
} _mm256_add_ps(
existing,
delta
),
maxVec
),
minVec
)
);
// x[i] += dt*s[i];
// if(x[i] < MIN_FLUID_VALUE){
// x[i] = MIN_FLUID_VALUE;
// } else if(x[i] > MAX_FLUID_VALUE){
// x[i] = MAX_FLUID_VALUE;
// }
} }
} }

54
src/main/c/src/fluid/sim/grid2/grid2.c
View File

@ -91,17 +91,17 @@ LIBRARY_API void fluid_grid2_simulate(
//solve vector diffusion //solve vector diffusion
fluid_grid2_solveVectorDiffuse(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,currentChunk->w0,timestep); fluid_grid2_solveVectorDiffuse(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,currentChunk->w0,timestep);
} // }
//time tracking // //time tracking
gettimeofday(&tv,NULL); // gettimeofday(&tv,NULL);
end = 1000000.0 * tv.tv_sec + tv.tv_usec; // end = 1000000.0 * tv.tv_sec + tv.tv_usec;
perMilli = (end - start) / 1000.0f; // perMilli = (end - start) / 1000.0f;
environment->state.timeTracking.velocityDiffuse = perMilli; // environment->state.timeTracking.velocityDiffuse = perMilli;
start = end; // start = end;
for(int i = 0; i < numChunks; i++){ // for(int i = 0; i < numChunks; i++){
Chunk * currentChunk = chunks[i]; // Chunk * currentChunk = chunks[i];
//setup projection //setup projection
fluid_grid2_setupProjection(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,timestep); fluid_grid2_setupProjection(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,timestep);
@ -116,31 +116,31 @@ LIBRARY_API void fluid_grid2_simulate(
fluid_grid2_flip_arrays(currentChunk->u,currentChunk->u0); fluid_grid2_flip_arrays(currentChunk->u,currentChunk->u0);
fluid_grid2_flip_arrays(currentChunk->v,currentChunk->v0); fluid_grid2_flip_arrays(currentChunk->v,currentChunk->v0);
fluid_grid2_flip_arrays(currentChunk->w,currentChunk->w0); fluid_grid2_flip_arrays(currentChunk->w,currentChunk->w0);
} // }
//time tracking // //time tracking
gettimeofday(&tv,NULL); // gettimeofday(&tv,NULL);
end = 1000000.0 * tv.tv_sec + tv.tv_usec; // end = 1000000.0 * tv.tv_sec + tv.tv_usec;
perMilli = (end - start) / 1000.0f; // perMilli = (end - start) / 1000.0f;
environment->state.timeTracking.velocityProject = perMilli; // environment->state.timeTracking.velocityProject = perMilli;
start = end; // start = end;
for(int i = 0; i < numChunks; i++){ // for(int i = 0; i < numChunks; i++){
Chunk * currentChunk = chunks[i]; // Chunk * currentChunk = chunks[i];
//advect //advect
fluid_grid2_advectVectors(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,currentChunk->w0,timestep); fluid_grid2_advectVectors(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,currentChunk->w0,timestep);
} // }
//time tracking // //time tracking
gettimeofday(&tv,NULL); // gettimeofday(&tv,NULL);
end = 1000000.0 * tv.tv_sec + tv.tv_usec; // end = 1000000.0 * tv.tv_sec + tv.tv_usec;
perMilli = (end - start) / 1000.0f; // perMilli = (end - start) / 1000.0f;
environment->state.timeTracking.velocityAdvect = perMilli; // environment->state.timeTracking.velocityAdvect = perMilli;
start = end; // start = end;
for(int i = 0; i < numChunks; i++){ // for(int i = 0; i < numChunks; i++){
Chunk * currentChunk = chunks[i]; // Chunk * currentChunk = chunks[i];
//setup projection //setup projection
fluid_grid2_setupProjection(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,timestep); fluid_grid2_setupProjection(currentChunk->u,currentChunk->v,currentChunk->w,currentChunk->u0,currentChunk->v0,timestep);
//Perform main projection solver //Perform main projection solver