smyalygames/FiniteVolumeGPU
1
0
Fork 0
mirror of https://github.com/smyalygames/FiniteVolumeGPU.git synced 2025-11-29 17:28:03 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Implemented variable timestep

Browse Source
This commit is contained in:
André R. Brodtkorb
2018-11-15 16:48:47 +01:00
parent 7592ad5b9f
commit ddac53271c
15 changed files with 224 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
GPUSimulators/cuda/SWE2D_FORCE.cu
View File

@@ -97,7 +97,10 @@ __global__ void FORCEKernel(
//Output h^{n+1}
float* h1_ptr_, int h1_pitch_,
float* hu1_ptr_, int hu1_pitch_,
float* hv1_ptr_, int hv1_pitch_) {
float* hv1_ptr_, int hv1_pitch_,
//Output CFL
float* cfl_) {
const unsigned int w = BLOCK_WIDTH;
const unsigned int h = BLOCK_HEIGHT;
@@ -130,6 +133,11 @@ __global__ void FORCEKernel(
writeBlock<w, h, gc_x, gc_y>( h1_ptr_, h1_pitch_, Q[0], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hu1_ptr_, hu1_pitch_, Q[1], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hv1_ptr_, hv1_pitch_, Q[2], nx_, ny_, 0, 1);
//Compute the CFL for this block
if (cfl_ != NULL) {
writeCfl<w, h, gc_x, gc_y, vars>(Q, F[0], nx_, ny_, dx_, dy_, g_, cfl_);
}
}
} // extern "C"

10
GPUSimulators/cuda/SWE2D_HLL.cu
View File

@@ -113,7 +113,10 @@ __global__ void HLLKernel(
//Output h^{n+1}
float* h1_ptr_, int h1_pitch_,
float* hu1_ptr_, int hu1_pitch_,
float* hv1_ptr_, int hv1_pitch_) {
float* hv1_ptr_, int hv1_pitch_,
//Output CFL
float* cfl_) {
const unsigned int w = BLOCK_WIDTH;
const unsigned int h = BLOCK_HEIGHT;
@@ -148,6 +151,11 @@ __global__ void HLLKernel(
writeBlock<w, h, gc_x, gc_y>( h1_ptr_, h1_pitch_, Q[0], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hu1_ptr_, hu1_pitch_, Q[1], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hv1_ptr_, hv1_pitch_, Q[2], nx_, ny_, 0, 1);
//Compute the CFL for this block
if (cfl_ != NULL) {
writeCfl<w, h, gc_x, gc_y, vars>(Q, F[0], nx_, ny_, dx_, dy_, g_, cfl_);
}
}
} // extern "C"

10
GPUSimulators/cuda/SWE2D_HLL2.cu
View File

@@ -141,7 +141,10 @@ __global__ void HLL2Kernel(
//Output h^{n+1}
float* h1_ptr_, int h1_pitch_,
float* hu1_ptr_, int hu1_pitch_,
float* hv1_ptr_, int hv1_pitch_) {
float* hv1_ptr_, int hv1_pitch_,
//Output CFL
float* cfl_) {
const unsigned int w = BLOCK_WIDTH;
const unsigned int h = BLOCK_HEIGHT;
@@ -203,6 +206,11 @@ __global__ void HLL2Kernel(
writeBlock<w, h, gc_x, gc_y>( h1_ptr_, h1_pitch_, Q[0], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hu1_ptr_, hu1_pitch_, Q[1], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hv1_ptr_, hv1_pitch_, Q[2], nx_, ny_, 0, 1);
//Compute the CFL for this block
if (cfl_ != NULL) {
writeCfl<w, h, gc_x, gc_y, vars>(Q, F[0], nx_, ny_, dx_, dy_, g_, cfl_);
}
}
} // extern "C"

11
GPUSimulators/cuda/SWE2D_KP07.cu
View File

@@ -151,8 +151,10 @@ __global__ void KP07Kernel(
//Output h^{n+1}
float* h1_ptr_, int h1_pitch_,
float* hu1_ptr_, int hu1_pitch_,
float* hv1_ptr_, int hv1_pitch_) {
float* hv1_ptr_, int hv1_pitch_,
//Output CFL
float* cfl_) {
const unsigned int w = BLOCK_WIDTH;
const unsigned int h = BLOCK_HEIGHT;
const unsigned int gc_x = 2;
@@ -222,5 +224,10 @@ __global__ void KP07Kernel(
hv_row[ti] = hv1;
}
}
//Compute the CFL for this block
if (cfl_ != NULL) {
writeCfl<w, h, gc_x, gc_y, vars>(Q, Q[0], nx_, ny_, dx_, dy_, g_, cfl_);
}
}
} //extern "C"

10
GPUSimulators/cuda/SWE2D_KP07_dimsplit.cu
View File

@@ -138,7 +138,10 @@ __global__ void KP07DimsplitKernel(
//Output h^{n+1}
float* h1_ptr_, int h1_pitch_,
float* hu1_ptr_, int hu1_pitch_,
float* hv1_ptr_, int hv1_pitch_) {
float* hv1_ptr_, int hv1_pitch_,
//Output CFL
float* cfl_) {
const unsigned int w = BLOCK_WIDTH;
const unsigned int h = BLOCK_HEIGHT;
const unsigned int gc_x = 2;
@@ -194,6 +197,11 @@ __global__ void KP07DimsplitKernel(
writeBlock<w, h, gc_x, gc_y>( h1_ptr_, h1_pitch_, Q[0], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hu1_ptr_, hu1_pitch_, Q[1], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hv1_ptr_, hv1_pitch_, Q[2], nx_, ny_, 0, 1);
//Compute the CFL for this block
if (cfl_ != NULL) {
writeCfl<w, h, gc_x, gc_y, vars>(Q, F[0], nx_, ny_, dx_, dy_, g_, cfl_);
}
}

10
GPUSimulators/cuda/SWE2D_LxF.cu
View File

@@ -114,7 +114,10 @@ void LxFKernel(
//Output h^{n+1}
float* h1_ptr_, int h1_pitch_,
float* hu1_ptr_, int hu1_pitch_,
float* hv1_ptr_, int hv1_pitch_) {
float* hv1_ptr_, int hv1_pitch_,
//Output CFL
float* cfl_) {
const unsigned int w = BLOCK_WIDTH;
const unsigned int h = BLOCK_HEIGHT;
@@ -154,6 +157,11 @@ void LxFKernel(
writeBlock<w, h, gc_x, gc_y>( h1_ptr_, h1_pitch_, Q[0], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hu1_ptr_, hu1_pitch_, Q[1], nx_, ny_, 0, 1);
writeBlock<w, h, gc_x, gc_y>(hv1_ptr_, hv1_pitch_, Q[2], nx_, ny_, 0, 1);
//Compute the CFL for this block
if (cfl_ != NULL) {
writeCfl<w, h, gc_x, gc_y, vars>(Q, Q[0], nx_, ny_, dx_, dy_, g_, cfl_);
}
}
} // extern "C"

63
GPUSimulators/cuda/SWECommon.h
View File

@@ -468,3 +468,66 @@ __device__ float3 FORCE_1D_flux(const float3 Q_l, const float3 Q_r, const float
const float3 F_lw2 = LxW2_1D_flux(Q_l, Q_r, g_, dx_, dt_);
return 0.5f*(F_lf + F_lw2);
}
template<int w, int h, int gc_x, int gc_y, int vars>
__device__ void writeCfl(float Q[vars][h+2*gc_y][w+2*gc_x],
float shmem[h+2*gc_y][w+2*gc_x],
const int nx_, const int ny_,
const float dx_, const float dy_, const float g_,
float* output_) {
//Index of thread within block
const int tx = threadIdx.x + gc_x;
const int ty = threadIdx.y + gc_y;
//Index of cell within domain
const int ti = blockDim.x*blockIdx.x + tx;
const int tj = blockDim.y*blockIdx.y + ty;
//Only internal cells
if (ti < nx_+gc_x && tj < ny_+gc_y) {
const float h = Q[0][ty][tx];
const float u = Q[1][ty][tx] / h;
const float v = Q[2][ty][tx] / h;
const float max_u = dx_ / (fabsf(u) + sqrtf(g_*h));
const float max_v = dy_ / (fabsf(v) + sqrtf(g_*h));
shmem[ty][tx] = fminf(max_u, max_v);
}
__syncthreads();
//One row of threads loop over all rows
if (ti < nx_+gc_x && tj < ny_+gc_y) {
if (ty == gc_y) {
float min_val = shmem[ty][tx];
const int max_y = min(h, ny_+gc_y - tj);
for (int j=gc_y; j<max_y+gc_y; j++) {
min_val = fminf(min_val, shmem[j][tx]);
}
shmem[ty][tx] = min_val;
}
}
__syncthreads();
//One thread loops over first row to find global max
if (tx == gc_x && ty == gc_y) {
float min_val = shmem[ty][tx];
const int max_x = min(w, nx_+gc_x - ti);
for (int i=gc_x; i<max_x+gc_x; ++i) {
min_val = fminf(min_val, shmem[ty][i]);
}
const int idx = gridDim.x*blockIdx.y + blockIdx.x;
output_[idx] = min_val;
}
}