mirror of
https://github.com/smyalygames/FiniteVolumeGPU.git
synced 2026-01-14 15:48:43 +01:00
Refactoring
This commit is contained in:
André R. Brodtkorb
@@ -48,6 +48,10 @@ inline __device__ __host__ float clamp(const float f, const float a, const float
|
||||
return fmaxf(a, fminf(f, b));
|
||||
}
|
||||
|
||||
inline __device__ __host__ int clamp(const int f, const int a, const int b) {
|
||||
return (f < b) ? ( (f > a) ? f : a) : b;
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
@@ -60,37 +64,27 @@ __device__ float desingularize(float x_, float eps_) {
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
/**
|
||||
* Reads a block of data with ghost cells
|
||||
*/
|
||||
template<int vars, int sm_width, int sm_height, int block_width, int block_height>
|
||||
inline __device__ void readBlock(float* ptr_[vars], int pitch_[vars],
|
||||
float shmem[vars][sm_height][sm_width],
|
||||
template<int block_width, int block_height, int ghost_cells>
|
||||
inline __device__ void readBlock(float* ptr_, int pitch_,
|
||||
float shmem[block_height+2*ghost_cells][block_width+2*ghost_cells],
|
||||
const int max_x_, const int max_y_) {
|
||||
|
||||
//Index of block within domain
|
||||
const int bx = blockDim.x * blockIdx.x;
|
||||
const int by = blockDim.y * blockIdx.y;
|
||||
|
||||
float* rows[3];
|
||||
|
||||
|
||||
//Read into shared memory
|
||||
for (int j=threadIdx.y; j<sm_height; j+=block_height) {
|
||||
const int l = clamp(by + j, 0, max_y_-1); // Clamp out of bounds
|
||||
//Loop over all variables
|
||||
for (int j=threadIdx.y; j<block_height+2*ghost_cells; j+=block_height) {
|
||||
const int l = min(by + j, max_y_-1);
|
||||
float* row = (float*) ((char*) ptr_ + pitch_*l);
|
||||
|
||||
//Compute the pointer to current row in the arrays
|
||||
for (int m=0; m<vars; ++m) {
|
||||
rows[m] = (float*) ((char*) ptr_[m] + pitch_[m]*l);
|
||||
}
|
||||
|
||||
for (int i=threadIdx.x; i<sm_width; i+=block_width) {
|
||||
const int k = clamp(bx + i, 0, max_x_-1); // Clamp out of bounds
|
||||
|
||||
for (int m=0; m<vars; ++m) {
|
||||
shmem[m][j][i] = rows[m][k];
|
||||
}
|
||||
for (int i=threadIdx.x; i<block_width+2*ghost_cells; i+=block_width) {
|
||||
const int k = min(bx + i, max_x_-1);
|
||||
|
||||
shmem[j][i] = row[k];
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -98,24 +92,23 @@ inline __device__ void readBlock(float* ptr_[vars], int pitch_[vars],
|
||||
|
||||
|
||||
|
||||
|
||||
/**
|
||||
* Writes a block of data to global memory for the shallow water equations.
|
||||
*/
|
||||
template<int sm_width, int sm_height, int offset_x=0, int offset_y=0>
|
||||
template<int block_width, int block_height, int ghost_cells>
|
||||
inline __device__ void writeBlock(float* ptr_, int pitch_,
|
||||
float shmem[sm_height][sm_width],
|
||||
float shmem[block_height+2*ghost_cells][block_width+2*ghost_cells],
|
||||
const int width, const int height) {
|
||||
|
||||
//Index of cell within domain
|
||||
const int ti = blockDim.x*blockIdx.x + threadIdx.x + offset_x;
|
||||
const int tj = blockDim.y*blockIdx.y + threadIdx.y + offset_y;
|
||||
const int ti = blockDim.x*blockIdx.x + threadIdx.x + ghost_cells;
|
||||
const int tj = blockDim.y*blockIdx.y + threadIdx.y + ghost_cells;
|
||||
|
||||
//Only write internal cells
|
||||
if (ti < width+offset_x && tj < height+offset_y) {
|
||||
if (ti < width+ghost_cells && tj < height+ghost_cells) {
|
||||
//Index of thread within block
|
||||
const int tx = threadIdx.x + offset_x;
|
||||
const int ty = threadIdx.y + offset_y;
|
||||
const int tx = threadIdx.x + ghost_cells;
|
||||
const int ty = threadIdx.y + ghost_cells;
|
||||
|
||||
float* const row = (float*) ((char*) ptr_ + pitch_*tj);
|
||||
row[ti] = shmem[ty][tx];
|
||||
@@ -129,60 +122,93 @@ inline __device__ void writeBlock(float* ptr_, int pitch_,
|
||||
|
||||
|
||||
|
||||
/**
|
||||
* Writes a block of data to global memory for the shallow water equations.
|
||||
*/
|
||||
__device__ void writeBlock2(float* h_ptr_, int h_pitch_,
|
||||
float* hu_ptr_, int hu_pitch_,
|
||||
float* hv_ptr_, int hv_pitch_,
|
||||
float Q[3][BLOCK_HEIGHT+4][BLOCK_WIDTH+4],
|
||||
const int nx_, const int ny_) {
|
||||
writeBlock<BLOCK_WIDTH+4, BLOCK_HEIGHT+4, 2, 2>( h_ptr_, h_pitch_, Q[0], nx_, ny_);
|
||||
writeBlock<BLOCK_WIDTH+4, BLOCK_HEIGHT+4, 2, 2>(hu_ptr_, hu_pitch_, Q[1], nx_, ny_);
|
||||
writeBlock<BLOCK_WIDTH+4, BLOCK_HEIGHT+4, 2, 2>(hv_ptr_, hv_pitch_, Q[2], nx_, ny_);
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
/**
|
||||
* No flow boundary conditions for the shallow water equations
|
||||
* with one ghost cell in each direction
|
||||
*/
|
||||
__device__ void noFlowBoundary1(float Q[3][BLOCK_HEIGHT+2][BLOCK_WIDTH+2], const int nx_, const int ny_) {
|
||||
//Global index
|
||||
const int ti = blockDim.x*blockIdx.x + threadIdx.x + 1; //Skip global ghost cells, i.e., +1
|
||||
const int tj = blockDim.y*blockIdx.y + threadIdx.y + 1;
|
||||
template<int block_width, int block_height, int ghost_cells, int scale_east_west=1, int scale_north_south=1>
|
||||
__device__ void noFlowBoundary(float Q[block_height+2*ghost_cells][block_width+2*ghost_cells], const int nx_, const int ny_) {
|
||||
const int ti = blockDim.x*blockIdx.x + threadIdx.x + ghost_cells;
|
||||
const int tj = blockDim.y*blockIdx.y + threadIdx.y + ghost_cells;
|
||||
|
||||
//Block-local indices
|
||||
const int tx = threadIdx.x;
|
||||
const int ty = threadIdx.y;
|
||||
const int i = threadIdx.x + ghost_cells;
|
||||
const int j = threadIdx.y + ghost_cells;
|
||||
|
||||
const int i = tx + 1; //Skip local ghost cells, i.e., +1
|
||||
const int j = ty + 1;
|
||||
|
||||
//Fix boundary conditions
|
||||
if (ti == 1) {
|
||||
Q[0][j][i-1] = Q[0][j][i];
|
||||
Q[1][j][i-1] = -Q[1][j][i];
|
||||
Q[2][j][i-1] = Q[2][j][i];
|
||||
// West boundary
|
||||
if (ti == ghost_cells) {
|
||||
Q[j][i-1] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (ti == nx_) {
|
||||
Q[0][j][i+1] = Q[0][j][i];
|
||||
Q[1][j][i+1] = -Q[1][j][i];
|
||||
Q[2][j][i+1] = Q[2][j][i];
|
||||
if (ghost_cells >= 2 && ti == ghost_cells + 1) {
|
||||
Q[j][i-3] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (tj == 1) {
|
||||
Q[0][j-1][i] = Q[0][j][i];
|
||||
Q[1][j-1][i] = Q[1][j][i];
|
||||
Q[2][j-1][i] = -Q[2][j][i];
|
||||
if (ghost_cells >= 3 && ti == ghost_cells + 2) {
|
||||
Q[j][i-5] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (tj == ny_) {
|
||||
Q[0][j+1][i] = Q[0][j][i];
|
||||
Q[1][j+1][i] = Q[1][j][i];
|
||||
Q[2][j+1][i] = -Q[2][j][i];
|
||||
if (ghost_cells >= 4 && ti == ghost_cells + 3) {
|
||||
Q[j][i-7] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 5 && ti == ghost_cells + 4) {
|
||||
Q[j][i-9] = scale_east_west*Q[j][i];
|
||||
}
|
||||
|
||||
|
||||
|
||||
// East boundary
|
||||
if (ti == nx_ + ghost_cells - 1) {
|
||||
Q[j][i+1] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 2 && ti == nx_ + ghost_cells - 2) {
|
||||
Q[j][i+3] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && ti == nx_ + ghost_cells - 3) {
|
||||
Q[j][i+5] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && ti == nx_ + ghost_cells - 4) {
|
||||
Q[j][i+7] = scale_east_west*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && ti == nx_ + ghost_cells - 5) {
|
||||
Q[j][i+9] = scale_east_west*Q[j][i];
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
// South boundary
|
||||
if (tj == ghost_cells) {
|
||||
Q[j-1][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 2 && tj == ghost_cells + 1) {
|
||||
Q[j-3][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && tj == ghost_cells + 2) {
|
||||
Q[j-5][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 4 && tj == ghost_cells + 3) {
|
||||
Q[j-7][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 5 && tj == ghost_cells + 4) {
|
||||
Q[j-9][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
|
||||
|
||||
|
||||
// North boundary
|
||||
if (tj == ny_ + ghost_cells - 1) {
|
||||
Q[j+1][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 2 && tj == ny_ + ghost_cells - 2) {
|
||||
Q[j+3][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && tj == ny_ + ghost_cells - 3) {
|
||||
Q[j+5][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && tj == ny_ + ghost_cells - 4) {
|
||||
Q[j+7][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
if (ghost_cells >= 3 && tj == ny_ + ghost_cells - 5) {
|
||||
Q[j+9][i] = scale_north_south*Q[j][i];
|
||||
}
|
||||
}
|
||||
|
||||
@@ -191,59 +217,9 @@ __device__ void noFlowBoundary1(float Q[3][BLOCK_HEIGHT+2][BLOCK_WIDTH+2], const
|
||||
|
||||
|
||||
|
||||
/**
|
||||
* No flow boundary conditions for the shallow water equations
|
||||
* with two ghost cells in each direction
|
||||
*/
|
||||
__device__ void noFlowBoundary2(float Q[3][BLOCK_HEIGHT+4][BLOCK_WIDTH+4], const int nx_, const int ny_) {
|
||||
//Global index
|
||||
const int ti = blockDim.x*blockIdx.x + threadIdx.x + 2; //Skip global ghost cells, i.e., +2
|
||||
const int tj = blockDim.y*blockIdx.y + threadIdx.y + 2;
|
||||
|
||||
//Block-local indices
|
||||
const int tx = threadIdx.x;
|
||||
const int ty = threadIdx.y;
|
||||
|
||||
const int i = tx + 2; //Skip local ghost cells, i.e., +2
|
||||
const int j = ty + 2;
|
||||
|
||||
if (ti == 2) {
|
||||
Q[0][j][i-1] = Q[0][j][i];
|
||||
Q[1][j][i-1] = -Q[1][j][i];
|
||||
Q[2][j][i-1] = Q[2][j][i];
|
||||
|
||||
Q[0][j][i-2] = Q[0][j][i+1];
|
||||
Q[1][j][i-2] = -Q[1][j][i+1];
|
||||
Q[2][j][i-2] = Q[2][j][i+1];
|
||||
}
|
||||
if (ti == nx_+1) {
|
||||
Q[0][j][i+1] = Q[0][j][i];
|
||||
Q[1][j][i+1] = -Q[1][j][i];
|
||||
Q[2][j][i+1] = Q[2][j][i];
|
||||
|
||||
Q[0][j][i+2] = Q[0][j][i-1];
|
||||
Q[1][j][i+2] = -Q[1][j][i-1];
|
||||
Q[2][j][i+2] = Q[2][j][i-1];
|
||||
}
|
||||
if (tj == 2) {
|
||||
Q[0][j-1][i] = Q[0][j][i];
|
||||
Q[1][j-1][i] = Q[1][j][i];
|
||||
Q[2][j-1][i] = -Q[2][j][i];
|
||||
|
||||
Q[0][j-2][i] = Q[0][j+1][i];
|
||||
Q[1][j-2][i] = Q[1][j+1][i];
|
||||
Q[2][j-2][i] = -Q[2][j+1][i];
|
||||
}
|
||||
if (tj == ny_+1) {
|
||||
Q[0][j+1][i] = Q[0][j][i];
|
||||
Q[1][j+1][i] = Q[1][j][i];
|
||||
Q[2][j+1][i] = -Q[2][j][i];
|
||||
|
||||
Q[0][j+2][i] = Q[0][j-1][i];
|
||||
Q[1][j+2][i] = Q[1][j-1][i];
|
||||
Q[2][j+2][i] = -Q[2][j-1][i];
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user