aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added workable .cu files for stencil & mandelbrot

Browse Source
This commit is contained in:
Evghenii
2013-11-08 10:00:49 +01:00
parent cb7cbec0d5
commit 426afc7377
10 changed files with 645 additions and 160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
examples_cuda/stencil/stencil.ispc
View File

@@ -34,13 +34,14 @@
#ifdef __NVPTX__
#warning "emitting DEVICE code"
#define taskIndex blockIndex0()
#define taskCount blockCount0()
#define programIndex laneIndex()
#define programCount warpSize()
#else
#warning "emitting HOST code"
#endif
static void
static inline void
stencil_step(uniform int x0, uniform int x1,
uniform int y0, uniform int y1,
uniform int z0, uniform int z1,
@@ -50,29 +51,62 @@ stencil_step(uniform int x0, uniform int x1,
const uniform int Nxy = Nx * Ny;
// foreach (z = z0 ... z1, y = y0 ... y1, x = x0 ... x1)
#if 0
#define VER1
#endif
#ifdef VER1
const uniform long x1o = 1;
const uniform long x2o = 2;
const uniform long x3o = 3;
const uniform long y1o = Nx;
const uniform long y2o = Nx*2;
const uniform long y3o = Nx*3;
const uniform long z1o = Nxy;
const uniform long z2o = Nxy*2;
const uniform long z3o = Nxy*3;
#endif
for (uniform int z = z0; z < z1; z++)
for (uniform int y = y0; y < y1; y++)
for (uniform int xb = x0; xb < x1; xb += programCount)
{
const int x = xb + programIndex;
int index = (z * Nxy) + (y * Nx) + x;
const int index_base = (z * Nxy) + (y * Nx);
for (uniform int xb = x0; xb < x1; xb += programCount)
{
const int x = xb + programIndex;
int index = index_base + x;
#ifndef VER1
#define A_cur(x, y, z) Ain[index + (x) + ((y) * Nx) + ((z) * Nxy)]
#define A_next(x, y, z) Aout[index + (x) + ((y) * Nx) + ((z) * Nxy)]
double div = coef[0] * A_cur(0, 0, 0) +
double div = coef[0] * A_cur(0, 0, 0) +
coef[1] * (A_cur(+1, 0, 0) + A_cur(-1, 0, 0) +
A_cur(0, +1, 0) + A_cur(0, -1, 0) +
A_cur(0, 0, +1) + A_cur(0, 0, -1)) +
A_cur(0, +1, 0) + A_cur(0, -1, 0) +
A_cur(0, 0, +1) + A_cur(0, 0, -1)) +
coef[2] * (A_cur(+2, 0, 0) + A_cur(-2, 0, 0) +
A_cur(0, +2, 0) + A_cur(0, -2, 0) +
A_cur(0, 0, +2) + A_cur(0, 0, -2)) +
A_cur(0, +2, 0) + A_cur(0, -2, 0) +
A_cur(0, 0, +2) + A_cur(0, 0, -2)) +
coef[3] * (A_cur(+3, 0, 0) + A_cur(-3, 0, 0) +
A_cur(0, +3, 0) + A_cur(0, -3, 0) +
A_cur(0, 0, +3) + A_cur(0, 0, -3));
A_cur(0, +3, 0) + A_cur(0, -3, 0) +
A_cur(0, 0, +3) + A_cur(0, 0, -3));
#else
#define A_cur(x, y, z) Ain [index + (x) + (y) + (z)]
#define A_next(x, y, z) Aout[index + (x) + (y) + (z)]
double div = coef[0] * A_cur(0, 0, 0) +
coef[1] * (A_cur(+x1o, 0, 0) + A_cur(-x1o, 0, 0) +
A_cur(0, +y1o, 0) + A_cur(0, -y1o, 0) +
A_cur(0, 0, +z1o) + A_cur(0, 0, -z1o)) +
coef[2] * (A_cur(+x2o, 0, 0) + A_cur(-x2o, 0, 0) +
A_cur(0, +y2o, 0) + A_cur(0, -y2o, 0) +
A_cur(0, 0, +z2o) + A_cur(0, 0, -z2o)) +
coef[3] * (A_cur(+x3o, 0, 0) + A_cur(-x3o, 0, 0) +
A_cur(0, +y3o, 0) + A_cur(0, -y3o, 0) +
A_cur(0, 0, +z3o) + A_cur(0, 0, -z3o));
#endif
if (x < x1)
A_next(0, 0, 0) = 2.0d0 * A_cur(0, 0, 0) - A_next(0, 0, 0) +
vsq[index] * div;
}
if (x < x1)
A_next(0, 0, 0) = 2.0d0 * A_cur(0, 0, 0) - A_next(0, 0, 0) +
vsq[index] * div;
}
}
}
@@ -83,6 +117,8 @@ stencil_step_task(uniform int x0, uniform int x1,
uniform int Nx, uniform int Ny, uniform int Nz,
uniform const double coef[4], uniform const double vsq[],
uniform const double Ain[], uniform double Aout[]) {
if(taskIndex >= taskCount) return;
stencil_step(x0, x1, y0, y1, z0+taskIndex, z0+taskIndex+1,
Nx, Ny, Nz, coef, vsq, Ain, Aout);
}