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

changes

Browse Source
This commit is contained in:
Evghenii
2013-11-18 11:58:19 +01:00
parent 4d278a654e
commit 8d4dd13750
38 changed files with 5481 additions and 4073 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples_cuda/stencil/stencil.cpp
View File

@@ -71,7 +71,6 @@ extern void loop_stencil_serial(int t0, int t1, int x0, int x1,
void InitData(int Nx, int Ny, int Nz, double *A[2], double *vsq) {
int offset = 0;
#pragma omp parallel for collapse(2) private(offset)
for (int z = 0; z < Nz; ++z)
for (int y = 0; y < Ny; ++y)
for (int x = 0; x < Nx; ++x, ++offset) {
@@ -129,7 +128,7 @@ int main() {
loop_stencil_ispc_tasks(0, 6, width, Nx - width, width, Ny - width,
width, Nz - width, Nx, Ny, Nz, coeff, vsq,
Aispc[0], Aispc[1]);
double dt = rtc() - t0; //get_elapsed_mcycles();
double dt = 1e3*(rtc() - t0); //get_elapsed_mcycles();
minTimeISPCTasks = std::min(minTimeISPCTasks, dt);
}