diff options
Optionally generate cell-list-based OpenCL dispatch functions
Requires different function naming as OpenCL 1.2 doesn't support overloads.
The OpenCL kernel code generated using this commit was successfully tested
on an actual GPU. Time to set up some automatic validation.
Diffstat (limited to 'boltzgen/kernel/template/collect_moments.cl.mako')
-rw-r--r-- | boltzgen/kernel/template/collect_moments.cl.mako | 15 |
1 files changed, 12 insertions, 3 deletions
diff --git a/boltzgen/kernel/template/collect_moments.cl.mako b/boltzgen/kernel/template/collect_moments.cl.mako index ef520da..0ab42d1 100644 --- a/boltzgen/kernel/template/collect_moments.cl.mako +++ b/boltzgen/kernel/template/collect_moments.cl.mako @@ -1,6 +1,6 @@ -__kernel void collect_moments(__global ${float_type}* f, - __global ${float_type}* moments, - unsigned int gid) +__kernel void collect_moments_gid(__global ${float_type}* f, + __global ${float_type}* moments, + unsigned int gid) { __global ${float_type}* preshifted_f = f + gid; @@ -16,3 +16,12 @@ __kernel void collect_moments(__global ${float_type}* f, moments[${layout.pop_offset(i)} + gid] = ${ccode(expr.rhs)}; % endfor } + +% if 'cell_list_dispatch' in extras: +__kernel void collect_moments_cells(__global ${float_type}* f, + __global ${float_type}* moments, + __global unsigned int* cells) +{ + collect_moments_gid(f, moments, cells[get_global_id(0)]); +} +% endif |