[REQ] Allow block dimension to be retrieved in a kernel with wp.block_dim() #695
Description
Description
def _ray_all_geom(
...
)
worldid, rayid, tid = wp.tid()
num_threads_in_thread_block = wp.block_dim()
ngeom = m.ngeom
upper = ((ngeom + num_threads_in_thread_block - 1) // num_threads_in_thread_block) * num_threads_in_thread_block
for geom_id in range(tid, upper, num_threads_in_thread_block):
cur_dist = max_dist
if geom_id < ngeom:
... compute cur_dist
t = wp.tile(cur_dist)
local_min_idx = wp.tile_argmin(t)
local_min_val = t[local_min_idx[0]]
Context
Improve support for strided loops when working with tiles to allow the same code to run on CPU and CUDA devices.