hofee/gen_data_curobo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #299 from AndyZe/cleanup

Browse Source 
minor cleanups to cpp files
This commit is contained in:
Balakumar Sundaralingam
2024-07-10 18:30:45 -07:00
committed by GitHub
parent 14b554bf36 991994581f
commit 0c122a3353
2 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/curobo/curobolib/cpp/geom_cuda.cpp
View File

@@ -32,8 +32,6 @@ std::vector<torch::Tensor>self_collision_distance(
const int ndpt = 8, // Does this need to match template? const int ndpt = 8, // Does this need to match template?
const bool debug = false); const bool debug = false);
// CUDA forward declarations
std::vector<torch::Tensor>swept_sphere_obb_clpt( std::vector<torch::Tensor>swept_sphere_obb_clpt(
const torch::Tensor sphere_position, // batch_size, 3 const torch::Tensor sphere_position, // batch_size, 3
torch::Tensor distance, // batch_size, 1 torch::Tensor distance, // batch_size, 1
@@ -83,6 +81,7 @@ sphere_obb_clpt(const torch::Tensor sphere_position, // batch_size, 4
const bool use_batch_env, const bool use_batch_env,
const bool sum_collisions, const bool sum_collisions,
const bool compute_esdf); const bool compute_esdf);
std::vector<torch::Tensor> std::vector<torch::Tensor>
sphere_voxel_clpt(const torch::Tensor sphere_position, // batch_size, 3 sphere_voxel_clpt(const torch::Tensor sphere_position, // batch_size, 3
torch::Tensor distance, torch::Tensor distance,
@@ -126,6 +125,7 @@ swept_sphere_voxel_clpt(const torch::Tensor sphere_position, // batch_size, 3
const bool compute_distance, const bool compute_distance,
const bool use_batch_env, const bool use_batch_env,
const bool sum_collisions); const bool sum_collisions);
std::vector<torch::Tensor>pose_distance( std::vector<torch::Tensor>pose_distance(
torch::Tensor out_distance, torch::Tensor out_distance,
torch::Tensor out_position_distance, torch::Tensor out_position_distance,

6
src/curobo/curobolib/cpp/self_collision_kernel.cu
View File

@@ -162,7 +162,7 @@ namespace Curobo
////////////////////////////////////////////////////// //////////////////////////////////////////////////////
// Compute distances and store the maximum per thread // Compute distances and store the maximum per thread
// in registers (max_d). // in registers (max_d).
// Each thread computes upto ndpt distances. // Each thread computes up to ndpt distances.
// two warps per row // two warps per row
////////////////////////////////////////////////////// //////////////////////////////////////////////////////
// int nspheres_2 = nspheres * nspheres; // int nspheres_2 = nspheres * nspheres;
@@ -175,8 +175,6 @@ namespace Curobo
{ {
sph2 = __rs_shared[j]; // we need not load sph2 in every iteration. sph2 = __rs_shared[j]; // we need not load sph2 in every iteration.
#pragma unroll 16
for (int k = 0; k < ndpt; k++, i++) // increment i also here for (int k = 0; k < ndpt; k++, i++) // increment i also here
{ {
if ((i < nspheres) && (j > i)) if ((i < nspheres) && (j > i))
@@ -230,6 +228,8 @@ namespace Curobo
if (threadIdx.x < blockDim.x) if (threadIdx.x < blockDim.x)
{ {
// dist_t max_d = dist_sh[threadIdx.x]; // dist_t max_d = dist_sh[threadIdx.x];
#pragma unroll 4
for (int offset = 16; offset > 0; offset /= 2) for (int offset = 16; offset > 0; offset /= 2)
{ {
uint64_t nd = __shfl_down_sync(mask, *(uint64_t *)&max_d, offset); uint64_t nd = __shfl_down_sync(mask, *(uint64_t *)&max_d, offset);