wassname/Pointnet2_PyTorch
1
0
Fork 0
mirror of https://github.com/wassname/Pointnet2_PyTorch.git synced 2026-06-30 16:30:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

Some updates

Browse Source
This commit is contained in:
erikwijmans
2018-01-30 16:22:43 -05:00
parent c4ddd6b099
commit 8bce353da4
4 changed files with 15 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
utils/cinclude/cuda_utils.h
+3 -15
View File
@@ -1,24 +1,12 @@
#ifndef _CUDA_UTILS_H
#define _CUDA_UTILS_H
#ifdef __cplusplus
extern "C" {
#endif
#include <cmath>
inline int opt_n_threads(int work_size) {
unsigned int n_threads = work_size;
n_threads--;
n_threads |= n_threads >> 1;
n_threads |= n_threads >> 2;
n_threads |= n_threads >> 4;
n_threads |= n_threads >> 8;
n_threads |= n_threads >> 16;
n_threads++;
const int pow_2 = std::log(static_cast<double>(work_size)) / std::log(2.0);
return max(min(n_threads / 2, 512), 2);
return max(min(1 << pow_2, 512), 32);
}
#ifdef __cplusplus
}
#endif
#endif
utils/csrc/sampling_gpu.cu
+5 -4
View File
@@ -13,9 +13,10 @@ __global__ void gather_points_kernel(int b, int n, int c, int m,
for (int i = blockIdx.x; i < b; i += gridDim.x) {
for (int j = blockIdx.y * blockDim.x + threadIdx.x; j < m;
j += blockDim.x * gridDim.y) {
int a = idx[i * m + j];
memcpy(out + (i * m + j) * c, points + (i * n + a) * c,
sizeof(float) * c);
const int jj = idx[i * m + j];
for (int l = 0; l < c; ++l) {
out[(i * m + j) * c + l] = points[(i * n + jj) * c + l];
}
}
}
}
@@ -25,7 +26,7 @@ void gather_points_kernel_wrapper(int b, int n, int c, int npoints,
float *out, cudaStream_t stream) {
cudaError_t err;
gather_points_kernel<<<dim3(2, 8, 1), opt_n_threads(npoints) / 4, 0,
gather_points_kernel<<<dim3(b, 8, 1), opt_n_threads(npoints), 0,
stream>>>(b, n, c, npoints, points, idx, out);
err = cudaGetLastError();