Skip to content

GitLab

Commit d1b9ae40 authored by zhangwenwei's avatar zhangwenwei
Browse files

clean c files

parent 16c3f6e1
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 1598 additions and 1385 deletions
.pre-commit-config.yaml
View file @ d1b9ae40
......@@ -10,11 +10,12 @@ repos:
- repo: https://github.com/timothycrosley/isort
rev: 4.3.21
hooks:
- id: isort
- id: isort
- repo: https://github.com/pre-commit/mirrors-yapf
rev: v0.30.0
hooks:
- id: yapf
- repo: https://github.com/pre-commit/pre-commit-hooks
rev: v2.5.0
hooks:
......
mmdet3d/ops/ball_query/src/ball_query.cpp
View file @ d1b9ae40
#include <torch/serialize/tensor.h>
#include <vector>
#include <THC/THC.h>
#include <cuda.h>
#include <cuda_runtime_api.h>
#include <torch/extension.h>
#include <torch/serialize/tensor.h>
#include <vector>
extern THCState *state;
#define CHECK_CUDA(x) AT_CHECK(x.type().is_cuda(), #x, " must be a CUDAtensor ")
#define CHECK_CONTIGUOUS(x) AT_CHECK(x.is_contiguous(), #x, " must be contiguous ")
#define CHECK_INPUT(x) CHECK_CUDA(x);CHECK_CONTIGUOUS(x)
#define CHECK_CUDA(x) \
TORCH_CHECK(x.type().is_cuda(), #x, " must be a CUDAtensor ")
#define CHECK_CONTIGUOUS(x) \
TORCH_CHECK(x.is_contiguous(), #x, " must be contiguous ")
#define CHECK_INPUT(x) \
CHECK_CUDA(x); \
CHECK_CONTIGUOUS(x)
int ball_query_wrapper(int b, int n, int m, float radius, int nsample,
at::Tensor new_xyz_tensor, at::Tensor xyz_tensor, at::Tensor idx_tensor);
at::Tensor new_xyz_tensor, at::Tensor xyz_tensor,
at::Tensor idx_tensor);
void ball_query_kernel_launcher(int b, int n, int m, float radius, int nsample,
const float *xyz, const float *new_xyz, int *idx, cudaStream_t stream);
const float *xyz, const float *new_xyz,
int *idx, cudaStream_t stream);
int ball_query_wrapper(int b, int n, int m, float radius, int nsample,
at::Tensor new_xyz_tensor, at::Tensor xyz_tensor, at::Tensor idx_tensor) {
CHECK_INPUT(new_xyz_tensor);
CHECK_INPUT(xyz_tensor);
const float *new_xyz = new_xyz_tensor.data<float>();
const float *xyz = xyz_tensor.data<float>();
int *idx = idx_tensor.data<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
ball_query_kernel_launcher(b, n, m, radius, nsample, new_xyz, xyz, idx, stream);
return 1;
at::Tensor new_xyz_tensor, at::Tensor xyz_tensor,
at::Tensor idx_tensor) {
CHECK_INPUT(new_xyz_tensor);
CHECK_INPUT(xyz_tensor);
const float *new_xyz = new_xyz_tensor.data_ptr<float>();
const float *xyz = xyz_tensor.data_ptr<float>();
int *idx = idx_tensor.data_ptr<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
ball_query_kernel_launcher(b, n, m, radius, nsample, new_xyz, xyz, idx,
stream);
return 1;
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("ball_query_wrapper", &ball_query_wrapper, "ball_query_wrapper");
m.def("ball_query_wrapper", &ball_query_wrapper, "ball_query_wrapper");
}
mmdet3d/ops/ball_query/src/ball_query_cuda.cu
View file @ d1b9ae40
......@@ -3,65 +3,70 @@
#include <stdlib.h>
#define THREADS_PER_BLOCK 256
#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
#define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
__global__ void ball_query_kernel(int b, int n, int m, float radius,
int nsample,
const float *__restrict__ new_xyz,
const float *__restrict__ xyz,
int *__restrict__ idx) {
// new_xyz: (B, M, 3)
// xyz: (B, N, 3)
// output:
// idx: (B, M, nsample)
int bs_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || pt_idx >= m) return;
__global__ void ball_query_kernel(int b, int n, int m, float radius, int nsample,
const float *__restrict__ new_xyz, const float *__restrict__ xyz, int *__restrict__ idx) {
// new_xyz: (B, M, 3)
// xyz: (B, N, 3)
// output:
// idx: (B, M, nsample)
int bs_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || pt_idx >= m) return;
new_xyz += bs_idx * m * 3 + pt_idx * 3;
xyz += bs_idx * n * 3;
idx += bs_idx * m * nsample + pt_idx * nsample;
new_xyz += bs_idx * m * 3 + pt_idx * 3;
xyz += bs_idx * n * 3;
idx += bs_idx * m * nsample + pt_idx * nsample;
float radius2 = radius * radius;
float new_x = new_xyz[0];
float new_y = new_xyz[1];
float new_z = new_xyz[2];
float radius2 = radius * radius;
float new_x = new_xyz[0];
float new_y = new_xyz[1];
float new_z = new_xyz[2];
int cnt = 0;
for (int k = 0; k < n; ++k) {
float x = xyz[k * 3 + 0];
float y = xyz[k * 3 + 1];
float z = xyz[k * 3 + 2];
float d2 = (new_x - x) * (new_x - x) + (new_y - y) * (new_y - y) + (new_z - z) * (new_z - z);
if (d2 < radius2){
if (cnt == 0){
for (int l = 0; l < nsample; ++l) {
idx[l] = k;
}
}
idx[cnt] = k;
++cnt;
if (cnt >= nsample) break;
int cnt = 0;
for (int k = 0; k < n; ++k) {
float x = xyz[k * 3 + 0];
float y = xyz[k * 3 + 1];
float z = xyz[k * 3 + 2];
float d2 = (new_x - x) * (new_x - x) + (new_y - y) * (new_y - y) +
(new_z - z) * (new_z - z);
if (d2 < radius2) {
if (cnt == 0) {
for (int l = 0; l < nsample; ++l) {
idx[l] = k;
}
}
idx[cnt] = k;
++cnt;
if (cnt >= nsample) break;
}
}
}
void ball_query_kernel_launcher(int b, int n, int m, float radius, int nsample,
const float *new_xyz, const float *xyz,
int *idx, cudaStream_t stream) {
// new_xyz: (B, M, 3)
// xyz: (B, N, 3)
// output:
// idx: (B, M, nsample)
void ball_query_kernel_launcher(int b, int n, int m, float radius, int nsample, \
const float *new_xyz, const float *xyz, int *idx, cudaStream_t stream) {
// new_xyz: (B, M, 3)
// xyz: (B, N, 3)
// output:
// idx: (B, M, nsample)
cudaError_t err;
cudaError_t err;
dim3 blocks(DIVUP(m, THREADS_PER_BLOCK), b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
dim3 blocks(DIVUP(m, THREADS_PER_BLOCK),
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
ball_query_kernel<<<blocks, threads, 0, stream>>>(b, n, m, radius, nsample, new_xyz, xyz, idx);
// cudaDeviceSynchronize(); // for using printf in kernel function
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
ball_query_kernel<<<blocks, threads, 0, stream>>>(b, n, m, radius, nsample,
new_xyz, xyz, idx);
// cudaDeviceSynchronize(); // for using printf in kernel function
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
mmdet3d/ops/furthest_point_sample/src/furthest_point_sample.cpp
View file @ d1b9ae40
#include <torch/serialize/tensor.h>
#include <ATen/cuda/CUDAContext.h>
#include <vector>
#include <THC/THC.h>
#include <torch/extension.h>
#include <torch/serialize/tensor.h>
#include <vector>
extern THCState *state;
int furthest_point_sampling_wrapper(int b, int n, int m,
at::Tensor points_tensor, at::Tensor temp_tensor, at::Tensor idx_tensor);
at::Tensor points_tensor,
at::Tensor temp_tensor,
at::Tensor idx_tensor);
void furthest_point_sampling_kernel_launcher(int b, int n, int m,
const float *dataset, float *temp, int *idxs, cudaStream_t stream);
const float *dataset, float *temp,
int *idxs, cudaStream_t stream);
int furthest_point_sampling_wrapper(int b, int n, int m,
at::Tensor points_tensor, at::Tensor temp_tensor, at::Tensor idx_tensor) {
const float *points = points_tensor.data<float>();
float *temp = temp_tensor.data<float>();
int *idx = idx_tensor.data<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
furthest_point_sampling_kernel_launcher(b, n, m, points, temp, idx, stream);
return 1;
at::Tensor points_tensor,
at::Tensor temp_tensor,
at::Tensor idx_tensor) {
const float *points = points_tensor.data_ptr<float>();
float *temp = temp_tensor.data_ptr<float>();
int *idx = idx_tensor.data_ptr<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
furthest_point_sampling_kernel_launcher(b, n, m, points, temp, idx, stream);
return 1;
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("furthest_point_sampling_wrapper", &furthest_point_sampling_wrapper, "furthest_point_sampling_wrapper");
m.def("furthest_point_sampling_wrapper", &furthest_point_sampling_wrapper,
"furthest_point_sampling_wrapper");
}
mmdet3d/ops/furthest_point_sample/src/furthest_point_sample_cuda.cu
View file @ d1b9ae40
......@@ -3,179 +3,204 @@
#define TOTAL_THREADS 1024
#define THREADS_PER_BLOCK 256
#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
#define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
inline int opt_n_threads(int work_size) {
const int pow_2 = std::log(static_cast<double>(work_size)) / std::log(2.0);
const int pow_2 = std::log(static_cast<double>(work_size)) / std::log(2.0);
return max(min(1 << pow_2, TOTAL_THREADS), 1);
return max(min(1 << pow_2, TOTAL_THREADS), 1);
}
__device__ void __update(float *__restrict__ dists, int *__restrict__ dists_i, int idx1, int idx2){
const float v1 = dists[idx1], v2 = dists[idx2];
const int i1 = dists_i[idx1], i2 = dists_i[idx2];
dists[idx1] = max(v1, v2);
dists_i[idx1] = v2 > v1 ? i2 : i1;
__device__ void __update(float *__restrict__ dists, int *__restrict__ dists_i,
int idx1, int idx2) {
const float v1 = dists[idx1], v2 = dists[idx2];
const int i1 = dists_i[idx1], i2 = dists_i[idx2];
dists[idx1] = max(v1, v2);
dists_i[idx1] = v2 > v1 ? i2 : i1;
}
template <unsigned int block_size>
__global__ void furthest_point_sampling_kernel(int b, int n, int m,
const float *__restrict__ dataset, float *__restrict__ temp, int *__restrict__ idxs) {
// dataset: (B, N, 3)
// tmp: (B, N)
// output:
// idx: (B, M)
if (m <= 0) return;
__shared__ float dists[block_size];
__shared__ int dists_i[block_size];
int batch_index = blockIdx.x;
dataset += batch_index * n * 3;
temp += batch_index * n;
idxs += batch_index * m;
int tid = threadIdx.x;
const int stride = block_size;
int old = 0;
if (threadIdx.x == 0)
idxs[0] = old;
__syncthreads();
for (int j = 1; j < m; j++) {
__global__ void furthest_point_sampling_kernel(
int b, int n, int m, const float *__restrict__ dataset,
float *__restrict__ temp, int *__restrict__ idxs) {
// dataset: (B, N, 3)
// tmp: (B, N)
// output:
// idx: (B, M)
if (m <= 0) return;
__shared__ float dists[block_size];
__shared__ int dists_i[block_size];
int batch_index = blockIdx.x;
dataset += batch_index * n * 3;
temp += batch_index * n;
idxs += batch_index * m;
int tid = threadIdx.x;
const int stride = block_size;
int old = 0;
if (threadIdx.x == 0) idxs[0] = old;
__syncthreads();
for (int j = 1; j < m; j++) {
int besti = 0;
float best = -1;
float x1 = dataset[old * 3 + 0];
float y1 = dataset[old * 3 + 1];
float z1 = dataset[old * 3 + 2];
for (int k = tid; k < n; k += stride) {
float x2, y2, z2;
x2 = dataset[k * 3 + 0];
y2 = dataset[k * 3 + 1];
z2 = dataset[k * 3 + 2];
// float mag = (x2 * x2) + (y2 * y2) + (z2 * z2);
// if (mag <= 1e-3)
// continue;
float d = (x2 - x1) * (x2 - x1) + (y2 - y1) * (y2 - y1) + (z2 - z1) * (z2 - z1);
float d2 = min(d, temp[k]);
temp[k] = d2;
besti = d2 > best ? k : besti;
best = d2 > best ? d2 : best;
float x2, y2, z2;
x2 = dataset[k * 3 + 0];
y2 = dataset[k * 3 + 1];
z2 = dataset[k * 3 + 2];
// float mag = (x2 * x2) + (y2 * y2) + (z2 * z2);
// if (mag <= 1e-3)
// continue;
float d =
(x2 - x1) * (x2 - x1) + (y2 - y1) * (y2 - y1) + (z2 - z1) * (z2 - z1);
float d2 = min(d, temp[k]);
temp[k] = d2;
besti = d2 > best ? k : besti;
best = d2 > best ? d2 : best;
}
dists[tid] = best;
dists_i[tid] = besti;
__syncthreads();
if (block_size >= 1024) {
if (tid < 512) {
__update(dists, dists_i, tid, tid + 512);
}
__syncthreads();
if (tid < 512) {
__update(dists, dists_i, tid, tid + 512);
}
__syncthreads();
}
if (block_size >= 512) {
if (tid < 256) {
__update(dists, dists_i, tid, tid + 256);
}
__syncthreads();
if (tid < 256) {
__update(dists, dists_i, tid, tid + 256);
}
__syncthreads();
}
if (block_size >= 256) {
if (tid < 128) {
__update(dists, dists_i, tid, tid + 128);
}
__syncthreads();
if (tid < 128) {
__update(dists, dists_i, tid, tid + 128);
}
__syncthreads();
}
if (block_size >= 128) {
if (tid < 64) {
__update(dists, dists_i, tid, tid + 64);
}
__syncthreads();
if (tid < 64) {
__update(dists, dists_i, tid, tid + 64);
}
__syncthreads();
}
if (block_size >= 64) {
if (tid < 32) {
__update(dists, dists_i, tid, tid + 32);
}
__syncthreads();
if (tid < 32) {
__update(dists, dists_i, tid, tid + 32);
}
__syncthreads();
}
if (block_size >= 32) {
if (tid < 16) {
__update(dists, dists_i, tid, tid + 16);
}
__syncthreads();
if (tid < 16) {
__update(dists, dists_i, tid, tid + 16);
}
__syncthreads();
}
if (block_size >= 16) {
if (tid < 8) {
__update(dists, dists_i, tid, tid + 8);
}
__syncthreads();
if (tid < 8) {
__update(dists, dists_i, tid, tid + 8);
}
__syncthreads();
}
if (block_size >= 8) {
if (tid < 4) {
__update(dists, dists_i, tid, tid + 4);
}
__syncthreads();
if (tid < 4) {
__update(dists, dists_i, tid, tid + 4);
}
__syncthreads();
}
if (block_size >= 4) {
if (tid < 2) {
__update(dists, dists_i, tid, tid + 2);
}
__syncthreads();
if (tid < 2) {
__update(dists, dists_i, tid, tid + 2);
}
__syncthreads();
}
if (block_size >= 2) {
if (tid < 1) {
__update(dists, dists_i, tid, tid + 1);
}
__syncthreads();
if (tid < 1) {
__update(dists, dists_i, tid, tid + 1);
}
__syncthreads();
}
old = dists_i[0];
if (tid == 0)
idxs[j] = old;
}
if (tid == 0) idxs[j] = old;
}
}
void furthest_point_sampling_kernel_launcher(int b, int n, int m,
const float *dataset, float *temp, int *idxs, cudaStream_t stream) {
// dataset: (B, N, 3)
// tmp: (B, N)
// output:
// idx: (B, M)
cudaError_t err;
unsigned int n_threads = opt_n_threads(n);
switch (n_threads) {
case 1024:
furthest_point_sampling_kernel<1024><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 512:
furthest_point_sampling_kernel<512><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 256:
furthest_point_sampling_kernel<256><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 128:
furthest_point_sampling_kernel<128><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 64:
furthest_point_sampling_kernel<64><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 32:
furthest_point_sampling_kernel<32><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 16:
furthest_point_sampling_kernel<16><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 8:
furthest_point_sampling_kernel<8><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 4:
furthest_point_sampling_kernel<4><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 2:
furthest_point_sampling_kernel<2><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
case 1:
furthest_point_sampling_kernel<1><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs); break;
default:
furthest_point_sampling_kernel<512><<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
}
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
const float *dataset, float *temp,
int *idxs, cudaStream_t stream) {
// dataset: (B, N, 3)
// tmp: (B, N)
// output:
// idx: (B, M)
cudaError_t err;
unsigned int n_threads = opt_n_threads(n);
switch (n_threads) {
case 1024:
furthest_point_sampling_kernel<1024>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 512:
furthest_point_sampling_kernel<512>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 256:
furthest_point_sampling_kernel<256>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 128:
furthest_point_sampling_kernel<128>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 64:
furthest_point_sampling_kernel<64>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 32:
furthest_point_sampling_kernel<32>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 16:
furthest_point_sampling_kernel<16>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 8:
furthest_point_sampling_kernel<8>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 4:
furthest_point_sampling_kernel<4>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 2:
furthest_point_sampling_kernel<2>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
case 1:
furthest_point_sampling_kernel<1>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
break;
default:
furthest_point_sampling_kernel<512>
<<<b, n_threads, 0, stream>>>(b, n, m, dataset, temp, idxs);
}
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
mmdet3d/ops/gather_points/src/gather_points.cpp
View file @ d1b9ae40
#include <torch/serialize/tensor.h>
#include <ATen/cuda/CUDAContext.h>
#include <vector>
#include <THC/THC.h>
#include <torch/extension.h>
#include <torch/serialize/tensor.h>
#include <vector>
extern THCState *state;
int gather_points_wrapper(int b, int c, int n, int npoints,
at::Tensor points_tensor, at::Tensor idx_tensor, at::Tensor out_tensor);
at::Tensor points_tensor, at::Tensor idx_tensor,
at::Tensor out_tensor);
void gather_points_kernel_launcher(int b, int c, int n, int npoints,
const float *points, const int *idx, float *out, cudaStream_t stream);
const float *points, const int *idx,
float *out, cudaStream_t stream);
int gather_points_grad_wrapper(int b, int c, int n, int npoints,
at::Tensor grad_out_tensor, at::Tensor idx_tensor, at::Tensor grad_points_tensor);
at::Tensor grad_out_tensor,
at::Tensor idx_tensor,
at::Tensor grad_points_tensor);
void gather_points_grad_kernel_launcher(int b, int c, int n, int npoints,
const float *grad_out, const int *idx, float *grad_points, cudaStream_t stream);
const float *grad_out, const int *idx,
float *grad_points,
cudaStream_t stream);
int gather_points_wrapper(int b, int c, int n, int npoints,
at::Tensor points_tensor, at::Tensor idx_tensor, at::Tensor out_tensor){
const float *points = points_tensor.data<float>();
const int *idx = idx_tensor.data<int>();
float *out = out_tensor.data<float>();
cudaStream_t stream = THCState_getCurrentStream(state);
gather_points_kernel_launcher(b, c, n, npoints, points, idx, out, stream);
return 1;
at::Tensor points_tensor, at::Tensor idx_tensor,
at::Tensor out_tensor) {
const float *points = points_tensor.data_ptr<float>();
const int *idx = idx_tensor.data_ptr<int>();
float *out = out_tensor.data_ptr<float>();
cudaStream_t stream = THCState_getCurrentStream(state);
gather_points_kernel_launcher(b, c, n, npoints, points, idx, out, stream);
return 1;
}
int gather_points_grad_wrapper(int b, int c, int n, int npoints,
at::Tensor grad_out_tensor, at::Tensor idx_tensor, at::Tensor grad_points_tensor) {
const float *grad_out = grad_out_tensor.data<float>();
const int *idx = idx_tensor.data<int>();
float *grad_points = grad_points_tensor.data<float>();
cudaStream_t stream = THCState_getCurrentStream(state);
gather_points_grad_kernel_launcher(b, c, n, npoints, grad_out, idx, grad_points, stream);
return 1;
at::Tensor grad_out_tensor,
at::Tensor idx_tensor,
at::Tensor grad_points_tensor) {
const float *grad_out = grad_out_tensor.data_ptr<float>();
const int *idx = idx_tensor.data_ptr<int>();
float *grad_points = grad_points_tensor.data_ptr<float>();
cudaStream_t stream = THCState_getCurrentStream(state);
gather_points_grad_kernel_launcher(b, c, n, npoints, grad_out, idx,
grad_points, stream);
return 1;
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("gather_points_wrapper", &gather_points_wrapper, "gather_points_wrapper");
m.def("gather_points_grad_wrapper", &gather_points_grad_wrapper, "gather_points_grad_wrapper");
m.def("gather_points_wrapper", &gather_points_wrapper,
"gather_points_wrapper");
m.def("gather_points_grad_wrapper", &gather_points_grad_wrapper,
"gather_points_grad_wrapper");
}
mmdet3d/ops/gather_points/src/gather_points_cuda.cu
View file @ d1b9ae40
......@@ -3,82 +3,92 @@
#define TOTAL_THREADS 1024
#define THREADS_PER_BLOCK 256
#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
#define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
__global__ void gather_points_kernel(int b, int c, int n, int m,
const float *__restrict__ points, const int *__restrict__ idx, float *__restrict__ out) {
// points: (B, C, N)
// idx: (B, M)
// output:
// out: (B, C, M)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= m) return;
out += bs_idx * c * m + c_idx * m + pt_idx;
idx += bs_idx * m + pt_idx;
points += bs_idx * c * n + c_idx * n;
out[0] = points[idx[0]];
const float *__restrict__ points,
const int *__restrict__ idx,
float *__restrict__ out) {
// points: (B, C, N)
// idx: (B, M)
// output:
// out: (B, C, M)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= m) return;
out += bs_idx * c * m + c_idx * m + pt_idx;
idx += bs_idx * m + pt_idx;
points += bs_idx * c * n + c_idx * n;
out[0] = points[idx[0]];
}
void gather_points_kernel_launcher(int b, int c, int n, int npoints,
const float *points, const int *idx, float *out, cudaStream_t stream) {
// points: (B, C, N)
// idx: (B, npoints)
// output:
// out: (B, C, npoints)
cudaError_t err;
dim3 blocks(DIVUP(npoints, THREADS_PER_BLOCK), c, b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
gather_points_kernel<<<blocks, threads, 0, stream>>>(b, c, n, npoints, points, idx, out);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
const float *points, const int *idx,
float *out, cudaStream_t stream) {
// points: (B, C, N)
// idx: (B, npoints)
// output:
// out: (B, C, npoints)
cudaError_t err;
dim3 blocks(DIVUP(npoints, THREADS_PER_BLOCK), c,
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
gather_points_kernel<<<blocks, threads, 0, stream>>>(b, c, n, npoints, points,
idx, out);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
__global__ void gather_points_grad_kernel(int b, int c, int n, int m, const float *__restrict__ grad_out,
const int *__restrict__ idx, float *__restrict__ grad_points) {
// grad_out: (B, C, M)
// idx: (B, M)
// output:
// grad_points: (B, C, N)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= m) return;
grad_out += bs_idx * c * m + c_idx * m + pt_idx;
idx += bs_idx * m + pt_idx;
grad_points += bs_idx * c * n + c_idx * n;
atomicAdd(grad_points + idx[0], grad_out[0]);
__global__ void gather_points_grad_kernel(int b, int c, int n, int m,
const float *__restrict__ grad_out,
const int *__restrict__ idx,
float *__restrict__ grad_points) {
// grad_out: (B, C, M)
// idx: (B, M)
// output:
// grad_points: (B, C, N)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= m) return;
grad_out += bs_idx * c * m + c_idx * m + pt_idx;
idx += bs_idx * m + pt_idx;
grad_points += bs_idx * c * n + c_idx * n;
atomicAdd(grad_points + idx[0], grad_out[0]);
}
void gather_points_grad_kernel_launcher(int b, int c, int n, int npoints,
const float *grad_out, const int *idx, float *grad_points, cudaStream_t stream) {
// grad_out: (B, C, npoints)
// idx: (B, npoints)
// output:
// grad_points: (B, C, N)
cudaError_t err;
dim3 blocks(DIVUP(npoints, THREADS_PER_BLOCK), c, b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
gather_points_grad_kernel<<<blocks, threads, 0, stream>>>(b, c, n, npoints, grad_out, idx, grad_points);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
const float *grad_out, const int *idx,
float *grad_points,
cudaStream_t stream) {
// grad_out: (B, C, npoints)
// idx: (B, npoints)
// output:
// grad_points: (B, C, N)
cudaError_t err;
dim3 blocks(DIVUP(npoints, THREADS_PER_BLOCK), c,
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
gather_points_grad_kernel<<<blocks, threads, 0, stream>>>(
b, c, n, npoints, grad_out, idx, grad_points);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
mmdet3d/ops/group_points/src/group_points.cpp
View file @ d1b9ae40
#include <torch/serialize/tensor.h>
#include <THC/THC.h>
#include <cuda.h>
#include <cuda_runtime_api.h>
#include <vector>
#include <THC/THC.h>
#include <torch/extension.h>
#include <torch/serialize/tensor.h>
#include <vector>
extern THCState *state;
int group_points_wrapper(int b, int c, int n, int npoints, int nsample,
at::Tensor points_tensor, at::Tensor idx_tensor, at::Tensor out_tensor);
at::Tensor points_tensor, at::Tensor idx_tensor,
at::Tensor out_tensor);
void group_points_kernel_launcher(int b, int c, int n, int npoints, int nsample,
const float *points, const int *idx, float *out, cudaStream_t stream);
const float *points, const int *idx,
float *out, cudaStream_t stream);
int group_points_grad_wrapper(int b, int c, int n, int npoints, int nsample,
at::Tensor grad_out_tensor, at::Tensor idx_tensor, at::Tensor grad_points_tensor);
void group_points_grad_kernel_launcher(int b, int c, int n, int npoints, int nsample,
const float *grad_out, const int *idx, float *grad_points, cudaStream_t stream);
at::Tensor grad_out_tensor, at::Tensor idx_tensor,
at::Tensor grad_points_tensor);
void group_points_grad_kernel_launcher(int b, int c, int n, int npoints,
int nsample, const float *grad_out,
const int *idx, float *grad_points,
cudaStream_t stream);
int group_points_grad_wrapper(int b, int c, int n, int npoints, int nsample,
at::Tensor grad_out_tensor, at::Tensor idx_tensor, at::Tensor grad_points_tensor) {
float *grad_points = grad_points_tensor.data<float>();
const int *idx = idx_tensor.data<int>();
const float *grad_out = grad_out_tensor.data<float>();
at::Tensor grad_out_tensor, at::Tensor idx_tensor,
at::Tensor grad_points_tensor) {
float *grad_points = grad_points_tensor.data_ptr<float>();
const int *idx = idx_tensor.data_ptr<int>();
const float *grad_out = grad_out_tensor.data_ptr<float>();
cudaStream_t stream = THCState_getCurrentStream(state);
cudaStream_t stream = THCState_getCurrentStream(state);
group_points_grad_kernel_launcher(b, c, n, npoints, nsample, grad_out, idx, grad_points, stream);
return 1;
group_points_grad_kernel_launcher(b, c, n, npoints, nsample, grad_out, idx,
grad_points, stream);
return 1;
}
int group_points_wrapper(int b, int c, int n, int npoints, int nsample,
at::Tensor points_tensor, at::Tensor idx_tensor, at::Tensor out_tensor) {
const float *points = points_tensor.data<float>();
const int *idx = idx_tensor.data<int>();
float *out = out_tensor.data<float>();
at::Tensor points_tensor, at::Tensor idx_tensor,
at::Tensor out_tensor) {
const float *points = points_tensor.data_ptr<float>();
const int *idx = idx_tensor.data_ptr<int>();
float *out = out_tensor.data_ptr<float>();
cudaStream_t stream = THCState_getCurrentStream(state);
cudaStream_t stream = THCState_getCurrentStream(state);
group_points_kernel_launcher(b, c, n, npoints, nsample, points, idx, out, stream);
return 1;
group_points_kernel_launcher(b, c, n, npoints, nsample, points, idx, out,
stream);
return 1;
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("forward", &group_points_wrapper, "group_points_wrapper");
m.def("backward", &group_points_grad_wrapper, "group_points_grad_wrapper");
m.def("forward", &group_points_wrapper, "group_points_wrapper");
m.def("backward", &group_points_grad_wrapper, "group_points_grad_wrapper");
}
mmdet3d/ops/group_points/src/group_points_cuda.cu
View file @ d1b9ae40
......@@ -2,84 +2,97 @@
#include <stdlib.h>
#define THREADS_PER_BLOCK 256
#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
#define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
__global__ void group_points_grad_kernel(int b, int c, int n, int npoints, int nsample,
const float *__restrict__ grad_out, const int *__restrict__ idx, float *__restrict__ grad_points) {
// grad_out: (B, C, npoints, nsample)
// idx: (B, npoints, nsample)
// output:
// grad_points: (B, C, N)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int index = blockIdx.x * blockDim.x + threadIdx.x;
int pt_idx = index / nsample;
if (bs_idx >= b || c_idx >= c || pt_idx >= npoints) return;
__global__ void group_points_grad_kernel(int b, int c, int n, int npoints,
int nsample,
const float *__restrict__ grad_out,
const int *__restrict__ idx,
float *__restrict__ grad_points) {
// grad_out: (B, C, npoints, nsample)
// idx: (B, npoints, nsample)
// output:
// grad_points: (B, C, N)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int index = blockIdx.x * blockDim.x + threadIdx.x;
int pt_idx = index / nsample;
if (bs_idx >= b || c_idx >= c || pt_idx >= npoints) return;
int sample_idx = index % nsample;
grad_out += bs_idx * c * npoints * nsample + c_idx * npoints * nsample + pt_idx * nsample + sample_idx;
idx += bs_idx * npoints * nsample + pt_idx * nsample + sample_idx;
int sample_idx = index % nsample;
grad_out += bs_idx * c * npoints * nsample + c_idx * npoints * nsample +
pt_idx * nsample + sample_idx;
idx += bs_idx * npoints * nsample + pt_idx * nsample + sample_idx;
atomicAdd(grad_points + bs_idx * c * n + c_idx * n + idx[0] , grad_out[0]);
atomicAdd(grad_points + bs_idx * c * n + c_idx * n + idx[0], grad_out[0]);
}
void group_points_grad_kernel_launcher(int b, int c, int n, int npoints, int nsample,
const float *grad_out, const int *idx, float *grad_points, cudaStream_t stream) {
// grad_out: (B, C, npoints, nsample)
// idx: (B, npoints, nsample)
// output:
// grad_points: (B, C, N)
cudaError_t err;
dim3 blocks(DIVUP(npoints * nsample, THREADS_PER_BLOCK), c, b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
void group_points_grad_kernel_launcher(int b, int c, int n, int npoints,
int nsample, const float *grad_out,
const int *idx, float *grad_points,
cudaStream_t stream) {
// grad_out: (B, C, npoints, nsample)
// idx: (B, npoints, nsample)
// output:
// grad_points: (B, C, N)
cudaError_t err;
dim3 blocks(DIVUP(npoints * nsample, THREADS_PER_BLOCK), c,
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
group_points_grad_kernel<<<blocks, threads, 0, stream>>>(b, c, n, npoints, nsample, grad_out, idx, grad_points);
group_points_grad_kernel<<<blocks, threads, 0, stream>>>(
b, c, n, npoints, nsample, grad_out, idx, grad_points);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
__global__ void group_points_kernel(int b, int c, int n, int npoints,
int nsample,
const float *__restrict__ points,
const int *__restrict__ idx,
float *__restrict__ out) {
// points: (B, C, N)
// idx: (B, npoints, nsample)
// output:
// out: (B, C, npoints, nsample)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int index = blockIdx.x * blockDim.x + threadIdx.x;
int pt_idx = index / nsample;
if (bs_idx >= b || c_idx >= c || pt_idx >= npoints) return;
__global__ void group_points_kernel(int b, int c, int n, int npoints, int nsample,
const float *__restrict__ points, const int *__restrict__ idx, float *__restrict__ out) {
// points: (B, C, N)
// idx: (B, npoints, nsample)
// output:
// out: (B, C, npoints, nsample)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int index = blockIdx.x * blockDim.x + threadIdx.x;
int pt_idx = index / nsample;
if (bs_idx >= b || c_idx >= c || pt_idx >= npoints) return;
int sample_idx = index % nsample;
int sample_idx = index % nsample;
idx += bs_idx * npoints * nsample + pt_idx * nsample + sample_idx;
int in_idx = bs_idx * c * n + c_idx * n + idx[0];
int out_idx = bs_idx * c * npoints * nsample + c_idx * npoints * nsample +
pt_idx * nsample + sample_idx;
idx += bs_idx * npoints * nsample + pt_idx * nsample + sample_idx;
int in_idx = bs_idx * c * n + c_idx * n + idx[0];
int out_idx = bs_idx * c * npoints * nsample + c_idx * npoints * nsample + pt_idx * nsample + sample_idx;
out[out_idx] = points[in_idx];
out[out_idx] = points[in_idx];
}
void group_points_kernel_launcher(int b, int c, int n, int npoints, int nsample,
const float *points, const int *idx, float *out, cudaStream_t stream) {
// points: (B, C, N)
// idx: (B, npoints, nsample)
// output:
// out: (B, C, npoints, nsample)
cudaError_t err;
dim3 blocks(DIVUP(npoints * nsample, THREADS_PER_BLOCK), c, b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
const float *points, const int *idx,
float *out, cudaStream_t stream) {
// points: (B, C, N)
// idx: (B, npoints, nsample)
// output:
// out: (B, C, npoints, nsample)
cudaError_t err;
dim3 blocks(DIVUP(npoints * nsample, THREADS_PER_BLOCK), c,
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
group_points_kernel<<<blocks, threads, 0, stream>>>(b, c, n, npoints, nsample, points, idx, out);
// cudaDeviceSynchronize(); // for using printf in kernel function
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
group_points_kernel<<<blocks, threads, 0, stream>>>(b, c, n, npoints, nsample,
points, idx, out);
// cudaDeviceSynchronize(); // for using printf in kernel function
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
mmdet3d/ops/interpolate/src/interpolate.cpp
View file @ d1b9ae40
#include <torch/serialize/tensor.h>
#include <vector>
#include <THC/THC.h>
#include <cuda.h>
#include <cuda_runtime_api.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <cuda.h>
#include <cuda_runtime_api.h>
#include <torch/extension.h>
#include <torch/serialize/tensor.h>
#include <vector>
extern THCState *state;
void three_nn_wrapper(int b, int n, int m, at::Tensor unknown_tensor,
at::Tensor known_tensor, at::Tensor dist2_tensor, at::Tensor idx_tensor);
at::Tensor known_tensor, at::Tensor dist2_tensor,
at::Tensor idx_tensor);
void three_nn_kernel_launcher(int b, int n, int m, const float *unknown,
const float *known, float *dist2, int *idx, cudaStream_t stream);
const float *known, float *dist2, int *idx,
cudaStream_t stream);
void three_interpolate_wrapper(int b, int c, int m, int n, at::Tensor points_tensor,
at::Tensor idx_tensor, at::Tensor weight_tensor, at::Tensor out_tensor);
void three_interpolate_wrapper(int b, int c, int m, int n,
at::Tensor points_tensor, at::Tensor idx_tensor,
at::Tensor weight_tensor, at::Tensor out_tensor);
void three_interpolate_kernel_launcher(int b, int c, int m, int n,
const float *points, const int *idx, const float *weight, float *out, cudaStream_t stream);
const float *points, const int *idx,
const float *weight, float *out,
cudaStream_t stream);
void three_interpolate_grad_wrapper(int b, int c, int n, int m,
at::Tensor grad_out_tensor,
at::Tensor idx_tensor,
at::Tensor weight_tensor,
at::Tensor grad_points_tensor);
void three_interpolate_grad_wrapper(int b, int c, int n, int m, at::Tensor grad_out_tensor,
at::Tensor idx_tensor, at::Tensor weight_tensor, at::Tensor grad_points_tensor);
void three_interpolate_grad_kernel_launcher(int b, int c, int n, int m, const float *grad_out,
const int *idx, const float *weight, float *grad_points, cudaStream_t stream);
void three_interpolate_grad_kernel_launcher(int b, int c, int n, int m,
const float *grad_out,
const int *idx, const float *weight,
float *grad_points,
cudaStream_t stream);
void three_nn_wrapper(int b, int n, int m, at::Tensor unknown_tensor,
at::Tensor known_tensor, at::Tensor dist2_tensor, at::Tensor idx_tensor) {
const float *unknown = unknown_tensor.data<float>();
const float *known = known_tensor.data<float>();
float *dist2 = dist2_tensor.data<float>();
int *idx = idx_tensor.data<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
three_nn_kernel_launcher(b, n, m, unknown, known, dist2, idx, stream);
at::Tensor known_tensor, at::Tensor dist2_tensor,
at::Tensor idx_tensor) {
const float *unknown = unknown_tensor.data_ptr<float>();
const float *known = known_tensor.data_ptr<float>();
float *dist2 = dist2_tensor.data_ptr<float>();
int *idx = idx_tensor.data_ptr<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
three_nn_kernel_launcher(b, n, m, unknown, known, dist2, idx, stream);
}
void three_interpolate_wrapper(int b, int c, int m, int n,
at::Tensor points_tensor,
at::Tensor idx_tensor,
at::Tensor weight_tensor,
at::Tensor out_tensor) {
const float *points = points_tensor.data<float>();
const float *weight = weight_tensor.data<float>();
float *out = out_tensor.data<float>();
const int *idx = idx_tensor.data<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
three_interpolate_kernel_launcher(b, c, m, n, points, idx, weight, out, stream);
at::Tensor points_tensor, at::Tensor idx_tensor,
at::Tensor weight_tensor,
at::Tensor out_tensor) {
const float *points = points_tensor.data_ptr<float>();
const float *weight = weight_tensor.data_ptr<float>();
float *out = out_tensor.data_ptr<float>();
const int *idx = idx_tensor.data_ptr<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
three_interpolate_kernel_launcher(b, c, m, n, points, idx, weight, out,
stream);
}
void three_interpolate_grad_wrapper(int b, int c, int n, int m,
at::Tensor grad_out_tensor,
at::Tensor idx_tensor,
at::Tensor weight_tensor,
at::Tensor grad_points_tensor) {
const float *grad_out = grad_out_tensor.data<float>();
const float *weight = weight_tensor.data<float>();
float *grad_points = grad_points_tensor.data<float>();
const int *idx = idx_tensor.data<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
three_interpolate_grad_kernel_launcher(b, c, n, m, grad_out, idx, weight, grad_points, stream);
at::Tensor grad_out_tensor,
at::Tensor idx_tensor,
at::Tensor weight_tensor,
at::Tensor grad_points_tensor) {
const float *grad_out = grad_out_tensor.data_ptr<float>();
const float *weight = weight_tensor.data_ptr<float>();
float *grad_points = grad_points_tensor.data_ptr<float>();
const int *idx = idx_tensor.data_ptr<int>();
cudaStream_t stream = THCState_getCurrentStream(state);
three_interpolate_grad_kernel_launcher(b, c, n, m, grad_out, idx, weight,
grad_points, stream);
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("three_nn_wrapper", &three_nn_wrapper, "three_nn_wrapper");
m.def("three_interpolate_wrapper", &three_interpolate_wrapper, "three_interpolate_wrapper");
m.def("three_interpolate_grad_wrapper", &three_interpolate_grad_wrapper, "three_interpolate_grad_wrapper");
m.def("three_nn_wrapper", &three_nn_wrapper, "three_nn_wrapper");
m.def("three_interpolate_wrapper", &three_interpolate_wrapper,
"three_interpolate_wrapper");
m.def("three_interpolate_grad_wrapper", &three_interpolate_grad_wrapper,
"three_interpolate_grad_wrapper");
}
mmdet3d/ops/interpolate/src/three_interpolate_cuda.cu
View file @ d1b9ae40
......@@ -3,91 +3,103 @@
#include <stdlib.h>
#define THREADS_PER_BLOCK 256
#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
__global__ void three_interpolate_kernel(int b, int c, int m, int n, const float *__restrict__ points,
const int *__restrict__ idx, const float *__restrict__ weight, float *__restrict__ out) {
// points: (B, C, M)
// idx: (B, N, 3)
// weight: (B, N, 3)
// output:
// out: (B, C, N)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= n) return;
weight += bs_idx * n * 3 + pt_idx * 3;
points += bs_idx * c * m + c_idx * m;
idx += bs_idx * n * 3 + pt_idx * 3;
out += bs_idx * c * n + c_idx * n;
out[pt_idx] = weight[0] * points[idx[0]] + weight[1] * points[idx[1]] + weight[2] * points[idx[2]];
#define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
__global__ void three_interpolate_kernel(int b, int c, int m, int n,
const float *__restrict__ points,
const int *__restrict__ idx,
const float *__restrict__ weight,
float *__restrict__ out) {
// points: (B, C, M)
// idx: (B, N, 3)
// weight: (B, N, 3)
// output:
// out: (B, C, N)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= n) return;
weight += bs_idx * n * 3 + pt_idx * 3;
points += bs_idx * c * m + c_idx * m;
idx += bs_idx * n * 3 + pt_idx * 3;
out += bs_idx * c * n + c_idx * n;
out[pt_idx] = weight[0] * points[idx[0]] + weight[1] * points[idx[1]] +
weight[2] * points[idx[2]];
}
void three_interpolate_kernel_launcher(int b, int c, int m, int n,
const float *points, const int *idx, const float *weight, float *out, cudaStream_t stream) {
// points: (B, C, M)
// idx: (B, N, 3)
// weight: (B, N, 3)
// output:
// out: (B, C, N)
cudaError_t err;
dim3 blocks(DIVUP(n, THREADS_PER_BLOCK), c, b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
three_interpolate_kernel<<<blocks, threads, 0, stream>>>(b, c, m, n, points, idx, weight, out);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
const float *points, const int *idx,
const float *weight, float *out,
cudaStream_t stream) {
// points: (B, C, M)
// idx: (B, N, 3)
// weight: (B, N, 3)
// output:
// out: (B, C, N)
cudaError_t err;
dim3 blocks(DIVUP(n, THREADS_PER_BLOCK), c,
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
three_interpolate_kernel<<<blocks, threads, 0, stream>>>(b, c, m, n, points,
idx, weight, out);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
__global__ void three_interpolate_grad_kernel(int b, int c, int n, int m, const float *__restrict__ grad_out,
const int *__restrict__ idx, const float *__restrict__ weight, float *__restrict__ grad_points) {
// grad_out: (B, C, N)
// weight: (B, N, 3)
// output:
// grad_points: (B, C, M)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= n) return;
grad_out += bs_idx * c * n + c_idx * n + pt_idx;
weight += bs_idx * n * 3 + pt_idx * 3;
grad_points += bs_idx * c * m + c_idx * m;
idx += bs_idx * n * 3 + pt_idx * 3;
atomicAdd(grad_points + idx[0], grad_out[0] * weight[0]);
atomicAdd(grad_points + idx[1], grad_out[0] * weight[1]);
atomicAdd(grad_points + idx[2], grad_out[0] * weight[2]);
__global__ void three_interpolate_grad_kernel(
int b, int c, int n, int m, const float *__restrict__ grad_out,
const int *__restrict__ idx, const float *__restrict__ weight,
float *__restrict__ grad_points) {
// grad_out: (B, C, N)
// weight: (B, N, 3)
// output:
// grad_points: (B, C, M)
int bs_idx = blockIdx.z;
int c_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || c_idx >= c || pt_idx >= n) return;
grad_out += bs_idx * c * n + c_idx * n + pt_idx;
weight += bs_idx * n * 3 + pt_idx * 3;
grad_points += bs_idx * c * m + c_idx * m;
idx += bs_idx * n * 3 + pt_idx * 3;
atomicAdd(grad_points + idx[0], grad_out[0] * weight[0]);
atomicAdd(grad_points + idx[1], grad_out[0] * weight[1]);
atomicAdd(grad_points + idx[2], grad_out[0] * weight[2]);
}
void three_interpolate_grad_kernel_launcher(int b, int c, int n, int m, const float *grad_out,
const int *idx, const float *weight, float *grad_points, cudaStream_t stream) {
// grad_out: (B, C, N)
// weight: (B, N, 3)
// output:
// grad_points: (B, C, M)
cudaError_t err;
dim3 blocks(DIVUP(n, THREADS_PER_BLOCK), c, b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
three_interpolate_grad_kernel<<<blocks, threads, 0, stream>>>(b, c, n, m, grad_out, idx, weight, grad_points);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
void three_interpolate_grad_kernel_launcher(int b, int c, int n, int m,
const float *grad_out,
const int *idx, const float *weight,
float *grad_points,
cudaStream_t stream) {
// grad_out: (B, C, N)
// weight: (B, N, 3)
// output:
// grad_points: (B, C, M)
cudaError_t err;
dim3 blocks(DIVUP(n, THREADS_PER_BLOCK), c,
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
three_interpolate_grad_kernel<<<blocks, threads, 0, stream>>>(
b, c, n, m, grad_out, idx, weight, grad_points);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
mmdet3d/ops/interpolate/src/three_nn_cuda.cu
View file @ d1b9ae40
......@@ -3,72 +3,84 @@
#include <stdlib.h>
#define THREADS_PER_BLOCK 256
#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
#define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
__global__ void three_nn_kernel(int b, int n, int m,
const float *__restrict__ unknown,
const float *__restrict__ known,
float *__restrict__ dist2,
int *__restrict__ idx) {
// unknown: (B, N, 3)
// known: (B, M, 3)
// output:
// dist2: (B, N, 3)
// idx: (B, N, 3)
__global__ void three_nn_kernel(int b, int n, int m, const float *__restrict__ unknown,
const float *__restrict__ known, float *__restrict__ dist2, int *__restrict__ idx) {
// unknown: (B, N, 3)
// known: (B, M, 3)
// output:
// dist2: (B, N, 3)
// idx: (B, N, 3)
int bs_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || pt_idx >= n) return;
int bs_idx = blockIdx.y;
int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
if (bs_idx >= b || pt_idx >= n) return;
unknown += bs_idx * n * 3 + pt_idx * 3;
known += bs_idx * m * 3;
dist2 += bs_idx * n * 3 + pt_idx * 3;
idx += bs_idx * n * 3 + pt_idx * 3;
unknown += bs_idx * n * 3 + pt_idx * 3;
known += bs_idx * m * 3;
dist2 += bs_idx * n * 3 + pt_idx * 3;
idx += bs_idx * n * 3 + pt_idx * 3;
float ux = unknown[0];
float uy = unknown[1];
float uz = unknown[2];
float ux = unknown[0];
float uy = unknown[1];
float uz = unknown[2];
double best1 = 1e40, best2 = 1e40, best3 = 1e40;
int besti1 = 0, besti2 = 0, besti3 = 0;
for (int k = 0; k < m; ++k) {
float x = known[k * 3 + 0];
float y = known[k * 3 + 1];
float z = known[k * 3 + 2];
float d = (ux - x) * (ux - x) + (uy - y) * (uy - y) + (uz - z) * (uz - z);
if (d < best1) {
best3 = best2; besti3 = besti2;
best2 = best1; besti2 = besti1;
best1 = d; besti1 = k;
}
else if (d < best2) {
best3 = best2; besti3 = besti2;
best2 = d; besti2 = k;
}
else if (d < best3) {
best3 = d; besti3 = k;
}
double best1 = 1e40, best2 = 1e40, best3 = 1e40;
int besti1 = 0, besti2 = 0, besti3 = 0;
for (int k = 0; k < m; ++k) {
float x = known[k * 3 + 0];
float y = known[k * 3 + 1];
float z = known[k * 3 + 2];
float d = (ux - x) * (ux - x) + (uy - y) * (uy - y) + (uz - z) * (uz - z);
if (d < best1) {
best3 = best2;
besti3 = besti2;
best2 = best1;
besti2 = besti1;
best1 = d;
besti1 = k;
} else if (d < best2) {
best3 = best2;
besti3 = besti2;
best2 = d;
besti2 = k;
} else if (d < best3) {
best3 = d;
besti3 = k;
}
dist2[0] = best1; dist2[1] = best2; dist2[2] = best3;
idx[0] = besti1; idx[1] = besti2; idx[2] = besti3;
}
dist2[0] = best1;
dist2[1] = best2;
dist2[2] = best3;
idx[0] = besti1;
idx[1] = besti2;
idx[2] = besti3;
}
void three_nn_kernel_launcher(int b, int n, int m, const float *unknown,
const float *known, float *dist2, int *idx, cudaStream_t stream) {
// unknown: (B, N, 3)
// known: (B, M, 3)
// output:
// dist2: (B, N, 3)
// idx: (B, N, 3)
const float *known, float *dist2, int *idx,
cudaStream_t stream) {
// unknown: (B, N, 3)
// known: (B, M, 3)
// output:
// dist2: (B, N, 3)
// idx: (B, N, 3)
cudaError_t err;
dim3 blocks(DIVUP(n, THREADS_PER_BLOCK), b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
cudaError_t err;
dim3 blocks(DIVUP(n, THREADS_PER_BLOCK),
b); // blockIdx.x(col), blockIdx.y(row)
dim3 threads(THREADS_PER_BLOCK);
three_nn_kernel<<<blocks, threads, 0, stream>>>(b, n, m, unknown, known, dist2, idx);
three_nn_kernel<<<blocks, threads, 0, stream>>>(b, n, m, unknown, known,
dist2, idx);
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
exit(-1);
}
}
mmdet3d/ops/iou3d/src/iou3d_kernel.cu
View file @ d1b9ae40
This diff is collapsed.
mmdet3d/ops/roiaware_pool3d/src/points_in_boxes_cuda.cu
View file @ d1b9ae40
......@@ -78,9 +78,9 @@ __global__ void points_in_boxes_kernel(int batch_size, int boxes_num,
}
__global__ void points_in_boxes_batch_kernel(int batch_size, int boxes_num,
int pts_num, const float *boxes,
const float *pts,
int *box_idx_of_points) {
int pts_num, const float *boxes,
const float *pts,
int *box_idx_of_points) {
// params boxes: (B, N, 7) [x, y, z, w, l, h, rz] in LiDAR coordinate, z is
// the bottom center, each box DO NOT overlaps params pts: (B, npoints, 3) [x,
// y, z] in LiDAR coordinate params boxes_idx_of_points: (B, npoints), default
......@@ -131,17 +131,17 @@ void points_in_boxes_launcher(int batch_size, int boxes_num, int pts_num,
}
void points_in_boxes_batch_launcher(int batch_size, int boxes_num, int pts_num,
const float *boxes, const float *pts,
int *box_idx_of_points) {
const float *boxes, const float *pts,
int *box_idx_of_points) {
// params boxes: (B, N, 7) [x, y, z, w, l, h, rz] in LiDAR coordinate, z is
// the bottom center, each box params pts: (B, npoints, 3) [x, y, z] in
//LiDAR coordinate params boxes_idx_of_points: (B, npoints), default -1
// LiDAR coordinate params boxes_idx_of_points: (B, npoints), default -1
cudaError_t err;
dim3 blocks(DIVUP(pts_num, THREADS_PER_BLOCK), batch_size);
dim3 threads(THREADS_PER_BLOCK);
points_in_boxes_batch_kernel<<<blocks, threads>>>(batch_size, boxes_num, pts_num,
boxes, pts, box_idx_of_points);
points_in_boxes_batch_kernel<<<blocks, threads>>>(
batch_size, boxes_num, pts_num, boxes, pts, box_idx_of_points);
err = cudaGetLastError();
if (cudaSuccess != err) {
......@@ -180,7 +180,7 @@ int points_in_boxes_gpu(at::Tensor boxes_tensor, at::Tensor pts_tensor,
}
int points_in_boxes_batch(at::Tensor boxes_tensor, at::Tensor pts_tensor,
at::Tensor box_idx_of_points_tensor) {
at::Tensor box_idx_of_points_tensor) {
// params boxes: (B, N, 7) [x, y, z, w, l, h, rz] in LiDAR coordinate, z is
// the bottom center. params pts: (B, npoints, 3) [x, y, z] in LiDAR
// coordinate params boxes_idx_of_points: (B, npoints), default -1
......
mmdet3d/ops/spconv/include/paramsgrid.h
View file @ d1b9ae40
......@@ -18,13 +18,19 @@
#include <vector>
namespace detail {
template <class T> int getTotalSize(std::vector<T> arg) { return arg.size(); }
template <class T>
int getTotalSize(std::vector<T> arg) {
return arg.size();
}
template <class T, class... TArgs>
int getTotalSize(std::vector<T> arg, std::vector<TArgs>... args) {
return arg.size() * getTotalSize(args...);
}
template <typename T> int getSize(std::vector<T> arg) { return arg.size(); }
template <typename T>
int getSize(std::vector<T> arg) {
return arg.size();
}
template <int Idx, class TT, class T>
void assigner(TT &src, std::vector<int> counter, std::vector<T> &arg) {
......@@ -37,7 +43,7 @@ void assigner(TT &src, std::vector<int> counter, std::vector<T> &arg,
std::get<Idx>(src) = arg[counter[Idx]];
assigner<Idx + 1>(src, counter, args...);
}
} // namespace detail
} // namespace detail
template <class... TArgs>
std::vector<std::tuple<TArgs...>> paramsGrid(std::vector<TArgs>... args) {
int length = detail::getTotalSize(args...);
......
mmdet3d/ops/spconv/include/prettyprint.h
View file @ d1b9ae40
This diff is collapsed.
mmdet3d/ops/spconv/include/spconv/box_iou.h
View file @ d1b9ae40
......@@ -12,15 +12,15 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef BOX_IOU_H
#define BOX_IOU_H
#include <pybind11/pybind11.h>
// must include pybind11/eigen.h if using eigen matrix as arguments.
#include <pybind11/numpy.h>
#include <algorithm>
#include <boost/geometry.hpp>
#include <pybind11/numpy.h>
namespace spconv {
// #include "voxelnet/core/cc/pybind11_helper.h"
......@@ -40,9 +40,10 @@ inline py::array_t<DType> zeros(std::vector<long int> shape) {
}
template <typename DType>
py::array_t<DType>
rbbox_iou(py::array_t<DType> box_corners, py::array_t<DType> qbox_corners,
py::array_t<DType> standup_iou, DType standup_thresh) {
py::array_t<DType> rbbox_iou(py::array_t<DType> box_corners,
py::array_t<DType> qbox_corners,
py::array_t<DType> standup_iou,
DType standup_thresh) {
namespace bg = boost::geometry;
typedef bg::model::point<DType, 2, bg::cs::cartesian> point_t;
typedef bg::model::polygon<point_t> polygon_t;
......@@ -61,8 +62,7 @@ rbbox_iou(py::array_t<DType> box_corners, py::array_t<DType> qbox_corners,
}
for (int k = 0; k < K; ++k) {
for (int n = 0; n < N; ++n) {
if (standup_iou_r(n, k) <= standup_thresh)
continue;
if (standup_iou_r(n, k) <= standup_thresh) continue;
bg::append(poly, point_t(box_corners_r(n, 0, 0), box_corners_r(n, 0, 1)));
bg::append(poly, point_t(box_corners_r(n, 1, 0), box_corners_r(n, 1, 1)));
bg::append(poly, point_t(box_corners_r(n, 2, 0), box_corners_r(n, 2, 1)));
......@@ -99,9 +99,10 @@ rbbox_iou(py::array_t<DType> box_corners, py::array_t<DType> qbox_corners,
}
template <typename DType>
py::array_t<DType>
rbbox_intersection(py::array_t<DType> box_corners, py::array_t<DType> qbox_corners,
py::array_t<DType> standup_iou, DType standup_thresh) {
py::array_t<DType> rbbox_intersection(py::array_t<DType> box_corners,
py::array_t<DType> qbox_corners,
py::array_t<DType> standup_iou,
DType standup_thresh) {
namespace bg = boost::geometry;
typedef bg::model::point<DType, 2, bg::cs::cartesian> point_t;
typedef bg::model::polygon<point_t> polygon_t;
......@@ -120,8 +121,7 @@ rbbox_intersection(py::array_t<DType> box_corners, py::array_t<DType> qbox_corne
}
for (int k = 0; k < K; ++k) {
for (int n = 0; n < N; ++n) {
if (standup_iou_r(n, k) <= standup_thresh)
continue;
if (standup_iou_r(n, k) <= standup_thresh) continue;
bg::append(poly, point_t(box_corners_r(n, 0, 0), box_corners_r(n, 0, 1)));
bg::append(poly, point_t(box_corners_r(n, 1, 0), box_corners_r(n, 1, 1)));
bg::append(poly, point_t(box_corners_r(n, 2, 0), box_corners_r(n, 2, 1)));
......@@ -152,6 +152,5 @@ rbbox_intersection(py::array_t<DType> box_corners, py::array_t<DType> qbox_corne
return overlaps;
}
} // namespace spconv
} // namespace spconv
#endif
mmdet3d/ops/spconv/include/spconv/geometry.h
View file @ d1b9ae40
......@@ -15,9 +15,10 @@
#ifndef SPCONV_GEOMETRY_H_
#define SPCONV_GEOMETRY_H_
#include <tensorview/tensorview.h>
#include <iostream>
#include <limits>
#include <tensorview/tensorview.h>
namespace spconv {
template <typename Index, unsigned NDim>
......@@ -70,8 +71,7 @@ TV_HOST_DEVICE Index getValidOutPos(const Index *input_pos,
}
out[pointCounter * (NDim + 1) + NDim] = offset;
if (valid)
++pointCounter;
if (valid) ++pointCounter;
counter[NDim - 1] += 1;
#pragma unroll
for (int c = NDim - 1; c >= 0; --c) {
......@@ -128,8 +128,7 @@ TV_HOST_DEVICE Index getValidOutPosTranspose(
m *= kernelSize[j];
}
out[pointCounter * (NDim + 1) + NDim] = offset;
if (valid)
++pointCounter;
if (valid) ++pointCounter;
counter[NDim - 1] += 1;
#pragma unroll
for (int c = NDim - 1; c >= 0; --c) {
......@@ -167,7 +166,7 @@ Index getIndicePairsConv(tv::TensorView<const Index> indicesIn,
}
Index numValidPoints = 0;
std::vector<Index> validPoints_(kernelVolume * (NDim + 1));
Index* validPoints = validPoints_.data();
Index *validPoints = validPoints_.data();
Index *pointPtr = nullptr;
for (int j = 0; j < numActIn; ++j) {
batchIdx = indicesIn(j, 0);
......@@ -218,7 +217,7 @@ Index getIndicePairsDeConv(tv::TensorView<const Index> indicesIn,
}
Index numValidPoints = 0;
std::vector<Index> validPoints_(kernelVolume * (NDim + 1));
Index* validPoints = validPoints_.data();
Index *validPoints = validPoints_.data();
Index *pointPtr = nullptr;
for (int j = 0; j < numActIn; ++j) {
batchIdx = indicesIn(j, 0);
......@@ -252,7 +251,8 @@ Index getIndicePairsSubM(tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indiceNum,
const Index *const kernelSize,
const Index *const stride, const Index *const padding,
const Index *dilation, const Index *const outSpatialShape) {
const Index *dilation,
const Index *const outSpatialShape) {
Index numAct = 0;
auto numActIn = indicesIn.dim(0);
Index batchIdx = 0;
......@@ -269,7 +269,7 @@ Index getIndicePairsSubM(tv::TensorView<const Index> indicesIn,
Index numValidPoints = 0;
// Index validPoints[kernelVolume * (NDim + 1)];
std::vector<Index> validPoints_(kernelVolume * (NDim + 1));
Index* validPoints = validPoints_.data();
Index *validPoints = validPoints_.data();
Index *pointPtr = nullptr;
Index index = 0;
for (int j = 0; j < numActIn; ++j) {
......@@ -296,6 +296,6 @@ Index getIndicePairsSubM(tv::TensorView<const Index> indicesIn,
return numActIn;
}
} // namespace spconv
} // namespace spconv
#endif
mmdet3d/ops/spconv/include/spconv/indice.cu.h
View file @ d1b9ae40
......@@ -14,9 +14,9 @@
#ifndef INDICE_CU_H_
#define INDICE_CU_H_
#include <tensorview/tensorview.h>
#include <tensorview/helper_kernel.cu.h>
#include <spconv/geometry.h>
#include <tensorview/helper_kernel.cu.h>
#include <tensorview/tensorview.h>
namespace spconv {
template <typename Index, typename IndexGrid, unsigned NDim,
......@@ -115,7 +115,6 @@ __global__ void assignGridAndIndiceOutKernel(
int numAct, tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> outSpatialShape, int batchSize) {
Index index;
auto indicesOutPtr = indicesOut.data();
for (int ix : tv::KernelLoopX<int>(numAct)) {
......@@ -128,13 +127,11 @@ __global__ void assignGridAndIndiceOutKernel(
}
template <typename Index, typename IndexGrid, unsigned NDim>
__global__ void
assignIndicePairsKernel(tv::TensorView<Index> indicesOut,
tv::TensorView<IndexGrid> gridsOut, int numActIn,
tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> outSpatialShape) {
__global__ void assignIndicePairsKernel(
tv::TensorView<Index> indicesOut, tv::TensorView<IndexGrid> gridsOut,
int numActIn, tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> outSpatialShape) {
Index index;
int kernelVolume = indicePairs.dim(0);
for (int ix : tv::KernelLoopX<int>(numActIn)) {
......@@ -148,10 +145,9 @@ assignIndicePairsKernel(tv::TensorView<Index> indicesOut,
}
template <typename Index, typename IndexGrid, unsigned NDim>
__global__ void
prepareSubMGridKernel(tv::TensorView<const Index> indicesIn,
tv::TensorView<IndexGrid> gridsOut,
const tv::SimpleVector<Index, NDim> outSpatialShape) {
__global__ void prepareSubMGridKernel(
tv::TensorView<const Index> indicesIn, tv::TensorView<IndexGrid> gridsOut,
const tv::SimpleVector<Index, NDim> outSpatialShape) {
auto numActIn = indicesIn.dim(0);
Index spatialVolume = 1;
#pragma unroll
......@@ -216,10 +212,9 @@ __global__ void resetGridKernel(const Index *indicePairUnique,
}
template <typename Index, typename IndexGrid, unsigned NDim>
__global__ void
resetGridSubMKernel(const Index *indices, tv::TensorView<IndexGrid> gridsOut,
const tv::SimpleVector<Index, NDim> outSpatialShape,
int numAct) {
__global__ void resetGridSubMKernel(
const Index *indices, tv::TensorView<IndexGrid> gridsOut,
const tv::SimpleVector<Index, NDim> outSpatialShape, int numAct) {
int outSpatialShapeReg[NDim];
for (int i = 0; i < NDim; ++i) {
outSpatialShapeReg[i] = outSpatialShape[i];
......@@ -238,6 +233,6 @@ resetGridSubMKernel(const Index *indices, tv::TensorView<IndexGrid> gridsOut,
}
}
} // namespace spconv
} // namespace spconv
#endif
mmdet3d/ops/spconv/include/spconv/indice.h
View file @ d1b9ae40
......@@ -16,64 +16,65 @@
#define SPARSE_CONV_INDICE_FUNCTOR_H_
#include <tensorview/tensorview.h>
namespace spconv
{
namespace functor
{
namespace spconv {
namespace functor {
template <typename Device, typename Index, typename IndexGrid, unsigned NDim>
struct CreateConvIndicePairFunctorP1
{
Index operator()(
const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indicesOut, tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs, tv::TensorView<Index> indiceNum,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> kernelSize,
const tv::SimpleVector<Index, NDim> stride,
const tv::SimpleVector<Index, NDim> padding,
const tv::SimpleVector<Index, NDim> dilation,
const tv::SimpleVector<Index, NDim> outSpatialShape, bool transpose);
struct CreateConvIndicePairFunctorP1 {
Index operator()(const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indicesOut,
tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indiceNum,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> kernelSize,
const tv::SimpleVector<Index, NDim> stride,
const tv::SimpleVector<Index, NDim> padding,
const tv::SimpleVector<Index, NDim> dilation,
const tv::SimpleVector<Index, NDim> outSpatialShape,
bool transpose);
};
template <typename Device, typename Index, typename IndexGrid, unsigned NDim>
struct CreateConvIndicePairFunctorP2
{
Index operator()(
const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indicesOut, tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs, tv::TensorView<Index> indiceNum,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> outSpatialShape, bool transpose,
bool resetGrid=false);
struct CreateConvIndicePairFunctorP2 {
Index operator()(const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indicesOut,
tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indiceNum,
tv::TensorView<Index> indicePairUnique,
const tv::SimpleVector<Index, NDim> outSpatialShape,
bool transpose, bool resetGrid = false);
};
template <typename Device, typename Index, typename IndexGrid, unsigned NDim>
struct CreateConvIndicePairFunctor
{
Index operator()(
const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indicesOut, tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs, tv::TensorView<Index> indiceNum,
const tv::SimpleVector<Index, NDim> kernelSize,
const tv::SimpleVector<Index, NDim> stride,
const tv::SimpleVector<Index, NDim> padding,
const tv::SimpleVector<Index, NDim> dilation,
const tv::SimpleVector<Index, NDim> outSpatialShape, bool transpose, bool resetGrid=false);
struct CreateConvIndicePairFunctor {
Index operator()(const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<Index> indicesOut,
tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indiceNum,
const tv::SimpleVector<Index, NDim> kernelSize,
const tv::SimpleVector<Index, NDim> stride,
const tv::SimpleVector<Index, NDim> padding,
const tv::SimpleVector<Index, NDim> dilation,
const tv::SimpleVector<Index, NDim> outSpatialShape,
bool transpose, bool resetGrid = false);
};
template <typename Device, typename Index, typename IndexGrid, unsigned NDim>
struct CreateSubMIndicePairFunctor
{
Index operator()(
const Device& d, tv::TensorView<const Index> indicesIn, tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs, tv::TensorView<Index> indiceNum,
const tv::SimpleVector<Index, NDim> kernelSize,
const tv::SimpleVector<Index, NDim> stride,
const tv::SimpleVector<Index, NDim> padding,
const tv::SimpleVector<Index, NDim> dilation,
const tv::SimpleVector<Index, NDim> outSpatialShape, bool transpose, bool resetGrid=false);
struct CreateSubMIndicePairFunctor {
Index operator()(const Device& d, tv::TensorView<const Index> indicesIn,
tv::TensorView<IndexGrid> gridsOut,
tv::TensorView<Index> indicePairs,
tv::TensorView<Index> indiceNum,
const tv::SimpleVector<Index, NDim> kernelSize,
const tv::SimpleVector<Index, NDim> stride,
const tv::SimpleVector<Index, NDim> padding,
const tv::SimpleVector<Index, NDim> dilation,
const tv::SimpleVector<Index, NDim> outSpatialShape,
bool transpose, bool resetGrid = false);
};
} // namespace functor
} // namespace spconv
} // namespace functor
} // namespace spconv
#endif
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment