Skip to content

T
tile_processor_gpu
Commit a51d6a77 authored by Andrey Filippov's avatar Andrey Filippov
Browse files
Options

More editing to make dynamic number of cameras

parent ee0cfc3b
Expand all Show whitespace changes
Inline Side-by-side
Showing with 231 additions and 120 deletions
src/TileProcessor.cuh
View file @ a51d6a77
This diff is collapsed.
src/TileProcessor.h
View file @ a51d6a77
......@@ -104,8 +104,9 @@ extern "C" __global__ void corr2D_combine(
float * gpu_corrs_combo); // combined correlation output (one per tile)
extern "C" __global__ void textures_nonoverlap(
int num_cams, // number of cameras used
struct tp_task * gpu_tasks,
int num_cams, // number of cameras
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats
// struct tp_task * gpu_tasks,
int num_tiles, // number of tiles in task list
// int num_tilesx, // number of tiles in a row
// declare arrays in device code?
......@@ -151,7 +152,8 @@ extern "C" __global__ void imclt_rbg(
extern "C" __global__ void generate_RBGA(
int num_cams, // number of cameras used
// Parameters to generate texture tasks
struct tp_task * gpu_tasks,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
// struct tp_task * gpu_tasks,
int num_tiles, // number of tiles in task list
// declare arrays in device code?
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
......
src/geometry_correction.cu
View file @ a51d6a77
......@@ -44,16 +44,18 @@
// Using NUM_CAMS threads per tile
#define THREADS_PER_BLOCK_GEOM (TILES_PER_BLOCK_GEOM * NUM_CAMS)
#define CYCLES_COPY_GC ((sizeof(struct gc)/sizeof(float) + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
#define CYCLES_COPY_CV ((sizeof(struct corr_vector)/sizeof(float) + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
#define CYCLES_COPY_RBRD ((RBYRDIST_LEN + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
///#define CYCLES_COPY_GC ((sizeof(struct gc)/sizeof(float) + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
///#define CYCLES_COPY_CV ((sizeof(struct corr_vector)/sizeof(float) + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
///#define CYCLES_COPY_RBRD ((RBYRDIST_LEN + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
//#define CYCLES_COPY_ROTS ((NUM_CAMS * 3 *3 + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
#define CYCLES_COPY_ROTS (((sizeof(trot_deriv)/sizeof(float)) + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
//#define CYCLES_COPY_ROTS (((sizeof(trot_deriv)/sizeof(float)) + THREADS_PER_BLOCK_GEOM - 1) / THREADS_PER_BLOCK_GEOM)
#define DBG_CAM 3
__device__ void printGeometryCorrection(struct gc * g);
__device__ void printExtrinsicCorrection(corr_vector * cv);
/**
* Calculate non-distorted radius from distorted using table approximation
* @param rDist distorted radius
......@@ -123,11 +125,28 @@ __constant__ int offset_derivs = 1; // 1..4 // should be nex
__constant__ int offset_matrices = 5; // 5..11
__constant__ int offset_tmp = 12; // 12..15
//inline __device__ int get_task_size_gc(int num_cams);
inline __device__ int get_task_task_gc(int num_tile, float * gpu_ftasks, int num_cams);
inline __device__ int get_task_txy_gc(int num_tile, float * gpu_ftasks, int num_cams);
//inline __device__ int get_task_size_gc(int num_cams){
// return sizeof(struct tp_task)/sizeof(float) - 6 * (NUM_CAMS - num_cams);
//}
inline __device__ int get_task_task_gc(int num_tile, float * gpu_ftasks, int num_cams) {
return *(int *) (gpu_ftasks + get_task_size(num_cams) * num_tile);
}
inline __device__ int get_task_txy_gc(int num_tile, float * gpu_ftasks, int num_cams) {
return *(int *) (gpu_ftasks + get_task_size(num_cams) * num_tile + 1);
}
/**
* Calculate rotation matrices and derivatives by az, tilt, roll, zoom
* NUM_CAMS blocks of 3,3,3 tiles
*/
extern "C" __global__ void calc_rot_deriv(
int num_cams,
struct corr_vector * gpu_correction_vector,
trot_deriv * gpu_rot_deriv)
{
......@@ -282,18 +301,27 @@ extern "C" __global__ void calc_rot_deriv(
extern "C" __global__ void calculate_tiles_offsets(
struct tp_task * gpu_tasks,
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
// struct tp_task * gpu_tasks,
int num_tiles, // number of tiles in task
struct gc * gpu_geometry_correction,
struct corr_vector * gpu_correction_vector,
float * gpu_rByRDist, // length should match RBYRDIST_LEN
trot_deriv * gpu_rot_deriv)
{
dim3 threads_geom(NUM_CAMS,TILES_PER_BLOCK_GEOM, 1);
dim3 grid_geom ((num_tiles+TILES_PER_BLOCK_GEOM-1)/TILES_PER_BLOCK_GEOM, 1, 1);
/// dim3 threads_geom(NUM_CAMS,TILES_PER_BLOCK_GEOM, 1);
/// dim3 grid_geom ((num_tiles+TILES_PER_BLOCK_GEOM-1)/TILES_PER_BLOCK_GEOM, 1, 1);
int tiles_per_block_geom = NUM_THREADS/ num_cams;
dim3 threads_geom(num_cams,tiles_per_block_geom, 1);
dim3 grid_geom ((num_tiles + tiles_per_block_geom - 1)/tiles_per_block_geom, 1, 1);
//#define NUM_THREADS 32
if (threadIdx.x == 0) { // always 1
get_tiles_offsets<<<grid_geom,threads_geom>>> (
gpu_tasks, // struct tp_task * gpu_tasks,
num_cams, // int num_cams,
gpu_ftasks, // float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
// gpu_tasks, // struct tp_task * gpu_tasks,
num_tiles, // int num_tiles, // number of tiles in task list
gpu_geometry_correction, // struct gc * gpu_geometry_correction,
gpu_correction_vector, // struct corr_vector * gpu_correction_vector,
......@@ -313,66 +341,76 @@ extern "C" __global__ void calculate_tiles_offsets(
*/
extern "C" __global__ void get_tiles_offsets(
struct tp_task * gpu_tasks,
int num_cams,
// struct tp_task * gpu_tasks,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task
struct gc * gpu_geometry_correction,
struct corr_vector * gpu_correction_vector,
float * gpu_rByRDist, // length should match RBYRDIST_LEN
trot_deriv * gpu_rot_deriv)
{
int task_size = get_task_size(num_cams);
int task_num = blockIdx.x * blockDim.y + threadIdx.y; // blockIdx.x * TILES_PER_BLOCK_GEOM + threadIdx.y
int thread_xy = blockDim.x * threadIdx.y + threadIdx.x;
int dim_xy = blockDim.x * blockDim.y; // number of parallel threads (<=32)
__shared__ struct gc geometry_correction;
__shared__ float rByRDist [RBYRDIST_LEN];
__shared__ struct corr_vector extrinsic_corr;
__shared__ trot_deriv rot_deriv;
__shared__ float pY_offsets[TILES_PER_BLOCK_GEOM][NUM_CAMS];
/// __shared__ float pY_offsets[TILES_PER_BLOCK_GEOM][NUM_CAMS];
__shared__ float pY_offsets[NUM_THREADS][NUM_CAMS]; // maximal dimensions, actual will be smaller
float pXY[2]; // result to be copied to task
//blockDim.y
// copy data common to all threads
{
int cycles_copy_gc = ((sizeof(struct gc)/sizeof(float) + dim_xy - 1) / dim_xy);
float * gcp_local = (float *) &geometry_correction;
float * gcp_global = (float *) gpu_geometry_correction;
int offset = thread_xy;
for (int i = 0; i < CYCLES_COPY_GC; i++){
for (int i = 0; i < cycles_copy_gc; i++){
if (offset < sizeof(struct gc)/sizeof(float)) {
*(gcp_local + offset) = *(gcp_global + offset);
}
offset += THREADS_PER_BLOCK_GEOM;
offset += dim_xy;
}
}
{
int cycles_copy_cv = ((sizeof(struct corr_vector)/sizeof(float) + dim_xy - 1) / dim_xy);
float * cvp_local = (float *) &extrinsic_corr;
float * cvp_global = (float *) gpu_correction_vector;
int offset = thread_xy;
for (int i = 0; i < CYCLES_COPY_CV; i++){
for (int i = 0; i < cycles_copy_cv; i++){
if (offset < sizeof(struct corr_vector)/sizeof(float)) {
*(cvp_local + offset) = *(cvp_global + offset);
}
offset += THREADS_PER_BLOCK_GEOM;
offset += dim_xy;
}
}
// TODO: maybe it is better to use system memory and not read all table?
{
int cycles_copy_rbrd = (RBYRDIST_LEN + dim_xy - 1) / dim_xy;
float * rByRDistp_local = (float *) rByRDist;
float * rByRDistp_global = (float *) gpu_rByRDist;
int offset = thread_xy;
for (int i = 0; i < CYCLES_COPY_RBRD; i++){
for (int i = 0; i < cycles_copy_rbrd; i++){
if (offset < RBYRDIST_LEN) {
*(rByRDistp_local + offset) = *(rByRDistp_global + offset);
}
offset += THREADS_PER_BLOCK_GEOM;
offset += dim_xy;
}
}
// copy rotational matrices (with their derivatives by azimuth, tilt, roll and zoom - for ERS correction)
{
int cycles_copy_rot = ((sizeof(trot_deriv)/sizeof(float)) + dim_xy - 1) / dim_xy;
float * rots_local = (float *) &rot_deriv;
float * rots_global = (float *) gpu_rot_deriv; // rot_matrices;
int offset = thread_xy;
for (int i = 0; i < CYCLES_COPY_ROTS; i++){
for (int i = 0; i < cycles_copy_rot; i++){
if (offset < sizeof(trot_deriv)/sizeof(float)) {
*(rots_local + offset) = *(rots_global + offset);
}
offset += THREADS_PER_BLOCK_GEOM;
offset += dim_xy;
}
}
__syncthreads();
......@@ -411,8 +449,10 @@ extern "C" __global__ void get_tiles_offsets(
*/
// common code, calculated in parallel
int cxy = gpu_tasks[task_num].txy;
float disparity = gpu_tasks[task_num].target_disparity;
/// int cxy = gpu_tasks[task_num].txy;
/// float disparity = gpu_tasks[task_num].target_disparity;
int cxy = *(int *) (gpu_ftasks + task_size * task_num + 1);
float disparity = * (gpu_ftasks + task_size * task_num + 2);
int tileX = (cxy & 0xffff);
int tileY = (cxy >> 16);
#ifdef DEBUG23
......@@ -638,11 +678,15 @@ extern "C" __global__ void get_tiles_offsets(
}
__syncthreads();// __syncwarp();
#endif // DEBUG21
gpu_tasks[task_num].disp_dist[ncam][0] = disp_dist[0];
gpu_tasks[task_num].disp_dist[ncam][1] = disp_dist[1];
gpu_tasks[task_num].disp_dist[ncam][2] = disp_dist[2];
gpu_tasks[task_num].disp_dist[ncam][3] = disp_dist[3];
/// gpu_tasks[task_num].disp_dist[ncam][0] = disp_dist[0];
/// gpu_tasks[task_num].disp_dist[ncam][1] = disp_dist[1];
/// gpu_tasks[task_num].disp_dist[ncam][2] = disp_dist[2];
/// gpu_tasks[task_num].disp_dist[ncam][3] = disp_dist[3];
float * disp_dist_p = gpu_ftasks + task_size * task_num + 3 + ncam * 4; // ncam = threadIdx.x, so each thread will have different offset
*(disp_dist_p++) = disp_dist[0]; // global memory
*(disp_dist_p++) = disp_dist[1];
*(disp_dist_p++) = disp_dist[2];
*(disp_dist_p++) = disp_dist[3];
// imu = extrinsic_corr.getIMU(i); // currently it is common for all channels
// float imu_rot [3]; // d_tilt/dt (rad/s), d_az/dt, d_roll/dt 13..15
......@@ -697,8 +741,11 @@ extern "C" __global__ void get_tiles_offsets(
}
}
// copy results to global memory pXY, disp_dist
gpu_tasks[task_num].xy[ncam][0] = pXY[0];
gpu_tasks[task_num].xy[ncam][1] = pXY[1];
// gpu_tasks[task_num].xy[ncam][0] = pXY[0];
// gpu_tasks[task_num].xy[ncam][1] = pXY[1];
float * tile_xy_p = gpu_ftasks + task_size * task_num + 3 + num_cams * 4 + ncam * 2; // ncam = threadIdx.x, so each thread will have different offset
*(tile_xy_p++) = pXY[0]; // global memory
*(tile_xy_p++) = pXY[1]; // global memory
}
extern "C" __global__ void calcReverseDistortionTable(
......
src/geometry_correction.h
View file @ a51d6a77
......@@ -41,6 +41,8 @@
#include "tp_defines.h"
#endif
#define get_task_size(x) (sizeof(struct tp_task)/sizeof(float) - 6 * (NUM_CAMS - x))
#define NVRTC_BUG 1
#ifndef M_PI
#define M_PI 3.14159265358979323846 /* pi */
......@@ -60,8 +62,9 @@ struct tp_task {
int txy;
unsigned short sxy[2];
};
float xy[NUM_CAMS][2];
float target_disparity;
float xy[NUM_CAMS][2];
// float target_disparity;
float disp_dist[NUM_CAMS][4]; // calculated with getPortsCoordinates()
};
......@@ -142,7 +145,9 @@ struct gc {
};
#define RAD_COEFF_LEN 7
extern "C" __global__ void get_tiles_offsets(
struct tp_task * gpu_tasks,
int num_cams,
// struct tp_task * gpu_tasks,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task
struct gc * gpu_geometry_correction,
struct corr_vector * gpu_correction_vector,
......@@ -150,7 +155,9 @@ extern "C" __global__ void get_tiles_offsets(
trot_deriv * gpu_rot_deriv);
extern "C" __global__ void calculate_tiles_offsets(
struct tp_task * gpu_tasks,
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
// struct tp_task * gpu_tasks,
int num_tiles, // number of tiles in task
struct gc * gpu_geometry_correction,
struct corr_vector * gpu_correction_vector,
......@@ -160,6 +167,7 @@ extern "C" __global__ void calculate_tiles_offsets(
// uses NUM_CAMS blocks, (3,3,3) threads
extern "C" __global__ void calc_rot_deriv(
int num_cams,
struct corr_vector * gpu_correction_vector,
trot_deriv * gpu_rot_deriv);
......@@ -170,3 +178,4 @@ extern "C" __global__ void calcReverseDistortionTable(
float * rByRDist);
src/test_tp.cu
View file @ a51d6a77
......@@ -715,6 +715,7 @@ int main(int argc, char **argv)
}
calc_rot_deriv<<<grid_rot,threads_rot>>> (
num_cams, // int num_cams,
gpu_correction_vector , // struct corr_vector * gpu_correction_vector,
gpu_rot_deriv); // union trot_deriv * gpu_rot_deriv);
......@@ -821,6 +822,7 @@ int main(int argc, char **argv)
}
/*
get_tiles_offsets<<<grid_geom,threads_geom>>> (
num_cams, // int num_cams,
gpu_tasks, // struct tp_task * gpu_tasks,
tp_task_size, // int num_tiles, // number of tiles in task list
gpu_geometry_correction, // struct gc * gpu_geometry_correction,
......@@ -829,7 +831,9 @@ int main(int argc, char **argv)
gpu_rot_deriv); // union trot_deriv * gpu_rot_deriv);
*/
calculate_tiles_offsets<<<1,1>>> (
gpu_tasks, // struct tp_task * gpu_tasks,
num_cams, // int num_cams,
gpu_ftasks, // float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
// gpu_tasks, // struct tp_task * gpu_tasks,
tp_task_size, // int num_tiles, // number of tiles in task list
gpu_geometry_correction, // struct gc * gpu_geometry_correction,
gpu_correction_vector, // struct corr_vector * gpu_correction_vector,
......@@ -1273,7 +1277,8 @@ int main(int argc, char **argv)
cudaFuncSetAttribute(textures_nonoverlap, cudaFuncAttributeMaxDynamicSharedMemorySize, 65536); // for CC 7.5
textures_nonoverlap<<<1,1>>> (
num_cams, // int num_cams, // number of cameras used
gpu_tasks, // struct tp_task * gpu_tasks,
gpu_ftasks, // float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats
// gpu_tasks, // struct tp_task * gpu_tasks,
tp_task_size, // int num_tiles, // number of tiles in task list
// declare arrays in device code?
gpu_texture_indices, // int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
......@@ -1365,7 +1370,8 @@ int main(int argc, char **argv)
generate_RBGA<<<1,1>>> (
num_cams, // int num_cams, // number of cameras used
// Parameters to generate texture tasks
gpu_tasks, // struct tp_task * gpu_tasks,
gpu_ftasks, // float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
// gpu_tasks, // struct tp_task * gpu_tasks,
tp_task_size, // int num_tiles, // number of tiles in task list
// Does not require initialized gpu_texture_indices to be initialized - just allocated, will generate.
gpu_texture_indices, // int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
......
src/tp_defines.h
View file @ a51d6a77
......@@ -56,7 +56,8 @@
#define CORR_TILES_PER_BLOCK 4
#define CORR_TILES_PER_BLOCK_NORMALIZE 4 // increase to 8?
#define CORR_TILES_PER_BLOCK_COMBINE 4 // increase to 16?
#define TEXTURE_THREADS 32 //
//#define TEXTURE_THREADS 32 //
#define NUM_THREADS 32
#define TEXTURE_THREADS_PER_TILE 8
#define TEXTURE_TILES_PER_BLOCK 1
#define IMCLT_THREADS_PER_TILE 16
......
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