Skip to content

T
tile_processor_gpu
Commit 1bf5e8bf authored by Andrey Filippov's avatar Andrey Filippov
Browse files
Options

Refactoring, working state

parent 13f515b9
Hide whitespace changes
Inline Side-by-side
Showing with 330 additions and 378 deletions
src/TileProcessor.cuh → src/TileProcessor.cu
View file @ 1bf5e8bf
...@@ -36,12 +36,17 @@ ...@@ -36,12 +36,17 @@
* \brief Top level of the Tile Processor for frequency domain * \brief Top level of the Tile Processor for frequency domain
*/ */
// Avoiding includes in jcuda, all source files will be merged // Avoiding includes in jcuda, all source files will be merged
#pragma once //#pragma once
#ifndef JCUDA #ifndef JCUDA
#include "tp_defines.h" #include "tp_defines.h"
#include "dtt8x8.h" #include "dtt8x8.h"
#include "geometry_correction.h" #include "geometry_correction.h"
#include "TileProcessor.h" #include "TileProcessor.h"
#include <cuda_runtime.h>
// #include <helper_cuda.h>
// #include <helper_functions.h>
#endif // #ifndef JCUDA #endif // #ifndef JCUDA
// CUDA fast math is slower! // CUDA fast math is slower!
...@@ -55,67 +60,16 @@ GPU run time =523.451927ms, (direct conversion: 24.080189999999998ms, imclt: 17. ...@@ -55,67 +60,16 @@ GPU run time =523.451927ms, (direct conversion: 24.080189999999998ms, imclt: 17.
//#define TASK_TEXTURE_BITS ((1 << TASK_TEXTURE_N_BIT) | (1 << TASK_TEXTURE_E_BIT) | (1 << TASK_TEXTURE_S_BIT) | (1 << TASK_TEXTURE_W_BIT)) //#define TASK_TEXTURE_BITS ((1 << TASK_TEXTURE_N_BIT) | (1 << TASK_TEXTURE_E_BIT) | (1 << TASK_TEXTURE_S_BIT) | (1 << TASK_TEXTURE_W_BIT))
#define TASK_TEXTURE_BITS ((1 << TASK_TEXT_N_BIT) | (1 << TASK_TEXT_NE_BIT) | (1 << TASK_TEXT_E_BIT) | (1 << TASK_TEXT_SE_BIT)\ /*
| (1 << TASK_TEXT_S_BIT) | (1 << TASK_TEXT_SW_BIT) | (1 << TASK_TEXT_W_BIT) | (1 << TASK_TEXT_NW_BIT))
//#define IMCLT14
//#define NOICLT 1
//#define TEST_IMCLT
//#define SAVE_CLT
// Not enough shared memory to have more threads per block,even just for the result clt tiles
// What to do:
// 1) make single image aberration correction: 1/4 of the result tiles
// With 4 cameras = calculate correlations (9x9), reusing kernel or just clt ones after color reducing, then output them to device memory
//Average run time =1308.124146 ms
//#define TILES_PER_BLOCK 2
//Average run time =12502.638672 - with 2 tiles/block it is longer!
///12129.268555 ms
//Average run time =4704.506348 ms (syncwarp)
//Average run time =4705.612305 ms (syncthreads)
//Average run time =1051.411255 ms
//Average run time =861.866577 ms
//Average run time =850.871277 ms had bugs
//Average run time =857.947632 ms fixed bugs
// Something broke, even w/o LPF: Average run time =1093.115112 ms
// without clt copying to device memory - Average run time =965.342407 ms - still worse
//Average run time =965.880554 ms
// combined tx and ty into a single int : Average run time =871.017944 ms
//Average run time =873.386597 ms (reduced number of registers)
//__umul24 : Average run time =879.125122 ms
// without __umul24 - back to Average run time =871.315552 ms
// Added copying clt to device memory - Average run time =942.071960 ms
// Removed rest of NOICLT : Average run time =943.456177 ms
// Added lpf: Average run time =1046.101318 ms (0.1 sec, 10%) - can be combined with the PSF kernel
//#define USE_UMUL24
////#define TILES_PER_BLOCK 4
//Average run time =5155.922852 ms
//Average run time =1166.388306 ms
//Average run time =988.750977 ms
//#define TILES_PER_BLOCK 8
//Average run time =9656.743164 ms
// Average run time =9422.057617 ms (reducing divergence)
//#define TILES_PER_BLOCK 1
//#define THREADS_PER_TILE 8
//#define IMCLT_THREADS_PER_TILE 16
//#define IMCLT_TILES_PER_BLOCK 4
/// #define KERNELS_STEP (1 << KERNELS_LSTEP)
//#define TILES-X (IMG-WIDTH / DTT_SIZE)
//#define TILES-Y (IMG-HEIGHT / DTT_SIZE)
#define CONVERT_DIRECT_INDEXING_THREADS_LOG2 5 #define CONVERT_DIRECT_INDEXING_THREADS_LOG2 5
#define CONVERT_DIRECT_INDEXING_THREADS (1 << CONVERT_DIRECT_INDEXING_THREADS_LOG2) // 32 #define CONVERT_DIRECT_INDEXING_THREADS (1 << CONVERT_DIRECT_INDEXING_THREADS_LOG2) // 32
*/
// Make TILES-YA >= TILES-X and a multiple of 4 // Make TILES-YA >= TILES-X and a multiple of 4
//#define TILES-YA ((TILES-Y +3) & (~3)) //#define TILES-YA ((TILES-Y +3) & (~3))
// increase row length by 1 so vertical passes will use different ports // increase row length by 1 so vertical passes will use different ports
#define MCLT_UNION_LEN (DTT_SIZE2 * (DTT_SIZE2 + 2)) //#define MCLT_UNION_LEN (DTT_SIZE2 * (DTT_SIZE2 + 2))
/*
struct CltExtra{ struct CltExtra{
float data_x; // kernel data is relative to this displacement X (0.5 pixel increments) float data_x; // kernel data is relative to this displacement X (0.5 pixel increments)
float data_y; // kernel data is relative to this displacement Y (0.5 pixel increments) float data_y; // kernel data is relative to this displacement Y (0.5 pixel increments)
...@@ -126,7 +80,7 @@ struct CltExtra{ ...@@ -126,7 +80,7 @@ struct CltExtra{
float dyc_dx; float dyc_dx;
float dyc_dy; float dyc_dy;
}; };
*/
/* /*
Python code to generate constant coefficients: Python code to generate constant coefficients:
def setup_hwindow(n=8, l=4): def setup_hwindow(n=8, l=4):
...@@ -735,48 +689,11 @@ __device__ void imclt_plane( // not implemented, not used ...@@ -735,48 +689,11 @@ __device__ void imclt_plane( // not implemented, not used
float * gpu_rbg, // WIDTH, HEIGHT float * gpu_rbg, // WIDTH, HEIGHT
const size_t dstride); // in floats (pixels) const size_t dstride); // in floats (pixels)
extern "C" __global__ void clear_texture_list(
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
int width, // <= TILES-X, use for faster processing of LWIR images
int height); // <= TILES-Y, use for faster processing of LWIR images
extern "C" __global__ void mark_texture_tiles(
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task list
int width, // number of tiles in a row
int * gpu_texture_indices);// packed tile + bits (now only (1 << 7)
extern "C" __global__ void mark_texture_neighbor_tiles( // TODO: remove __global__?
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task list
int width, // number of tiles in a row
int height, // number of tiles rows
int * gpu_texture_indices, // packed tile + bits (now only (1 << 7)
int * woi); // x,y,width,height of the woi
extern "C" __global__ void gen_texture_list(
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task list
int width, // number of tiles in a row
int height, // number of tiles rows
int * gpu_texture_indices, // packed tile + bits (now only (1 << 7)
int * num_texture_tiles, // number of texture tiles to process
int * woi); // min_x, min_y, max_x, max_y input
extern "C" __global__ void update_woi( extern "C" __global__ void update_woi(
int texture_slices, int texture_slices,
int * woi, // min_x, min_y, max_x, max_y input, not modified, max_x - not used int * woi, // min_x, min_y, max_x, max_y input, not modified, max_x - not used
int * twh); // 2-element in device global memory int * twh); // 2-element in device global memory
extern "C" __global__ void clear_texture_rbga(
int texture_width,
int texture_slice_height,
const size_t texture_rbga_stride, // in floats 8*stride
float * gpu_texture_tiles); // (number of colors +1 + ?)*16*16 rgba texture tiles
extern "C" __global__ void clear_texture_rbga2( // version for CDP2 extern "C" __global__ void clear_texture_rbga2( // version for CDP2
int * twh, // texture_width, // aligned to DTT_SIZE int * twh, // texture_width, // aligned to DTT_SIZE
const size_t texture_rbga_stride, // in floats 8*stride const size_t texture_rbga_stride, // in floats 8*stride
...@@ -818,15 +735,6 @@ __global__ void index_inter_correlate( ...@@ -818,15 +735,6 @@ __global__ void index_inter_correlate(
int * gpu_corr_indices, // array of correlation tasks int * gpu_corr_indices, // array of correlation tasks
int * pnum_corr_tiles); // pointer to the length of correlation tasks array int * pnum_corr_tiles); // pointer to the length of correlation tasks array
extern "C" __global__ void create_nonoverlap_list(
int num_cams,
float * gpu_ftasks , // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task
int width, // number of tiles in a row
int * nonoverlap_list, // pointer to the calculated number of non-zero tiles
int * pnonoverlap_length); // indices to gpu_tasks // should be initialized to zero
__global__ void convert_correct_tiles( __global__ void convert_correct_tiles(
int num_cams, // actual number of cameras int num_cams, // actual number of cameras
int num_colors, // actual number of colors: 3 for RGB, 1 for LWIR/mono int num_colors, // actual number of colors: 3 for RGB, 1 for LWIR/mono
...@@ -906,34 +814,6 @@ extern "C" __global__ void corr2D_combine_inner( ...@@ -906,34 +814,6 @@ extern "C" __global__ void corr2D_combine_inner(
const size_t corr_stride_combo, // (in floats) stride for the output TD correlations (same as input) const size_t corr_stride_combo, // (in floats) stride for the output TD correlations (same as input)
float * gpu_corrs_combo); // combined correlation output (one per tile) float * gpu_corrs_combo); // combined correlation output (one per tile)
extern "C" __global__ void textures_accumulate( // (8,4,1) (N,1,1)
int num_cams, // number of cameras used
int * woi, // x, y, width,height
float ** gpu_clt, // [num_cams] ->[TILES-Y][TILES-X][colors][DTT_SIZE*DTT_SIZE]
/// size_t num_texture_tiles, // number of texture tiles to process
int * pnum_texture_tiles, // pointer to a number of texture tiles to process
int gpu_texture_indices_offset,// add to gpu_texture_indices
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
// TODO: use geometry_correction rXY !
struct gc * gpu_geometry_correction,
int colors, // number of colors (3/1)
int is_lwir, // do not perform shot correction
float min_shot, // 10.0
float scale_shot, // 3.0
float diff_sigma, // pixel value/pixel change
float diff_threshold, // pixel value/pixel change
float min_agree, // minimal number of channels to agree on a point (real number to work with fuzzy averages)
const float weights[3], // scale for R,B,G
int dust_remove, // Do not reduce average weight when only one image differs much from the average
int keep_weights, // return channel weights after A in RGBA (was removed) (should be 0 if gpu_texture_rbg)?
// combining both non-overlap and overlap (each calculated if pointer is not null )
size_t texture_rbg_stride, // in floats
float * gpu_texture_rbg, // (number of colors +1 + ?)*16*16 rgba texture tiles
size_t texture_stride, // in floats (now 256*4 = 1024)
float * gpu_texture_tiles, // (number of colors +1 + ?)*16*16 rgba texture tiles
int linescan_order, // if !=0 then output gpu_diff_rgb_combo in linescan order, else - in gpu_texture_indices order
float * gpu_diff_rgb_combo, //) // diff[num_cams], R[num_cams], B[num_cams],G[num_cams]
int tilesx);
__device__ int get_textures_shared_size( // in bytes __device__ int get_textures_shared_size( // in bytes
int num_cams, // actual number of cameras int num_cams, // actual number of cameras
......
src/TileProcessor.h
View file @ 1bf5e8bf
...@@ -37,10 +37,34 @@ ...@@ -37,10 +37,34 @@
*/ */
#pragma once #pragma once
#ifndef TILE_PROCESSOR_H_
#define TILE_PROCESSOR_H_
#ifndef NUM_CAMS #ifndef NUM_CAMS
#include "tp_defines.h" #include "tp_defines.h"
#endif #endif
#define TASK_TEXTURE_BITS ((1 << TASK_TEXT_N_BIT) | (1 << TASK_TEXT_NE_BIT) | (1 << TASK_TEXT_E_BIT) | (1 << TASK_TEXT_SE_BIT)\
| (1 << TASK_TEXT_S_BIT) | (1 << TASK_TEXT_SW_BIT) | (1 << TASK_TEXT_W_BIT) | (1 << TASK_TEXT_NW_BIT))
#define CONVERT_DIRECT_INDEXING_THREADS_LOG2 5
#define CONVERT_DIRECT_INDEXING_THREADS (1 << CONVERT_DIRECT_INDEXING_THREADS_LOG2) // 32
#define MCLT_UNION_LEN (DTT_SIZE2 * (DTT_SIZE2 + 2))
struct CltExtra{
float data_x; // kernel data is relative to this displacement X (0.5 pixel increments)
float data_y; // kernel data is relative to this displacement Y (0.5 pixel increments)
float center_x; // actual center X (use to find derivatives)
float center_y; // actual center X (use to find derivatives)
float dxc_dx; // add this to data_x per each pixel X-shift relative to the kernel center location
float dxc_dy; // same per each Y-shift pixel
float dyc_dx;
float dyc_dy;
};
extern "C" __global__ void convert_direct( // called with a single block, single thread extern "C" __global__ void convert_direct( // called with a single block, single thread
// struct CltExtra ** gpu_kernel_offsets, // [NUM_CAMS], // changed for jcuda to avoid struct parameters // struct CltExtra ** gpu_kernel_offsets, // [NUM_CAMS], // changed for jcuda to avoid struct parameters
...@@ -219,4 +243,81 @@ extern "C" __global__ void generate_RBGA( ...@@ -219,4 +243,81 @@ extern "C" __global__ void generate_RBGA(
float * gpu_texture_tiles, // (number of colors +1 + ?)*16*16 rgba texture tiles float * gpu_texture_tiles, // (number of colors +1 + ?)*16*16 rgba texture tiles
int * twh); int * twh);
extern "C" __global__ void textures_accumulate( // (8,4,1) (N,1,1)
int num_cams, // number of cameras used
int * woi, // x, y, width,height
float ** gpu_clt, // [num_cams] ->[TILES-Y][TILES-X][colors][DTT_SIZE*DTT_SIZE]
/// size_t num_texture_tiles, // number of texture tiles to process
int * pnum_texture_tiles, // pointer to a number of texture tiles to process
int gpu_texture_indices_offset,// add to gpu_texture_indices
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
// TODO: use geometry_correction rXY !
struct gc * gpu_geometry_correction,
int colors, // number of colors (3/1)
int is_lwir, // do not perform shot correction
float min_shot, // 10.0
float scale_shot, // 3.0
float diff_sigma, // pixel value/pixel change
float diff_threshold, // pixel value/pixel change
float min_agree, // minimal number of channels to agree on a point (real number to work with fuzzy averages)
const float weights[3], // scale for R,B,G
int dust_remove, // Do not reduce average weight when only one image differs much from the average
int keep_weights, // return channel weights after A in RGBA (was removed) (should be 0 if gpu_texture_rbg)?
// combining both non-overlap and overlap (each calculated if pointer is not null )
size_t texture_rbg_stride, // in floats
float * gpu_texture_rbg, // (number of colors +1 + ?)*16*16 rgba texture tiles
size_t texture_stride, // in floats (now 256*4 = 1024)
float * gpu_texture_tiles, // (number of colors +1 + ?)*16*16 rgba texture tiles
int linescan_order, // if !=0 then output gpu_diff_rgb_combo in linescan order, else - in gpu_texture_indices order
float * gpu_diff_rgb_combo, //) // diff[num_cams], R[num_cams], B[num_cams],G[num_cams]
int tilesx);
extern "C" __global__ void clear_texture_list(
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
int width, // <= TILES-X, use for faster processing of LWIR images
int height); // <= TILES-Y, use for faster processing of LWIR images
extern "C" __global__ void clear_texture_rbga(
int texture_width,
int texture_slice_height,
const size_t texture_rbga_stride, // in floats 8*stride
float * gpu_texture_tiles); // (number of colors +1 + ?)*16*16 rgba texture tiles
extern "C" __global__ void create_nonoverlap_list(
int num_cams,
float * gpu_ftasks , // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task
int width, // number of tiles in a row
int * nonoverlap_list, // pointer to the calculated number of non-zero tiles
int * pnonoverlap_length); // indices to gpu_tasks // should be initialized to zero
extern "C" __global__ void mark_texture_tiles(
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task list
int width, // number of tiles in a row
int * gpu_texture_indices);// packed tile + bits (now only (1 << 7)
extern "C" __global__ void mark_texture_neighbor_tiles( // TODO: remove __global__?
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task list
int width, // number of tiles in a row
int height, // number of tiles rows
int * gpu_texture_indices, // packed tile + bits (now only (1 << 7)
int * woi); // x,y,width,height of the woi
extern "C" __global__ void gen_texture_list(
int num_cams,
float * gpu_ftasks, // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
int num_tiles, // number of tiles in task list
int width, // number of tiles in a row
int height, // number of tiles rows
int * gpu_texture_indices, // packed tile + bits (now only (1 << 7)
int * num_texture_tiles, // number of texture tiles to process
int * woi); // min_x, min_y, max_x, max_y input
#endif
src/geometry_correction.cu
View file @ 1bf5e8bf
...@@ -43,6 +43,11 @@ ...@@ -43,6 +43,11 @@
// #include "TileProcessor.h" // #include "TileProcessor.h"
#include <cuda_runtime.h>
// #include <helper_cuda.h>
// #include <helper_functions.h>
#endif // #ifndef JCUDA #endif // #ifndef JCUDA
#ifndef get_task_size #ifndef get_task_size
......
src/test_tp.cu
View file @ 1bf5e8bf
...@@ -33,9 +33,9 @@ ...@@ -33,9 +33,9 @@
// all of the next 5 were disabled // all of the next 5 were disabled
#define NOCORR #define NOCORR
#define NOCORR_TD #define NOCORR_TD
//#define NOTEXTURES #define NOTEXTURES //
//#define NOTEXTURE_RGBA #define NOTEXTURE_RGBA //
//#define NOTEXTURE_RGBAXXX //#define NOTEXTURE_RGBAXXX //
#define SAVE_CLT #define SAVE_CLT
...@@ -55,11 +55,15 @@ ...@@ -55,11 +55,15 @@
#include <iterator> #include <iterator>
#include <vector> #include <vector>
#include "tp_defines.h" // was not here
#include "dtt8x8.h" #include "dtt8x8.h"
#include "geometry_correction.h" #include "geometry_correction.h"
#include "TileProcessor.cuh" //#include "TileProcessor.cuh"
#include "TileProcessor.h"
#include "tp_utils.h" #include "tp_utils.h"
#include "tp_files.h" #include "tp_files.h"
//#include "tp_paths.cuh"
#include "tp_paths.h" #include "tp_paths.h"
#if TEST_LWIR #if TEST_LWIR
...@@ -2130,6 +2134,7 @@ int main(int argc, char **argv) ...@@ -2130,6 +2134,7 @@ int main(int argc, char **argv)
cudaMemcpyDeviceToHost)); cudaMemcpyDeviceToHost));
#ifndef NSAVE_TEXTURES #ifndef NSAVE_TEXTURES
printf("Import raw, real, little-endian, 18 images 648x520\n");
#ifdef NO_DP #ifdef NO_DP
printf("Writing RBGA texture slices to %s\n", tpPaths.result_textures_rgba_file); printf("Writing RBGA texture slices to %s\n", tpPaths.result_textures_rgba_file);
writeFloatsToFile( writeFloatsToFile(
......
src/tp_files.cu
View file @ 1bf5e8bf
...@@ -17,7 +17,7 @@ ...@@ -17,7 +17,7 @@
#include "tp_files.h" #include "tp_files.h"
int get_file_size(std::string filename) // path to file int get_file_size(const std::string filename) // path to file
{ {
FILE *p_file = NULL; FILE *p_file = NULL;
p_file = fopen(filename.c_str(),"rb"); p_file = fopen(filename.c_str(),"rb");
...@@ -26,6 +26,7 @@ int get_file_size(std::string filename) // path to file ...@@ -26,6 +26,7 @@ int get_file_size(std::string filename) // path to file
fclose(p_file); fclose(p_file);
return size; return size;
} }
int readFloatsFromFile(float * data, // allocated array int readFloatsFromFile(float * data, // allocated array
const char * path) // file path const char * path) // file path
{ {
...@@ -42,6 +43,8 @@ int readFloatsFromFile(float * data, // allocated array ...@@ -42,6 +43,8 @@ int readFloatsFromFile(float * data, // allocated array
return 0; return 0;
} }
float * readAllFloatsFromFile(const char * path, float * readAllFloatsFromFile(const char * path,
int * len_in_floats) // int * len_in_floats) //
{ {
...@@ -55,7 +58,6 @@ float * readAllFloatsFromFile(const char * path, ...@@ -55,7 +58,6 @@ float * readAllFloatsFromFile(const char * path,
printf("---- Bytes read: %d from %s\n", fsize, path); printf("---- Bytes read: %d from %s\n", fsize, path);
* len_in_floats = fsize/sizeof(float); * len_in_floats = fsize/sizeof(float);
return data; return data;
} }
int writeFloatsToFile(float * data, // allocated array int writeFloatsToFile(float * data, // allocated array
...@@ -68,5 +70,46 @@ int writeFloatsToFile(float * data, // allocated array ...@@ -68,5 +70,46 @@ int writeFloatsToFile(float * data, // allocated array
return 0; return 0;
} }
/*
int readFloatsFromFile(float * data, // allocated array
const std::string path) // file path
{
printf("readFloatsFromFile(%s)\n", path);
int fsize = get_file_size(path);
std::ifstream input(path, std::ios::binary );
// copies all data into buffer
std::vector<char> buffer((
std::istreambuf_iterator<char>(input)),
(std::istreambuf_iterator<char>()));
std::copy( buffer.begin(), buffer.end(), (char *) data);
printf("---- Bytes read: %d from %s\n", fsize, path);
return 0;
}
float * readAllFloatsFromFile(const std::string path,
int * len_in_floats) //
{
int fsize = get_file_size(path);
float * data = (float *) malloc(fsize);
std::ifstream input(path, std::ios::binary );
std::vector<char> buffer((
std::istreambuf_iterator<char>(input)),
(std::istreambuf_iterator<char>()));
std::copy( buffer.begin(), buffer.end(), (char *) data);
printf("---- Bytes read: %d from %s\n", fsize, path);
* len_in_floats = fsize/sizeof(float);
return data;
}
int writeFloatsToFile(float * data, // allocated array
int size, // length in elements
const std::string path) // file path
{
std::ofstream ofile(path, std::ios::binary);
ofile.write((char *) data, size * sizeof(float));
return 0;
}
*/
src/tp_files.h
View file @ 1bf5e8bf
...@@ -11,7 +11,7 @@ ...@@ -11,7 +11,7 @@
#include <iterator> #include <iterator>
#include <vector> #include <vector>
int get_file_size(std::string filename); // path to file int get_file_size(const std::string filename); // path to file
int readFloatsFromFile(float * data, // allocated array int readFloatsFromFile(float * data, // allocated array
const char * path); // file path const char * path); // file path
float * readAllFloatsFromFile(const char * path, float * readAllFloatsFromFile(const char * path,
...@@ -19,4 +19,14 @@ float * readAllFloatsFromFile(const char * path, ...@@ -19,4 +19,14 @@ float * readAllFloatsFromFile(const char * path,
int writeFloatsToFile(float * data, // allocated array int writeFloatsToFile(float * data, // allocated array
int size, // length in elements int size, // length in elements
const char * path); // file path const char * path); // file path
/*
int readFloatsFromFile(float * data, // allocated array
const std::string filename); // file path
float * readAllFloatsFromFile(const std::string filename,
int * len_in_floats);
int writeFloatsToFile(float * data, // allocated array
int size, // length in elements
const std::string filename); // file path
*/
#endif /* SRC_TP_FILES_H_ */ #endif /* SRC_TP_FILES_H_ */
src/tp_paths.cu
View file @ 1bf5e8bf
...@@ -4,243 +4,17 @@ ...@@ -4,243 +4,17 @@
* Created on: Mar 26, 2025 * Created on: Mar 26, 2025
* Author: elphel * Author: elphel
*/ */
#include <vector>
#include <string>
#include "tp_paths.h" #include "tp_paths.h"
TpPaths::TpPaths(int lwir){ TpPaths::TpPaths(int lwir){
// values for LWIR sensors m_lwir = lwir;
const char* kernel_file_lwir[16] = { kernel_file = lwir? m_kernel_file_lwir : m_kernel_file_rgb;
"clt/aux_chn0_transposed.kernel", kernel_offs_file = lwir? m_kernel_offs_file_lwir : m_kernel_offs_file_rgb;
"clt/aux_chn1_transposed.kernel", image_files = lwir? m_image_files_lwir : m_image_files_rgb;
"clt/aux_chn2_transposed.kernel", ports_offs_xy_file = lwir? m_ports_offs_xy_file_lwir : m_ports_offs_xy_file_rgb;
"clt/aux_chn3_transposed.kernel", ports_clt_file = lwir? m_ports_clt_file_lwir : m_ports_clt_file_rgb;
"clt/aux_chn4_transposed.kernel", result_rbg_file = lwir? m_result_rbg_file_lwir : m_result_rbg_file_rgb;
"clt/aux_chn5_transposed.kernel",
"clt/aux_chn6_transposed.kernel",
"clt/aux_chn7_transposed.kernel",
"clt/aux_chn8_transposed.kernel",
"clt/aux_chn9_transposed.kernel",
"clt/aux_chn10_transposed.kernel",
"clt/aux_chn11_transposed.kernel",
"clt/aux_chn12_transposed.kernel",
"clt/aux_chn13_transposed.kernel",
"clt/aux_chn14_transposed.kernel",
"clt/aux_chn15_transposed.kernel"};
const char* kernel_offs_file_lwir[16] = {
"clt/aux_chn0_transposed.kernel_offsets",
"clt/aux_chn1_transposed.kernel_offsets",
"clt/aux_chn2_transposed.kernel_offsets",
"clt/aux_chn3_transposed.kernel_offsets",
"clt/aux_chn4_transposed.kernel_offsets",
"clt/aux_chn5_transposed.kernel_offsets",
"clt/aux_chn6_transposed.kernel_offsets",
"clt/aux_chn7_transposed.kernel_offsets",
"clt/aux_chn8_transposed.kernel_offsets",
"clt/aux_chn9_transposed.kernel_offsets",
"clt/aux_chn10_transposed.kernel_offsets",
"clt/aux_chn11_transposed.kernel_offsets",
"clt/aux_chn12_transposed.kernel_offsets",
"clt/aux_chn13_transposed.kernel_offsets",
"clt/aux_chn14_transposed.kernel_offsets",
"clt/aux_chn15_transposed.kernel_offsets"};
const char* image_files_lwir[16] = {
"clt/aux_chn0.bayer",
"clt/aux_chn1.bayer",
"clt/aux_chn2.bayer",
"clt/aux_chn3.bayer",
"clt/aux_chn4.bayer",
"clt/aux_chn5.bayer",
"clt/aux_chn6.bayer",
"clt/aux_chn7.bayer",
"clt/aux_chn8.bayer",
"clt/aux_chn9.bayer",
"clt/aux_chn10.bayer",
"clt/aux_chn11.bayer",
"clt/aux_chn12.bayer",
"clt/aux_chn13.bayer",
"clt/aux_chn14.bayer",
"clt/aux_chn15.bayer"};
const char* ports_offs_xy_file_lwir[16] = {
"clt/aux_chn0.portsxy",
"clt/aux_chn1.portsxy",
"clt/aux_chn2.portsxy",
"clt/aux_chn3.portsxy",
"clt/aux_chn4.portsxy",
"clt/aux_chn5.portsxy",
"clt/aux_chn6.portsxy",
"clt/aux_chn7.portsxy",
"clt/aux_chn8.portsxy",
"clt/aux_chn9.portsxy",
"clt/aux_chn10.portsxy",
"clt/aux_chn11.portsxy",
"clt/aux_chn12.portsxy",
"clt/aux_chn13.portsxy",
"clt/aux_chn14.portsxy",
"clt/aux_chn15.portsxy"};
//#ifndef DBG_TILE
const char* ports_clt_file_lwir[16] = { // never referenced
"clt/aux_chn0.clt",
"clt/aux_chn1.clt",
"clt/aux_chn2.clt",
"clt/aux_chn3.clt",
"clt/aux_chn4.clt",
"clt/aux_chn5.clt",
"clt/aux_chn6.clt",
"clt/aux_chn7.clt",
"clt/aux_chn8.clt",
"clt/aux_chn9.clt",
"clt/aux_chn10.clt",
"clt/aux_chn11.clt",
"clt/aux_chn12.clt",
"clt/aux_chn13.clt",
"clt/aux_chn14.clt",
"clt/aux_chn15.clt"};
const char* result_rbg_file_lwir[16] = {
"clt/aux_chn0.rbg",
"clt/aux_chn1.rbg",
"clt/aux_chn2.rbg",
"clt/aux_chn3.rbg",
"clt/aux_chn4.rbg",
"clt/aux_chn5.rbg",
"clt/aux_chn6.rbg",
"clt/aux_chn7.rbg",
"clt/aux_chn8.rbg",
"clt/aux_chn9.rbg",
"clt/aux_chn10.rbg",
"clt/aux_chn11.rbg",
"clt/aux_chn12.rbg",
"clt/aux_chn13.rbg",
"clt/aux_chn14.rbg",
"clt/aux_chn15.rbg"};
//#endif
/*
const char* result_corr_file_lwir = "clt/aux_corr.corr";
const char* result_corr_quad_file_lwir = "clt/aux_corr-quad.corr";
const char* result_corr_td_norm_file_lwir = "clt/aux_corr-td-norm.corr";
const char* result_inter_td_norm_file_lwir = "clt/aux_inter-td-norm.corr";
const char* result_textures_file_lwir = "clt/aux_texture_nodp.rgba";
const char* result_diff_rgb_combo_file_lwir ="clt/aux_diff_rgb_combo_nodp.drbg";
const char* result_textures_rgba_file_lwir = "clt/aux_texture_rgba_nodp.rgba";
const char* result_textures_file_dp_lwir = "clt/aux_texture_dp.rgba";
const char* result_diff_rgb_combo_file_dp_lwir ="clt/aux_diff_rgb_combo_dp.drbg";
const char* result_textures_rgba_file_dp_lwir = "clt/aux_texture_rgba_dp.rgba";
const char* rByRDist_file_lwir = "clt/aux.rbyrdist";
const char* correction_vector_file_lwir = "clt/aux.correction_vector";
const char* geometry_correction_file_lwir = "clt/aux.geometry_correction";
float color_weights_lwir [3] = {
1.0, // float weight0, // scale for R 0.5 / (1.0 + 0.5 +0.2)
1.0, // float weight1, // scale for B 0.2 / (1.0 + 0.5 +0.2)
1.0}; // float weight2, // scale for G 1.0 / (1.0 + 0.5 +0.2)
float generate_RBGA_params_lwir[5]={
10.0, // float min_shot, // 10.0
3.0, // float scale_shot, // 3.0
10.0, // 1.5f, // float diff_sigma, // pixel value/pixel change
10.0f, // float diff_threshold, // pixel value/pixel change
12.0 // 3.0 // float min_agree, // minimal number of channels to agree on a point (real number to work with fuzzy averages)
};
*/
// values for RGB sensors
const char* kernel_file_rgb[4] = {
"clt/main_chn0_transposed.kernel",
"clt/main_chn1_transposed.kernel",
"clt/main_chn2_transposed.kernel",
"clt/main_chn3_transposed.kernel"};
const char* kernel_offs_file_rgb[4] = {
"clt/main_chn0_transposed.kernel_offsets",
"clt/main_chn1_transposed.kernel_offsets",
"clt/main_chn2_transposed.kernel_offsets",
"clt/main_chn3_transposed.kernel_offsets"};
const char* image_files_rgb[4] = {
"clt/main_chn0.bayer",
"clt/main_chn1.bayer",
"clt/main_chn2.bayer",
"clt/main_chn3.bayer"};
const char* ports_offs_xy_file_rgb[4] = {
"clt/main_chn0.portsxy",
"clt/main_chn1.portsxy",
"clt/main_chn2.portsxy",
"clt/main_chn3.portsxy"};
const char* ports_clt_file_rgb[4] = { // never referenced
"clt/main_chn0.clt",
"clt/main_chn1.clt",
"clt/main_chn2.clt",
"clt/main_chn3.clt"};
const char* result_rbg_file_rgb[4] = {
"clt/main_chn0.rbg",
"clt/main_chn1.rbg",
"clt/main_chn2.rbg",
"clt/main_chn3.rbg"};
//#endif
/*
const char* result_corr_file_rgb = "clt/main_corr.corr";
const char* result_corr_quad_file_rgb = "clt/main_corr-quad.corr";
const char* result_corr_td_norm_file_rgb = "clt/aux_corr-td-norm.corr";
/// const char* result_corr_cross_file = "clt/main_corr-cross.corr";
const char* result_inter_td_norm_file_rgb = "clt/aux_inter-td-norm.corr";
const char* result_textures_file_rgb = "clt/main_texture_nodp.rgba";
const char* result_diff_rgb_combo_file_rgb = "clt/main_diff_rgb_combo_nodp.drbg";
const char* result_textures_rgba_file_rgb = "clt/main_texture_rgba_nodp.rgba";
const char* result_textures_file_dp_rgb = "clt/main_texture_dp.rgba";
const char* result_diff_rgb_combo_file_dp_rgb = "clt/main_diff_rgb_combo_dp.drbg";
const char* result_textures_rgba_file_dp_rgb = "clt/main_texture_rgba_dp.rgba";
const char* rByRDist_file_rgb = "clt/main.rbyrdist";
const char* correction_vector_file_rgb = "clt/main.correction_vector";
const char* geometry_correction_file_rgb = "clt/main.geometry_correction";
*/
/*
float color_weights_rgb [3] = {
0.294118, // float weight0, // scale for R 0.5 / (1.0 + 0.5 +0.2)
0.117647, // float weight1, // scale for B 0.2 / (1.0 + 0.5 +0.2)
0.588235}; // float weight2, // scale for G 1.0 / (1.0 + 0.5 +0.2)
float generate_RBGA_params_rgb[5]={
10.0, // float min_shot, // 10.0
3.0, // float scale_shot, // 3.0
1.5f, // float diff_sigma, // pixel value/pixel change
10.0f, // float diff_threshold, // pixel value/pixel change
3.0 // float min_agree, // minimal number of channels to agree on a point (real number to work with fuzzy averages)
};
*/
/*
const char ** kernel_file;
const char ** kernel_offs_file;
const char ** image_files;
const char ** ports_offs_xy_file;
const char ** ports_clt_file;
const char ** result_rbg_file;
const char* result_corr_file;
const char* result_corr_quad_file;
const char* result_corr_td_norm_file;
const char* result_inter_td_norm_file;
const char* result_textures_file;
const char* result_diff_rgb_combo_file;
const char* result_textures_rgba_file;
const char* result_textures_file_dp;
const char* result_diff_rgb_combo_file_dp;
const char* result_textures_rgba_file_dp;
const char* rByRDist_file;
const char* correction_vector_file;
const char* geometry_correction_file;
const float* color_weights;
const float* generate_RBGA_params;
*/
kernel_file = lwir? kernel_file_lwir : kernel_file_rgb;
kernel_offs_file = lwir? kernel_offs_file_lwir : kernel_offs_file_rgb;
image_files = lwir? image_files_lwir : image_files_rgb;
ports_offs_xy_file = lwir? ports_offs_xy_file_lwir : ports_offs_xy_file_rgb;
ports_clt_file = lwir? ports_clt_file_lwir : ports_clt_file_rgb;
result_rbg_file = lwir? result_rbg_file_lwir : result_rbg_file_rgb;
result_corr_file = lwir ? "clt/aux_corr.corr" : "clt/main_corr.corr"; result_corr_file = lwir ? "clt/aux_corr.corr" : "clt/main_corr.corr";
result_corr_quad_file = lwir ? "clt/aux_corr-quad.corr" : "clt/main_corr-quad.corr"; result_corr_quad_file = lwir ? "clt/aux_corr-quad.corr" : "clt/main_corr-quad.corr";
result_corr_td_norm_file = lwir ? "clt/aux_corr-td-norm.corr": "clt/aux_corr-td-norm.corr"; result_corr_td_norm_file = lwir ? "clt/aux_corr-td-norm.corr": "clt/aux_corr-td-norm.corr";
...@@ -254,8 +28,6 @@ TpPaths::TpPaths(int lwir){ ...@@ -254,8 +28,6 @@ TpPaths::TpPaths(int lwir){
rByRDist_file = lwir ? "clt/aux.rbyrdist" : "clt/main.rbyrdist"; rByRDist_file = lwir ? "clt/aux.rbyrdist" : "clt/main.rbyrdist";
correction_vector_file = lwir ? "clt/aux.correction_vector" : "clt/main.correction_vector"; correction_vector_file = lwir ? "clt/aux.correction_vector" : "clt/main.correction_vector";
geometry_correction_file = lwir ? "clt/aux.geometry_correction" : "clt/main.geometry_correction"; geometry_correction_file = lwir ? "clt/aux.geometry_correction" : "clt/main.geometry_correction";
// color_weights = lwir ? color_weights_lwir : color_weights_rgb;
// generate_RBGA_params = lwir ? generate_RBGA_params_lwir : generate_RBGA_params_rgb;
} }
......
src/tp_paths.h
View file @ 1bf5e8bf
...@@ -11,7 +11,9 @@ ...@@ -11,7 +11,9 @@
class TpPaths{ class TpPaths{
public: public:
TpPaths(int lwir); TpPaths(int lwir);
const char ** kernel_file; int m_lwir;
const char ** kernel_file;
const char ** kernel_offs_file; const char ** kernel_offs_file;
const char ** image_files; const char ** image_files;
const char ** ports_offs_xy_file; const char ** ports_offs_xy_file;
...@@ -31,7 +33,141 @@ public: ...@@ -31,7 +33,141 @@ public:
const char* rByRDist_file; const char* rByRDist_file;
const char* correction_vector_file; const char* correction_vector_file;
const char* geometry_correction_file; const char* geometry_correction_file;
// float * color_weights;// [3];
// float * generate_RBGA_params; // [5]; private:
const char * m_kernel_file_lwir[16] ={
"clt/aux_chn0_transposed.kernel",
"clt/aux_chn1_transposed.kernel",
"clt/aux_chn2_transposed.kernel",
"clt/aux_chn3_transposed.kernel",
"clt/aux_chn4_transposed.kernel",
"clt/aux_chn5_transposed.kernel",
"clt/aux_chn6_transposed.kernel",
"clt/aux_chn7_transposed.kernel",
"clt/aux_chn8_transposed.kernel",
"clt/aux_chn9_transposed.kernel",
"clt/aux_chn10_transposed.kernel",
"clt/aux_chn11_transposed.kernel",
"clt/aux_chn12_transposed.kernel",
"clt/aux_chn13_transposed.kernel",
"clt/aux_chn14_transposed.kernel",
"clt/aux_chn15_transposed.kernel"};
const char * m_kernel_offs_file_lwir[16] = {
"clt/aux_chn0_transposed.kernel_offsets",
"clt/aux_chn1_transposed.kernel_offsets",
"clt/aux_chn2_transposed.kernel_offsets",
"clt/aux_chn3_transposed.kernel_offsets",
"clt/aux_chn4_transposed.kernel_offsets",
"clt/aux_chn5_transposed.kernel_offsets",
"clt/aux_chn6_transposed.kernel_offsets",
"clt/aux_chn7_transposed.kernel_offsets",
"clt/aux_chn8_transposed.kernel_offsets",
"clt/aux_chn9_transposed.kernel_offsets",
"clt/aux_chn10_transposed.kernel_offsets",
"clt/aux_chn11_transposed.kernel_offsets",
"clt/aux_chn12_transposed.kernel_offsets",
"clt/aux_chn13_transposed.kernel_offsets",
"clt/aux_chn14_transposed.kernel_offsets",
"clt/aux_chn15_transposed.kernel_offsets"};
const char * m_image_files_lwir[16]= {
"clt/aux_chn0.bayer",
"clt/aux_chn1.bayer",
"clt/aux_chn2.bayer",
"clt/aux_chn3.bayer",
"clt/aux_chn4.bayer",
"clt/aux_chn5.bayer",
"clt/aux_chn6.bayer",
"clt/aux_chn7.bayer",
"clt/aux_chn8.bayer",
"clt/aux_chn9.bayer",
"clt/aux_chn10.bayer",
"clt/aux_chn11.bayer",
"clt/aux_chn12.bayer",
"clt/aux_chn13.bayer",
"clt/aux_chn14.bayer",
"clt/aux_chn15.bayer"};
const char * m_ports_offs_xy_file_lwir[16]= {
"clt/aux_chn0.portsxy",
"clt/aux_chn1.portsxy",
"clt/aux_chn2.portsxy",
"clt/aux_chn3.portsxy",
"clt/aux_chn4.portsxy",
"clt/aux_chn5.portsxy",
"clt/aux_chn6.portsxy",
"clt/aux_chn7.portsxy",
"clt/aux_chn8.portsxy",
"clt/aux_chn9.portsxy",
"clt/aux_chn10.portsxy",
"clt/aux_chn11.portsxy",
"clt/aux_chn12.portsxy",
"clt/aux_chn13.portsxy",
"clt/aux_chn14.portsxy",
"clt/aux_chn15.portsxy"};
const char * m_ports_clt_file_lwir[16]= { // never referenced
"clt/aux_chn0.clt",
"clt/aux_chn1.clt",
"clt/aux_chn2.clt",
"clt/aux_chn3.clt",
"clt/aux_chn4.clt",
"clt/aux_chn5.clt",
"clt/aux_chn6.clt",
"clt/aux_chn7.clt",
"clt/aux_chn8.clt",
"clt/aux_chn9.clt",
"clt/aux_chn10.clt",
"clt/aux_chn11.clt",
"clt/aux_chn12.clt",
"clt/aux_chn13.clt",
"clt/aux_chn14.clt",
"clt/aux_chn15.clt"};
const char * m_result_rbg_file_lwir[16] = {
"clt/aux_chn0.rbg",
"clt/aux_chn1.rbg",
"clt/aux_chn2.rbg",
"clt/aux_chn3.rbg",
"clt/aux_chn4.rbg",
"clt/aux_chn5.rbg",
"clt/aux_chn6.rbg",
"clt/aux_chn7.rbg",
"clt/aux_chn8.rbg",
"clt/aux_chn9.rbg",
"clt/aux_chn10.rbg",
"clt/aux_chn11.rbg",
"clt/aux_chn12.rbg",
"clt/aux_chn13.rbg",
"clt/aux_chn14.rbg",
"clt/aux_chn15.rbg"};
const char * m_kernel_file_rgb[4] = {
"clt/main_chn0_transposed.kernel",
"clt/main_chn1_transposed.kernel",
"clt/main_chn2_transposed.kernel",
"clt/main_chn3_transposed.kernel"};
const char * m_kernel_offs_file_rgb[4] = {
"clt/main_chn0_transposed.kernel_offsets",
"clt/main_chn1_transposed.kernel_offsets",
"clt/main_chn2_transposed.kernel_offsets",
"clt/main_chn3_transposed.kernel_offsets"};
const char * m_image_files_rgb[4] = {
"clt/main_chn0.bayer",
"clt/main_chn1.bayer",
"clt/main_chn2.bayer",
"clt/main_chn3.bayer"};
const char * m_ports_offs_xy_file_rgb[4] = {
"clt/main_chn0.portsxy",
"clt/main_chn1.portsxy",
"clt/main_chn2.portsxy",
"clt/main_chn3.portsxy"};
const char * m_ports_clt_file_rgb[4] = { // never referenced
"clt/main_chn0.clt",
"clt/main_chn1.clt",
"clt/main_chn2.clt",
"clt/main_chn3.clt"};
const char * m_result_rbg_file_rgb[4] = {
"clt/main_chn0.rbg",
"clt/main_chn1.rbg",
"clt/main_chn2.rbg",
"clt/main_chn3.rbg"};
}; };
#endif /* SRC_TP_PATHS_H_ */ #endif /* SRC_TP_PATHS_H_ */
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