Skip to content

I
imagej-elphel
Commit b0ba6ef5 authored by Andrey Filippov's avatar Andrey Filippov
Browse files
Options

more threads per block, limited by the shared memory size

parent 54b1382d
Show whitespace changes
Inline Side-by-side
Showing with 50 additions and 43 deletions
src/main/resources/TileProcessor.cuh
View file @ b0ba6ef5
...@@ -39,20 +39,17 @@ ...@@ -39,20 +39,17 @@
#pragma once #pragma once
#include "dtt8x8.cuh" #include "dtt8x8.cuh"
// Using 1 tile per block with 32 threads per tile. // Not enough shared memory to have more threads per block,even just for the result clt tiles
// some subtasks use 8 threads per 2d DTT #define TILES_PER_BLOCK 2
//#define TILES_PER_BLOCK 1
#define THREADS_PER_TILE 32 #define THREADS_PER_TILE 32
#define IMG_WIDTH 2592 #define IMG_WIDTH 2592
#define IMG_HEIGHT 1936 #define IMG_HEIGHT 1936
#define NUM_CAMS 4 #define NUM_CAMS 4
#define NUM_COLORS 3 #define NUM_COLORS 3
//#define KERNELS_STEP 16
#define KERNELS_LSTEP 4 #define KERNELS_LSTEP 4
#define KERNELS_HOR 164 #define KERNELS_HOR 164
#define KERNELS_VERT 123 #define KERNELS_VERT 123
#define IMAGE_TILE_SIDE 18 #define IMAGE_TILE_SIDE 18
//#define KERNEL_OFFSETS 8
#define KERNELS_STEP (1 << KERNELS_LSTEP) #define KERNELS_STEP (1 << KERNELS_LSTEP)
#define TILESX (IMG_WIDTH / DTT_SIZE) #define TILESX (IMG_WIDTH / DTT_SIZE)
#define TILESY (IMG_HEIGHT / DTT_SIZE) #define TILESY (IMG_HEIGHT / DTT_SIZE)
...@@ -219,7 +216,10 @@ __device__ void convertCorrectTile( ...@@ -219,7 +216,10 @@ __device__ void convertCorrectTile(
float clt_kernels [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports float clt_kernels [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports
// float bayer_tiles [IMAGE_TILE_SIDE][IMAGE_TILE_SIDE], // float bayer_tiles [IMAGE_TILE_SIDE][IMAGE_TILE_SIDE],
int int_topleft [NUM_COLORS][2], int int_topleft [NUM_COLORS][2],
float residual_shift [NUM_COLORS][2]); float residual_shift [NUM_COLORS][2],
float window_hor_cos [NUM_COLORS][2*DTT_SIZE],
float window_hor_sin [NUM_COLORS][2*DTT_SIZE],
float window_vert_cos [NUM_COLORS][2*DTT_SIZE]);
// Fractional pixel shift (phase rotation), horizontal. In-place. // Fractional pixel shift (phase rotation), horizontal. In-place.
__device__ void shiftTileHor( __device__ void shiftTileHor(
...@@ -250,28 +250,29 @@ __global__ void tileProcessor( ...@@ -250,28 +250,29 @@ __global__ void tileProcessor(
// struct CltExtra * dbg_ce_h1= &gpu_kernel_offsets[0][14328 + (164*123)]; // struct CltExtra * dbg_ce_h1= &gpu_kernel_offsets[0][14328 + (164*123)];
// struct CltExtra * dbg_ce_h2= &gpu_kernel_offsets[0][14328 + 2* (164*123)]; // struct CltExtra * dbg_ce_h2= &gpu_kernel_offsets[0][14328 + 2* (164*123)];
int task_num = blockIdx.x; // * TILES_PER_BLOCK + threadIdx.y; int tile_in_block = threadIdx.z;
int task_num = blockIdx.x * TILES_PER_BLOCK + tile_in_block;
if (task_num >= num_tiles) return; // nothing to do if (task_num >= num_tiles) return; // nothing to do
struct tp_task * gpu_task = &gpu_tasks[task_num]; struct tp_task * gpu_task = &gpu_tasks[task_num];
if (!gpu_task->task) return; // NOP tile if (!gpu_task->task) return; // NOP tile
__shared__ struct tp_task tt; // [TILES_PER_BLOCK]; __shared__ struct tp_task tt [TILES_PER_BLOCK];
// Copy task data to shared memory // Copy task data to shared memory
int nc = (threadIdx.x >> 1) + (threadIdx.y << 2) - 1; int nc = (threadIdx.x >> 1) + (threadIdx.y << 2) - 1;
if (nc < 0) { if (nc < 0) {
tt.task = gpu_task -> task; tt[tile_in_block].task = gpu_task -> task;
tt.tx = gpu_task -> tx; tt[tile_in_block].tx = gpu_task -> tx;
tt.ty = gpu_task -> ty; tt[tile_in_block].ty = gpu_task -> ty;
} else { } else {
if (nc < NUM_CAMS) { if (nc < NUM_CAMS) {
tt.xy[nc][0] = gpu_task -> xy[nc][0]; tt[tile_in_block].xy[nc][0] = gpu_task -> xy[nc][0];
tt.xy[nc][1] = gpu_task -> xy[nc][1]; tt[tile_in_block].xy[nc][1] = gpu_task -> xy[nc][1];
} }
} }
if (NUM_CAMS > 31){ // unlikely if (NUM_CAMS > 31){ // unlikely
nc += 32; nc += 32;
while (nc < NUM_CAMS){ while (nc < NUM_CAMS){
tt.xy[nc][0] = gpu_task -> xy[nc][0]; tt[tile_in_block].xy[nc][0] = gpu_task -> xy[nc][0];
tt.xy[nc][1] = gpu_task -> xy[nc][1]; tt[tile_in_block].xy[nc][1] = gpu_task -> xy[nc][1];
nc += 32; nc += 32;
} }
} }
...@@ -282,14 +283,18 @@ __global__ void tileProcessor( ...@@ -282,14 +283,18 @@ __global__ void tileProcessor(
// clt_tile[][0] - before rotation, [][0][0] - R:DCT/DCT, [][0][1] - B:DCT/DCT, [][0][2] - G:DCT/DCT, [][0][3] - G:DST/DCT, // clt_tile[][0] - before rotation, [][0][0] - R:DCT/DCT, [][0][1] - B:DCT/DCT, [][0][2] - G:DCT/DCT, [][0][3] - G:DST/DCT,
// clt_tile[][1], clt_tile[][2], and clt_tile[][3] - after rotation, 4 quadrants each // clt_tile[][1], clt_tile[][2], and clt_tile[][3] - after rotation, 4 quadrants each
// changed, above is wrong now // changed, above is wrong now
__shared__ float clt_tile [NUM_CAMS][NUM_COLORS][4][DTT_SIZE][DTT_SIZE1]; __shared__ float clt_tile [TILES_PER_BLOCK][NUM_CAMS][NUM_COLORS][4][DTT_SIZE][DTT_SIZE1];
// sharing for cameras as they are corrected one after another // sharing shared memory for cameras as they are corrected one after another
__shared__ float clt_kernels [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1]; // +1 to alternate column ports // TODO: evaluate total shared memory usage, maybe this sharing is not needed
__shared__ int int_topleft [NUM_COLORS][2];
__shared__ float residual_shift [NUM_COLORS][2]; __shared__ float clt_kernels [TILES_PER_BLOCK][NUM_COLORS][4][DTT_SIZE][DTT_SIZE1]; // +1 to alternate column ports
// __shared__ float window_hor_cos [NUM_COLORS][2*DTT_SIZE]; __shared__ int int_topleft [TILES_PER_BLOCK][NUM_COLORS][2];
// __shared__ float window_hor_sin [NUM_COLORS][2*DTT_SIZE]; __shared__ float residual_shift [TILES_PER_BLOCK][NUM_COLORS][2];
// __shared__ float window_vert_cos [NUM_COLORS][2*DTT_SIZE];
__shared__ float window_hor_cos [TILES_PER_BLOCK][NUM_COLORS][2*DTT_SIZE];
__shared__ float window_hor_sin [TILES_PER_BLOCK][NUM_COLORS][2*DTT_SIZE];
__shared__ float window_vert_cos [TILES_PER_BLOCK][NUM_COLORS][2*DTT_SIZE];
//IMAGE_TILE_SIDE //IMAGE_TILE_SIDE
// process each camera in series // process each camera in series
for (int ncam = 0; ncam < NUM_CAMS; ncam++){ for (int ncam = 0; ncam < NUM_CAMS; ncam++){
...@@ -297,14 +302,16 @@ __global__ void tileProcessor( ...@@ -297,14 +302,16 @@ __global__ void tileProcessor(
gpu_kernel_offsets[ncam], // float * gpu_kernel_offsets, gpu_kernel_offsets[ncam], // float * gpu_kernel_offsets,
gpu_kernels[ncam], // float * gpu_kernels, gpu_kernels[ncam], // float * gpu_kernels,
gpu_images[ncam], // float * gpu_images, gpu_images[ncam], // float * gpu_images,
// &tt[threadIdx.y], // struct tp_task * tt, tt[tile_in_block].xy[ncam][0], // float centerX,
tt.xy[ncam][0], // float centerX, tt[tile_in_block].xy[ncam][1], // float centerY,
tt.xy[ncam][1], // float centerY,
dstride, // size_t dstride, // in floats (pixels) dstride, // size_t dstride, // in floats (pixels)
clt_tile [ncam], // float clt_tile [TILES_PER_BLOCK][NUM_CAMS][NUM_COLORS][4][DTT_SIZE][DTT_SIZE]) clt_tile [tile_in_block][ncam], // float clt_tile [TILES_PER_BLOCK][NUM_CAMS][NUM_COLORS][4][DTT_SIZE][DTT_SIZE])
clt_kernels, // float clt_tile [NUM_COLORS][4][DTT_SIZE][DTT_SIZE], clt_kernels[tile_in_block], // float clt_tile [NUM_COLORS][4][DTT_SIZE][DTT_SIZE],
int_topleft, // int int_topleft [NUM_COLORS][2], int_topleft[tile_in_block], // int int_topleft [NUM_COLORS][2],
residual_shift); // float frac_topleft [NUM_COLORS][2], residual_shift[tile_in_block], // float frac_topleft [NUM_COLORS][2],
window_hor_cos[tile_in_block], // float window_hor_cos [NUM_COLORS][2*DTT_SIZE],
window_hor_sin[tile_in_block], //float window_hor_sin [NUM_COLORS][2*DTT_SIZE],
window_vert_cos[tile_in_block]); //float window_vert_cos [NUM_COLORS][2*DTT_SIZE]);
} }
} }
...@@ -438,16 +445,16 @@ __device__ void convertCorrectTile( ...@@ -438,16 +445,16 @@ __device__ void convertCorrectTile(
float clt_tile [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports float clt_tile [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports
float clt_kernels [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports float clt_kernels [NUM_COLORS][4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports
int int_topleft [NUM_COLORS][2], int int_topleft [NUM_COLORS][2],
float residual_shift [NUM_COLORS][2]) float residual_shift [NUM_COLORS][2],
float window_hor_cos [NUM_COLORS][2*DTT_SIZE],
float window_hor_sin [NUM_COLORS][2*DTT_SIZE],
float window_vert_cos [NUM_COLORS][2*DTT_SIZE])
{ {
// struct CltExtra * dbg_ce0= &gpu_kernel_offsets[14328];
// struct CltExtra * dbg_ce1= &gpu_kernel_offsets[14328 + (164*123)]; /// __shared__ float window_hor_cos [NUM_COLORS][2*DTT_SIZE];
// struct CltExtra * dbg_ce2= &gpu_kernel_offsets[14328 + 2* (164*123)]; /// __shared__ float window_hor_sin [NUM_COLORS][2*DTT_SIZE];
/// __shared__ float window_vert_cos [NUM_COLORS][2*DTT_SIZE];
__shared__ float window_hor_cos [NUM_COLORS][2*DTT_SIZE];
__shared__ float window_hor_sin [NUM_COLORS][2*DTT_SIZE];
__shared__ float window_vert_cos [NUM_COLORS][2*DTT_SIZE];
// __shared__ float rot_hvcs [NUM_COLORS][4][DTT_SIZE1]; // rotation cosine/sines: CH,SH,CV,SV for each color
// get correct kernel tile, then use 2 threads per kernel and image // get correct kernel tile, then use 2 threads per kernel and image
int ktileX, ktileY; int ktileX, ktileY;
int kernel_index; // common for all coors int kernel_index; // common for all coors
......
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