Index: /soft/giet_vm/applications/convol/main.c =================================================================== --- /soft/giet_vm/applications/convol/main.c (revision 443) +++ /soft/giet_vm/applications/convol/main.c (revision 444) @@ -3,6 +3,21 @@ // Date : june 2014 // author : Alain Greiner +//////////////////////////////////////////////////////////////////////////////////////////// +// This multi-threaded application application implements a 2D convolution product. +// The convolution kernel is [201]*[35] pixels, but it can be factored in two +// independant line and column convolution products. +// It can run on a multi-processors, multi-clusters architecture, with one thread +// per processor. It uses the he following hardware parameters, that must be defined +// in the hard_config.h file: +// - X_SIZE : number of clusters in a row +// - Y_SIZE : number of clusters in a column +// - NB_PROCS_MAX : number of processors per cluster +// - FBUF_X_SIZE : number of pixels per line in frame buffer +// - FBUF_Y_SIZE : number of lines in frame buffer +// +// The (1024 * 1024) pixels image is read from a file (2 bytes per pixel). // -// The "convol" application implements a 2D convolution product. +// - The number of clusters containing processors must be a power of 2. +// - The number of processors per cluster must be a power of 2. //////////////////////////////////////////////////////////////////////////////////////////// @@ -106,5 +121,4 @@ unsigned int task_id = (cluster_id * nprocs) + lpid; // continuous task index unsigned int ntasks = nclusters * nprocs; // number of tasks - unsigned int npixels = NB_PIXELS; // pixels per frame unsigned int frame_size = FRAME_SIZE; // total size (bytes) unsigned int nblocks = frame_size / 512; // number of blocks per frame @@ -126,4 +140,8 @@ // parameters checking + if ( (NP != FBUF_X_SIZE) || (NL != FBUF_Y_SIZE) ) + { + giet_exit("[TRANSPOSE ERROR] Frame buffer size does not fit image size"); + } if ((nprocs != 1) && (nprocs != 2) && (nprocs != 4)) giet_exit( "[CONVOL ERROR] NB_PROCS_MAX must be 1, 2 or 4\n"); @@ -148,13 +166,30 @@ if ( (x==0) && (y==0) && (lpid==0) ) { + // parameters checking + if ( (NP != FBUF_X_SIZE) || (NL != FBUF_Y_SIZE) ) + giet_exit("[TRANSPOSE ERROR] Frame buffer size does not fit image size"); + + if ((nprocs != 1) && (nprocs != 2) && (nprocs != 4)) + giet_exit( "[CONVOL ERROR] NB_PROCS_MAX must be 1, 2 or 4\n"); + + if ((X_SIZE!=1) && (X_SIZE!=2) && (X_SIZE!=4) && (X_SIZE!=8) && (X_SIZE!=16)) + giet_exit( "[CONVOL ERROR] X_SIZE must be 1, 2, 4, 8, 16\n"); + + if ((Y_SIZE!=1) && (Y_SIZE!=2) && (Y_SIZE!=4) && (Y_SIZE!=8) && (Y_SIZE!=16)) + giet_exit( "[CONVOL ERROR] Y_SIZE must be 1, 2, 4, 8, 16\n"); + + if ( NL % nclusters != 0 ) + giet_exit( "[CONVOL ERROR] NB_CLUSTERS must be a divider of NL"); + + if ( NP % nclusters != 0 ) + giet_exit( "[CONVOL ERROR] NB_CLUSTERS must be a divider of NP"); + + giet_shr_printf("\n[CONVOL] task[0,0,0] starts barrier init at cycle %d\n" "- NB_CLUSTERS = %d\n" - "- NB_LOCAL_PROCS = %d\n" + "- NB_PROCS_MAX = %d\n" "- NB_TASKS = %d\n" - "- NB_PIXELS = %x\n" - "- FRAME_SIZE = %x\n" "- NB_BLOCKS = %x\n", - giet_proctime(), nclusters, nprocs, ntasks, - npixels, frame_size, nblocks ); + giet_proctime(), nclusters, nprocs, ntasks, nblocks ); #if USE_SBT_BARRIER sbt_barrier_init( &barrier, ntasks ); @@ -294,7 +329,7 @@ } - giet_fb_sync_write( NP*(l + (task_id * lines_per_task) ), - &TZ(cluster_id, line, 0), - NP); + giet_fbf_sync_write( NP*(l + (task_id * lines_per_task) ), + &TZ(cluster_id, line, 0), + NP); } @@ -572,7 +607,7 @@ } - giet_fb_sync_write( NP*(l + (task_id * lines_per_task) ), - &TZ(cluster_id, line, 0), - NP); + giet_fbf_sync_write( NP*(l + (task_id * lines_per_task) ), + &TZ(cluster_id, line, 0), + NP); } Index: /soft/giet_vm/applications/display/display.ld =================================================================== --- /soft/giet_vm/applications/display/display.ld (revision 443) +++ /soft/giet_vm/applications/display/display.ld (revision 444) @@ -3,6 +3,6 @@ *****************************************************************************/ -seg_code_base = 0x00400000; -seg_data_base = 0x00500000; +seg_code_base = 0x10000000; +seg_data_base = 0x20000000; /*************************************************************************** Index: /soft/giet_vm/applications/display/display.py =================================================================== --- /soft/giet_vm/applications/display/display.py (revision 444) +++ /soft/giet_vm/applications/display/display.py (revision 444) @@ -0,0 +1,83 @@ +#!/usr/bin/env python + +from mapping import * + +###################################################################################### +# file : display.py +# date : may 2014 +# author : Alain Greiner +####################################################################################### +# This file describes the mapping of the singl-threaded "display" application +# on processor[0][0][0] of a multi-clusters, multi-processors architecture. +#################################################################################### + +###################### +def display( mapping ): + + nprocs = mapping.nprocs + x_width = mapping.x_width + y_width = mapping.y_width + + # define vsegs base & size + code_base = 0x10000000 + code_size = 0x00010000 # 64 Kbytes + + data_base = 0x20000000 + data_size = 0x00010000 # 64 Kbytes + + stack_base = 0x40000000 + stack_size = 0x00200000 # 2 Mbytes + + heap_base = 0x60000000 + heap_size = 0x00001000 # 4 Kbytes + + # create vspace + vspace = mapping.addVspace( name = 'display', startname = 'disp_data' ) + + # data vseg + mapping.addVseg( vspace, 'disp_data', data_base , data_size, + 'C_WU', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM', + binpath = 'build/display/display.elf', + local = False ) + + # code vseg + mapping.addVseg( vspace, 'disp_code', code_base , code_size, + 'CXWU', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM', + binpath = 'build/display/display.elf', + local = False ) + + # stack vseg + mapping.addVseg( vspace, 'disp_stack', stack_base, stack_size, + 'C_WU', vtype = 'BUFFER', x = 0 , y = 0 , pseg = 'RAM', + local = False, big = True ) + + # heap vseg (unused) + mapping.addVseg( vspace, 'disp_heap', heap_base, heap_size, + 'C_WU', vtype = 'BUFFER', x = 0 , y = 0 , pseg = 'RAM', + local = False ) + + # task + mapping.addTask( vspace, 'disp', 0, 0, 0, 0, 'disp_stack', 'disp_heap', 0 ) + + # extend mapping name + mapping.name += '_display' + + return vspace # useful for test + +################################ test ###################################################### + +if __name__ == '__main__': + + vspace = display( Mapping( 'test', 2, 2, 4 ) ) + print vspace.xml() + + +# Local Variables: +# tab-width: 4; +# c-basic-offset: 4; +# c-file-offsets:((innamespace . 0)(inline-open . 0)); +# indent-tabs-mode: nil; +# End: +# +# vim: filetype=python:expandtab:shiftwidth=4:tabstop=4:softtabstop=4 + Index: /soft/giet_vm/applications/display/main.c =================================================================== --- /soft/giet_vm/applications/display/main.c (revision 443) +++ /soft/giet_vm/applications/display/main.c (revision 444) @@ -1,50 +1,100 @@ -#include "stdio.h" -#include "hard_config.h" + /////////////////////////////////////////////////////////////////////////////// +// file : main.c (for display application) +// date : may 2014 +// author : Alain Greiner +/////////////////////////////////////////////////////////////////////////////////////// +// This file describes the single thread "display" application. +// It uses the external chained buffer DMA to display a stream +// of images on the frame buffer. +/////////////////////////////////////////////////////////////////////////////////////// -#define NBLOCKS 32 // (128 * 128) / 512 +#include +#include -unsigned char buf[128*128] __attribute__((aligned(512))); +#define FILENAME "misc/images.raw" +#define NPIXELS 128 +#define NLINES 128 +#define NIMAGES 10 +#define NBLOCKS (NPIXELS*NLINES/512) // number of blocks per image + +unsigned char buf0[NPIXELS*NLINES] __attribute__((aligned(512))); +unsigned char buf1[NPIXELS*NLINES] __attribute__((aligned(512))); + +//////////////////////////////////////////// __attribute__((constructor)) void main() -{ +//////////////////////////////////////////// +{ + // get processor identifiers + unsigned int x; + unsigned int y; + unsigned int lpid; + giet_proc_xyp( &x, &y, &lpid ); + int fd; - int blocks; - unsigned int ko; - unsigned int blocks_to_skip = 8*NBLOCKS; + unsigned int image = 0; - unsigned int procid = giet_procid(); - unsigned int cluster_xy = procid/NB_PROCS_MAX; - unsigned int lpid = procid%NB_PROCS_MAX; - unsigned int x = cluster_xy >> Y_WIDTH; - unsigned int y = cluster_xy & ((1<>5), giet_proctime()); - - // transfert vers le frame buffer - giet_fb_sync_write( 0, buf, 128 * 128 ); - - giet_tty_printf("\ndisplay completed for image %d at cycle %d \n", - (blocks_to_skip>>5), giet_proctime()); - - blocks_to_skip = blocks_to_skip + NBLOCKS; + giet_exit("[DISPLAY ERROR] Frame buffer size does not fit image size"); } - giet_exit("completed"); + giet_shr_printf("\n[DISPLAY] Processor[%d,%d,%d] starts at cycle %d\n", + x, y, lpid, giet_proctime() ); + + // open file + fd = giet_fat_open( FILENAME , 0 ); + if ( fd < 0 ) + { + giet_exit("echec giet_fat_open for misc/images.raw"); + } + else + { + giet_shr_printf("\n[DISPLAY] Proc[%d,%d,%d] open file %s at cycle %d\n", + x, y, lpid, FILENAME, giet_proctime() ); + } + + // get a Chained Buffer DMA channel + giet_fbf_cma_alloc(); + + // start Chained Buffer DMA channel + giet_fbf_cma_start( buf0, buf1, NPIXELS*NLINES ); + + giet_shr_printf("\n[DISPLAY] Proc[%d,%d,%d] starts CMA at cycle %d\n", + x, y, lpid, giet_proctime() ); + + // Main loop (on images) + while ( image < NIMAGES ) + { + giet_fat_read( fd, buf0, NBLOCKS, image*NBLOCKS ); + + giet_shr_printf("\n[DISPLAY] Proc[%d,%d,%d] load image %d to buf0 at cycle %d\n", + x, y, lpid, image, giet_proctime() ); + + giet_fbf_cma_display( 0 ); + + giet_shr_printf("\n[DISPLAY] Proc[%d,%d,%d] display image %d from buf0 at cycle %d\n", + x, y, lpid, image, giet_proctime() ); + + image++; + + giet_fat_read( fd, buf1, NBLOCKS, image*NBLOCKS ); + + giet_shr_printf("\n[DISPLAY] Proc[%d,%d,%d] load image %d to buf1 at cycle %d\n", + x, y, lpid, image, giet_proctime() ); + + giet_fbf_cma_display( 1 ); + + giet_shr_printf("\n[DISPLAY] Proc[%d,%d,%d] display image %d from buf1 at cycle %d\n", + x, y, lpid, image, giet_proctime() ); + + + image++; + } + + // stop Chained buffer DMA channel + giet_fbf_cma_stop(); + + giet_exit("display completed"); } Index: ft/giet_vm/applications/display/main_cma.c =================================================================== --- /soft/giet_vm/applications/display/main_cma.c (revision 443) +++ (revision ) @@ -1,64 +1,0 @@ -#include "stdio.h" -#include "hard_config.h" - -#define NBLOCKS 32 - -unsigned char buf0[128*128] __attribute__((aligned(512))); -unsigned char buf1[128*128] __attribute__((aligned(512))); - -__attribute__((constructor)) void main(void) -{ - // get processor identifiers - unsigned int x; - unsigned int y; - unsigned int lpid; - giet_proc_xyp( &x, &y, &lpid ); - - int fd; - unsigned int blocks_to_skip = 0; - - giet_tty_printf( "*** Starting task display on processor[%d,%d,%d] at cycle %d\n\n", - x, y, lpid, giet_proctime() ); - - fd = giet_fat_open( "misc/images.raw", 0 ); - if ( fd < 0 ) giet_exit("echec giet_fat_open for misc/images.raw"); - - giet_tty_printf("\ngiet_fat_open completed for misc/images.raw at cycle %d\n", - giet_proctime() ); - - giet_fb_cma_init( buf0, buf1, 128*128 ); - - giet_tty_printf("\ngiet_cma_init completed at cycle %d\n", - giet_proctime() ); - - while ( blocks_to_skip < 10 * NBLOCKS ) - { - giet_fat_read( fd, buf0, NBLOCKS, blocks_to_skip ); - - giet_tty_printf("\ngiet_fat_read to buf0 completed at cycle = %d\n", - giet_proctime() ); - - giet_fb_cma_write( 0 ); - - giet_tty_printf("giet_cma_write for buf0 completed at cycle = %d\n", - giet_proctime() ); - - blocks_to_skip = blocks_to_skip + NBLOCKS; - - giet_fat_read( fd, buf1, NBLOCKS, blocks_to_skip ); - - giet_tty_printf("giet_fat_read to buf1 completed at cycle = %d\n", - giet_proctime() ); - - giet_fb_cma_write( 1 ); - - giet_tty_printf("giet_cma_write for buf1 completed at cycle = %d\n", - giet_proctime() ); - - blocks_to_skip = blocks_to_skip + NBLOCKS; - } - - giet_fb_cma_stop(); - - giet_exit("display completed"); -} Index: /soft/giet_vm/applications/gameoflife/main.c =================================================================== --- /soft/giet_vm/applications/gameoflife/main.c (revision 443) +++ /soft/giet_vm/applications/gameoflife/main.c (revision 444) @@ -106,5 +106,5 @@ } - giet_fb_sync_write( base_line * WIDTH , + giet_fbf_sync_write( base_line * WIDTH , &world[NEW][base_line][0], nb_line * WIDTH); Index: /soft/giet_vm/applications/transpose/main.c =================================================================== --- /soft/giet_vm/applications/transpose/main.c (revision 443) +++ /soft/giet_vm/applications/transpose/main.c (revision 444) @@ -1,19 +1,28 @@ -/////////////////////////////////////////////////////////////////////////////////////////////// +///////////////////////////////////////////////////////////////////////////////////////////// // File : main.c (for transpose application) // Date : february 2014 // author : Alain Greiner -// -// This application makes a transpose for a NN*NN pixels sequence of images. +///////////////////////////////////////////////////////////////////////////////////////////// +// This multi-threaded application makes a transpose for a NN*NN pixels sequence of images. +// It can run on a multi-processors, multi-clusters architecture, with one thread +// per processor. It uses the he following hardware parameters, that must be defined +// in the hard_config.h file: +// - X_SIZE : number of clusters in a row +// - Y_SIZE : number of clusters in a column +// - NB_PROCS_MAX : number of processors per cluster +// - FBUF_X_SIZE : number of pixels per line in frame buffer +// - FBUF_Y_SIZE : number of lines in frame buffer +// // The image sequence is read from a file (one byte per pixel). // The input and output buffers containing the image are distributed in all clusters. // -// - The image size NN must be a power of 2. +// - The image size NN must be a power of 2 and must fit the frame buffer size. // - The number of clusters containing processors must be a power of 2. -// - The number of processors per cluster (NB_PROCS_MAX) must be a power of 2. +// - The number of processors per cluster must be a power of 2. // - The image size NN must be larger or equal to the total number of processor. // // For each image the application makes a self test (checksum for each line). // The actual display on the frame buffer depends on frame buffer availability. -/////////////////////////////////////////////////////////////////////////////////////////////// +///////////////////////////////////////////////////////////////////////////////////////////// #include "hard_config.h" @@ -23,5 +32,5 @@ #define NN 128 // image size : nlines = npixels = 128 -#define NB_IMAGES 2 // number of images to be handled +#define NB_IMAGES 5 // number of images to be handled #define FILE_PATHNAME "misc/images.raw" // file pathname on disk #define NB_CLUSTERS (X_SIZE * Y_SIZE) // number of clusters @@ -54,5 +63,5 @@ giet_barrier_t barrier; -volatile unsigned int init_ok = 1; +volatile unsigned int init_ok = 0; ////////////////////////////////////////// @@ -86,4 +95,8 @@ { // Parameters checking + if ( (NN != FBUF_X_SIZE) || (NN != FBUF_Y_SIZE) ) + { + giet_exit("[TRANSPOSE ERROR] Frame buffer size does not fit image size"); + } if ((NB_PROCS_MAX != 1) && (NB_PROCS_MAX != 2) && (NB_PROCS_MAX != 4)) { @@ -149,10 +162,11 @@ giet_shr_printf("\n[TRANSPOSE] Proc [0,0,0] open file misc/images.raw\n"); } - init_ok = 0; + init_ok = 1; } else // others processors wait initialisation completion { - while ( init_ok == 1 ); - giet_shr_printf("\n[TRANSPOSE] Processor[%d,%d,%d] starts at cycle %d\n", x, y, lpid); + while ( init_ok == 0 ); + giet_shr_printf("\n[TRANSPOSE] Processor[%d,%d,%d] starts at cycle %d\n", + x, y, lpid, giet_proctime() ); } @@ -247,7 +261,7 @@ unsigned int npt = npixels / ntasks; // number of pixels per task - giet_fb_sync_write( npt * task_id, - &buf_out[cluster_id][lpid*npt], - npt ); + giet_fbf_sync_write( npt * task_id, + &buf_out[cluster_id][lpid*npt], + npt ); if ( lpid == 0 ) Index: /soft/giet_vm/applications/transpose/transpose.py =================================================================== --- /soft/giet_vm/applications/transpose/transpose.py (revision 443) +++ /soft/giet_vm/applications/transpose/transpose.py (revision 444) @@ -25,5 +25,5 @@ x_size = mapping.x_size y_size = mapping.y_size - nprocs = mapping.nprocs + nprocs = mapping.nprocs x_width = mapping.x_width y_width = mapping.y_width