Index: /soft/giet_vm/applications/classif/classif.ld =================================================================== --- /soft/giet_vm/applications/classif/classif.ld (revision 472) +++ /soft/giet_vm/applications/classif/classif.ld (revision 473) @@ -1,27 +1,8 @@ /**************************************************************************** -* Definition of the base address for all virtual segments +* Definition of the base address for code and data vsegs *****************************************************************************/ seg_code_base = 0x10000000; - -seg_data_0_0_base = 0x20000000; -seg_data_0_1_base = 0x20010000; -seg_data_0_2_base = 0x20020000; -seg_data_0_3_base = 0x20030000; - -seg_data_1_0_base = 0x20040000; -seg_data_1_1_base = 0x20050000; -seg_data_1_2_base = 0x20060000; -seg_data_1_3_base = 0x20070000; - -seg_data_2_0_base = 0x20080000; -seg_data_2_1_base = 0x20090000; -seg_data_2_2_base = 0x200A0000; -seg_data_2_3_base = 0x200B0000; - -seg_data_3_0_base = 0x200C0000; -seg_data_3_1_base = 0x200D0000; -seg_data_3_2_base = 0x200E0000; -seg_data_3_3_base = 0x200F0000; +seg_data_base = 0x20000000; /*************************************************************************** @@ -37,6 +18,6 @@ } - . = seg_data_0_0_base; - seg_data_0_0 : + . = seg_data_base; + seg_data : { *(.ctors) @@ -51,83 +32,4 @@ *(.sbss) *(.scommon) - *(.data_0_0) - } - . = seg_data_0_1_base; - seg_data_0_1 : - { - *(.data_0_1) - } - . = seg_data_0_2_base; - seg_data_0_2 : - { - *(.data_0_2) - } - . = seg_data_0_3_base; - seg_data_0_3 : - { - *(.data_0_3) - } - - . = seg_data_1_0_base; - seg_data_1_0 : - { - *(.data_1_0) - } - . = seg_data_1_1_base; - seg_data_1_1 : - { - *(.data_1_1) - } - . = seg_data_1_2_base; - seg_data_1_2 : - { - *(.data_1_2) - } - . = seg_data_1_3_base; - seg_data_1_3 : - { - *(.data_1_3) - } - - . = seg_data_2_0_base; - seg_data_2_0 : - { - *(.data_2_0) - } - . = seg_data_2_1_base; - seg_data_2_1 : - { - *(.data_2_1) - } - . = seg_data_2_2_base; - seg_data_2_2 : - { - *(.data_2_2) - } - . = seg_data_2_3_base; - seg_data_2_3 : - { - *(.data_2_3) - } - - . = seg_data_3_0_base; - seg_data_3_0 : - { - *(.data_3_0) - } - . = seg_data_3_1_base; - seg_data_3_1 : - { - *(.data_3_1) - } - . = seg_data_3_2_base; - seg_data_3_2 : - { - *(.data_3_2) - } - . = seg_data_3_3_base; - seg_data_3_3 : - { - *(.data_3_3) } } Index: /soft/giet_vm/applications/classif/classif.py =================================================================== --- /soft/giet_vm/applications/classif/classif.py (revision 472) +++ /soft/giet_vm/applications/classif/classif.py (revision 473) @@ -15,5 +15,6 @@ # The mapping of virtual segments on the clusters is the following: # - The code vsegs are replicated on all clusters. -# - There is one shared data vseg per cluster. +# - There is one shared data vseg in cluster[0][0] +# - There is one heap vseg per cluster. # - The stacks vsegs are distibuted on all clusters. # This mapping uses 5 platform parameters, (obtained from the "mapping" argument) @@ -37,5 +38,4 @@ y_width = mapping.y_width - assert (x_size <= 4) and (y_size <= 4) assert (nprocs >= 2) @@ -45,5 +45,8 @@ data_base = 0x20000000 - data_size = 0x00010000 # 64 Kbytes (per cluster) + data_size = 0x00010000 # 64 Kbytes + + heap_base = 0x30000000 + heap_size = 0x00008000 # 32 Kbytes (per cluster) stack_base = 0x40000000 @@ -51,14 +54,19 @@ # create vspace - vspace = mapping.addVspace( name = 'classif', startname = 'classif_data_0_0' ) + vspace = mapping.addVspace( name = 'classif', startname = 'classif_data' ) - # data_x_y vsegs : shared / one per cluster + # data vseg : shared / cluster[0][0] + mapping.addVseg( vspace, 'classif_data', data_base , data_size, + 'C_WU', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM', + binpath = 'build/classif/classif.elf', + local = False ) + + # heap_x_y vsegs : shared / one per cluster for x in xrange (x_size): for y in xrange (y_size): - base = data_base + ( (4*x + y) * data_size ) + base = heap_base + ( (4*x + y) * heap_size ) - mapping.addVseg( vspace, 'classif_data_%d_%d' %(x,y), base , data_size, - 'C_WU', vtype = 'ELF', x = x, y = y, pseg = 'RAM', - binpath = 'build/classif/classif.elf', + mapping.addVseg( vspace, 'classif_heap_%d_%d' %(x,y), base , heap_size, + 'C_WU', vtype = 'HEAP', x = x, y = y, pseg = 'RAM', local = False ) @@ -94,8 +102,10 @@ else : # task analyse task_index = 1 - task_name = 'analyse_%d_%d_%d' % (x,y,p -) + task_name = 'analyse_%d_%d_%d' % (x,y,p) + mapping.addTask( vspace, task_name, trdid, x, y, p, - 'classif_stack_%d_%d_%d' % (x,y,p), '' , task_index ) + 'classif_stack_%d_%d_%d' % (x,y,p), + 'classif_heap_%d_%d' % (x,y), + task_index ) # extend mapping name Index: /soft/giet_vm/applications/classif/main.c =================================================================== --- /soft/giet_vm/applications/classif/main.c (revision 472) +++ /soft/giet_vm/applications/classif/main.c (revision 473) @@ -17,11 +17,19 @@ // as global variables distributed in (up to) 16 clusters. // -// WARNING: the platform cannot contain more than 16 clusters: (X_SIZE < 4) && (Y_SIZE < 4) -// -// 1) The "load" task transfer one container from the kernel chbuf associated to the -// NIC_RX channel, to a private buffer. Then it copies this bufferer to the local MWMR fifo. +// Initialisation is done in two steps by the "load" tasks: +// - Task "load" in cluster[0][0] initialises NIC & CMA channel, and initialises +// the barrier between all "load" tasks. Other "load" tasks are waiting on the +// global_sync synchronisation variable. +// - In each cluster[x][y], the load" task allocates the MWMR fifo descriptor & the data +// buffer in the local heap, and store the pointers on a global array of pointers. +// The "analyse" tasks are waiting on the sync[x][y] variables. +// +// When initialisation is completed, all "load and "analyse" tasks loop on containers: +// 1) The "load" task transfer containers from the kernel chbuf associated to the +// NIC_RX channel (in cluster[0][0]), to the local MWMR fifo (in cluster[x][y]), +// after an in termediate copy in a private stack buffer. // Each "load" task loads CONTAINERS_MAX containers before exit, and the // task in cluster[0,0] displays the results stored in global counters filled -// by the "analyse" tasks. +// by the "analyse" tasks when all "load" tasks reach the barrier. // // 2) The "analyse" task transfer one container from the local MWMR fifo to a private @@ -47,105 +55,25 @@ #define CONTAINERS_MAX 10 -/////////// distributed data ///////////////////////////////////////////////////////////// -// - fifo_x_y is the local MWMR fifo descriptor -// - data_x_y is the local MWMR fifo data buffer -// - sync_x_y is the local variable signaling MWMR fifo initialisation -/////////// distributed data ///////////////////////////////////////////////////////////// - -#if ( (X_SIZE > 0) && (Y_SIZE > 0) ) -__attribute__((section (".data_0_0"))) mwmr_channel_t fifo_0_0; -__attribute__((section (".data_0_0"))) unsigned int data_0_0[NB_PROCS_MAX<<10]; -__attribute__((section (".data_0_0"))) unsigned int sync_0_0 = 0; -#endif -#if ( (X_SIZE > 0) && (Y_SIZE > 1) ) -__attribute__((section (".data_0_1"))) mwmr_channel_t fifo_0_1; -__attribute__((section (".data_0_1"))) unsigned int data_0_1[NB_PROCS_MAX<<10]; -__attribute__((section (".data_0_1"))) unsigned int sync_0_1 = 0; -#endif -#if ( (X_SIZE > 0) && (Y_SIZE > 2) ) -__attribute__((section (".data_0_2"))) mwmr_channel_t fifo_0_2; -__attribute__((section (".data_0_2"))) unsigned int data_0_2[NB_PROCS_MAX<<10]; -__attribute__((section (".data_0_2"))) unsigned int sync_0_2 = 0; -#endif -#if ( (X_SIZE > 0) && (Y_SIZE > 3) ) -__attribute__((section (".data_0_3"))) mwmr_channel_t fifo_0_3; -__attribute__((section (".data_0_3"))) unsigned int data_0_3[NB_PROCS_MAX<<10]; -__attribute__((section (".data_0_3"))) unsigned int sync_0_3 = 0; -#endif -#if ( (X_SIZE > 1) && (Y_SIZE > 0) ) -__attribute__((section (".data_1_0"))) mwmr_channel_t fifo_1_0; -__attribute__((section (".data_1_0"))) unsigned int data_1_0[NB_PROCS_MAX<<10]; -__attribute__((section (".data_1_0"))) unsigned int sync_1_0 = 0; -#endif -#if ( (X_SIZE > 1) && (Y_SIZE > 1) ) -__attribute__((section (".data_1_1"))) mwmr_channel_t fifo_1_1; -__attribute__((section (".data_1_1"))) unsigned int data_1_1[NB_PROCS_MAX<<10]; -__attribute__((section (".data_1_1"))) unsigned int sync_1_1 = 0; -#endif -#if ( (X_SIZE > 1) && (Y_SIZE > 2) ) -__attribute__((section (".data_1_2"))) mwmr_channel_t fifo_1_2; -__attribute__((section (".data_1_2"))) unsigned int data_1_2[NB_PROCS_MAX<<10]; -__attribute__((section (".data_1_2"))) unsigned int sync_1_2 = 0; -#endif -#if ( (X_SIZE > 1) && (Y_SIZE > 3) ) -__attribute__((section (".data_1_3"))) mwmr_channel_t fifo_1_3; -__attribute__((section (".data_1_3"))) unsigned int data_1_3[NB_PROCS_MAX<<10]; -__attribute__((section (".data_1_3"))) unsigned int sync_1_3 = 0; -#endif -#if ( (X_SIZE > 2) && (Y_SIZE > 0) ) -__attribute__((section (".data_2_0"))) mwmr_channel_t fifo_2_0; -__attribute__((section (".data_2_0"))) unsigned int data_2_0[NB_PROCS_MAX<<10]; -__attribute__((section (".data_2_0"))) unsigned int sync_2_0 = 0; -#endif -#if ( (X_SIZE > 2) && (Y_SIZE > 1) ) -__attribute__((section (".data_2_1"))) mwmr_channel_t fifo_2_1; -__attribute__((section (".data_2_1"))) unsigned int data_2_1[NB_PROCS_MAX<<10]; -__attribute__((section (".data_2_1"))) unsigned int sync_2_1 = 0; -#endif -#if ( (X_SIZE > 2) && (Y_SIZE > 2) ) -__attribute__((section (".data_2_2"))) mwmr_channel_t fifo_2_2; -__attribute__((section (".data_2_2"))) unsigned int data_2_2[NB_PROCS_MAX<<10]; -__attribute__((section (".data_2_2"))) unsigned int sync_2_2 = 0; -#endif -#if ( (X_SIZE > 2) && (Y_SIZE > 3) ) -__attribute__((section (".data_2_3"))) mwmr_channel_t fifo_2_3; -__attribute__((section (".data_2_3"))) unsigned int data_2_3[NB_PROCS_MAX<<10]; -__attribute__((section (".data_2_3"))) unsigned int sync_2_3 = 0; -#endif -#if ( (X_SIZE > 3) && (Y_SIZE > 0) ) -__attribute__((section (".data_3_0"))) mwmr_channel_t fifo_3_0; -__attribute__((section (".data_3_0"))) unsigned int data_3_0[NB_PROCS_MAX<<10]; -__attribute__((section (".data_3_0"))) unsigned int sync_3_0 = 0; -#endif -#if ( (X_SIZE > 3) && (Y_SIZE > 1) ) -__attribute__((section (".data_3_1"))) mwmr_channel_t fifo_3_1; -__attribute__((section (".data_3_1"))) unsigned int data_3_1[NB_PROCS_MAX<<10]; -__attribute__((section (".data_3_1"))) unsigned int sync_3_1 = 0; -#endif -#if ( (X_SIZE > 3) && (Y_SIZE > 2) ) -__attribute__((section (".data_3_2"))) mwmr_channel_t fifo_3_2; -__attribute__((section (".data_3_2"))) unsigned int data_3_2[NB_PROCS_MAX<<10]; -__attribute__((section (".data_3_2"))) unsigned int sync_3_2 = 0; -#endif -#if ( (X_SIZE > 3) && (Y_SIZE > 3) ) -__attribute__((section (".data_3_3"))) mwmr_channel_t fifo_3_3; -__attribute__((section (".data_3_3"))) unsigned int data_3_3[NB_PROCS_MAX<<10]; -__attribute__((section (".data_3_3"))) unsigned int sync_3_3 = 0; -#endif - -/////////// shared variables in cluster[0,0] ////////////////////////// - -__attribute__((section (".data_0_0"))) unsigned int count[16]; - -__attribute__((section (".data_0_0"))) giet_barrier_t barrier; - -__attribute__((section (".data_0_0"))) unsigned int global_init_ok = 0; - -__attribute__((section (".data_0_0"))) unsigned int nic_channel; - - -///////////////////////////////////////// +/////////////////////////////////////////////////////////////////////////////////////////// +// Global variables +// The communication channels are distributed in the clusters, +// but the pointers arrays are global variables in cluster[0][0] +/////////////////////////////////////////////////////////////////////////////////////////// + +mwmr_channel_t* mwmr[X_SIZE][Y_SIZE]; // distributed MWMR fifos pointers + +unsigned int local_sync[X_SIZE][Y_SIZE]; // distributed synchros "load" / "analyse" + +unsigned int global_sync = 0; // global synchro between "load" tasks + +unsigned int count[16]; // instrumentation counters + +giet_barrier_t barrier; // barrier between "load" (instrumentation) + +unsigned int nic_channel; // allocated NIC channel index + +/////////////////////////////////////////////////////////////////////////////////////////// __attribute__ ((constructor)) void load() -///////////////////////////////////////// +/////////////////////////////////////////////////////////////////////////////////////////// { // get processor identifiers @@ -158,18 +86,33 @@ if (Y_SIZE > 4 ) giet_exit("The Y_SIZE parameter cannot be larger than 4\n"); - // local buffer to store one container + // local buffer to store one container in private stack unsigned int temp[1024]; - // get pointer on local MWMR fifo descriptor and data buffer - unsigned int offset = ((x * 4) + y) * 0x10000; - mwmr_channel_t* fifo = (mwmr_channel_t*)(((unsigned int)&fifo_0_0) + offset); - unsigned int* data = (unsigned int*) (((unsigned int)data_0_0) + offset); - unsigned int* sync = (unsigned int*) (((unsigned int)&sync_0_0) + offset); - +// giet_shr_printf("\n@@@ P[%d,%d,%d] enters load task at cycle %d\n", +// x, y, l, giet_proctime() ); + + // allocates data buffer for MWMR fifo in local heap + unsigned int* data = malloc( NB_PROCS_MAX<<12 ); + +//giet_shr_printf("\n@@@ P[%d,%d,%d] completes data malloc at cycle %d " +// "/ &data = %x\n", x, y, l, giet_proctime(), (unsigned int)data ); + + // allocates MWMR fifo descriptor in local heap + mwmr_channel_t* fifo = malloc( sizeof(mwmr_channel_t) ); + +//giet_shr_printf("\n@@@ P[%d,%d,%d] completes mwmr malloc at cycle %d " +// "/ &mwmr = %x\n", x, y, l, giet_proctime(), (unsigned int)mwmr ); + + // makes copy of pointer in global array for "analyse" tasks + mwmr[x][y] = fifo; + + // display status for cluster[X_SIZE-1][Y_SIZE-1] if ( (x==X_SIZE-1) && (y==Y_SIZE-1) ) giet_shr_printf("\n*** Task load starts on P[%d,%d,%d] at cycle %d\n" - " &fifo = %x / &data = %x / &sync = %x\n", - x, y, l, giet_proctime(), - (unsigned int)fifo, (unsigned int)data, (unsigned int)sync ); + " &fifo = %x / &data = %x / &sync = %x\n" + " x_size = %d / y_size = %d / nprocs = %d\n", + x , y , l , giet_proctime() , + (unsigned int)fifo , (unsigned int)data, (unsigned int)(&local_sync[x][y]) , + X_SIZE, Y_SIZE, NB_PROCS_MAX ); // Task load on cluster[0,0] makes global initialisation: @@ -188,9 +131,12 @@ barrier_init( &barrier, X_SIZE * Y_SIZE ); - global_init_ok = 1; + // clear NIC RX channels stats + giet_nic_rx_clear(); + + global_sync = 1; } else { - while ( global_init_ok == 0 ) asm volatile ("nop"); + while ( global_sync == 0 ) asm volatile ("nop"); } @@ -198,6 +144,6 @@ mwmr_init( fifo , data , 1024 , NB_PROCS_MAX ); - // signal MWMR fifo initialisation completion - *sync = 1; + // signal MWMR fifo initialisation completion to analyse tasks + local_sync[x][y] = 1; // main loop (on containers) @@ -226,7 +172,9 @@ barrier_wait( &barrier ); - // Task load in cluster[0,0] displays counters and stops NIC / CMA transfer + // Task load in cluster[0,0] stops NIC and displays results if ( (x==0) && (y==0) ) { + giet_nic_rx_stop(); + giet_shr_printf("\n@@@@ Clasification Results @@@\n" " - TYPE 0 : %d packets\n" @@ -246,10 +194,8 @@ " - TYPE E : %d packets\n" " - TYPE F : %d packets\n", - count[0x0], count[0x1], count[0x2], count[0x3], - count[0x4], count[0x5], count[0x6], count[0x7], - count[0x8], count[0x9], count[0xA], count[0xB], - count[0xC], count[0xD], count[0xE], count[0xF] ); - - giet_nic_rx_stop(); + count[0x0], count[0x1], count[0x2], count[0x3], + count[0x4], count[0x5], count[0x6], count[0x7], + count[0x8], count[0x9], count[0xA], count[0xB], + count[0xC], count[0xD], count[0xE], count[0xF] ); giet_nic_rx_stats(); @@ -277,19 +223,18 @@ unsigned int temp[1024]; + // wait MWMR channel initialisation (done by task load) + while ( local_sync[x][y] == 0 ) asm volatile ("nop"); + // get pointer on MWMR channel descriptor - unsigned int offset = ((x * 4) + y) * 0x10000; - mwmr_channel_t* fifo = (mwmr_channel_t*)(((unsigned int)&fifo_0_0) + offset); - unsigned int* sync = (unsigned int*) (((unsigned int)&sync_0_0) + offset); - + mwmr_channel_t* fifo = mwmr[x][y]; + + // display status for cluster[X_SIZE-1][Y_SIZE-1] if ( (x==X_SIZE-1) && (y==Y_SIZE-1) ) giet_shr_printf("\n*** Task analyse starts on P[%d,%d,%d] at cycle %d\n" " &fifo = %x / &sync = %x\n", x, y, l, giet_proctime(), - (unsigned int)fifo, (unsigned int)sync ); + (unsigned int)fifo, (unsigned int)(&local_sync[x][y]) ); - // wait MWMR channel initialisation (done by task load) - while ( *sync == 0 ) asm volatile ("nop"); - - // infinite loop (on containers) + // main loop (on containers) unsigned int nwords; // number of words in container unsigned int npackets; // number of packets in container