Index: /soft/giet_vm/transpose/main.c =================================================================== --- /soft/giet_vm/transpose/main.c (revision 316) +++ /soft/giet_vm/transpose/main.c (revision 317) @@ -100,12 +100,17 @@ } - barrier_init( &barrier_0, ntasks ); - barrier_init( &barrier_1, ntasks ); - barrier_init( &barrier_2, ntasks ); - barrier_init( &barrier_3, ntasks ); - barrier_init( &barrier_4, ntasks ); - - giet_shr_printf("\n*** Proc [%d,%d,%d] completes barrier init at cycle %d\n", + if ( ntasks > 1 ) + { + barrier_init( &barrier_0, ntasks ); + barrier_init( &barrier_1, ntasks ); + barrier_init( &barrier_2, ntasks ); + barrier_init( &barrier_3, ntasks ); + barrier_init( &barrier_4, ntasks ); + } + + giet_shr_printf("\n[TRANSPOSE] Proc [%d,%d,%d] completes barrier init at cycle %d\n", 0, 0, 0, giet_proctime() ); + giet_shr_printf(" - nclusters = %d\n", nclusters ); + giet_shr_printf(" - ntasks = %d\n", ntasks ); init_ok = 0; @@ -124,5 +129,5 @@ // get heap vaddr in cluster[0,0] unsigned int heap_base; - giet_vobj_get_vbase( "transpose", "heap_0_0", &heap_base ); + giet_vobj_get_vbase( "transpose", "trsp_heap_0_0", &heap_base ); // allocate buffers in cluster[x,y] @@ -130,5 +135,5 @@ buf_out[cluster_id] = buf_in[cluster_id] + NN*NN/nclusters; - giet_shr_printf("\n*** Proc [%d,%d,%d] completes buffer allocation at cycle %d\n" + giet_shr_printf("\n[TRANSPOSE] Proc [%d,%d,%d] completes buffer allocation at cycle %d\n" " - buf_in = %x\n" " - buf_out = %x\n", @@ -147,8 +152,14 @@ giet_exit(" open() failure"); } + else + { + giet_shr_printf("\n[TRANSPOSE] Processor[%d,%d,%d]" + " open file misc/images.raw\n", + x, y, lpid ); + } } - /////////////////////////// - barrier_wait( &barrier_0 ); + ///////////////////////////////////////////// + if ( ntasks > 1 ) barrier_wait( &barrier_0 ); // Main loop (on images) @@ -167,5 +178,6 @@ ((image*nblocks) + ((nblocks*cluster_id)/nclusters)) ); - giet_shr_printf("\n*** Proc [%d,%d,0] completes load for image %d at cycle %d\n", + giet_shr_printf("\n[TRANSPOSE] Proc [%d,%d,0] completes load" + " for image %d at cycle %d\n", x, y, image, giet_proctime() ); } @@ -173,6 +185,6 @@ LOAD_END[cluster_id][lpid] = giet_proctime(); - /////////////////////////// - barrier_wait( &barrier_1 ); + ///////////////////////////////////////////// + if ( ntasks > 1 ) barrier_wait( &barrier_1 ); // parallel transpose from buf_in to buf_out @@ -221,12 +233,13 @@ if ( lpid == 0 ) { - giet_shr_printf("\n*** proc [%d,%d,0] completes transpose for image %d at cycle %d\n", - x, y, image, giet_proctime() ); + giet_shr_printf("\n[TRANSPOSE] proc [%d,%d,0] completes transpose" + " for image %d at cycle %d\n", + x, y, image, giet_proctime() ); } TRSP_END[cluster_id][lpid] = giet_proctime(); - /////////////////////////// - barrier_wait( &barrier_2 ); + ///////////////////////////////////////////// + if ( ntasks > 1 ) barrier_wait( &barrier_2 ); // optional parallel display from local buf_out to frame buffer @@ -245,5 +258,6 @@ if ( lpid == 0 ) { - giet_shr_printf("\n*** Proc [%d,%d,0] completes display for image %d at cycle %d\n", + giet_shr_printf("\n[TRANSPOSE] Proc [%d,%d,0] completes display" + " for image %d at cycle %d\n", x, y, image, giet_proctime() ); } @@ -251,6 +265,6 @@ DISP_END[cluster_id][lpid] = giet_proctime(); - /////////////////////////// - barrier_wait( &barrier_3 ); + ///////////////////////////////////////////// + if ( ntasks > 1 ) barrier_wait( &barrier_3 ); } @@ -283,10 +297,12 @@ if ( success ) { - giet_shr_printf("\n*** proc [%d,%d,0] : checksum OK for image %d\n", + giet_shr_printf("\n[TRANSPOSE] proc [%d,%d,0] checksum OK" + " for image %d\n", x, y, image ); } else { - giet_shr_printf("\n*** proc [%d,%d,0] : checksum KO for image %d\n", + giet_shr_printf("\n[TRANSPOSE] proc [%d,%d,0] checksum KO" + " for image %d\n", x, y, image ); } @@ -360,5 +376,5 @@ // all tasks wait instrumentation completion ////////////////////////////////////////////////// - barrier_wait( &barrier_4 ); + if ( ntasks > 1 ) barrier_wait( &barrier_4 ); } // end while image Index: /soft/giet_vm/transpose/transpose.py =================================================================== --- /soft/giet_vm/transpose/transpose.py (revision 317) +++ /soft/giet_vm/transpose/transpose.py (revision 317) @@ -0,0 +1,105 @@ +#!/usr/bin/env python + +from mapping import * + +###################################################################################### +# file : transpose.py +# date : may 2014 +# author : Alain Greiner +####################################################################################### +# This file describes the mapping of the multi-threaded "transpose" +# application on a multi-clusters, multi-processors architecture. +# This include both the mapping of virtual segments on the clusters, +# and the mapping of tasks on processors. +# This mapping uses 5 platform parameters, (obtained from the "mapping" argument) +# - x_size : number of clusters in a row +# - y_size : number of clusters in a column +# - x_width : number of bits coding x coordinate +# - y_width : number of bits coding y coordinate +# - procs_max : number of processors per cluster +#################################################################################### + +######################### +def transpose( mapping ): + + x_size = mapping.x_size + y_size = mapping.y_size + procs_max = mapping.procs_max + 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 + + ptab_base = 0x30000000 + ptab_size = 0x00040000 # 256 Kbytes + + stack_base = 0x40000000 + stack_size = 0x00010000 # 64 Kbytes + + heap_base = 0x50000000 + heap_size = 0x00010000 # 64 Kbytes + + # create Vspace + vspace = mapping.addVspace( name = 'transpose', startname = 'trsp_data' ) + + # non replicated vsegs in cluster[0,0] + mapping.addVseg( vspace, 'trsp_code', code_base , code_size, 'CXWU', vtype = 'ELF', + x = 0, y = 0, pseg = 'RAM', binpath = 'build/transpose/transpose.elf' ) + + mapping.addVseg( vspace, 'trsp_data', data_base , data_size, 'C_WU', vtype = 'ELF', + x = 0, y = 0, pseg = 'RAM', binpath = 'build/transpose/transpose.elf' ) + + mapping.addVseg( vspace, 'trsp_ptab', ptab_base , ptab_size, 'C_WU', vtype = 'PTAB', + x = 0, y = 0, pseg = 'RAM', align = 13 ) + + # distributed vsegs: one stack per processor/task, one heap per cluster + for x_rep in xrange (x_size): + for y_rep in xrange (y_size): + cluster_offset = ((x_rep << y_width) + y_rep) << 20 # 1 Mbytes per cluster + mapping.addVseg( vspace, 'trsp_heap_%d_%d' % (x_rep, y_rep), + heap_base + cluster_offset, heap_size, 'C_WU', + vtype = 'BUFFER', x = x_rep, y = y_rep, pseg = 'RAM' ) + + for p in xrange( procs_max ): + proc_offset = cluster_offset + (p << 18) # 256 Kbytes per proc + mapping.addVseg( vspace, 'trsp_stack_%d_%d_%d' % (x_rep, y_rep, p), + stack_base + proc_offset, stack_size, 'C_WU', + vtype = 'BUFFER', x = x_rep, y = y_rep, pseg = 'RAM' ) + + # distributed tasks / one task per processor + for x in xrange (x_size): + for y in xrange (y_size): + for p in xrange( procs_max ): + + trdid = (((x * y_size) + y) * procs_max) + p + mapping.addTask( vspace, 'sort_%d_%d_%d' % (x,y,p), trdid, x, y, p, + 'trsp_stack_%d_%d_%d' % (x,y,p), + 'trsp_heap_%d_%d' % (x,y), 0 ) + + # extend mapping name + mapping.name += '_transpose' + + return vspace # useful for test + +################################ test ###################################################### + +if __name__ == '__main__': + + vspace = transpose( 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 +