X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/d2658ff1a37cc89f85a5fd92c5425e17f4045caa..c836edd2518e98d7666b73277ced612cc526f3f4:/src/simix/smx_context_raw.c diff --git a/src/simix/smx_context_raw.c b/src/simix/smx_context_raw.c index a7267ed0ea..e829551d13 100644 --- a/src/simix/smx_context_raw.c +++ b/src/simix/smx_context_raw.c @@ -6,8 +6,9 @@ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -#include "xbt/threadpool.h" #include "simix/private.h" +#include "xbt/parmap.h" + #ifdef HAVE_VALGRIND_VALGRIND_H # include @@ -20,12 +21,6 @@ #include "ucontext.h" #endif -/* lower this if you want to reduce the memory consumption */ -#ifndef CONTEXT_STACK_SIZE /* allow lua to override this */ -#define CONTEXT_STACK_SIZE 128*1024 -#endif /*CONTEXT_STACK_SIZE */ - - typedef char * raw_stack_t; typedef void (*rawctx_entry_point_t)(void *); @@ -33,38 +28,43 @@ typedef struct s_smx_ctx_raw { s_smx_ctx_base_t super; /* Fields of super implementation */ char *malloced_stack; /* malloced area containing the stack */ raw_stack_t stack_top; /* pointer to stack top (within previous area) */ + raw_stack_t old_stack_top; /* to whom I should return the control */ #ifdef HAVE_VALGRIND_VALGRIND_H unsigned int valgrind_stack_id; /* the valgrind stack id */ #endif } s_smx_ctx_raw_t, *smx_ctx_raw_t; -smx_ctx_raw_t maestro_context; +smx_ctx_raw_t maestro_raw_context; extern raw_stack_t raw_makecontext(char* malloced_stack, int stack_size, rawctx_entry_point_t entry_point, void* arg); -extern void raw_swapcontext(raw_stack_t* old, raw_stack_t* new); +extern void raw_swapcontext(raw_stack_t* old, raw_stack_t new); #ifdef PROCESSOR_i686 __asm__ ( -".globl raw_makecontext\n" -"raw_makecontext:\n" -" movl 4(%esp),%eax\n" /* stack */ -" addl 8(%esp),%eax\n" /* size */ -" movl 12(%esp),%ecx\n" /* func */ -" movl 16(%esp),%edx\n" /* arg */ -" movl %edx, -4(%eax)\n" -" movl $0, -8(%eax)\n" -" movl %ecx,-12(%eax)\n" -" movl $0, -16(%eax)\n" -" movl $0, -20(%eax)\n" -" movl $0, -24(%eax)\n" -" movl $0, -28(%eax)\n" -" subl $28,%eax\n" -" ret\n" + ".text\n" + ".globl raw_makecontext\n" + ".type raw_makecontext,@function\n" + "raw_makecontext:\n" + " movl 4(%esp),%eax\n" /* stack */ + " addl 8(%esp),%eax\n" /* size */ + " movl 12(%esp),%ecx\n" /* func */ + " movl 16(%esp),%edx\n" /* arg */ + " movl %edx, -4(%eax)\n" + " movl $0, -8(%eax)\n" /* @return for func */ + " movl %ecx,-12(%eax)\n" + " movl $0, -16(%eax)\n" /* ebp */ + " movl $0, -20(%eax)\n" /* ebx */ + " movl $0, -24(%eax)\n" /* esi */ + " movl $0, -28(%eax)\n" /* edi */ + " subl $28,%eax\n" + " retl\n" ); __asm__ ( + ".text\n" ".globl raw_swapcontext\n" + ".type raw_swapcontext,@function\n" "raw_swapcontext:\n" " movl 4(%esp),%eax\n" /* old */ " movl 8(%esp),%edx\n" /* new */ @@ -73,40 +73,111 @@ __asm__ ( " pushl %esi\n" " pushl %edi\n" " movl %esp,(%eax)\n" - " movl (%edx),%esp\n" + " movl %edx,%esp\n" " popl %edi\n" " popl %esi\n" " popl %ebx\n" " popl %ebp\n" - " ret\n" + " retl\n" +); +#elif PROCESSOR_x86_64 +__asm__ ( + ".text\n" + ".globl raw_makecontext\n" + ".type raw_makecontext,@function\n" + "raw_makecontext:\n" /* Calling convention sets the arguments in rdi, rsi, rdx and rcx, respectively */ + " movq %rdi,%rax\n" /* stack */ + " addq %rsi,%rax\n" /* size */ + " movq $0, -8(%rax)\n" /* @return for func */ + " movq %rdx,-16(%rax)\n" /* func */ + " movq %rcx,-24(%rax)\n" /* arg/rdi */ + " movq $0, -32(%rax)\n" /* rsi */ + " movq $0, -40(%rax)\n" /* rdx */ + " movq $0, -48(%rax)\n" /* rcx */ + " movq $0, -56(%rax)\n" /* r8 */ + " movq $0, -64(%rax)\n" /* r9 */ + " movq $0, -72(%rax)\n" /* rbp */ + " movq $0, -80(%rax)\n" /* rbx */ + " movq $0, -88(%rax)\n" /* r12 */ + " movq $0, -96(%rax)\n" /* r13 */ + " movq $0, -104(%rax)\n" /* r14 */ + " movq $0, -112(%rax)\n" /* r15 */ + " subq $112,%rax\n" + " retq\n" +); + +__asm__ ( + ".text\n" + ".globl raw_swapcontext\n" + ".type raw_swapcontext,@function\n" + "raw_swapcontext:\n" /* Calling convention sets the arguments in rdi and rsi, respectively */ + " pushq %rdi\n" + " pushq %rsi\n" + " pushq %rdx\n" + " pushq %rcx\n" + " pushq %r8\n" + " pushq %r9\n" + " pushq %rbp\n" + " pushq %rbx\n" + " pushq %r12\n" + " pushq %r13\n" + " pushq %r14\n" + " pushq %r15\n" + " movq %rsp,(%rdi)\n" /* old */ + " movq %rsi,%rsp\n" /* new */ + " popq %r15\n" + " popq %r14\n" + " popq %r13\n" + " popq %r12\n" + " popq %rbx\n" + " popq %rbp\n" + " popq %r9\n" + " popq %r8\n" + " popq %rcx\n" + " popq %rdx\n" + " popq %rsi\n" + " popq %rdi\n" + " retq\n" ); #else -#error "Cannot compile raw contexts for your architecture" +raw_stack_t raw_makecontext(char* malloced_stack, int stack_size, + rawctx_entry_point_t entry_point, void* arg) { + THROW_UNIMPLEMENTED; +} + +void raw_swapcontext(raw_stack_t* old, raw_stack_t new) { + THROW_UNIMPLEMENTED; +} + #endif XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_context); -static xbt_tpool_t tpool; +#ifdef CONTEXT_THREADS +static xbt_parmap_t parmap; +#endif -static void smx_ctx_raw_wrapper(smx_ctx_raw_t context); +static smx_context_factory_t raw_factory; +static void smx_ctx_raw_wrapper(smx_ctx_raw_t context); static int smx_ctx_raw_factory_finalize(smx_context_factory_t *factory) { - if(tpool) - xbt_tpool_destroy(tpool); + if(parmap) + xbt_parmap_destroy(parmap); return smx_ctx_base_factory_finalize(factory); } + static smx_context_t -smx_ctx_raw_create_context_sized(size_t size, xbt_main_func_t code, - int argc, char **argv, +smx_ctx_raw_create_context(xbt_main_func_t code, int argc, char **argv, void_pfn_smxprocess_t cleanup_func, void *data) { smx_ctx_raw_t context = - (smx_ctx_raw_t) smx_ctx_base_factory_create_context_sized(size, + (smx_ctx_raw_t) smx_ctx_base_factory_create_context_sized( + sizeof(s_smx_ctx_raw_t), code, argc, argv, @@ -116,35 +187,22 @@ smx_ctx_raw_create_context_sized(size_t size, xbt_main_func_t code, /* If the user provided a function for the process then use it otherwise is the context for maestro */ if (code) { - context->malloced_stack = xbt_malloc0(CONTEXT_STACK_SIZE); + context->malloced_stack = xbt_malloc0(smx_context_stack_size); context->stack_top = - raw_makecontext(context->malloced_stack,CONTEXT_STACK_SIZE, + raw_makecontext(context->malloced_stack, smx_context_stack_size, (void(*)(void*))smx_ctx_raw_wrapper,context); #ifdef HAVE_VALGRIND_VALGRIND_H context->valgrind_stack_id = VALGRIND_STACK_REGISTER(context->malloced_stack, - context->malloced_stack + CONTEXT_STACK_SIZE); + context->malloced_stack + smx_context_stack_size); #endif /* HAVE_VALGRIND_VALGRIND_H */ }else{ - maestro_context = context; + maestro_raw_context = context; } return (smx_context_t) context; - -} - -static smx_context_t -smx_ctx_raw_create_context(xbt_main_func_t code, int argc, char **argv, - void_pfn_smxprocess_t cleanup_func, - void *data) -{ - - return smx_ctx_raw_create_context_sized(sizeof(s_smx_ctx_raw_t), - code, argc, argv, cleanup_func, - data); - } static void smx_ctx_raw_free(smx_context_t context) @@ -157,14 +215,17 @@ static void smx_ctx_raw_free(smx_context_t context) context)->valgrind_stack_id); #endif /* HAVE_VALGRIND_VALGRIND_H */ + free(((smx_ctx_raw_t)context)->malloced_stack); } smx_ctx_base_free(context); } static void smx_ctx_raw_suspend(smx_context_t context) { - smx_current_context = (smx_context_t)maestro_context; - raw_swapcontext(&((smx_ctx_raw_t) context)->stack_top, &maestro_context->stack_top); + smx_current_context = (smx_context_t)maestro_raw_context; + raw_swapcontext( + &((smx_ctx_raw_t) context)->stack_top, + ((smx_ctx_raw_t) context)->old_stack_top); } static void smx_ctx_raw_stop(smx_context_t context) @@ -180,43 +241,61 @@ static void smx_ctx_raw_wrapper(smx_ctx_raw_t context) smx_ctx_raw_stop((smx_context_t) context); } -static void smx_ctx_raw_resume(smx_context_t context) +static void smx_ctx_raw_resume(smx_process_t process) { - smx_current_context = context; - raw_swapcontext(&maestro_context->stack_top, &((smx_ctx_raw_t) context)->stack_top); + smx_ctx_raw_t context = (smx_ctx_raw_t)process->context; + smx_current_context = (smx_context_t)context; + raw_swapcontext( + &((smx_ctx_raw_t) context)->old_stack_top, + ((smx_ctx_raw_t) context)->stack_top); } -static void smx_ctx_raw_runall(xbt_swag_t processes) + +static void smx_ctx_raw_runall_serial(xbt_dynar_t processes) { smx_process_t process; + unsigned int cursor; - while ((process = xbt_swag_extract(processes))) - smx_ctx_raw_resume(process->context); + xbt_dynar_foreach(processes, cursor, process) { + DEBUG2("Schedule item %u of %lu",cursor,xbt_dynar_length(processes)); + smx_ctx_raw_resume(process); + } + xbt_dynar_reset(processes); } -static void smx_ctx_raw_resume_parallel(smx_context_t context) +static void smx_ctx_raw_runall_parallel(xbt_dynar_t processes) { - xbt_os_thread_set_extra_data(context); - raw_swapcontext(&maestro_context->stack_top, &((smx_ctx_raw_t) context)->stack_top); + xbt_parmap_apply(parmap, (void_f_pvoid_t)smx_ctx_raw_resume, processes); + xbt_dynar_reset(processes); } -static void smx_ctx_raw_runall_parallel(xbt_swag_t processes) +static smx_context_t smx_ctx_raw_self_parallel(void) { - smx_process_t process; - while((process = xbt_swag_extract(processes))){ - xbt_tpool_queue_job(tpool, (void_f_pvoid_t)smx_ctx_raw_resume_parallel, process->context); - } - xbt_tpool_wait_all(tpool); + return smx_current_context; } -static smx_context_t smx_ctx_raw_self_parallel(void) +static int smx_ctx_raw_get_thread_id(){ + return (int)(unsigned long)xbt_os_thread_get_extra_data(); +} + +static void smx_ctx_raw_runall(xbt_dynar_t processes) { - smx_context_t self_context = (smx_context_t) xbt_os_thread_get_extra_data(); - return self_context ? self_context : (smx_context_t) maestro_context; + if (xbt_dynar_length(processes) >= SIMIX_context_get_parallel_threshold()) { + DEBUG1("Runall // %lu", xbt_dynar_length(processes)); + raw_factory->self = smx_ctx_raw_self_parallel; + raw_factory->get_thread_id = smx_ctx_raw_get_thread_id; + smx_ctx_raw_runall_parallel(processes); + } else { + DEBUG1("Runall serial %lu", xbt_dynar_length(processes)); + raw_factory->self = smx_ctx_base_self; + raw_factory->get_thread_id = smx_ctx_base_get_thread_id; + smx_ctx_raw_runall_serial(processes); + } } void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory) { + VERB0("Using raw contexts. Because the glibc is just not good enough for us."); smx_ctx_base_factory_init(factory); (*factory)->finalize = smx_ctx_raw_factory_finalize; @@ -227,15 +306,26 @@ void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory) (*factory)->suspend = smx_ctx_raw_suspend; (*factory)->name = "smx_raw_context_factory"; - if(_surf_parallel_contexts){ -#ifdef CONTEXT_THREADS /* To use parallel ucontexts a thread pool is needed */ - tpool = xbt_tpool_new(2, 10); - (*factory)->runall = smx_ctx_raw_runall_parallel; - (*factory)->self = smx_ctx_raw_self_parallel; -#else - THROW0(arg_error, 0, "No thread support for parallel context execution"); -#endif - }else{ - (*factory)->runall = smx_ctx_raw_runall; + parmap = xbt_parmap_new(2); + + if (SIMIX_context_is_parallel()) { + if (SIMIX_context_get_parallel_threshold() > 1) { + /* choose dynamically */ + (*factory)->runall = smx_ctx_raw_runall; + } + else { + /* always parallel */ + (*factory)->self = smx_ctx_raw_self_parallel; + (*factory)->get_thread_id = smx_ctx_raw_get_thread_id; + (*factory)->runall = smx_ctx_raw_runall_parallel; + } } + else { + /* always serial */ + (*factory)->self = smx_ctx_base_self; + (*factory)->get_thread_id = smx_ctx_base_get_thread_id; + (*factory)->runall = smx_ctx_raw_runall_serial; + } + + raw_factory = *factory; }