X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/d51e11eeca324570cf038a3bf6a5d05109903633..5880a1bbbf150def1664cfefd6ec7f36f56edfaa:/src/simix/smx_context_raw.c diff --git a/src/simix/smx_context_raw.c b/src/simix/smx_context_raw.c index 55b5d80d9e..df733cda81 100644 --- a/src/simix/smx_context_raw.c +++ b/src/simix/smx_context_raw.c @@ -1,6 +1,6 @@ -/* context_raw - fast context switching inspired from System V ucontextes */ +/* context_raw - fast context switching inspired from System V ucontexts */ -/* Copyright (c) 2009, 2010. The SimGrid Team. +/* Copyright (c) 2009-2014. The SimGrid Team. * All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it @@ -31,7 +31,7 @@ typedef struct s_smx_ctx_raw { #ifdef CONTEXT_THREADS static xbt_parmap_t raw_parmap; -static raw_stack_t* raw_workers_stacks; /* space to save the worker stack in each thread */ +static smx_ctx_raw_t* raw_workers_context; /* space to save the worker context in each thread */ static unsigned long raw_threads_working; /* number of threads that have started their work */ static xbt_os_thread_key_t raw_worker_id_key; /* thread-specific storage for the thread id */ #endif @@ -44,7 +44,86 @@ 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); -#ifdef PROCESSOR_i686 +#if PROCESSOR_x86_64 +__asm__ ( +#if defined(APPLE) + ".text\n" + ".globl _raw_makecontext\n" + "_raw_makecontext:\n" +#elif defined(_WIN32) + ".text\n" + ".globl raw_makecontext\n" + "raw_makecontext:\n" +#else + ".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 */ +#endif + " mov %rdi,%rax\n" /* stack */ + " add %rsi,%rax\n" /* size */ + " movq $0, -8(%rax)\n" /* @return for func */ + " mov %rdx,-16(%rax)\n" /* func */ + " mov %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 */ + " sub $112,%rax\n" + " ret\n" +); + +__asm__ ( +#if defined(APPLE) + ".text\n" + ".globl _raw_swapcontext\n" + "_raw_swapcontext:\n" +#elif defined(_WIN32) + ".text\n" + ".globl raw_swapcontext\n" + "raw_swapcontext:\n" +#else + ".text\n" + ".globl raw_swapcontext\n" + ".type raw_swapcontext,@function\n" + "raw_swapcontext:\n" /* Calling convention sets the arguments in rdi and rsi, respectively */ +#endif + " push %rdi\n" + " push %rsi\n" + " push %rdx\n" + " push %rcx\n" + " push %r8\n" + " push %r9\n" + " push %rbp\n" + " push %rbx\n" + " push %r12\n" + " push %r13\n" + " push %r14\n" + " push %r15\n" + " mov %rsp,(%rdi)\n" /* old */ + " mov %rsi,%rsp\n" /* new */ + " pop %r15\n" + " pop %r14\n" + " pop %r13\n" + " pop %r12\n" + " pop %rbx\n" + " pop %rbp\n" + " pop %r9\n" + " pop %r8\n" + " pop %rcx\n" + " pop %rdx\n" + " pop %rsi\n" + " pop %rdi\n" + " ret\n" +); +#elif PROCESSOR_i686 __asm__ ( #if defined(APPLE) || defined(_WIN32) ".text\n" @@ -96,86 +175,8 @@ __asm__ ( " popl %ebp\n" " retl\n" ); -#elif PROCESSOR_x86_64 -__asm__ ( -#if defined(APPLE) - ".text\n" - ".globl _raw_makecontext\n" - "_raw_makecontext:\n" -#elif defined(_WIN32) - ".text\n" - ".globl raw_makecontext\n" - "raw_makecontext:\n" #else - ".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 */ -#endif - " 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__ ( -#if defined(APPLE) - ".text\n" - ".globl _raw_swapcontext\n" - "_raw_swapcontext:\n" -#elif defined(_WIN32) - ".text\n" - ".globl raw_swapcontext\n" - "raw_swapcontext:\n" -#else - ".text\n" - ".globl raw_swapcontext\n" - ".type raw_swapcontext,@function\n" - "raw_swapcontext:\n" /* Calling convention sets the arguments in rdi and rsi, respectively */ -#endif - " 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 /* If you implement raw contexts for other processors, don't forget to update the definition of HAVE_RAWCTX in buildtools/Cmake/CompleteInFiles.cmake */ @@ -209,7 +210,7 @@ static char new_sr = 0; static void smx_ctx_raw_wrapper(smx_ctx_raw_t context); static int smx_ctx_raw_factory_finalize(smx_context_factory_t *factory); 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); + char **argv, void_pfn_smxprocess_t cleanup_func, smx_process_t process); static void smx_ctx_raw_free(smx_context_t context); static void smx_ctx_raw_wrapper(smx_ctx_raw_t context); static void smx_ctx_raw_stop(smx_context_t context); @@ -241,9 +242,11 @@ void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory) if (SIMIX_context_is_parallel()) { #ifdef CONTEXT_THREADS int nthreads = SIMIX_context_get_nthreads(); - raw_parmap = xbt_parmap_new(nthreads, SIMIX_context_get_parallel_mode()); - raw_workers_stacks = xbt_new(raw_stack_t, nthreads); xbt_os_thread_key_create(&raw_worker_id_key); + raw_parmap = xbt_parmap_new(nthreads, SIMIX_context_get_parallel_mode()); + raw_workers_context = xbt_new(smx_ctx_raw_t, nthreads); + raw_maestro_context=NULL; + #endif if (SIMIX_context_get_parallel_threshold() > 1) { /* choose dynamically */ @@ -273,14 +276,14 @@ void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory) static int smx_ctx_raw_factory_finalize(smx_context_factory_t *factory) { #ifdef TIME_BENCH_PER_SR - XBT_CRITICAL("Total wasted time in %u SR: %lf", sr_count, time_wasted_sr); - XBT_CRITICAL("Total wasted time in %u SSR: %lf", ssr_count, time_wasted_ssr); + XBT_CRITICAL("Total wasted time in %u SR: %f", sr_count, time_wasted_sr); + XBT_CRITICAL("Total wasted time in %u SSR: %f", ssr_count, time_wasted_ssr); #endif #ifdef CONTEXT_THREADS if (raw_parmap) xbt_parmap_destroy(raw_parmap); - xbt_free(raw_workers_stacks); + xbt_free(raw_workers_context); #endif return smx_ctx_base_factory_finalize(factory); } @@ -293,12 +296,12 @@ static int smx_ctx_raw_factory_finalize(smx_context_factory_t *factory) * \param argv arguments to pass to the main function * \param cleanup_func a function to call to free the user data when the * context finished - * \param data user data + * \param process SIMIX process */ 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) + void_pfn_smxprocess_t cleanup_func, + smx_process_t process) { smx_ctx_raw_t context = @@ -308,12 +311,12 @@ smx_ctx_raw_create_context(xbt_main_func_t code, int argc, char **argv, argc, argv, cleanup_func, - data); + process); /* if the user provided a function for the process then use it, otherwise it is the context for maestro */ if (code) { - context->malloced_stack = xbt_malloc0(smx_context_stack_size); + context->malloced_stack = SIMIX_context_stack_new(); context->stack_top = raw_makecontext(context->malloced_stack, smx_context_stack_size, (void_f_pvoid_t) smx_ctx_raw_wrapper, context); @@ -325,10 +328,11 @@ smx_ctx_raw_create_context(xbt_main_func_t code, int argc, char **argv, #endif /* HAVE_VALGRIND_VALGRIND_H */ } else { - raw_maestro_context = context; + if(process != NULL && raw_maestro_context==NULL) + raw_maestro_context = context; - if(MC_IS_ENABLED) - MC_ignore(&(raw_maestro_context->stack_top), sizeof(raw_maestro_context->stack_top)); + if(MC_is_active()) + MC_ignore_heap(&(raw_maestro_context->stack_top), sizeof(raw_maestro_context->stack_top)); } @@ -348,7 +352,8 @@ 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); + SIMIX_context_stack_delete(((smx_ctx_raw_t) context)->malloced_stack); + } smx_ctx_base_free(context); } @@ -444,9 +449,9 @@ static void smx_ctx_raw_runall_serial(xbt_dynar_t processes) time_thread_sr[t] = 0; } - xbt_os_timer_start(timer); + xbt_os_cputimer_start(timer); smx_ctx_raw_resume(process); - xbt_os_timer_stop(timer); + xbt_os_cputimer_stop(timer); elapsed = xbt_os_timer_elapsed(timer); time_thread_ssr[t] += elapsed; time_thread_sr[((smx_ctx_raw_t)process->context)->thread] += elapsed; @@ -459,7 +464,7 @@ static void smx_ctx_raw_runall_serial(xbt_dynar_t processes) tmax = time_thread_ssr[t]; for(cursor=0; cursor <= t; cursor++){ - XBT_VERB("Time SSR thread %u = %lf (max %lf)", cursor, time_thread_ssr[cursor], tmax); + XBT_VERB("Time SSR thread %u = %f (max %f)", cursor, time_thread_ssr[cursor], tmax); time_wasted_ssr += tmax - time_thread_ssr[cursor]; } } @@ -477,7 +482,7 @@ void smx_ctx_raw_new_sr(void) } for(i=0; i < NUM_THREADS; i++){ - XBT_VERB("Time SR thread %u = %lf (max %lf)", i, time_thread_sr[i], tmax); + XBT_VERB("Time SR thread %u = %f (max %f)", i, time_thread_sr[i], tmax); time_wasted_sr += tmax - time_thread_sr[i]; } @@ -490,14 +495,12 @@ void smx_ctx_raw_new_sr(void) */ static void smx_ctx_raw_runall_serial(void) { - if (!xbt_dynar_is_empty(simix_global->process_to_run)) { - smx_process_t first_process = - xbt_dynar_get_as(simix_global->process_to_run, 0, smx_process_t); - raw_process_index = 1; + smx_process_t first_process = + xbt_dynar_get_as(simix_global->process_to_run, 0, smx_process_t); + raw_process_index = 1; - /* execute the first process */ - smx_ctx_raw_resume_serial(first_process); - } + /* execute the first process */ + smx_ctx_raw_resume_serial(first_process); } #endif @@ -523,12 +526,14 @@ static void smx_ctx_raw_suspend_parallel(smx_context_t context) else { /* all processes were run, go to the barrier */ XBT_DEBUG("No more processes to run"); - next_context = (smx_context_t) raw_maestro_context; + unsigned long worker_id = (unsigned long) xbt_os_thread_get_specific(raw_worker_id_key); + + next_context = (smx_context_t)raw_workers_context[worker_id]; XBT_DEBUG("Restoring worker stack %lu (working threads = %lu)", worker_id, raw_threads_working); - next_stack = raw_workers_stacks[worker_id]; + next_stack = ((smx_ctx_raw_t)next_context)->stack_top; } SIMIX_context_set_current(next_context); @@ -546,8 +551,11 @@ static void smx_ctx_raw_resume_parallel(smx_process_t first_process) #ifdef CONTEXT_THREADS unsigned long worker_id = __sync_fetch_and_add(&raw_threads_working, 1); xbt_os_thread_set_specific(raw_worker_id_key, (void*) worker_id); + smx_ctx_raw_t worker_context = (smx_ctx_raw_t)SIMIX_context_self(); + raw_workers_context[worker_id] = worker_context; XBT_DEBUG("Saving worker stack %lu", worker_id); - raw_stack_t* worker_stack = &raw_workers_stacks[worker_id]; + raw_stack_t* worker_stack = &(worker_context)->stack_top; + smx_context_t context = first_process->context; SIMIX_context_set_current(context); @@ -564,6 +572,8 @@ static void smx_ctx_raw_runall_parallel(void) raw_threads_working = 0; xbt_parmap_apply(raw_parmap, (void_f_pvoid_t) smx_ctx_raw_resume_parallel, simix_global->process_to_run); +#else + xbt_die("You asked for a parallel execution, but you don't have any threads."); #endif }