Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(2)

Unified Diff: gdb/osf-share/cma_sched.h

Issue 124383005: GDB 7.6.50 (Closed) Base URL: http://git.chromium.org/native_client/nacl-gdb.git@upstream
Patch Set: Created 6 years, 11 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « gdb/osf-share/cma_mutex.h ('k') | gdb/osf-share/cma_semaphore_defs.h » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: gdb/osf-share/cma_sched.h
diff --git a/gdb/osf-share/cma_sched.h b/gdb/osf-share/cma_sched.h
deleted file mode 100644
index 6eb78743bafd3cd1ce5e42c6a965fc6621fc936b..0000000000000000000000000000000000000000
--- a/gdb/osf-share/cma_sched.h
+++ /dev/null
@@ -1,279 +0,0 @@
-/*
- * (c) Copyright 1990-1996 OPEN SOFTWARE FOUNDATION, INC.
- * (c) Copyright 1990-1996 HEWLETT-PACKARD COMPANY
- * (c) Copyright 1990-1996 DIGITAL EQUIPMENT CORPORATION
- * (c) Copyright 1991, 1992 Siemens-Nixdorf Information Systems
- * To anyone who acknowledges that this file is provided "AS IS" without
- * any express or implied warranty: permission to use, copy, modify, and
- * distribute this file for any purpose is hereby granted without fee,
- * provided that the above copyright notices and this notice appears in
- * all source code copies, and that none of the names listed above be used
- * in advertising or publicity pertaining to distribution of the software
- * without specific, written prior permission. None of these organizations
- * makes any representations about the suitability of this software for
- * any purpose.
- */
-/*
- * Header file for priority scheduling
- */
-
-
-#ifndef CMA_SCHED
-#define CMA_SCHED
-
-/*
- * INCLUDE FILES
- */
-
-/*
- * CONSTANTS AND MACROS
- */
-
-/*
- * Scaling factor for integer priority calculations
- */
-#define cma__c_prio_scale 8
-
-#if _CMA_VENDOR_ == _CMA__APOLLO
-/*
- * FIX-ME: Apollo cc 6.8 blows contant folded "<<" and ">>"
- */
-# define cma__scale_up(exp) ((exp) * 256)
-# define cma__scale_dn(exp) ((exp) / 256)
-#else
-# define cma__scale_up(exp) ((exp) << cma__c_prio_scale)
-# define cma__scale_dn(exp) ((exp) >> cma__c_prio_scale)
-#endif
-
-
-/*
- * Min. num. of ticks between self-adjustments for priority adjusting policies.
- */
-#define cma__c_prio_interval 10
-
-
-/*
- * Number of queues in each class of queues
- */
-#define cma__c_prio_n_id 1 /* Very-low-priority class threads */
-#define cma__c_prio_n_bg 8 /* Background class threads */
-#define cma__c_prio_n_0 1 /* Very low priority throughput quartile */
-#define cma__c_prio_n_1 2 /* Low priority throughput quartile */
-#define cma__c_prio_n_2 3 /* Medium priority throughput quartile */
-#define cma__c_prio_n_3 4 /* High priority throughput quartile */
-#define cma__c_prio_n_rt 1 /* Real Time priority queues */
-
-/*
- * Number of queues to skip (offset) to get to the queues in this section of LA
- */
-#define cma__c_prio_o_id 0
-#define cma__c_prio_o_bg cma__c_prio_o_id + cma__c_prio_n_id
-#define cma__c_prio_o_0 cma__c_prio_o_bg + cma__c_prio_n_bg
-#define cma__c_prio_o_1 cma__c_prio_o_0 + cma__c_prio_n_0
-#define cma__c_prio_o_2 cma__c_prio_o_1 + cma__c_prio_n_1
-#define cma__c_prio_o_3 cma__c_prio_o_2 + cma__c_prio_n_2
-#define cma__c_prio_o_rt cma__c_prio_o_3 + cma__c_prio_n_3
-
-/*
- * Ada_low: These threads are queued in the background queues, thus there
- * must be enough queues to allow one queue for each Ada priority below the
- * Ada default.
- */
-#define cma__c_prio_o_al cma__c_prio_o_bg
-
-/*
- * Total number of ready queues, for declaration purposes
- */
-#define cma__c_prio_n_tot \
- cma__c_prio_n_id + cma__c_prio_n_bg + cma__c_prio_n_rt \
- + cma__c_prio_n_0 + cma__c_prio_n_1 + cma__c_prio_n_2 + cma__c_prio_n_3
-
-/*
- * Formulae for determining a thread's priority. Variable priorities (such
- * as foreground and background) are scaled values.
- */
-#define cma__sched_priority(tcb) \
- ((tcb)->sched.class == cma__c_class_fore ? cma__sched_prio_fore (tcb) \
- :((tcb)->sched.class == cma__c_class_back ? cma__sched_prio_back (tcb) \
- :((tcb)->sched.class == cma__c_class_rt ? cma__sched_prio_rt (tcb) \
- :((tcb)->sched.class == cma__c_class_idle ? cma__sched_prio_idle (tcb) \
- :(cma__bugcheck ("cma__sched_priority: unrecognized class"), 0) ))))
-
-#define cma__sched_prio_fore(tcb) cma__sched_prio_fore_var (tcb)
-#define cma__sched_prio_back(tcb) ((tcb)->sched.fixed_prio \
- ? cma__sched_prio_back_fix (tcb) : cma__sched_prio_back_var (tcb) )
-#define cma__sched_prio_rt(tcb) ((tcb)->sched.priority)
-#define cma__sched_prio_idle(tcb) ((tcb)->sched.priority)
-
-#define cma__sched_prio_back_fix(tcb) \
- (cma__g_prio_bg_min + (cma__g_prio_bg_max - cma__g_prio_bg_min) \
- * ((tcb)->sched.priority + cma__c_prio_o_al - cma__c_prio_o_bg) \
- / cma__c_prio_n_bg)
-
-/*
- * FIX-ME: Enable after modeling (if we like it)
- */
-#if 1
-# define cma__sched_prio_fore_var(tcb) \
- ((cma__g_prio_fg_max + cma__g_prio_fg_min)/2)
-# define cma__sched_prio_back_var(tcb) \
- ((cma__g_prio_bg_max + cma__g_prio_bg_min)/2)
-#else
-# define cma__sched_prio_back_var(tcb) cma__sched_prio_fore_var (tcb)
-
-# if 1
-/*
- * Re-scale, since the division removes the scale factor.
- * Scale and multiply before dividing to avoid loss of precision.
- */
-# define cma__sched_prio_fore_var(tcb) \
- ((cma__g_vp_count * cma__scale_up((tcb)->sched.tot_time)) \
- / (tcb)->sched.cpu_time)
-# else
-/*
- * Re-scale, since the division removes the scale factor.
- * Scale and multiply before dividing to avoid loss of precision.
- * Left shift the numerator to multiply by two.
- */
-# define cma__sched_prio_fore_var(tcb) \
- (((cma__g_vp_count * cma__scale_up((tcb)->sched.tot_time) \
- * (tcb)->sched.priority * cma__g_init_frac_sum) << 1) \
- / ((tcb)->sched.cpu_time * (tcb)->sched.priority * cma__g_init_frac_sum \
- + (tcb)->sched.tot_time))
-# endif
-#endif
-
-/*
- * Update weighted-averaged, scaled tick counters
- */
-#define cma__sched_update_time(ave, new) \
- (ave) = (ave) - ((cma__scale_dn((ave)) - (new)) << (cma__c_prio_scale - 4))
-
-#define cma__sched_parameterize(tcb, policy) { \
- switch (policy) { \
- case cma_c_sched_fifo : { \
- (tcb)->sched.rtb = cma_c_true; \
- (tcb)->sched.spp = cma_c_true; \
- (tcb)->sched.fixed_prio = cma_c_true; \
- (tcb)->sched.class = cma__c_class_rt; \
- break; \
- } \
- case cma_c_sched_rr : { \
- (tcb)->sched.rtb = cma_c_false; \
- (tcb)->sched.spp = cma_c_true; \
- (tcb)->sched.fixed_prio = cma_c_true; \
- (tcb)->sched.class = cma__c_class_rt; \
- break; \
- } \
- case cma_c_sched_throughput : { \
- (tcb)->sched.rtb = cma_c_false; \
- (tcb)->sched.spp = cma_c_false; \
- (tcb)->sched.fixed_prio = cma_c_false; \
- (tcb)->sched.class = cma__c_class_fore; \
- break; \
- } \
- case cma_c_sched_background : { \
- (tcb)->sched.rtb = cma_c_false; \
- (tcb)->sched.spp = cma_c_false; \
- (tcb)->sched.fixed_prio = cma_c_false; \
- (tcb)->sched.class = cma__c_class_back; \
- break; \
- } \
- case cma_c_sched_ada_low : { \
- (tcb)->sched.rtb = cma_c_false; \
- (tcb)->sched.spp = cma_c_true; \
- (tcb)->sched.fixed_prio = cma_c_true; \
- (tcb)->sched.class = cma__c_class_back; \
- break; \
- } \
- case cma_c_sched_idle : { \
- (tcb)->sched.rtb = cma_c_false; \
- (tcb)->sched.spp = cma_c_false; \
- (tcb)->sched.fixed_prio = cma_c_false; \
- (tcb)->sched.class = cma__c_class_idle; \
- break; \
- } \
- default : { \
- cma__bugcheck ("cma__sched_parameterize: bad scheduling Policy"); \
- break; \
- } \
- } \
- }
-
-/*
- * TYPEDEFS
- */
-
-/*
- * Scheduling classes
- */
-typedef enum CMA__T_SCHED_CLASS {
- cma__c_class_rt,
- cma__c_class_fore,
- cma__c_class_back,
- cma__c_class_idle
- } cma__t_sched_class;
-
-/*
- * GLOBAL DATA
- */
-
-/*
- * Minimuma and maximum prioirities, for foreground and background threads,
- * as of the last time the scheduler ran. (Scaled once.)
- */
-extern cma_t_integer cma__g_prio_fg_min;
-extern cma_t_integer cma__g_prio_fg_max;
-extern cma_t_integer cma__g_prio_bg_min;
-extern cma_t_integer cma__g_prio_bg_max;
-
-/*
- * The "m" values are the slopes of the four sections of linear approximation.
- *
- * cma__g_prio_m_I = 4*N(I)/cma__g_prio_range (Scaled once.)
- */
-extern cma_t_integer cma__g_prio_m_0,
- cma__g_prio_m_1,
- cma__g_prio_m_2,
- cma__g_prio_m_3;
-
-/*
- * The "b" values are the intercepts of the four sections of linear approx.
- * (Not scaled.)
- *
- * cma__g_prio_b_I = -N(I)*(I*prio_max + (4-I)*prio_min)/prio_range + prio_o_I
- */
-extern cma_t_integer cma__g_prio_b_0,
- cma__g_prio_b_1,
- cma__g_prio_b_2,
- cma__g_prio_b_3;
-
-/*
- * The "p" values are the end points of the four sections of linear approx.
- *
- * cma__g_prio_p_I = cma__g_prio_fg_min + (I/4)*cma__g_prio_range
- *
- * [cma__g_prio_p_0 is not defined since it is not used (also, it is the same
- * as cma__g_prio_fg_min).] (Scaled once.)
- */
-extern cma_t_integer cma__g_prio_p_1,
- cma__g_prio_p_2,
- cma__g_prio_p_3;
-
-/*
- * Points to the next queue for the dispatcher to check for ready threads.
- */
-extern cma_t_integer cma__g_next_ready_queue;
-
-/*
- * Points to the queues of virtual processors (for preempt victim search)
- */
-extern cma__t_queue cma__g_run_vps;
-extern cma__t_queue cma__g_susp_vps;
-extern cma_t_integer cma__g_vp_count;
-
-/*
- * INTERNAL INTERFACES
- */
-
-#endif
« no previous file with comments | « gdb/osf-share/cma_mutex.h ('k') | gdb/osf-share/cma_semaphore_defs.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698