Changes in libcfa/src/concurrency/kernel.hfa [1eb239e4:71c8b7e]

File:

• libcfa/src/concurrency/kernel.hfa (modified) (10 diffs)

libcfa/src/concurrency/kernel.hfa

@@ -16 +16 @@
 #pragma once
 
+#include <stdbool.h>
+#include <stdint.h>
+
 #include "invoke.h"
 #include "time_t.hfa"
 #include "coroutine.hfa"
 
-#include "containers/list.hfa"
-
 extern "C" {
-#include <bits/pthreadtypes.h>
+#include <pthread.h>
+#include <semaphore.h>
 }
 
@@ -45 +47 @@
 extern struct cluster * mainCluster;
 
-// Processor id, required for scheduling threads
-struct __processor_id_t {
-        unsigned id;
-
-        #if !defined(__CFA_NO_STATISTICS__)
-                struct __stats_t * stats;
-        #endif
-};
-
+// Processor
 coroutine processorCtx_t {
         struct processor * proc;
@@ -59 +53 @@
 
 // Wrapper around kernel threads
-struct __attribute__((aligned(128))) processor {
+struct processor {
         // Main state
-        inline __processor_id_t;
+        // Coroutine ctx who does keeps the state of the processor
+        struct processorCtx_t runner;
 
         // Cluster from which to get threads
         struct cluster * cltr;
-
-        // Set to true to notify the processor should terminate
-        volatile bool do_terminate;
-
-        // Coroutine ctx who does keeps the state of the processor
-        struct processorCtx_t runner;
 
         // Name of the processor
@@ -92 +81 @@
         __bin_sem_t idle;
 
+        // Termination
+        // Set to true to notify the processor should terminate
+        volatile bool do_terminate;
+
         // Termination synchronisation (user semaphore)
         semaphore terminated;
@@ -99 +92 @@
 
         // Link lists fields
-        DLISTED_MGD_IMPL_IN(processor)
-
-        #if !defined(__CFA_NO_STATISTICS__)
-                int print_stats;
-                bool print_halts;
-        #endif
+        struct __dbg_node_proc {
+                struct processor * next;
+                struct processor * prev;
+        } node;
 
 #ifdef __CFA_DEBUG__
@@ -119 +110 @@
 static inline void  ?{}(processor & this, const char name[]) { this{name, *mainCluster }; }
 
-DLISTED_MGD_IMPL_OUT(processor)
+static inline [processor *&, processor *& ] __get( processor & this ) __attribute__((const)) { return this.node.[next, prev]; }
 
 //-----------------------------------------------------------------------------
@@ -125 +116 @@
 struct __io_data;
 
-// IO poller user-thread
-// Not using the "thread" keyword because we want to control
-// more carefully when to start/stop it
-struct $io_ctx_thread {
-        struct __io_data * ring;
-        single_sem sem;
-        volatile bool done;
-        $thread self;
-};
-
-
-struct io_context {
-        $io_ctx_thread thrd;
-};
-
-struct io_context_params {
-        int num_entries;
-        int num_ready;
-        int submit_aff;
-        bool eager_submits:1;
-        bool poller_submits:1;
-        bool poll_submit:1;
-        bool poll_complete:1;
-};
-
-void  ?{}(io_context_params & this);
-
-void  ?{}(io_context & this, struct cluster & cl);
-void  ?{}(io_context & this, struct cluster & cl, const io_context_params & params);
-void ^?{}(io_context & this);
-
-struct io_cancellation {
-        uint32_t target;
-};
-
-static inline void  ?{}(io_cancellation & this) { this.target = -1u; }
-static inline void ^?{}(io_cancellation & this) {}
-bool cancel(io_cancellation & this);
-
-//-----------------------------------------------------------------------------
-// Cluster Tools
-
-// Intrusives lanes which are used by the relaxed ready queue
-struct __attribute__((aligned(128))) __intrusive_lane_t;
-void  ?{}(__intrusive_lane_t & this);
-void ^?{}(__intrusive_lane_t & this);
-
-// Counter used for wether or not the lanes are all empty
-struct __attribute__((aligned(128))) __snzi_node_t;
-struct __snzi_t {
-        unsigned mask;
-        int root;
-        __snzi_node_t * nodes;
-};
-
-void  ?{}( __snzi_t & this, unsigned depth );
-void ^?{}( __snzi_t & this );
-
-//TODO adjust cache size to ARCHITECTURE
-// Structure holding the relaxed ready queue
-struct __ready_queue_t {
-        // Data tracking how many/which lanes are used
-        // Aligned to 128 for cache locality
-        __snzi_t snzi;
-
-        // Data tracking the actual lanes
-        // On a seperate cacheline from the used struct since
-        // used can change on each push/pop but this data
-        // only changes on shrink/grow
-        struct {
-                // Arary of lanes
-                __intrusive_lane_t * volatile data;
-
-                // Number of lanes (empty or not)
-                volatile size_t count;
-        } lanes;
-};
-
-void  ?{}(__ready_queue_t & this);
-void ^?{}(__ready_queue_t & this);
-
-// Idle Sleep
-struct __cluster_idles {
-        // Spin lock protecting the queue
-        volatile uint64_t lock;
-
-        // Total number of processors
-        unsigned total;
-
-        // Total number of idle processors
-        unsigned idle;
-
-        // List of idle processors
-        dlist(processor, processor) list;
-};
+#define CFA_CLUSTER_IO_POLLER_USER_THREAD 1 << 0
+// #define CFA_CLUSTER_IO_POLLER_KERNEL_SIDE 1 << 1
 
 //-----------------------------------------------------------------------------
 // Cluster
-struct __attribute__((aligned(128))) cluster {
+struct cluster {
+        // Ready queue locks
+        __spinlock_t ready_queue_lock;
+
         // Ready queue for threads
-        __ready_queue_t ready_queue;
+        __queue_t($thread) ready_queue;
 
         // Name of the cluster
@@ -233 +134 @@
         Duration preemption_rate;
 
-        // List of idle processors
-        __cluster_idles idles;
+        // List of processors
+        __spinlock_t idle_lock;
+        __dllist_t(struct processor) procs;
+        __dllist_t(struct processor) idles;
+        unsigned int nprocessors;
 
         // List of threads
@@ -247 +151 @@
         } node;
 
-        struct {
-                io_context * ctxs;
-                unsigned cnt;
-        } io;
+        struct __io_data * io;
 
         #if !defined(__CFA_NO_STATISTICS__)
-                struct __stats_t * stats;
-                int print_stats;
+                bool print_stats;
         #endif
 };
 extern Duration default_preemption();
 
-void ?{} (cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params);
+void ?{} (cluster & this, const char name[], Duration preemption_rate, int flags);
 void ^?{}(cluster & this);
 
-static inline void ?{} (cluster & this)                                            { io_context_params default_params;    this{"Anonymous Cluster", default_preemption(), 1, default_params}; }
-static inline void ?{} (cluster & this, Duration preemption_rate)                  { io_context_params default_params;    this{"Anonymous Cluster", preemption_rate, 1, default_params}; }
-static inline void ?{} (cluster & this, const char name[])                         { io_context_params default_params;    this{name, default_preemption(), 1, default_params}; }
-static inline void ?{} (cluster & this, unsigned num_io)                           { io_context_params default_params;    this{"Anonymous Cluster", default_preemption(), num_io, default_params}; }
-static inline void ?{} (cluster & this, Duration preemption_rate, unsigned num_io) { io_context_params default_params;    this{"Anonymous Cluster", preemption_rate, num_io, default_params}; }
-static inline void ?{} (cluster & this, const char name[], unsigned num_io)        { io_context_params default_params;    this{name, default_preemption(), num_io, default_params}; }
-static inline void ?{} (cluster & this, const io_context_params & io_params)                                            { this{"Anonymous Cluster", default_preemption(), 1, io_params}; }
-static inline void ?{} (cluster & this, Duration preemption_rate, const io_context_params & io_params)                  { this{"Anonymous Cluster", preemption_rate, 1, io_params}; }
-static inline void ?{} (cluster & this, const char name[], const io_context_params & io_params)                         { this{name, default_preemption(), 1, io_params}; }
-static inline void ?{} (cluster & this, unsigned num_io, const io_context_params & io_params)                           { this{"Anonymous Cluster", default_preemption(), num_io, io_params}; }
-static inline void ?{} (cluster & this, Duration preemption_rate, unsigned num_io, const io_context_params & io_params) { this{"Anonymous Cluster", preemption_rate, num_io, io_params}; }
-static inline void ?{} (cluster & this, const char name[], unsigned num_io, const io_context_params & io_params)        { this{name, default_preemption(), num_io, io_params}; }
+static inline void ?{} (cluster & this)                                      { this{"Anonymous Cluster", default_preemption(), 0}; }
+static inline void ?{} (cluster & this, Duration preemption_rate)            { this{"Anonymous Cluster", preemption_rate, 0}; }
+static inline void ?{} (cluster & this, const char name[])                   { this{name, default_preemption(), 0}; }
+static inline void ?{} (cluster & this, int flags)                           { this{"Anonymous Cluster", default_preemption(), flags}; }
+static inline void ?{} (cluster & this, Duration preemption_rate, int flags) { this{"Anonymous Cluster", preemption_rate, flags}; }
+static inline void ?{} (cluster & this, const char name[], int flags)        { this{name, default_preemption(), flags}; }
 
 static inline [cluster *&, cluster *& ] __get( cluster & this ) __attribute__((const)) { return this.node.[next, prev]; }
@@ -281 +175 @@
 
 #if !defined(__CFA_NO_STATISTICS__)
-        static inline void print_stats_at_exit( cluster & this, int flags ) {
-                this.print_stats |= flags;
+        static inline void print_stats_at_exit( cluster & this ) {
+                this.print_stats = true;
         }
-
-        static inline void print_stats_at_exit( processor & this, int flags ) {
-                this.print_stats |= flags;
-        }
-
-        void print_halts( processor & this );
 #endif