Changeset 637 for trunk/kernel/kern/dqdt.c

Timestamp:

Jul 18, 2019, 2:06:55 PM (5 years ago)

Author:

alain

Message:

Introduce the non-standard pthread_parallel_create() system call
and re-write the <fft> and <sort> applications to improve the
intrinsic paralelism in applications.

File:

• trunk/kernel/kern/dqdt.c (modified) (10 diffs)

trunk/kernel/kern/dqdt.c

@@ -2 +2 @@
  * dqdt.c - Distributed Quaternary Decision Tree implementation.
  *
- * Author : Alain Greiner (2016,2017,2018)
+ * Author : Alain Greiner (2016,2017,2018,2019)
  *
  * Copyright (c)  UPMC Sorbonne Universites
@@ -55 +55 @@
 
     // display node content
-        nolock_printk("- level %d / cluster %x : threads = %x / pages = %x / clusters %d / cores %d\n",
-    node.level, GET_CXY( node_xp ), node.threads, node.pages, node.clusters, node.cores );
+        nolock_printk("- [%d,%x] : threads %x / pages %x / clusters %d / cores %d / parent_cxy %x\n",
+                  node.level, GET_CXY( node_xp ),
+                  node.threads, node.pages,
+                  node.clusters, node.cores,
+                  GET_CXY( node.parent ) );
 
     // recursive call on children if node is not terminal
@@ -116 +119 @@
                                   xptr_t   parent_xp )
 {
-    assert( (level < 5) , __FUNCTION__, "illegal DQDT level %d\n", level );
+    assert( (level <= 5) , __FUNCTION__, "illegal DQDT level %d\n", level );
   
     uint32_t node_x;         // node X coordinate
@@ -147 +150 @@
 
 #if DEBUG_DQDT_INIT
-printk("\n[DBG] %s : cxy(%d,%d) / level %d / mask %x / half %d / ptr %x\n",
+printk("\n[%s] thread[%x,%x] : cxy(%d,%d) / level %d / mask %x / half %d / ptr %x\n",
 __FUNCTION__, node_x, node_y, level, mask, half, node_ptr );
 #endif
@@ -336 +339 @@
 void dqdt_init( void )
 {
-    // get x_size & y_size from cluster manager
-    cluster_t * cluster = &cluster_manager;
+    // get x_size & y_size
+    cluster_t * cluster = LOCAL_CLUSTER;
     uint32_t    x_size  = cluster->x_size;
     uint32_t    y_size  = cluster->y_size;
@@ -349 +352 @@
     uint32_t  level_max  = bits_log2( size_ext );
 
-    // each CP0 register the DQDT root in local cluster manager
+    // all CP0s register the DQDT root in local cluster manager
     cluster->dqdt_root_xp = XPTR( 0 , &cluster->dqdt_tbl[level_max] );
 
+    // only CP0 in cluster 0 build the DQDT
+    if( local_cxy == 0 )
+    {
+
 #if DEBUG_DQDT_INIT
-if( local_cxy == 0 )
-printk("\n[DBG] %s : x_size = %d / y_size = %d / level_max = %d\n",
-__FUNCTION__, x_size, y_size, level_max );
+thread_t * this = CURRENT_THREAD;
+printk("\n[%s] thread[%x,%x] enters : x_size = %d / y_size = %d / level_max = %d\n",
+__FUNCTION__, this->process->pid, this->trdid, x_size, y_size, level_max );
 #endif
     
@@ -362 +369 @@
 
 #if DEBUG_DQDT_INIT
-if( local_cxy == 0 ) dqdt_display();
-#endif
-
+dqdt_display();
+#endif
+
+    }
 }  // end dqdt_init()
 
@@ -516 +524 @@
 }
 
+///////////////////////////////////
+xptr_t dqdt_get_root( cxy_t    cxy,
+                      uint32_t level )
+{
+    xptr_t        node_xp;
+    cxy_t         node_cxy;
+    dqdt_node_t * node_ptr;
+    uint32_t      current_level;
+
+assert( (level <= 5) , __FUNCTION__, "illegal DQDT level %d\n", level );
+
+#if DEBUG_DQDT_GET_ROOT
+thread_t * this = CURRENT_THREAD;
+printk("\n[%s] thread[%x,%x] enters / cxy %x / level %d\n",
+__FUNCTION__, this->process->pid, this->trdid, cxy, level );
+#endif
+
+    // check macro-cluster 
+    if( cluster_is_active( cxy ) )
+    {    
+        // initialise node_xp and current_level 
+        node_xp       = XPTR( cxy , &LOCAL_CLUSTER->dqdt_tbl[0] );
+        current_level = 0;
+
+        // traverse the quad-tree from bottom to root
+        while( current_level < level )
+        {
+            node_cxy = GET_CXY( node_xp );
+            node_ptr = GET_PTR( node_xp );
+
+            node_xp = hal_remote_l64( XPTR( node_cxy , &node_ptr->parent ) );
+            current_level++;
+        }
+    }
+    else
+    {
+        node_xp =  XPTR_NULL;
+    }
+
+#if DEBUG_DQDT_GET_ROOT
+printk("\n[%s] thread[%x,%x] exit / root_xp[%x,%x]\n",
+__FUNCTION__, this->process->pid, this->trdid, GET_CXY( node_xp ), GET_PTR( node_xp ) );
+#endif
+
+    return node_xp;
+    
+}
 
 /////////////////////////////////////////////////////////////////////////////////////
@@ -584 +639 @@
 
 
-//////////////////////////////////////////
-cxy_t dqdt_get_cluster_for_process( void )
+///////////////////////////////////////////////////
+cxy_t dqdt_get_cluster_for_thread( xptr_t root_xp )
 {
     // call recursive function
-    cxy_t cxy = dqdt_select_cluster( LOCAL_CLUSTER->dqdt_root_xp , false );
-
-#if DEBUG_DQDT_SELECT_FOR_PROCESS
+    cxy_t cxy = dqdt_select_cluster( root_xp , false );
+
+#if DEBUG_DQDT_SELECT_FOR_THREAD
 uint32_t cycle = hal_get_cycles();
 if( cycle > DEBUG_DQDT_SELECT_FOR_PROCESS )
@@ -600 +655 @@
 }
 
-/////////////////////////////////////////
-cxy_t dqdt_get_cluster_for_memory( void )
+///////////////////////////////////////////////////
+cxy_t dqdt_get_cluster_for_memory( xptr_t root_xp )
 {
     // call recursive function
-    cxy_t cxy = dqdt_select_cluster( LOCAL_CLUSTER->dqdt_root_xp , true );
+    cxy_t cxy = dqdt_select_cluster( root_xp , true );
 
 #if DEBUG_DQDT_SELECT_FOR_MEMORY