Changeset 2210cfc

Timestamp:

Jan 10, 2022, 5:43:42 PM (3 years ago)

Author:

Peter A. Buhr <pabuhr@…>

Branches:

ADT, ast-experimental, enum, forall-pointer-decay, master, pthread-emulation, qualifiedEnum

Children:

c52f033

Parents:

4177592f

Message:

second attempt at specialized PRNG

Location:

libcfa/src

Files:

• concurrency/invoke.h (modified) (2 diffs)
• concurrency/thread.cfa (modified) (4 diffs)
• stdlib.cfa (modified) (3 diffs)
• stdlib.hfa (modified) (4 diffs)

libcfa/src/concurrency/invoke.h

@@ -10 +10 @@
 // Created On       : Tue Jan 17 12:27:26 2016
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Thu Jan  6 16:37:40 2022
-// Update Count     : 47
+// Last Modified On : Sun Jan  9 19:06:45 2022
+// Update Count     : 48
 //
 
@@ -211 +211 @@
                 struct processor * last_proc;
 
+                uint32_t random_state;                                                  // fast random numbers
+
                 #if defined( __CFA_WITH_VERIFY__ )
                         void * canary;

libcfa/src/concurrency/thread.cfa

@@ -10 +10 @@
 // Created On       : Tue Jan 17 12:27:26 2017
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Wed Dec  4 09:17:49 2019
-// Update Count     : 9
+// Last Modified On : Mon Jan 10 17:05:35 2022
+// Update Count     : 28
 //
 
@@ -27 +27 @@
 uint64_t thread_rand();
 
+extern uint32_t __thread_seed;                                                  // global thread seed
+
 //-----------------------------------------------------------------------------
 // Thread ctors and dtors
-void ?{}(thread$ & this, const char * const name, cluster & cl, void * storage, size_t storageSize ) with( this ) {
+void ?{}( thread$ & this, const char * const name, cluster & cl, void * storage, size_t storageSize ) with( this ) {
         context{ 0p, 0p };
         self_cor{ name, storage, storageSize };
@@ -39 +41 @@
         self_mon.owner = &this;
         self_mon.recursion = 1;
+        random_state = __thread_seed;
         self_mon_p = &self_mon;
         curr_cluster = &cl;
@@ -178 +181 @@
 }
 
+#define GENERATOR LCG
+
+inline uint32_t MarsagliaXor( uint32_t & state ) {
+        state ^= state << 6;
+        state ^= state >> 21;
+        state ^= state << 7;
+        return state;
+} // MarsagliaXor
+
+inline uint32_t LCG( uint32_t & state ) {                               // linear congruential generator
+        return state = 36969 * (state & 65535) + (state >> 16); // 36969 is NOT prime!
+} // LCG
+
+void set_seed( uint32_t seed ) { active_thread()->random_state = seed; __thread_seed = seed; }
+uint32_t prng( void ) { return GENERATOR( active_thread()->random_state ); } // [0,UINT_MAX]
+
 // Local Variables: //
 // mode: c //

libcfa/src/stdlib.cfa

@@ -10 +10 @@
 // Created On       : Thu Jan 28 17:10:29 2016
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Mon Jan  3 09:36:27 2022
-// Update Count     : 519
+// Last Modified On : Mon Jan 10 17:07:21 2022
+// Update Count     : 572
 //
 
 #include "stdlib.hfa"
+//#include "concurrency/kernel/fwd.hfa"
+#include "concurrency/invoke.h"                                                 // random_state
 
 //---------------------------------------
@@ -221 +223 @@
 //---------------------------------------
 
-static uint32_t seed = 0;                                                               // current seed
-static thread_local uint32_t state;                                             // random state
-
-void set_seed( uint32_t seed_ ) { state = seed = seed_; }
-uint32_t get_seed() { return seed; }
-
 #define GENERATOR LCG
 
 inline uint32_t MarsagliaXor( uint32_t & state ) {
-        if ( unlikely( seed == 0 ) ) set_seed( rdtscl() );
-        else if ( unlikely( state == 0 ) ) state = seed;
         state ^= state << 6;
         state ^= state >> 21;
@@ -239 +233 @@
 
 inline uint32_t LCG( uint32_t & state ) {                               // linear congruential generator
-        if ( unlikely( seed == 0 ) ) set_seed( rdtscl() );
-        else if ( unlikely( state == 0 ) ) state = seed;
         return state = 36969 * (state & 65535) + (state >> 16); // 36969 is NOT prime!
 } // LCG
 
+uint32_t __thread_seed = rdtscl();                                              // global thread seed
+
+void set_seed( uint32_t seed ) { __thread_seed = seed; }
+uint32_t get_seed() { return  __thread_seed; }
 uint32_t prng( PRNG & prng ) with( prng ) { callcnt += 1; return GENERATOR( state ); }
 
-uint32_t prng( void ) { return GENERATOR( state ); }
+uint32_t prng( void ) { return GENERATOR( __thread_seed ); } // [0,UINT_MAX]
 
 //---------------------------------------

libcfa/src/stdlib.hfa

@@ -10 +10 @@
 // Created On       : Thu Jan 28 17:12:35 2016
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Sun Jan  2 22:53:57 2022
-// Update Count     : 594
+// Last Modified On : Mon Jan 10 17:03:18 2022
+// Update Count     : 619
 //
 
@@ -21 +21 @@
 #include <stdlib.h>                                                                             // *alloc, strto*, ato*
 #include <heap.hfa>
+
 
 // Reduce includes by explicitly defining these routines.
@@ -391 +392 @@
 }; // PRNG
 
-extern uint32_t prng( PRNG & prng ) __attribute__(( warn_unused_result )); // [0,UINT_MAX]
+uint32_t prng( PRNG & prng ) __attribute__(( warn_unused_result )); // [0,UINT_MAX]
 static inline {
         void set_seed( PRNG & prng, uint32_t seed_ ) with( prng ) { state = seed = seed_; } // set seed
@@ -402 +403 @@
 } // distribution
 
-extern void set_seed( uint32_t seed );                                  // set per thread seed
-extern uint32_t get_seed();                                                             // get seed
-extern uint32_t prng( void ) __attribute__(( warn_unused_result )); // [0,UINT_MAX]
+void set_seed( uint32_t seed_ ) OPTIONAL_THREAD;
+uint32_t get_seed() __attribute__(( warn_unused_result ));
+uint32_t prng( void ) __attribute__(( warn_unused_result )) OPTIONAL_THREAD; // [0,UINT_MAX]
 static inline {
-        uint32_t prng( uint32_t u ) __attribute__(( warn_unused_result ));
-        uint32_t prng( uint32_t u ) { return prng() % u; }      // [0,u)
-        uint32_t prng( uint32_t l, uint32_t u ) __attribute__(( warn_unused_result ));
-        uint32_t prng( uint32_t l, uint32_t u ) { return prng( u - l + 1 ) + l; } // [l,u]
+        uint32_t prng( uint32_t u ) __attribute__(( warn_unused_result )) { return prng() % u; } // [0,u)
+        uint32_t prng( uint32_t l, uint32_t u ) __attribute__(( warn_unused_result )) { return prng( u - l + 1 ) + l; } // [l,u]
 } // distribution