Larson sized deallocation test

[jq-rs]

Larson sized deallocation test. This will call delete[] operator with size to test sized deallocation path.

The test can be run with following:

../../bench.sh alla larson-sized

[mjp41]

Looks interesting to me as a test. I have posted a diff here to make it easier to see what you have changed.

--- larson\larson.cpp   2021-02-26 10:16:01.360644200 +0000
+++ larson_sized\larson.cpp     2021-03-04 11:20:20.151509500 +0000
@@ -29,7 +29,6 @@
 #include <pthread.h>
 #endif

-
 typedef void * LPVOID;
 typedef long long LONGLONG;
 typedef long DWORD;
@@ -177,7 +176,7 @@
   int    max_size ;

   char * *array ;
-  int    *blksize ;
+  size_t *blksize ;
   int     asize ;

   long   cAllocs ;
@@ -200,7 +199,7 @@
 ULONG CountReservedSpace() ;

 char *          blkp[MAX_BLOCKS] ;
-int             blksize[MAX_BLOCKS] ;
+size_t          blksize[MAX_BLOCKS] ;
 long            seqlock=0 ;
 struct lran2_st rgen ;
 int             min_size=10, max_size=500 ;
@@ -353,7 +352,7 @@
 {
   int     cblks ;
   int     victim ;
-  int     blk_size ;
+  size_t  blk_size ;
 #ifdef __WIN32__
        _LARGE_INTEGER ticks_per_sec, start_cnt, end_cnt;
 #else
@@ -388,7 +387,7 @@
     for( cblks=0; cblks<num_chunks; cblks++){
       victim = lran2(&rgen)%num_chunks ;
 #if defined(CPP)
-    delete[] blkp[victim] ;
+    operator delete[] (blkp[victim], blksize[victim]);
 #elif defined(USE_MALLOC)
     free(blkp[victim]);
 #else
@@ -578,7 +577,7 @@
   thread_data  *pdea;
   int           cblks=0 ;
   int           victim ;
-  long          blk_size ;
+  size_t        blk_size;
   int           range ;

   if( stopflag ) return 0;
@@ -592,7 +591,7 @@
   for( cblks=0; cblks<pdea->NumBlocks; cblks++){
     victim = lran2(&pdea->rgen)%pdea->asize ;
 #ifdef CPP
-    delete[] pdea->array[victim] ;
+    operator delete[] (pdea->array[victim], pdea->blksize[victim]);
 #else
     CUSTOM_FREE(pdea->array[victim]) ;
 #endif
@@ -652,8 +651,9 @@
 {
   int     cblks ;
   int     victim ;
-  int     blk_size ;
+  size_t  blk_size ;
   LPVOID  tmp ;
+  size_t  tmp_sz;


   for( cblks=0; cblks<num_chunks; cblks++){
@@ -675,14 +675,17 @@
   for( cblks=num_chunks; cblks > 0 ; cblks--){
     victim = lran2(&rgen)%cblks ;
     tmp = blkp[victim] ;
+    tmp_sz = blksize[victim];
     blkp[victim]  = blkp[cblks-1] ;
+    blksize[victim] = blksize[cblks-1];
     blkp[cblks-1] = (char *) tmp ;
+    blksize[cblks-1] = tmp_sz;
   }

   for( cblks=0; cblks<4*num_chunks; cblks++){
     victim = lran2(&rgen)%num_chunks ;
 #ifdef CPP
-    delete[] blkp[victim] ;
+    operator delete[] (blkp[victim], blksize[victim]);
 #else
     CUSTOM_FREE(blkp[victim]) ;
 #endif

[jq-rs]

Just realized that we could have used the same source file even though there are two tests. I'll fix this, then the diff will be obvious.

[jq-rs]

Below is an example test run. Especially tc gets a boost with sized deallocation in my environment.

larsonN-sized sys 3.917 256496 194.51 2.79 0 68867
larsonN-sized mi 3.085 322616 198.25 0.58 0 86489
larsonN-sized smi 4.060 314540 198.56 0.51 0 82905
larsonN-sized tc 2.987 171368 198.02 0.42 0 42174
larsonN-sized je 3.113 263092 196.99 1.04 0 126367
larsonN-sized tbb 5.473 352276 198.70 0.82 0 109019
larsonN-sized sn 2.959 336252 199.19 0.42 0 83373
larsonN-sized rp 3.226 303708 197.73 0.37 0 75941
larsonN-sized hd 6.198 209420 198.90 0.29 0 51741

larsonN sys 3.792 262788 193.92 3.12 0 72135
larsonN mi 2.985 321508 198.27 0.57 0 86728
larsonN smi 4.023 317756 198.01 0.50 0 84125
larsonN tc 3.347 170608 198.28 0.41 0 41942
larsonN je 2.927 265116 198.16 1.04 0 130669
larsonN tbb 5.355 352916 198.90 0.84 0 109238
larsonN sn 2.907 332660 198.27 0.43 0 82496
larsonN rp 3.188 305736 198.86 0.42 0 76505
larsonN hd 6.381 204472 198.82 0.24 0 50544

[daanx]

Apologies for the super late reply :-( I remember being a tad hesitant when I first saw this as Larson is used in many papers and if we change it we should perhaps rename it? So, I think it would be better as larson2 or something? not sure

[jq-rs]

I have already renamed this to Larson-sized. The original Larson-test remains there. I can change this to Larson2, if you prefer that one, but I personally cannot see how it is better. Naming is always hard :)

[daanx]

I have already renamed this to Larson-sized. The original Larson-test remains there. I can change this to Larson2, if you prefer that one, but I personally cannot see how it is better. Naming is always hard :)

Ah, I did not notice that -- this is great. Thanks so much!