Merge pull request #25 from jq-rs/larson_sized_test

Larson sized deallocation test

Author: daanx

README.md

@@ -126,6 +126,8 @@ The first set of benchmarks are real world programs and consist of:
    objects to be freed by other threads. Larson and Krishnan observe this
    behavior (which they call _bleeding_) in actual server applications,
    and the benchmark simulates this.
+- __larsonN-sized__: same as the __larsonN__ except it uses sized deallocation calls which
+   have a fast path in some allocators. 
 
 The second set of  benchmarks are stress tests and consist of:
 

bench.sh

@@ -32,6 +32,7 @@ run_sc=0
 
 run_cfrac=0
 run_larson=0
+run_larson_sized=0
 run_ebizzy=0
 run_sh6bench=0
 run_sh8bench=0
@@ -177,6 +178,7 @@ while : ; do
         run_lean=1
         run_xmalloc_test=1
         run_larson=1        
+        run_larson_sized=1
         run_cscratch=1
 	      run_mstress=1
         if [ -z "$darwin" ]; then
@@ -242,6 +244,8 @@ while : ; do
         run_barnes=1;;
     larson)
         run_larson=1;;
+    larson-sized)
+        run_larson_sized=1;;
     ebizzy)
         run_ebizzy=1;;
     sh6bench)
@@ -323,6 +327,7 @@ while : ; do
         echo "  redis                        run redis benchmark"
         echo "  spec=<num>                   run selected spec2017 benchmarks (if available)"
         echo "  larson                       run larsonN"
+        echo "  larson-sized                 run larsonN sized deallocation test"
         echo "  alloc-test                   run alloc-testN"
         echo "  xmalloc-test                 run xmalloc-testN"
         echo "  sh6bench                     run sh6benchN"
@@ -648,6 +653,9 @@ fi
 if test "$run_larson" = "1"; then
   run_test "larsonN" "./larson 5 8 1000 5000 100 4141 $procs"
 fi
+if test "$run_larson_sized" = "1"; then
+  run_test "larsonN-sized" "./larson-sized 5 8 1000 5000 100 4141 $procs"
+fi
 if test "$run_ebizzy" = "1"; then
   run_test "ebizzy" "./ebizzy -t $procs -M -S 2 -s 128"
 fi

bench/CMakeLists.txt

@@ -60,6 +60,11 @@ target_compile_options(larson PRIVATE -Wno-unused-result)
 target_compile_definitions(larson PRIVATE CPP=1)
 target_link_libraries(larson pthread)
 
+add_executable(larson-sized larson/larson.cpp)
+target_compile_options(larson-sized PRIVATE -Wno-unused-result -fsized-deallocation)
+target_compile_definitions(larson-sized PRIVATE CPP=1 SIZED=1)
+target_link_libraries(larson-sized pthread)
+
 if(NOT APPLE)
 add_executable(alloc-test alloc-test/test_common.cpp alloc-test/allocator_tester.cpp)
 target_compile_definitions(alloc-test PRIVATE BENCH=4)

bench/larson/README.md

@@ -1,2 +1,4 @@
 Test driver for memory allocators           
 Author: Paul Larson, [email protected]
+
+Updated to support C++14 sized deallocation with CPP.

bench/larson/larson.cpp

@@ -29,7 +29,6 @@
 #include <pthread.h>
 #endif
 
-
 typedef void * LPVOID;
 typedef long long LONGLONG;
 typedef long DWORD;
@@ -177,7 +176,7 @@ typedef struct thr_data {
   int    max_size ;
 
   char * *array ;
-  int    *blksize ;
+  size_t *blksize ;
   int     asize ;
 
   long   cAllocs ;
@@ -200,7 +199,7 @@ static long lran2(struct lran2_st* d) ;
 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 ;
@@ -280,7 +279,11 @@ int main (int argc, char *argv[])
   printf( "\nSingle-threaded test driver \n") ;
 #endif
 #ifdef CPP
+#if defined(SIZED)
+  printf("C++ version (new and sized delete)\n") ;
+#else
   printf("C++ version (new and delete)\n") ;
+#endif
 #else
   printf("C version (malloc and free)\n") ;
 #endif
@@ -353,7 +356,7 @@ void runloops(long sleep_cnt, int num_chunks )
 {
   int     cblks ;
   int     victim ;
-  int     blk_size ;
+  size_t  blk_size ;
 #ifdef __WIN32__
 	_LARGE_INTEGER ticks_per_sec, start_cnt, end_cnt;
 #else
@@ -388,9 +391,11 @@ void runloops(long sleep_cnt, int num_chunks )
     for( cblks=0; cblks<num_chunks; cblks++){
       victim = lran2(&rgen)%num_chunks ;
 #if defined(CPP)
+#if defined(SIZED)
+    operator delete[] (blkp[victim], blksize[victim]);
+#else
     delete[] blkp[victim] ;
-#elif defined(USE_MALLOC)
-    free(blkp[victim]);
+#endif
 #else
     CUSTOM_FREE(blkp[victim]) ;
 #endif
@@ -578,7 +583,7 @@ static void * exercise_heap( void *pinput)
   thread_data  *pdea;
   int           cblks=0 ;
   int           victim ;
-  long          blk_size ;
+  size_t        blk_size;
   int           range ;
 
   if( stopflag ) return 0;
@@ -592,7 +597,11 @@ static void * exercise_heap( void *pinput)
   for( cblks=0; cblks<pdea->NumBlocks; cblks++){
     victim = lran2(&pdea->rgen)%pdea->asize ;
 #ifdef CPP
+#if defined(SIZED)
+    operator delete[] (pdea->array[victim], pdea->blksize[victim]);
+#else
     delete[] pdea->array[victim] ;
+#endif
 #else
     CUSTOM_FREE(pdea->array[victim]) ;
 #endif
@@ -652,8 +661,9 @@ static void warmup(char **blkp, int num_chunks )
 {
   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 +685,21 @@ static void warmup(char **blkp, int num_chunks )
   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
+#if defined(SIZED)
+    operator delete[] (blkp[victim], blksize[victim]);
+#else
     delete[] blkp[victim] ;
+#endif
 #else
     CUSTOM_FREE(blkp[victim]) ;
 #endif