int main(void)
{
double std=0, ned=0;
#if 0
{
usCount start, end;
start=GetUsCount();
THREADSLEEP(5000);
end=GetUsCount();
printf("Wait was %lf\n", (end-start)/1000000000000.0);
}
#endif
#ifdef WIN32
{ /* Force load of user32.dll so we can debug */
BOOL v;
SystemParametersInfo(SPI_GETBEEP, 0, &v, 0);
}
#endif
if(0)
{
printf("\nTesting standard allocator with %d threads ...\n", THREADS);
std=runtest();
}
if(1)
{
printf("\nTesting nedmalloc with %d threads ...\n", THREADS);
whichmalloc=1;
ned=runtest();
}
#ifdef WIN32
if(0)
{
ULONG data=2;
win32heap=HeapCreate(0, 0, 0);
HeapSetInformation(win32heap, HeapCompatibilityInformation, &data, sizeof(data));
HeapQueryInformation(win32heap, HeapCompatibilityInformation, &data, sizeof(data), NULL);
if(2!=data)
{
printf("The win32 low frag allocator won't work under a debugger!\n");
}
else
{
printf("Testing win32 low frag allocator with %d threads ...\n\n", THREADS);
whichmalloc=2;
runtest();
}
HeapDestroy(win32heap);
}
#endif
if(std && ned)
{ // ned should have more ops/sec
printf("\n\nnedmalloc allocator is %lf times faster than standard\n", ned/std);
}
printf("\nPress a key to trim\n");
getchar();
nedmalloc_trim(0);
#ifdef _MSC_VER
printf("\nPress a key to end\n");
getchar();
#endif
return 0;
}
static double runtest()
{
unsigned int seed=1;
int n, i;
double opspersec=0;
THREADVAR threads[THREADS];
for(n=0; n<THREADS; n++)
{
unsigned int *toallocptr;
int m;
threadstuff[n].ops=0;
times[n]=0;
threadstuff[n].toalloc=toallocptr=calloc(RECORDS, sizeof(unsigned int));
threadstuff[n].allocs=calloc(RECORDS, sizeof(void *));
for(m=0; m<RECORDS; m++)
{
unsigned int size=myrandom(&seed);
if(size<(1<<30))
{ /* Make it two power multiple of less than 512 bytes to
model frequent C++ new's */
size=4<<(size & 7);
}
else
{
size&=0x3FFF; /* < 16Kb */
/*size&=0x1FFF;*/ /* < 8Kb */
/*size=(1<<6)<<(size & 7);*/ /* < 8Kb */
}
*toallocptr++=size;
}
}
#ifdef TORTURETEST
for(n=0; n<THREADS; n++)
{
THREADINIT(&threads[n], n);
}
for(i=0; i<32; i++)
{
int found=-1;
do
{
for(n=0; n<THREADS; n++)
{
THREADSLEEP(100);
if(threadstuff[n].done)
{
found=n;
break;
}
}
} while(found<0);
THREADWAIT(threads[found]);
threads[found]=0;
THREADINIT(&threads[found], found);
printf("Relaunched thread %d\n", found);
}
for(n=THREADS-1; n>=0; n--)
{
THREADWAIT(threads[n]);
threads[n]=0;
}
#else
#if 1
for(n=0; n<THREADS; n++)
{
THREADINIT(&threads[n], n);
}
for(n=THREADS-1; n>=0; n--)
{
THREADWAIT(threads[n]);
threads[n]=0;
}
#else
/* Quick realloc() test */
doRealloc=1;
for(n=0; n<THREADS; n++)
{
THREADINIT(&threads[n], n);
}
for(n=THREADS-1; n>=0; n--)
{
THREADWAIT(threads[n]);
threads[n]=0;
}
#endif
#endif
{
usCount totaltime=0;
int totalops=0;
for(n=0; n<THREADS; n++)
{
totaltime+=times[n];
totalops+=threadstuff[n].ops;
}
opspersec=1000000000000.0*totalops/totaltime*THREADS;
printf("This allocator achieves %lfops/sec under %d threads\n", opspersec, THREADS);
}
for(n=THREADS-1; n>=0; n--)
{
free(threadstuff[n].allocs); threadstuff[n].allocs=0;
free(threadstuff[n].toalloc); threadstuff[n].toalloc=0;
}
return opspersec;
}
static void threadcode(int threadidx)
{
int n;
unsigned int *toallocptr=threadstuff[threadidx].toalloc;
void **allocptr=threadstuff[threadidx].allocs;
unsigned int seed=threadidx;
usCount start;
threadstuff[threadidx].done=0;
/*neddisablethreadcache(0);*/
THREADSLEEP(100);
start=GetUsCount();
#ifdef TORTURETEST
/* A randomised malloc/realloc/free test (torture test) */
for(n=0; n<RECORDS*100; n++)
{
unsigned int r=myrandom(&seed), i;
i=(int)(r % RECORDS);
if(!allocptr[i])
{
allocptr[i]=mallocs[whichmalloc](r & 0x1FFF);
threadstuff[threadidx].ops++;
}
else if(r & (1<<31))
{
allocptr[i]=reallocs[whichmalloc](allocptr[i], r & 0x1FFF);
threadstuff[threadidx].ops++;
}
else
{
frees[whichmalloc](allocptr[i]);
allocptr[i]=0;
}
}
for(n=0; n<RECORDS; n++)
{
if(allocptr[n])
{
frees[whichmalloc](allocptr[n]);
allocptr[n]=0;
}
}
#else
/* A simple stack which allocates and deallocates off the top (speed test) */
for(n=0; n<RECORDS;)
{
#if 1
r=myrandom(&seed);
if(allocptr>threadstuff[threadidx].allocs && (r & 65535)<32760) /*<32760)*/
{ /* free */
--toallocptr;
--allocptr;
--n;
frees[whichmalloc](*allocptr);
*allocptr=0;
}
else
#endif
{
if(doRealloc && allocptr>threadstuff[threadidx].allocs && (r & 1))
{
allocptr[-1]=reallocs[whichmalloc](allocptr[-1], *toallocptr);
}
else
{
allocptr[0]=mallocs[whichmalloc](*toallocptr);
allocptr++;
}
n++;
toallocptr++;
threadstuff[threadidx].ops++;
}
}
while(allocptr>threadstuff[threadidx].allocs)
{
frees[whichmalloc](*--allocptr);
}
#endif
times[threadidx]+=GetUsCount()-start;
neddisablethreadcache(0);
threadstuff[threadidx].done=1;
}
int main(void)
{
double std=0, ned=0;
#if defined(WIN32) && defined(USE_NEDMALLOC_DLL)
/*PatchInNedmallocDLL();*/
#endif
#if 0
{
usCount start, end;
start=GetUsCount();
THREADSLEEP(5000);
end=GetUsCount();
printf("Wait was %lf\n", (end-start)/1000000000000.0);
}
#endif
#ifdef WIN32
#pragma comment(lib, "user32.lib")
{ /* Force load of user32.dll so we can debug */
BOOL v;
SystemParametersInfo(SPI_GETBEEP, 0, &v, 0);
}
#endif
#if 2==TESTTYPE
printf("Running torture test\n"
"-=-=-=-=-=-=-=-=-=-=\n");
#elif 1==TESTTYPE
printf("Running speed test\n"
"-=-=-=-=-=-=-=-=-=\n");
#endif
printf("Block size <= %u, C++ test mode is %s\n", BLOCKSIZE, TESTCPLUSPLUS ? "on" : "off");
if(0)
{
printf("\nTesting standard allocator with %d threads ...\n", THREADS);
std=runtest();
}
if(1)
{
printf("\nTesting nedmalloc with %d threads ...\n", THREADS);
whichmalloc=1;
ned=runtest();
}
#ifdef WIN32
if(0)
{
ULONG data=2;
win32heap=HeapCreate(0, 0, 0);
HeapSetInformation(win32heap, HeapCompatibilityInformation, &data, sizeof(data));
HeapQueryInformation(win32heap, HeapCompatibilityInformation, &data, sizeof(data), NULL);
if(2!=data)
{
printf("The win32 low frag allocator won't work under a debugger!\n");
}
else
{
printf("Testing win32 low frag allocator with %d threads ...\n\n", THREADS);
whichmalloc=2;
runtest();
}
HeapDestroy(win32heap);
}
#endif
if(std && ned)
{ // ned should have more ops/sec
printf("\n\nnedmalloc allocator is %lf times faster than standard\n", ned/std);
}
printf("\nPress a key to trim\n");
getchar();
nedmalloc_trim(0);
#ifdef _MSC_VER
printf("\nPress a key to end\n");
getchar();
#endif
return 0;
}
static double runtest()
{
unsigned int seed=1;
int n, i;
double opspersec=0;
THREADVAR threads[THREADS];
for(n=0; n<THREADS; n++)
{
unsigned int *toallocptr;
int m;
memset(&threadstuff[n].ops, 0, sizeof(threadstuff[n].ops));
times[n]=0;
threadstuff[n].toalloc=toallocptr=calloc(RECORDS, sizeof(unsigned int));
threadstuff[n].allocs=calloc(RECORDS, sizeof(void *));
for(m=0; m<RECORDS; m++)
{
unsigned int size=myrandom(&seed);
#if TESTCPLUSPLUS
if(size&(1<<31))
{ /* Make it two power multiple of less than 512 bytes to
model frequent C++ new's */
size=4<<(size & 7);
}
else
#endif
{
size&=BLOCKSIZE-1;
}
*toallocptr++=size;
}
}
#if 2==TESTTYPE
for(n=0; n<THREADS; n++)
{
THREADINIT(&threads[n], n);
}
for(i=0; i<8; i++)
{
int found=-1;
do
{
for(n=0; n<THREADS; n++)
{
THREADSLEEP(100);
if(threadstuff[n].done)
{
found=n;
break;
}
}
} while(found<0);
THREADWAIT(threads[found]);
threads[found]=0;
#if DEBUG
{
usCount totaltime=0;
int totalops=0, totalmallocs=0, totalreallocs=0;
for(n=0; n<THREADS; n++)
{
totaltime+=times[n];
totalmallocs+=threadstuff[n].ops.mallocs;
totalreallocs+=threadstuff[n].ops.reallocs;
totalops+=threadstuff[n].ops.mallocs+threadstuff[n].ops.reallocs;
}
opspersec=1000000000000.0*totalops/totaltime*THREADS;
printf("This test spent %f%% of its time doing reallocs\n", 100.0*totalreallocs/totalops);
printf("This allocator achieves %lfops/sec under %d threads\n\n", opspersec, THREADS);
}
#endif
THREADINIT(&threads[found], found);
printf("Relaunched thread %d\n", found);
}
for(n=THREADS-1; n>=0; n--)
{
THREADWAIT(threads[n]);
threads[n]=0;
}
#else
#if 1
for(n=0; n<THREADS; n++)
{
THREADINIT(&threads[n], n);
}
for(n=THREADS-1; n>=0; n--)
{
THREADWAIT(threads[n]);
threads[n]=0;
}
#else
/* Quick realloc() test */
doRealloc=1;
for(n=0; n<THREADS; n++)
{
THREADINIT(&threads[n], n);
}
for(n=THREADS-1; n>=0; n--)
{
THREADWAIT(threads[n]);
threads[n]=0;
}
#endif
#endif
{
usCount totaltime=0;
int totalops=0, totalmallocs=0, totalreallocs=0;
for(n=0; n<THREADS; n++)
{
totaltime+=times[n];
totalmallocs+=threadstuff[n].ops.mallocs;
totalreallocs+=threadstuff[n].ops.reallocs;
totalops+=threadstuff[n].ops.mallocs+threadstuff[n].ops.reallocs;
}
opspersec=1000000000000.0*totalops/totaltime*THREADS;
printf("This test spent %f%% of its time doing reallocs\n", 100.0*totalreallocs/totalops);
printf("This allocator achieves %lfops/sec under %d threads\n", opspersec, THREADS);
}
for(n=THREADS-1; n>=0; n--)
{
free(threadstuff[n].allocs); threadstuff[n].allocs=0;
free(threadstuff[n].toalloc); threadstuff[n].toalloc=0;
}
return opspersec;
}
static void threadcode(int threadidx)
{
int n;
unsigned int *toallocptr=threadstuff[threadidx].toalloc;
void **allocptr=threadstuff[threadidx].allocs;
unsigned int r, seed=threadidx;
usCount start;
size_t allocated=0, size;
threadstuff[threadidx].done=0;
/*neddisablethreadcache(0);*/
THREADSLEEP(100);
start=GetUsCount();
#if 2==TESTTYPE
/* A randomised malloc/realloc/free test (torture test) */
for(n=0; n<RECORDS*100; n++)
{
static int reallocflip;
unsigned int i, dorealloc=(reallocflip=!reallocflip);
r=myrandom(&seed);
i=(int)(r % RECORDS);
#if TESTCPLUSPLUS
dorealloc=!(r&(15<<28));
if(r&(1<<31))
{ /* Make it two power multiple of less than 512 bytes to
model frequent C++ new's */
size=4<<(r & 7);
dorealloc=0;
}
else
#endif
size=(size_t)(r & (BLOCKSIZE-1));
if(allocated<MAXMEMORY2 && !allocptr[i])
{
if(!(allocptr[i]=mallocs[whichmalloc](size))) abort();
#if TOUCH
{
volatile char *mem=(volatile char *)allocptr[i];
volatile char *end=mem+size;
for(; mem<end; mem+=4096) *mem;
}
#endif
allocated+=memsizes[whichmalloc](allocptr[i]);
threadstuff[threadidx].ops.mallocs++;
}
else if(allocated<MAXMEMORY2 && dorealloc) /* If not TESTCPLUSPLUS, then how often realloc() gets called depends on how small RECORDS is. */
{
allocated-=memsizes[whichmalloc](allocptr[i]);
if(!(allocptr[i]=reallocs[whichmalloc](allocptr[i], size))) abort();
#if TOUCH
{
volatile char *mem=(volatile char *)allocptr[i];
volatile char *end=mem+size;
for(; mem<end; mem+=4096) *mem;
}
#endif
allocated+=memsizes[whichmalloc](allocptr[i]);
threadstuff[threadidx].ops.reallocs++;
}
else if(allocptr[i])
{
allocated-=memsizes[whichmalloc](allocptr[i]);
frees[whichmalloc](allocptr[i]);
allocptr[i]=0;
threadstuff[threadidx].ops.frees++;
}
}
for(n=0; n<RECORDS; n++)
{
if(allocptr[n])
{
allocated-=memsizes[whichmalloc](allocptr[n]);
frees[whichmalloc](allocptr[n]);
allocptr[n]=0;
threadstuff[threadidx].ops.frees++;
}
}
assert(!allocated);
#elif 1==TESTTYPE
/* A simple stack which allocates and deallocates off the top (speed test) */
for(n=0; n<RECORDS;)
{
#if 1
r=myrandom(&seed);
if(allocptr>threadstuff[threadidx].allocs && (r & 65535)<32760) /*<32760)*/
{ /* free */
--toallocptr;
--allocptr;
--n;
frees[whichmalloc](*allocptr);
*allocptr=0;
threadstuff[threadidx].ops.frees++;
}
else
#endif
{
if(doRealloc && allocptr>threadstuff[threadidx].allocs && (r & 1))
{
if(!(allocptr[-1]=reallocs[whichmalloc](allocptr[-1], *toallocptr))) abort();
#if TOUCH
{
volatile char *mem=(volatile char *)allocptr[-1];
volatile char *end=mem+*toallocptr;
for(; mem<end; mem+=4096) *mem;
}
#endif
threadstuff[threadidx].ops.reallocs++;
}
else
{
if(!(allocptr[0]=mallocs[whichmalloc](*toallocptr))) abort();
#if TOUCH
{
volatile char *mem=(volatile char *)allocptr[0];
volatile char *end=mem+*toallocptr;
for(; mem<end; mem+=4096) *mem;
}
#endif
threadstuff[threadidx].ops.mallocs++;
allocptr++;
}
n++;
toallocptr++;
/*if(!(threadstuff[threadidx].ops & 0xff))
nedtrimthreadcache(0,0);*/
}
}
while(allocptr>threadstuff[threadidx].allocs)
{
frees[whichmalloc](*--allocptr);
threadstuff[threadidx].ops.frees++;
}
#endif
times[threadidx]+=GetUsCount()-start;
neddisablethreadcache(0);
threadstuff[threadidx].done=1;
}