floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #3794 from bartoldeman/benchmark-align-malloc 

Benchmarks: align malloc'ed buffers.
This commit is contained in:
Martin Kroeker 2022-10-21 16:13:58 +02:00 committed by GitHub
parent ad424fce08 9e6b060bf3
commit 8c10f0abba
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 18 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
benchmark/bench.h
View File

@ -74,6 +74,24 @@ static void *huge_malloc(BLASLONG size){
#endif #endif
/* Benchmarks should allocate with cacheline (often 64 bytes) alignment
to avoid unreliable results. This technique, storing the allocated
pointer value just before the aligned memory, doesn't require
C11's aligned_alloc for compatibility with older compilers. */
static void *aligned_alloc_cacheline(size_t n)
{
void *p = malloc((size_t)(void *) + n + L1_DATA_LINESIZE - 1);
if (p) {
void **newp = (void **)
(((uintptr_t)p + L1_DATA_LINESIZE) & (uintptr_t)-L1_DATA_LINESIZE);
newp[-1] = p;
p = newp;
}
return p;
}
#define malloc aligned_alloc_cacheline
#define free(p) free((p) ? ((void **)(p))[-1] : (p))
#if defined(__WIN32__) || defined(__WIN64__) || !defined(_POSIX_TIMERS) #if defined(__WIN32__) || defined(__WIN64__) || !defined(_POSIX_TIMERS)
struct timeval start, stop; struct timeval start, stop;
#elif defined(__APPLE__) #elif defined(__APPLE__)