OmniNX/Horizon-OC-pro
2
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

benchmark-toolbox: make cpu readings *actually* match memtoolkit

Browse Source
This commit is contained in:
Lightos1 committed 2026-10-03 17:01:21 +02:00
1 parent 119a9a6ea4
commit 82d3f2a93c
1 file changed
+10 -14
Download Patch File Download Diff File Expand all files Collapse all files
Source/Benchmark-Toolbox/source/bench.c
+10 -14
View File
@@ -39,6 +39,7 @@ typedef struct {
long size; long size;
int core; int core;
bool created; bool created;
uint64_t ticks;
} bw_worker_t; } bw_worker_t;
typedef struct bw_pool { typedef struct bw_pool {
@@ -58,7 +59,9 @@ static void bw_worker_entry(void *arg) {
bench_kernel_fn f = w->func; bench_kernel_fn f = w->func;
if (!f) if (!f)
break; break;
uint64_t t0 = armGetSystemTick();
f(w->dst, w->src, w->size); f(w->dst, w->src, w->size);
w->ticks = armGetSystemTick() - t0;
semaphoreSignal(&p->done); semaphoreSignal(&p->done);
} }
} }
@@ -115,13 +118,14 @@ static double bw_run_once(bw_pool_t *p, bench_kernel_fn f, int64_t *dstbuf, int6
w->src = srcbuf + (size * i) / (long)sizeof(int64_t); w->src = srcbuf + (size * i) / (long)sizeof(int64_t);
w->size = size; w->size = size;
} }
uint64_t t0 = armGetSystemTick();
for (int i = 0; i < threads; i++) for (int i = 0; i < threads; i++)
semaphoreSignal(&p->workers[i].start); semaphoreSignal(&p->workers[i].start);
for (int i = 0; i < threads; i++) for (int i = 0; i < threads; i++)
semaphoreWait(&p->done); semaphoreWait(&p->done);
uint64_t t1 = armGetSystemTick(); uint64_t sum = 0;
return (double)armTicksToNs(t1 - t0) / 1000000000.0; for (int i = 0; i < threads; i++)
sum += p->workers[i].ticks;
return (double)armTicksToNs(sum) / 1000000000.0 / threads;
} }
#define BW_GPR_CLOBBERS \ #define BW_GPR_CLOBBERS \
@@ -201,14 +205,14 @@ static double gettime(void) {
static double bandwidth_bench_helper(bw_pool_t *pool, int threads, int64_t *dstbuf, int64_t *srcbuf, long size, bench_kernel_fn f) { static double bandwidth_bench_helper(bw_pool_t *pool, int threads, int64_t *dstbuf, int64_t *srcbuf, long size, bench_kernel_fn f) {
int i, loopcount, innerloopcount, n; int i, loopcount, innerloopcount, n;
double t, speed, maxspeed, s, s0, s1, s2; double t, speed, maxspeed;
if (!pool || !pool->ok) if (!pool || !pool->ok)
return 0.; return 0.;
s = s0 = s1 = s2 = 0.;
maxspeed = 0.; maxspeed = 0.;
for (n = 0; n < MAXREPEATS; n++) { for (n = 0; n < MAXREPEATS; n++) {
bw_run_once(pool, f, dstbuf, srcbuf, size);
loopcount = 0; loopcount = 0;
innerloopcount = 1; innerloopcount = 1;
t = 0.; t = 0.;
@@ -217,19 +221,11 @@ static double bandwidth_bench_helper(bw_pool_t *pool, int threads, int64_t *dstb
for (i = 0; i < innerloopcount; i++) for (i = 0; i < innerloopcount; i++)
t += bw_run_once(pool, f, dstbuf, srcbuf, size); t += bw_run_once(pool, f, dstbuf, srcbuf, size);
innerloopcount *= 2; innerloopcount *= 2;
} while (t < 0.5); } while (t < 0.25);
speed = (double)size * threads * loopcount / t / 1000000.; speed = (double)size * threads * loopcount / t / 1000000.;
s0 += 1.;
s1 += speed;
s2 += speed * speed;
if (speed > maxspeed) if (speed > maxspeed)
maxspeed = speed; maxspeed = speed;
if (s0 > 2.) {
s = sqrt((s0 * s2 - s1 * s1) / (s0 * (s0 - 1)));
if (s < maxspeed / 1000.)
break;
}
} }
return maxspeed; return maxspeed;
} }