static void testPerformance(const char *pszSub, uint8_t const *pabInstrs, uintptr_t uEndPtr, DISCPUMODE enmDisCpuMode)
{
RTTestISubF("Performance - %s", pszSub);
size_t const cbInstrs = uEndPtr - (uintptr_t)pabInstrs;
uint64_t cInstrs = 0;
uint64_t nsStart = RTTimeNanoTS();
for (uint32_t i = 0; i < _512K; i++) /* the samples are way to small. :-) */
{
for (size_t off = 0; off < cbInstrs; cInstrs++)
{
uint32_t cb = 1;
DISSTATE Dis;
DISInstrWithReader((uintptr_t)&pabInstrs[off], enmDisCpuMode, testReadBytes, NULL, &Dis, &cb);
off += cb;
}
}
uint64_t cNsElapsed = RTTimeNanoTS() - nsStart;
RTTestIValueF(cNsElapsed, RTTESTUNIT_NS, "%s-Total", pszSub);
RTTestIValueF(cNsElapsed / cInstrs, RTTESTUNIT_NS_PER_CALL, "%s-per-instruction", pszSub);
}
static void doTest(RTTEST hTest)
{
NOREF(hTest);
uint32_t iAllocCpu = 0;
while (iAllocCpu < RTCPUSET_MAX_CPUS)
{
const uint32_t cbTestSet = _1M * 32;
const uint32_t cIterations = 384;
/*
* Change CPU and allocate a chunk of memory.
*/
RTTESTI_CHECK_RC_OK_RETV(RTThreadSetAffinityToCpu(RTMpCpuIdFromSetIndex(iAllocCpu)));
void *pvTest = RTMemPageAlloc(cbTestSet); /* may be leaked, who cares */
RTTESTI_CHECK_RETV(pvTest != NULL);
memset(pvTest, 0xef, cbTestSet);
/*
* Do the tests.
*/
uint32_t iAccessCpu = 0;
while (iAccessCpu < RTCPUSET_MAX_CPUS)
{
RTTESTI_CHECK_RC_OK_RETV(RTThreadSetAffinityToCpu(RTMpCpuIdFromSetIndex(iAccessCpu)));
/*
* The write test.
*/
RTTimeNanoTS(); RTThreadYield();
uint64_t u64StartTS = RTTimeNanoTS();
for (uint32_t i = 0; i < cIterations; i++)
{
ASMCompilerBarrier(); /* paranoia */
memset(pvTest, i, cbTestSet);
}
uint64_t const cNsElapsedWrite = RTTimeNanoTS() - u64StartTS;
uint64_t cMBPerSec = (uint64_t)( ((uint64_t)cIterations * cbTestSet) /* bytes */
/ ((long double)cNsElapsedWrite / RT_NS_1SEC_64) /* seconds */
/ _1M /* MB */ );
RTTestIValueF(cMBPerSec, RTTESTUNIT_MEGABYTES_PER_SEC, "cpu%02u-mem%02u-write", iAllocCpu, iAccessCpu);
/*
* The read test.
*/
memset(pvTest, 0, cbTestSet);
RTTimeNanoTS(); RTThreadYield();
u64StartTS = RTTimeNanoTS();
for (uint32_t i = 0; i < cIterations; i++)
{
#if 1
size_t register u = 0;
size_t volatile *puCur = (size_t volatile *)pvTest;
size_t volatile *puEnd = puCur + cbTestSet / sizeof(size_t);
while (puCur != puEnd)
u += *puCur++;
#else
ASMCompilerBarrier(); /* paranoia */
void *pvFound = memchr(pvTest, (i & 127) + 1, cbTestSet);
RTTESTI_CHECK(pvFound == NULL);
#endif
}
uint64_t const cNsElapsedRead = RTTimeNanoTS() - u64StartTS;
cMBPerSec = (uint64_t)( ((uint64_t)cIterations * cbTestSet) /* bytes */
/ ((long double)cNsElapsedRead / RT_NS_1SEC_64) /* seconds */
/ _1M /* MB */ );
RTTestIValueF(cMBPerSec, RTTESTUNIT_MEGABYTES_PER_SEC, "cpu%02u-mem%02u-read", iAllocCpu, iAccessCpu);
/*
* The read/write test.
*/
RTTimeNanoTS(); RTThreadYield();
u64StartTS = RTTimeNanoTS();
for (uint32_t i = 0; i < cIterations; i++)
{
ASMCompilerBarrier(); /* paranoia */
memcpy(pvTest, (uint8_t *)pvTest + cbTestSet / 2, cbTestSet / 2);
}
uint64_t const cNsElapsedRW = RTTimeNanoTS() - u64StartTS;
cMBPerSec = (uint64_t)( ((uint64_t)cIterations * cbTestSet) /* bytes */
/ ((long double)cNsElapsedRW / RT_NS_1SEC_64) /* seconds */
/ _1M /* MB */ );
RTTestIValueF(cMBPerSec, RTTESTUNIT_MEGABYTES_PER_SEC, "cpu%02u-mem%02u-read-write", iAllocCpu, iAccessCpu);
/*
* Total time.
*/
RTTestIValueF(cNsElapsedRead + cNsElapsedWrite + cNsElapsedRW, RTTESTUNIT_NS,
"cpu%02u-mem%02u-time", iAllocCpu, iAccessCpu);
/* advance */
iAccessCpu = getNextCpu(iAccessCpu);
}
/*
* Clean up and advance to the next CPU.
*/
RTMemPageFree(pvTest, cbTestSet);
iAllocCpu = getNextCpu(iAllocCpu);
}
}
