關(guān)鍵詞：
ESM6802
iMX6-WEC7
摘要：WEC7一個(gè)最重要的特性就是對(duì)多核處理器的支持(Symmetric Multi-Processing(SMP))，下面將通過應(yīng)用程序來測(cè)試在單核和多核情況下系統(tǒng)的執(zhí)行情況，為了更直觀的比較，同時(shí)參與測(cè)試的還有ESM3354，ESM3354是基于TI Coertex-A8處理器的工控主板，CPU主頻1GHz，同樣預(yù)裝WEC7操作系統(tǒng)。
ESM6802是英創(chuàng)公司推出的基于Freescale i.MX6DL雙核處理器(ARM Cortex-A9，主頻1GHz)的高性能工控主板，預(yù)裝正版Windows Embedded Compact 7(WEC7)嵌入式操作系統(tǒng)，WEC7一個(gè)最重要的特性就是對(duì)多核處理器的支持(Symmetric Multi-Processing(SMP))，下面將通過應(yīng)用程序來測(cè)試在單核和多核情況下系統(tǒng)的執(zhí)行情況，為了更直觀的比較，同時(shí)參與測(cè)試的還有ESM3354，ESM3354是基于TI Coertex-A8處理器的工控主板，CPU主頻1GHz，同樣預(yù)裝WEC7操作系統(tǒng)。
所設(shè)計(jì)的測(cè)試程序代碼如下，其中的TestSmp函數(shù)有兩個(gè)輸入?yún)?shù)，第一參數(shù)表示要?jiǎng)?chuàng)建測(cè)試線程的數(shù)量，第二個(gè)參數(shù)為所創(chuàng)建線程的運(yùn)行時(shí)長。cbTestSmp是被創(chuàng)建的測(cè)試線程，測(cè)試線程主要是在一個(gè)while循環(huán)中，反復(fù)讀取內(nèi)存變量然后與預(yù)設(shè)值進(jìn)行比較，在運(yùn)行設(shè)定的時(shí)間后自動(dòng)退出循環(huán)，其中的threadParam->loops變量會(huì)記錄下while循環(huán)總共執(zhí)行的次數(shù)。
typedef struct _SMP_THREAD_PARAM
{
UINT32 durationMs;
UINT32 threadId;
UINT64 loops;
BOOL bSetAffinity;
UINT32 sandBoxSize;
LPVOID sandBoxStart;
}SMP_THREAD_PARAM, *PSMP_THREAD_PARAM;
ULONG cbTestSmp(LPVOID param)
{
PSMP_THREAD_PARAM threadParam = (PSMP_THREAD_PARAM)param;
DWORD tStart = GetTickCount();
UINT8 *buffer = (UINT8 *)threadParam->sandBoxStart;
wprintf(L"Ahou, Thread %d, running for %d ms
", threadParam->threadId,
threadParam->durationMs);
// Write to sandbox
for (UINT32 i = 0; i < threadParam->sandBoxSize; i++)
{
buffer[i] = (UINT8)(i);
}
while ( (GetTickCount() - tStart) < threadParam->durationMs)
{
// Read back from sandbox
for (UINT32 i = 0; i < threadParam->sandBoxSize; i++)
{
if (buffer[i] != (UINT8)(i))
{
wprintf(L"Thread %d : error at byte %d for loop %I64d !!
",
threadParam->threadId, i, threadParam->loops);
}
}
threadParam->loops++;
}
wprintf(L"Thread %d : terminating
", threadParam->threadId);
return 0;
}
void TestSmp(UINT32 nNumOfThread, UINT32 durationMs)
{
UINT32 i;
PSMP_THREAD_PARAM threadParams;
HANDLE *threadHandles;
UINT64 totalLoops = 0;
UINT32 sandBoxSize = 1024 * 128; // 128 kB
HANDLE h_array[1];
threadParams = (PSMP_THREAD_PARAM)malloc(nNumOfThread * sizeof(SMP_THREAD_PARAM));
if (threadParams == NULL)
{
wprintf(L"Failed allocating thread params !
");
return;
}
threadHandles = (HANDLE *)malloc(nNumOfThread * sizeof(HANDLE));
if (threadHandles == NULL)
{
wprintf(L"Failed allocating thread handles !
");
return;
}
for (i = 0; i < nNumOfThread; i++)
{
threadParams[i].bSetAffinity = TRUE;
threadParams[i].threadId = i;
threadParams[i].durationMs = durationMs;
threadParams[i].loops = 0;
threadParams[i].sandBoxSize = sandBoxSize;
threadParams[i].sandBoxStart = malloc(sandBoxSize);
threadHandles[i] = CreateThread(NULL, 0, cbTestSmp, &threadParams[i], 0, NULL);
wprintf(L"Thread handle %d : 0x%x
", i, threadHandles[i]);
}
h_array[0] = threadHandles[0];
DWORD res = WaitForSingleObject(h_array[0], INFINITE);
Sleep(500);
if (res == WAIT_TIMEOUT)
{
wprintf(L"Timeout waiting for threads !
");
}
else
{
wprintf(L"All threads exited
");
}
for (i = 0; i < nNumOfThread; i++)
{
wprintf(L"Thread %d did run %I64d loops
", i, threadParams[i].loops);
totalLoops += threadParams[i].loops;
free(threadParams[i].sandBoxStart);
CloseHandle(threadHandles[i]);
}
wprintf(L"Total number of loops %I64d (%I64d millions)
", totalLoops,
totalLoops / 1000000);
free(threadHandles);
free(threadParams);
}
將上述測(cè)試代碼編譯生成為exe文件，分別在ESM3354和ESM6802上運(yùn)行，設(shè)置while循環(huán)的執(zhí)行時(shí)間均為10000ms，測(cè)試結(jié)果如下：
1、創(chuàng)建單個(gè)線程
測(cè)試主板與線程ESM3354（1GHz單核 Cortex-A8）ESM6802（1GHz雙核Cortex-A9）
循環(huán)次數(shù)67917493
當(dāng)測(cè)試程序只創(chuàng)建一個(gè)測(cè)試線程時(shí)，ESM3354的while循環(huán)執(zhí)行了6791次，ESM6802執(zhí)行7493次，雖然ESM6802為雙核處理器，但由于程序只有一個(gè)線程，即同一時(shí)刻只有一個(gè)線程在運(yùn)行，所以在相同的時(shí)間內(nèi)，循環(huán)的次數(shù)僅略多于ESM3354。由于ESM3354和ESM6802的CPU主頻同樣都是1GHz，所以可以認(rèn)為ESM6802多出的循環(huán)次數(shù)也就是Cortex-A8與Cortex-A9在代碼執(zhí)行效率上的差別。
2、創(chuàng)建兩個(gè)線程
測(cè)試主板與線程ESM3354（1GHz單核 Cortex-A8）ESM6802（1GHz雙核Cortex-A9）
線程1循環(huán)次數(shù)33907438
線程2循環(huán)次數(shù)34427452
總循環(huán)次數(shù)683214890
當(dāng)測(cè)試程序創(chuàng)建了兩個(gè)線程時(shí)，ESM3354會(huì)將CPU資源大約平均的分配給兩個(gè)線程，如上表中線程1執(zhí)行了3390次，線程2執(zhí)行了3442次，兩個(gè)線程總共執(zhí)行的次數(shù)與只創(chuàng)建單個(gè)線程測(cè)試時(shí)的循環(huán)次數(shù)相當(dāng)。ESM6802為雙核CPU，在測(cè)試程序有兩個(gè)線程的情況下，在同一時(shí)刻兩個(gè)線程可以同時(shí)運(yùn)行，所以總的循環(huán)次數(shù)大約是單個(gè)線程測(cè)試時(shí)的兩倍。
通過上面的測(cè)試可以看到，在多線程情況下，如果操作系統(tǒng)支持多核處理器，那么雙核CPU的運(yùn)算能力將是單核CPU的兩倍。
