try to auto-detect whether to use 128-bit 4-way SSE2

git-svn-id: https://bitcoin.svn.sourceforge.net/svnroot/bitcoin/trunk@150 1a98c847-1fd6-4fd8-948a-caf3550aa51b
This commit is contained in:
s_nakamoto 2010-09-09 01:00:40 +00:00
parent 14002c8eba
commit 8555125a1a
4 changed files with 71 additions and 5 deletions

View File

@ -184,8 +184,8 @@ public:
void deallocate(void *p, size_type n)
{
//// Bitcoin: can't figure out why this is tripping on a few compiles.
//assert(false);
//// Bitcoin: don't know why this trips, probably a false alarm, depends on the compiler used.
//assert(false);
}
size_type max_size() const {return 0;}

View File

@ -2767,6 +2767,68 @@ inline void SHA256Transform(void* pstate, void* pinput, const void* pinit)
static const int NPAR = 32;
extern void Double_BlockSHA256(const void* pin, void* pout, const void* pinit, unsigned int hash[8][NPAR], const void* init2);
#ifdef __GNUC__
void CallCPUID(int in, int& aret, int& cret)
{
int a, c;
asm (
"mov %2, %%eax; " // in into eax
"cpuid;"
"mov %%eax, %0;" // eax into ret
"mov %%ecx, %1;" // eax into ret
:"=r"(a),"=r"(c) /* output */
:"r"(in) /* input */
:"%eax","%ecx" /* clobbered register */
);
aret = a;
cret = c;
}
bool Detect128BitSSE2()
{
int a, c, nBrand;
CallCPUID(0, a, nBrand);
bool fIntel = (nBrand == 0x6c65746e); // ntel
bool fAMD = (nBrand == 0x444d4163); // cAMD
struct
{
unsigned int nStepping : 4;
unsigned int nModel : 4;
unsigned int nFamily : 4;
unsigned int nProcessorType : 2;
unsigned int nUnused : 2;
unsigned int nExtendedModel : 4;
unsigned int nExtendedFamily : 8;
}
cpu;
CallCPUID(1, a, c);
memcpy(&cpu, &a, sizeof(cpu));
int nFamily = cpu.nExtendedFamily + cpu.nFamily;
int nModel = cpu.nExtendedModel*16 + cpu.nModel;
// We need Intel Nehalem or AMD K10 or better for 128bit SSE2
// Nehalem = i3/i5/i7 and some Xeon
// K10 = Opterons with 4 or more cores, Phenom, Phenom II, Athlon II
// Intel Core i5 family 6, model 26 or 30
// Intel Core i7 family 6, model 26 or 30
// Intel Core i3 family 6, model 37
// AMD Phenom family 16, model 10
bool fUseSSE2 = ((fIntel && nFamily * 10000 + nModel >= 60026) ||
(fAMD && nFamily * 10000 + nModel >= 160010));
static bool fPrinted;
if (!fPrinted)
{
fPrinted = true;
printf("CPUID %08x family %d, model %d, stepping %d, fUseSSE2=%d\n", nBrand, nFamily, nModel, cpu.nStepping, fUseSSE2);
}
return fUseSSE2;
}
#else
bool Detect128BitSSE2() { return false; }
#endif
@ -2774,6 +2836,9 @@ void BitcoinMiner()
{
printf("BitcoinMiner started\n");
SetThreadPriority(THREAD_PRIORITY_LOWEST);
bool f4WaySSE2 = Detect128BitSSE2();
if (mapArgs.count("-4way"))
f4WaySSE2 = (mapArgs["-4way"] != "0");
CKey key;
key.MakeNewKey();
@ -2913,7 +2978,6 @@ void BitcoinMiner()
//
// Search
//
bool f4WaySSE2 = mapArgs.count("-4way");
int64 nStart = GetTime();
uint256 hashTarget = CBigNum().SetCompact(pblock->nBits).getuint256();
uint256 hashbuf[2];

View File

@ -1160,7 +1160,9 @@ void ThreadMessageHandler2(void* parg)
pnode->Release();
}
// Wait and allow messages to bunch up
// Wait and allow messages to bunch up.
// Reduce vnThreadsRunning so StopNode has permission to exit while
// we're sleeping, but we must always check fShutdown after doing this.
vnThreadsRunning[2]--;
Sleep(100);
if (fRequestShutdown)

View File

@ -23,7 +23,7 @@ class CAutoFile;
static const unsigned int MAX_SIZE = 0x02000000;
static const int VERSION = 312;
static const char* pszSubVer = ".1";
static const char* pszSubVer = ".2";