LBRYCommunity/cpuminer
1
0
Fork 0
Code Issues 3 Pull requests Projects Releases Packages Wiki Activity

sha256_4way minor optimizations

Browse source 
http://www.bitcoin.org/smf/index.php?topic=820.msg40022#msg40022

Credit: BeeCee1
This commit is contained in:
Jeff Garzik 2011-01-29 00:36:32 -05:00 committed by Jeff Garzik
parent e41171424a
commit 8504cf4803
1 changed files with 5 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
sha256_4way.c
View file

@ -73,7 +73,7 @@ static inline void store_epi32(const __m128i x, unsigned int *x0, unsigned int *
*x0 = box.ret[3]; *x1 = box.ret[2]; *x2 = box.ret[1]; *x3 = box.ret[0]; *x0 = box.ret[3]; *x1 = box.ret[2]; *x2 = box.ret[1]; *x3 = box.ret[0];
} }
#define add4(x0, x1, x2, x3) _mm_add_epi32(_mm_add_epi32(_mm_add_epi32(x0, x1), x2), x3) #define add4(x0, x1, x2, x3) _mm_add_epi32(_mm_add_epi32(x0, x1),_mm_add_epi32( x2,x3))
#define add5(x0, x1, x2, x3, x4) _mm_add_epi32(add4(x0, x1, x2, x3), x4) #define add5(x0, x1, x2, x3, x4) _mm_add_epi32(add4(x0, x1, x2, x3), x4)
#define SHA256ROUND(a, b, c, d, e, f, g, h, i, w) \ #define SHA256ROUND(a, b, c, d, e, f, g, h, i, w) \
@ -150,12 +150,14 @@ static void DoubleBlockSHA256(const void* pin, void* pad, const void *pre, unsig
__m128i w8, w9, w10, w11, w12, w13, w14, w15; __m128i w8, w9, w10, w11, w12, w13, w14, w15;
__m128i T1; __m128i T1;
__m128i a, b, c, d, e, f, g, h; __m128i a, b, c, d, e, f, g, h;
__m128i nonce; __m128i nonce, preNonce;
/* nonce offset for vector */ /* nonce offset for vector */
__m128i offset = _mm_set_epi32(0x00000003, 0x00000002, 0x00000001, 0x00000000); __m128i offset = _mm_set_epi32(0x00000003, 0x00000002, 0x00000001, 0x00000000);
preNonce = _mm_add_epi32(_mm_set1_epi32(In[3]), offset);
for(k = 0; k<NPAR; k+=4) { for(k = 0; k<NPAR; k+=4) {
w0 = _mm_set1_epi32(In[0]); w0 = _mm_set1_epi32(In[0]);
w1 = _mm_set1_epi32(In[1]); w1 = _mm_set1_epi32(In[1]);
@ -175,9 +177,7 @@ static void DoubleBlockSHA256(const void* pin, void* pad, const void *pre, unsig
w15 = _mm_set1_epi32(In[15]); w15 = _mm_set1_epi32(In[15]);
/* hack nonce into lowest byte of w3 */ /* hack nonce into lowest byte of w3 */
nonce = _mm_set1_epi32(In[3]); nonce = _mm_add_epi32(preNonce, _mm_set1_epi32(k));
nonce = _mm_add_epi32(nonce, offset);
nonce = _mm_add_epi32(nonce, _mm_set1_epi32(k));
w3 = nonce; w3 = nonce;
a = _mm_set1_epi32(hPre[0]); a = _mm_set1_epi32(hPre[0]);