popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2025-09-17 23:44:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

v3.5.13

Browse Source
This commit is contained in:
Jay D Dee
2017-03-10 11:38:58 -05:00
parent 38c6f23b66
commit f1f9e821a2
18 changed files with 139 additions and 342 deletions
Download Patch File Download Diff File Expand all files Collapse all files

190
algo/luffa/sse2/luffa_for_sse2.c
View File

@@ -23,22 +23,6 @@
#include "avxdefs.h"
#include "luffa_for_sse2.h"
#if defined (__AVX2__)
#define MULT256(a) \
a = _mm256_xor_si256( \
_mm256_and_si256( _mm256_srli_si256( a, 4 ), \
_mm256_set_epi32( \
0, 0xffffffff, 0xffffffff, 0xffffffff, \
0, 0xffffffff, 0xffffffff, 0xffffffff ) ), \
_mm256_permutevar8x32_epi32( \
_mm256_and_si256( _mm256_srli_si256( a, 4 ), \
_mm256_set_epi32( 0xffffffff, 0, 0, 0, \
0xffffffff, 0,0, 0 ) ), \
_mm256_set_epi32( 0, 0, 0, 0, 0, 0, 0, 0x00800800 ) ) )
#endif // __AVX2__
#define MULT2(a0,a1) do \
{ \
__m128i b = _mm_xor_si128( a0, _mm_shuffle_epi32( _mm_and_si128(a1,MASK), 16 ) ); \
@@ -46,17 +30,6 @@ _mm256_set_epi32( 0, 0, 0, 0, 0, 0, 0, 0x00800800 ) ) )
a1 = _mm_or_si128( _mm_srli_si128(a1,4), _mm_slli_si128(b,12) ); \
} while(0)
/*
#define MULT2(a0,a1) do \
{ \
__m128i b; \
a0 = _mm_xor_si128( a0, _mm_shuffle_epi32( _mm_and_si128(a1,MASK), 16 ) ); \
b = a0; \
a0 = _mm_or_si128( _mm_srli_si128(a0,4), _mm_slli_si128(a1,12) ); \
a1 = _mm_or_si128( _mm_srli_si128(a1,4), _mm_slli_si128(b,12) ); \
} while(0)
*/
#define STEP_PART(x,c,t)\
SUBCRUMB(*x,*(x+1),*(x+2),*(x+3),*t);\
SUBCRUMB(*(x+5),*(x+6),*(x+7),*(x+4),*t);\
@@ -213,17 +186,10 @@ _mm256_set_epi32( 0, 0, 0, 0, 0, 0, 0, 0x00800800 ) ) )
#define MIXTON1024(r0,r1,r2,r3,s0,s1,s2,s3,p0,p1,p2,p3,q0,q1,q2,q3)\
NMLTOM1024(r0,r1,r2,r3,s0,s1,s2,s3,p0,p1,p2,p3,q0,q1,q2,q3);
//#if defined (__AVX2__)
// static void rnd512( hashState_luffa *state, __m256i msg );
//#else
static void rnd512( hashState_luffa *state, __m128i msg1, __m128i msg0 );
//static void rnd512( hashState_luffa *state );
//#endif
static void rnd512( hashState_luffa *state, __m128i msg1, __m128i msg0 );
static void finalization512( hashState_luffa *state, uint32 *b );
/* initial values of chaining variables */
static const uint32 IV[40] __attribute((aligned(16))) = {
0xdbf78465,0x4eaa6fb4,0x44b051e0,0x6d251e69,
@@ -306,12 +272,8 @@ HashReturn update_luffa( hashState_luffa *state, const BitSequence *data,
// full blocks
for ( i = 0; i < blocks; i++ )
{
//#if defined (__AVX2__)
// rnd512( state, mm256_byteswap_epi32( cast_m256i( data ) ) ),
//#else
rnd512( state, mm_byteswap_epi32( casti_m128i( data, 1 ) ),
mm_byteswap_epi32( casti_m128i( data, 0 ) ) );
//#endif
data += MSG_BLOCK_BYTE_LEN;
}
@@ -335,23 +297,14 @@ HashReturn final_luffa(hashState_luffa *state, BitSequence *hashval)
if ( state->rembytes )
{
// not empty, data is in buffer
//#if defined (__AVX2__)
// rnd512( state, cast_m256i( state->buffer ) );
//#else
rnd512( state, casti_m128i( state->buffer, 1 ),
casti_m128i( state->buffer, 0 ) );
//#endif
}
else
{
// empty pad block, constant data
//#if defined (__AVX2__)
// rnd512( state, _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0,
// 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 ) );
//#else
rnd512( state, _mm_setzero_si128(),
_mm_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 ) );
//#endif
}
finalization512(state, (uint32*) hashval);
@@ -371,41 +324,23 @@ HashReturn update_and_final_luffa( hashState_luffa *state, BitSequence* output,
// full blocks
for ( i = 0; i < blocks; i++ )
{
//#if defined (__AVX2__)
// rnd512( state, mm256_byteswap_epi32( cast_m256i( data ) ) ),
//#else
rnd512( state, mm_byteswap_epi32( casti_m128i( data, 1 ) ),
mm_byteswap_epi32( casti_m128i( data, 0 ) ) );
//#endif
data += MSG_BLOCK_BYTE_LEN;
}
// 16 byte partial block exists for 80 byte len
if ( state->rembytes )
{
// remaining 16 data bytes + 16 bytes padding
//#if defined (__AVX2__)
// use buffer to manage 16 bytes of data in 32 byte world
// casti_m128i( state->buffer, 0 ) = mm_byteswap_epi32( cast_m128i( data ) );
// padding of partial block
// casti_m128i( state->buffer, 1 ) =
// _mm_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
// rnd512( state, cast_m256i( state->buffer ) );
//#else
rnd512( state, _mm_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 ),
mm_byteswap_epi32( cast_m128i( data ) ) );
//#endif
}
else
{
// empty pad block
//#if defined (__AVX2__)
// rnd512( state, _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0,
// 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 ) );
//#else
rnd512( state, _mm_setzero_si128(),
_mm_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 ) );
//#endif
}
finalization512( state, (uint32*) output );
@@ -419,109 +354,6 @@ HashReturn update_and_final_luffa( hashState_luffa *state, BitSequence* output,
/* Round function */
/* state: hash context */
/*
#if defined (__AVX2__)
// AVX2 only
static void rnd512( hashState_luffa *state, __m256i msg )
{
do
{
area256 t;
area256 *chainv;
chainv.v256 = (__m256i*)state->chainv;
area256 Msg;
Msg.v256 = Msg
// __m256i t;
// __m256i *chainv = (__m256i*)state->chainv;
t.v256 = chainv[0];
t.v256 = _mm256_xor_si256( t.v256, chainv.v256[1] );
t.v256 = _mm256_xor_si256( t.v256, chainv.v256[2] );
t.v256 = _mm256_xor_si256( t.v256, chainv.v256[3] );
t.v256 = _mm256_xor_si256( t.v256, chainv.v256[4] );
MULT2( t.v128[0], t.v128[1] );
// MULT256( t );
Msg.v256 = _mm256_shuffle_epi32( Msg.v256, 27 );
chainv.v256[0] = _mm256_xor_si256( chainv.v256[0], t.v256 );
chainv.v256[1] = _mm256_xor_si256( chainv.v256[1], t.v256 );
chainv.v256[2] = _mm256_xor_si256( chainv.v256[2], t.v256 );
chainv.v256[3] = _mm256_xor_si256( chainv.v256[3], t.v256 );
chainv.v256[4] = _mm256_xor_si256( chainv.v256[4], t.v256 );
t.v256 = chainv[0];
MULT2( chainv.v128[0], chainv.v128[1]);
// MULT256( chainv[0] );
chainv[0] = _mm256_xor_si256( chainv.v256[0], chainv.v256[1] );
MULT2( chainv.v128[2], chainv.v128[3]);
// MULT256( chainv[1] );
chainv.v256[1] = _mm256_xor_si256( chainv.v256[1], chainv.v256[2] );
MULT2( chainv.v128[4], chainv.v128[5]);
// MULT256( chainv[2] );
chainv.v256[2] = _mm256_xor_si256( chainv.v256[2], chainv.v256[3] );
MULT2( chainv.v128[6], chainv.v128[7]);
// MULT256( chainv[3] );
chainv.v256[3] = _mm256_xor_si256( chainv.v256[3], chainv.v256[4] );
MULT2( chainv.v128[8], chainv.v128[9]);
// MULT256( chainv[4] );
chainv.v256[4] = _mm256_xor_si256( chainv.v256[4], chainv.v256[5] );
t.v256 = chainv.v256[4];
MULT2( chainv.v128[8], chainv.v128[9]);
// MULT256( chainv[4] );
chainv.v256[4] = _mm256_xor_si256( chainv.v256[4], chainv.v256[3] );
MULT2( chainv.v128[6], chainv.v128[7]);
// MULT256( chainv[3] );
chainv.v256[3] = _mm256_xor_si256( chainv.v256[3], chainv.v256[2] );
MULT2( chainv.v128[4], chainv.v128[5]);
// MULT256( chainv[2] );
chainv.v256[2] = _mm256_xor_si256( chainv.v256[2], chainv.v256[1] );
MULT2( chainv.v128[2], chainv.v128[3]);
// MULT256( chainv[1] );
chainv.v256[1] = _mm256_xor_si256( chainv.v256[1], chainv.v256[0] );
MULT2( chainv.v128[0], chainv.v128[1]);
// MULT256( chainv[0] );
chainv.v256[0] = _mm256_xor_si256( _mm256_xor_si256( chainv.v256[0], t ), Msg.v256 );
MULT2( Msg.v128[0], Msg.v128[1] );
// MULT256( msg );
chainv.v256[1] = _mm256_xor_si256( chainv.v256[1], Msg.v256 );
MULT2( Msg.v128[0], Msg.v128[1] );
// MULT256( msg );
chainv.v256[2] = _mm256_xor_si256( chainv.v256[2], Msg.v256 );
MULT2( Msg.v128[0], Msg.v128[1] );
// MULT256( msg );
chainv.v256[3] = _mm256_xor_si256( chainv.v256[3], Msg.v256 );
MULT2( Msg.v128[0], Msg.v128[1] );
// MULT256( msg );
chainv.v256[4] = _mm256_xor_si256( chainv.v256[4], Msg.v256 );
MULT2( Msg.v128[0], Msg.v128[1] );
// MULT256( msg );
} while (0);
// new set of __m128i vars for the rest
__m128i t[2];
__m128i *chainv = state->chainv;
__m128i tmp[2];
__m128i x[8];
__m128i msg0 = Msg.v128[0];
__m128i msg1 = Msg.v128[1];
// remainder common with SSE2
#else
// SSE2 only
*/
static void rnd512( hashState_luffa *state, __m128i msg1, __m128i msg0 )
{
__m128i t[2];
@@ -635,10 +467,6 @@ static void rnd512( hashState_luffa *state, __m128i msg1, __m128i msg0 )
MULT2( msg0, msg1);
//#endif
// common to SSE2 and AVX2
chainv[3] = _mm_or_si128( _mm_slli_epi32(chainv[3], 1),
_mm_srli_epi32(chainv[3], 31) );
chainv[5] = _mm_or_si128( _mm_slli_epi32(chainv[5], 2),
@@ -693,7 +521,6 @@ static void rnd512( hashState_luffa *state, __m128i msg1, __m128i msg0 )
/* state: hash context */
/* b[8]: hash values */
//*
#if defined (__AVX2__)
static void finalization512( hashState_luffa *state, uint32 *b )
@@ -701,9 +528,9 @@ static void finalization512( hashState_luffa *state, uint32 *b )
uint32 hash[8] __attribute((aligned(64)));
__m256i* chainv = (__m256i*)state->chainv;
__m256i t;
const __m128i zero = _mm_setzero_si128();
rnd512( state, _mm_setzero_si128(), _mm_setzero_si128() );
// rnd512( state, _mm256_setzero_si256() );
rnd512( state, zero, zero );
t = chainv[0];
t = _mm256_xor_si256( t, chainv[1] );
@@ -717,8 +544,7 @@ static void finalization512( hashState_luffa *state, uint32 *b )
casti_m256i( b, 0 ) = mm256_byteswap_epi32( casti_m256i( hash, 0 ) );
rnd512( state, _mm_setzero_si128(), _mm_setzero_si128() );
// rnd512( state, _mm256_setzero_si256() );
rnd512( state, zero, zero );
t = chainv[0];
t = _mm256_xor_si256( t, chainv[1] );
@@ -734,17 +560,15 @@ static void finalization512( hashState_luffa *state, uint32 *b )
#else
static void finalization512( hashState_luffa *state, uint32 *b )
{
uint32 hash[8] __attribute((aligned(64)));
__m128i* chainv = state->chainv;
__m128i t[2];
const __m128i zero = _mm_setzero_si128();
/*---- blank round with m=0 ----*/
rnd512( state, _mm_setzero_si128(), _mm_setzero_si128() );
// _mm_prefetch( b, _MM_HINT_T0 );
rnd512( state, zero, zero );
t[0] = chainv[0];
t[1] = chainv[1];
@@ -766,7 +590,7 @@ static void finalization512( hashState_luffa *state, uint32 *b )
casti_m128i( b, 0 ) = mm_byteswap_epi32( casti_m128i( hash, 0 ) );
casti_m128i( b, 1 ) = mm_byteswap_epi32( casti_m128i( hash, 1 ) );
rnd512( state, _mm_setzero_si128(), _mm_setzero_si128() );
rnd512( state, zero, zero );
t[0] = chainv[0];
t[1] = chainv[1];