static inline unsigned int evaluatePopcount(INT_TYPE v_N, char *precomputed)
{
#ifdef __AVX
unsigned long int
res[4] __attribute__ ((aligned (BYTE_ALIGNMENT)));
unsigned int a, b;
_mm256_store_pd((double*)res, v_N);
a = __builtin_popcountl(res[0]) + __builtin_popcountl(res[1]);
b = __builtin_popcountl(res[2]) + __builtin_popcountl(res[3]);
return (a + b);
#else
unsigned int
sum = 0,
counts[INTS_PER_VECTOR] __attribute__ ((aligned (BYTE_ALIGNMENT)));
VECTOR_STORE((CAST)counts, v_N);
sum += BIT_COUNT(counts[0], precomputed) + BIT_COUNT(counts[1], precomputed);
sum += BIT_COUNT(counts[2], precomputed) + BIT_COUNT(counts[3], precomputed);
return sum;
#endif
}
static inline unsigned int vectorPopcount(INT_TYPE v)
{
unsigned long
counts[LONG_INTS_PER_VECTOR] __attribute__ ((aligned (PLL_BYTE_ALIGNMENT)));
int
i,
sum = 0;
VECTOR_STORE((CAST)counts, v);
for(i = 0; i < LONG_INTS_PER_VECTOR; i++)
sum += __builtin_popcountl(counts[i]);
return ((unsigned int)sum);
}
static inline void storePerSiteScores (partitionList * pr, int model, INT_TYPE v, unsigned int offset)
{
unsigned long
counts[LONG_INTS_PER_VECTOR] __attribute__ ((aligned (PLL_BYTE_ALIGNMENT)));
parsimonyNumber * buf;
int
i,
j;
VECTOR_STORE((CAST)counts, v);
for (i = 0; i < LONG_INTS_PER_VECTOR; ++i)
{
buf = &(pr->partitionData[model]->perSiteParsScores[offset * PLL_PCF + i * ULINT_SIZE]);
for (j = 0; j < ULINT_SIZE; ++ j)
buf[j] += ((counts[i] >> j) & 1);
}
}
static void newviewParsimonyIterativeFast(tree *tr)
{
INT_TYPE
allOne = SET_ALL_BITS_ONE;
int
model,
*ti = tr->ti,
count = ti[0],
index;
for(index = 4; index < count; index += 4)
{
unsigned int
totalScore = 0;
size_t
pNumber = (size_t)ti[index],
qNumber = (size_t)ti[index + 1],
rNumber = (size_t)ti[index + 2];
for(model = 0; model < tr->NumberOfModels; model++)
{
size_t
k,
states = tr->partitionData[model].states,
width = tr->partitionData[model].parsimonyLength;
unsigned int
i;
switch(states)
{
case 2:
{
parsimonyNumber
*left[2],
*right[2],
*this[2];
for(k = 0; k < 2; k++)
{
left[k] = &(tr->partitionData[model].parsVect[(width * 2 * qNumber) + width * k]);
right[k] = &(tr->partitionData[model].parsVect[(width * 2 * rNumber) + width * k]);
this[k] = &(tr->partitionData[model].parsVect[(width * 2 * pNumber) + width * k]);
}
for(i = 0; i < width; i += INTS_PER_VECTOR)
{
INT_TYPE
s_r, s_l, v_N,
l_A, l_C,
v_A, v_C;
s_l = VECTOR_LOAD((CAST)(&left[0][i]));
s_r = VECTOR_LOAD((CAST)(&right[0][i]));
l_A = VECTOR_BIT_AND(s_l, s_r);
v_A = VECTOR_BIT_OR(s_l, s_r);
s_l = VECTOR_LOAD((CAST)(&left[1][i]));
s_r = VECTOR_LOAD((CAST)(&right[1][i]));
l_C = VECTOR_BIT_AND(s_l, s_r);
v_C = VECTOR_BIT_OR(s_l, s_r);
v_N = VECTOR_BIT_OR(l_A, l_C);
VECTOR_STORE((CAST)(&this[0][i]), VECTOR_BIT_OR(l_A, VECTOR_AND_NOT(v_N, v_A)));
VECTOR_STORE((CAST)(&this[1][i]), VECTOR_BIT_OR(l_C, VECTOR_AND_NOT(v_N, v_C)));
v_N = VECTOR_AND_NOT(v_N, allOne);
totalScore += populationCount(v_N);
}
}
break;
case 4:
{
parsimonyNumber
*left[4],
*right[4],
*this[4];
for(k = 0; k < 4; k++)
{
left[k] = &(tr->partitionData[model].parsVect[(width * 4 * qNumber) + width * k]);
right[k] = &(tr->partitionData[model].parsVect[(width * 4 * rNumber) + width * k]);
this[k] = &(tr->partitionData[model].parsVect[(width * 4 * pNumber) + width * k]);
}
for(i = 0; i < width; i += INTS_PER_VECTOR)
{
INT_TYPE
s_r, s_l, v_N,
l_A, l_C, l_G, l_T,
v_A, v_C, v_G, v_T;
s_l = VECTOR_LOAD((CAST)(&left[0][i]));
s_r = VECTOR_LOAD((CAST)(&right[0][i]));
l_A = VECTOR_BIT_AND(s_l, s_r);
v_A = VECTOR_BIT_OR(s_l, s_r);
s_l = VECTOR_LOAD((CAST)(&left[1][i]));
s_r = VECTOR_LOAD((CAST)(&right[1][i]));
l_C = VECTOR_BIT_AND(s_l, s_r);
v_C = VECTOR_BIT_OR(s_l, s_r);
s_l = VECTOR_LOAD((CAST)(&left[2][i]));
s_r = VECTOR_LOAD((CAST)(&right[2][i]));
l_G = VECTOR_BIT_AND(s_l, s_r);
v_G = VECTOR_BIT_OR(s_l, s_r);
s_l = VECTOR_LOAD((CAST)(&left[3][i]));
s_r = VECTOR_LOAD((CAST)(&right[3][i]));
l_T = VECTOR_BIT_AND(s_l, s_r);
v_T = VECTOR_BIT_OR(s_l, s_r);
v_N = VECTOR_BIT_OR(VECTOR_BIT_OR(l_A, l_C), VECTOR_BIT_OR(l_G, l_T));
VECTOR_STORE((CAST)(&this[0][i]), VECTOR_BIT_OR(l_A, VECTOR_AND_NOT(v_N, v_A)));
VECTOR_STORE((CAST)(&this[1][i]), VECTOR_BIT_OR(l_C, VECTOR_AND_NOT(v_N, v_C)));
VECTOR_STORE((CAST)(&this[2][i]), VECTOR_BIT_OR(l_G, VECTOR_AND_NOT(v_N, v_G)));
VECTOR_STORE((CAST)(&this[3][i]), VECTOR_BIT_OR(l_T, VECTOR_AND_NOT(v_N, v_T)));
v_N = VECTOR_AND_NOT(v_N, allOne);
totalScore += populationCount(v_N);
}
}
break;
case 20:
{
parsimonyNumber
*left[20],
*right[20],
*this[20];
for(k = 0; k < 20; k++)
{
left[k] = &(tr->partitionData[model].parsVect[(width * 20 * qNumber) + width * k]);
right[k] = &(tr->partitionData[model].parsVect[(width * 20 * rNumber) + width * k]);
this[k] = &(tr->partitionData[model].parsVect[(width * 20 * pNumber) + width * k]);
}
for(i = 0; i < width; i += INTS_PER_VECTOR)
{
size_t j;
INT_TYPE
s_r, s_l,
v_N = SET_ALL_BITS_ZERO,
l_A[20],
v_A[20];
for(j = 0; j < 20; j++)
{
s_l = VECTOR_LOAD((CAST)(&left[j][i]));
s_r = VECTOR_LOAD((CAST)(&right[j][i]));
l_A[j] = VECTOR_BIT_AND(s_l, s_r);
v_A[j] = VECTOR_BIT_OR(s_l, s_r);
v_N = VECTOR_BIT_OR(v_N, l_A[j]);
}
for(j = 0; j < 20; j++)
VECTOR_STORE((CAST)(&this[j][i]), VECTOR_BIT_OR(l_A[j], VECTOR_AND_NOT(v_N, v_A[j])));
v_N = VECTOR_AND_NOT(v_N, allOne);
totalScore += populationCount(v_N);
}
}
break;
default:
{
parsimonyNumber
*left[32],
*right[32],
*this[32];
assert(states <= 32);
for(k = 0; k < states; k++)
{
left[k] = &(tr->partitionData[model].parsVect[(width * states * qNumber) + width * k]);
right[k] = &(tr->partitionData[model].parsVect[(width * states * rNumber) + width * k]);
this[k] = &(tr->partitionData[model].parsVect[(width * states * pNumber) + width * k]);
}
for(i = 0; i < width; i += INTS_PER_VECTOR)
{
size_t j;
INT_TYPE
s_r, s_l,
v_N = SET_ALL_BITS_ZERO,
l_A[32],
v_A[32];
for(j = 0; j < states; j++)
{
s_l = VECTOR_LOAD((CAST)(&left[j][i]));
s_r = VECTOR_LOAD((CAST)(&right[j][i]));
l_A[j] = VECTOR_BIT_AND(s_l, s_r);
v_A[j] = VECTOR_BIT_OR(s_l, s_r);
v_N = VECTOR_BIT_OR(v_N, l_A[j]);
}
for(j = 0; j < states; j++)
VECTOR_STORE((CAST)(&this[j][i]), VECTOR_BIT_OR(l_A[j], VECTOR_AND_NOT(v_N, v_A[j])));
v_N = VECTOR_AND_NOT(v_N, allOne);
totalScore += populationCount(v_N);
}
}
}
}
tr->parsimonyScore[pNumber] = totalScore + tr->parsimonyScore[rNumber] + tr->parsimonyScore[qNumber];
}
}
bool CDecoder_OMS_fixed_SSE::decode_8bits(char Intrinsic_fix[], char Rprime_fix[], int nombre_iterations)
{
////////////////////////////////////////////////////////////////////////////
//
// Initilisation des espaces memoire
//
const TYPE zero = VECTOR_ZERO;
for (int i=0; i<MESSAGE; i++){
var_mesgs[i] = zero;
}
//
////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////
//
// ENTRELACEMENT DES DONNEES D'ENTREE POUR POUVOIR EXPLOITER LE MODE SIMD
//
if( NOEUD%16 == 0 ){
uchar_transpose_sse((TYPE*)Intrinsic_fix, (TYPE*)var_nodes, NOEUD);
}else{
char *ptrVar = (char*) var_nodes;
for (int i=0; i<NOEUD; i++){
for (int z=0; z<16; z++){
ptrVar[16 * i + z] = Intrinsic_fix[z * NOEUD + i];
}
}
}
//
////////////////////////////////////////////////////////////////////////////
// unsigned int arret = 0;
while ( nombre_iterations-- ) {
TYPE *p_msg1r = var_mesgs;
TYPE *p_msg1w = var_mesgs;
#if PETIT == 1
TYPE **p_indice_nod1 = p_vn_adr;
TYPE **p_indice_nod2 = p_vn_adr;
#else
const unsigned short *p_indice_nod1 = PosNoeudsVariable;
const unsigned short *p_indice_nod2 = PosNoeudsVariable;
#endif
// arret = 0;
const TYPE min_var = VECTOR_SET1( vSAT_NEG_VAR );
const TYPE max_msg = VECTOR_SET1( vSAT_POS_MSG );
for (int i=0; i<DEG_1_COMPUTATIONS; i++){
//IACA_START
TYPE tab_vContr[DEG_1];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#if (DEG_1 & 0x01) == 1
const unsigned char sign8 = 0x80;
const unsigned char isign8 = 0xC0;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8 = VECTOR_SET1( isign8 );
#else
const unsigned char sign8 = 0x80;
const unsigned char isign8b = 0x40;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8b = VECTOR_SET1( isign8b );
#endif
#if PETIT == 1
#if MANUAL_PREFETCH == 1
_mm_prefetch((const char*)(p_indice_nod1[DEG_1]), _MM_HINT_T0);
_mm_prefetch((const char*)(&p_msg1r[DEG_1]), _MM_HINT_T0);
#endif
#endif
#pragma unroll(DEG_1)
for(int j=0; j<DEG_1; j++){
#if PETIT == 1
TYPE vNoeud = VECTOR_LOAD( *p_indice_nod1 );
#else
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
#endif
TYPE vMessg = VECTOR_LOAD(p_msg1r);
TYPE vContr = VECTOR_SUB_AND_SATURATE_VAR_8bits(vNoeud, vMessg, min_var);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr, msign8);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_MIN( VECTOR_ABS( vContr), max_msg);
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
#if PETIT == 1
#if MANUAL_PREFETCH == 1
for(int j=0 ; j<DEG_1 ; j++){
_mm_prefetch((const char*)(p_indice_nod1[j]), _MM_HINT_T0);
}
_mm_prefetch((const char*)(p_indice_nod1[DEG_1]), _MM_HINT_T0);
#endif
#endif
TYPE cste_1 = VECTOR_MIN(VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
TYPE cste_2 = VECTOR_MIN(VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
#if (DEG_1 & 0x01) == 1
sign = VECTOR_XOR(sign, misign8);
#else
sign = VECTOR_XOR(sign, misign8b);
#endif
#pragma unroll(DEG_1)
for(int j=0 ; j<DEG_1 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_MIN(VECTOR_ABS(vContr), max_msg );
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr, msign8));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD_AND_SATURATE_VAR_8bits(vContr, v2St, min_var);
VECTOR_STORE( p_msg1w, v2St);
#if PETIT == 1
VECTOR_STORE( *p_indice_nod2, v2Sr);
#else
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
#endif
p_msg1w += 1;
p_indice_nod2 += 1;
}
// arret = arret || VECTOR_XOR_REDUCE( sign );
//IACA_END
}
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 2
for (int i=0; i<DEG_2_COMPUTATIONS; i++){
#if (DEG_2 & 0x01) == 1
const unsigned char sign8 = 0x80;
const unsigned char isign8 = 0xC0;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8 = VECTOR_SET1( isign8 );
#else
const unsigned char sign8 = 0x80;
const unsigned char isign8b = 0x40;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8b = VECTOR_SET1( isign8b );
#endif
TYPE tab_vContr[DEG_2];
TYPE sign = zero;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#pragma unroll(DEG_2)
for(int j=0 ; j<DEG_2 ; j++)
{
#if PETIT == 1
TYPE vNoeud = VECTOR_LOAD( *p_indice_nod1 );
#else
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
#endif
TYPE vMessg = VECTOR_LOAD( p_msg1r );
TYPE vContr = VECTOR_SUB_AND_SATURATE_VAR_8bits(vNoeud, vMessg, min_var);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr, msign8);
sign = VECTOR_XOR (sign, cSign);
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1 );
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
#if (DEG_2 & 0x01) == 1
sign = VECTOR_XOR(sign, misign8);
#else
sign = VECTOR_XOR(sign, misign8b);
#endif
#pragma unroll(DEG_2)
for(int j=0 ; j<DEG_2 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr, msign8));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD_AND_SATURATE_VAR_8bits(vContr, v2St, min_var);
VECTOR_STORE( p_msg1w, v2St);
#if PETIT == 1
VECTOR_STORE( *p_indice_nod2, v2Sr);
#else
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
#endif
p_msg1w += 1;
p_indice_nod2 += 1;
}
// arret = arret || VECTOR_XOR_REDUCE( sign );
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 3
for (int i=0; i<DEG_3_COMPUTATIONS; i++){
#if (DEG_3 & 0x01) == 1
const unsigned char sign8 = 0x80;
const unsigned char isign8 = 0xC0;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8 = VECTOR_SET1( isign8 );
#else
const unsigned char sign8 = 0x80;
const unsigned char isign8b = 0x40;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8b = VECTOR_SET1( isign8b );
#endif
TYPE tab_vContr[DEG_3];
TYPE sign = zero;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
for(int j=0 ; j<DEG_3 ; j++)
{
#if PETIT == 1
TYPE vNoeud = VECTOR_LOAD( *p_indice_nod1 );
#else
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
#endif
TYPE vMessg = VECTOR_LOAD( p_msg1r );
TYPE vContr = VECTOR_SUB_AND_SATURATE_VAR_8bits(vNoeud, vMessg, min_var);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr, msign8);
sign = VECTOR_XOR (sign, cSign);
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1 );
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
#if (DEG_3 & 0x01) == 1
sign = VECTOR_XOR(sign, misign8);
#else
sign = VECTOR_XOR(sign, misign8b);
#endif
for(int j=0 ; j<DEG_3 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr, msign8));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD_AND_SATURATE_VAR_8bits(vContr, v2St, min_var);
VECTOR_STORE( p_msg1w, v2St);
#if PETIT == 1
VECTOR_STORE( *p_indice_nod2, v2Sr);
#else
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
#endif
p_msg1w += 1;
p_indice_nod2 += 1;
}
// arret = arret || VECTOR_XOR_REDUCE( sign );
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 4
for (int i=0; i<DEG_4_COMPUTATIONS; i++){
#if (DEG_4 & 0x01) == 1
const unsigned char sign8 = 0x80;
const unsigned char isign8 = 0xC0;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8 = VECTOR_SET1( isign8 );
#else
const unsigned char sign8 = 0x80;
const unsigned char isign8b = 0x40;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8b = VECTOR_SET1( isign8b );
#endif
TYPE tab_vContr[DEG_4];
TYPE sign = zero;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
for(int j=0 ; j<DEG_4 ; j++)
{
#if PETIT == 1
TYPE vNoeud = VECTOR_LOAD( *p_indice_nod1 );
#else
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
#endif
TYPE vMessg = VECTOR_LOAD( p_msg1r );
TYPE vContr = VECTOR_SUB_AND_SATURATE_VAR_8bits(vNoeud, vMessg, min_var);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr, msign8);
sign = VECTOR_XOR (sign, cSign);
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1 );
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
#if (DEG_4 & 0x01) == 1
sign = VECTOR_XOR(sign, misign8);
#else
sign = VECTOR_XOR(sign, misign8b);
#endif
for(int j=0 ; j<DEG_4 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr, msign8));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD_AND_SATURATE_VAR_8bits(vContr, v2St, min_var);
VECTOR_STORE( p_msg1w, v2St);
#if PETIT == 1
VECTOR_STORE( *p_indice_nod2, v2Sr);
#else
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
#endif
p_msg1w += 1;
p_indice_nod2 += 1;
}
// arret = arret || VECTOR_XOR_REDUCE( sign );
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 5
for (int i=0; i<DEG_5_COMPUTATIONS; i++){
#if (DEG_5 & 0x01) == 1
const unsigned char sign8 = 0x80;
const unsigned char isign8 = 0xC0;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8 = VECTOR_SET1( isign8 );
#else
const unsigned char sign8 = 0x80;
const unsigned char isign8b = 0x40;
const TYPE msign8 = VECTOR_SET1( sign8 );
const TYPE misign8b = VECTOR_SET1( isign8b );
#endif
TYPE tab_vContr[DEG_5];
TYPE sign = zero;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
for(int j=0 ; j<DEG_5 ; j++)
{
#if PETIT == 1
TYPE vNoeud = VECTOR_LOAD( *p_indice_nod1 );
#else
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
#endif
TYPE vMessg = VECTOR_LOAD( p_msg1r );
TYPE vContr = VECTOR_SUB_AND_SATURATE_VAR_8bits(vNoeud, vMessg, min_var);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr, msign8);
sign = VECTOR_XOR (sign, cSign);
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1 );
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg); // ON SATURE DIREECTEMENT AU FORMAT MSG
#if (DEG_5 & 0x01) == 1
sign = VECTOR_XOR(sign, misign8);
#else
sign = VECTOR_XOR(sign, misign8b);
#endif
for(int j=0 ; j<DEG_5 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS ( VECTOR_MIN(vContr, max_msg) );
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr, msign8));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD_AND_SATURATE_VAR_8bits(vContr, v2St, min_var);
VECTOR_STORE( p_msg1w, v2St);
#if PETIT == 1
VECTOR_STORE( *p_indice_nod2, v2Sr);
#else
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
#endif
p_msg1w += 1;
p_indice_nod2 += 1;
}
// arret = arret || VECTOR_XOR_REDUCE( sign );
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES > 5
printf("The number of DEGREE(Cn) IS HIGHER THAN 5. YOU NEED TO PERFORM A COPY PASTE IN SOURCE CODE...\n");
exit( 0 );
#endif
/////////////////////////////////////////////////////////////////////////////////
//
// GESTION DU CRITERE D'ARRET
//
// if( (arret == 0) && (fast_stop == 1) ){
// break;
// }
}
////////////////////////////////////////////////////////////////////////////
//
// ON REMET EN FORME LES DONNEES DE SORTIE POUR LA SUITE DU PROCESS
//
if( NOEUD%16 == 0 ){
uchar_itranspose_sse((TYPE*)var_nodes, (TYPE*)Rprime_fix, NOEUD);
}else{
char* ptr = (char*) var_nodes;
for (int i=0; i<NOEUD; i+=1){
for (int j=0; j<16; j+=1){
Rprime_fix[j*NOEUD +i] = (ptr[16*i+j] > 0);
}
}
}
//
////////////////////////////////////////////////////////////////////////////
return 1;
}
bool CDecoder_OMS_fixed_NEON16_v3::decode_8bits(signed char Intrinsic_fix[], signed char Rprime_fix[], int nombre_iterations)
{
////////////////////////////////////////////////////////////////////////////
//
// Initilisation des espaces memoire
//
// for (int i=0; i<MESSAGE; i++){
// var_mesgs[i] = VECTOR_ZERO;
// }
//
////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////
//
// ENTRELACEMENT DES DONNEES D'ENTREE POUR POUVOIR EXPLOITER LE MODE SIMD
//
if( NOEUD%16 == 0 ){
uchar_transpose_neon((trans_TYPE*)Intrinsic_fix, (trans_TYPE*)var_nodes, NOEUD);
}else{
signed char* ptrVar = (signed char*) var_nodes;
for (int i=0; i<NOEUD; i++){
for (int z=0; z<16; z++){
ptrVar[16 * i + z] = Intrinsic_fix[z * NOEUD + i];
}
}
}
//
////////////////////////////////////////////////////////////////////////////
nombre_iterations--;
if( 1 )
{
TYPE *p_msg1w = var_mesgs;
const unsigned short *p_indice_nod1 = PosNoeudsVariable;
const unsigned short *p_indice_nod2 = PosNoeudsVariable;
//const TYPE min_var = VECTOR_SET1( -127 );
const TYPE max_msg = VECTOR_SET1( 31 );
#if NB_DEGRES >= 1
for (int i=0; i<DEG_1_COMPUTATIONS; i++){
TYPE tab_vContr[DEG_1];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#ifdef _PREFETCH_
__builtin_prefetch (p_indice_nod1 + DEG_1, 0, 3);
#endif
for(int j=0; j<DEG_1; j++){
TYPE vContr = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
//#ifdef _PREFETCH_
// if( (j & 0x01) == 0 ) __builtin_prefetch (p_msg1r+DEG_1, 0, 0);
//#endif
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_ABS( vContr );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
}
#ifdef _PREFETCH_
for(int j=0; j<DEG_1; j++){
__builtin_prefetch (&var_nodes[p_indice_nod1[j]], 0, 3);
}
#endif
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg);
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg);
for(int j=0 ; j<DEG_1 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS (vContr);
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
vRes = VECTOR_MIN(vRes, max_msg); // BLG
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD(vContr, v2St);
VECTOR_STORE( p_msg1w, v2St);
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
p_msg1w += 1;
p_indice_nod2 += 1;
}
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 2
for (int i=0; i<DEG_2_COMPUTATIONS; i++){
TYPE tab_vContr[DEG_2];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#ifdef _PREFETCH_
__builtin_prefetch (p_indice_nod1 + DEG_2, 0, 3);
#endif
for(int j=0; j<DEG_2; j++){
TYPE vContr = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
//#ifdef _PREFETCH_
// if( (j & 0x01) == 0 ) __builtin_prefetch (p_msg1r+DEG_2, 0, 0);
//#endif
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_ABS( vContr );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
}
#ifdef _PREFETCH_
for(int j=0; j<DEG_2; j++){
__builtin_prefetch (&var_nodes[p_indice_nod1[j]], 0, 3);
}
#endif
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg);
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg);
for(int j=0 ; j<DEG_2 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS (vContr);
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
vRes = VECTOR_MIN(vRes, max_msg); // BLG
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD(vContr, v2St);
VECTOR_STORE( p_msg1w, v2St);
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
p_msg1w += 1;
p_indice_nod2 += 1;
}
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES > 2
printf("The number of DEGREE(Cn) IS HIGHER THAN 5. YOU NEED TO PERFORM A COPY PASTE IN SOURCE CODE...\n");
exit( 0 );
#endif
}
//
//
// ON REPREND LE TRAITEMENT NORMAL DE L'INFORMATION
//
//
while (nombre_iterations-- != 1) {
TYPE *p_msg1r = var_mesgs;
TYPE *p_msg1w = var_mesgs;
const unsigned short *p_indice_nod1 = PosNoeudsVariable;
const unsigned short *p_indice_nod2 = PosNoeudsVariable;
// const TYPE min_var = VECTOR_SET1( -127 );
const TYPE max_msg = VECTOR_SET1( 31 );
#if NB_DEGRES >= 1
for (int i=0; i<DEG_1_COMPUTATIONS; i++){
TYPE tab_vContr[DEG_1];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#ifdef _PREFETCH_
__builtin_prefetch (p_indice_nod1 + DEG_1, 0, 3);
#endif
for(int j=0; j<DEG_1; j++){
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
TYPE vMessg = VECTOR_LOAD(p_msg1r);
#ifdef _PREFETCH_
if( (j & 0x01) == 0 ) __builtin_prefetch (p_msg1r+DEG_1, 0, 0);
#endif
TYPE vContr = VECTOR_SUB(vNoeud, vMessg);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_ABS( vContr );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
#ifdef _PREFETCH_
for(int j=0; j<DEG_1; j++){
__builtin_prefetch (&var_nodes[p_indice_nod1[j]], 0, 3);
}
#endif
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg);
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg);
for(int j=0 ; j<DEG_1 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS (vContr);
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
vRes = VECTOR_MIN(vRes, max_msg); // BLG
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD(vContr, v2St);
VECTOR_STORE( p_msg1w, v2St);
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
p_msg1w += 1;
p_indice_nod2 += 1;
}
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 2
for (int i=0; i<DEG_2_COMPUTATIONS; i++){
TYPE tab_vContr[DEG_2];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#ifdef _PREFETCH_
__builtin_prefetch (p_indice_nod1 + DEG_2, 0, 3);
#endif
for(int j=0; j<DEG_2; j++){
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
TYPE vMessg = VECTOR_LOAD(p_msg1r);
#ifdef _PREFETCH_
if( (j & 0x01) == 0 ) __builtin_prefetch (p_msg1r+DEG_2, 0, 0);
#endif
TYPE vContr = VECTOR_SUB(vNoeud, vMessg);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_ABS( vContr );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
#ifdef _PREFETCH_
for(int j=0; j<DEG_2; j++){
__builtin_prefetch (&var_nodes[p_indice_nod1[j]], 0, 3);
}
#endif
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg);
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg);
for(int j=0 ; j<DEG_2 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS (vContr);
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
vRes = VECTOR_MIN(vRes, max_msg); // BLG
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD(vContr, v2St);
VECTOR_STORE( p_msg1w, v2St);
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
p_msg1w += 1;
p_indice_nod2 += 1;
}
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES > 2
printf("The number of DEGREE(Cn) IS HIGHER THAN 5. YOU NEED TO PERFORM A COPY PASTE IN SOURCE CODE...\n");
exit( 0 );
#endif
}
{
TYPE *p_msg1r = var_mesgs;
const unsigned short *p_indice_nod1 = PosNoeudsVariable;
const unsigned short *p_indice_nod2 = PosNoeudsVariable;
// const TYPE min_var = VECTOR_SET1( -127 );
const TYPE max_msg = VECTOR_SET1( 31 );
#if NB_DEGRES >= 1
for (int i=0; i<DEG_1_COMPUTATIONS; i++){
TYPE tab_vContr[DEG_1];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#ifdef _PREFETCH_
__builtin_prefetch (p_indice_nod1 + DEG_1, 0, 3);
#endif
for(int j=0; j<DEG_1; j++){
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
TYPE vMessg = VECTOR_LOAD(p_msg1r);
#ifdef _PREFETCH_
if( (j & 0x01) == 0 ) __builtin_prefetch (p_msg1r+DEG_1, 0, 0);
#endif
TYPE vContr = VECTOR_SUB(vNoeud, vMessg);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_ABS( vContr );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
#ifdef _PREFETCH_
for(int j=0; j<DEG_1; j++){
__builtin_prefetch (&var_nodes[p_indice_nod1[j]], 0, 3);
}
#endif
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg);
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg);
for(int j=0 ; j<DEG_1 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS (vContr);
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
vRes = VECTOR_MIN(vRes, max_msg); // BLG
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD(vContr, v2St);
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
p_indice_nod2 += 1;
}
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES >= 2
for (int i=0; i<DEG_2_COMPUTATIONS; i++){
TYPE tab_vContr[DEG_2];
TYPE sign = VECTOR_ZERO;
TYPE min1 = VECTOR_SET1(vSAT_POS_VAR);
TYPE min2 = min1;
#ifdef _PREFETCH_
__builtin_prefetch (p_indice_nod1 + DEG_2, 0, 3);
#endif
for(int j=0; j<DEG_2; j++){
TYPE vNoeud = VECTOR_LOAD(&var_nodes[(*p_indice_nod1)]);
TYPE vMessg = VECTOR_LOAD(p_msg1r);
#ifdef _PREFETCH_
if( (j & 0x01) == 0 ) __builtin_prefetch (p_msg1r+DEG_2, 0, 0);
#endif
TYPE vContr = VECTOR_SUB(vNoeud, vMessg);
TYPE cSign = VECTOR_GET_SIGN_BIT(vContr);
sign = VECTOR_XOR(sign, cSign);
TYPE vAbs = VECTOR_ABS( vContr );
tab_vContr[j] = vContr;
TYPE vTemp = min1;
min1 = VECTOR_MIN_1(vAbs, min1);
min2 = VECTOR_MIN_2(vAbs, vTemp, min2);
p_indice_nod1 += 1;
p_msg1r += 1;
}
#ifdef _PREFETCH_
for(int j=0; j<DEG_1; j++){
__builtin_prefetch (&var_nodes[p_indice_nod1[j]], 0, 3);
}
#endif
TYPE cste_1 = VECTOR_MIN( VECTOR_SBU(min2, VECTOR_SET1(offset)), max_msg);
TYPE cste_2 = VECTOR_MIN( VECTOR_SBU(min1, VECTOR_SET1(offset)), max_msg);
for(int j=0 ; j<DEG_2 ; j++) {
TYPE vContr = tab_vContr[j];
TYPE vAbs = VECTOR_ABS (vContr);
TYPE vRes = VECTOR_CMOV (vAbs, min1, cste_1, cste_2);
vRes = VECTOR_MIN(vRes, max_msg); // BLG
TYPE vSig = VECTOR_XOR (sign, VECTOR_GET_SIGN_BIT(vContr));
TYPE v2St = VECTOR_invSIGN2(vRes, vSig);
TYPE v2Sr = VECTOR_ADD(vContr, v2St);
VECTOR_STORE( &var_nodes[(*p_indice_nod2)], v2Sr);
p_indice_nod2 += 1;
}
}
#endif
/////////////////////////////////////////////////////////////////////////////////
#if NB_DEGRES > 2
printf("The number of DEGREE(Cn) IS HIGHER THAN 5. YOU NEED TO PERFORM A COPY PASTE IN SOURCE CODE...\n");
exit( 0 );
#endif
}
////////////////////////////////////////////////////////////////////////////
//
// ON REMET EN FORME LES DONNEES DE SORTIE POUR LA SUITE DU PROCESS
//
if( NOEUD%16 == 0 ){
uchar_itranspose_neon((trans_TYPE*)var_nodes, (trans_TYPE*)Rprime_fix, NOEUD);
}else{
signed char* ptr = (signed char*) var_nodes;
for (int i=0; i<NOEUD; i+=1){
for (int j=0; j<16; j+=1){
Rprime_fix[j*NOEUD +i] = (ptr[16*i+j] > 0);
}
}
}
//
////////////////////////////////////////////////////////////////////////////
return 0;
}
