/** * @file * * @author jeffrey.daily@gmail.com * * Copyright (c) 2015 Battelle Memorial Institute. */ #include "config.h" #include #if defined(_MSC_VER) #include #else #include #endif #include "parasail.h" #include "parasail/memory.h" #include "parasail/internal_sse.h" static inline __m128i _mm_blendv_epi8_rpl(__m128i a, __m128i b, __m128i mask) { a = _mm_andnot_si128(mask, a); a = _mm_or_si128(a, _mm_and_si128(mask, b)); return a; } static inline __m128i _mm_insert_epi32_rpl(__m128i a, int32_t i, const int imm) { __m128i_32_t A; A.m = a; A.v[imm] = i; return A.m; } static inline __m128i _mm_max_epi32_rpl(__m128i a, __m128i b) { __m128i mask = _mm_cmpgt_epi32(a, b); a = _mm_and_si128(a, mask); b = _mm_andnot_si128(mask, b); return _mm_or_si128(a, b); } static inline int32_t _mm_extract_epi32_rpl(__m128i a, const int imm) { __m128i_32_t A; A.m = a; return A.v[imm]; } static inline __m128i _mm_min_epi32_rpl(__m128i a, __m128i b) { __m128i mask = _mm_cmpgt_epi32(b, a); a = _mm_and_si128(a, mask); b = _mm_andnot_si128(mask, b); return _mm_or_si128(a, b); } static inline void arr_store_si128( int8_t *array, __m128i vWH, int32_t i, int32_t s1Len, int32_t j, int32_t s2Len) { if (0 <= i+0 && i+0 < s1Len && 0 <= j-0 && j-0 < s2Len) { array[1LL*(i+0)*s2Len + (j-0)] = (int8_t)_mm_extract_epi32_rpl(vWH, 3); } if (0 <= i+1 && i+1 < s1Len && 0 <= j-1 && j-1 < s2Len) { array[1LL*(i+1)*s2Len + (j-1)] = (int8_t)_mm_extract_epi32_rpl(vWH, 2); } if (0 <= i+2 && i+2 < s1Len && 0 <= j-2 && j-2 < s2Len) { array[1LL*(i+2)*s2Len + (j-2)] = (int8_t)_mm_extract_epi32_rpl(vWH, 1); } if (0 <= i+3 && i+3 < s1Len && 0 <= j-3 && j-3 < s2Len) { array[1LL*(i+3)*s2Len + (j-3)] = (int8_t)_mm_extract_epi32_rpl(vWH, 0); } } #define FNAME parasail_sw_trace_diag_sse2_128_32 parasail_result_t* FNAME( const char * const restrict _s1, const int _s1Len, const char * const restrict _s2, const int s2Len, const int open, const int gap, const parasail_matrix_t *matrix) { /* declare local variables */ int32_t N = 0; int32_t PAD = 0; int32_t PAD2 = 0; int32_t s1Len_PAD = 0; int32_t s2Len_PAD = 0; int32_t * restrict s1 = NULL; int32_t * restrict s2B = NULL; int32_t * restrict _H_pr = NULL; int32_t * restrict _F_pr = NULL; int32_t * restrict s2 = NULL; int32_t * restrict H_pr = NULL; int32_t * restrict F_pr = NULL; parasail_result_t *result = NULL; int32_t i = 0; int32_t j = 0; int32_t s1Len = 0; int32_t end_query = 0; int32_t end_ref = 0; int32_t NEG_LIMIT = 0; int32_t POS_LIMIT = 0; int32_t score = 0; __m128i vNegLimit; __m128i vPosLimit; __m128i vSaturationCheckMin; __m128i vSaturationCheckMax; __m128i vNegInf; __m128i vNegInf0; __m128i vOpen; __m128i vGap; __m128i vZero; __m128i vOne; __m128i vN; __m128i vNegOne; __m128i vI; __m128i vJreset; __m128i vMaxH; __m128i vEndI; __m128i vEndJ; __m128i vILimit; __m128i vJLimit; __m128i vTDiag; __m128i vTIns; __m128i vTDel; __m128i vTZero; __m128i vTDiagE; __m128i vTInsE; __m128i vTDiagF; __m128i vTDelF; /* validate inputs */ PARASAIL_CHECK_NULL(_s2); PARASAIL_CHECK_GT0(s2Len); PARASAIL_CHECK_GE0(open); PARASAIL_CHECK_GE0(gap); PARASAIL_CHECK_NULL(matrix); if (matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) { PARASAIL_CHECK_NULL(_s1); PARASAIL_CHECK_GT0(_s1Len); } /* initialize stack variables */ N = 4; /* number of values in vector */ PAD = N-1; PAD2 = PAD*2; s1Len = matrix->type == PARASAIL_MATRIX_TYPE_SQUARE ? _s1Len : matrix->length; s1Len_PAD = s1Len+PAD; s2Len_PAD = s2Len+PAD; i = 0; j = 0; end_query = 0; end_ref = 0; NEG_LIMIT = (-open < matrix->min ? INT32_MIN + open : INT32_MIN - matrix->min) + 1; POS_LIMIT = INT32_MAX - matrix->max - 1; score = NEG_LIMIT; vNegLimit = _mm_set1_epi32(NEG_LIMIT); vPosLimit = _mm_set1_epi32(POS_LIMIT); vSaturationCheckMin = vPosLimit; vSaturationCheckMax = vNegLimit; vNegInf = _mm_set1_epi32(NEG_LIMIT); vNegInf0 = _mm_srli_si128(vNegInf, 4); /* shift in a 0 */ vOpen = _mm_set1_epi32(open); vGap = _mm_set1_epi32(gap); vZero = _mm_set1_epi32(0); vOne = _mm_set1_epi32(1); vN = _mm_set1_epi32(N); vNegOne = _mm_set1_epi32(-1); vI = _mm_set_epi32(0,1,2,3); vJreset = _mm_set_epi32(0,-1,-2,-3); vMaxH = vNegInf; vEndI = vNegInf; vEndJ = vNegInf; vILimit = _mm_set1_epi32(s1Len); vJLimit = _mm_set1_epi32(s2Len); vTDiag = _mm_set1_epi32(PARASAIL_DIAG); vTIns = _mm_set1_epi32(PARASAIL_INS); vTDel = _mm_set1_epi32(PARASAIL_DEL); vTZero = _mm_set1_epi32(PARASAIL_ZERO); vTDiagE = _mm_set1_epi32(PARASAIL_DIAG_E); vTInsE = _mm_set1_epi32(PARASAIL_INS_E); vTDiagF = _mm_set1_epi32(PARASAIL_DIAG_F); vTDelF = _mm_set1_epi32(PARASAIL_DEL_F); /* initialize result */ result = parasail_result_new_trace(s1Len, s2Len, 16, sizeof(int8_t)); if (!result) return NULL; /* set known flags */ result->flag |= PARASAIL_FLAG_SW | PARASAIL_FLAG_DIAG | PARASAIL_FLAG_TRACE | PARASAIL_FLAG_BITS_32 | PARASAIL_FLAG_LANES_4; /* initialize heap variables */ s2B= parasail_memalign_int32_t(16, s2Len+PAD2); _H_pr = parasail_memalign_int32_t(16, s2Len+PAD2); _F_pr = parasail_memalign_int32_t(16, s2Len+PAD2); s2 = s2B+PAD; /* will allow later for negative indices */ H_pr = _H_pr+PAD; F_pr = _F_pr+PAD; /* validate heap variables */ if (!s2B) return NULL; if (!_H_pr) return NULL; if (!_F_pr) return NULL; /* convert _s1 from char to int in range 0-23 */ if (matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) { s1 = parasail_memalign_int32_t(16, s1Len+PAD); if (!s1) return NULL; for (i=0; imapper[(unsigned char)_s1[i]]; } /* pad back of s1 with dummy values */ for (i=s1Len; imapper[(unsigned char)_s2[j]]; } /* pad front of s2 with dummy values */ for (j=-PAD; j<0; ++j) { s2[j] = 0; /* point to first matrix row because we don't care */ } /* pad back of s2 with dummy values */ for (j=s2Len; jmatrix[matrix->size * ((matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) ? s1[i+0] : ((i+0 >= s1Len) ? s1Len-1 : i+0))]; const int * const restrict matrow1 = &matrix->matrix[matrix->size * ((matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) ? s1[i+1] : ((i+1 >= s1Len) ? s1Len-1 : i+1))]; const int * const restrict matrow2 = &matrix->matrix[matrix->size * ((matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) ? s1[i+2] : ((i+2 >= s1Len) ? s1Len-1 : i+2))]; const int * const restrict matrow3 = &matrix->matrix[matrix->size * ((matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) ? s1[i+3] : ((i+3 >= s1Len) ? s1Len-1 : i+3))]; __m128i vIltLimit = _mm_cmplt_epi32(vI, vILimit); /* iterate over database sequence */ for (j=0; j PAD) { vSaturationCheckMin = _mm_min_epi32_rpl(vSaturationCheckMin, vWH); vSaturationCheckMax = _mm_max_epi32_rpl(vSaturationCheckMax, vWH); } /* trace table */ { __m128i cond_zero = _mm_cmpeq_epi32(vWH, vZero); __m128i case1 = _mm_cmpeq_epi32(vWH, vNWH); __m128i case2 = _mm_cmpeq_epi32(vWH, vF); __m128i vT = _mm_blendv_epi8_rpl( _mm_blendv_epi8_rpl(vTIns, vTDel, case2), _mm_blendv_epi8_rpl(vTDiag, vTZero, cond_zero), case1); __m128i condE = _mm_cmpgt_epi32(vE_opn, vE_ext); __m128i condF = _mm_cmpgt_epi32(vF_opn, vF_ext); __m128i vET = _mm_blendv_epi8_rpl(vTInsE, vTDiagE, condE); __m128i vFT = _mm_blendv_epi8_rpl(vTDelF, vTDiagF, condF); vT = _mm_or_si128(vT, vET); vT = _mm_or_si128(vT, vFT); arr_store_si128(result->trace->trace_table, vT, i, s1Len, j, s2Len); } H_pr[j-3] = (int32_t)_mm_extract_epi32_rpl(vWH,0); F_pr[j-3] = (int32_t)_mm_extract_epi32_rpl(vF,0); /* as minor diagonal vector passes across table, extract * max values within the i,j bounds */ { __m128i cond_valid_J = _mm_and_si128( _mm_cmpgt_epi32(vJ, vNegOne), _mm_cmplt_epi32(vJ, vJLimit)); __m128i cond_valid_IJ = _mm_and_si128(cond_valid_J, vIltLimit); __m128i cond_eq = _mm_cmpeq_epi32(vWH, vMaxH); __m128i cond_max = _mm_cmpgt_epi32(vWH, vMaxH); __m128i cond_all = _mm_and_si128(cond_max, cond_valid_IJ); __m128i cond_Jlt = _mm_cmplt_epi32(vJ, vEndJ); vMaxH = _mm_blendv_epi8_rpl(vMaxH, vWH, cond_all); vEndI = _mm_blendv_epi8_rpl(vEndI, vI, cond_all); vEndJ = _mm_blendv_epi8_rpl(vEndJ, vJ, cond_all); cond_all = _mm_and_si128(cond_Jlt, cond_eq); cond_all = _mm_and_si128(cond_all, cond_valid_IJ); vEndI = _mm_blendv_epi8_rpl(vEndI, vI, cond_all); vEndJ = _mm_blendv_epi8_rpl(vEndJ, vJ, cond_all); } vJ = _mm_add_epi32(vJ, vOne); } vI = _mm_add_epi32(vI, vN); } /* alignment ending position */ { int32_t *t = (int32_t*)&vMaxH; int32_t *i = (int32_t*)&vEndI; int32_t *j = (int32_t*)&vEndJ; int32_t k; for (k=0; k score) { score = *t; end_query = *i; end_ref = *j; } else if (*t == score) { if (*j < end_ref) { end_query = *i; end_ref = *j; } else if (*j == end_ref && *i < end_query) { end_query = *i; end_ref = *j; } } } } if (_mm_movemask_epi8(_mm_or_si128( _mm_cmplt_epi32(vSaturationCheckMin, vNegLimit), _mm_cmpgt_epi32(vSaturationCheckMax, vPosLimit)))) { result->flag |= PARASAIL_FLAG_SATURATED; score = 0; end_query = 0; end_ref = 0; } result->score = score; result->end_query = end_query; result->end_ref = end_ref; parasail_free(_F_pr); parasail_free(_H_pr); parasail_free(s2B); if (matrix->type == PARASAIL_MATRIX_TYPE_SQUARE) { parasail_free(s1); } return result; }