#include "config.h"
#include <stdint.h>
#include <stdlib.h>
#include "parasail.h"
#include "parasail/memory.h"
#include "parasail/internal_neon.h"
#ifdef PARASAIL_TABLE
static inline void arr_store_si128(
int *array,
simde__m128i vH,
int32_t t,
int32_t seglen,
int32_t d,
int32_t dlen)
{
array[1LL*(0*seglen+t)*dlen + d] = (int64_t)simde_mm_extract_epi64(vH, 0);
array[1LL*(1*seglen+t)*dlen + d] = (int64_t)simde_mm_extract_epi64(vH, 1);
}
#endif
#ifdef PARASAIL_ROWCOL
static inline void arr_store_col(
int *col,
simde__m128i vH,
int32_t t,
int32_t seglen)
{
col[0*seglen+t] = (int64_t)simde_mm_extract_epi64(vH, 0);
col[1*seglen+t] = (int64_t)simde_mm_extract_epi64(vH, 1);
}
#endif
#ifdef PARASAIL_TABLE
#define FNAME parasail_nw_table_striped_neon_128_64
#define PNAME parasail_nw_table_striped_profile_neon_128_64
#else
#ifdef PARASAIL_ROWCOL
#define FNAME parasail_nw_rowcol_striped_neon_128_64
#define PNAME parasail_nw_rowcol_striped_profile_neon_128_64
#else
#define FNAME parasail_nw_striped_neon_128_64
#define PNAME parasail_nw_striped_profile_neon_128_64
#endif
#endif
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)
{
parasail_profile_t *profile = NULL;
parasail_result_t *result = NULL;
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);
}
profile = parasail_profile_create_neon_128_64(s1, s1Len, matrix);
if (!profile) return NULL;
result = PNAME(profile, s2, s2Len, open, gap);
parasail_profile_free(profile);
return result;
}
parasail_result_t* PNAME(
const parasail_profile_t * const restrict profile,
const char * const restrict s2, const int s2Len,
const int open, const int gap)
{
int32_t i = 0;
int32_t j = 0;
int32_t k = 0;
int s1Len = 0;
int32_t end_query = 0;
int32_t end_ref = 0;
const parasail_matrix_t *matrix = NULL;
int32_t segWidth = 0;
int32_t segLen = 0;
int32_t offset = 0;
int32_t position = 0;
simde__m128i* restrict vProfile = NULL;
simde__m128i* restrict pvHStore = NULL;
simde__m128i* restrict pvHLoad = NULL;
simde__m128i* restrict pvE = NULL;
int64_t* restrict boundary = NULL;
simde__m128i vGapO;
simde__m128i vGapE;
int64_t NEG_LIMIT = 0;
int64_t POS_LIMIT = 0;
int64_t score = 0;
simde__m128i vNegLimit;
simde__m128i vPosLimit;
simde__m128i vSaturationCheckMin;
simde__m128i vSaturationCheckMax;
parasail_result_t *result = NULL;
PARASAIL_CHECK_NULL(profile);
PARASAIL_CHECK_NULL(profile->profile64.score);
PARASAIL_CHECK_NULL(profile->matrix);
PARASAIL_CHECK_GT0(profile->s1Len);
PARASAIL_CHECK_NULL(s2);
PARASAIL_CHECK_GT0(s2Len);
PARASAIL_CHECK_GE0(open);
PARASAIL_CHECK_GE0(gap);
i = 0;
j = 0;
k = 0;
s1Len = profile->s1Len;
end_query = s1Len-1;
end_ref = s2Len-1;
matrix = profile->matrix;
segWidth = 2;
segLen = (s1Len + segWidth - 1) / segWidth;
offset = (s1Len - 1) % segLen;
position = (segWidth - 1) - (s1Len - 1) / segLen;
vProfile = (simde__m128i*)profile->profile64.score;
vGapO = simde_mm_set1_epi64x(open);
vGapE = simde_mm_set1_epi64x(gap);
NEG_LIMIT = (-open < matrix->min ? INT64_MIN + open : INT64_MIN - matrix->min) + 1;
POS_LIMIT = INT64_MAX - matrix->max - 1;
score = NEG_LIMIT;
vNegLimit = simde_mm_set1_epi64x(NEG_LIMIT);
vPosLimit = simde_mm_set1_epi64x(POS_LIMIT);
vSaturationCheckMin = vPosLimit;
vSaturationCheckMax = vNegLimit;
#ifdef PARASAIL_TABLE
result = parasail_result_new_table1(segLen*segWidth, s2Len);
#else
#ifdef PARASAIL_ROWCOL
result = parasail_result_new_rowcol1(segLen*segWidth, s2Len);
#else
result = parasail_result_new();
#endif
#endif
if (!result) return NULL;
result->flag |= PARASAIL_FLAG_NW | PARASAIL_FLAG_STRIPED
| PARASAIL_FLAG_BITS_64 | PARASAIL_FLAG_LANES_2;
#ifdef PARASAIL_TABLE
result->flag |= PARASAIL_FLAG_TABLE;
#endif
#ifdef PARASAIL_ROWCOL
result->flag |= PARASAIL_FLAG_ROWCOL;
#endif
pvHStore = parasail_memalign_simde__m128i(16, segLen);
pvHLoad = parasail_memalign_simde__m128i(16, segLen);
pvE = parasail_memalign_simde__m128i(16, segLen);
boundary = parasail_memalign_int64_t(16, s2Len+1);
if (!pvHStore) return NULL;
if (!pvHLoad) return NULL;
if (!pvE) return NULL;
if (!boundary) return NULL;
{
int32_t index = 0;
for (i=0; i<segLen; ++i) {
int32_t segNum = 0;
simde__m128i h;
simde__m128i e;
for (segNum=0; segNum<segWidth; ++segNum) {
int64_t tmp = -open-gap*(segNum*segLen+i);
h.i64[segNum] = tmp < INT64_MIN ? INT64_MIN : tmp;
tmp = tmp - open;
e.i64[segNum] = tmp < INT64_MIN ? INT64_MIN : tmp;
}
simde_mm_store_si128(&pvHStore[index], h);
simde_mm_store_si128(&pvE[index], e);
++index;
}
}
{
boundary[0] = 0;
for (i=1; i<=s2Len; ++i) {
int64_t tmp = -open-gap*(i-1);
boundary[i] = tmp < INT64_MIN ? INT64_MIN : tmp;
}
}
for (j=0; j<s2Len; ++j) {
simde__m128i vE;
simde__m128i vF = vNegLimit;
simde__m128i vH = simde_mm_slli_si128(pvHStore[segLen - 1], 8);
const simde__m128i* vP = vProfile + matrix->mapper[(unsigned char)s2[j]] * segLen;
simde__m128i* pv = pvHLoad;
pvHLoad = pvHStore;
pvHStore = pv;
vH = simde_mm_insert_epi64(vH, boundary[j], 0);
for (i=0; i<segLen; ++i) {
vH = simde_mm_add_epi64(vH, simde_mm_load_si128(vP + i));
vE = simde_mm_load_si128(pvE + i);
vH = simde_mm_max_epi64(vH, vE);
vH = simde_mm_max_epi64(vH, vF);
simde_mm_store_si128(pvHStore + i, vH);
vSaturationCheckMax = simde_mm_max_epi64(vSaturationCheckMax, vH);
vSaturationCheckMin = simde_mm_min_epi64(vSaturationCheckMin, vH);
vSaturationCheckMin = simde_mm_min_epi64(vSaturationCheckMin, vE);
vSaturationCheckMin = simde_mm_min_epi64(vSaturationCheckMin, vF);
#ifdef PARASAIL_TABLE
arr_store_si128(result->tables->score_table, vH, i, segLen, j, s2Len);
#endif
vH = simde_mm_sub_epi64(vH, vGapO);
vE = simde_mm_sub_epi64(vE, vGapE);
vE = simde_mm_max_epi64(vE, vH);
simde_mm_store_si128(pvE + i, vE);
vF = simde_mm_sub_epi64(vF, vGapE);
vF = simde_mm_max_epi64(vF, vH);
vH = simde_mm_load_si128(pvHLoad + i);
}
for (k=0; k<segWidth; ++k) {
int64_t tmp = boundary[j+1]-open;
int64_t tmp2 = tmp < INT64_MIN ? INT64_MIN : tmp;
vF = simde_mm_slli_si128(vF, 8);
vF = simde_mm_insert_epi64(vF, tmp2, 0);
for (i=0; i<segLen; ++i) {
vH = simde_mm_load_si128(pvHStore + i);
vH = simde_mm_max_epi64(vH,vF);
simde_mm_store_si128(pvHStore + i, vH);
vSaturationCheckMin = simde_mm_min_epi64(vSaturationCheckMin, vH);
vSaturationCheckMax = simde_mm_max_epi64(vSaturationCheckMax, vH);
#ifdef PARASAIL_TABLE
arr_store_si128(result->tables->score_table, vH, i, segLen, j, s2Len);
#endif
vH = simde_mm_sub_epi64(vH, vGapO);
vF = simde_mm_sub_epi64(vF, vGapE);
if (! simde_mm_movemask_epi8(simde_mm_cmpgt_epi64(vF, vH))) goto end;
}
}
end:
{
}
#ifdef PARASAIL_ROWCOL
{
vH = simde_mm_load_si128(pvHStore + offset);
for (k=0; k<position; ++k) {
vH = simde_mm_slli_si128(vH, 8);
}
result->rowcols->score_row[j] = (int64_t) simde_mm_extract_epi64 (vH, 1);
}
#endif
}
#ifdef PARASAIL_ROWCOL
for (i=0; i<segLen; ++i) {
simde__m128i vH = simde_mm_load_si128(pvHStore+i);
arr_store_col(result->rowcols->score_col, vH, i, segLen);
}
#endif
{
simde__m128i vH = simde_mm_load_si128(pvHStore + offset);
for (k=0; k<position; ++k) {
vH = simde_mm_slli_si128 (vH, 8);
}
score = (int64_t) simde_mm_extract_epi64 (vH, 1);
}
if (simde_mm_movemask_epi8(simde_mm_or_si128(
simde_mm_cmplt_epi64(vSaturationCheckMin, vNegLimit),
simde_mm_cmpgt_epi64(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(boundary);
parasail_free(pvE);
parasail_free(pvHLoad);
parasail_free(pvHStore);
return result;
}