fork/minimap2
1
0
Fork 0
mirror of https://github.com/lh3/minimap2.git synced 2026-10-06 04:18:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

Default compilation without AVX2

Browse Source
This commit is contained in:
Saurabh
2021-07-20 12:32:34 -04:00
committed by Heng Li
parent f68b4b22df
commit 34e273c8ee
6 changed files with 70 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Makefile
+2 -3
View File
@@ -38,13 +38,12 @@ ifeq ($(lhash), 1)
OPT_FLAGS+= -DLISA_HASH -DUINT64 -DVECTORIZE OPT_FLAGS+= -DLISA_HASH -DUINT64 -DVECTORIZE
endif endif
ifeq ($(manual_profile), 1) ifeq ($(manual_profile), 1)
CPPFLAGS+= -DMANUAL_PROFILING CPPFLAGS+= -DMANUAL_PROFILING
endif endif
ifeq ($(arch), avx512) ifeq ($(use_avx2), 1)
CPPFLAGS+= -xCORE-AVX512 OPT_FLAGS+= -DAPPLY_AVX2
endif endif
ifeq ($(disable_output), 1) ifeq ($(disable_output), 1)
align.c
+2 -1
View File
@@ -364,8 +364,9 @@ static void mm_align_pair(void *km, const mm_mapopt_t *opt, int qlen, const uint
else if (opt->q == opt->q2 && opt->e == opt->e2) else if (opt->q == opt->q2 && opt->e == opt->e2)
ksw_extz2_sse(km, qlen, qseq, tlen, tseq, 5, mat, opt->q, opt->e, w, zdrop, end_bonus, flag, ez); ksw_extz2_sse(km, qlen, qseq, tlen, tseq, 5, mat, opt->q, opt->e, w, zdrop, end_bonus, flag, ez);
else{ else{
#ifdef ALIGN_AVX && (__AVX512BW__ || __AVX2__) #if defined (ALIGN_AVX) && (defined(__AVX512BW__) || (defined(__AVX2__) && defined(APPLY_AVX2)))
#ifdef __AVX512BW__ #ifdef __AVX512BW__
ksw_extd2_avx512(km, qlen, qseq, tlen, tseq, 5, mat, opt->q, opt->e, opt->q2, opt->e2, w, zdrop, end_bonus, flag, ez); ksw_extd2_avx512(km, qlen, qseq, tlen, tseq, 5, mat, opt->q, opt->e, opt->q2, opt->e2, w, zdrop, end_bonus, flag, ez);
#elif __AVX2__ #elif __AVX2__
ksw_extd2_avx2(km, qlen, qseq, tlen, tseq, 5, mat, opt->q, opt->e, opt->q2, opt->e2, w, zdrop, end_bonus, flag, ez); ksw_extd2_avx2(km, qlen, qseq, tlen, tseq, 5, mat, opt->q, opt->e, opt->q2, opt->e2, w, zdrop, end_bonus, flag, ez);
chain.c
+10 -5
View File
@@ -33,7 +33,11 @@ Modified Copyright (C) 2021 Intel Corporation
#include "minimap.h" #include "minimap.h"
#include "mmpriv.h" #include "mmpriv.h"
#include "kalloc.h" #include "kalloc.h"
#if defined(VECTORIZED_CHAINING) && defined(__AVX512BW__)
#include "parallel_chaining_32_bit.h" #include "parallel_chaining_32_bit.h"
#endif
static const char LogTable256[256] = { static const char LogTable256[256] = {
#define LT(n) n, n, n, n, n, n, n, n, n, n, n, n, n, n, n, n #define LT(n) n, n, n, n, n, n, n, n, n, n, n, n, n, n, n, n
-1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, -1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
@@ -53,9 +57,8 @@ mm128_t *mm_chain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int m
{ // TODO: make sure this works when n has more than 32 bits { // TODO: make sure this works when n has more than 32 bits
int32_t k, *p, *t, *v, n_u, n_v; int32_t k, *p, *t, *v, n_u, n_v;
uint32_t *f; uint32_t *f;
int64_t i, j, st = 0; int64_t i, j;
uint64_t *u, *u2, sum_qspan = 0; uint64_t *u, *u2;
float avg_qspan;
mm128_t *b, *w; mm128_t *b, *w;
if (_u) *_u = 0, *n_u_ = 0; if (_u) *_u = 0, *n_u_ = 0;
@@ -68,7 +71,7 @@ mm128_t *mm_chain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int m
t = (int32_t*)kmalloc(km, n * 4); t = (int32_t*)kmalloc(km, n * 4);
v = (int32_t*)kmalloc(km, n * 4); v = (int32_t*)kmalloc(km, n * 4);
memset(t, 0, n * 4); memset(t, 0, n * 4);
#if VECTORIZED_CHAINING && (__AVX512BW__ || __AVX2__) #if defined(VECTORIZED_CHAINING) && defined(__AVX512BW__)
/* Allocation for debugging /* Allocation for debugging
f_avx = (uint32_t*)kmalloc(km, n * 4); f_avx = (uint32_t*)kmalloc(km, n * 4);
p_avx = (int32_t*)kmalloc(km, n * 4); p_avx = (int32_t*)kmalloc(km, n * 4);
@@ -94,7 +97,9 @@ mm128_t *mm_chain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int m
free(anchor_l); free(anchor_l);
free(anchors); free(anchors);
#else #else
int64_t st = 0;
uint64_t sum_qspan = 0;
float avg_qspan;
for (i = 0; i < n; ++i) sum_qspan += a[i].y>>32&0xff; for (i = 0; i < n; ++i) sum_qspan += a[i].y>>32&0xff;
avg_qspan = (float)sum_qspan / n; avg_qspan = (float)sum_qspan / n;
ksw2_extd2_avx.c
+25 -31
View File
@@ -33,8 +33,6 @@ Modified Copyright (C) 2021 Intel Corporation
void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const uint8_t *target, int8_t m, const int8_t *mat, void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const uint8_t *target, int8_t m, const int8_t *mat,
int8_t q, int8_t e, int8_t q2, int8_t e2, int w, int zdrop, int end_bonus, int flag, ksw_extz_t *ez) int8_t q, int8_t e, int8_t q2, int8_t e2, int w, int zdrop, int end_bonus, int flag, ksw_extz_t *ez)
{ {
__m512i ind512_srli = _mm512_set1_epi8(15);
__mmask8 msk_srli = 0x0003;
__m512i bt32_ = _mm512_setr_epi32(0,0,0,0,4,4,4,4,8,8,8,8,12,12,12,12); __m512i bt32_ = _mm512_setr_epi32(0,0,0,0,4,4,4,4,8,8,8,8,12,12,12,12);
int8_t index[64] __attribute((aligned(64))); int8_t index[64] __attribute((aligned(64)));
@@ -105,7 +103,6 @@ void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const
a2= _mm512_sub_epi8(a2, tmp); \ a2= _mm512_sub_epi8(a2, tmp); \
b2= _mm512_sub_epi8(b2, tmp); b2= _mm512_sub_epi8(b2, tmp);
__mmask64 msk_ar[5] = {0xFFFF, 0xFFFFFFFF, 0xFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF};
__mmask64 msk_ar2[5] = {0xFFFF, 0xFFFF, 0xFFFFFFFF, 0xFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF}; __mmask64 msk_ar2[5] = {0xFFFF, 0xFFFF, 0xFFFFFFFF, 0xFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF};
int r, t, qe = q + e, n_col_, *off = 0, *off_end = 0, tlen_, qlen_, last_st, last_en, wl, wr, max_sc, min_sc, long_thres, long_diff; int r, t, qe = q + e, n_col_, *off = 0, *off_end = 0, tlen_, qlen_, last_st, last_en, wl, wr, max_sc, min_sc, long_thres, long_diff;
@@ -201,9 +198,9 @@ void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const
st0 = st, en0 = en; st0 = st, en0 = en;
int st_new = st / 16 * 16, en_new = (en + 16) / 16 * 16 - 1; int st_new = st / 16 * 16, en_new = (en + 16) / 16 * 16 - 1;
// int st_new = st / 64 * 64, en_new = (en + 64) / 64 * 64 - 1; // int st_new = st / 64 * 64, en_new = (en + 64) / 64 * 64 - 1;
int stb = st, enb = en; //int stb = st, enb = en;
st = st / 64 * 64, en = (en + 64) / 64 * 64 - 1; st = st / 64 * 64, en = (en + 64) / 64 * 64 - 1;
int stn = stb / 16 * 16, enn = (enb + 16) / 16 * 16 - 1; //int stn = stb / 16 * 16, enn = (enb + 16) / 16 * 16 - 1;
// set boundary conditions // set boundary conditions
if (st_new > 0) { if (st_new > 0) {
if (st_new - 1 >= last_st && st_new - 1 <= last_en) { if (st_new - 1 >= last_st && st_new - 1 <= last_en) {
@@ -224,7 +221,7 @@ void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const
// loop fission: set scores first // loop fission: set scores first
if (!(flag & KSW_EZ_GENERIC_SC)) { if (!(flag & KSW_EZ_GENERIC_SC)) {
for (t = st0; t <= en0; t += 64) { for (t = st0; t <= en0; t += 64) {
__m512i sq, st, tmp_512, mask_512; __m512i sq, st, tmp_512;
__mmask64 tmp, mask; __mmask64 tmp, mask;
sq = _mm512_loadu_si512((__m512i*)&sf[t]); sq = _mm512_loadu_si512((__m512i*)&sf[t]);
st = _mm512_loadu_si512((__m512i*)&qrr[t]); st = _mm512_loadu_si512((__m512i*)&qrr[t]);
@@ -446,7 +443,6 @@ void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const
// __dp_code_block2_pcl; // __dp_code_block2_pcl;
__mmask64 msk;
{ {
_mm512_store_si512(&u[t], _mm512_sub_epi8(z, vt1)); _mm512_store_si512(&u[t], _mm512_sub_epi8(z, vt1));
_mm512_store_si512(&v[t], _mm512_sub_epi8(z, ut)); _mm512_store_si512(&v[t], _mm512_sub_epi8(z, ut));
@@ -551,7 +547,7 @@ void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const
max_H_ = _mm512_set1_epi32(max_H); max_H_ = _mm512_set1_epi32(max_H);
max_t_ = _mm512_set1_epi32(max_t); max_t_ = _mm512_set1_epi32(max_t);
for (t = st0; t < en1; t += /*4*/16) { // this implements: H[t]+=v8[t]-qe; if(H[t]>max_H) max_H=H[t],max_t=t; for (t = st0; t < en1; t += /*4*/16) { // this implements: H[t]+=v8[t]-qe; if(H[t]>max_H) max_H=H[t],max_t=t;
__m512i H1, tmp, t_; __m512i H1, t_;
__mmask16 tmp_mask; __mmask16 tmp_mask;
H1 = _mm512_loadu_si512((__m512i*)&H[t]); H1 = _mm512_loadu_si512((__m512i*)&H[t]);
__m128i t__ = _mm_load_si128((__m128i*) &v8[t]); __m128i t__ = _mm_load_si128((__m128i*) &v8[t]);
@@ -659,6 +655,8 @@ void ksw_extd2_avx512(void *km, int qlen, const uint8_t *query, int tlen, const
} }
kfree(km, mem2); kfree(km, off); kfree(km, mem2); kfree(km, off);
} }
#undef __dp_code_block1_pcl
#undef __dp_code_block2_pcl
} }
#endif #endif
@@ -692,14 +690,14 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
__m256i shf256a, shf256b, slli256; __m256i shf256a, shf256b, slli256;
__m256i ind256_slli = _mm256_load_si256((__m256i*) index); __m256i ind256_slli = _mm256_load_si256((__m256i*) index);
__mmask8 mska = 0x00;//0x90 //__mmask8 mska = 0x00;//0x90
__mmask32 mskb = 0x00010000;//0000 0000 0000 0001 0000 0000 0000 0000 //__mmask32 mskb = 0x00010000;//0000 0000 0000 0001 0000 0000 0000 0000
__mmask32 mskc = 0x1; //__mmask32 mskc = 0x1;
__m256i mskb_v = _mm256_set_epi32(0,0,0,255,0,0,0,0); __m256i mskb_v = _mm256_set_epi32(0,0,0,255,0,0,0,0);
__m256i mskc_v = _mm256_set_epi32(0,0,0,0,0,0,0,255); __m256i mskc_v = _mm256_set_epi32(0,0,0,0,0,0,0,255);
__mmask32 mskc_ar[2] = {0x1, 0x10000}; //__mmask32 mskc_ar[2] = {0x1, 0x10000};
__m256i mskc_ar_v[2];// = {0x1, 0x10000}; __m256i mskc_ar_v[2];// = {0x1, 0x10000};
mskc_ar_v[0] = _mm256_set_epi32(0,0,0,0,0,0,0,255); mskc_ar_v[0] = _mm256_set_epi32(0,0,0,0,0,0,0,255);
mskc_ar_v[1] = _mm256_set_epi32(0,0,0,255,0,0,0,0); mskc_ar_v[1] = _mm256_set_epi32(0,0,0,255,0,0,0,0);
@@ -760,7 +758,7 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
a2= _mm256_sub_epi8(a2, tmp); \ a2= _mm256_sub_epi8(a2, tmp); \
b2= _mm256_sub_epi8(b2, tmp); b2= _mm256_sub_epi8(b2, tmp);
__mmask32 msk_ar2[3] = {0xFFFF, 0xFFFF, 0xFFFFFFFF}; //__mmask32 msk_ar2[3] = {0xFFFF, 0xFFFF, 0xFFFFFFFF};
__m256i msk_ar2_v[3]; __m256i msk_ar2_v[3];
msk_ar2_v[0] = _mm256_set_epi32(0,0,0,0,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF); msk_ar2_v[0] = _mm256_set_epi32(0,0,0,0,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF);
msk_ar2_v[1] = _mm256_set_epi32(0,0,0,0,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF); msk_ar2_v[1] = _mm256_set_epi32(0,0,0,0,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF);
@@ -861,9 +859,9 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
st0 = st, en0 = en; st0 = st, en0 = en;
int st_new = st / 16 * 16, en_new = (en + 16) / 16 * 16 - 1; int st_new = st / 16 * 16, en_new = (en + 16) / 16 * 16 - 1;
// int st_new = st / 64 * 64, en_new = (en + 64) / 64 * 64 - 1; // int st_new = st / 64 * 64, en_new = (en + 64) / 64 * 64 - 1;
int stb = st, enb = en; //int stb = st, enb = en;
st = st / 32 * 32, en = (en + 32) / 32 * 32 - 1; st = st / 32 * 32, en = (en + 32) / 32 * 32 - 1;
int stn = stb / 16 * 16, enn = (enb + 16) / 16 * 16 - 1; //int stn = stb / 16 * 16, enn = (enb + 16) / 16 * 16 - 1;
// set boundary conditions // set boundary conditions
if (st_new > 0) { if (st_new > 0) {
if (st_new - 1 >= last_st && st_new - 1 <= last_en) { if (st_new - 1 >= last_st && st_new - 1 <= last_en) {
@@ -885,7 +883,6 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
if (!(flag & KSW_EZ_GENERIC_SC)) { if (!(flag & KSW_EZ_GENERIC_SC)) {
for (t = st0; t <= en0; t += 32) { for (t = st0; t <= en0; t += 32) {
__m256i sq, st, tmp_256, mask_256; __m256i sq, st, tmp_256, mask_256;
__mmask32 tmp, mask;
sq = _mm256_loadu_si256((__m256i*)&sf[t]); sq = _mm256_loadu_si256((__m256i*)&sf[t]);
st = _mm256_loadu_si256((__m256i*)&qrr[t]); st = _mm256_loadu_si256((__m256i*)&qrr[t]);
// mask = (_mm256_cmpeq_epi8_mask(sq, m1_) | _mm256_cmpeq_epi8_mask(st, m1_)); // mask = (_mm256_cmpeq_epi8_mask(sq, m1_) | _mm256_cmpeq_epi8_mask(st, m1_));
@@ -897,10 +894,8 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
tmp_256 = _mm256_blendv_epi8(tmp_256, sc_N_, mask_256); tmp_256 = _mm256_blendv_epi8(tmp_256, sc_N_, mask_256);
if (t + 32 > en0) if (t + 32 > en0)
{ {
__mmask32 msk;
int ind = (en0 - t + 16)/16; int ind = (en0 - t + 16)/16;
//assert(ind >= 0 && ind < 3); //assert(ind >= 0 && ind < 3);
msk = msk_ar2[ind];
__m256i msk_v = msk_ar2_v[ind]; __m256i msk_v = msk_ar2_v[ind];
__m256i str = (get_mask_store(msk_v, ((int8_t*)s + t), tmp_256));// msk_ar2_v[ind]; __m256i str = (get_mask_store(msk_v, ((int8_t*)s + t), tmp_256));// msk_ar2_v[ind];
//__m256i str =_mm256_and_si256(get_mask_store(msk_v,(int8_t*)s + t), tmp_256);// msk_ar2_v[ind]; //__m256i str =_mm256_and_si256(get_mask_store(msk_v,(int8_t*)s + t), tmp_256);// msk_ar2_v[ind];
@@ -939,9 +934,7 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
z = _mm256_min_epi8(z, sc_mch_); z = _mm256_min_epi8(z, sc_mch_);
// __dp_code_block2_pcl; // save u[] and v[]; update a, b, a2 and b2 // __dp_code_block2_pcl; // save u[] and v[]; update a, b, a2 and b2
if (t == en_) { if (t == en_) {
__mmask32 msk;
int ind = (en0 - t*32 + 16)/16;//doubt int ind = (en0 - t*32 + 16)/16;//doubt
msk = msk_ar2[ind];
// fprintf(stderr, "en0: %d, t: %d, ind: %d, msk: %d\n", en0, t, ind, msk); // fprintf(stderr, "en0: %d, t: %d, ind: %d, msk: %d\n", en0, t, ind, msk);
_mm256_storeu_si256(&u[t], (get_mask_store(msk_ar2_v[ind], &u[t], _mm256_sub_epi8(z, vt1)))); _mm256_storeu_si256(&u[t], (get_mask_store(msk_ar2_v[ind], &u[t], _mm256_sub_epi8(z, vt1))));
_mm256_storeu_si256(&v[t], (get_mask_store(msk_ar2_v[ind], &v[t], _mm256_sub_epi8(z, ut)))); _mm256_storeu_si256(&v[t], (get_mask_store(msk_ar2_v[ind], &v[t], _mm256_sub_epi8(z, ut))));
@@ -965,10 +958,10 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
} }
if (t == en_) { if (t == en_) {
__mmask32 msk; //__mmask32 msk;
int ind = (en0 - t*32 + 16)/16;//doubt int ind = (en0 - t*32 + 16)/16;//doubt
//assert(ind >= 0); //assert(ind >= 0);
msk = msk_ar2[ind]; //msk = msk_ar2[ind];
// fprintf(stderr, "en0: %d, t: %d, ind: %d, msk: %d\n", en0, t, ind, msk); // fprintf(stderr, "en0: %d, t: %d, ind: %d, msk: %d\n", en0, t, ind, msk);
_mm256_storeu_si256(&x[t], (get_mask_store(msk_ar2_v[ind], &x[t], _mm256_sub_epi8(_mm256_max_epi8(a, zero_), qe_)))); _mm256_storeu_si256(&x[t], (get_mask_store(msk_ar2_v[ind], &x[t], _mm256_sub_epi8(_mm256_max_epi8(a, zero_), qe_))));
_mm256_storeu_si256(&y[t], (get_mask_store(msk_ar2_v[ind],&y[t], _mm256_sub_epi8(_mm256_max_epi8(b, zero_), qe_)))); _mm256_storeu_si256(&y[t], (get_mask_store(msk_ar2_v[ind],&y[t], _mm256_sub_epi8(_mm256_max_epi8(b, zero_), qe_))));
@@ -1047,10 +1040,10 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
z = _mm256_min_epi8(z, sc_mch_); z = _mm256_min_epi8(z, sc_mch_);
// __dp_code_block2_pcl; // __dp_code_block2_pcl;
{ {
__mmask32 msk; //__mmask32 msk;
int ind = (en0 - t*32 + 16)/16;//doubt int ind = (en0 - t*32 + 16)/16;//doubt
// //assert(ind >= 0 && ind < 5); // //assert(ind >= 0 && ind < 5);
msk = msk_ar2[ind]; //msk = msk_ar2[ind];
_mm256_storeu_si256(&u[t], (get_mask_store(msk_ar2_v[ind], &u[t], _mm256_sub_epi8(z, vt1)))); _mm256_storeu_si256(&u[t], (get_mask_store(msk_ar2_v[ind], &u[t], _mm256_sub_epi8(z, vt1))));
_mm256_storeu_si256(&v[t], (get_mask_store(msk_ar2_v[ind],&v[t], _mm256_sub_epi8(z, ut)))); _mm256_storeu_si256(&v[t], (get_mask_store(msk_ar2_v[ind],&v[t], _mm256_sub_epi8(z, ut))));
@@ -1063,9 +1056,9 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
} }
{ {
__mmask32 msk; //__mmask32 msk;
int ind = (en0 - t*32 + 16)/16;//doubt int ind = (en0 - t*32 + 16)/16;//doubt
msk = msk_ar2[ind]; //msk = msk_ar2[ind];
__m256i msk_v= msk_ar2_v[ind]; __m256i msk_v= msk_ar2_v[ind];
off_end[r] -= (2-ind)*16;//doubt off_end[r] -= (2-ind)*16;//doubt
@@ -1116,7 +1109,7 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
// __dp_code_block2_pcl; // __dp_code_block2_pcl;
__mmask32 msk; //__mmask32 msk;
{ {
_mm256_storeu_si256(&u[t], _mm256_sub_epi8(z, vt1)); _mm256_storeu_si256(&u[t], _mm256_sub_epi8(z, vt1));
_mm256_storeu_si256(&v[t], _mm256_sub_epi8(z, ut)); _mm256_storeu_si256(&v[t], _mm256_sub_epi8(z, ut));
@@ -1172,11 +1165,11 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
// __dp_code_block2_pcl; // __dp_code_block2_pcl;
__mmask32 msk; //__mmask32 msk;
// __mmask64 msk; // __mmask64 msk;
int ind = (en0 - t*32 + 16)/16;//doubt int ind = (en0 - t*32 + 16)/16;//doubt
msk = msk_ar2[ind]; //msk = msk_ar2[ind];
off_end[r] -= (2-ind)*16;//doubt off_end[r] -= (2-ind)*16;//doubt
_mm256_storeu_si256(&u[t], (get_mask_store(msk_ar2_v[ind], &u[t], _mm256_sub_epi8(z, vt1)))); _mm256_storeu_si256(&u[t], (get_mask_store(msk_ar2_v[ind], &u[t], _mm256_sub_epi8(z, vt1))));
@@ -1222,7 +1215,7 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
max_H_ = _mm256_set1_epi32(max_H); max_H_ = _mm256_set1_epi32(max_H);
max_t_ = _mm256_set1_epi32(max_t); max_t_ = _mm256_set1_epi32(max_t);
for (t = st0; t < en1; t += /*4*/8) { // this implements: H[t]+=v8[t]-qe; if(H[t]>max_H) max_H=H[t],max_t=t; for (t = st0; t < en1; t += /*4*/8) { // this implements: H[t]+=v8[t]-qe; if(H[t]>max_H) max_H=H[t],max_t=t;
__m256i H1, tmp, t_; __m256i H1, t_;
H1 = _mm256_loadu_si256((__m256i*)&H[t]); H1 = _mm256_loadu_si256((__m256i*)&H[t]);
//__m128i t__ = _mm_load_si128((__m128i*) &v8[t]); //__m128i t__ = _mm_load_si128((__m128i*) &v8[t]);
//t_ = _mm256_cvtepi8_epi32(t__); //t_ = _mm256_cvtepi8_epi32(t__);
@@ -1232,7 +1225,6 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
// making it 4 lanes to match accuracy // making it 4 lanes to match accuracy
__m256i shfH, shft, max1, max2; __m256i shfH, shft, max1, max2;
__mmask8 tmp_mask;
__m256i tmp_mask_v; __m256i tmp_mask_v;
t_ = _mm256_set1_epi32(t); t_ = _mm256_set1_epi32(t);
t_ = _mm256_add_epi32(t_, bt32_); t_ = _mm256_add_epi32(t_, bt32_);
@@ -1340,6 +1332,8 @@ void ksw_extd2_avx2(void *km, int qlen, const uint8_t *query, int tlen, const ui
} }
kfree(km, mem2); kfree(km, off); kfree(km, mem2); kfree(km, off);
} }
#undef __dp_code_block1_pcl
#undef __dp_code_block2_pcl
} }
#endif #endif
main.c
+31
View File
@@ -149,6 +149,33 @@ static inline void yes_or_no(mm_mapopt_t *opt, int flag, int long_idx, const cha
int main(int argc, char *argv[]) int main(int argc, char *argv[])
{ {
#ifdef LISA_HASH
#if VECTORIZE && __AVX512BW__
fprintf(stderr, "Using LISA hash with AVX512-vectorized last-mile search.\n");
#else
fprintf(stderr, "Using LISA hash with sequential last-mile search.\n");
#endif
#else
fprintf(stderr, "Using default hash lookup.\n");
#endif
#if defined(VECTORIZED_CHAINING) && defined(__AVX512BW__)
fprintf(stderr, "Using AVX512-vectorized chaining.\n");
#else
fprintf(stderr, "Using default chaining.\n");
#endif
#if defined (ALIGN_AVX) && (defined(__AVX512BW__) || (defined(__AVX2__) && defined(APPLY_AVX2)))
#ifdef __AVX512BW__
fprintf(stderr, "Using AVX512-vectorized alignment.\n");
#elif __AVX2__
fprintf(stderr, "Using AVX2-vectorized alignment.\n");
#endif
#else
fprintf(stderr, "Using default SSE-vectorized alignment.\n");
#endif
const char *opt_str = "2aSDw:k:K:t:r:f:Vv:g:G:I:d:XT:s:x:Hcp:M:n:z:A:B:O:E:m:N:Qu:R:hF:LC:yYPo:Z:"; const char *opt_str = "2aSDw:k:K:t:r:f:Vv:g:G:I:d:XT:s:x:Hcp:M:n:z:A:B:O:E:m:N:Qu:R:hF:LC:yYPo:Z:";
ketopt_t o = KETOPT_INIT; ketopt_t o = KETOPT_INIT;
mm_mapopt_t opt; mm_mapopt_t opt;
@@ -163,6 +190,7 @@ int main(int argc, char *argv[])
mm_verbose = 3; mm_verbose = 3;
liftrlimit(); liftrlimit();
mm_realtime0 = realtime(); mm_realtime0 = realtime();
double mapping_time = realtime();
mm_set_opt(0, &ipt, &opt); mm_set_opt(0, &ipt, &opt);
string preset_arg = ""; string preset_arg = "";
@@ -453,7 +481,9 @@ int main(int argc, char *argv[])
lh = new lisa_hash<uint64_t, uint64_t>(preset_arg, prefix); lh = new lisa_hash<uint64_t, uint64_t>(preset_arg, prefix);
fprintf(stderr, "Loading done.\n"); fprintf(stderr, "Loading done.\n");
total_time = __rdtsc(); total_time = __rdtsc();
fprintf(stderr, "\nIndexing Real time: %.3f sec;\n", realtime() - mapping_time);
#endif #endif
mapping_time = realtime();
if (!(opt.flag & MM_F_FRAG_MODE)) { if (!(opt.flag & MM_F_FRAG_MODE)) {
for (i = o.ind + 1; i < argc; ++i) { for (i = o.ind + 1; i < argc; ++i) {
ret = mm_map_file(mi, argv[i], &opt, n_threads); ret = mm_map_file(mi, argv[i], &opt, n_threads);
@@ -495,6 +525,7 @@ int main(int argc, char *argv[])
fprintf(stderr, "\n Number of reads = %lld Minimizer hit time = %lld dp_chaining time = %lld alignment time = %lld total time = %lld \n", num_reads, minimizer_hit_time, dp_chaining_time, alignment_time, __rdtsc() - total_time); fprintf(stderr, "\n Number of reads = %lld Minimizer hit time = %lld dp_chaining time = %lld alignment time = %lld total time = %lld \n", num_reads, minimizer_hit_time, dp_chaining_time, alignment_time, __rdtsc() - total_time);
#endif #endif
fprintf(stderr, "Total ticks: %lld \n",__rdtsc() - total_time); fprintf(stderr, "Total ticks: %lld \n",__rdtsc() - total_time);
fprintf(stderr, "\nMapping Real time: %.3f sec;\n", realtime() - mapping_time);
#ifdef LISA_HASH #ifdef LISA_HASH
delete lh; delete lh;
#endif #endif
map.c
-2
View File
@@ -269,8 +269,6 @@ static mm128_t *collect_seed_hits(void *km, const mm_mapopt_t *opt, int max_occ,
#else #else
m = collect_matches_lisa_hash(km, &n_m, max_occ, mi, mv, n_a, rep_len, n_mini_pos, mini_pos); m = collect_matches_lisa_hash(km, &n_m, max_occ, mi, mv, n_a, rep_len, n_mini_pos, mini_pos);
#endif #endif
uint64_t mapping_count = 0;
mapping_count = 0;
a = (mm128_t*)kmalloc(km, *n_a * sizeof(mm128_t)); a = (mm128_t*)kmalloc(km, *n_a * sizeof(mm128_t));
for (i = 0, *n_a = 0; i < n_m; ++i) { for (i = 0, *n_a = 0; i < n_m; ++i) {