fork/minimap2
1
0
Fork 0
mirror of https://github.com/lh3/minimap2.git synced 2026-10-08 07:28:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

disable chaining for avx2 - clr and hifi

Browse Source
This commit is contained in:
Saurabh
2021-11-27 10:42:22 -05:00
committed by Heng Li
parent 3b1deac0a5
commit 03540c47b3
3 changed files with 25 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lchain.c
+13 -28
View File
@@ -10,9 +10,11 @@
//#include "parallel_chaining_32_bit.h" //#include "parallel_chaining_32_bit.h"
#include "parallel_chaining_v2_22.h" #include "parallel_chaining_v2_22.h"
#ifdef MANUAL_PROFILING
extern uint64_t dp_time, rmq_time, rmq_t1, rmq_t2, rmq_t3, rmq_t4; extern uint64_t dp_time, rmq_time, rmq_t1, rmq_t2, rmq_t3, rmq_t4;
#endif
extern bool enable_vect_dp_chaining;
uint64_t *mg_chain_backtrack(void *km, int64_t n, const int32_t *f, const int64_t *p, int32_t *v, int32_t *t, int32_t min_cnt, int32_t min_sc, int32_t *n_u_, int32_t *n_v_) uint64_t *mg_chain_backtrack(void *km, int64_t n, const int32_t *f, const int64_t *p, int32_t *v, int32_t *t, int32_t min_cnt, int32_t min_sc, int32_t *n_u_, int32_t *n_v_)
{ {
mm128_t *z; mm128_t *z;
@@ -215,9 +217,9 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
KMALLOC(km, v_1, n); KMALLOC(km, v_1, n);
KCALLOC(km, t, n); KCALLOC(km, t, n);
#ifdef PARALLEL_CHAINING //#ifdef PARALLEL_CHAINING
if(enable_vect_dp_chaining){
// Parallel chaining data-structures // Parallel chaining data-structures
anchor_t* anchors = (anchor_t*)malloc(n* sizeof(anchor_t)); anchor_t* anchors = (anchor_t*)malloc(n* sizeof(anchor_t));
for (i = 0; i < n; ++i) { for (i = 0; i < n; ++i) {
uint64_t ri = a[i].x; uint64_t ri = a[i].x;
@@ -228,11 +230,11 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
} }
num_bits_t *anchor_r, *anchor_q, *anchor_l; num_bits_t *anchor_r, *anchor_q, *anchor_l;
create_SoA_Anchors_32_bit(anchors, n, anchor_r, anchor_q, anchor_l); create_SoA_Anchors_32_bit(anchors, n, anchor_r, anchor_q, anchor_l);
//dp_chain obj(max_dist_x, max_dist_y, bw, max_skip, max_iter, 0, is_cdna, n_seg);
dp_chain obj(max_dist_x, max_dist_y, bw, max_skip, max_iter, min_cnt, min_sc, chn_pen_gap, chn_pen_skip, is_cdna, n_seg); dp_chain obj(max_dist_x, max_dist_y, bw, max_skip, max_iter, min_cnt, min_sc, chn_pen_gap, chn_pen_skip, is_cdna, n_seg);
#ifdef PARALLEL_CHAINING
obj.mm_dp_vectorized(n, &anchors[0], anchor_r, anchor_q, anchor_l, f_1, p_1, v_1, max_dist_x, max_dist_y, NULL, NULL); obj.mm_dp_vectorized(n, &anchors[0], anchor_r, anchor_q, anchor_l, f_1, p_1, v_1, max_dist_x, max_dist_y, NULL, NULL);
#endif
// -16 is due to extra padding at the start of arrays // -16 is due to extra padding at the start of arrays
anchor_r -= 16; anchor_q -= 16; anchor_l -= 16; anchor_r -= 16; anchor_q -= 16; anchor_l -= 16;
free(anchor_r); free(anchor_r);
@@ -240,10 +242,6 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
free(anchor_l); free(anchor_l);
free(anchors); free(anchors);
for(int i = 0; i < n; i++){ for(int i = 0; i < n; i++){
// if(f[i] != f_1[i] || p[i] != p_1[i] || v[i] !=v_1[i])
// {
// fprintf(stderr, "i:%d %d %d %d %d %d %d\n",i, f[i], f_1[i], p[i], p_1[i], v[i], v_1[i] );
// }
#if 1 #if 1
f[i] = f_1[i]; f[i] = f_1[i];
p[i] = p_1[i]; p[i] = p_1[i];
@@ -252,7 +250,8 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
} }
// //
#else } else {
//#else
// fill the score and backtrack arrays // fill the score and backtrack arrays
for (i = 0, max_ii = -1; i < n; ++i) { for (i = 0, max_ii = -1; i < n; ++i) {
@@ -265,10 +264,6 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
int32_t sc; int32_t sc;
sc = comput_sc(&a[i], &a[j], max_dist_x, max_dist_y, bw, chn_pen_gap, chn_pen_skip, is_cdna, n_seg); sc = comput_sc(&a[i], &a[j], max_dist_x, max_dist_y, bw, chn_pen_gap, chn_pen_skip, is_cdna, n_seg);
++n_iter; ++n_iter;
// if(i == 177){
//fprintf(stderr, "args: %d %d %d %d %d\n", a[i].x, a[i].y, a[j].x, a[j].y, a[j].y>>32&0xff);
//fprintf(stderr, "j_th %d score: %d\n", ++my_cnt, sc);
// }
if (sc == INT32_MIN) continue; if (sc == INT32_MIN) continue;
sc += f[j]; sc += f[j];
if (sc > max_f) { if (sc > max_f) {
@@ -282,9 +277,7 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
} }
end_j = j; end_j = j;
int debug_iter = 2057329; int debug_iter = 2057329;
//if (i == debug_iter) fprintf(stderr, "mm2 -- endj: %d max_ii: %d max_f: %d \n", end_j, max_ii, max_f);
#if 1
if (max_ii < 0 || a[i].x - a[max_ii].x > (int64_t)max_dist_x) { if (max_ii < 0 || a[i].x - a[max_ii].x > (int64_t)max_dist_x) {
int32_t max = INT32_MIN; int32_t max = INT32_MIN;
max_ii = -1; max_ii = -1;
@@ -292,33 +285,26 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
if (max < (int32_t)f[j]) max = f[j], max_ii = j; if (max < (int32_t)f[j]) max = f[j], max_ii = j;
} }
} }
#endif
#if 1
if (max_ii >= 0 && max_ii < end_j) { if (max_ii >= 0 && max_ii < end_j) {
int32_t tmp; int32_t tmp;
tmp = comput_sc(&a[i], &a[max_ii], max_dist_x, max_dist_y, bw, chn_pen_gap, chn_pen_skip, is_cdna, n_seg); tmp = comput_sc(&a[i], &a[max_ii], max_dist_x, max_dist_y, bw, chn_pen_gap, chn_pen_skip, is_cdna, n_seg);
// if (i == debug_iter) fprintf(stderr, "mm2: endj: %d max_ii: %d max_f: %d tmp_score: %d \n", end_j, max_ii, max_f, tmp); // if (i == debug_iter) fprintf(stderr, "mm2: endj: %d max_ii: %d max_f: %d tmp_score: %d \n", end_j, max_ii, max_f, tmp);
if (tmp != INT32_MIN && max_f < tmp + f[max_ii]) { if (tmp != INT32_MIN && max_f < tmp + f[max_ii]){
// if (i == debug_iter) fprintf(stderr, "mm2: endj: %d max_ii: %d max_f: %d tmp_score: %d \n", end_j, max_ii, max_f, tmp);
max_f = tmp + f[max_ii], max_j = max_ii; max_f = tmp + f[max_ii], max_j = max_ii;
// if (i == debug_iter) fprintf(stderr, "mm2: endj: %d max_ii: %d max_f: %d tmp_score: %d sum : %d \n", end_j, max_ii, max_f, tmp, tmp + f[max_ii]);
} }
} }
#endif
f[i] = max_f, p[i] = max_j; f[i] = max_f, p[i] = max_j;
v[i] = max_j >= 0 && v[max_j] > max_f? v[max_j] : max_f; // v[] keeps the peak score up to i; f[] is the score ending at i, not always the peak v[i] = max_j >= 0 && v[max_j] > max_f? v[max_j] : max_f; // v[] keeps the peak score up to i; f[] is the score ending at i, not always the peak
#if 1
if (max_ii < 0 || (a[i].x - a[max_ii].x <= (int64_t)max_dist_x && f[max_ii] < f[i])) if (max_ii < 0 || (a[i].x - a[max_ii].x <= (int64_t)max_dist_x && f[max_ii] < f[i]))
max_ii = i; max_ii = i;
if (mmax_f < max_f) mmax_f = max_f; if (mmax_f < max_f) mmax_f = max_f;
#endif
} }
#endif }
//#endif
#ifdef CHAIN_DEBUG #ifdef CHAIN_DEBUG
@@ -333,7 +319,6 @@ mm128_t *mg_lchain_dp(int max_dist_x, int max_dist_y, int bw, int max_skip, int
v[i] = v_1[i]; v[i] = v_1[i];
#endif #endif
} }
#endif #endif
u = mg_chain_backtrack(km, n, f, p, v, t, min_cnt, min_sc, &n_u, &n_v); u = mg_chain_backtrack(km, n, f, p, v, t, min_cnt, min_sc, &n_u, &n_v);
*n_u_ = n_u, *_u = u; // NB: note that u[] may not be sorted by score here *n_u_ = n_u, *_u = u; // NB: note that u[] may not be sorted by score here
main.c
+5
View File
@@ -29,6 +29,8 @@ using namespace std;
uint64_t avg; uint64_t avg;
uint64_t minimizer_lookup_time, alignment_time, dp_time, rmq_time, rmq_t1, rmq_t2, rmq_t3, rmq_t4; uint64_t minimizer_lookup_time, alignment_time, dp_time, rmq_time, rmq_t1, rmq_t2, rmq_t3, rmq_t4;
bool enable_vect_dp_chaining = false;
#ifdef LISA_HASH #ifdef LISA_HASH
#include "lisa_hash.h" #include "lisa_hash.h"
lisa_hash<uint64_t, uint64_t> *lh; lisa_hash<uint64_t, uint64_t> *lh;
@@ -186,6 +188,9 @@ int main(int argc, char *argv[])
{ {
// Memory allocation for alignment optimizations // Memory allocation for alignment optimizations
//km1 = calloc(km_size, 1); // 10 MB init contg. alloc //km1 = calloc(km_size, 1); // 10 MB init contg. alloc
#ifdef PARALLEL_CHAINING
enable_vect_dp_chaining = true;
#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:e:U:"; 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:e:U:";
ketopt_t o = KETOPT_INIT; ketopt_t o = KETOPT_INIT;
options.c
+7 -1
View File
@@ -1,7 +1,7 @@
#include <stdio.h> #include <stdio.h>
#include <limits.h> #include <limits.h>
#include "mmpriv.h" #include "mmpriv.h"
extern bool enable_vect_dp_chaining;
void mm_idxopt_init(mm_idxopt_t *opt) void mm_idxopt_init(mm_idxopt_t *opt)
{ {
memset(opt, 0, sizeof(mm_idxopt_t)); memset(opt, 0, sizeof(mm_idxopt_t));
@@ -94,6 +94,9 @@ int mm_set_opt(const char *preset, mm_idxopt_t *io, mm_mapopt_t *mo)
mo->bw = mo->bw_long = 2000; mo->bw = mo->bw_long = 2000;
mo->occ_dist = 0; mo->occ_dist = 0;
} else if (strcmp(preset, "map10k") == 0 || strcmp(preset, "map-pb") == 0) { } else if (strcmp(preset, "map10k") == 0 || strcmp(preset, "map-pb") == 0) {
#if defined (PARALLEL_CHAINING) && (defined(__AVX2__))
enable_vect_dp_chaining = false;
#endif
io->flag |= MM_I_HPC, io->k = 19; io->flag |= MM_I_HPC, io->k = 19;
} else if (strcmp(preset, "ava-pb") == 0) { } else if (strcmp(preset, "ava-pb") == 0) {
io->flag |= MM_I_HPC, io->k = 19, io->w = 5; io->flag |= MM_I_HPC, io->k = 19, io->w = 5;
@@ -102,6 +105,9 @@ int mm_set_opt(const char *preset, mm_idxopt_t *io, mm_mapopt_t *mo)
mo->bw_long = mo->bw; mo->bw_long = mo->bw;
mo->occ_dist = 0; mo->occ_dist = 0;
} else if (strcmp(preset, "map-hifi") == 0 || strcmp(preset, "map-ccs") == 0) { } else if (strcmp(preset, "map-hifi") == 0 || strcmp(preset, "map-ccs") == 0) {
#if defined (PARALLEL_CHAINING) && (defined(__AVX2__))
enable_vect_dp_chaining = false;
#endif
io->flag = 0, io->k = 19, io->w = 19; io->flag = 0, io->k = 19, io->w = 19;
mo->max_gap = 10000; mo->max_gap = 10000;
mo->a = 1, mo->b = 4, mo->q = 6, mo->q2 = 26, mo->e = 2, mo->e2 = 1; mo->a = 1, mo->b = 4, mo->q = 6, mo->q2 = 26, mo->e = 2, mo->e2 = 1;