From 4badf2fcbf4e1fa8906bd2b28350b54371223fe7 Mon Sep 17 00:00:00 2001
From: Heng Li <lh3@me.com>
Date: Tue, 8 Aug 2017 11:29:46 -0400
Subject: [PATCH] avoid wasted memory allocation for backtrack

---
 ksw2_extd2_sse.c | 5 ++++-
 ksw2_extz2_sse.c | 5 ++++-
 2 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/ksw2_extd2_sse.c b/ksw2_extd2_sse.c
index e0f5337..c7f5843 100644
--- a/ksw2_extd2_sse.c
+++ b/ksw2_extd2_sse.c
@@ -1,5 +1,6 @@
 #include <string.h>
 #include <stdio.h>
+#include <assert.h>
 #include "ksw2.h"
 
 #ifdef __SSE2__
@@ -66,7 +67,8 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 	if (w < 0) w = tlen > qlen? tlen : qlen;
 	wl = wr = w;
 	tlen_ = (tlen + 15) / 16;
-	n_col_ = ((w + 1 < tlen? (w + 1 < qlen? w + 1 : qlen): tlen) + 15) / 16 + 1;
+	n_col_ = qlen < tlen? qlen : tlen;
+	n_col_ = ((n_col_ < w + 1? n_col_ : w + 1) + 15) / 16 + 1;
 	qlen_ = (qlen + 15) / 16;
 	for (t = 1, max_sc = mat[0], min_sc = mat[1]; t < m * m; ++t) {
 		max_sc = max_sc > mat[t]? max_sc : mat[t];
@@ -161,6 +163,7 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 		x21_ = _mm_cvtsi32_si128((uint8_t)x21);
 		v1_  = _mm_cvtsi32_si128((uint8_t)v1);
 		st_ = st / 16, en_ = en / 16;
+		assert(en_ - st_ + 1 <= n_col_);
 		if (!with_cigar) { // score only
 			for (t = st_; t <= en_; ++t) {
 				__m128i z, a, b, a2, b2, xt1, x2t1, vt1, ut, tmp;
diff --git a/ksw2_extz2_sse.c b/ksw2_extz2_sse.c
index d665c6f..04669b9 100644
--- a/ksw2_extz2_sse.c
+++ b/ksw2_extz2_sse.c
@@ -1,4 +1,5 @@
 #include <string.h>
+#include <assert.h>
 #include "ksw2.h"
 
 #ifdef __SSE2__
@@ -58,7 +59,8 @@ void ksw_extz2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 	if (w < 0) w = tlen > qlen? tlen : qlen;
 	wl = wr = w;
 	tlen_ = (tlen + 15) / 16;
-	n_col_ = ((w + 1 < tlen? (w + 1 < qlen? w + 1 : qlen): tlen) + 15) / 16 + 1;
+	n_col_ = qlen < tlen? qlen : tlen;
+	n_col_ = ((n_col_ < w + 1? n_col_ : w + 1) + 15) / 16 + 1;
 	qlen_ = (qlen + 15) / 16;
 	for (t = 1, max_sc = mat[0], min_sc = mat[1]; t < m * m; ++t) {
 		max_sc = max_sc > mat[t]? max_sc : mat[t];
@@ -130,6 +132,7 @@ void ksw_extz2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 		x1_ = _mm_cvtsi32_si128(x1);
 		v1_ = _mm_cvtsi32_si128(v1);
 		st_ = st / 16, en_ = en / 16;
+		assert(en_ - st_ + 1 <= n_col_);
 		if (!with_cigar) { // score only
 			for (t = st_; t <= en_; ++t) {
 				__m128i z, a, b, xt1, vt1, ut, tmp;