diff --git a/Makefile b/Makefile
index ed341f6..ee7bd8e 100644
--- a/Makefile
+++ b/Makefile
@@ -8,7 +8,11 @@ LIBS=		-lm -lz -lpthread
 
 ifeq ($(arm_neon),) # if arm_neon is not defined
 ifeq ($(sse2only),) # if sse2only is not defined
+ifeq ($(avx2),)
 	OBJS+=ksw2_extz2_sse41.o ksw2_extd2_sse41.o ksw2_exts2_sse41.o ksw2_extz2_sse2.o ksw2_extd2_sse2.o ksw2_exts2_sse2.o ksw2_dispatch.o
+else
+	OBJS+=ksw2_extd2_avx2.o ksw2_extz2_sse41.o ksw2_extd2_sse41.o ksw2_exts2_sse41.o ksw2_extz2_sse2.o ksw2_extd2_sse2.o ksw2_exts2_sse2.o ksw2_dispatch.o
+endif
 else                # if sse2only is defined
 	OBJS+=ksw2_extz2_sse.o ksw2_extd2_sse.o ksw2_exts2_sse.o
 endif
@@ -57,11 +61,14 @@ ksw2_extz2_sse41.o:ksw2_extz2_sse.c ksw2.h kalloc.h
 ksw2_extz2_sse2.o:ksw2_extz2_sse.c ksw2.h kalloc.h
 		$(CC) -c $(CFLAGS) -msse2 -mno-sse4.1 $(CPPFLAGS) -DKSW_CPU_DISPATCH -DKSW_SSE2_ONLY $(INCLUDES) $< -o $@
 
+ksw2_extd2_avx2.o:ksw2_extd2_sse.c ksw2.h kalloc.h
+		$(CC) -c $(CFLAGS) -mavx2 $(CPPFLAGS) -DKSW_CPU_DISPATCH $(INCLUDES) $< -o $@
+
 ksw2_extd2_sse41.o:ksw2_extd2_sse.c ksw2.h kalloc.h
-		$(CC) -c $(CFLAGS) -msse4.1 $(CPPFLAGS) -DKSW_CPU_DISPATCH $(INCLUDES) $< -o $@
+		$(CC) -c $(CFLAGS) -msse4.1 -mno-avx2 $(CPPFLAGS) -DKSW_CPU_DISPATCH $(INCLUDES) $< -o $@
 
 ksw2_extd2_sse2.o:ksw2_extd2_sse.c ksw2.h kalloc.h
-		$(CC) -c $(CFLAGS) -msse2 -mno-sse4.1 $(CPPFLAGS) -DKSW_CPU_DISPATCH -DKSW_SSE2_ONLY $(INCLUDES) $< -o $@
+		$(CC) -c $(CFLAGS) -msse2 -mno-sse4.1 -mno-avx2 $(CPPFLAGS) -DKSW_CPU_DISPATCH -DKSW_SSE2_ONLY $(INCLUDES) $< -o $@
 
 ksw2_exts2_sse41.o:ksw2_exts2_sse.c ksw2.h kalloc.h
 		$(CC) -c $(CFLAGS) -msse4.1 $(CPPFLAGS) -DKSW_CPU_DISPATCH $(INCLUDES) $< -o $@
@@ -69,8 +76,13 @@ ksw2_exts2_sse41.o:ksw2_exts2_sse.c ksw2.h kalloc.h
 ksw2_exts2_sse2.o:ksw2_exts2_sse.c ksw2.h kalloc.h
 		$(CC) -c $(CFLAGS) -msse2 -mno-sse4.1 $(CPPFLAGS) -DKSW_CPU_DISPATCH -DKSW_SSE2_ONLY $(INCLUDES) $< -o $@
 
+ifeq ($(avx2),)
 ksw2_dispatch.o:ksw2_dispatch.c ksw2.h
 		$(CC) -c $(CFLAGS) -msse4.1 $(CPPFLAGS) -DKSW_CPU_DISPATCH $(INCLUDES) $< -o $@
+else
+ksw2_dispatch.o:ksw2_dispatch.c ksw2.h
+		$(CC) -c $(CFLAGS) -mavx2 $(CPPFLAGS) -DKSW_CPU_DISPATCH $(INCLUDES) $< -o $@
+endif
 
 # NEON-specific targets on ARM
 
diff --git a/ksw2_dispatch.c b/ksw2_dispatch.c
index 22ca9bc..79bf191 100644
--- a/ksw2_dispatch.c
+++ b/ksw2_dispatch.c
@@ -71,7 +71,14 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 				   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);
 	extern void ksw_extd2_sse41(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);
+	extern void ksw_extd2_avx2(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);
 	if (ksw_simd < 0) ksw_simd = x86_simd();
+#ifdef __AVX2__
+	if (ksw_simd & SIMD_AVX2)
+		ksw_extd2_avx2(km, qlen, query, tlen, target, m, mat, q, e, q2, e2, w, zdrop, end_bonus, flag, ez);
+	else
+#endif
 	if (ksw_simd & SIMD_SSE4_1)
 		ksw_extd2_sse41(km, qlen, query, tlen, target, m, mat, q, e, q2, e2, w, zdrop, end_bonus, flag, ez);
 	else if (ksw_simd & SIMD_SSE2)
diff --git a/ksw2_extd2_sse.c b/ksw2_extd2_sse.c
index b578274..f19a51d 100644
--- a/ksw2_extd2_sse.c
+++ b/ksw2_extd2_sse.c
@@ -4,7 +4,20 @@
 #include "ksw2.h"
 
 #ifdef __SSE2__
+
+#if defined(__AVX2__)
+#include <immintrin.h>
+#define SIMD_INT __m256i
+#define SIMD_SHIFT 5
+#define simd_func(func) _mm256_##func
+#define simd_funcw(func) _mm256_##func##_si256
+
+#elif defined(__SSE2__)
 #include <emmintrin.h>
+#define SIMD_INT __m128i
+#define SIMD_SHIFT 4
+#define simd_func(func) _mm_##func
+#define simd_funcw(func) _mm_##func##_si128
 
 #ifdef KSW_SSE2_ONLY
 #undef __SSE4_1__
@@ -13,12 +26,19 @@
 #ifdef __SSE4_1__
 #include <smmintrin.h>
 #endif
+#endif // defined(__SSE2__)
+
+#define SIMD_WIDTH (1<<SIMD_SHIFT)
+
 
 #ifdef KSW_CPU_DISPATCH
-#ifdef __SSE4_1__
+#if defined(__AVX2__)
+void ksw_extd2_avx2(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)
+#elif defined(__SSE4_1__)
 void ksw_extd2_sse41(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)
-#else
+#elif defined(__SSE2__)
 void ksw_extd2_sse2(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)
 #endif
@@ -28,63 +48,63 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 #endif // ~KSW_CPU_DISPATCH
 {
 #define __dp_code_block1 \
-	z = _mm_load_si128(&s[t]); \
-	xt1 = _mm_load_si128(&x[t]);                     /* xt1 <- x[r-1][t..t+15] */ \
-	tmp = _mm_srli_si128(xt1, 15);                   /* tmp <- x[r-1][t+15] */ \
-	xt1 = _mm_or_si128(_mm_slli_si128(xt1, 1), x1_); /* xt1 <- x[r-1][t-1..t+14] */ \
+	z = simd_funcw(load)(&s[t]); \
+	xt1 = simd_funcw(load)(&x[t]);                        /* xt1 <- x[r-1][t..t+15] */ \
+	tmp = simd_funcw(srli)(xt1, SIMD_WIDTH-1);            /* tmp <- x[r-1][t+15] */ \
+	xt1 = simd_funcw(or)(simd_funcw(slli)(xt1, 1), x1_);  /* xt1 <- x[r-1][t-1..t+14] */ \
 	x1_ = tmp; \
-	vt1 = _mm_load_si128(&v[t]);                     /* vt1 <- v[r-1][t..t+15] */ \
-	tmp = _mm_srli_si128(vt1, 15);                   /* tmp <- v[r-1][t+15] */ \
-	vt1 = _mm_or_si128(_mm_slli_si128(vt1, 1), v1_); /* vt1 <- v[r-1][t-1..t+14] */ \
+	vt1 = simd_funcw(load)(&v[t]);                        /* vt1 <- v[r-1][t..t+15] */ \
+	tmp = simd_funcw(srli)(vt1, SIMD_WIDTH-1);            /* tmp <- v[r-1][t+15] */ \
+	vt1 = simd_funcw(or)(simd_funcw(slli)(vt1, 1), v1_);  /* vt1 <- v[r-1][t-1..t+14] */ \
 	v1_ = tmp; \
-	a = _mm_add_epi8(xt1, vt1);                      /* a <- x[r-1][t-1..t+14] + v[r-1][t-1..t+14] */ \
-	ut = _mm_load_si128(&u[t]);                      /* ut <- u[t..t+15] */ \
-	b = _mm_add_epi8(_mm_load_si128(&y[t]), ut);     /* b <- y[r-1][t..t+15] + u[r-1][t..t+15] */ \
-	x2t1= _mm_load_si128(&x2[t]); \
-	tmp = _mm_srli_si128(x2t1, 15); \
-	x2t1= _mm_or_si128(_mm_slli_si128(x2t1, 1), x21_); \
+	a = simd_func(add_epi8)(xt1, vt1);                    /* a <- x[r-1][t-1..t+14] + v[r-1][t-1..t+14] */ \
+	ut = simd_funcw(load)(&u[t]);                         /* ut <- u[t..t+15] */ \
+	b = simd_func(add_epi8)(simd_funcw(load)(&y[t]), ut); /* b <- y[r-1][t..t+15] + u[r-1][t..t+15] */ \
+	x2t1= simd_funcw(load)(&x2[t]); \
+	tmp = simd_funcw(srli)(x2t1, SIMD_WIDTH-1); \
+	x2t1= simd_funcw(or)(simd_funcw(slli)(x2t1, 1), x21_); \
 	x21_= tmp; \
-	a2= _mm_add_epi8(x2t1, vt1); \
-	b2= _mm_add_epi8(_mm_load_si128(&y2[t]), ut);
+	a2= simd_func(add_epi8)(x2t1, vt1); \
+	b2= simd_func(add_epi8)(simd_funcw(load)(&y2[t]), ut);
 
 #define __dp_code_block2 \
-	_mm_store_si128(&u[t], _mm_sub_epi8(z, vt1));    /* u[r][t..t+15] <- z - v[r-1][t-1..t+14] */ \
-	_mm_store_si128(&v[t], _mm_sub_epi8(z, ut));     /* v[r][t..t+15] <- z - u[r-1][t..t+15] */ \
-	tmp = _mm_sub_epi8(z, q_); \
-	a = _mm_sub_epi8(a, tmp); \
-	b = _mm_sub_epi8(b, tmp); \
-	tmp = _mm_sub_epi8(z, q2_); \
-	a2= _mm_sub_epi8(a2, tmp); \
-	b2= _mm_sub_epi8(b2, tmp);
+	simd_funcw(store)(&u[t], simd_func(sub_epi8)(z, vt1));    /* u[r][t..t+15] <- z - v[r-1][t-1..t+14] */ \
+	simd_funcw(store)(&v[t], simd_func(sub_epi8)(z, ut));     /* v[r][t..t+15] <- z - u[r-1][t..t+15] */ \
+	tmp = simd_func(sub_epi8)(z, q_); \
+	a = simd_func(sub_epi8)(a, tmp); \
+	b = simd_func(sub_epi8)(b, tmp); \
+	tmp = simd_func(sub_epi8)(z, q2_); \
+	a2= simd_func(sub_epi8)(a2, tmp); \
+	b2= simd_func(sub_epi8)(b2, tmp);
 
 	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 with_cigar = !(flag&KSW_EZ_SCORE_ONLY), approx_max = !!(flag&KSW_EZ_APPROX_MAX);
 	int32_t *H = 0, H0 = 0, last_H0_t = 0;
 	uint8_t *qr, *sf, *mem, *mem2 = 0;
-	__m128i q_, q2_, qe_, qe2_, zero_, sc_mch_, sc_mis_, m1_, sc_N_;
-	__m128i *u, *v, *x, *y, *x2, *y2, *s, *p = 0;
+	SIMD_INT q_, q2_, qe_, qe2_, zero_, sc_mch_, sc_mis_, m1_, sc_N_;
+	SIMD_INT *u, *v, *x, *y, *x2, *y2, *s, *p = 0;
 
 	ksw_reset_extz(ez);
 	if (m <= 1 || qlen <= 0 || tlen <= 0) return;
 
 	if (q2 + e2 < q + e) t = q, q = q2, q2 = t, t = e, e = e2, e2 = t; // make sure q+e no larger than q2+e2
 
-	zero_   = _mm_set1_epi8(0);
-	q_      = _mm_set1_epi8(q);
-	q2_     = _mm_set1_epi8(q2);
-	qe_     = _mm_set1_epi8(q + e);
-	qe2_    = _mm_set1_epi8(q2 + e2);
-	sc_mch_ = _mm_set1_epi8(mat[0]);
-	sc_mis_ = _mm_set1_epi8(mat[1]);
-	sc_N_   = mat[m*m-1] == 0? _mm_set1_epi8(-e2) : _mm_set1_epi8(mat[m*m-1]);
-	m1_     = _mm_set1_epi8(m - 1); // wildcard
+	zero_   = simd_func(set1_epi8)(0);
+	q_      = simd_func(set1_epi8)(q);
+	q2_     = simd_func(set1_epi8)(q2);
+	qe_     = simd_func(set1_epi8)(q + e);
+	qe2_    = simd_func(set1_epi8)(q2 + e2);
+	sc_mch_ = simd_func(set1_epi8)(mat[0]);
+	sc_mis_ = simd_func(set1_epi8)(mat[1]);
+	sc_N_   = mat[m*m-1] == 0? simd_func(set1_epi8)(-e2) : simd_func(set1_epi8)(mat[m*m-1]);
+	m1_     = simd_func(set1_epi8)(m - 1); // wildcard
 
 	if (w < 0) w = tlen > qlen? tlen : qlen;
 	wl = wr = w;
-	tlen_ = (tlen + 15) / 16;
+	tlen_ = (tlen + SIMD_WIDTH - 1) / SIMD_WIDTH;
 	n_col_ = qlen < tlen? qlen : tlen;
-	n_col_ = ((n_col_ < w + 1? n_col_ : w + 1) + 15) / 16 + 1;
-	qlen_ = (qlen + 15) / 16;
+	n_col_ = ((n_col_ < w + 1? n_col_ : w + 1) + SIMD_WIDTH - 1) / SIMD_WIDTH + 1;
+	qlen_ = (qlen + SIMD_WIDTH - 1) / SIMD_WIDTH;
 	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];
 		min_sc = min_sc < mat[t]? min_sc : mat[t];
@@ -96,23 +116,23 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 		++long_thres;
 	long_diff = long_thres * (e - e2) - (q2 - q) - e2;
 
-	mem = (uint8_t*)kcalloc(km, tlen_ * 8 + qlen_ + 1, 16);
-	u = (__m128i*)(((size_t)mem + 15) >> 4 << 4); // 16-byte aligned
+	mem = (uint8_t*)kcalloc(km, tlen_ * 8 + qlen_ + 1, SIMD_WIDTH);
+	u = (SIMD_INT*)(((size_t)mem + SIMD_WIDTH - 1) >> SIMD_SHIFT << SIMD_SHIFT); // 16-byte aligned
 	v = u + tlen_, x = v + tlen_, y = x + tlen_, x2 = y + tlen_, y2 = x2 + tlen_;
-	s = y2 + tlen_, sf = (uint8_t*)(s + tlen_), qr = sf + tlen_ * 16;
-	memset(u,  -q  - e,  tlen_ * 16);
-	memset(v,  -q  - e,  tlen_ * 16);
-	memset(x,  -q  - e,  tlen_ * 16);
-	memset(y,  -q  - e,  tlen_ * 16);
-	memset(x2, -q2 - e2, tlen_ * 16);
-	memset(y2, -q2 - e2, tlen_ * 16);
+	s = y2 + tlen_, sf = (uint8_t*)(s + tlen_), qr = sf + tlen_ * SIMD_WIDTH;
+	memset(u,  -q  - e,  tlen_ * SIMD_WIDTH);
+	memset(v,  -q  - e,  tlen_ * SIMD_WIDTH);
+	memset(x,  -q  - e,  tlen_ * SIMD_WIDTH);
+	memset(y,  -q  - e,  tlen_ * SIMD_WIDTH);
+	memset(x2, -q2 - e2, tlen_ * SIMD_WIDTH);
+	memset(y2, -q2 - e2, tlen_ * SIMD_WIDTH);
 	if (!approx_max) {
-		H = (int32_t*)kmalloc(km, tlen_ * 16 * 4);
-		for (t = 0; t < tlen_ * 16; ++t) H[t] = KSW_NEG_INF;
+		H = (int32_t*)kmalloc(km, tlen_ * SIMD_WIDTH * 4);
+		for (t = 0; t < tlen_ * SIMD_WIDTH; ++t) H[t] = KSW_NEG_INF;
 	}
 	if (with_cigar) {
-		mem2 = (uint8_t*)kmalloc(km, ((size_t)(qlen + tlen - 1) * n_col_ + 1) * 16);
-		p = (__m128i*)(((size_t)mem2 + 15) >> 4 << 4);
+		mem2 = (uint8_t*)kmalloc(km, ((size_t)(qlen + tlen - 1) * n_col_ + 1) * SIMD_WIDTH);
+		p = (SIMD_INT*)(((size_t)mem2 + SIMD_WIDTH - 1) >> SIMD_SHIFT << SIMD_SHIFT);
 		off = (int*)kmalloc(km, (qlen + tlen - 1) * sizeof(int) * 2);
 		off_end = off + qlen + tlen - 1;
 	}
@@ -125,7 +145,7 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 		int8_t x1, x21, v1;
 		uint8_t *qrr = qr + (qlen - 1 - r);
 		int8_t *u8 = (int8_t*)u, *v8 = (int8_t*)v, *x8 = (int8_t*)x, *x28 = (int8_t*)x2;
-		__m128i x1_, x21_, v1_;
+		SIMD_INT x1_, x21_, v1_;
 		// find the boundaries
 		if (st < r - qlen + 1) st = r - qlen + 1;
 		if (en > r) en = r;
@@ -136,7 +156,7 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 			break;
 		}
 		st0 = st, en0 = en;
-		st = st / 16 * 16, en = (en + 16) / 16 * 16 - 1;
+		st = st / SIMD_WIDTH * SIMD_WIDTH, en = (en + SIMD_WIDTH) / SIMD_WIDTH * SIMD_WIDTH - 1;
 		// set boundary conditions
 		if (st > 0) {
 			if (st - 1 >= last_st && st - 1 <= last_en) {
@@ -155,47 +175,53 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 		}
 		// loop fission: set scores first
 		if (!(flag & KSW_EZ_GENERIC_SC)) {
-			for (t = st0; t <= en0; t += 16) {
-				__m128i sq, st, tmp, mask;
-				sq = _mm_loadu_si128((__m128i*)&sf[t]);
-				st = _mm_loadu_si128((__m128i*)&qrr[t]);
-				mask = _mm_or_si128(_mm_cmpeq_epi8(sq, m1_), _mm_cmpeq_epi8(st, m1_));
-				tmp = _mm_cmpeq_epi8(sq, st);
-#ifdef __SSE4_1__
-				tmp = _mm_blendv_epi8(sc_mis_, sc_mch_, tmp);
-				tmp = _mm_blendv_epi8(tmp,     sc_N_,   mask);
-#else
+			for (t = st0; t <= en0; t += SIMD_WIDTH) {
+				SIMD_INT sq, st, tmp, mask;
+				sq = simd_funcw(loadu)((SIMD_INT*)&sf[t]);
+				st = simd_funcw(loadu)((SIMD_INT*)&qrr[t]);
+				mask = simd_funcw(or)(simd_func(cmpeq_epi8)(sq, m1_), simd_func(cmpeq_epi8)(st, m1_));
+				tmp = simd_func(cmpeq_epi8)(sq, st);
+#if defined(__SSE4_1__) || defined(__AVX2__)
+				tmp = simd_func(blendv_epi8)(sc_mis_, sc_mch_, tmp);
+				tmp = simd_func(blendv_epi8)(tmp,     sc_N_,   mask);
+#elif defined(__SSE2__) // emulate blendv
 				tmp = _mm_or_si128(_mm_andnot_si128(tmp,  sc_mis_), _mm_and_si128(tmp,  sc_mch_));
 				tmp = _mm_or_si128(_mm_andnot_si128(mask, tmp),     _mm_and_si128(mask, sc_N_));
 #endif
-				_mm_storeu_si128((__m128i*)((int8_t*)s + t), tmp);
+				simd_funcw(storeu)((SIMD_INT*)((int8_t*)s + t), tmp);
 			}
 		} else {
 			for (t = st0; t <= en0; ++t)
 				((uint8_t*)s)[t] = mat[sf[t] * m + qrr[t]];
 		}
 		// core loop
-		x1_  = _mm_cvtsi32_si128((uint8_t)x1);
-		x21_ = _mm_cvtsi32_si128((uint8_t)x21);
-		v1_  = _mm_cvtsi32_si128((uint8_t)v1);
-		st_ = st / 16, en_ = en / 16;
+#if defined(__AVX2__)
+		x1_  = _mm256_setr_epi32((uint8_t)x1,  0, 0, 0, 0, 0, 0, 0);
+		x21_ = _mm256_setr_epi32((uint8_t)x21, 0, 0, 0, 0, 0, 0, 0);
+		v1_  = _mm256_setr_epi32((uint8_t)v1,  0, 0, 0, 0, 0, 0, 0);
+#elif defined(__SSE2__)
+		x1_  = _mm_setr_epi32((uint8_t)x1,  0, 0, 0);
+		x21_ = _mm_setr_epi32((uint8_t)x21, 0, 0, 0);
+		v1_  = _mm_setr_epi32((uint8_t)v1,  0, 0, 0);
+#endif
+		st_ = st / SIMD_WIDTH, en_ = en / SIMD_WIDTH;
 		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;
+				SIMD_INT z, a, b, a2, b2, xt1, x2t1, vt1, ut, tmp;
 				__dp_code_block1;
-#ifdef __SSE4_1__
-				z = _mm_max_epi8(z, a);
-				z = _mm_max_epi8(z, b);
-				z = _mm_max_epi8(z, a2);
-				z = _mm_max_epi8(z, b2);
-				z = _mm_min_epi8(z, sc_mch_);
+#if defined(__SSE4_1__) || defined(__AVX2__)
+				z = simd_func(max_epi8)(z, a);
+				z = simd_func(max_epi8)(z, b);
+				z = simd_func(max_epi8)(z, a2);
+				z = simd_func(max_epi8)(z, b2);
+				z = simd_func(min_epi8)(z, sc_mch_);
 				__dp_code_block2; // save u[] and v[]; update a, b, a2 and b2
-				_mm_store_si128(&x[t],  _mm_sub_epi8(_mm_max_epi8(a,  zero_), qe_));
-				_mm_store_si128(&y[t],  _mm_sub_epi8(_mm_max_epi8(b,  zero_), qe_));
-				_mm_store_si128(&x2[t], _mm_sub_epi8(_mm_max_epi8(a2, zero_), qe2_));
-				_mm_store_si128(&y2[t], _mm_sub_epi8(_mm_max_epi8(b2, zero_), qe2_));
-#else
+				simd_funcw(store)(&x[t],  simd_func(sub_epi8)(simd_func(max_epi8)(a,  zero_), qe_));
+				simd_funcw(store)(&y[t],  simd_func(sub_epi8)(simd_func(max_epi8)(b,  zero_), qe_));
+				simd_funcw(store)(&x2[t], simd_func(sub_epi8)(simd_func(max_epi8)(a2, zero_), qe2_));
+				simd_funcw(store)(&y2[t], simd_func(sub_epi8)(simd_func(max_epi8)(b2, zero_), qe2_));
+#elif defined(__SSE2__)
 				tmp = _mm_cmpgt_epi8(a,  z);
 				z = _mm_or_si128(_mm_andnot_si128(tmp, z), _mm_and_si128(tmp, a));
 				tmp = _mm_cmpgt_epi8(b,  z);
@@ -218,22 +244,22 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 #endif
 			}
 		} else if (!(flag&KSW_EZ_RIGHT)) { // gap left-alignment
-			__m128i *pr = p + (size_t)r * n_col_ - st_;
+			SIMD_INT *pr = p + (size_t)r * n_col_ - st_;
 			off[r] = st, off_end[r] = en;
 			for (t = st_; t <= en_; ++t) {
-				__m128i d, z, a, b, a2, b2, xt1, x2t1, vt1, ut, tmp;
+				SIMD_INT d, z, a, b, a2, b2, xt1, x2t1, vt1, ut, tmp;
 				__dp_code_block1;
-#ifdef __SSE4_1__
-				d = _mm_and_si128(_mm_cmpgt_epi8(a, z), _mm_set1_epi8(1));       // d = a  > z? 1 : 0
-				z = _mm_max_epi8(z, a);
-				d = _mm_blendv_epi8(d, _mm_set1_epi8(2), _mm_cmpgt_epi8(b,  z)); // d = b  > z? 2 : d
-				z = _mm_max_epi8(z, b);
-				d = _mm_blendv_epi8(d, _mm_set1_epi8(3), _mm_cmpgt_epi8(a2, z)); // d = a2 > z? 3 : d
-				z = _mm_max_epi8(z, a2);
-				d = _mm_blendv_epi8(d, _mm_set1_epi8(4), _mm_cmpgt_epi8(b2, z)); // d = a2 > z? 3 : d
-				z = _mm_max_epi8(z, b2);
-				z = _mm_min_epi8(z, sc_mch_);
-#else // we need to emulate SSE4.1 intrinsics _mm_max_epi8() and _mm_blendv_epi8()
+#if defined(__SSE4_1__) || defined(__AVX2__)
+				d = simd_funcw(and)(simd_func(cmpgt_epi8)(a, z), simd_func(set1_epi8)(1));       // d = a  > z? 1 : 0
+				z = simd_func(max_epi8)(z, a);
+				d = simd_func(blendv_epi8)(d, simd_func(set1_epi8)(2), simd_func(cmpgt_epi8)(b,  z)); // d = b  > z? 2 : d
+				z = simd_func(max_epi8)(z, b);
+				d = simd_func(blendv_epi8)(d, simd_func(set1_epi8)(3), simd_func(cmpgt_epi8)(a2, z)); // d = a2 > z? 3 : d
+				z = simd_func(max_epi8)(z, a2);
+				d = simd_func(blendv_epi8)(d, simd_func(set1_epi8)(4), simd_func(cmpgt_epi8)(b2, z)); // d = a2 > z? 3 : d
+				z = simd_func(max_epi8)(z, b2);
+				z = simd_func(min_epi8)(z, sc_mch_);
+#elif defined(__SSE2__) // emulate SSE4.1 intrinsics _mm_max_epi8() and _mm_blendv_epi8()
 				tmp = _mm_cmpgt_epi8(a,  z);
 				d = _mm_and_si128(tmp, _mm_set1_epi8(1));
 				z = _mm_or_si128(_mm_andnot_si128(tmp, z), _mm_and_si128(tmp, a));
@@ -250,37 +276,37 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 				z = _mm_or_si128(_mm_and_si128(tmp, sc_mch_), _mm_andnot_si128(tmp, z));
 #endif
 				__dp_code_block2;
-				tmp = _mm_cmpgt_epi8(a, zero_);
-				_mm_store_si128(&x[t],  _mm_sub_epi8(_mm_and_si128(tmp, a),  qe_));
-				d = _mm_or_si128(d, _mm_and_si128(tmp, _mm_set1_epi8(0x08))); // d = a > 0? 1<<3 : 0
-				tmp = _mm_cmpgt_epi8(b, zero_);
-				_mm_store_si128(&y[t],  _mm_sub_epi8(_mm_and_si128(tmp, b),  qe_));
-				d = _mm_or_si128(d, _mm_and_si128(tmp, _mm_set1_epi8(0x10))); // d = b > 0? 1<<4 : 0
-				tmp = _mm_cmpgt_epi8(a2, zero_);
-				_mm_store_si128(&x2[t], _mm_sub_epi8(_mm_and_si128(tmp, a2), qe2_));
-				d = _mm_or_si128(d, _mm_and_si128(tmp, _mm_set1_epi8(0x20))); // d = a > 0? 1<<5 : 0
-				tmp = _mm_cmpgt_epi8(b2, zero_);
-				_mm_store_si128(&y2[t], _mm_sub_epi8(_mm_and_si128(tmp, b2), qe2_));
-				d = _mm_or_si128(d, _mm_and_si128(tmp, _mm_set1_epi8(0x40))); // d = b > 0? 1<<6 : 0
-				_mm_store_si128(&pr[t], d);
+				tmp = simd_func(cmpgt_epi8)(a, zero_);
+				simd_funcw(store)(&x[t],  simd_func(sub_epi8)(simd_funcw(and)(tmp, a),  qe_));
+				d = simd_funcw(or)(d, simd_funcw(and)(tmp, simd_func(set1_epi8)(0x08))); // d = a > 0? 1<<3 : 0
+				tmp = simd_func(cmpgt_epi8)(b, zero_);
+				simd_funcw(store)(&y[t],  simd_func(sub_epi8)(simd_funcw(and)(tmp, b),  qe_));
+				d = simd_funcw(or)(d, simd_funcw(and)(tmp, simd_func(set1_epi8)(0x10))); // d = b > 0? 1<<4 : 0
+				tmp = simd_func(cmpgt_epi8)(a2, zero_);
+				simd_funcw(store)(&x2[t], simd_func(sub_epi8)(simd_funcw(and)(tmp, a2), qe2_));
+				d = simd_funcw(or)(d, simd_funcw(and)(tmp, simd_func(set1_epi8)(0x20))); // d = a > 0? 1<<5 : 0
+				tmp = simd_func(cmpgt_epi8)(b2, zero_);
+				simd_funcw(store)(&y2[t], simd_func(sub_epi8)(simd_funcw(and)(tmp, b2), qe2_));
+				d = simd_funcw(or)(d, simd_funcw(and)(tmp, simd_func(set1_epi8)(0x40))); // d = b > 0? 1<<6 : 0
+				simd_funcw(store)(&pr[t], d);
 			}
 		} else { // gap right-alignment
-			__m128i *pr = p + (size_t)r * n_col_ - st_;
+			SIMD_INT *pr = p + (size_t)r * n_col_ - st_;
 			off[r] = st, off_end[r] = en;
 			for (t = st_; t <= en_; ++t) {
-				__m128i d, z, a, b, a2, b2, xt1, x2t1, vt1, ut, tmp;
+				SIMD_INT d, z, a, b, a2, b2, xt1, x2t1, vt1, ut, tmp;
 				__dp_code_block1;
-#ifdef __SSE4_1__
-				d = _mm_andnot_si128(_mm_cmpgt_epi8(z, a), _mm_set1_epi8(1));    // d = z > a?  0 : 1
-				z = _mm_max_epi8(z, a);
-				d = _mm_blendv_epi8(_mm_set1_epi8(2), d, _mm_cmpgt_epi8(z, b));  // d = z > b?  d : 2
-				z = _mm_max_epi8(z, b);
-				d = _mm_blendv_epi8(_mm_set1_epi8(3), d, _mm_cmpgt_epi8(z, a2)); // d = z > a2? d : 3
-				z = _mm_max_epi8(z, a2);
-				d = _mm_blendv_epi8(_mm_set1_epi8(4), d, _mm_cmpgt_epi8(z, b2)); // d = z > b2? d : 4
-				z = _mm_max_epi8(z, b2);
-				z = _mm_min_epi8(z, sc_mch_);
-#else // we need to emulate SSE4.1 intrinsics _mm_max_epi8() and _mm_blendv_epi8()
+#if defined(__SSE4_1__) || defined(__AVX2__)
+				d = simd_funcw(andnot)(simd_func(cmpgt_epi8)(z, a), simd_func(set1_epi8)(1));    // d = z > a?  0 : 1
+				z = simd_func(max_epi8)(z, a);
+				d = simd_func(blendv_epi8)(simd_func(set1_epi8)(2), d, simd_func(cmpgt_epi8)(z, b));  // d = z > b?  d : 2
+				z = simd_func(max_epi8)(z, b);
+				d = simd_func(blendv_epi8)(simd_func(set1_epi8)(3), d, simd_func(cmpgt_epi8)(z, a2)); // d = z > a2? d : 3
+				z = simd_func(max_epi8)(z, a2);
+				d = simd_func(blendv_epi8)(simd_func(set1_epi8)(4), d, simd_func(cmpgt_epi8)(z, b2)); // d = z > b2? d : 4
+				z = simd_func(max_epi8)(z, b2);
+				z = simd_func(min_epi8)(z, sc_mch_);
+#elif defined(__SSE2__)
 				tmp = _mm_cmpgt_epi8(z, a);
 				d = _mm_andnot_si128(tmp, _mm_set1_epi8(1));
 				z = _mm_or_si128(_mm_and_si128(tmp, z), _mm_andnot_si128(tmp, a));
@@ -297,50 +323,54 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 				z = _mm_or_si128(_mm_and_si128(tmp, sc_mch_), _mm_andnot_si128(tmp, z));
 #endif
 				__dp_code_block2;
-				tmp = _mm_cmpgt_epi8(zero_, a);
-				_mm_store_si128(&x[t],  _mm_sub_epi8(_mm_andnot_si128(tmp, a),  qe_));
-				d = _mm_or_si128(d, _mm_andnot_si128(tmp, _mm_set1_epi8(0x08))); // d = a > 0? 1<<3 : 0
-				tmp = _mm_cmpgt_epi8(zero_, b);
-				_mm_store_si128(&y[t],  _mm_sub_epi8(_mm_andnot_si128(tmp, b),  qe_));
-				d = _mm_or_si128(d, _mm_andnot_si128(tmp, _mm_set1_epi8(0x10))); // d = b > 0? 1<<4 : 0
-				tmp = _mm_cmpgt_epi8(zero_, a2);
-				_mm_store_si128(&x2[t], _mm_sub_epi8(_mm_andnot_si128(tmp, a2), qe2_));
-				d = _mm_or_si128(d, _mm_andnot_si128(tmp, _mm_set1_epi8(0x20))); // d = a > 0? 1<<5 : 0
-				tmp = _mm_cmpgt_epi8(zero_, b2);
-				_mm_store_si128(&y2[t], _mm_sub_epi8(_mm_andnot_si128(tmp, b2), qe2_));
-				d = _mm_or_si128(d, _mm_andnot_si128(tmp, _mm_set1_epi8(0x40))); // d = b > 0? 1<<6 : 0
-				_mm_store_si128(&pr[t], d);
+				tmp = simd_func(cmpgt_epi8)(zero_, a);
+				simd_funcw(store)(&x[t],  simd_func(sub_epi8)(simd_funcw(andnot)(tmp, a),  qe_));
+				d = simd_funcw(or)(d, simd_funcw(andnot)(tmp, simd_func(set1_epi8)(0x08))); // d = a > 0? 1<<3 : 0
+				tmp = simd_func(cmpgt_epi8)(zero_, b);
+				simd_funcw(store)(&y[t],  simd_func(sub_epi8)(simd_funcw(andnot)(tmp, b),  qe_));
+				d = simd_funcw(or)(d, simd_funcw(andnot)(tmp, simd_func(set1_epi8)(0x10))); // d = b > 0? 1<<4 : 0
+				tmp = simd_func(cmpgt_epi8)(zero_, a2);
+				simd_funcw(store)(&x2[t], simd_func(sub_epi8)(simd_funcw(andnot)(tmp, a2), qe2_));
+				d = simd_funcw(or)(d, simd_funcw(andnot)(tmp, simd_func(set1_epi8)(0x20))); // d = a > 0? 1<<5 : 0
+				tmp = simd_func(cmpgt_epi8)(zero_, b2);
+				simd_funcw(store)(&y2[t], simd_func(sub_epi8)(simd_funcw(andnot)(tmp, b2), qe2_));
+				d = simd_funcw(or)(d, simd_funcw(andnot)(tmp, simd_func(set1_epi8)(0x40))); // d = b > 0? 1<<6 : 0
+				simd_funcw(store)(&pr[t], d);
 			}
 		}
 		if (!approx_max) { // find the exact max with a 32-bit score array
 			int32_t max_H, max_t;
 			// compute H[], max_H and max_t
 			if (r > 0) {
-				int32_t HH[4], tt[4], en1 = st0 + (en0 - st0) / 4 * 4, i;
-				__m128i max_H_, max_t_;
+				int32_t HH[SIMD_WIDTH/4], tt[SIMD_WIDTH/4], en1 = st0 + (en0 - st0) / (SIMD_WIDTH/4) * (SIMD_WIDTH/4), i;
+				SIMD_INT max_H_, max_t_;
 				max_H = H[en0] = en0 > 0? H[en0-1] + u8[en0] : H[en0] + v8[en0]; // special casing the last element
 				max_t = en0;
-				max_H_ = _mm_set1_epi32(max_H);
-				max_t_ = _mm_set1_epi32(max_t);
-				for (t = st0; t < en1; t += 4) { // this implements: H[t]+=v8[t]-qe; if(H[t]>max_H) max_H=H[t],max_t=t;
-					__m128i H1, tmp, t_;
-					H1 = _mm_loadu_si128((__m128i*)&H[t]);
+				max_H_ = simd_func(set1_epi32)(max_H);
+				max_t_ = simd_func(set1_epi32)(max_t);
+				for (t = st0; t < en1; t += SIMD_WIDTH/4) { // this implements: H[t]+=v8[t]-qe; if(H[t]>max_H) max_H=H[t],max_t=t;
+					SIMD_INT H1, tmp, t_;
+					H1 = simd_funcw(loadu)((SIMD_INT*)&H[t]);
+#if defined(__AVX2__)
+					t_ = _mm256_setr_epi32(v8[t], v8[t+1], v8[t+2], v8[t+3], v8[t+4], v8[t+5], v8[t+6], v8[t+7]);
+#elif defined(__SSE2__)
 					t_ = _mm_setr_epi32(v8[t], v8[t+1], v8[t+2], v8[t+3]);
-					H1 = _mm_add_epi32(H1, t_);
-					_mm_storeu_si128((__m128i*)&H[t], H1);
-					t_ = _mm_set1_epi32(t);
-					tmp = _mm_cmpgt_epi32(H1, max_H_);
-#ifdef __SSE4_1__
-					max_H_ = _mm_blendv_epi8(max_H_, H1, tmp);
-					max_t_ = _mm_blendv_epi8(max_t_, t_, tmp);
-#else
-					max_H_ = _mm_or_si128(_mm_and_si128(tmp, H1), _mm_andnot_si128(tmp, max_H_));
-					max_t_ = _mm_or_si128(_mm_and_si128(tmp, t_), _mm_andnot_si128(tmp, max_t_));
+#endif
+					H1 = simd_func(add_epi32)(H1, t_);
+					simd_funcw(storeu)((SIMD_INT*)&H[t], H1);
+					t_ = simd_func(set1_epi32)(t);
+					tmp = simd_func(cmpgt_epi32)(H1, max_H_);
+#if defined(__SSE4_1__) || defined(__AVX2__)
+					max_H_ = simd_func(blendv_epi8)(max_H_, H1, tmp);
+					max_t_ = simd_func(blendv_epi8)(max_t_, t_, tmp);
+#elif defined(__SSE2__)
+					max_H_ = simd_funcw(or)(simd_funcw(and)(tmp, H1), simd_funcw(andnot)(tmp, max_H_));
+					max_t_ = simd_funcw(or)(simd_funcw(and)(tmp, t_), simd_funcw(andnot)(tmp, max_t_));
 #endif
 				}
-				_mm_storeu_si128((__m128i*)HH, max_H_);
-				_mm_storeu_si128((__m128i*)tt, max_t_);
-				for (i = 0; i < 4; ++i)
+				simd_funcw(storeu)((SIMD_INT*)HH, max_H_);
+				simd_funcw(storeu)((SIMD_INT*)tt, max_t_);
+				for (i = 0; i < SIMD_WIDTH/4; ++i)
 					if (max_H < HH[i]) max_H = HH[i], max_t = tt[i] + i;
 				for (; t < en0; ++t) { // for the rest of values that haven't been computed with SSE
 					H[t] += (int32_t)v8[t];
@@ -381,12 +411,12 @@ void ksw_extd2_sse(void *km, int qlen, const uint8_t *query, int tlen, const uin
 	if (with_cigar) { // backtrack
 		int rev_cigar = !!(flag & KSW_EZ_REV_CIGAR);
 		if (!ez->zdropped && !(flag&KSW_EZ_EXTZ_ONLY)) {
-			ksw_backtrack(km, 1, rev_cigar, 0, (uint8_t*)p, off, off_end, n_col_*16, tlen-1, qlen-1, &ez->m_cigar, &ez->n_cigar, &ez->cigar);
+			ksw_backtrack(km, 1, rev_cigar, 0, (uint8_t*)p, off, off_end, n_col_*SIMD_WIDTH, tlen-1, qlen-1, &ez->m_cigar, &ez->n_cigar, &ez->cigar);
 		} else if (!ez->zdropped && (flag&KSW_EZ_EXTZ_ONLY) && ez->mqe + end_bonus > (int)ez->max) {
 			ez->reach_end = 1;
-			ksw_backtrack(km, 1, rev_cigar, 0, (uint8_t*)p, off, off_end, n_col_*16, ez->mqe_t, qlen-1, &ez->m_cigar, &ez->n_cigar, &ez->cigar);
+			ksw_backtrack(km, 1, rev_cigar, 0, (uint8_t*)p, off, off_end, n_col_*SIMD_WIDTH, ez->mqe_t, qlen-1, &ez->m_cigar, &ez->n_cigar, &ez->cigar);
 		} else if (ez->max_t >= 0 && ez->max_q >= 0) {
-			ksw_backtrack(km, 1, rev_cigar, 0, (uint8_t*)p, off, off_end, n_col_*16, ez->max_t, ez->max_q, &ez->m_cigar, &ez->n_cigar, &ez->cigar);
+			ksw_backtrack(km, 1, rev_cigar, 0, (uint8_t*)p, off, off_end, n_col_*SIMD_WIDTH, ez->max_t, ez->max_q, &ez->m_cigar, &ez->n_cigar, &ez->cigar);
 		}
 		kfree(km, mem2); kfree(km, off);
 	}