diff --git a/Makefile b/Makefile
index de45ff1..98d0eb6 100644
--- a/Makefile
+++ b/Makefile
@@ -28,14 +28,19 @@ bwa:libbwa.a $(AOBJS) main.o
 libbwa.a:$(LOBJS)
 		$(AR) -csru $@ $(LOBJS)
 
-QSufSort.o:QSufSort.h
-bwt_gen.o:QSufSort.h
-
 ksw.o:ksw.h
+kstring.o:kstring.h
 utils.o:utils.h ksort.h kseq.h
 bntseq.o:bntseq.h
 bwt.o:bwt.h utils.h
-bwa.o:bwa.h
+bwa.o:bwa.h bwt.h bntseq.h
+bwamem.o:ksw.h kbtree.h ksort.h kvec.h kstring.h utils.h bwamem.h
+bwamem_pair.o:ksw.h kvec.h kstring.h utils.h bwamem.h
+
+QSufSort.o:QSufSort.h
+bwt_gen.o:QSufSort.h
+
+fastmap.o:bwt.h bwamem.h
 
 bwtaln.o:bwt.h bwtaln.h kseq.h
 bwtgap.o:bwtgap.h bwtaln.h bwt.h
@@ -44,9 +49,5 @@ bwtsw2_core.o:bwtsw2.h bwt.h bwt_lite.h stdaln.h
 bwtsw2_aux.o:bwtsw2.h bwt.h bwt_lite.h stdaln.h
 bwtsw2_main.o:bwtsw2.h
 
-bwamem.o:bwamem.h
-bwamem_pair.o:bwamem.h
-fastmap.o:bwt.h bwamem.h
-
 clean:
 		rm -f gmon.out *.o a.out $(PROG) *~ *.a
diff --git a/ksw.h b/ksw.h
index 40216d9..d2975de 100644
--- a/ksw.h
+++ b/ksw.h
@@ -30,7 +30,7 @@ extern "C" {
 	 * @param tlen    length of the target sequence
 	 * @param target  target sequence
 	 * @param m       number of residue types
-	 * @param mat     m*m scoring matrix in one-dimention array
+	 * @param mat     m*m scoring matrix in one-dimension array
 	 * @param gapo    gap open penalty; a gap of length l cost "-(gapo+l*gape)"
 	 * @param gape    gap extension penalty
 	 * @param xtra    extra information (see below)
@@ -62,8 +62,47 @@ extern "C" {
 	 */
 	kswr_t ksw_align(int qlen, uint8_t *query, int tlen, uint8_t *target, int m, const int8_t *mat, int gapo, int gape, int xtra, kswq_t **qry);
 
-	int ksw_extend(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int m, const int8_t *mat, int gapo, int gape, int w, int h0, int *_qle, int *_tle, int *_gtle, int *_gscore);
-	int ksw_global(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int m, const int8_t *mat, int gapo, int gape, int w, int *_n_cigar, uint32_t **_cigar);
+	/**
+	 * Banded global alignment
+	 *
+	 * @param qlen    query length
+	 * @param query   query sequence with 0 <= query[i] < m
+	 * @param tlen    target length
+	 * @param target  target sequence with 0 <= target[i] < m
+	 * @param m       number of residue types
+	 * @param mat     m*m scoring mattrix in one-dimension array
+	 * @param gapo    gap open penalty; a gap of length l cost "-(gapo+l*gape)"
+	 * @param gape    gap extension penalty
+	 * @param w       band width
+	 * @param n_cigar (out) number of CIGAR elements
+	 * @param cigar   (out) BAM-encoded CIGAR; caller need to deallocate with free()
+	 *
+	 * @return        score of the alignment
+	 */
+	int ksw_global(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int m, const int8_t *mat, int gapo, int gape, int w, int *n_cigar, uint32_t **cigar);
+
+	/**
+	 * Extend alignment
+	 *
+	 * The routine aligns $query and $target, assuming their upstream sequences,
+	 * which are not provided, have been aligned with score $h0. In return,
+	 * region [0,*qle) on the query and [0,*tle) on the target sequences are
+	 * aligned together. If *gscore>=0, *gscore keeps the best score such that
+	 * the entire query sequence is aligned; *gtle keeps the position on the
+	 * target where *gscore is achieved. Returning *gscore and *gtle helps the
+	 * caller to decide whether an end-to-end hit or a partial hit is preferred.
+	 *
+	 * The first 9 parameters are identical to those in ksw_global()
+	 *
+	 * @param h0      alignment score of upstream sequences
+	 * @param _qle    (out) length of the query in the alignment
+	 * @param _tle    (out) length of the target in the alignment
+	 * @param _gtle   (out) length of the target if query is fully aligned
+	 * @param _gscore (out) score of the best end-to-end alignment; negative if not found
+	 *
+	 * @return        best semi-local alignment score
+	 */
+	int ksw_extend(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int m, const int8_t *mat, int gapo, int gape, int w, int h0, int *qle, int *tle, int *gtle, int *gscore);
 
 #ifdef __cplusplus
 }