diff --git a/src/liblzma/lz/lz_encoder.c b/src/liblzma/lz/lz_encoder.c
index 76954e4d..2033844f 100644
--- a/src/liblzma/lz/lz_encoder.c
+++ b/src/liblzma/lz/lz_encoder.c
@@ -20,6 +20,8 @@
 #	include "lz_encoder_hash_table.h"
 #endif
 
+#include "memcmplen.h"
+
 
 struct lzma_coder_s {
 	/// LZ-based encoder e.g. LZMA
@@ -363,9 +365,18 @@ lz_encoder_init(lzma_mf *mf, const lzma_allocator *allocator,
 {
 	// Allocate the history buffer.
 	if (mf->buffer == NULL) {
-		mf->buffer = lzma_alloc(mf->size, allocator);
+		// lzma_memcmplen() is used for the dictionary buffer
+		// so we need to allocate a few extra bytes to prevent
+		// it from reading past the end of the buffer.
+		mf->buffer = lzma_alloc(mf->size + LZMA_MEMCMPLEN_EXTRA,
+				allocator);
 		if (mf->buffer == NULL)
 			return true;
+
+		// Keep Valgrind happy with lzma_memcmplen() and initialize
+		// the extra bytes whose value may get read but which will
+		// effectively get ignored.
+		memzero(mf->buffer + mf->size, LZMA_MEMCMPLEN_EXTRA);
 	}
 
 	// Use cyclic_size as initial mf->offset. This allows
diff --git a/src/liblzma/lz/lz_encoder_mf.c b/src/liblzma/lz/lz_encoder_mf.c
index bf787f45..6dc39f4a 100644
--- a/src/liblzma/lz/lz_encoder_mf.c
+++ b/src/liblzma/lz/lz_encoder_mf.c
@@ -13,6 +13,7 @@
 
 #include "lz_encoder.h"
 #include "lz_encoder_hash.h"
+#include "memcmplen.h"
 
 
 /// \brief      Find matches starting from the current byte
@@ -65,9 +66,7 @@ lzma_mf_find(lzma_mf *mf, uint32_t *count_ptr, lzma_match *matches)
 			// here because the match distances are zero based.
 			const uint8_t *p2 = p1 - matches[count - 1].dist - 1;
 
-			while (len_best < limit
-					&& p1[len_best] == p2[len_best])
-				++len_best;
+			len_best = lzma_memcmplen(p1, p2, len_best, limit);
 		}
 	}
 
@@ -272,10 +271,7 @@ hc_find_func(
 				+ (delta > cyclic_pos ? cyclic_size : 0)];
 
 		if (pb[len_best] == cur[len_best] && pb[0] == cur[0]) {
-			uint32_t len = 0;
-			while (++len != len_limit)
-				if (pb[len] != cur[len])
-					break;
+			uint32_t len = lzma_memcmplen(pb, cur, 1, len_limit);
 
 			if (len_best < len) {
 				len_best = len;
@@ -321,9 +317,8 @@ lzma_mf_hc3_find(lzma_mf *mf, lzma_match *matches)
 	uint32_t len_best = 2;
 
 	if (delta2 < mf->cyclic_size && *(cur - delta2) == *cur) {
-		for ( ; len_best != len_limit; ++len_best)
-			if (*(cur + len_best - delta2) != cur[len_best])
-				break;
+		len_best = lzma_memcmplen(cur - delta2, cur,
+				len_best, len_limit);
 
 		matches[0].len = len_best;
 		matches[0].dist = delta2 - 1;
@@ -400,9 +395,8 @@ lzma_mf_hc4_find(lzma_mf *mf, lzma_match *matches)
 	}
 
 	if (matches_count != 0) {
-		for ( ; len_best != len_limit; ++len_best)
-			if (*(cur + len_best - delta2) != cur[len_best])
-				break;
+		len_best = lzma_memcmplen(cur - delta2, cur,
+				len_best, len_limit);
 
 		matches[matches_count - 1].len = len_best;
 
@@ -487,9 +481,7 @@ bt_find_func(
 		uint32_t len = my_min(len0, len1);
 
 		if (pb[len] == cur[len]) {
-			while (++len != len_limit)
-				if (pb[len] != cur[len])
-					break;
+			len = lzma_memcmplen(pb, cur, len + 1, len_limit);
 
 			if (len_best < len) {
 				len_best = len;
@@ -552,9 +544,7 @@ bt_skip_func(
 		uint32_t len = my_min(len0, len1);
 
 		if (pb[len] == cur[len]) {
-			while (++len != len_limit)
-				if (pb[len] != cur[len])
-					break;
+			len = lzma_memcmplen(pb, cur, len + 1, len_limit);
 
 			if (len == len_limit) {
 				*ptr1 = pair[0];
@@ -715,9 +705,8 @@ lzma_mf_bt4_find(lzma_mf *mf, lzma_match *matches)
 	}
 
 	if (matches_count != 0) {
-		for ( ; len_best != len_limit; ++len_best)
-			if (*(cur + len_best - delta2) != cur[len_best])
-				break;
+		len_best = lzma_memcmplen(
+				cur, cur - delta2, len_best, len_limit);
 
 		matches[matches_count - 1].len = len_best;