diff --git a/src/hwlm/noodle_engine_simd.hpp b/src/hwlm/noodle_engine_simd.hpp
index dfe7eea1..c49bfc7e 100644
--- a/src/hwlm/noodle_engine_simd.hpp
+++ b/src/hwlm/noodle_engine_simd.hpp
@@ -58,12 +58,10 @@ hwlm_error_t double_zscan(const struct noodTable *n,const u8 *d, const u8 *buf,
     return HWLM_SUCCESS;
 }
 
-// The short scan routine. It is used both to scan data up to an
-// alignment boundary if needed and to finish off data that the aligned scan
-// function can't handle (due to small/unaligned chunk at end)
+
 template<uint16_t S>
 static really_inline
-hwlm_error_t scanSingleUnaligned(const struct noodTable *n, const u8 *buf,
+hwlm_error_t scanSingleShort(const struct noodTable *n, const u8 *buf,
                                  SuperVector<S> caseMask, SuperVector<S> mask1,
                                  const struct cb_info *cbi, size_t len, size_t start,
                                  size_t end) {
@@ -76,7 +74,36 @@ hwlm_error_t scanSingleUnaligned(const struct noodTable *n, const u8 *buf,
         return HWLM_SUCCESS;
     }
 
+    SuperVector<S> v = SuperVector<S>::Zeroes();
+    memcpy(&v.u, d, l);
+
     typename SuperVector<S>::movemask_type mask = SINGLE_LOAD_MASK(l);
+    v = v & caseMask;
+    typename SuperVector<S>::movemask_type z = mask & mask1.eqmask(v);
+
+    return single_zscan(n, d, buf, z, len, cbi);
+}
+
+// The short scan routine. It is used both to scan data up to an
+// alignment boundary if needed and to finish off data that the aligned scan
+// function can't handle (due to small/unaligned chunk at end)
+template<uint16_t S>
+static really_inline
+hwlm_error_t scanSingleUnaligned(const struct noodTable *n, const u8 *buf,
+                                 SuperVector<S> caseMask, SuperVector<S> mask1,
+                                 const struct cb_info *cbi, size_t len, size_t offset,
+                                     size_t start,
+                                 size_t end) {
+    const u8 *d = buf + offset;
+    DEBUG_PRINTF("start %zu end %zu offset %zu\n", start, end, offset);
+    const size_t l = end - start;
+    DEBUG_PRINTF("l = %ld\n", l);
+    assert(l <= 64);
+    if (!l) {
+        return HWLM_SUCCESS;
+    }
+    size_t buf_off = start - offset;
+    typename SuperVector<S>::movemask_type mask = SINGLE_LOAD_MASK(l) << buf_off;
     SuperVector<S> v = SuperVector<S>::loadu(d) & caseMask;
     typename SuperVector<S>::movemask_type z = mask & mask1.eqmask(v);
 
@@ -85,8 +112,8 @@ hwlm_error_t scanSingleUnaligned(const struct noodTable *n, const u8 *buf,
 
 template<uint16_t S>
 static really_inline
-hwlm_error_t scanDoubleUnaligned(const struct noodTable *n, const u8 *buf,
-                                 SuperVector<S> caseMask, SuperVector<S> mask1, SuperVector<S> mask2, typename SuperVector<S>::movemask_type *lastz1,
+hwlm_error_t scanDoubleShort(const struct noodTable *n, const u8 *buf,
+                                 SuperVector<S> caseMask, SuperVector<S> mask1, SuperVector<S> mask2,
                                  const struct cb_info *cbi, size_t len, size_t start, size_t end) {
     const u8 *d = buf + start;
     DEBUG_PRINTF("start %zu end %zu\n", start, end);
@@ -95,13 +122,36 @@ hwlm_error_t scanDoubleUnaligned(const struct noodTable *n, const u8 *buf,
     if (!l) {
         return HWLM_SUCCESS;
     }
-    SuperVector<S> v = SuperVector<S>::loadu(d) & caseMask;
+    SuperVector<S> v = SuperVector<S>::Zeroes();
+    memcpy(&v.u, d, l);
+    v = v & caseMask;
 
     typename SuperVector<S>::movemask_type mask = DOUBLE_LOAD_MASK(l);
     typename SuperVector<S>::movemask_type z1 = mask1.eqmask(v);
     typename SuperVector<S>::movemask_type z2 = mask2.eqmask(v);
-    typename SuperVector<S>::movemask_type z = mask & (*lastz1 | z1 << 1) & z2;
-    *lastz1 = z1 >> (l -1);
+    typename SuperVector<S>::movemask_type z = mask & (z1 << 1) & z2;
+
+    return double_zscan(n, d, buf, z, len, cbi);
+}
+
+template<uint16_t S>
+static really_inline
+hwlm_error_t scanDoubleUnaligned(const struct noodTable *n, const u8 *buf,
+                                 SuperVector<S> caseMask, SuperVector<S> mask1, SuperVector<S> mask2,
+                                 const struct cb_info *cbi, size_t len, size_t offset, size_t start, size_t end) {
+    const u8 *d = buf + offset;
+    DEBUG_PRINTF("start %zu end %zu offset %zu\n", start, end, offset);
+    const size_t l = end - start;
+    assert(l <= S);
+    if (!l) {
+        return HWLM_SUCCESS;
+    }
+    SuperVector<S> v = SuperVector<S>::loadu(d) & caseMask;
+    size_t buf_off = start - offset;
+    typename SuperVector<S>::movemask_type mask = DOUBLE_LOAD_MASK(l) << buf_off;
+    typename SuperVector<S>::movemask_type z1 = mask1.eqmask(v);
+    typename SuperVector<S>::movemask_type z2 = mask2.eqmask(v);
+    typename SuperVector<S>::movemask_type z = mask & (z1 << 1) & z2;
 
     return double_zscan(n, d, buf, z, len, cbi);
 }
@@ -119,11 +169,14 @@ hwlm_error_t scanSingleMain(const struct noodTable *n, const u8 *buf,
     const u8 *e = buf + end;
     DEBUG_PRINTF("start %p end %p \n", d, e);
     assert(d < e);
+    if (e - d < S) {
+      return scanSingleShort(n, buf, caseMask, mask1, cbi, len, start, end);
+    }
     if (d + S <= e) {
         // peel off first part to cacheline boundary
         const u8 *d1 = ROUNDUP_PTR(d, S);
         DEBUG_PRINTF("until aligned %p \n", d1);
-        if (scanSingleUnaligned(n, buf, caseMask, mask1, cbi, len, start, d1 - buf) == HWLM_TERMINATED) {
+        if (scanSingleUnaligned(n, buf, caseMask, mask1, cbi, len, start, start, d1 - buf) == HWLM_TERMINATED) {
             return HWLM_TERMINATED;
         }
         d = d1;
@@ -147,8 +200,12 @@ hwlm_error_t scanSingleMain(const struct noodTable *n, const u8 *buf,
 
     DEBUG_PRINTF("d %p e %p \n", d, e);
     // finish off tail
+    size_t s2End = ROUNDDOWN_PTR(e, S) - buf;
+    if (s2End == end) {
+      return HWLM_SUCCESS;
+    }
 
-    return scanSingleUnaligned(n, buf, caseMask, mask1, cbi, len, d - buf, end);
+    return scanSingleUnaligned(n, buf, caseMask, mask1, cbi, len, end - S, s2End, len);
 }
 
 template <uint16_t S>
@@ -169,14 +226,17 @@ hwlm_error_t scanDoubleMain(const struct noodTable *n, const u8 *buf,
     const u8 *e = buf + end;
     DEBUG_PRINTF("start %p end %p \n", d, e);
     assert(d < e);
+    if (e - d < S) {
+      return scanDoubleShort(n, buf, caseMask, mask1, mask2, cbi, len, d - buf, end);
+    }
     if (d + S <= e) {
         // peel off first part to cacheline boundary
-        const u8 *d1 = ROUNDUP_PTR(d, S);
+        const u8 *d1 = ROUNDUP_PTR(d, S) + 1;
         DEBUG_PRINTF("until aligned %p \n", d1);
-        if (scanDoubleUnaligned(n, buf, caseMask, mask1, mask2, &lastz1, cbi, len, start, d1 - buf) == HWLM_TERMINATED) {
+        if (scanDoubleUnaligned(n, buf, caseMask, mask1, mask2, cbi, len, start, start, d1 - buf) == HWLM_TERMINATED) {
             return HWLM_TERMINATED;
         }
-        d = d1;
+        d = d1 - 1;
 
         size_t loops = (end - (d - buf)) / S;
         DEBUG_PRINTF("loops %ld \n", loops);
@@ -196,12 +256,16 @@ hwlm_error_t scanDoubleMain(const struct noodTable *n, const u8 *buf,
             hwlm_error_t rv = double_zscan(n, d, buf, z, len, cbi);
             RETURN_IF_TERMINATED(rv);
         }
+        if (loops == 0) {
+          d = d1;
+        }
     }
-
-    DEBUG_PRINTF("d %p e %p \n", d, e);
     // finish off tail
-
-    return scanDoubleUnaligned(n, buf, caseMask, mask1, mask2, &lastz1, cbi, len, d - buf, end);
+    size_t s2End = ROUNDDOWN_PTR(e, S) - buf;
+    if (s2End == end) {
+      return HWLM_SUCCESS;
+    }
+    return scanDoubleUnaligned(n, buf, caseMask, mask1, mask2, cbi, len, end - S, d - buf, end);
 }
 
 // Single-character specialisation, used when keyLen = 1
diff --git a/src/nfa/arm/vermicelli.hpp b/src/nfa/arm/vermicelli.hpp
index d790fa1f..496468e0 100644
--- a/src/nfa/arm/vermicelli.hpp
+++ b/src/nfa/arm/vermicelli.hpp
@@ -67,7 +67,7 @@ const u8 *rvermicelliBlockNeg(SuperVector<S> const data, SuperVector<S> const ch
     return last_zero_match_inverted<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                 u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len) {
@@ -78,14 +78,16 @@ const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     SuperVector<S> mask = mask1 & (mask2 >> 1);
 
     DEBUG_PRINTF("rv[0] = %02hhx, rv[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
+    bool partial_match = (check_partial && ((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
-    if (partial_match) return buf - 1;
+    if (partial_match) {
+        mask = mask | ((SuperVector<S>::Ones() >> (S-1)) << (S-1));
+    }
 
     return first_non_zero_match<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                  u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len) {
@@ -96,7 +98,7 @@ const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     SuperVector<S> mask = (mask1 << 1)& mask2;
 
     DEBUG_PRINTF("buf[0] = %02hhx, buf[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
+    bool partial_match = (check_partial && ((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
     if (partial_match) {
         mask = mask | (SuperVector<S>::Ones() >> (S-1));
@@ -105,7 +107,7 @@ const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     return last_non_zero_match<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2,
                                       SuperVector<S> const mask1, SuperVector<S> const mask2,
@@ -116,9 +118,11 @@ const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S>
     SuperVector<S> mask = v1 & (v2 >> 1);
 
     DEBUG_PRINTF("rv[0] = %02hhx, rv[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & m1) == c2) && ((buf[-1] & m2) == c1));
+    bool partial_match = (check_partial && ((buf[0] & m2) == c2) && ((buf[-1] & m1) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
-    if (partial_match) return buf - 1;
+    if (partial_match) {
+        mask = mask | ((SuperVector<S>::Ones() >> (S-1)) << (S-1));
+    }
 
     return first_non_zero_match<S>(buf, mask, len);
 }
diff --git a/src/nfa/ppc64el/vermicelli.hpp b/src/nfa/ppc64el/vermicelli.hpp
index eeaad6a1..1f3de25f 100644
--- a/src/nfa/ppc64el/vermicelli.hpp
+++ b/src/nfa/ppc64el/vermicelli.hpp
@@ -67,7 +67,7 @@ const u8 *rvermicelliBlockNeg(SuperVector<S> const data, SuperVector<S> const ch
     return last_zero_match_inverted<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                 u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len) {
@@ -78,14 +78,16 @@ const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     SuperVector<S> mask = mask1 & (mask2 >> 1);
 
     DEBUG_PRINTF("rv[0] = %02hhx, rv[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
+    bool partial_match = (check_partial && ((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
-    if (partial_match) return buf - 1;
+    if (partial_match) {
+        mask = mask | ((SuperVector<S>::Ones() >> (S-1)) << (S-1));
+    }
 
     return first_non_zero_match<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                  u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len) {
@@ -96,7 +98,7 @@ const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     SuperVector<S> mask = (mask1 << 1)& mask2;
 
     DEBUG_PRINTF("buf[0] = %02hhx, buf[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
+    bool partial_match = (check_partial && ((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
     if (partial_match) {
         mask = mask | (SuperVector<S>::Ones() >> (S-1));
@@ -105,7 +107,7 @@ const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     return last_non_zero_match<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2,
                                       SuperVector<S> const mask1, SuperVector<S> const mask2,
@@ -116,9 +118,11 @@ const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S>
     SuperVector<S> mask = v1 & (v2 >> 1);
 
     DEBUG_PRINTF("rv[0] = %02hhx, rv[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & m1) == c2) && ((buf[-1] & m2) == c1));
+    bool partial_match = (check_partial && ((buf[0] & m2) == c2) && ((buf[-1] & m1) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
-    if (partial_match) return buf - 1;
+    if (partial_match) {
+        mask = mask | ((SuperVector<S>::Ones() >> (S-1)) << (S-1));
+    }
 
     return first_non_zero_match<S>(buf, mask, len);
 }
diff --git a/src/nfa/shufti_simd.hpp b/src/nfa/shufti_simd.hpp
index 887f2468..0f8e2a7b 100644
--- a/src/nfa/shufti_simd.hpp
+++ b/src/nfa/shufti_simd.hpp
@@ -128,8 +128,8 @@ const u8 *shuftiExecReal(m128 mask_lo, m128 mask_hi, const u8 *buf, const u8 *bu
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> chars = SuperVector<S>::loadu_maskz(d, buf_end - d);
-        rv = fwdBlock(wide_mask_lo, wide_mask_hi, chars, d);
+        SuperVector<S> chars = SuperVector<S>::loadu(buf_end - S);
+        rv = fwdBlock(wide_mask_lo, wide_mask_hi, chars, buf_end - S);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -240,22 +240,36 @@ const u8 *shuftiDoubleExecReal(m128 mask1_lo, m128 mask1_hi, m128 mask2_lo, m128
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> chars = SuperVector<S>::loadu(d);
-        rv = fwdBlockDouble(wide_mask1_lo, wide_mask1_hi, wide_mask2_lo, wide_mask2_hi, chars, d);
+        SuperVector<S> chars = SuperVector<S>::Zeroes();
+        const u8 *end_buf;
+        if (buf_end - buf < S) {
+          memcpy(&chars.u, buf, buf_end - buf);
+          end_buf = buf;
+        } else {
+          chars = SuperVector<S>::loadu(buf_end - S);
+          end_buf = buf_end - S;
+        }
+        rv = fwdBlockDouble(wide_mask1_lo, wide_mask1_hi, wide_mask2_lo, wide_mask2_hi, chars, end_buf);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
-    
+
     return buf_end;
 }
 
 const u8 *shuftiExec(m128 mask_lo, m128 mask_hi, const u8 *buf,
                       const u8 *buf_end) {
-    return shuftiExecReal<VECTORSIZE>(mask_lo, mask_hi, buf, buf_end);
+  if (buf_end - buf < VECTORSIZE) {
+    return shuftiFwdSlow((const u8 *)&mask_lo, (const u8 *)&mask_hi, buf, buf_end);
+  }
+  return shuftiExecReal<VECTORSIZE>(mask_lo, mask_hi, buf, buf_end);
 }
 
 const u8 *rshuftiExec(m128 mask_lo, m128 mask_hi, const u8 *buf,
                        const u8 *buf_end) {
+    if (buf_end - buf < VECTORSIZE) {
+      return shuftiRevSlow((const u8 *)&mask_lo, (const u8 *)&mask_hi, buf, buf_end);
+    }
     return rshuftiExecReal<VECTORSIZE>(mask_lo, mask_hi, buf, buf_end);
 }
 
diff --git a/src/nfa/truffle_simd.hpp b/src/nfa/truffle_simd.hpp
index 8d9911fd..e07e92f6 100644
--- a/src/nfa/truffle_simd.hpp
+++ b/src/nfa/truffle_simd.hpp
@@ -107,8 +107,16 @@ const u8 *truffleExecReal(m128 &shuf_mask_lo_highclear, m128 shuf_mask_lo_highse
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> chars = SuperVector<S>::loadu_maskz(d, buf_end - d);
-        rv = fwdBlock(wide_shuf_mask_lo_highclear, wide_shuf_mask_lo_highset, chars, d);
+        SuperVector<S> chars = SuperVector<S>::Zeroes();
+        const u8* end_buf;
+        if (buf_end - buf < S) {
+          memcpy(&chars.u, buf, buf_end - buf);
+          end_buf = buf;
+        } else {
+          chars = SuperVector<S>::loadu(buf_end - S);
+          end_buf = buf_end - S;
+        }
+        rv = fwdBlock(wide_shuf_mask_lo_highclear, wide_shuf_mask_lo_highset, chars, end_buf);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -171,7 +179,12 @@ const u8 *rtruffleExecReal(m128 shuf_mask_lo_highclear, m128 shuf_mask_lo_highse
     // finish off head
 
     if (d != buf) {
-        SuperVector<S> chars = SuperVector<S>::loadu(buf);
+        SuperVector<S> chars = SuperVector<S>::Zeroes();
+        if (buf_end - buf < S) {
+          memcpy(&chars.u, buf, buf_end - buf);
+        } else {
+          chars = SuperVector<S>::loadu(buf);
+        }
         rv = revBlock(wide_shuf_mask_lo_highclear, wide_shuf_mask_lo_highset, chars, buf);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
diff --git a/src/nfa/vermicelli_simd.cpp b/src/nfa/vermicelli_simd.cpp
index d790d137..17d99d55 100644
--- a/src/nfa/vermicelli_simd.cpp
+++ b/src/nfa/vermicelli_simd.cpp
@@ -55,17 +55,17 @@ template <uint16_t S>
 static really_inline
 const u8 *rvermicelliBlockNeg(SuperVector<S> const data, SuperVector<S> const chars, SuperVector<S> const casemask, const u8 *buf, u16 const len);
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial = true>
 static really_inline
 const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                 u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len);
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial = true>
 static really_inline
 const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                  u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len);
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial = true>
 static really_inline
 const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2,
                                       SuperVector<S> const mask1, SuperVector<S> const mask2,
@@ -120,8 +120,8 @@ static const u8 *vermicelliExecReal(SuperVector<S> const chars, SuperVector<S> c
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> data = SuperVector<S>::loadu_maskz(d, buf_end - d);
-        rv = vermicelliBlock(data, chars, casemask, d, buf_end - d);
+        SuperVector<S> data = SuperVector<S>::loadu(buf_end - S);
+        rv = vermicelliBlock(data, chars, casemask, buf_end - S, buf_end - d);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -170,8 +170,8 @@ static const u8 *nvermicelliExecReal(SuperVector<S> const chars, SuperVector<S>
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> data = SuperVector<S>::loadu_maskz(d, buf_end - d);
-        rv = vermicelliBlockNeg(data, chars, casemask, d, buf_end - d);
+        SuperVector<S> data = SuperVector<S>::loadu(buf_end - S);
+        rv = vermicelliBlockNeg(data, chars, casemask, buf_end - S, buf_end - d);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -316,17 +316,17 @@ static const u8 *vermicelliDoubleExecReal(u8 const c1, u8 const c2, SuperVector<
         if (!ISALIGNED_N(d, S)) {
             u8 const *d1 = ROUNDUP_PTR(d, S);
             SuperVector<S> data = SuperVector<S>::loadu(d);
-            rv = vermicelliDoubleBlock(data, chars1, chars2, casemask, c1, c2, casechar, d, S);
-            if (rv) return rv;
+            rv = vermicelliDoubleBlock(data, chars1, chars2, casemask, c1, c2, casechar, d + S, S);
+            if (rv) return rv - S;
             d = d1;
         }
 
-        while(d + S <= buf_end) {
+        while(d + S < buf_end) {
             __builtin_prefetch(d + 64);
             DEBUG_PRINTF("d %p \n", d);
             SuperVector<S> data = SuperVector<S>::load(d);
-            rv = vermicelliDoubleBlock(data, chars1, chars2, casemask, c1, c2, casechar, d, S);
-            if (rv) return rv;
+            rv = vermicelliDoubleBlock(data, chars1, chars2, casemask, c1, c2, casechar, d + S, S);
+            if (rv) return rv - S;
             d += S;
         }
     }
@@ -335,8 +335,16 @@ static const u8 *vermicelliDoubleExecReal(u8 const c1, u8 const c2, SuperVector<
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> data = SuperVector<S>::loadu_maskz(d, buf_end - d);
-        rv = vermicelliDoubleBlock(data, chars1, chars2, casemask, c1, c2, casechar, d, buf_end - d);
+        SuperVector<S> data = SuperVector<S>::Zeroes();
+        const u8* end_buf;
+        if (buf_end - buf < S) {
+          memcpy(&data.u, buf, buf_end - buf);
+          end_buf = buf;
+        } else {
+          data = SuperVector<S>::loadu(buf_end - S);
+          end_buf = buf_end - S;
+        }
+        rv = vermicelliDoubleBlock<S, false>(data, chars1, chars2, casemask, c1, c2, casechar, end_buf, buf_end - d);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -403,8 +411,13 @@ const u8 *rvermicelliDoubleExecReal(char c1, char c2, SuperVector<S> const casem
     // finish off head
 
     if (d != buf) {
-        SuperVector<S> data = SuperVector<S>::loadu(buf);
-        rv = rvermicelliDoubleBlock(data, chars1, chars2, casemask, c1, c2, casechar, buf, d - buf);
+        SuperVector<S> data = SuperVector<S>::Zeroes();
+        if (d - buf < S) {
+          memcpy(&data.u, buf, d - buf);
+        } else {
+          data = SuperVector<S>::loadu(buf);
+        }
+        rv = rvermicelliDoubleBlock<S, false>(data, chars1, chars2, casemask, c1, c2, casechar, buf, d - buf);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -440,17 +453,17 @@ static const u8 *vermicelliDoubleMaskedExecReal(u8 const c1, u8 const c2, u8 con
         if (!ISALIGNED_N(d, S)) {
             u8 const *d1 = ROUNDUP_PTR(d, S);
             SuperVector<S> data = SuperVector<S>::loadu(d);
-            rv = vermicelliDoubleMaskedBlock(data, chars1, chars2, mask1, mask2, c1, c2, m1, m2, d, S);
-            if (rv) return rv;
+            rv = vermicelliDoubleMaskedBlock(data, chars1, chars2, mask1, mask2, c1, c2, m1, m2, d + S, S);
+            if (rv) return rv - S;
             d = d1;
         }
 
-        while(d + S <= buf_end) {
+        while(d + S < buf_end) {
             __builtin_prefetch(d + 64);
             DEBUG_PRINTF("d %p \n", d);
             SuperVector<S> data = SuperVector<S>::load(d);
-            rv = vermicelliDoubleMaskedBlock(data, chars1, chars2, mask1, mask2, c1, c2, m1, m2, d, S);
-            if (rv) return rv;
+            rv = vermicelliDoubleMaskedBlock(data, chars1, chars2, mask1, mask2, c1, c2, m1, m2, d + S, S);
+            if (rv) return rv - S;
             d += S;
         }
     }
@@ -459,8 +472,16 @@ static const u8 *vermicelliDoubleMaskedExecReal(u8 const c1, u8 const c2, u8 con
     // finish off tail
 
     if (d != buf_end) {
-        SuperVector<S> data = SuperVector<S>::loadu_maskz(d, buf_end - d);
-        rv = vermicelliDoubleMaskedBlock(data, chars1, chars2, mask1, mask2, c1, c2, m1, m2, d, buf_end - d);
+        SuperVector<S> data = SuperVector<S>::Zeroes();
+        const u8* end_buf;
+        if (buf_end - buf < S) {
+          memcpy(&data.u, buf, buf_end - buf);
+          end_buf = buf;
+        } else {
+          data = SuperVector<S>::loadu(buf_end - S);
+          end_buf = buf_end - S;
+        }
+        rv = vermicelliDoubleMaskedBlock<S, false>(data, chars1, chars2, mask1, mask2, c1, c2, m1, m2, end_buf, buf_end - d);
         DEBUG_PRINTF("rv %p \n", rv);
         if (rv && rv < buf_end) return rv;
     }
@@ -480,6 +501,20 @@ extern "C" const u8 *vermicelliExec(char c, char nocase, const u8 *buf, const u8
                  nocase ? "nocase " : "", c, (size_t)(buf_end - buf));
     assert(buf < buf_end);
 
+    // Small ranges.
+    if (buf_end - buf < VECTORSIZE) {
+        for (; buf < buf_end; buf++) {
+            char cur = (char)*buf;
+            if (nocase) {
+                cur &= CASE_CLEAR;
+            }
+            if (cur == c) {
+                break;
+            }
+        }
+        return buf;
+    }
+
     const SuperVector<VECTORSIZE> chars = SuperVector<VECTORSIZE>::dup_u8(c);
     const SuperVector<VECTORSIZE> casemask{nocase ? getCaseMask<VECTORSIZE>() : SuperVector<VECTORSIZE>::Ones()};
 
@@ -493,6 +528,20 @@ extern "C" const u8 *nvermicelliExec(char c, char nocase, const u8 *buf, const u
                  nocase ? "nocase " : "", c, (size_t)(buf_end - buf));
     assert(buf < buf_end);
 
+    // Small ranges.
+    if (buf_end - buf < VECTORSIZE) {
+        for (; buf < buf_end; buf++) {
+            char cur = *buf;
+            if (nocase) {
+                cur &= CASE_CLEAR;
+            }
+            if (cur != c) {
+                break;
+            }
+        }
+        return buf;
+    }
+
     const SuperVector<VECTORSIZE> chars = SuperVector<VECTORSIZE>::dup_u8(c);
     const SuperVector<VECTORSIZE> casemask{nocase ? getCaseMask<VECTORSIZE>() : SuperVector<VECTORSIZE>::Ones()};
 
@@ -504,6 +553,20 @@ extern "C" const u8 *rvermicelliExec(char c, char nocase, const u8 *buf, const u
                  nocase ? "nocase " : "", c, (size_t)(buf_end - buf));
     assert(buf < buf_end);
 
+    // Small ranges.
+    if (buf_end - buf < VECTORSIZE) {
+        for (buf_end--; buf_end >= buf; buf_end--) {
+            char cur = (char)*buf_end;
+            if (nocase) {
+                cur &= CASE_CLEAR;
+            }
+            if (cur == c) {
+                break;
+            }
+        }
+        return buf_end;
+    }
+
     const SuperVector<VECTORSIZE> chars = SuperVector<VECTORSIZE>::dup_u8(c);
     const SuperVector<VECTORSIZE> casemask{nocase ? getCaseMask<VECTORSIZE>() : SuperVector<VECTORSIZE>::Ones()};
 
@@ -515,6 +578,20 @@ extern "C" const u8 *rnvermicelliExec(char c, char nocase, const u8 *buf, const
                   nocase ? "nocase " : "", c, (size_t)(buf_end - buf));
     assert(buf < buf_end);
 
+    // Small ranges.
+    if (buf_end - buf < VECTORSIZE) {
+        for (buf_end--; buf_end >= buf; buf_end--) {
+            char cur = (char)*buf_end;
+            if (nocase) {
+                cur &= CASE_CLEAR;
+            }
+            if (cur != c) {
+                break;
+            }
+        }
+        return buf_end;
+    }
+
     const SuperVector<VECTORSIZE> chars = SuperVector<VECTORSIZE>::dup_u8(c);
     const SuperVector<VECTORSIZE> casemask{nocase ? getCaseMask<VECTORSIZE>() : SuperVector<VECTORSIZE>::Ones()};
 
diff --git a/src/nfa/x86/vermicelli.hpp b/src/nfa/x86/vermicelli.hpp
index 8b461dfe..2f219f31 100644
--- a/src/nfa/x86/vermicelli.hpp
+++ b/src/nfa/x86/vermicelli.hpp
@@ -67,7 +67,7 @@ const u8 *rvermicelliBlockNeg(SuperVector<S> const data, SuperVector<S> const ch
     return last_zero_match_inverted<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                 u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len) {
@@ -78,14 +78,16 @@ const u8 *vermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     SuperVector<S> mask = mask1 & (mask2 >> 1);
 
     DEBUG_PRINTF("rv[0] = %02hhx, rv[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
+    bool partial_match = (check_partial && ((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
-    if (partial_match) return buf - 1;
+    if (partial_match) {
+        mask = mask | ((SuperVector<S>::Ones() >> (S-1)) << (S-1));
+    }
 
     return first_non_zero_match<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2, SuperVector<S> const casemask,
                                  u8 const c1, u8 const c2, u8 const casechar, u8 const *buf, u16 const len) {
@@ -96,7 +98,7 @@ const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     SuperVector<S> mask = (mask1 << 1)& mask2;
 
     DEBUG_PRINTF("buf[0] = %02hhx, buf[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
+    bool partial_match = (check_partial && ((buf[0] & casechar) == c2) && ((buf[-1] & casechar) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
     if (partial_match) {
         mask = mask | (SuperVector<S>::Ones() >> (S-1));
@@ -105,7 +107,7 @@ const u8 *rvermicelliDoubleBlock(SuperVector<S> const data, SuperVector<S> const
     return last_non_zero_match<S>(buf, mask, len);
 }
 
-template <uint16_t S>
+template <uint16_t S, bool check_partial>
 static really_inline
 const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S> const chars1, SuperVector<S> const chars2,
                                       SuperVector<S> const mask1, SuperVector<S> const mask2,
@@ -116,9 +118,11 @@ const u8 *vermicelliDoubleMaskedBlock(SuperVector<S> const data, SuperVector<S>
     SuperVector<S> mask = v1 & (v2 >> 1);
 
     DEBUG_PRINTF("rv[0] = %02hhx, rv[-1] = %02hhx\n", buf[0], buf[-1]);
-    bool partial_match = (((buf[0] & m1) == c2) && ((buf[-1] & m2) == c1));
+    bool partial_match = (check_partial && ((buf[0] & m2) == c2) && ((buf[-1] & m1) == c1));
     DEBUG_PRINTF("partial = %d\n", partial_match);
-    if (partial_match) return buf - 1;
+    if (partial_match) {
+        mask = mask | ((SuperVector<S>::Ones() >> (S-1)) << (S-1));
+    }
 
     return first_non_zero_match<S>(buf, mask, len);
 }
diff --git a/unit/hyperscan/allocators.cpp b/unit/hyperscan/allocators.cpp
index 40c45072..a30a3702 100644
--- a/unit/hyperscan/allocators.cpp
+++ b/unit/hyperscan/allocators.cpp
@@ -99,7 +99,7 @@ TEST(CustomAllocator, TwoAlignedCompileError) {
     ASSERT_NE(nullptr, compile_err);
     EXPECT_STREQ("Allocator returned misaligned memory.", compile_err->message);
     hs_free_compile_error(compile_err);
-    hs_set_database_allocator(nullptr, nullptr);
+    hs_set_misc_allocator(nullptr, nullptr);
 }
 
 TEST(CustomAllocator, TwoAlignedDatabaseInfo) {
diff --git a/unit/internal/shuffle.cpp b/unit/internal/shuffle.cpp
index f1a03d5a..deb85e9f 100644
--- a/unit/internal/shuffle.cpp
+++ b/unit/internal/shuffle.cpp
@@ -36,6 +36,9 @@
 #include"util/supervector/supervector.hpp"
 #include "nfa/limex_shuffle.hpp"
 
+#ifdef setbit
+#undef setbit
+#endif
 
 namespace {
 
diff --git a/unit/internal/simd_utils.cpp b/unit/internal/simd_utils.cpp
index bc2421dc..69f1a64c 100644
--- a/unit/internal/simd_utils.cpp
+++ b/unit/internal/simd_utils.cpp
@@ -33,6 +33,10 @@
 #include "util/bytecode_ptr.h"
 #include "util/simd_utils.h"
 
+#ifdef setbit
+#undef setbit
+#endif
+
 using namespace std;
 using namespace ue2;