Add: Big-endian SWAR backends

Enhances performance for 2, 3, and 4-byte substring searches on big endian
architectures by implementing endianness-aware SWAR techniques. This bridges
the performance gap between little and big endian systems by making optimized
SWAR functions work on both architectures, eliminating the need for separate
fallback paths to slower Horspool algorithms.

Co-authored-by: Semyon Danilov <samvimes@yandex.ru>
Co-authored-by: Semyon Danilov <4058545+SammyVimes@users.noreply.github.com>

Author: ashvardanian

include/stringzilla/find.h

@@ -439,16 +439,29 @@ SZ_INTERNAL sz_cptr_t sz_find_2byte_serial_(sz_cptr_t h, sz_size_t h_length, sz_
 
     // This code simulates hyper-scalar execution, analyzing 8 offsets at a time.
     for (; h + 9 <= h_end; h += 8) {
+#if !SZ_IS_BIG_ENDIAN_
         h_even_vec.u64 = *(sz_u64_t *)h;
         h_odd_vec.u64 = (h_even_vec.u64 >> 8) | ((sz_u64_t)h[8] << 56);
+#else
+        h_even_vec.u64 = *(sz_u64_t *)h;
+        h_odd_vec.u64 = (h_even_vec.u64 << 8) | ((sz_u64_t)h[8] >> 56);
+#endif
         matches_even_vec = sz_u64_each_2byte_equal_(h_even_vec, n_vec);
         matches_odd_vec = sz_u64_each_2byte_equal_(h_odd_vec, n_vec);
 
+#if !SZ_IS_BIG_ENDIAN_
         matches_even_vec.u64 >>= 8;
         if (matches_even_vec.u64 + matches_odd_vec.u64) {
             sz_u64_t match_indicators = matches_even_vec.u64 | matches_odd_vec.u64;
             return h + sz_u64_ctz(match_indicators) / 8;
         }
+#else
+        matches_even_vec.u64 <<= 8;
+        if (matches_even_vec.u64 + matches_odd_vec.u64) {
+            sz_u64_t match_indicators = matches_even_vec.u64 | matches_odd_vec.u64;
+            return h + sz_u64_clz(match_indicators) / 8;
+        }
+#endif
     }
 
     for (; h + 2 <= h_end; ++h)
@@ -498,21 +511,36 @@ SZ_INTERNAL sz_cptr_t sz_find_4byte_serial_(sz_cptr_t h, sz_size_t h_length, sz_
     for (; h + sizeof(sz_u64_t) + sizeof(sz_u32_t) <= h_end; h += sizeof(sz_u64_t)) {
         h_page_current = *(sz_u64_t *)h;
         h_page_next = *(sz_u32_t *)(h + 8);
+#if !SZ_IS_BIG_ENDIAN_
         h0_vec.u64 = (h_page_current);
         h1_vec.u64 = (h_page_current >> 8) | (h_page_next << 56);
         h2_vec.u64 = (h_page_current >> 16) | (h_page_next << 48);
         h3_vec.u64 = (h_page_current >> 24) | (h_page_next << 40);
+#else
+        h0_vec.u64 = (h_page_current);
+        h1_vec.u64 = (h_page_current << 8) | (h_page_next >> 24);
+        h2_vec.u64 = (h_page_current << 16) | (h_page_next >> 16);
+        h3_vec.u64 = (h_page_current << 24) | (h_page_next >> 8);
+#endif
         matches0_vec = sz_u64_each_4byte_equal_(h0_vec, n_vec);
         matches1_vec = sz_u64_each_4byte_equal_(h1_vec, n_vec);
         matches2_vec = sz_u64_each_4byte_equal_(h2_vec, n_vec);
         matches3_vec = sz_u64_each_4byte_equal_(h3_vec, n_vec);
 
         if (matches0_vec.u64 | matches1_vec.u64 | matches2_vec.u64 | matches3_vec.u64) {
+#if !SZ_IS_BIG_ENDIAN_
             matches0_vec.u64 >>= 24;
             matches1_vec.u64 >>= 16;
             matches2_vec.u64 >>= 8;
             sz_u64_t match_indicators = matches0_vec.u64 | matches1_vec.u64 | matches2_vec.u64 | matches3_vec.u64;
             return h + sz_u64_ctz(match_indicators) / 8;
+#else
+            matches0_vec.u64 <<= 24;
+            matches1_vec.u64 <<= 16;
+            matches2_vec.u64 <<= 8;
+            sz_u64_t match_indicators = matches0_vec.u64 | matches1_vec.u64 | matches2_vec.u64 | matches3_vec.u64;
+            return h + sz_u64_clz(match_indicators) / 8;
+#endif
         }
     }
 
@@ -567,24 +595,41 @@ SZ_INTERNAL sz_cptr_t sz_find_3byte_serial_(sz_cptr_t h, sz_size_t h_length, sz_
         h_page_current = *(sz_u64_t *)h;
         h_page_next = *(sz_u16_t *)(h + 8);
         h0_vec.u64 = (h_page_current);
+#if !SZ_IS_BIG_ENDIAN_
         h1_vec.u64 = (h_page_current >> 8) | (h_page_next << 56);
         h2_vec.u64 = (h_page_current >> 16) | (h_page_next << 48);
         h3_vec.u64 = (h_page_current >> 24) | (h_page_next << 40);
         h4_vec.u64 = (h_page_current >> 32) | (h_page_next << 32);
+#else
+        h1_vec.u64 = (h_page_current << 8) | (h_page_next >> 8);
+        h2_vec.u64 = (h_page_current << 16) | (h_page_next >> 16);
+        h3_vec.u64 = (h_page_current << 24) | (h_page_next >> 24);
+        h4_vec.u64 = (h_page_current << 32) | (h_page_next >> 32);
+#endif
         matches0_vec = sz_u64_each_3byte_equal_(h0_vec, n_vec);
         matches1_vec = sz_u64_each_3byte_equal_(h1_vec, n_vec);
         matches2_vec = sz_u64_each_3byte_equal_(h2_vec, n_vec);
         matches3_vec = sz_u64_each_3byte_equal_(h3_vec, n_vec);
         matches4_vec = sz_u64_each_3byte_equal_(h4_vec, n_vec);
 
         if (matches0_vec.u64 | matches1_vec.u64 | matches2_vec.u64 | matches3_vec.u64 | matches4_vec.u64) {
+#if !SZ_IS_BIG_ENDIAN_
             matches0_vec.u64 >>= 16;
             matches1_vec.u64 >>= 8;
             matches3_vec.u64 <<= 8;
             matches4_vec.u64 <<= 16;
             sz_u64_t match_indicators =
                 matches0_vec.u64 | matches1_vec.u64 | matches2_vec.u64 | matches3_vec.u64 | matches4_vec.u64;
             return h + sz_u64_ctz(match_indicators) / 8;
+#else
+            matches0_vec.u64 <<= 16;
+            matches1_vec.u64 <<= 8;
+            matches3_vec.u64 >>= 8;
+            matches4_vec.u64 >>= 16;
+            sz_u64_t match_indicators =
+                matches0_vec.u64 | matches1_vec.u64 | matches2_vec.u64 | matches3_vec.u64 | matches4_vec.u64;
+            return h + sz_u64_clz(match_indicators) / 8;
+#endif
         }
     }
 
@@ -768,17 +813,8 @@ SZ_PUBLIC sz_cptr_t sz_find_serial(sz_cptr_t h, sz_size_t h_length, sz_cptr_t n,
     // This almost never fires, but it's better to be safe than sorry.
     if (h_length < n_length || !n_length) return SZ_NULL_CHAR;
 
-#if SZ_IS_BIG_ENDIAN_
-    sz_find_t backends[] = {
-        sz_find_1byte_serial_,
-        sz_find_horspool_upto_256bytes_serial_,
-        sz_find_horspool_over_256bytes_serial_,
-    };
-
-    return backends[(n_length > 1) + (n_length > 256)](h, h_length, n, n_length);
-#else
     sz_find_t backends[] = {
-        // For very short strings brute-force SWAR makes sense.
+        // For very short strings brute-force SWAR makes sense - now optimized for both endianness!
         sz_find_1byte_serial_,
         sz_find_2byte_serial_,
         sz_find_3byte_serial_,
@@ -797,7 +833,6 @@ SZ_PUBLIC sz_cptr_t sz_find_serial(sz_cptr_t h, sz_size_t h_length, sz_cptr_t n,
         (n_length > 4) +
         // For longer needles - use skip tables.
         (n_length > 8) + (n_length > 256)](h, h_length, n, n_length);
-#endif
 }
 
 SZ_PUBLIC sz_cptr_t sz_rfind_serial(sz_cptr_t h, sz_size_t h_length, sz_cptr_t n, sz_size_t n_length) {