arm64 implementation for scanning for non-normal chars in strings

src/parser3.h

@@ -14,6 +14,9 @@
 #ifdef __x86_64__
 #include <immintrin.h>
 #endif
+#ifdef __aarch64__
+#include <arm_neon.h>
+#endif
 
 #include "musttail.h"
 #include "tables.h"
@@ -429,8 +432,37 @@ inline WeaselJsonStatus n_string2(Parser3 *self) {
     }
     self->buf += 16;
   }
+#elif defined(__aarch64__)
+  for (;;) {
+    if (self->bufEnd - self->buf < 16) [[unlikely]] {
+      while (self->buf != self->bufEnd &&
+             tables.stringByteMeaning[uint8_t(*self->buf)] == Tables::NORMAL) {
+        ++self->buf;
+      }
+      break;
+    }
+    uint8x16_t x;
+    memcpy(&x, self->buf, 16);
+    const auto dubquote = vceqq_s8(vdupq_n_u8('"'), x);
+    const auto backslash = vceqq_s8(vdupq_n_u8('\\'), x);
+    const auto control_or_negative = vcgtq_s8(vdupq_n_u8(0x20), x);
+    const auto non_normal = vget_lane_u64(
+        vreinterpret_u64_u16(vshrn_n_u16(
+            vreinterpretq_u16_u8(
+                vorrq_s8(vorrq_s8(dubquote, backslash), control_or_negative)),
+            4)),
+        0);
+    if (non_normal) {
+      self->buf += std::countr_zero(non_normal) / 4;
+      break;
+    }
+    self->buf += 16;
+  }
 #else
-#error "port me"
+  while (self->buf != self->bufEnd &&
+         tables.stringByteMeaning[uint8_t(*self->buf)] == Tables::NORMAL) {
+    ++self->buf;
+  }
 #endif
 
   int len = self->buf - before;