From 337d93bcea2ccd19218ed465775884200d5a3a06 Mon Sep 17 00:00:00 2001
From: Andrew Noyes <andrew@weaselab.dev>
Date: Sat, 21 Jun 2025 21:59:21 -0400
Subject: [PATCH] simd scan with dfa as fallback

---
 src/parser3.h | 53 +++++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 49 insertions(+), 4 deletions(-)

diff --git a/src/parser3.h b/src/parser3.h
index 7da709e..311d2bb 100644
--- a/src/parser3.h
+++ b/src/parser3.h
@@ -782,10 +782,27 @@ inline PRESERVE_NONE WeaselJsonStatus n_number(Parser3 *self, char *buf,
 
 // Advance buf until double quote, backslash, invalid utf8, or codepoint <
 // 0x20
-inline PRESERVE_NONE WeaselJsonStatus n_scan_string(Parser3 *self, char *&buf,
-                                                    char *bufEnd) {
+template <class V>
+inline PRESERVE_NONE WeaselJsonStatus scan_string_impl(Parser3 *self,
+                                                       char *&buf,
+                                                       char *bufEnd) {
   const auto before = buf;
 
+  // Advance buf past normal characters
+  for (;;) {
+    if (bufEnd - buf < V::lanes) [[unlikely]] {
+      break;
+    }
+    auto v = V{(int8_t *)buf};
+    int normal =
+        (v != V::splat('"') & v != V::splat('\\') & v >= V::splat(0x20))
+            .count_leading_nonzero_lanes();
+    buf += normal;
+    if (normal < V::lanes) {
+      break;
+    }
+  }
+
   buf = (char *)self->strDfa.scan(buf, bufEnd);
 
   int len = buf - before;
@@ -805,6 +822,34 @@ inline PRESERVE_NONE WeaselJsonStatus n_scan_string(Parser3 *self, char *&buf,
   return WeaselJson_OK;
 }
 
+#ifdef __x86_64__
+constexpr int kLanes = 32;
+template WeaselJsonStatus
+scan_string_impl<simd<int8_t, kLanes, sse::Simd_x86_SSE>>(Parser3 *, char *&,
+                                                          char *);
+
+template __attribute__((target("avx2"))) WeaselJsonStatus
+scan_string_impl<simd<int8_t, kLanes, sse::Simd_x86_AVX2>>(Parser3 *, char *&,
+                                                           char *);
+
+__attribute__((target("default"))) inline PRESERVE_NONE WeaselJsonStatus
+scan_string(Parser3 *self, char *&buf, char *bufEnd) {
+  MUSTTAIL return scan_string_impl<simd<int8_t, kLanes, sse::Simd_x86_SSE>>(
+      self, buf, bufEnd);
+}
+
+__attribute__((target("avx2"))) inline PRESERVE_NONE WeaselJsonStatus
+scan_string(Parser3 *self, char *&buf, char *bufEnd) {
+  MUSTTAIL return scan_string_impl<simd<int8_t, kLanes, sse::Simd_x86_AVX2>>(
+      self, buf, bufEnd);
+}
+#else
+inline PRESERVE_NONE WeaselJsonStatus scan_string(Parser3 *self, char *buf,
+                                                  char *bufEnd) {
+  MUSTTAIL return scan_string_impl<simd<int8_t, 32>>(self, buf, bufEnd);
+}
+#endif
+
 inline PRESERVE_NONE WeaselJsonStatus n_value(Parser3 *self, char *buf,
                                               char *bufEnd) {
   assert(bufEnd - buf != 0);
@@ -836,7 +881,7 @@ inline PRESERVE_NONE WeaselJsonStatus n_value(Parser3 *self, char *buf,
     self->dataBegin = self->writeBuf = buf;
     self->pop();
     self->strDfa.reset();
-    if (auto s = n_scan_string(self, buf, bufEnd)) {
+    if (auto s = scan_string(self, buf, bufEnd)) {
       if (s == WeaselJson_AGAIN) {
         if (auto s2 = self->push({N_STRING2})) {
           return s2;
@@ -1072,7 +1117,7 @@ inline PRESERVE_NONE WeaselJsonStatus n_string(Parser3 *self, char *buf,
 
 inline PRESERVE_NONE WeaselJsonStatus n_string2(Parser3 *self, char *buf,
                                                 char *bufEnd) {
-  if (auto s = n_scan_string(self, buf, bufEnd)) {
+  if (auto s = scan_string(self, buf, bufEnd)) {
     return s;
   }
   switch (*buf) {