optimize skipstring with sve2 match

optimize-skipstring.diff

diff --git a/cmake/set_arch_flags.cmake b/cmake/set_arch_flags.cmake
index 538ddfe..6dc7754 100644
--- a/cmake/set_arch_flags.cmake
+++ b/cmake/set_arch_flags.cmake
@@ -2,8 +2,8 @@ function(set_arch_flags target arch)
   message(STATUS "Setting architecture flags for ${arch}")
   if(arch MATCHES "x86_64")
     target_compile_options(${target} PRIVATE -mavx2 -mpclmul -mbmi -mlzcnt)
-  elseif(arch MATCHES "arm")
-    target_compile_options(${target} PRIVATE -march=armv8-a)
+  elseif(arch MATCHES "arm|aarch64")
+    target_compile_options(${target} PRIVATE -march=armv8.5-a+sve2 -msve-vector-bits=128)
   else()
     message(FATAL_ERROR "Unsupported architecture: ${arch}")
   endif()
diff --git a/include/sonic/internal/arch/neon/skip.h b/include/sonic/internal/arch/neon/skip.h
index 8be0926..f048b68 100644
--- a/include/sonic/internal/arch/neon/skip.h
+++ b/include/sonic/internal/arch/neon/skip.h
@@ -26,6 +26,10 @@
 #include "sonic/macro.h"
 #include "unicode.h"
 
+#include <arm_sve.h>
+typedef svuint8_t svuint8x16_t __attribute__((arm_sve_vector_bits(128)));
+typedef svbool_t svbool8x16_t __attribute__((arm_sve_vector_bits(128)));
+
 namespace sonic_json {
 namespace internal {
 namespace neon {
@@ -79,6 +83,37 @@ sonic_force_inline int SkipString(const uint8_t *data, size_t &pos,
   const static int kEscaped = 2;
   const static int kNormal = 1;
   const static int kUnclosed = 0;
+#if 1
+  int ret = kNormal;
+  const svbool8x16_t ptrue = svptrue_b8();
+  const uint8_t needle_array[] = R"(\""""""""""""""")";
+  const svuint8x16_t needle = svld1(ptrue, needle_array);
+  while (pos + VEC_LEN <= len) {
+    const svuint8x16_t v = svld1(ptrue, data + pos);
+    const svbool8x16_t mask = svmatch_u8(ptrue, v, needle);
+    if (!svptest_any(mask, mask)) {
+      pos += VEC_LEN;
+      continue;
+    }
+    for (int i = 0; i < VEC_LEN; ++i) {
+      if (data[pos + i] == '"') {
+        pos += i + 1;
+        return ret;
+      } else if (data[pos + i] == '\\') {
+        ret = kEscaped;
+        pos += i + 2;
+        while (pos < len) {
+          if (data[pos] == '\\') {
+            pos += 2;
+          } else {
+            i = VEC_LEN;
+            break;
+          }
+        }
+      }
+    }
+  }
+#else
   uint64_t quote_bits = 0;
   uint64_t bs_bits = 0;
   int ret = kNormal;
@@ -105,6 +140,7 @@ sonic_force_inline int SkipString(const uint8_t *data, size_t &pos,
       pos += VEC_LEN;
     }
   }
+#endif
   while (pos < len) {
     if (data[pos] == '\\') {
       if (pos + 1 >= len) {