Just do a simple lookup table

src/test.cpp

@@ -3,7 +3,6 @@
 #include <cstdint>
 #include <cstdio>
 #include <cstring>
-#include <immintrin.h>
 
 #include <initializer_list>
 #include <string>
@@ -158,59 +157,18 @@ const char *symbolNames[] = {
 };
 
 int leadingWhitespaceCount(const char *buf, int len) {
-  // Based on
-  // http://0x80.pl/articles/simd-byte-lookup.html#special-case-1-small-sets
-  constexpr uint8_t charSet[] = {' ', '\t', '\n', '\r'};
   constexpr static struct Table {
     constexpr Table() {
-      static_assert(sizeof(charSet) < 8);
-      uint8_t bitElement[sizeof(charSet)]{};
-      for (int i = 0; i < int(sizeof(charSet)); ++i) {
-        bitElement[i] = 1 << i;
-      }
-      for (int i = 0; i < int(sizeof(charSet)); ++i) {
-        uint8_t c = charSet[i];
-        int low = c & 0xf;
-        int high = c >> 4;
-        lowNibbleTable[low] |= bitElement[i];
-        highNibbleTable[high] |= bitElement[i];
+      for (int i = 0; i < 256; ++i) {
+        notWhitespace[i] = !(i == ' ' || i == '\n' || i == '\t' || i == '\r');
       }
     }
-    // lowNibbleTable[i] is the set of chars with i as its low nibble
-    alignas(16) uint8_t lowNibbleTable[16]{};
-    // highNibbleTable[i] is the set of chars with i as its high nibble
-    alignas(16) uint8_t highNibbleTable[16]{};
+    alignas(16) bool notWhitespace[256]{};
   } table;
 
   int i = 0;
-  for (; i + 16 <= len; i += 16) {
-    __m128i copy;
-    memcpy(&copy, &buf[i], sizeof(copy));
-    const __m128i input = _mm_loadu_si128((const __m128i *)&copy);
-    const __m128i lower_nibbles = _mm_and_si128(input, _mm_set1_epi8(0x0f));
-    const __m128i higher_nibbles =
-        _mm_and_si128(_mm_srli_epi16(input, 4), _mm_set1_epi8(0x0f));
-    const __m128i lo_translated = _mm_shuffle_epi8(
-        _mm_load_si128((const __m128i *)table.lowNibbleTable), lower_nibbles);
-    const __m128i hi_translated = _mm_shuffle_epi8(
-        _mm_load_si128((const __m128i *)table.highNibbleTable), higher_nibbles);
-    const __m128i intersection = _mm_and_si128(lo_translated, hi_translated);
-    uint32_t notInCharSet =
-        _mm_movemask_epi8(_mm_cmpeq_epi8(intersection, _mm_setzero_si128()));
-    if (notInCharSet != 0) {
-      return i + __builtin_ctz(notInCharSet);
-    }
-  }
-
   for (; i < len; ++i) {
-    bool any = false;
-    for (auto c : charSet) {
-      if (buf[i] == c) {
-        any = true;
-        break;
-      }
-    }
-    if (!any) {
+    if (table.notWhitespace[buf[i]]) {
       break;
     }
   }
@@ -242,10 +200,7 @@ private:
   int len() const { return bufEnd - buf; }
 
   // Helpers
-  void maybeSkipWs() {
-    int leadingWs = leadingWhitespaceCount(buf, len());
-    buf += leadingWs;
-  }
+  void maybeSkipWs() { buf += leadingWhitespaceCount(buf, len()); }
   bool parseLiteral(const char *literal) {
     const int litLen = strlen(literal);
     if (len() < litLen) {
@@ -481,11 +436,7 @@ struct Parser2 {
 
 private:
   // Helpers
-  void maybeSkipWs() {
-    int leadingWs = leadingWhitespaceCount(buf, len());
-    // printf("ws: %d\n", leadingWs);
-    buf += leadingWs;
-  }
+  void maybeSkipWs() { buf += leadingWhitespaceCount(buf, len()); }
   bool parseLiteral(const char *literal) {
     const int litLen = strlen(literal);
     if (len() < litLen) {
@@ -717,13 +668,6 @@ Callbacks printCallbacks() {
 
 } // namespace
 
-TEST_CASE("leadingWhitespaceCount") {
-  {
-    const char *s = " \r\t\n x                                       ";
-    CHECK(leadingWhitespaceCount(s, strlen(s)) == 5);
-  }
-}
-
 TEST_CASE("parser1") {
   Callbacks c = printCallbacks();
   auto copy = json;