feat: improving parser performance using SSE Instructions

helintongh · Jul 20, 2023 · 063cf9c · 063cf9c
1 parent 1e1cdb9
commit 063cf9c
Show file tree

Hide file tree

Showing 3 changed files with 78 additions and 2 deletions.
diff --git a/cmake/develop.cmake b/cmake/develop.cmake
@@ -70,6 +70,23 @@ if(ENABLE_CLIENT_SSL)
 	add_definitions(-DCINATRA_ENABLE_CLIENT_SSL)
 endif()
 
+
+if(ENABLE_SIMD STREQUAL "SSE42" OR ENABLE_SIMD STREQUAL "AVX2" OR ENABLE_SIMD STREQUAL "AARCH64")
+	if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
+		message(STATUS "Build with simd in aarch64")
+		add_definitions(-DCINATRA_AARCH64)
+	elseif (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "x86_64")
+		message(STATUS "Build with simd in X86_64")
+		if (ENABLE_SIMD STREQUAL "SSE42")
+			message(STATUS "Build with SSE4.2 ISA")
+			add_definitions(-DCINATRA_SSE)
+		elseif (ENABLE_SIMD STREQUAL "AVX2")
+			message(STATUS "Build with AVX2 ISA")
+			add_definitions(-DCINATRA_AVX2)
+		endif ()
+	endif ()
+endif()
+
 add_definitions(-DASIO_STANDALONE)
 
 if (ENABLE_SSL)

diff --git a/include/cinatra/picohttpparser.h b/include/cinatra/picohttpparser.h
@@ -30,6 +30,16 @@
 #include <stddef.h>
 #include <string.h>
 
+#ifdef CINATRA_SSE
+#ifdef _MSC_VER
+#include <nmmintrin.h>
+#else
+#include <x86intrin.h>
+#endif
+#endif
+
+#include <sys/types.h>
+
 #ifdef _MSC_VER
 #define ssize_t intptr_t
 #endif
@@ -163,18 +173,52 @@ static const char *findchar_fast(const char *buf, const char *buf_end,
                                  const char *ranges, int ranges_size,
                                  int *found) {
   *found = 0;
+#ifdef CINATRA_SSE
+  if (likely(buf_end - buf >= 16)) {
+    __m128i ranges16 = _mm_loadu_si128((const __m128i *)ranges);
+
+    size_t left = (buf_end - buf) & ~15;
+    do {
+      __m128i b16 = _mm_loadu_si128((const __m128i *)buf);
+      int r = _mm_cmpestri(
+          ranges16, ranges_size, b16, 16,
+          _SIDD_LEAST_SIGNIFICANT | _SIDD_CMP_RANGES | _SIDD_UBYTE_OPS);
+      if (unlikely(r != 16)) {
+        buf += r;
+        *found = 1;
+        break;
+      }
+      buf += 16;
+      left -= 16;
+    } while (likely(left != 0));
+  }
+#else
   /* suppress unused parameter warning */
   (void)buf_end;
   (void)ranges;
   (void)ranges_size;
+#endif
   return buf;
 }
 
 static const char *get_token_to_eol(const char *buf, const char *buf_end,
                                     const char **token, size_t *token_len,
                                     int *ret) {
   const char *token_start = buf;
-
+#ifdef CINATRA_SSE
+  static const char ranges1[] =
+      "\0\010"
+      /* allow HT */
+      "\012\037"
+      /* allow SP and up to but not including DEL */
+      "\177\177"
+      /* allow chars w. MSB set */
+      ;
+  int found;
+  buf = findchar_fast(buf, buf_end, ranges1, sizeof(ranges1) - 1, &found);
+  if (found)
+    goto FOUND_CTL;
+#else
   /* find non-printable char within the next 8 bytes, this is the hottest code;
    * manually inlined */
   while (likely(buf_end - buf >= 8)) {
@@ -201,7 +245,7 @@ static const char *get_token_to_eol(const char *buf, const char *buf_end,
     }
     ++buf;
   }
-
+#endif
   for (;; ++buf) {
     CHECK_EOF();
     if (unlikely(!IS_PRINTABLE_ASCII(*buf))) {

diff --git a/press_tool/CMakeLists.txt b/press_tool/CMakeLists.txt
@@ -22,6 +22,21 @@ endif()
 if (ENABLE_GZIP)
 	target_link_libraries(${project_name} ${ZLIB_LIBRARIES})
 endif()
+
+if (ENABLE_SIMD STREQUAL "AARCH64")
+	if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
+		#TODO
+	endif ()
+elseif (ENABLE_SIMD STREQUAL "SSE42")
+	if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "x86_64")
+		target_link_libraries(${project_name} sse4.2)
+	endif ()
+elseif (ENABLE_SIMD STREQUAL "AVX2")
+	if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "x86_64")
+		target_link_libraries(${project_name} avx2)
+	endif ()
+endif ()
+
 install(TARGETS ${project_name} DESTINATION include)
 
 set(unittest_press_tool test_cinatra_press_tool)