feature: x86: Runtime detection of SSE4.2 support when compiled with '-msse4.2'.

Co-authored-by: iSage <isage.dna@gmail.com>
Co-authored-by: Siddhesh Poyarekar <siddhesh@gotplt.org>

Author: 3 people

src/Makefile

@@ -490,9 +490,9 @@ LJLIB_O= lib_base.o lib_math.o lib_bit.o lib_string.o lib_table.o \
 LJLIB_C= $(LJLIB_O:.o=.c)
 
 LJCORE_O= lj_gc.o lj_err.o lj_char.o lj_bc.o lj_obj.o lj_buf.o \
-	  lj_str.o lj_tab.o lj_func.o lj_udata.o lj_meta.o lj_debug.o \
-	  lj_state.o lj_dispatch.o lj_vmevent.o lj_vmmath.o lj_strscan.o \
-	  lj_strfmt.o lj_strfmt_num.o lj_api.o lj_profile.o \
+	  lj_str.o lj_str_hash.o lj_tab.o lj_func.o lj_udata.o lj_meta.o \
+	  lj_debug.o lj_state.o lj_dispatch.o lj_vmevent.o lj_vmmath.o \
+	  lj_strscan.o lj_strfmt.o lj_strfmt_num.o lj_api.o lj_profile.o \
 	  lj_lex.o lj_parse.o lj_bcread.o lj_bcwrite.o lj_load.o \
 	  lj_ir.o lj_opt_mem.o lj_opt_fold.o lj_opt_narrow.o \
 	  lj_opt_dce.o lj_opt_loop.o lj_opt_split.o lj_opt_sink.o \

src/lj_arch.h

@@ -615,4 +615,13 @@ extern void *LJ_WIN_LOADLIBA(const char *path);
 #define LJ_52			0
 #endif
 
+/* Optimized string hashing, added by OpenResty. */
+#if LUAJIT_TARGET == LUAJIT_ARCH_X64 && defined(__GNUC__) && defined(__SSE4_2__)
+#ifndef LJ_OR_DISABLE_STRHASHCRC32
+#define LJ_OR_STRHASHCRC32 	1
+#endif
+#else
+#define LJ_OR_STRHASHCRC32 	0
+#endif
+
 #endif

src/lj_obj.h

@@ -160,6 +160,11 @@ typedef struct SBuf {
   MRef L;		/* lua_State, used for buffer resizing. */
 } SBuf;
 
+#if LJ_OR_STRHASHCRC32
+/* String hashing functions, added by OpenResty. */
+typedef MSize (*StrHashFunction)(const char *, size_t);
+#endif
+
 /* -- Tags and values ----------------------------------------------------- */
 
 /* Frame link. */
@@ -622,6 +627,9 @@ typedef struct global_State {
   MRef saved_jit_base;  /* saved jit_base for lj_err_throw */
   MRef ctype_state;	/* Pointer to C type state. */
   GCRef gcroot[GCROOT_MAX];  /* GC roots. */
+#if LJ_OR_STRHASHCRC32
+  StrHashFunction strhashfn; /* String hashing function, added by OpenResty */
+#endif
 } global_State;
 
 #define mainthread(g)	(&gcref(g->mainthref)->th)

src/lj_state.c

@@ -204,6 +204,9 @@ LUA_API lua_State *lua_newstate(lua_Alloc f, void *ud)
   setgcref(g->uvhead.prev, obj2gco(&g->uvhead));
   setgcref(g->uvhead.next, obj2gco(&g->uvhead));
   g->strmask = ~(MSize)0;
+#if LJ_OR_STRHASHCRC32
+  lj_init_strhashfn(g);
+#endif
   setnilV(registry(L));
   setnilV(&g->nilnode.val);
   setnilV(&g->nilnode.key);

src/lj_str.c

@@ -130,49 +130,6 @@ void lj_str_resize(lua_State *L, MSize newmask)
   g->strhash = newhash;
 }
 
-static MSize
-lj_str_original_hash(const char *str, size_t lenx)
-{
-  MSize len = (MSize)lenx;
-  MSize a, b, h = len;
-
-  /* Compute string hash. Constants taken from lookup3 hash by Bob Jenkins. */
-  if (len >= 4) {  /* Caveat: unaligned access! */
-    a = lj_getu32(str);
-    h ^= lj_getu32(str+len-4);
-    b = lj_getu32(str+(len>>1)-2);
-    h ^= b; h -= lj_rol(b, 14);
-    b += lj_getu32(str+(len>>2)-1);
-  } else if (len > 0) {
-    a = *(const uint8_t *)str;
-    h ^= *(const uint8_t *)(str+len-1);
-    b = *(const uint8_t *)(str+(len>>1));
-    h ^= b; h -= lj_rol(b, 14);
-  } else {
-    return 0;
-  }
-
-  a ^= h; a -= lj_rol(h, 11);
-  b ^= a; b -= lj_rol(a, 25);
-  h ^= b; h -= lj_rol(b, 16);
-
-  return h;
-}
-
-MSize
-lj_str_indep_hash(GCstr *str)
-{
-  return lj_str_original_hash(strdata(str), str->len);
-}
-
-#include "x64/src/lj_str_hash_x64.h"
-
-#if defined(LJ_ARCH_STR_HASH)
-#define LJ_STR_HASH LJ_ARCH_STR_HASH
-#else
-#define LJ_STR_HASH lj_str_original_hash
-#endif
-
 /* Intern a string and return string object. */
 GCstr *lj_str_new(lua_State *L, const char *str, size_t lenx)
 {
@@ -189,7 +146,12 @@ GCstr *lj_str_new(lua_State *L, const char *str, size_t lenx)
     return &g->strempty;
   }
 
-  h = LJ_STR_HASH(str, lenx);
+#if LJ_OR_STRHASHCRC32
+  lua_assert(g->strhashfn != NULL);
+  h = g->strhashfn(str, lenx);
+#else
+  h = lj_str_hash_orig(str, lenx);
+#endif
 
   /* Check if the string has already been interned. */
   o = gcref(g->strhash[h & g->strmask]);

src/lj_str.h

@@ -9,6 +9,7 @@
 #include <stdarg.h>
 
 #include "lj_obj.h"
+#include "lj_str_hash.h"
 
 /* String helpers. */
 LJ_FUNC int32_t LJ_FASTCALL lj_str_cmp(GCstr *a, GCstr *b);
@@ -23,7 +24,6 @@ LJ_FUNC void LJ_FASTCALL lj_str_free(global_State *g, GCstr *s);
 
 #define lj_str_newz(L, s)	(lj_str_new(L, s, strlen(s)))
 #define lj_str_newlit(L, s)	(lj_str_new(L, "" s, sizeof(s)-1))
-
-MSize lj_str_indep_hash(GCstr *str);
+#define lj_str_indep_hash(s)	(lj_str_hash_orig(strdata(s), s->len))
 
 #endif