feature: we added the '-msse4.2' and '-march' CC options to always dynamically enable the CRC32 string hashing optimization.

Author: thibaultcha

README.md

@@ -163,10 +163,8 @@ local newstate = jit.prngstate(123456)
 **syntax:** *ok = jit.crc32()*
 
 Returns a boolean value indicating if the current architecture supports a CRC32
-instruction set.
-
-CRC32 support will be checked at runtime if LuaJIT was compiled with `-msse4.2`
-on x64 architectures, or `-march=armv8-a+crc` on ARM64.
+instruction set. CRC32 support will be checked at runtime on x64 and ARM64
+platforms.
 
 CRC32 support allows for this branch to use an optimized string hashing
 algorithm. See the [String hashing](#string-hashing) section for details on
@@ -285,10 +283,8 @@ but makes hash collision attacks harder for strings up to 127 bytes of size
 (see `lj_str_new()`).
 
 This optimization is only available for x64 and ARM64 architectures, and will
-be enabled if:
-
-1. LuaJIT is compiled with `-msse4.2` on x64, or `-march=armv8-a+crc` on ARM64.
-2. A CRC32 instruction set is detected at runtime (see [jit.crc32](#jitcrc32)).
+be enabled if a CRC32 instruction set is detected at runtime (see
+[jit.crc32](#jitcrc32)).
 
 **Note:** This optimization can be disabled by compiling LuaJIT with
 `-DLJ_OR_DISABLE_STRHASHCRC32`.

src/Makefile

@@ -47,10 +47,10 @@ CCOPT= -O2 -fomit-frame-pointer
 # x86/x64 only: For GCC 4.2 or higher and if you don't intend to distribute
 # the binaries to a different machine you could also use: -march=native
 #
-CCOPT_x86= -march=i686 -msse -msse2 -mfpmath=sse
+CCOPT_x86= -march=i686 -msse -msse2 -msse4.2 -mfpmath=sse
 CCOPT_x64=
 CCOPT_arm=
-CCOPT_arm64=
+CCOPT_arm64= -march=armv8-a+crc
 CCOPT_ppc=
 CCOPT_mips=
 #