x86: add SSE4_2 PCLMUL version for crc32

Based on:
 "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction"
 V. Gopal, E. Ozturk, et al., 2009, http://intel.ly/2ySEwL0

Signed-off-by: Frank Du <[email protected]>

Author: frankdjx

ext/hash/hash_crc32.c

@@ -18,6 +18,7 @@
 #include "php_hash.h"
 #include "php_hash_crc32.h"
 #include "php_hash_crc32_tables.h"
+#include "ext/standard/crc32_x86.h"
 
 PHP_HASH_API void PHP_CRC32Init(PHP_CRC32_CTX *context)
 {
@@ -28,7 +29,9 @@ PHP_HASH_API void PHP_CRC32Update(PHP_CRC32_CTX *context, const unsigned char *i
 {
 	size_t i;
 
-	for (i = 0; i < len; ++i) {
+	i = crc32_x86_simd_update(X86_CRC32, &context->state, input, len);
+
+	for (; i < len; ++i) {
 		context->state = (context->state << 8) ^ crc32_table[(context->state >> 24) ^ (input[i] & 0xff)];
 	}
 }
@@ -37,7 +40,9 @@ PHP_HASH_API void PHP_CRC32BUpdate(PHP_CRC32_CTX *context, const unsigned char *
 {
 	size_t i;
 
-	for (i = 0; i < len; ++i) {
+	i = crc32_x86_simd_update(X86_CRC32B, &context->state, input, len);
+
+	for (; i < len; ++i) {
 		context->state = (context->state >> 8) ^ crc32b_table[(context->state ^ input[i]) & 0xff];
 	}
 }
@@ -46,7 +51,9 @@ PHP_HASH_API void PHP_CRC32CUpdate(PHP_CRC32_CTX *context, const unsigned char *
 {
 	size_t i;
 
-	for (i = 0; i < len; ++i) {
+	i = crc32_x86_simd_update(X86_CRC32C, &context->state, input, len);
+
+	for (; i < len; ++i) {
 		context->state = (context->state >> 8) ^ crc32c_table[(context->state ^ input[i]) & 0xff];
 	}
 }

ext/standard/basic_functions.c

@@ -33,6 +33,7 @@
 #include "ext/standard/php_dns.h"
 #include "ext/standard/php_uuencode.h"
 #include "ext/standard/php_mt_rand.h"
+#include "ext/standard/crc32_x86.h"
 
 #ifdef PHP_WIN32
 #include "win32/php_win32_globals.h"
@@ -366,6 +367,10 @@ PHP_MINIT_FUNCTION(basic) /* {{{ */
 	BASIC_MINIT_SUBMODULE(string_intrin)
 #endif
 
+#if ZEND_INTRIN_SSE4_2_PCLMUL_FUNC_PTR
+	BASIC_MINIT_SUBMODULE(crc32_x86_intrin)
+#endif
+
 #if ZEND_INTRIN_AVX2_FUNC_PTR || ZEND_INTRIN_SSSE3_FUNC_PTR
 	BASIC_MINIT_SUBMODULE(base64_intrin)
 #endif

ext/standard/config.m4

@@ -449,7 +449,7 @@ PHP_NEW_EXTENSION(standard, array.c base64.c basic_functions.c browscap.c crc32.
                             http_fopen_wrapper.c php_fopen_wrapper.c credits.c css.c \
                             var_unserializer.c ftok.c sha1.c user_filters.c uuencode.c \
                             filters.c proc_open.c streamsfuncs.c http.c password.c \
-                            random.c net.c hrtime.c,,,
+                            random.c net.c hrtime.c crc32_x86.c,,,
 			    -DZEND_ENABLE_STATIC_TSRMLS_CACHE=1)
 
 PHP_ADD_MAKEFILE_FRAGMENT

ext/standard/config.w32

@@ -25,7 +25,7 @@ ADD_FLAG("LIBS_STANDARD", "iphlpapi.lib");
 
 EXTENSION("standard", "array.c base64.c basic_functions.c browscap.c \
 	crc32.c crypt.c crypt_freesec.c crypt_blowfish.c crypt_sha256.c \
-	crypt_sha512.c  php_crypt_r.c \
+	crypt_sha512.c  php_crypt_r.c crc32_x86.c \
 	datetime.c dir.c dl.c dns.c dns_win32.c exec.c \
 	file.c filestat.c formatted_print.c fsock.c head.c html.c image.c \
 	info.c iptc.c lcg.c link.c mail.c math.c md5.c metaphone.c microtime.c \

ext/standard/crc32.c

@@ -17,6 +17,7 @@
 #include "php.h"
 #include "basic_functions.h"
 #include "crc32.h"
+#include "crc32_x86.h"
 
 #if HAVE_AARCH64_CRC32
 # include <arm_acle.h>
@@ -72,9 +73,9 @@ static uint32_t crc32_aarch64(uint32_t crc, char *p, size_t nr) {
 PHP_FUNCTION(crc32)
 {
 	char *p;
-	size_t nr;
+	size_t nr, nr_simd;
 	uint32_t crcinit = 0;
-	register uint32_t crc;
+	uint32_t crc;
 
 	ZEND_PARSE_PARAMETERS_START(1, 1)
 		Z_PARAM_STRING(p, nr)
@@ -89,6 +90,10 @@ PHP_FUNCTION(crc32)
 	}
 #endif
 
+	nr_simd = crc32_x86_simd_update(X86_CRC32B, &crc, (const unsigned char *)p, nr);
+	nr -= nr_simd;
+	p += nr_simd;
+
 	for (; nr--; ++p) {
 		crc = ((crc >> 8) & 0x00FFFFFF) ^ crc32tab[(crc ^ (*p)) & 0xFF ];
 	}