base64: add avx512 and vbmi version. (#6361)

1. Implementation based on https://github.com/WojciechMula/base64simd
2. Only runtime path is added to reduce the complexity of SIMD variants.
3. Expand test case to cover SIMD implementation.

Signed-off-by: Frank Du <frank.du@intel.com>

Author: frankdjx

Zend/zend_cpuinfo.h

@@ -61,6 +61,11 @@ typedef enum _zend_cpu_feature {
 
 	/* EBX */
 	ZEND_CPU_FEATURE_AVX2			= (1<<5 | ZEND_CPU_EBX_MASK),
+	ZEND_CPU_FEATURE_AVX512F		= (1<<16 | ZEND_CPU_EBX_MASK),
+	ZEND_CPU_FEATURE_AVX512DQ		= (1<<17 | ZEND_CPU_EBX_MASK),
+	ZEND_CPU_FEATURE_AVX512CD		= (1<<28 | ZEND_CPU_EBX_MASK),
+	/* intentionally don't support		= (1<<30 | ZEND_CPU_EBX_MASK) */
+	/* intentionally don't support		= (1<<31 | ZEND_CPU_EBX_MASK) */
 
 	/* EDX */
 	ZEND_CPU_FEATURE_FPU			= (1<<0 | ZEND_CPU_EDX_MASK),
@@ -174,6 +179,29 @@ static inline int zend_cpu_supports_avx2(void) {
 #endif
 	return __builtin_cpu_supports("avx2");
 }
+
+#if PHP_HAVE_AVX512_SUPPORTS
+ZEND_NO_SANITIZE_ADDRESS
+static inline int zend_cpu_supports_avx512(void) {
+#if PHP_HAVE_BUILTIN_CPU_INIT
+	__builtin_cpu_init();
+#endif
+	return __builtin_cpu_supports("avx512f") && __builtin_cpu_supports("avx512dq")
+		&& __builtin_cpu_supports("avx512cd") && __builtin_cpu_supports("avx512bw")
+		&& __builtin_cpu_supports("avx512vl");
+}
+#endif
+
+#if PHP_HAVE_AVX512_VBMI_SUPPORTS
+ZEND_NO_SANITIZE_ADDRESS
+static inline int zend_cpu_supports_avx512_vbmi(void) {
+#if PHP_HAVE_BUILTIN_CPU_INIT
+	__builtin_cpu_init();
+#endif
+	return zend_cpu_supports_avx512() && __builtin_cpu_supports("avx512vbmi");
+}
+#endif
+
 #else
 
 static inline int zend_cpu_supports_sse2(void) {
@@ -203,6 +231,16 @@ static inline int zend_cpu_supports_avx(void) {
 static inline int zend_cpu_supports_avx2(void) {
 	return zend_cpu_supports(ZEND_CPU_FEATURE_AVX2);
 }
+
+static inline int zend_cpu_supports_avx512(void) {
+	/* TODO: avx512_bw/avx512_vl use bit 30/31 which are reserved for mask */
+	return 0;
+}
+
+static zend_always_inline int zend_cpu_supports_avx512_vbmi(void) {
+	/* TODO: avx512_vbmi use ECX of cpuid 7 */
+	return 0;
+}
 #endif
 
 /* __builtin_cpu_supports has pclmul from gcc9 */

Zend/zend_portability.h

@@ -651,6 +651,46 @@ extern "C++" {
 # define ZEND_INTRIN_AVX2_FUNC_DECL(func)
 #endif
 
+#if PHP_HAVE_AVX512_SUPPORTS && defined(HAVE_FUNC_ATTRIBUTE_TARGET) || defined(ZEND_WIN32)
+#define ZEND_INTRIN_AVX512_RESOLVER 1
+#endif
+
+#if defined(ZEND_INTRIN_AVX512_RESOLVER) && defined(ZEND_INTRIN_HAVE_IFUNC_TARGET)
+# define ZEND_INTRIN_AVX512_FUNC_PROTO 1
+#elif defined(ZEND_INTRIN_AVX512_RESOLVER)
+# define ZEND_INTRIN_AVX512_FUNC_PTR 1
+#endif
+
+#ifdef ZEND_INTRIN_AVX512_RESOLVER
+# ifdef HAVE_FUNC_ATTRIBUTE_TARGET
+#  define ZEND_INTRIN_AVX512_FUNC_DECL(func) ZEND_API func __attribute__((target("avx512f,avx512cd,avx512vl,avx512dq,avx512bw")))
+# else
+#  define ZEND_INTRIN_AVX512_FUNC_DECL(func) func
+# endif
+#else
+# define ZEND_INTRIN_AVX512_FUNC_DECL(func)
+#endif
+
+#if PHP_HAVE_AVX512_VBMI_SUPPORTS && defined(HAVE_FUNC_ATTRIBUTE_TARGET)
+#define ZEND_INTRIN_AVX512_VBMI_RESOLVER 1
+#endif
+
+#if defined(ZEND_INTRIN_AVX512_VBMI_RESOLVER) && defined(ZEND_INTRIN_HAVE_IFUNC_TARGET)
+# define ZEND_INTRIN_AVX512_VBMI_FUNC_PROTO 1
+#elif defined(ZEND_INTRIN_AVX512_VBMI_RESOLVER)
+# define ZEND_INTRIN_AVX512_VBMI_FUNC_PTR 1
+#endif
+
+#ifdef ZEND_INTRIN_AVX512_VBMI_RESOLVER
+# ifdef HAVE_FUNC_ATTRIBUTE_TARGET
+#  define ZEND_INTRIN_AVX512_VBMI_FUNC_DECL(func) ZEND_API func __attribute__((target("avx512f,avx512cd,avx512vl,avx512dq,avx512bw,avx512vbmi")))
+# else
+#  define ZEND_INTRIN_AVX512_VBMI_FUNC_DECL(func) func
+# endif
+#else
+# define ZEND_INTRIN_AVX512_VBMI_FUNC_DECL(func)
+#endif
+
 /* Intrinsics macros end. */
 
 #ifdef ZEND_WIN32

build/php.m4

@@ -2807,3 +2807,58 @@ AC_DEFUN([PHP_CHECK_PROCCTL],
     AC_MSG_RESULT([no])
   ])
 ])
+
+dnl
+dnl PHP_CHECK_AVX512_SUPPORTS
+dnl
+AC_DEFUN([PHP_CHECK_AVX512_SUPPORTS], [
+  AC_MSG_CHECKING([for avx512 supports in compiler])
+  save_CFLAGS="$CFLAGS"
+  CFLAGS="-mavx512f -mavx512cd -mavx512vl -mavx512dq -mavx512bw $CFLAGS"
+
+  AC_LINK_IFELSE([AC_LANG_SOURCE([[
+    #include <immintrin.h>
+      int main() {
+        __m512i mask = _mm512_set1_epi32(0x1);
+        char out[32];
+        _mm512_storeu_si512(out, _mm512_shuffle_epi8(mask, mask));
+        return 0;
+    }]])], [
+    have_avx512_supports=1
+    AC_MSG_RESULT([yes])
+  ], [
+    have_avx512_supports=0
+    AC_MSG_RESULT([no])
+  ])
+
+  CFLAGS="$save_CFLAGS"
+
+  AC_DEFINE_UNQUOTED([PHP_HAVE_AVX512_SUPPORTS],
+   [$have_avx512_supports], [Whether the compiler supports AVX512])
+])
+
+dnl
+dnl PHP_CHECK_AVX512_VBMI_SUPPORTS
+dnl
+AC_DEFUN([PHP_CHECK_AVX512_VBMI_SUPPORTS], [
+  AC_MSG_CHECKING([for avx512 vbmi supports in compiler])
+  save_CFLAGS="$CFLAGS"
+  CFLAGS="-mavx512f -mavx512cd -mavx512vl -mavx512dq -mavx512bw -mavx512vbmi $CFLAGS"
+  AC_LINK_IFELSE([AC_LANG_SOURCE([[
+    #include <immintrin.h>
+      int main() {
+        __m512i mask = _mm512_set1_epi32(0x1);
+        char out[32];
+        _mm512_storeu_si512(out, _mm512_permutexvar_epi8(mask, mask));
+        return 0;
+    }]])], [
+    have_avx512_vbmi_supports=1
+    AC_MSG_RESULT([yes])
+  ], [
+    have_avx512_vbmi_supports=0
+    AC_MSG_RESULT([no])
+  ])
+  CFLAGS="$save_CFLAGS"
+  AC_DEFINE_UNQUOTED([PHP_HAVE_AVX512_VBMI_SUPPORTS],
+   [$have_avx512_vbmi_supports], [Whether the compiler supports AVX512 VBMI])
+])

configure.ac

@@ -520,6 +520,10 @@ dnl Check prctl
 PHP_CHECK_PRCTL
 dnl Check procctl
 PHP_CHECK_PROCCTL
+dnl Check AVX512
+PHP_CHECK_AVX512_SUPPORTS
+dnl Check AVX512 VBMI
+PHP_CHECK_AVX512_VBMI_SUPPORTS
 
 dnl Check for __alignof__ support in the compiler
 AC_CACHE_CHECK(whether the compiler supports __alignof__, ac_cv_alignof_exists,[