Optimize HT_HASH_RESET

Commit d835de1 added support for AVX2 in hash table initialization
code. The same kind of code also occurs for HT_HASH_RESET. However, this
place was forgotten in that patch. That is unfortunate, because a loop
is just when there may be the most benefit from this SIMD sequence.
Furthermore, the NEON special handling exists in the initialization code
but is also missing from HT_HASH_RESET, so add this as well.

Author: nielsdos

Zend/zend_types.h

@@ -31,6 +31,12 @@
 # include <mmintrin.h>
 # include <emmintrin.h>
 #endif
+#if defined(__AVX2__)
+# include <immintrin.h>
+#endif
+#if defined(__aarch64__) || defined(_M_ARM64)
+# include <arm_neon.h>
+#endif
 
 #ifdef WORDS_BIGENDIAN
 # define ZEND_ENDIAN_LOHI(lo, hi)          hi; lo;
@@ -460,7 +466,21 @@ struct _zend_array {
 	HT_PACKED_SIZE_EX((ht)->nTableSize, (ht)->nTableMask)
 #define HT_PACKED_USED_SIZE(ht) \
 	(HT_HASH_SIZE((ht)->nTableMask) + ((size_t)(ht)->nNumUsed * sizeof(zval)))
-#ifdef __SSE2__
+#if defined(__AVX2__)
+# define HT_HASH_RESET(ht) do { \
+		char *p = (char*)&HT_HASH(ht, (ht)->nTableMask); \
+		size_t size = HT_HASH_SIZE((ht)->nTableMask); \
+		__m256i ymm0 = _mm256_setzero_si256(); \
+		ymm0 = _mm256_cmpeq_epi64(ymm0, ymm0); \
+		ZEND_ASSERT(size >= 64 && ((size & 0x3f) == 0)); \
+		do { \
+			_mm256_storeu_si256((__m256i*)p, ymm0); \
+			_mm256_storeu_si256((__m256i*)(p+32), ymm0); \
+			p += 64; \
+			size -= 64; \
+		} while (size != 0); \
+	} while (0)
+#elif defined(__SSE2__)
 # define HT_HASH_RESET(ht) do { \
 		char *p = (char*)&HT_HASH(ht, (ht)->nTableMask); \
 		size_t size = HT_HASH_SIZE((ht)->nTableMask); \
@@ -476,6 +496,21 @@ struct _zend_array {
 			size -= 64; \
 		} while (size != 0); \
 	} while (0)
+#elif defined(__aarch64__) || defined(_M_ARM64)
+# define HT_HASH_RESET(ht) do { \
+		char *p = (char*)&HT_HASH(ht, (ht)->nTableMask); \
+		size_t size = HT_HASH_SIZE((ht)->nTableMask); \
+		int32x4_t t = vdupq_n_s32(-1); \
+		ZEND_ASSERT(size >= 64 && ((size & 0x3f) == 0)); \
+		do { \
+			vst1q_s32((int32_t*)p, t); \
+			vst1q_s32((int32_t*)(p+16), t); \
+			vst1q_s32((int32_t*)(p+32), t); \
+			vst1q_s32((int32_t*)(p+48), t); \
+			p += 64; \
+			size -= 64; \
+		} while (size != 0); \
+	} while (0)
 #else
 # define HT_HASH_RESET(ht) \
 	memset(&HT_HASH(ht, (ht)->nTableMask), HT_INVALID_IDX, HT_HASH_SIZE((ht)->nTableMask))