Use bulk conversion in BCMath of BCD/CHAR where possible (#14103)

On my i7-4790 with benchmark from #14076, on top of #14101 I obtain the
following results:

before (with #14101):
```
1.672737121582
2.3618471622467
2.3474779129028
```

after (with #14101 + this):
```
1.5878579616547
2.0568618774414
2.0204811096191
```

Author: nielsdos

ext/bcmath/config.m4

@@ -8,7 +8,7 @@ if test "$PHP_BCMATH" != "no"; then
 libbcmath/src/add.c libbcmath/src/div.c libbcmath/src/init.c libbcmath/src/neg.c libbcmath/src/raisemod.c libbcmath/src/sub.c \
 libbcmath/src/compare.c libbcmath/src/divmod.c libbcmath/src/int2num.c libbcmath/src/num2long.c libbcmath/src/output.c libbcmath/src/recmul.c \
 libbcmath/src/sqrt.c libbcmath/src/zero.c libbcmath/src/doaddsub.c libbcmath/src/floor_or_ceil.c libbcmath/src/nearzero.c libbcmath/src/num2str.c \
-libbcmath/src/raise.c libbcmath/src/rmzero.c libbcmath/src/round.c libbcmath/src/str2num.c,
+libbcmath/src/raise.c libbcmath/src/rmzero.c libbcmath/src/round.c libbcmath/src/str2num.c libbcmath/src/convert.c,
           $ext_shared,, -DZEND_ENABLE_STATIC_TSRMLS_CACHE=1)
   PHP_ADD_BUILD_DIR($ext_builddir/libbcmath/src)
   AC_DEFINE(HAVE_BCMATH, 1, [Whether you have bcmath])

ext/bcmath/config.w32

@@ -8,7 +8,7 @@ if (PHP_BCMATH == "yes") {
 		raisemod.c sub.c compare.c divmod.c int2num.c \
 		num2long.c output.c recmul.c sqrt.c zero.c doaddsub.c \
 		floor_or_ceil.c nearzero.c num2str.c raise.c rmzero.c str2num.c \
-		round.c", "bcmath");
+		round.c convert.c", "bcmath");
 
 	AC_DEFINE('HAVE_BCMATH', 1, 'Have BCMATH library');
 }

ext/bcmath/libbcmath/src/convert.c

@@ -0,0 +1,66 @@
+/*
+   +----------------------------------------------------------------------+
+   | Copyright (c) The PHP Group                                          |
+   +----------------------------------------------------------------------+
+   | This source file is subject to version 3.01 of the PHP license,      |
+   | that is bundled with this package in the file LICENSE, and is        |
+   | available through the world-wide-web at the following url:           |
+   | https://www.php.net/license/3_01.txt                                 |
+   | If you did not receive a copy of the PHP license and are unable to   |
+   | obtain it through the world-wide-web, please send a note to          |
+   | [email protected] so we can mail you a copy immediately.               |
+   +----------------------------------------------------------------------+
+   | Authors: Niels Dossche <[email protected]>                            |
+   +----------------------------------------------------------------------+
+*/
+
+#include "bcmath.h"
+#include "convert.h"
+
+/* This will be 0x01010101 for 32-bit and 0x0101010101010101 */
+#define SWAR_ONES (~((size_t) 0) / 0xFF)
+/* This repeats a byte `x` into an entire 32/64-bit word.
+ * Example: SWAR_REPEAT(0xAB) will be 0xABABABAB for 32-bit and 0xABABABABABABABAB for 64-bit. */
+#define SWAR_REPEAT(x) (SWAR_ONES * (x))
+
+static char *bc_copy_and_shift_numbers(char *restrict dest, const char *source, const char *source_end, unsigned char shift, bool add)
+{
+	size_t bulk_shift = SWAR_REPEAT(shift);
+	if (!add) {
+		bulk_shift = -bulk_shift;
+		shift = -shift;
+	}
+
+	/* Handle sizeof(size_t) (i.e. 4/8) bytes at once.
+	 * We know that adding/subtracting an individual byte cannot overflow,
+	 * so it is possible to add/subtract an entire word of bytes at once
+	 * by using SWAR_REPEAT. */
+	while (source + sizeof(size_t) <= source_end) {
+		size_t bytes;
+		memcpy(&bytes, source, sizeof(bytes));
+
+		bytes += bulk_shift;
+		memcpy(dest, &bytes, sizeof(bytes));
+
+		source += sizeof(size_t);
+		dest += sizeof(size_t);
+	}
+
+	while (source < source_end) {
+		*dest = *source + shift;
+		dest++;
+		source++;
+	}
+
+	return dest;
+}
+
+char *bc_copy_ch_val(char *restrict dest, const char *source, const char *source_end)
+{
+	return bc_copy_and_shift_numbers(dest, source, source_end, '0', false);
+}
+
+char *bc_copy_bcd_val(char *restrict dest, const char *source, const char *source_end)
+{
+	return bc_copy_and_shift_numbers(dest, source, source_end, '0', true);
+}

ext/bcmath/libbcmath/src/convert.h

@@ -0,0 +1,23 @@
+/*
+   +----------------------------------------------------------------------+
+   | Copyright (c) The PHP Group                                          |
+   +----------------------------------------------------------------------+
+   | This source file is subject to version 3.01 of the PHP license,      |
+   | that is bundled with this package in the file LICENSE, and is        |
+   | available through the world-wide-web at the following url:           |
+   | https://www.php.net/license/3_01.txt                                 |
+   | If you did not receive a copy of the PHP license and are unable to   |
+   | obtain it through the world-wide-web, please send a note to          |
+   | [email protected] so we can mail you a copy immediately.               |
+   +----------------------------------------------------------------------+
+   | Authors: Niels Dossche <[email protected]>                            |
+   +----------------------------------------------------------------------+
+*/
+
+#ifndef BCMATH_CONVERT_H
+#define BCMATH_CONVERT_H
+
+char *bc_copy_ch_val(char *restrict dest, const char *source, const char *source_end);
+char *bc_copy_bcd_val(char *restrict dest, const char *source, const char *source_end);
+
+#endif

ext/bcmath/libbcmath/src/num2str.c

@@ -30,7 +30,7 @@
 *************************************************************************/
 
 #include "bcmath.h"
-#include <stddef.h>
+#include "convert.h"
 #include "zend_string.h"
 
 /* Convert a numbers to a string.  Base 10 only.*/
@@ -40,9 +40,10 @@ zend_string *bc_num2str_ex(bc_num num, size_t scale)
 	char *sptr;
 	size_t index;
 	bool signch;
+	size_t min_scale = MIN(num->n_scale, scale);
 
 	/* Number of sign chars. */
-	signch = num->n_sign != PLUS && !bc_is_zero_for_scale(num, MIN(num->n_scale, scale));
+	signch = num->n_sign != PLUS && !bc_is_zero_for_scale(num, min_scale);
 	/* Allocate the string memory. */
 	if (scale > 0) {
 		str = zend_string_alloc(num->n_len + scale + signch + 1, 0);
@@ -56,16 +57,13 @@ zend_string *bc_num2str_ex(bc_num num, size_t scale)
 
 	/* Load the whole number. */
 	const char *nptr = num->n_value;
-	for (index = num->n_len; index > 0; index--) {
-		*sptr++ = BCD_CHAR(*nptr++);
-	}
+	sptr = bc_copy_bcd_val(sptr, nptr, nptr + num->n_len);
+	nptr += num->n_len;
 
 	/* Now the fraction. */
 	if (scale > 0) {
 		*sptr++ = '.';
-		for (index = 0; index < scale && index < num->n_scale; index++) {
-			*sptr++ = BCD_CHAR(*nptr++);
-		}
+		sptr = bc_copy_bcd_val(sptr, nptr, nptr + min_scale);
 		for (index = num->n_scale; index < scale; index++) {
 			*sptr++ = BCD_CHAR(0);
 		}

ext/bcmath/libbcmath/src/str2num.c

@@ -30,6 +30,7 @@
 *************************************************************************/
 
 #include "bcmath.h"
+#include "convert.h"
 #include <stdbool.h>
 #include <stddef.h>
 
@@ -124,24 +125,12 @@ bool bc_str2num(bc_num *num, char *str, size_t scale, bool auto_scale)
 		 * If zero_int is true and the str_scale is 0, there is an early return,
 		 * so here str_scale is always greater than 0.
 		 */
-		while (fractional_ptr < fractional_end) {
-			*nptr = CH_VAL(*fractional_ptr);
-			nptr++;
-			fractional_ptr++;
-		}
+		nptr = bc_copy_ch_val(nptr, fractional_ptr, fractional_end);
 	} else {
 		const char *integer_end = integer_ptr + digits;
-		while (integer_ptr < integer_end) {
-			*nptr = CH_VAL(*integer_ptr);
-			nptr++;
-			integer_ptr++;
-		}
+		nptr = bc_copy_ch_val(nptr, integer_ptr, integer_end);
 		if (str_scale > 0) {
-			while (fractional_ptr < fractional_end) {
-				*nptr = CH_VAL(*fractional_ptr);
-				nptr++;
-				fractional_ptr++;
-			}
+			nptr = bc_copy_ch_val(nptr, fractional_ptr, fractional_end);
 		}
 	}