From fc8e3d09717eaac3f2854640f91edfe55ede923f Mon Sep 17 00:00:00 2001
From: Delyan Angelov <delian66@gmail.com>
Date: Thu, 8 Sep 2022 11:09:13 +0300
Subject: [PATCH] builtin,strconv: speedup `str.int()` conversions (without
 -prod)

---
 vlib/builtin/string.v                    |  2 +
 vlib/strconv/atof.c.v                    |  5 ++-
 vlib/strconv/atofq.c.v                   |  1 +
 vlib/strconv/atoi.v                      | 49 +++++++++++++-----------
 vlib/strconv/number_to_base.c.v          |  2 +-
 vlib/strconv/utilities.c.v               |  4 +-
 vlib/strconv/vprintf.c.v                 |  4 +-
 vlib/strings/strings.c.v                 |  2 +
 vlib/v/tests/bench/bench_string_int.v    | 17 ++++++++
 vlib/v/tests/big_array_allocation_test.v |  7 ++++
 10 files changed, 63 insertions(+), 30 deletions(-)
 create mode 100644 vlib/v/tests/bench/bench_string_int.v

diff --git a/vlib/builtin/string.v b/vlib/builtin/string.v
index b2bf9c1946..da1b25eb2d 100644
--- a/vlib/builtin/string.v
+++ b/vlib/builtin/string.v
@@ -1818,6 +1818,7 @@ pub fn (s []string) join_lines() string {
 
 // reverse returns a reversed string.
 // Example: assert 'Hello V'.reverse() == 'V olleH'
+[direct_array_access]
 pub fn (s string) reverse() string {
 	if s.len == 0 || s.len == 1 {
 		return s.clone()
@@ -1870,6 +1871,7 @@ pub fn (s string) bytes() []u8 {
 }
 
 // repeat returns a new string with `count` number of copies of the string it was called on.
+[direct_array_access]
 pub fn (s string) repeat(count int) string {
 	if count < 0 {
 		panic('string.repeat: count is negative: $count')
diff --git a/vlib/strconv/atof.c.v b/vlib/strconv/atof.c.v
index c51bd3f294..14241cdcd3 100644
--- a/vlib/strconv/atof.c.v
+++ b/vlib/strconv/atof.c.v
@@ -101,7 +101,7 @@ fn sub96(s2 u32, s1 u32, s0 u32, d2 u32, d1 u32, d0 u32) (u32, u32, u32) {
 
 // Utility functions
 fn is_digit(x u8) bool {
-	return (x >= strconv.c_zero && x <= strconv.c_nine) == true
+	return x >= strconv.c_zero && x <= strconv.c_nine
 }
 
 fn is_space(x u8) bool {
@@ -109,7 +109,7 @@ fn is_space(x u8) bool {
 }
 
 fn is_exp(x u8) bool {
-	return (x == `E` || x == `e`) == true
+	return x == `E` || x == `e`
 }
 
 // Possible parser return values.
@@ -124,6 +124,7 @@ enum ParserState {
 
 // parser tries to parse the given string into a number
 // NOTE: #TOFIX need one char after the last char of the number
+[direct_array_access]
 fn parser(s string) (ParserState, PrepNumber) {
 	mut digx := 0
 	mut result := ParserState.ok
diff --git a/vlib/strconv/atofq.c.v b/vlib/strconv/atofq.c.v
index 07bb7b33bd..d7b2b5e6a0 100644
--- a/vlib/strconv/atofq.c.v
+++ b/vlib/strconv/atofq.c.v
@@ -16,6 +16,7 @@ Know limitation:
 */
 
 // atof_quick return a f64 number from a string in a quick way
+[direct_array_access]
 pub fn atof_quick(s string) f64 {
 	mut f := Float64u{} // result
 	mut sign := f64(1.0) // result sign
diff --git a/vlib/strconv/atoi.v b/vlib/strconv/atoi.v
index 2babb2c1a2..6418d1989d 100644
--- a/vlib/strconv/atoi.v
+++ b/vlib/strconv/atoi.v
@@ -12,8 +12,9 @@ const (
 	max_u64  = u64(18446744073709551615) // as u64 // use this until we add support
 )
 
+[inline]
 pub fn byte_to_lower(c u8) u8 {
-	return c | (`x` - `X`)
+	return c | 32
 }
 
 // common_parse_uint is called by parse_uint and allows the parsing
@@ -34,14 +35,14 @@ pub fn common_parse_uint(s string, _base int, _bit_size int, error_on_non_digit
 
 // the first returned value contains the parsed value,
 // the second returned value contains the error code (0 = OK, >1 = index of first non-parseable character + 1, -1 = wrong base, -2 = wrong bit size, -3 = overflow)
+[direct_array_access]
 pub fn common_parse_uint2(s string, _base int, _bit_size int) (u64, int) {
-	mut bit_size := _bit_size
-	mut base := _base
 	if s.len < 1 || !underscore_ok(s) {
 		// return error('parse_uint: syntax error $s')
 		return u64(0), 1
 	}
-	base0 := base == 0
+	mut bit_size := _bit_size
+	mut base := _base
 	mut start_index := 0
 	if 2 <= base && base <= 36 {
 		// valid base; nothing to do
@@ -49,13 +50,13 @@ pub fn common_parse_uint2(s string, _base int, _bit_size int) (u64, int) {
 		// Look for octal, hex prefix.
 		base = 10
 		if s[0] == `0` {
-			if s.len >= 3 && byte_to_lower(s[1]) == `b` {
+			if s.len >= 3 && s[1] | 32 == `b` {
 				base = 2
 				start_index += 2
-			} else if s.len >= 3 && byte_to_lower(s[1]) == `o` {
+			} else if s.len >= 3 && s[1] | 32 == `o` {
 				base = 8
 				start_index += 2
-			} else if s.len >= 3 && byte_to_lower(s[1]) == `x` {
+			} else if s.len >= 3 && s[1] | 32 == `x` {
 				base = 16
 				start_index += 2
 			}
@@ -85,10 +86,10 @@ pub fn common_parse_uint2(s string, _base int, _bit_size int) (u64, int) {
 	mut n := u64(0)
 	for i in start_index .. s.len {
 		c := s[i]
-		cl := byte_to_lower(c)
+		cl := c | 32
 
 		mut d := u8(0)
-		if c == `_` && base0 {
+		if c == `_` && _base == 0 {
 			// underscore_ok already called
 			continue
 		} else if `0` <= c && c <= `9` {
@@ -125,13 +126,17 @@ pub fn parse_uint(s string, _base int, _bit_size int) ?u64 {
 
 // common_parse_int is called by parse int and allows the parsing
 // to stop on non or invalid digit characters and return with an error
+[direct_array_access]
 pub fn common_parse_int(_s string, base int, _bit_size int, error_on_non_digit bool, error_on_high_digit bool) ?i64 {
-	mut s := _s
-	mut bit_size := _bit_size
-	if s.len < 1 {
+	if _s.len < 1 {
 		// return error('parse_int: syntax error $s')
 		return i64(0)
 	}
+	mut bit_size := _bit_size
+	if bit_size == 0 {
+		bit_size = strconv.int_size
+	}
+	mut s := _s
 	// Pick off leading sign.
 	mut neg := false
 	if s[0] == `+` {
@@ -148,9 +153,6 @@ pub fn common_parse_int(_s string, base int, _bit_size int, error_on_non_digit b
 	if un == 0 {
 		return i64(0)
 	}
-	if bit_size == 0 {
-		bit_size = strconv.int_size
-	}
 	// TODO: check should u64(bit_size-1) be size of int (32)?
 	cutoff := u64(1) << u64(bit_size - 1)
 	if !neg && un >= cutoff {
@@ -181,9 +183,10 @@ pub fn parse_int(_s string, base int, _bit_size int) ?i64 {
 }
 
 // atoi is equivalent to parse_int(s, 10, 0), converted to type int.
+[direct_array_access]
 pub fn atoi(s string) ?int {
 	if s == '' {
-		return error('strconv.atoi: parsing "$s": invalid syntax ')
+		return error('strconv.atoi: parsing "": invalid syntax')
 	}
 	if (strconv.int_size == 32 && (0 < s.len && s.len < 10))
 		|| (strconv.int_size == 64 && (0 < s.len && s.len < 19)) {
@@ -193,7 +196,7 @@ pub fn atoi(s string) ?int {
 			start_idx++
 			if s.len - start_idx < 1 {
 				// return 0, &NumError{fnAtoi, s0, ErrSyntax}
-				return error('strconv.atoi: parsing "$s": invalid syntax ')
+				return error('strconv.atoi: parsing "$s": invalid syntax')
 			}
 		}
 		mut n := 0
@@ -201,7 +204,7 @@ pub fn atoi(s string) ?int {
 			ch := s[i] - `0`
 			if ch > 9 {
 				// return 0, &NumError{fnAtoi, s0, ErrSyntax}
-				return error('strconv.atoi: parsing "$s": invalid syntax ')
+				return error('strconv.atoi: parsing "$s": invalid syntax')
 			}
 			n = n * 10 + int(ch)
 		}
@@ -215,6 +218,7 @@ pub fn atoi(s string) ?int {
 // underscore_ok reports whether the underscores in s are allowed.
 // Checking them in this one function lets all the parsers skip over them simply.
 // Underscore must appear only between digits or between a base prefix and a digit.
+[direct_array_access]
 fn underscore_ok(s string) bool {
 	// saw tracks the last character (class) we saw:
 	// ^ for beginning of number,
@@ -229,17 +233,16 @@ fn underscore_ok(s string) bool {
 	}
 	// Optional base prefix.
 	mut hex := false
-	if s.len - i >= 2 && s[i] == `0` && (byte_to_lower(s[i + 1]) == `b`
-		|| byte_to_lower(s[i + 1]) == `o` || byte_to_lower(s[i + 1]) == `x`) {
+	if (s.len - i >= 2) && (s[i] == `0`) && (((s[i + 1] | 32) == `b`)
+		|| ((s[i + 1] | 32) == `o`) || ((s[i + 1] | 32) == `x`)) {
 		saw = `0` // base prefix counts as a digit for "underscore as digit separator"
-		hex = byte_to_lower(s[i + 1]) == `x`
+		hex = (s[i + 1] | 32) == `x`
 		i += 2
 	}
 	// Number proper.
 	for ; i < s.len; i++ {
 		// Digits are always okay.
-		if (`0` <= s[i] && s[i] <= `9`) || (hex && `a` <= byte_to_lower(s[i])
-			&& byte_to_lower(s[i]) <= `f`) {
+		if (`0` <= s[i] && s[i] <= `9`) || ((hex && `a` <= (s[i] | 32)) && ((s[i] | 32) <= `f`)) {
 			saw = `0`
 			continue
 		}
diff --git a/vlib/strconv/number_to_base.c.v b/vlib/strconv/number_to_base.c.v
index afce53be79..69409f7e21 100644
--- a/vlib/strconv/number_to_base.c.v
+++ b/vlib/strconv/number_to_base.c.v
@@ -41,7 +41,7 @@ pub fn format_int(n i64, radix int) string {
 
 // format_uint returns the string representation of the number n in base `radix`
 // for digit values > 10, this function uses the small latin leters a-z.
-[manualfree]
+[direct_array_access; manualfree]
 pub fn format_uint(n u64, radix int) string {
 	unsafe {
 		if radix < 2 || radix > 36 {
diff --git a/vlib/strconv/utilities.c.v b/vlib/strconv/utilities.c.v
index 5cdfbf153d..d2c33a3a0f 100644
--- a/vlib/strconv/utilities.c.v
+++ b/vlib/strconv/utilities.c.v
@@ -75,7 +75,7 @@ pub fn f64_to_str_l_no_dot(f f64) string {
 // floating-point `string` in scientific notation.
 //
 // Example: assert strconv.fxx_to_str_l_parse('34.22e+00') == '34.22'
-[manualfree]
+[direct_array_access; manualfree]
 pub fn fxx_to_str_l_parse(s string) string {
 	// check for +inf -inf Nan
 	if s.len > 2 && (s[0] == `n` || s[1] == `i`) {
@@ -202,7 +202,7 @@ pub fn fxx_to_str_l_parse(s string) string {
 // The decimal digits after the dot can be omitted.
 //
 // Example: assert strconv.fxx_to_str_l_parse_no_dot ('34.e+01') == '340'
-[manualfree]
+[direct_array_access; manualfree]
 pub fn fxx_to_str_l_parse_no_dot(s string) string {
 	// check for +inf -inf Nan
 	if s.len > 2 && (s[0] == `n` || s[1] == `i`) {
diff --git a/vlib/strconv/vprintf.c.v b/vlib/strconv/vprintf.c.v
index dbea0e1490..85b68a2fcb 100644
--- a/vlib/strconv/vprintf.c.v
+++ b/vlib/strconv/vprintf.c.v
@@ -36,7 +36,7 @@ pub fn v_printf(str string, pt ...voidptr) {
 // assert strconv.v_sprintf('aaa %G', x) == 'aaa 3.141516'
 // ```
 [deprecated: 'use string interpolation instead']
-[manualfree]
+[direct_array_access; manualfree]
 pub fn v_sprintf(str string, pt ...voidptr) string {
 	mut res := strings.new_builder(pt.len * 16)
 	defer {
@@ -560,7 +560,7 @@ fn fabs(x f64) f64 {
 }
 
 // strings.Builder version of format_fl
-[manualfree]
+[direct_array_access; manualfree]
 pub fn format_fl_old(f f64, p BF_param) string {
 	unsafe {
 		mut s := ''
diff --git a/vlib/strings/strings.c.v b/vlib/strings/strings.c.v
index b475786ba9..ba14afcd74 100644
--- a/vlib/strings/strings.c.v
+++ b/vlib/strings/strings.c.v
@@ -1,6 +1,7 @@
 module strings
 
 // strings.repeat - fill a string with `n` repetitions of the character `c`
+[direct_array_access]
 pub fn repeat(c u8, n int) string {
 	if n <= 0 {
 		return ''
@@ -16,6 +17,7 @@ pub fn repeat(c u8, n int) string {
 // strings.repeat_string - gives you `n` repetitions of the substring `s`
 // Note: strings.repeat, that repeats a single byte, is between 2x
 // and 24x faster than strings.repeat_string called for a 1 char string.
+[direct_array_access]
 pub fn repeat_string(s string, n int) string {
 	if n <= 0 || s.len == 0 {
 		return ''
diff --git a/vlib/v/tests/bench/bench_string_int.v b/vlib/v/tests/bench/bench_string_int.v
new file mode 100644
index 0000000000..0b780012d0
--- /dev/null
+++ b/vlib/v/tests/bench/bench_string_int.v
@@ -0,0 +1,17 @@
+import benchmark
+
+const maxn = 999_999
+
+fn main() {
+	mut snumbers := []string{cap: maxn}
+	for i in 0 .. maxn {
+		snumbers << i.str()
+	}
+	mut sum := i64(0)
+	mut bmark := benchmark.start()
+	for s in snumbers {
+		sum += s.int()
+	}
+	bmark.measure('s.int()')
+	dump(sum)
+}
diff --git a/vlib/v/tests/big_array_allocation_test.v b/vlib/v/tests/big_array_allocation_test.v
index 1a6e5a1581..fc34a6f89e 100644
--- a/vlib/v/tests/big_array_allocation_test.v
+++ b/vlib/v/tests/big_array_allocation_test.v
@@ -1,3 +1,5 @@
+import os
+
 [direct_array_access]
 fn test_big_int_array() {
 	dump(sizeof(isize))
@@ -5,6 +7,11 @@ fn test_big_int_array() {
 	if sizeof(isize) > 4 {
 		maxn = 1_000_000_000 // 1 billion integers, when each is 4 bytes => require ~4GB
 	}
+	// NB: this test requires RAM that many people do not have, so only run it in full, when VTEST_BIGMEM is 1
+	vtest_bigmem := os.getenv('VTEST_BIGMEM').int()
+	if vtest_bigmem == 0 {
+		maxn = 10_000_000
+	}
 	dump(maxn)
 	mut data := []int{len: maxn}