Use a branchless check for unicode range

gen/wcwidth.py

@@ -517,8 +517,7 @@ def get_types(sz: int) -> tuple[str, str]:
 
 
 def gen_multistage_table(
-    c: Callable[..., None], g: Callable[..., None], t1: Sequence[int], t2: Sequence[int], t3: Sequence[Property], shift: int,
+    c: Callable[..., None], g: Callable[..., None], t1: Sequence[int], t2: Sequence[int], t3: Sequence[Property], shift: int, input_sz: int
-    for_go_type: str, maxval: int = 0
 ) -> None:
     t1_type_sz = getsize(t1)
     ctype_t1, gotype_t1 = get_types(t1_type_sz)
@@ -554,11 +553,10 @@ def gen_multistage_table(
     g(f'\t{items}')
     g('}')
 
-    check = f'x = max(0, min(x, {maxval}))' if maxval else ''
+    input_type = get_types(getsize((input_sz,)))[1]
     g(f'''
 // Array accessor function that avoids bounds checking
-func {name}For(x {for_go_type}) {name} {{
+func {lname}_for(x {input_type}) {name} {{
-    {check}
 	t1 := uintptr(*(*{gotype_t1})(unsafe.Pointer(uintptr(unsafe.Pointer(&{lname}_t1[0])) + uintptr(x>>{lname}_shift)*{t1_type_sz})))
     t1_shifted := (t1 << {lname}_shift) + (uintptr(x) & {lname}_mask)
 	t2 := uintptr(*(*{gotype_t2})(unsafe.Pointer(uintptr(unsafe.Pointer(&{lname}_t2[0])) + t1_shifted*{t2_type_sz})))
@@ -1146,6 +1144,8 @@ def gen_char_props() -> None:
         gp = partial(print, file=gof)
         gp('package wcswidth')
         gp('import "unsafe"')
+        gp(f'const MAX_UNICODE = {sys.maxunicode}')
+        gp(f'const UNICODE_LIMIT = {sys.maxunicode + 1}')
         generate_enum(c, gp, 'GraphemeBreakProperty', *grapheme_segmentation_maps, prefix='GBP_')
         generate_enum(c, gp, 'IndicConjunctBreak', *incb_map, prefix='ICB_')
         cen('// UCBDeclaration {{''{')
@@ -1156,8 +1156,8 @@ def gen_char_props() -> None:
         gp(make_bitfield('tools/wcswidth', 'GraphemeSegmentationResult', *GraphemeSegmentationResult.go_fields(), add_package=False)[1])
         gp(CharProps.go_extra())
         gp(GraphemeSegmentationResult.go_extra())
-        gen_multistage_table(c, gp, t1, t2, t3, t_shift, 'rune', sys.maxunicode)
+        gen_multistage_table(c, gp, t1, t2, t3, t_shift, len(prop_array)-1)
-        gen_multistage_table(c, gp, g1, g2, g3, g_shift, 'uint16')
+        gen_multistage_table(c, gp, g1, g2, g3, g_shift, len(gseg_results)-1)
         c(GraphemeSegmentationKey.code_to_convert_to_int())
         c(GraphemeSegmentationState.c_declaration())
         gp(GraphemeSegmentationKey.code_to_convert_to_int(for_go=True))

kitty/char-props.c

@@ -8,10 +8,18 @@
 #include "char-props.h"
 #include "char-props-data.h"
 
+static char_type
+ensure_char_in_range(const char_type value) {
+    // Branchless: if (value > MAX_UNICODE) value = 0
+    const int64_t diff = ((int64_t)value) - ((int64_t)(MAX_UNICODE + 1u));
+    // The right shift gives all ones for negative diff and all zeros for positive diff
+    const char_type mask = diff >> 63;
+    return value & mask;
+}
 
 CharProps
 char_props_for(char_type ch) {
-    if (ch > MAX_UNICODE) ch = 0;
+    ch = ensure_char_in_range(ch);
     return CharProps_t3[CharProps_t2[(CharProps_t1[ch >> CharProps_shift] << CharProps_shift) + (ch & CharProps_mask)]];
 }
 

tools/wcswidth/char-props-data.go

@@ -2,6 +2,9 @@ package wcswidth
 
 import "unsafe"
 
+const MAX_UNICODE = 1114111
+const UNICODE_LIMIT = 1114112
+
 type GraphemeBreakProperty uint8
 
 const (
@@ -387,8 +390,7 @@ var charprops_t3 = [109]CharProps{
 }
 
 // Array accessor function that avoids bounds checking
-func CharPropsFor(x rune) CharProps {
+func charprops_for(x uint32) CharProps {
-	x = max(0, min(x, 1114111))
 	t1 := uintptr(*(*uint8)(unsafe.Pointer(uintptr(unsafe.Pointer(&charprops_t1[0])) + uintptr(x>>charprops_shift)*1)))
 	t1_shifted := (t1 << charprops_shift) + (uintptr(x) & charprops_mask)
 	t2 := uintptr(*(*uint8)(unsafe.Pointer(uintptr(unsafe.Pointer(&charprops_t2[0])) + t1_shifted*1)))
@@ -1038,8 +1040,7 @@ var graphemesegmentationresult_t3 = [630]GraphemeSegmentationResult{
 }
 
 // Array accessor function that avoids bounds checking
-func GraphemeSegmentationResultFor(x uint16) GraphemeSegmentationResult {
+func graphemesegmentationresult_for(x uint16) GraphemeSegmentationResult {
-
 	t1 := uintptr(*(*uint8)(unsafe.Pointer(uintptr(unsafe.Pointer(&graphemesegmentationresult_t1[0])) + uintptr(x>>graphemesegmentationresult_shift)*1)))
 	t1_shifted := (t1 << graphemesegmentationresult_shift) + (uintptr(x) & graphemesegmentationresult_mask)
 	t2 := uintptr(*(*uint16)(unsafe.Pointer(uintptr(unsafe.Pointer(&graphemesegmentationresult_t2[0])) + t1_shifted*2)))

tools/wcswidth/char-props.go

@@ -7,6 +7,19 @@ import (
 
 var _ = fmt.Print
 
+func ensure_char_in_range(value uint32) uint32 {
+	// Branchless: if (value > MAX_UNICODE) value = 0
+	diff := int64(value) - UNICODE_LIMIT
+	// The right shift gives all ones for negative diff and all zeros for positive diff
+	mask := uint32(diff >> 63)
+	return value & mask
+}
+
+func CharPropsFor(ch rune) CharProps {
+	q := ensure_char_in_range(uint32(ch))
+	return charprops_for(q)
+}
+
 func IteratorOverGraphemes(text string) iter.Seq[string] {
 	var s GraphemeSegmentationResult
 	start_pos := 0
@@ -39,7 +52,7 @@ func (s *GraphemeSegmentationResult) Reset() {
 
 func (s GraphemeSegmentationResult) Step(ch CharProps) GraphemeSegmentationResult {
 	key := grapheme_segmentation_key(s, ch)
-	return GraphemeSegmentationResultFor(key)
+	return graphemesegmentationresult_for(key)
 }
 
 func Runewidth(code rune) int {