From 3c6ca49b23fd6a2df735e0eaf93432bfef3cba97 Mon Sep 17 00:00:00 2001
From: Thomas Voss <mail@thomasvoss.com>
Date: Sat, 4 May 2024 20:48:57 +0200
Subject: More 2-stage lookup tables

---
 gen/prop/wb | 207 ++++++++++++++++++++++++++++++++++++------------------------
 1 file changed, 125 insertions(+), 82 deletions(-)

(limited to 'gen/prop/wb')

diff --git a/gen/prop/wb b/gen/prop/wb
index a6b47f2..f6621f5 100755
--- a/gen/prop/wb
+++ b/gen/prop/wb
@@ -1,86 +1,129 @@
-#!/bin/sh
-
-set -e
-cd "${0%/*}/../.."
-exec >lib/unicode/prop/uprop_get_wb.c
-
-gawk '
-BEGIN {
-	FS = " *(; *|#.*)"
-
-	map["ALetter"]            = "LE"
-	map["CR"]                 = "CR"
-	map["Double_Quote"]       = "DQ"
-	map["E_Base"]             = "EB"
-	map["E_Base_GAZ"]         = "EBG"
-	map["E_Modifier"]         = "EM"
-	map["Extend"]             = "EXTEND"
-	map["ExtendNumLet"]       = "EX"
-	map["Format"]             = "FO"
-	map["Glue_After_Zwj"]     = "GAZ"
-	map["Hebrew_Letter"]      = "HL"
-	map["Katakana"]           = "KA"
-	map["LF"]                 = "LF"
-	map["MidLetter"]          = "ML"
-	map["MidNumLet"]          = "MB"
-	map["MidNum"]             = "MN"
-	map["Newline"]            = "NL"
-	map["Numeric"]            = "NU"
-	map["Other"]              = "XX"
-	map["Regional_Indicator"] = "RI"
-	map["Single_Quote"]       = "SQ"
-	map["WSegSpace"]          = "WSEGSPACE"
-	map["ZWJ"]                = "ZWJ"
-
-	print "/* This file is autogenerated by gen/prop/wb; DO NOT EDIT. */"
-	print ""
-	print "#include \"_bsearch.h\""
-	print "#include \"macros.h\""
-	print "#include \"rune.h\""
-	print "#include \"unicode/prop.h\""
-	print ""
-}
+#!/usr/bin/python3
 
-/^[A-F0-9]/ {
-	n = split($1, a, /\.\./)
-	lo = strtonum("0X" a[1])
-	hi = strtonum("0X" a[n])
+import math
 
-	for (i = lo; i <= hi; i++)
-		props[i] = "WB_" map[$2]
-}
+from lib import *
 
-END {
-	print "static constexpr enum uprop_wb lookup_lat1[] = {"
-	for (i = 0; i < 0x100; i++) {
-		if (i % 4 == 0)
-			printf "\t"
-		printf "%-13s%s", (props[i] ? props[i] : "WB_XX") ",", \
-			i % 4 == 3 ? "\n" : " "
-	}
-	print "};"
-	print ""
-	print "static const struct {"
-	print "\trune lo, hi;"
-	print "\tenum uprop_wb val;"
-	print "} lookup[] = {"
-
-	for (i = 0x100; i <= 0x10FFFF; i++) {
-		if (!props[i])
-			continue
-		for (lo = i; props[lo] == props[i + 1]; i++)
-			;
-		printf "\t{RUNE_C(0x%06X), RUNE_C(0x%06X), %s},\n", lo, i, props[i]
-	}
-
-	print "};"
-	print ""
-	print "_MLIB_DEFINE_BSEARCH(enum uprop_wb, lookup, WB_XX)"
-	print ""
-	print "enum uprop_wb"
-	print "uprop_get_wb(rune ch)"
-	print "{"
-	print "\treturn ch < lengthof(lookup_lat1) ? lookup_lat1[ch] : mlib_lookup(ch);"
-	print "}"
+
+MAP = {
+	'ALetter':            'LE',
+	'CR':                 'CR',
+	'Double_Quote':       'DQ',
+	'E_Base':             'EB',
+	'E_Base_GAZ':         'EBG',
+	'E_Modifier':         'EM',
+	'Extend':             'EXTEND',
+	'ExtendNumLet':       'EX',
+	'Format':             'FO',
+	'Glue_After_Zwj':     'GAZ',
+	'Hebrew_Letter':      'HL',
+	'Katakana':           'KA',
+	'LF':                 'LF',
+	'MidLetter':          'ML',
+	'MidNumLet':          'MB',
+	'MidNum':             'MN',
+	'Newline':            'NL',
+	'Numeric':            'NU',
+	'Other':              'XX',
+	'Regional_Indicator': 'RI',
+	'Single_Quote':       'SQ',
+	'WSegSpace':          'WSEGSPACE',
+	'ZWJ':                'ZWJ',
 }
-' data/WordBreakProperty | sed 's/\s*$//'
+
+longest = 0
+
+def parse(file: str) -> list[bool]:
+	global longest
+
+	xs = ['WB_XX'] * 0x110000
+	with open(file, 'r') as f:
+		for line in f.readlines():
+			if len(line.strip()) == 0 or line[0] == '#':
+				continue
+
+			parts = line.split(';')
+			ranges = [int(x, 16) for x in parts[0].strip().split('..')]
+			prop = 'WB_' + MAP[parts[1].split('#')[0].strip()]
+			longest = max(longest, len(prop))
+
+			for i in range(ranges[0], ranges[len(ranges) - 1] + 1):
+				xs[i] = prop
+	return xs
+
+def genfile(cs: list[tuple[bool, ...]], blksize: int) -> None:
+	Cs = cs
+	cs = list(dict.fromkeys(Cs))
+
+	print('''\
+/* This file is autogenerated by gen/prop/wb; DO NOT EDIT. */
+
+#include <stdint.h>
+
+#include "unicode/prop.h"
+''')
+
+	print(f'static constexpr {typename(len(cs) - 1)} stage1[] = {{')
+	for i, c in enumerate(Cs):
+		print(f'%c%{len(str(len(cs) - 1))}d,' % ('\t' if i % 16 == 0 else ' ', cs.index(c)), end='')
+		if i % 16 == 15:
+			print()
+	print('};')
+
+	print()
+
+	ppc = columns(blksize, longest + 1)
+	print(f'static constexpr enum uprop_wb stage2[][{blksize}] = {{')
+	for c in cs:
+		for i in range(blksize // ppc):
+			print('\t{' if i == 0 else '\t ', end='')
+			for j in range(ppc):
+				print(c[i*ppc + j], end='')
+				if i < blksize // ppc - 1 or j < ppc - 1:
+					print(',', end='')
+				if j < ppc - 1:
+					print(' ' * (longest + 1 - len(c[i*ppc + j])), end='')
+			if i < blksize // ppc - 1:
+				print()
+		print('},')
+	print('};')
+
+	print()
+
+	print(f'''\
+enum uprop_wb
+uprop_get_wb(rune ch)
+{{
+	return stage2[stage1[ch / {blksize}]][ch % {blksize}];
+}}''')
+
+def main() -> None:
+	cwd_init()
+	xs = parse('data/WordBreakProperty')
+
+	blksize = -1
+	smallest = math.inf
+
+	for bs in powers_of_2():
+		if bs > len(xs):
+			break
+		Cs = [tuple(x) for x in chunks(xs, bs)]
+		cs = set(Cs)
+
+		sz_s1 = len(Cs) * isize(len(cs) - 1)
+		sz_s2 = len(cs) * bs
+		sz = sz_s1 + sz_s2
+
+		if sz < smallest:
+			smallest = sz
+			blksize = bs
+
+	Cs = [tuple(x) for x in chunks(xs, blksize)]
+	with open('lib/unicode/prop/uprop_get_wb.c', 'w') as f:
+		sys.stdout = f
+		genfile(Cs, blksize)
+
+	report_size(len(xs), smallest)
+
+if __name__ == '__main__':
+	main()
-- 
cgit v1.2.3