Begin optimizing tokenizer; Replace `gb_utf8_decode` with `utf8_decode` (CC but easier to change later)

9 files changed, 169 insertions, 44 deletions

diff --git a/src/checker.cpp b/src/checker.cpp
index de69bd854..f81017c38 100644
--- a/src/checker.cpp
+++ b/src/checker.cpp
@@ -3512,7 +3512,7 @@ bool is_string_an_identifier(String s) {
 	while (offset < s.len) {
 		bool ok = false;
 		Rune r = -1;
-		isize size = gb_utf8_decode(s.text+offset, s.len-offset, &r);
+		isize size = utf8_decode(s.text+offset, s.len-offset, &r);
 		if (offset == 0) {
 			ok = rune_is_letter(r);
 		} else {
diff --git a/src/exact_value.cpp b/src/exact_value.cpp
index cd5ba6579..c8e1ae99f 100644
--- a/src/exact_value.cpp
+++ b/src/exact_value.cpp
@@ -317,7 +317,7 @@ ExactValue exact_value_from_basic_literal(Token token) {
 	}
 	case Token_Rune: {
 		Rune r = GB_RUNE_INVALID;
-		gb_utf8_decode(token.string.text, token.string.len, &r);
+		utf8_decode(token.string.text, token.string.len, &r);
 		return exact_value_i64(r);
 	}
 	default:
diff --git a/src/gb/gb.h b/src/gb/gb.h
index 187b6628b..848ad8afc 100644
--- a/src/gb/gb.h
+++ b/src/gb/gb.h
@@ -8232,7 +8232,9 @@ gbFileContents gb_file_read_contents(gbAllocator a, b32 zero_terminate, char con
 	if (gb_file_open(&file, filepath) == gbFileError_None) {
 		isize file_size = cast(isize)gb_file_size(&file);
 		if (file_size > 0) {
-			result.data = gb_alloc(a, zero_terminate ? file_size+1 : file_size);
+			isize total_size = file_size + !!zero_terminate;
+			total_size = (total_size+7)&~7;
+			result.data = gb_alloc(a, total_size);
 			result.size = file_size;
 			gb_file_read_at(&file, result.data, result.size, 0);
 			if (zero_terminate) {
diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp
index a59e1baa9..743d465e8 100644
--- a/src/llvm_backend.cpp
+++ b/src/llvm_backend.cpp
@@ -4712,7 +4712,7 @@ void lb_build_unroll_range_stmt(lbProcedure *p, AstUnrollRangeStmt *rs, Scope *s
 				Rune codepoint = 0;
 				isize offset = 0;
 				do {
-					isize width = gb_utf8_decode(str.text+offset, str.len-offset, &codepoint);
+					isize width = utf8_decode(str.text+offset, str.len-offset, &codepoint);
 					if (val0_type) lb_addr_store(p, val0_addr, lb_const_value(m, val0_type, exact_value_i64(codepoint)));
 					if (val1_type) lb_addr_store(p, val1_addr, lb_const_value(m, val1_type, exact_value_i64(offset)));
 					lb_build_stmt(p, rs->body);
@@ -6563,7 +6563,7 @@ lbValue lb_const_value(lbModule *m, Type *type, ExactValue value, bool allow_loc
 			LLVMValueRef *elems = gb_alloc_array(permanent_allocator(), LLVMValueRef, cast(isize)count);
 
 			for (i64 i = 0; i < count && offset < s.len; i++) {
-				width = gb_utf8_decode(s.text+offset, s.len-offset, &rune);
+				width = utf8_decode(s.text+offset, s.len-offset, &rune);
 				offset += width;
 
 				elems[i] = LLVMConstInt(et, rune, true);
diff --git a/src/main.cpp b/src/main.cpp
index 27d69d428..3b3cb49d3 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -561,7 +561,7 @@ bool string_is_valid_identifier(String str) {
 	isize offset = 0;
 	while (offset < str.len) {
 		Rune r = 0;
-		w = gb_utf8_decode(str.text, str.len, &r);
+		w = utf8_decode(str.text, str.len, &r);
 		if (r == GB_RUNE_INVALID) {
 			return false;
 		}
diff --git a/src/parser.cpp b/src/parser.cpp
index f20c29558..911082b01 100644
--- a/src/parser.cpp
+++ b/src/parser.cpp
@@ -1705,7 +1705,7 @@ bool is_foreign_name_valid(String name) {
 	while (offset < name.len) {
 		Rune rune;
 		isize remaining = name.len - offset;
-		isize width = gb_utf8_decode(name.text+offset, remaining, &rune);
+		isize width = utf8_decode(name.text+offset, remaining, &rune);
 		if (rune == GB_RUNE_INVALID && width == 1) {
 			return false;
 		} else if (rune == GB_RUNE_BOM && remaining > 0) {
@@ -4612,7 +4612,7 @@ ParseFileError init_ast_file(AstFile *f, String fullpath, TokenPos *err_pos) {
 
 	u64 start = time_stamp_time_now();
 
-	while (f->curr_token.kind != Token_EOF) {
+	for (;;) {
 		Token *token = array_add_and_get(&f->tokens);
 		tokenizer_get_token(&f->tokenizer, token);
 		if (token->kind == Token_Invalid) {
@@ -4887,7 +4887,7 @@ bool is_import_path_valid(String path) {
 			isize width = 1;
 			Rune r = *curr;
 			if (r >= 0x80) {
-				width = gb_utf8_decode(curr, end-curr, &r);
+				width = utf8_decode(curr, end-curr, &r);
 				if (r == GB_RUNE_INVALID && width == 1) {
 					return false;
 				}
@@ -4920,7 +4920,7 @@ bool is_build_flag_path_valid(String path) {
 			isize width = 1;
 			Rune r = *curr;
 			if (r >= 0x80) {
-				width = gb_utf8_decode(curr, end-curr, &r);
+				width = utf8_decode(curr, end-curr, &r);
 				if (r == GB_RUNE_INVALID && width == 1) {
 					return false;
 				}
@@ -5170,7 +5170,7 @@ String build_tag_get_token(String s, String *out) {
 	isize n = 0;
 	while (n < s.len) {
 		Rune rune = 0;
-		isize width = gb_utf8_decode(&s[n], s.len-n, &rune);
+		isize width = utf8_decode(&s[n], s.len-n, &rune);
 		if (n == 0 && rune == '!') {
 
 		} else if (!rune_is_letter(rune) && !rune_is_digit(rune)) {
diff --git a/src/string.cpp b/src/string.cpp
index 4e8273f60..0764009bc 100644
--- a/src/string.cpp
+++ b/src/string.cpp
@@ -500,7 +500,7 @@ String quote_to_ascii(gbAllocator a, String str, u8 quote='"') {
 		Rune r = cast(Rune)s[0];
 		width = 1;
 		if (r >= 0x80) {
-			width = gb_utf8_decode(s, n, &r);
+			width = utf8_decode(s, n, &r);
 		}
 		if (width == 1 && r == GB_RUNE_INVALID) {
 			array_add(&buf, cast(u8)'\\');
@@ -576,7 +576,7 @@ bool unquote_char(String s, u8 quote, Rune *rune, bool *multiple_bytes, String *
 		return false;
 	} else if (s[0] >= 0x80) {
 		Rune r = -1;
-		isize size = gb_utf8_decode(s.text, s.len, &r);
+		isize size = utf8_decode(s.text, s.len, &r);
 		*rune = r;
 		*multiple_bytes = true;
 		*tail_string = make_string(s.text+size, s.len-size);
@@ -736,7 +736,7 @@ i32 unquote_string(gbAllocator a, String *s_, u8 quote=0, bool has_carriage_retu
 			return 1;
 		} else if (quote == '\'') {
 			Rune r = GB_RUNE_INVALID;
-			isize size = gb_utf8_decode(s.text, s.len, &r);
+			isize size = utf8_decode(s.text, s.len, &r);
 			if ((size == s.len) && (r != -1 || size != 1)) {
 				*s_ = s;
 				return 1;
diff --git a/src/tokenizer.cpp b/src/tokenizer.cpp
index a073abc37..d375ca05d 100644
--- a/src/tokenizer.cpp
+++ b/src/tokenizer.cpp
@@ -789,26 +789,27 @@ void tokenizer_err(Tokenizer *t, TokenPos const &pos, char const *msg, ...) {
 
 void advance_to_next_rune(Tokenizer *t) {
 	if (t->read_curr < t->end) {
-		Rune rune;
-		isize width = 1;
-
 		t->curr = t->read_curr;
 		if (t->curr_rune == '\n') {
 			t->line = t->curr;
 			t->line_count++;
 		}
-		rune = *t->read_curr;
+
+		Rune rune = *t->read_curr;
 		if (rune == 0) {
 			tokenizer_err(t, "Illegal character NUL");
-		} else if (rune >= 0x80) { // not ASCII
-			width = gb_utf8_decode(t->read_curr, t->end-t->read_curr, &rune);
+			t->read_curr++;
+		} else if (rune & 0x80) { // not ASCII
+			isize width = utf8_decode(t->read_curr, t->end-t->read_curr, &rune);
+			t->read_curr += width;
 			if (rune == GB_RUNE_INVALID && width == 1) {
 				tokenizer_err(t, "Illegal UTF-8 encoding");
 			} else if (rune == GB_RUNE_BOM && t->curr-t->start > 0){
 				tokenizer_err(t, "Illegal byte order mark");
 			}
+		} else {
+			t->read_curr++;
 		}
-		t->read_curr += width;
 		t->curr_rune = rune;
 	} else {
 		t->curr = t->end;
@@ -820,7 +821,28 @@ void advance_to_next_rune(Tokenizer *t) {
 	}
 }
 
-TokenizerInitError init_tokenizer(Tokenizer *t, String fullpath, TokenizerFlags flags = TokenizerFlag_None) {
+void init_tokenizer_with_file_contents(Tokenizer *t, String const &fullpath, gbFileContents *fc, TokenizerFlags flags) {
+	t->flags = flags;
+	t->fullpath = fullpath;
+	t->line_count = 1;
+
+	t->start = cast(u8 *)fc->data;
+	t->line = t->read_curr = t->curr = t->start;
+	t->end = t->start + fc->size;
+
+	advance_to_next_rune(t);
+	if (t->curr_rune == GB_RUNE_BOM) {
+		advance_to_next_rune(t); // Ignore BOM at file beginning
+	}
+
+	if (t->allocated_strings.count != 0) {
+		array_clear(&t->allocated_strings);
+	} else {
+		array_init(&t->allocated_strings, heap_allocator());
+	}
+}
+
+TokenizerInitError init_tokenizer(Tokenizer *t, String const &fullpath, TokenizerFlags flags = TokenizerFlag_None) {
 	TokenizerInitError err = TokenizerInit_None;
 
 	char *c_str = alloc_cstring(heap_allocator(), fullpath);
@@ -829,25 +851,18 @@ TokenizerInitError init_tokenizer(Tokenizer *t, String fullpath, TokenizerFlags
 	// TODO(bill): Memory map rather than copy contents
 	gbFileContents fc = gb_file_read_contents(heap_allocator(), true, c_str);
 
-	t->flags = flags;
-	t->fullpath = fullpath;
-	t->line_count = 1;
-
 	if (fc.size > I32_MAX) {
+		t->flags = flags;
+		t->fullpath = fullpath;
+		t->line_count = 1;
 		err = TokenizerInit_FileTooLarge;
 		gb_file_free_contents(&fc);
 	} else if (fc.data != nullptr) {
-		t->start = cast(u8 *)fc.data;
-		t->line = t->read_curr = t->curr = t->start;
-		t->end = t->start + fc.size;
-
-		advance_to_next_rune(t);
-		if (t->curr_rune == GB_RUNE_BOM) {
-			advance_to_next_rune(t); // Ignore BOM at file beginning
-		}
-
-		array_init(&t->allocated_strings, heap_allocator());
+		init_tokenizer_with_file_contents(t, fullpath, &fc, flags);
 	} else {
+		t->flags = flags;
+		t->fullpath = fullpath;
+		t->line_count = 1;
 		gbFile f = {};
 		gbFileError file_err = gb_file_open(&f, c_str);
 		defer (gb_file_close(&f));
@@ -1093,8 +1108,24 @@ bool scan_escape(Tokenizer *t) {
 }
 
 
-void tokenizer_get_token(Tokenizer *t, Token *token, int repeat=0) {
+gb_inline void tokenizer_skip_line(Tokenizer *t) {
+#if 0
+	while (t->curr_rune != '\n' && t->curr_rune != GB_RUNE_EOF) {
+		advance_to_next_rune(t);
+	}
+#else
+	while (t->read_curr != t->end && t->curr_rune != '\n' && t->curr_rune != GB_RUNE_EOF) {
+		t->curr = t->read_curr;
+		t->curr_rune = *t->read_curr;
+		if (t->curr_rune == 0) {
+			tokenizer_err(t, "Illegal character NUL");
+		}
+		t->read_curr++;
+	}
+#endif
+}
 
+void tokenizer_get_token(Tokenizer *t, Token *token, int repeat=0) {
 	// Skip whitespace
 	if (t->flags & TokenizerFlag_InsertSemicolon && t->insert_semicolon) {
 		for (;;) {
@@ -1405,10 +1436,7 @@ void tokenizer_get_token(Tokenizer *t, Token *token, int repeat=0) {
 			token->kind = Token_Hash;
 			if (t->curr_rune == '!') {
 				token->kind = Token_Comment;
-
-				while (t->curr_rune != '\n' && t->curr_rune != GB_RUNE_EOF) {
-					advance_to_next_rune(t);
-				}
+				tokenizer_skip_line(t);
 			}
 			break;
 		case '/':
@@ -1416,9 +1444,7 @@ void tokenizer_get_token(Tokenizer *t, Token *token, int repeat=0) {
 			switch (t->curr_rune) {
 			case '/':
 				token->kind = Token_Comment;
-				while (t->curr_rune != '\n' && t->curr_rune != GB_RUNE_EOF) {
-					advance_to_next_rune(t);
-				}
+				tokenizer_skip_line(t);
 				break;
 			case '*':
 				token->kind = Token_Comment;
diff --git a/src/unicode.cpp b/src/unicode.cpp
index 83aa8deef..e33fb793b 100644
--- a/src/unicode.cpp
+++ b/src/unicode.cpp
@@ -65,3 +65,100 @@ bool rune_is_whitespace(Rune r) {
 	}
 	return false;
 }
+
+
+gb_global u8 const global__utf8_first[256] = {
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x00-0x0F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x10-0x1F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x20-0x2F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x30-0x3F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x40-0x4F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x50-0x5F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x60-0x6F
+	0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x70-0x7F
+	0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, // 0x80-0x8F
+	0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, // 0x90-0x9F
+	0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, // 0xA0-0xAF
+	0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, // 0xB0-0xBF
+	0xf1, 0xf1, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, // 0xC0-0xCF
+	0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, // 0xD0-0xDF
+	0x13, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x23, 0x03, 0x03, // 0xE0-0xEF
+	0x34, 0x04, 0x04, 0x04, 0x44, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, // 0xF0-0xFF
+};
+
+typedef struct Utf8AcceptRange {
+	u8 lo, hi;
+} Utf8AcceptRange;
+
+gb_global Utf8AcceptRange const global__utf8_accept_ranges[] = {
+	{0x80, 0xbf},
+	{0xa0, 0xbf},
+	{0x80, 0x9f},
+	{0x90, 0xbf},
+	{0x80, 0x8f},
+};
+
+
+isize utf8_decode(u8 const *str, isize str_len, Rune *codepoint_out) {
+	isize width = 0;
+	Rune codepoint = GB_RUNE_INVALID;
+
+	if (str_len > 0) {
+		u8 s0 = str[0];
+		u8 x = global__utf8_first[s0], sz;
+		u8 b1, b2, b3;
+		Utf8AcceptRange accept;
+		if (x >= 0xf0) {
+			Rune mask = (cast(Rune)x << 31) >> 31;
+			codepoint = (cast(Rune)s0 & (~mask)) | (GB_RUNE_INVALID & mask);
+			width = 1;
+			goto end;
+		}
+		if (s0 < 0x80) {
+			codepoint = s0;
+			width = 1;
+			goto end;
+		}
+
+		sz = x&7;
+		accept = global__utf8_accept_ranges[x>>4];
+		if (str_len < gb_size_of(sz))
+			goto invalid_codepoint;
+
+		b1 = str[1];
+		if (b1 < accept.lo || accept.hi < b1)
+			goto invalid_codepoint;
+
+		if (sz == 2) {
+			codepoint = (cast(Rune)s0&0x1f)<<6 | (cast(Rune)b1&0x3f);
+			width = 2;
+			goto end;
+		}
+
+		b2 = str[2];
+		if (!gb_is_between(b2, 0x80, 0xbf))
+			goto invalid_codepoint;
+
+		if (sz == 3) {
+			codepoint = (cast(Rune)s0&0x1f)<<12 | (cast(Rune)b1&0x3f)<<6 | (cast(Rune)b2&0x3f);
+			width = 3;
+			goto end;
+		}
+
+		b3 = str[3];
+		if (!gb_is_between(b3, 0x80, 0xbf))
+			goto invalid_codepoint;
+
+		codepoint = (cast(Rune)s0&0x07)<<18 | (cast(Rune)b1&0x3f)<<12 | (cast(Rune)b2&0x3f)<<6 | (cast(Rune)b3&0x3f);
+		width = 4;
+		goto end;
+
+	invalid_codepoint:
+		codepoint = GB_RUNE_INVALID;
+		width = 1;
+	}
+
+end:
+	if (codepoint_out) *codepoint_out = codepoint;
+	return width;
+}