1 \input texinfo @c -*- texinfo -*-
3 @setfilename tcc-doc.info
4 @settitle Tiny C Compiler Reference Documentation
5 @dircategory Software development
7 * TCC: (tcc-doc). The Tiny C Compiler.
17 @center @titlefont{Tiny C Compiler Reference Documentation}
25 @node Top, Introduction, (dir), (dir)
26 @top Tiny C Compiler Reference Documentation
28 This manual documents version @value{VERSION} of the Tiny C Compiler.
31 * Introduction:: Introduction to tcc.
32 * Invoke:: Invocation of tcc (command line, options).
33 * Clang:: ANSI C and extensions.
34 * asm:: Assembler syntax.
35 * linker:: Output file generation and supported targets.
36 * Bounds:: Automatic bounds-checking of C code.
37 * Libtcc:: The libtcc library.
38 * devel:: Guide for Developers.
45 TinyCC (aka TCC) is a small but hyper fast C compiler. Unlike other C
46 compilers, it is meant to be self-relying: you do not need an
47 external assembler or linker because TCC does that for you.
49 TCC compiles so @emph{fast} that even for big projects @code{Makefile}s may
52 TCC not only supports ANSI C, but also most of the new ISO C99
53 standard and many GNUC extensions including inline assembly.
55 TCC can also be used to make @emph{C scripts}, i.e. pieces of C source
56 that you run as a Perl or Python script. Compilation is so fast that
57 your script will be as fast as if it was an executable.
59 TCC can also automatically generate memory and bound checks
60 (@pxref{Bounds}) while allowing all C pointers operations. TCC can do
61 these checks even if non patched libraries are used.
63 With @code{libtcc}, you can use TCC as a backend for dynamic code
64 generation (@pxref{Libtcc}).
66 TCC mainly supports the i386 target on Linux and Windows. There are alpha
67 ports for the ARM (@code{arm-tcc}) and the TMS320C67xx targets
68 (@code{c67-tcc}). More information about the ARM port is available at
69 @url{http://lists.gnu.org/archive/html/tinycc-devel/2003-10/msg00044.html}.
71 For usage on Windows, see also tcc-win32.txt.
74 @chapter Command line invocation
80 usage: tcc [options] [@var{infile1} @var{infile2}@dots{}] [@option{-run} @var{infile} @var{args}@dots{}]
85 @c man begin DESCRIPTION
86 TCC options are a very much like gcc options. The main difference is that TCC
87 can also execute directly the resulting program and give it runtime
90 Here are some examples to understand the logic:
93 @item @samp{tcc -run a.c}
94 Compile @file{a.c} and execute it directly
96 @item @samp{tcc -run a.c arg1}
97 Compile a.c and execute it directly. arg1 is given as first argument to
98 the @code{main()} of a.c.
100 @item @samp{tcc a.c -run b.c arg1}
101 Compile @file{a.c} and @file{b.c}, link them together and execute them. arg1 is given
102 as first argument to the @code{main()} of the resulting program.
104 Because multiple C files are specified, @option{--} are necessary to clearly
105 separate the program arguments from the TCC options.
108 @item @samp{tcc -o myprog a.c b.c}
109 Compile @file{a.c} and @file{b.c}, link them and generate the executable @file{myprog}.
111 @item @samp{tcc -o myprog a.o b.o}
112 link @file{a.o} and @file{b.o} together and generate the executable @file{myprog}.
114 @item @samp{tcc -c a.c}
115 Compile @file{a.c} and generate object file @file{a.o}.
117 @item @samp{tcc -c asmfile.S}
118 Preprocess with C preprocess and assemble @file{asmfile.S} and generate
119 object file @file{asmfile.o}.
121 @item @samp{tcc -c asmfile.s}
122 Assemble (but not preprocess) @file{asmfile.s} and generate object file
125 @item @samp{tcc -r -o ab.o a.c b.c}
126 Compile @file{a.c} and @file{b.c}, link them together and generate the object file @file{ab.o}.
132 TCC can be invoked from @emph{scripts}, just as shell scripts. You just
133 need to add @code{#!/usr/local/bin/tcc -run} at the start of your C source:
136 #!/usr/local/bin/tcc -run
141 printf("Hello World\n");
146 TCC can read C source code from @emph{standard input} when @option{-} is used in
147 place of @option{infile}. Example:
150 echo 'main()@{puts("hello");@}' | tcc -run -
154 @section Option summary
161 Display current TCC version, increase verbosity.
163 @item -print-search-dirs
164 Print the name of the configured installation directory and a list
165 of program and library directories tcc will search.
168 Generate an object file.
171 Put object file, executable, or dll into output file @file{outfile}.
174 Set the path where the tcc internal libraries can be found (default is
175 @file{PREFIX/lib/tcc}).
178 Output compilation statistics.
180 @item -run source [args...]
181 Compile file @var{source} and run it with the command line arguments
182 @var{args}. In order to be able to give more than one argument to a
183 script, several TCC options can be given @emph{after} the
184 @option{-run} option, separated by spaces. Example:
187 tcc "-run -L/usr/X11R6/lib -lX11" ex4.c
190 In a script, it gives the following header:
193 #!/usr/local/bin/tcc -run -L/usr/X11R6/lib -lX11
195 int main(int argc, char **argv)
203 Preprocessor options:
207 Specify an additional include path. Include paths are searched in the
208 order they are specified.
210 System include paths are always searched after. The default system
211 include paths are: @file{/usr/local/include}, @file{/usr/include}
212 and @file{PREFIX/lib/tcc/include}. (@file{PREFIX} is usually
213 @file{/usr} or @file{/usr/local}).
216 Define preprocessor symbol @samp{sym} to
217 val. If val is not present, its value is @samp{1}. Function-like macros can
218 also be defined: @option{-DF(a)=a+1}
221 Undefine preprocessor symbol @samp{sym}.
226 Note: each of the following warning options has a negative form beginning with
230 @item -funsigned-char
231 Let the @code{char} type be unsigned.
234 Let the @code{char} type be signed.
237 Do not generate common symbols for uninitialized data.
239 @item -fleading-underscore
240 Add a leading underscore at the beginning of each C symbol.
248 Disable all warnings.
252 Note: each of the following warning options has a negative form beginning with
256 @item -Wimplicit-function-declaration
257 Warn about implicit function declaration.
260 Warn about unsupported GCC features that are ignored by TCC.
262 @item -Wwrite-strings
263 Make string constants be of type @code{const char *} instead of @code{char
267 Abort compilation if warnings are issued.
270 Activate all warnings, except @option{-Werror}, @option{-Wunusupported} and
271 @option{-Wwrite-strings}.
279 Specify an additional static library path for the @option{-l} option. The
280 default library paths are @file{/usr/local/lib}, @file{/usr/lib} and @file{/lib}.
283 Link your program with dynamic library libxxx.so or static library
284 libxxx.a. The library is searched in the paths specified by the
288 Generate a shared library instead of an executable.
291 set name for shared library to be used at runtime
294 Generate a statically linked executable (default is a shared linked
298 Export global symbols to the dynamic linker. It is useful when a library
299 opened with @code{dlopen()} needs to access executable symbols.
302 Generate an object file combining all input files.
304 @item -Wl,-Ttext,address
305 Set the start of the .text section to @var{address}.
307 @item -Wl,--oformat,fmt
308 Use @var{fmt} as output format. The supported output formats are:
311 ELF output format (default)
313 Binary image (only for executable output)
315 COFF output format (only for executable output for TMS320C67xx target)
318 @item -Wl,-rpath=path
319 Set custom library search path
327 Generate run time debug information so that you get clear run time
328 error messages: @code{ test.c:68: in function 'test5()': dereferencing
329 invalid pointer} instead of the laconic @code{Segmentation
333 Generate additional support code to check
334 memory allocations and array/pointer bounds. @option{-g} is implied. Note
335 that the generated code is slower and bigger in this case.
338 Display N callers in stack traces. This is useful with @option{-g} or
343 Note: GCC options @option{-Ox}, @option{-fx} and @option{-mx} are
350 @settitle Tiny C Compiler
363 @chapter C language support
367 TCC implements all the ANSI C standard, including structure bit fields
368 and floating point numbers (@code{long double}, @code{double}, and
369 @code{float} fully supported).
371 @section ISOC99 extensions
373 TCC implements many features of the new C standard: ISO C99. Currently
374 missing items are: complex and imaginary numbers and variable length
377 Currently implemented ISOC99 features:
381 @item 64 bit @code{long long} types are fully supported.
383 @item The boolean type @code{_Bool} is supported.
385 @item @code{__func__} is a string variable containing the current
388 @item Variadic macros: @code{__VA_ARGS__} can be used for
389 function-like macros:
391 #define dprintf(level, __VA_ARGS__) printf(__VA_ARGS__)
395 @code{dprintf} can then be used with a variable number of parameters.
397 @item Declarations can appear anywhere in a block (as in C++).
399 @item Array and struct/union elements can be initialized in any order by
402 struct @{ int x, y; @} st[10] = @{ [0].x = 1, [0].y = 2 @};
404 int tab[10] = @{ 1, 2, [5] = 5, [9] = 9@};
407 @item Compound initializers are supported:
409 int *p = (int [])@{ 1, 2, 3 @};
411 to initialize a pointer pointing to an initialized array. The same
412 works for structures and strings.
414 @item Hexadecimal floating point constants are supported:
416 double d = 0x1234p10;
420 is the same as writing
422 double d = 4771840.0;
425 @item @code{inline} keyword is ignored.
427 @item @code{restrict} keyword is ignored.
430 @section GNU C extensions
432 TCC implements some GNU C extensions:
436 @item array designators can be used without '=':
438 int a[10] = @{ [0] 1, [5] 2, 3, 4 @};
441 @item Structure field designators can be a label:
443 struct @{ int x, y; @} st = @{ x: 1, y: 1@};
447 struct @{ int x, y; @} st = @{ .x = 1, .y = 1@};
450 @item @code{\e} is ASCII character 27.
452 @item case ranges : ranges can be used in @code{case}s:
456 printf("range 1 to 9\n");
459 printf("unexpected\n");
464 @cindex aligned attribute
465 @cindex packed attribute
466 @cindex section attribute
467 @cindex unused attribute
468 @cindex cdecl attribute
469 @cindex stdcall attribute
470 @cindex regparm attribute
471 @cindex dllexport attribute
473 @item The keyword @code{__attribute__} is handled to specify variable or
474 function attributes. The following attributes are supported:
477 @item @code{aligned(n)}: align a variable or a structure field to n bytes
478 (must be a power of two).
480 @item @code{packed}: force alignment of a variable or a structure field to
483 @item @code{section(name)}: generate function or data in assembly section
484 name (name is a string containing the section name) instead of the default
487 @item @code{unused}: specify that the variable or the function is unused.
489 @item @code{cdecl}: use standard C calling convention (default).
491 @item @code{stdcall}: use Pascal-like calling convention.
493 @item @code{regparm(n)}: use fast i386 calling convention. @var{n} must be
494 between 1 and 3. The first @var{n} function parameters are respectively put in
495 registers @code{%eax}, @code{%edx} and @code{%ecx}.
497 @item @code{dllexport}: export function from dll/executable (win32 only)
501 Here are some examples:
503 int a __attribute__ ((aligned(8), section(".mysection")));
507 align variable @code{a} to 8 bytes and put it in section @code{.mysection}.
510 int my_add(int a, int b) __attribute__ ((section(".mycodesection")))
517 generate function @code{my_add} in section @code{.mycodesection}.
519 @item GNU style variadic macros:
521 #define dprintf(fmt, args@dots{}) printf(fmt, ## args)
524 dprintf("one arg %d\n", 1);
527 @item @code{__FUNCTION__} is interpreted as C99 @code{__func__}
528 (so it has not exactly the same semantics as string literal GNUC
529 where it is a string literal).
531 @item The @code{__alignof__} keyword can be used as @code{sizeof}
532 to get the alignment of a type or an expression.
534 @item The @code{typeof(x)} returns the type of @code{x}.
535 @code{x} is an expression or a type.
537 @item Computed gotos: @code{&&label} returns a pointer of type
538 @code{void *} on the goto label @code{label}. @code{goto *expr} can be
539 used to jump on the pointer resulting from @code{expr}.
541 @item Inline assembly with asm instruction:
542 @cindex inline assembly
543 @cindex assembly, inline
546 static inline void * my_memcpy(void * to, const void * from, size_t n)
549 __asm__ __volatile__(
554 "1:\ttestb $1,%b4\n\t"
558 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
559 :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
567 TCC includes its own x86 inline assembler with a @code{gas}-like (GNU
568 assembler) syntax. No intermediate files are generated. GCC 3.x named
569 operands are supported.
571 @item @code{__builtin_types_compatible_p()} and @code{__builtin_constant_p()}
574 @item @code{#pragma pack} is supported for win32 compatibility.
578 @section TinyCC extensions
582 @item @code{__TINYC__} is a predefined macro to @code{1} to
583 indicate that you use TCC.
585 @item @code{#!} at the start of a line is ignored to allow scripting.
587 @item Binary digits can be entered (@code{0b101} instead of
590 @item @code{__BOUNDS_CHECKING_ON} is defined if bound checking is activated.
595 @chapter TinyCC Assembler
597 Since version 0.9.16, TinyCC integrates its own assembler. TinyCC
598 assembler supports a gas-like syntax (GNU assembler). You can
599 desactivate assembler support if you want a smaller TinyCC executable
600 (the C compiler does not rely on the assembler).
602 TinyCC Assembler is used to handle files with @file{.S} (C
603 preprocessed assembler) and @file{.s} extensions. It is also used to
604 handle the GNU inline assembler with the @code{asm} keyword.
608 TinyCC Assembler supports most of the gas syntax. The tokens are the
613 @item C and C++ comments are supported.
615 @item Identifiers are the same as C, so you cannot use '.' or '$'.
617 @item Only 32 bit integer numbers are supported.
625 @item Integers in decimal, octal and hexa are supported.
627 @item Unary operators: +, -, ~.
629 @item Binary operators in decreasing priority order:
637 @item A value is either an absolute number or a label plus an offset.
638 All operators accept absolute values except '+' and '-'. '+' or '-' can be
639 used to add an offset to a label. '-' supports two labels only if they
640 are the same or if they are both defined and in the same section.
648 @item All labels are considered as local, except undefined ones.
650 @item Numeric labels can be used as local @code{gas}-like labels.
651 They can be defined several times in the same source. Use 'b'
652 (backward) or 'f' (forward) as suffix to reference them:
656 jmp 1b /* jump to '1' label before */
657 jmp 1f /* jump to '1' label after */
664 @cindex assembler directives
665 @cindex directives, assembler
666 @cindex align directive
667 @cindex skip directive
668 @cindex space directive
669 @cindex byte directive
670 @cindex word directive
671 @cindex short directive
672 @cindex int directive
673 @cindex long directive
674 @cindex quad directive
675 @cindex globl directive
676 @cindex global directive
677 @cindex section directive
678 @cindex text directive
679 @cindex data directive
680 @cindex bss directive
681 @cindex fill directive
682 @cindex org directive
683 @cindex previous directive
684 @cindex string directive
685 @cindex asciz directive
686 @cindex ascii directive
688 All directives are preceeded by a '.'. The following directives are
692 @item .align n[,value]
693 @item .skip n[,value]
694 @item .space n[,value]
695 @item .byte value1[,...]
696 @item .word value1[,...]
697 @item .short value1[,...]
698 @item .int value1[,...]
699 @item .long value1[,...]
700 @item .quad immediate_value1[,...]
703 @item .section section
707 @item .fill repeat[,size[,value]]
710 @item .string string[,...]
711 @item .asciz string[,...]
712 @item .ascii string[,...]
715 @section X86 Assembler
718 All X86 opcodes are supported. Only ATT syntax is supported (source
719 then destination operand order). If no size suffix is given, TinyCC
720 tries to guess it from the operand sizes.
722 Currently, MMX opcodes are supported but not SSE ones.
725 @chapter TinyCC Linker
728 @section ELF file generation
731 TCC can directly output relocatable ELF files (object files),
732 executable ELF files and dynamic ELF libraries without relying on an
735 Dynamic ELF libraries can be output but the C compiler does not generate
736 position independent code (PIC). It means that the dynamic library
737 code generated by TCC cannot be factorized among processes yet.
739 TCC linker eliminates unreferenced object code in libraries. A single pass is
740 done on the object and library list, so the order in which object files and
741 libraries are specified is important (same constraint as GNU ld). No grouping
742 options (@option{--start-group} and @option{--end-group}) are supported.
744 @section ELF file loader
746 TCC can load ELF object files, archives (.a files) and dynamic
749 @section PE-i386 file generation
752 TCC for Windows supports the native Win32 executable file format (PE-i386). It
753 generates EXE files (console and gui) and DLL files.
755 For usage on Windows, see also tcc-win32.txt.
757 @section GNU Linker Scripts
758 @cindex scripts, linker
759 @cindex linker scripts
760 @cindex GROUP, linker command
761 @cindex FILE, linker command
762 @cindex OUTPUT_FORMAT, linker command
763 @cindex TARGET, linker command
765 Because on many Linux systems some dynamic libraries (such as
766 @file{/usr/lib/libc.so}) are in fact GNU ld link scripts (horrible!),
767 the TCC linker also supports a subset of GNU ld scripts.
769 The @code{GROUP} and @code{FILE} commands are supported. @code{OUTPUT_FORMAT}
770 and @code{TARGET} are ignored.
772 Example from @file{/usr/lib/libc.so}:
775 Use the shared library, but some functions are only in
776 the static library, so try that secondarily. */
777 GROUP ( /lib/libc.so.6 /usr/lib/libc_nonshared.a )
781 @chapter TinyCC Memory and Bound checks
783 @cindex memory checks
785 This feature is activated with the @option{-b} (@pxref{Invoke}).
787 Note that pointer size is @emph{unchanged} and that code generated
788 with bound checks is @emph{fully compatible} with unchecked
789 code. When a pointer comes from unchecked code, it is assumed to be
790 valid. Even very obscure C code with casts should work correctly.
792 For more information about the ideas behind this method, see
793 @url{http://www.doc.ic.ac.uk/~phjk/BoundsChecking.html}.
795 Here are some examples of caught errors:
799 @item Invalid range with standard string function:
807 @item Out of bounds-error in global or local arrays:
817 @item Out of bounds-error in malloc'ed data:
821 tab = malloc(20 * sizeof(int));
829 @item Access of freed memory:
833 tab = malloc(20 * sizeof(int));
845 tab = malloc(20 * sizeof(int));
854 @chapter The @code{libtcc} library
856 The @code{libtcc} library enables you to use TCC as a backend for
857 dynamic code generation.
859 Read the @file{libtcc.h} to have an overview of the API. Read
860 @file{libtcc_test.c} to have a very simple example.
862 The idea consists in giving a C string containing the program you want
863 to compile directly to @code{libtcc}. Then you can access to any global
864 symbol (function or variable) defined.
867 @chapter Developer's guide
869 This chapter gives some hints to understand how TCC works. You can skip
870 it if you do not intend to modify the TCC code.
872 @section File reading
874 The @code{BufferedFile} structure contains the context needed to read a
875 file, including the current line number. @code{tcc_open()} opens a new
876 file and @code{tcc_close()} closes it. @code{inp()} returns the next
881 @code{next()} reads the next token in the current
882 file. @code{next_nomacro()} reads the next token without macro
885 @code{tok} contains the current token (see @code{TOK_xxx})
886 constants. Identifiers and keywords are also keywords. @code{tokc}
887 contains additional infos about the token (for example a constant value
888 if number or string token).
892 The parser is hardcoded (yacc is not necessary). It does only one pass,
897 @item For initialized arrays with unknown size, a first pass
898 is done to count the number of elements.
900 @item For architectures where arguments are evaluated in
901 reverse order, a first pass is done to reverse the argument order.
907 The types are stored in a single 'int' variable. It was choosen in the
908 first stages of development when tcc was much simpler. Now, it may not
909 be the best solution.
912 #define VT_INT 0 /* integer type */
913 #define VT_BYTE 1 /* signed byte type */
914 #define VT_SHORT 2 /* short type */
915 #define VT_VOID 3 /* void type */
916 #define VT_PTR 4 /* pointer */
917 #define VT_ENUM 5 /* enum definition */
918 #define VT_FUNC 6 /* function type */
919 #define VT_STRUCT 7 /* struct/union definition */
920 #define VT_FLOAT 8 /* IEEE float */
921 #define VT_DOUBLE 9 /* IEEE double */
922 #define VT_LDOUBLE 10 /* IEEE long double */
923 #define VT_BOOL 11 /* ISOC99 boolean type */
924 #define VT_LLONG 12 /* 64 bit integer */
925 #define VT_LONG 13 /* long integer (NEVER USED as type, only
927 #define VT_BTYPE 0x000f /* mask for basic type */
928 #define VT_UNSIGNED 0x0010 /* unsigned type */
929 #define VT_ARRAY 0x0020 /* array type (also has VT_PTR) */
930 #define VT_VLA 0x20000 /* VLA type (also has VT_PTR and VT_ARRAY) */
931 #define VT_BITFIELD 0x0040 /* bitfield modifier */
932 #define VT_CONSTANT 0x0800 /* const modifier */
933 #define VT_VOLATILE 0x1000 /* volatile modifier */
934 #define VT_SIGNED 0x2000 /* signed type */
936 #define VT_STRUCT_SHIFT 18 /* structure/enum name shift (14 bits left) */
939 When a reference to another type is needed (for pointers, functions and
940 structures), the @code{32 - VT_STRUCT_SHIFT} high order bits are used to
941 store an identifier reference.
943 The @code{VT_UNSIGNED} flag can be set for chars, shorts, ints and long
946 Arrays are considered as pointers @code{VT_PTR} with the flag
947 @code{VT_ARRAY} set. Variable length arrays are considered as special
948 arrays and have flag @code{VT_VLA} set instead of @code{VT_ARRAY}.
950 The @code{VT_BITFIELD} flag can be set for chars, shorts, ints and long
951 longs. If it is set, then the bitfield position is stored from bits
952 VT_STRUCT_SHIFT to VT_STRUCT_SHIFT + 5 and the bit field size is stored
953 from bits VT_STRUCT_SHIFT + 6 to VT_STRUCT_SHIFT + 11.
955 @code{VT_LONG} is never used except during parsing.
957 During parsing, the storage of an object is also stored in the type
961 #define VT_EXTERN 0x00000080 /* extern definition */
962 #define VT_STATIC 0x00000100 /* static variable */
963 #define VT_TYPEDEF 0x00000200 /* typedef definition */
964 #define VT_INLINE 0x00000400 /* inline definition */
965 #define VT_IMPORT 0x00004000 /* win32: extern data imported from dll */
966 #define VT_EXPORT 0x00008000 /* win32: data exported from dll */
967 #define VT_WEAK 0x00010000 /* win32: data exported from dll */
972 All symbols are stored in hashed symbol stacks. Each symbol stack
973 contains @code{Sym} structures.
975 @code{Sym.v} contains the symbol name (remember
976 an idenfier is also a token, so a string is never necessary to store
977 it). @code{Sym.t} gives the type of the symbol. @code{Sym.r} is usually
978 the register in which the corresponding variable is stored. @code{Sym.c} is
979 usually a constant associated to the symbol like its address for normal
980 symbols, and the number of entries for symbols representing arrays.
981 Variable length array types use @code{Sym.c} as a location on the stack
982 which holds the runtime sizeof for the type.
984 Four main symbol stacks are defined:
989 for the macros (@code{#define}s).
992 for the global variables, functions and types.
995 for the local variables, functions and types.
997 @item global_label_stack
998 for the local labels (for @code{goto}).
1001 for GCC block local labels (see the @code{__label__} keyword).
1005 @code{sym_push()} is used to add a new symbol in the local symbol
1006 stack. If no local symbol stack is active, it is added in the global
1009 @code{sym_pop(st,b)} pops symbols from the symbol stack @var{st} until
1010 the symbol @var{b} is on the top of stack. If @var{b} is NULL, the stack
1013 @code{sym_find(v)} return the symbol associated to the identifier
1014 @var{v}. The local stack is searched first from top to bottom, then the
1019 The generated code and datas are written in sections. The structure
1020 @code{Section} contains all the necessary information for a given
1021 section. @code{new_section()} creates a new section. ELF file semantics
1022 is assumed for each section.
1024 The following sections are predefined:
1029 is the section containing the generated code. @var{ind} contains the
1030 current position in the code section.
1033 contains initialized data
1036 contains uninitialized data
1038 @item bounds_section
1039 @itemx lbounds_section
1040 are used when bound checking is activated
1043 @itemx stabstr_section
1044 are used when debugging is actived to store debug information
1046 @item symtab_section
1047 @itemx strtab_section
1048 contain the exported symbols (currently only used for debugging).
1052 @section Code generation
1053 @cindex code generation
1055 @subsection Introduction
1057 The TCC code generator directly generates linked binary code in one
1058 pass. It is rather unusual these days (see gcc for example which
1059 generates text assembly), but it can be very fast and surprisingly
1062 The TCC code generator is register based. Optimization is only done at
1063 the expression level. No intermediate representation of expression is
1064 kept except the current values stored in the @emph{value stack}.
1066 On x86, three temporary registers are used. When more registers are
1067 needed, one register is spilled into a new temporary variable on the stack.
1069 @subsection The value stack
1070 @cindex value stack, introduction
1072 When an expression is parsed, its value is pushed on the value stack
1073 (@var{vstack}). The top of the value stack is @var{vtop}. Each value
1074 stack entry is the structure @code{SValue}.
1076 @code{SValue.t} is the type. @code{SValue.r} indicates how the value is
1077 currently stored in the generated code. It is usually a CPU register
1078 index (@code{REG_xxx} constants), but additional values and flags are
1082 #define VT_CONST 0x00f0
1083 #define VT_LLOCAL 0x00f1
1084 #define VT_LOCAL 0x00f2
1085 #define VT_CMP 0x00f3
1086 #define VT_JMP 0x00f4
1087 #define VT_JMPI 0x00f5
1088 #define VT_LVAL 0x0100
1089 #define VT_SYM 0x0200
1090 #define VT_MUSTCAST 0x0400
1091 #define VT_MUSTBOUND 0x0800
1092 #define VT_BOUNDED 0x8000
1093 #define VT_LVAL_BYTE 0x1000
1094 #define VT_LVAL_SHORT 0x2000
1095 #define VT_LVAL_UNSIGNED 0x4000
1096 #define VT_LVAL_TYPE (VT_LVAL_BYTE | VT_LVAL_SHORT | VT_LVAL_UNSIGNED)
1102 indicates that the value is a constant. It is stored in the union
1103 @code{SValue.c}, depending on its type.
1106 indicates a local variable pointer at offset @code{SValue.c.i} in the
1110 indicates that the value is actually stored in the CPU flags (i.e. the
1111 value is the consequence of a test). The value is either 0 or 1. The
1112 actual CPU flags used is indicated in @code{SValue.c.i}.
1114 If any code is generated which destroys the CPU flags, this value MUST be
1115 put in a normal register.
1119 indicates that the value is the consequence of a conditional jump. For VT_JMP,
1120 it is 1 if the jump is taken, 0 otherwise. For VT_JMPI it is inverted.
1122 These values are used to compile the @code{||} and @code{&&} logical
1125 If any code is generated, this value MUST be put in a normal
1126 register. Otherwise, the generated code won't be executed if the jump is
1130 is a flag indicating that the value is actually an lvalue (left value of
1131 an assignment). It means that the value stored is actually a pointer to
1134 Understanding the use @code{VT_LVAL} is very important if you want to
1135 understand how TCC works.
1138 @itemx VT_LVAL_SHORT
1139 @itemx VT_LVAL_UNSIGNED
1140 if the lvalue has an integer type, then these flags give its real
1141 type. The type alone is not enough in case of cast optimisations.
1144 is a saved lvalue on the stack. @code{VT_LLOCAL} should be eliminated
1145 ASAP because its semantics are rather complicated.
1148 indicates that a cast to the value type must be performed if the value
1149 is used (lazy casting).
1152 indicates that the symbol @code{SValue.sym} must be added to the constant.
1156 are only used for optional bound checking.
1160 @subsection Manipulating the value stack
1163 @code{vsetc()} and @code{vset()} pushes a new value on the value
1164 stack. If the previous @var{vtop} was stored in a very unsafe place(for
1165 example in the CPU flags), then some code is generated to put the
1166 previous @var{vtop} in a safe storage.
1168 @code{vpop()} pops @var{vtop}. In some cases, it also generates cleanup
1169 code (for example if stacked floating point registers are used as on
1172 The @code{gv(rc)} function generates code to evaluate @var{vtop} (the
1173 top value of the stack) into registers. @var{rc} selects in which
1174 register class the value should be put. @code{gv()} is the @emph{most
1175 important function} of the code generator.
1177 @code{gv2()} is the same as @code{gv()} but for the top two stack
1180 @subsection CPU dependent code generation
1181 @cindex CPU dependent
1182 See the @file{i386-gen.c} file to have an example.
1187 must generate the code needed to load a stack value into a register.
1190 must generate the code needed to store a register into a stack value
1194 @itemx gfunc_param()
1196 should generate a function call
1198 @item gfunc_prolog()
1199 @itemx gfunc_epilog()
1200 should generate a function prolog/epilog.
1203 must generate the binary integer operation @var{op} on the two top
1204 entries of the stack which are guaranted to contain integer types.
1206 The result value should be put on the stack.
1209 same as @code{gen_opi()} for floating point operations. The two top
1210 entries of the stack are guaranted to contain floating point values of
1213 @item gen_cvt_itof()
1214 integer to floating point conversion.
1216 @item gen_cvt_ftoi()
1217 floating point to integer conversion.
1219 @item gen_cvt_ftof()
1220 floating point to floating point of different size conversion.
1222 @item gen_bounded_ptr_add()
1223 @item gen_bounded_ptr_deref()
1224 are only used for bounds checking.
1228 @section Optimizations done
1229 @cindex optimizations
1230 @cindex constant propagation
1231 @cindex strength reduction
1232 @cindex comparison operators
1233 @cindex caching processor flags
1234 @cindex flags, caching
1235 @cindex jump optimization
1236 Constant propagation is done for all operations. Multiplications and
1237 divisions are optimized to shifts when appropriate. Comparison
1238 operators are optimized by maintaining a special cache for the
1239 processor flags. &&, || and ! are optimized by maintaining a special
1240 'jump target' value. No other jump optimization is currently performed
1241 because it would require to store the code in a more abstract fashion.
1243 @unnumbered Concept Index
1250 @c texinfo-column-for-description: 32