the real difference is in decl0 where we can use external_sym
just fine also for function definitions, we don't have to use
external_global_sym. Setting VT_EXTERN in external_sym isn't
necessary either (the type will have it set if necessary).
The rest is tidying: removing unused arguments and moving
some code around.
the ABIs (and other compilers) extend sub-int return values in the
caller. TCC extends them in the callee. For compatibility with
those other compilers we have extend them in the caller as well.
That introduces a useless double extension in pure TCC-compiled code,
but fixing that generally requires that the code generator of TCC would
understand sub-int types. For the time being bite the bullet.
In gfunc_call, regisger will be saved before gcall_or_jmp. The register
stored the function will be saved too, though in some generator the SValue
of this function will be immediately poped after gcall_or_jmp, and no need to be saved. So I modify some generator to avoid save redundant SValue before gcall_or_jmp.
which requires being able to emit an arbitrary number of NOP
instructions, which is also implemented here. For x86 we
could emit other sequences but these are the easiest.
The code generated for "((void (*)(void))0x12345678)()" will be a single "CALL 0x12345678" in previous code.
However, this will not work for DLLs, because "CALL imm" is PC related, DLL relocation will break the code.
This commit fixed the problem by forcing TCC generates indirect CALLs in this situation.
AL/AX should be extended to EAX when calling functions. However, the previous code did this only for direct calls, indirect calls were ignored.
New code also avoid redundant code when generating JMP instruction. (i.e. expanding code should be generated with CALL instruction only)
tcc.h:
* cleanup struct 'Sym'
* include some 'Attributes' into 'Sym'
* in turn get rid of VT_IM/EXPORT, VT_WEAK
* re-number VT_XXX flags
* replace some 'long' function args by 'int'
tccgen.c:
* refactor parse_btype()
- configure
* use aarch64 instead of arm64
- Makefile
* rename the custom include file to "config-extra.mak"
* Also avoid "rm -r /*" if $(tccdir) is empty
- pp/Makefile
* fix .expect generation with gcc
- tcc.h
* cleanup #defines for _MSC_VER
- tccgen.c:
* fix const-propagation for &,|
* fix anonymous named struct (ms-extension) and enable
-fms-extension by default
- i386-gen.c
* clear VT_DEFSIGN
- x86_64-gen.c/win64:
* fix passing structs in registers
* fix alloca (need to keep "func_scratch" below each alloca area on stack)
(This allows to compile a working gnu-make on win64)
- tccpp.c
* alternative approach to 37999a4fbf
This is to avoid some slowdown with ## token pasting.
* get_tok_str() : return <eof> for TOK_EOF
* -funsigned-char: apply to "string" literals as well
- tccpe/tools.c: -impdef: support both 32 and 64 bit dlls anyway
- configure:
- add --config-uClibc,-musl switch and suggest to use
it if uClibc/musl is detected
- make warning options magic clang compatible
- simplify (use $confvars instead of individual options)
- Revert "Remove some unused-parameter lint"
7443db0d5f
rather use -Wno-unused-parameter (or just not -Wextra)
- #ifdef functions that are unused on some targets
- tccgen.c: use PTR_SIZE==8 instead of (X86_64 || ARM64)
- tccpe.c: fix some warnings
- integrate dummy arm-asm better
- tccgen.c/tcc.h: allow function declaration after use:
int f() { return g(); }
int g() { return 1; }
may be a warning but not an error
see also 76cb1144ef
- tccgen.c: redundant code related to inline functions removed
(functions used anywhere have sym->c set automatically)
- tccgen.c: make 32bit llop non-equal test portable
(probably not on C67)
- dynarray_add: change prototype to possibly avoid aliasing
problems or at least warnings
- lib/alloca*.S: ".section .note.GNU-stack,"",%progbits" removed
(has no effect)
- tccpe: set SizeOfCode field (for correct upx decompression)
- libtcc.c: fixed alternative -run invocation
tcc "-run -lxxx ..." file.c
(meant to load the library after file).
Also supported now:
tcc files ... options ... -run @ arguments ...
Also:
- on windows i386 and x86-64, structures of size <= 8 are
NOT returned in registers if size is not one of 1,2,4,8.
- cleanup: put all tv-push/pop/swap/rot into one place
There seems nothing wrong. With
int t1 = 176401255;
float f = 0.25;
int t2 = t1 * f; // 176401255 * 0.25 = 44100313.75
according to the arithmetic conversion rules, the number
176401255 needs to be converted to float, and the compiler
can choose either the nearest higher or nearest lower
representable number "in an implementation-defined manner".
Which may be 176401248 or 176401264. So as result both
44100312 and 44100313 are correct.
This reverts commit 664c19ad5e.
Restore ebx from *ebp because alloca might change esp.
Also disable USE_EBX for upcoming release.
Actually the benefit is less than one would expect, it
appears that tcc can't do much with more than 3 registers
except with extensive use of long longs where the disassembly
looks much prettier (and shorter also).
Also: tccgen/expr_cond() : fix wrong gv/save_regs order
tccgen.c: remove any 'nocode_wanted' checks, except in
- greloca(), disables output elf symbols and relocs
- get_reg(), will return just the first suitable reg)
- save_regs(), will do nothing
Some minor adjustments were made where nocode_wanted is set.
xxx-gen.c: disable code output directly where it happens
in functions:
- g(), output disabled
- gjmp(), will do nothing
- gtst(), dto.
on 32bit long long support was sometimes broken. This fixes
code-gen for long long values in switches, disables a x86-64 specific
testcase and avoid an undefined shift amount. It comments out
a bitfield test involving long long bitfields > 32 bit; with GCC layout
they can straddle multiple words and code generation isn't prepared
for this.
Some routines were using the wrong type (int) in passing addends,
truncating it. This matters when bit 31 isn't set and the high
32 bits are set: the truncation would make it unsigned where in
reality it's signed (happen e.g. on the x86-64 with it's load
address at top-2GB).
C standard specifies that array should be declared with a non null size
or with * for standard array. Declaration of relocs_info in tcc.h was
not respecting this rule. This commit add a R_NUM macro that maps to the
R_<ARCH>_NUM macros and declare relocs_info using it. This commit also
moves all linker-related macros from <arch>-gen.c files to <arch>-link.c
ones.
With -b, this produces garbage. Code to call __bound_local_new
is put at wrong place, overwriting the regparam setup code.
Fix copied from x86_64-gen.c.
void __attribute__((regparm(3)))
fun(int unused)
{
char local[1];
}
Previously, long longs were 'lexpand'ed into two registers
always.
Now, it expands
- constants into two constants (lo-part, hi-part)
- variables into two lvalues with offset+4 for the hi-part.
This makes long long operations look a bit nicer.
Also: don't apply i386 'inc/dec' optimization if carry
generation is wanted.
tccgen.c:gv() when loading long long from lvalue, before
was saving all registers which caused problems in the arm
function call register parameter preparation, as with
void foo(long long y, int x);
int main(void)
{
unsigned int *xx[1], x;
unsigned long long *yy[1], y;
foo(**yy, **xx);
return 0;
}
Now only the modified register is saved if necessary,
as in this case where it is used to store the result
of the post-inc:
long long *p, v, **pp;
v = 1;
p = &v;
p[0]++;
printf("another long long spill test : %lld\n", *p);
i386-gen.c :
- found a similar problem with TOK_UMULL caused by the
vstack juggle in tccgen:gen_opl()
(bug seen only when using EBX as 4th register)
BUGZILLA:
interfacing with other compilers
extend the return value to the whole register if necessary.
visual studio and gcc do not always set the whole eax register
when assigning the return value of a function.
We've encountered wrong execution results on i386 platforms with an
application that uses both code compiled with TCC and code compiled
with other compilers (namely: Visual Studio on Windows, and GCC on
Linux).
When calling a function that returns an integer value shorter than 32
bits, TCC reads the return value from the whole EAX register,
although the code generated by the other compilers can only sets AL
for 8 bit values or AX for 16 bits values, and the rest of EAX can be
anything.
We worked around this with the attached patch on i386 for the version
0.9.26, but we did not look at other platforms to find if there are
similar issues.
* Documentation is now in "docs".
* Source code is now in "src".
* Misc. fixes here and there so that everything still works.
I think I got everything in this commit, but I only tested this
on Linux (Make) and Windows (CMake), so I might've messed
something up on other platforms...
Jsut for testing. It works for me (don't break anything)
Small fixes for x86_64-gen.c in "tccpp: fix issues, add tests"
are dropped in flavor of this patch.
Pip Cet:
Okay, here's a first patch that fixes the problem (but I've found
another bug, yet unfixed, in the process), though it's not
particularly pretty code (I tried hard to keep the changes to the
minimum necessary). If we decide to actually get rid of VT_QLONG and
VT_QFLOAT (please, can we?), there are some further simplifications in
tccgen.c that might offset some of the cost of this patch.
The idea is that an integer is no longer enough to describe how an
argument is stored in registers. There are a number of possibilities
(none, integer register, two integer registers, float register, two
float registers, integer register plus float register, float register
plus integer register), and instead of enumerating them I've
introduced a RegArgs type that stores the offsets for each of our
registers (for the other architectures, it's simply an int specifying
the number of registers). If someone strongly prefers an enum, we
could do that instead, but I believe this is a place where keeping
things general is worth it, because this way it should be doable to
add SSE or AVX support.
There is one line in the patch that looks suspicious:
} else {
addr = (addr + align - 1) & -align;
param_addr = addr;
addr += size;
- sse_param_index += reg_count;
}
break;
However, this actually fixes one half of a bug we have when calling a
function with eight double arguments "interrupted" by a two-double
structure after the seventh double argument:
f(double,double,double,double,double,double,double,struct { double
x,y; },double);
In this case, the last argument should be passed in %xmm7. This patch
fixes the problem in gfunc_prolog, but not the corresponding problem
in gfunc_call, which I'll try tackling next.
* fix some macro expansion issues
* add some pp tests in tests/pp
* improved tcc -E output for better diff'ability
* remove -dD feature (quirky code, exotic feature,
didn't work well)
Based partially on ideas / researches from PipCet
Some issues remain with VA_ARGS macros (if used in a
rather tricky way).
Also, to keep it simple, the pp doesn't automtically
add any extra spaces to separate tokens which otherwise
would form wrong tokens if re-read from tcc -E output
(such as '+' '=') GCC does that, other compilers don't.
* cleanups
- #line 01 "file" / # 01 "file" processing
- #pragma comment(lib,"foo")
- tcc -E: forward some pragmas to output (pack, comment(lib))
- fix macro parameter list parsing mess from
a3fc543459a715d7143d
(some coffee might help, next time ;)
- introduce TOK_PPSTR - to have character constants as
written in the file (similar to TOK_PPNUM)
- allow '\' appear in macros
- new functions begin/end_macro to:
- fix switching macro levels during expansion
- allow unget_tok to unget more than one tok
- slight speedup by using bitflags in isidnum_table
Also:
- x86_64.c : fix decl after statements
- i386-gen,c : fix a vstack leak with VLA on windows
- configure/Makefile : build on windows (MSYS) was broken
- tcc_warning: fflush stderr to keep output order (win32)
Author: Philip <pipcet@gmail.com>
Our VLA code can be made a lot simpler (simple enough for
even me to understand it) by giving up on the optimization idea, which
is very tempting. There's a patch to do that attached, feel free to
test and commit it if you like. (It passes all the tests, at least
On Linux 32: sizeof(long)=32 == sizeof(void *)=32
on Linux 64: sizeof(long)=64 == sizeof(void *)=64
on Windows 64: sizeof(long)=32 != sizeof(void *)=64
The common code to move a returned structure packed into
registers into memory on the caller side didn't take the
register size into account when allocating local storage,
so sometimes that lead to stack overwrites (e.g. in 73_arm64.c),
on x86_64. This fixes it by generally making gfunc_sret also return
the register size.
A test program:
/* result of the new version inroduced in 4ad186c5ef: t2a = 44100312 */
#include<stdio.h>
int main() {
int t1 = 176401255;
float f = 0.25f;
int t2a = (int)(t1 * f); // must be 44100313
int t2b = (int)(t1 * (float)0.25f);
printf("t2a=%d t2b=%d \n",t2a,t2b);
return 0;
}
Author: Thomas Preud'homme <robotux@celest.fr>
Date: Tue Dec 31 23:51:20 2013 +0800
Move logic for if (int value) to tccgen.c
Move the logic to do a test of an integer value (ex if (0)) out of
arch-specific code to tccgen.c to avoid code duplication. This also
fixes test of long long value which was only testing the bottom half of
such values on 32 bits architectures.
I don't understand why if () in gtst(i) was removed.
This patch allows to compile a linux kernel v.2.4.26
W/o this patch a tcc simply crashes.
Refactoring (no logical changes):
- use memcpy in tccgen.c:ieee_finite(double d)
- use union to store attribute flags in Sym
Makefile: "CFLAGS+=-fno-strict-aliasing" basically not necessary
anymore but I left it for now because gcc sometimes behaves
unexpectedly without.
Also:
- configure: back to mode 100755
- tcc.h: remove unused variables tdata/tbss_section
- x86_64-gen.c: adjust gfunc_sret for prototype
Variants __fixsfdi/__fixxfdi are not needed for now because
the value is converted to double always.
Also:
- remove __tcc_fpinit for unix as it seems redundant by the
__setfpucw call in the startup code
- avoid reference to s->runtime_main in cross compilers
- configure: fix --with-libgcc help
- tcctok.h: cleanup
The procedure calling standard for ARM architecture mandate the use of
the base standard for variadic function. Therefore, hgen float aggregate
must be returned via stack when greater than 4 bytes and via core
registers else in case of variadic function.
This patch improve gfunc_sret() to take into account whether the
function is variadic or not and make use of gfunc_sret() return value to
determine whether to pass a structure via stack in gfunc_prolog(). It
also take advantage of knowing if a function is variadic or not move
float result value from VFP register to core register in gfunc_epilog().
Move the logic to do a test of an integer value (ex if (0)) out of
arch-specific code to tccgen.c to avoid code duplication. This also
fixes test of long long value which was only testing the bottom half of
such values on 32 bits architectures.
