https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119718
Bug ID: 119718
Summary: __attribute__((musttail)) affects whether
-foptimize-tail-calls will in fact optimize a tail
call
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: lucier at math dot purdue.edu
Target Milestone: ---
I built the Gambit Scheme system with this compiler:
/pkgs/gcc-mainline/bin/gcc -v
Using built-in specs.
COLLECT_GCC=/pkgs/gcc-mainline/bin/gcc
COLLECT_LTO_WRAPPER=/pkgs/gcc-mainline/libexec/gcc/x86_64-pc-linux-gnu/15.0.1/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../../gcc-mainline/configure --prefix=/pkgs/gcc-mainline/
--enable-languages=c --enable-checking=release --disable-multilib
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 15.0.1 20250409 (experimental) (GCC)
There are lots of files, this is a typical compile call
/pkgs/gcc-mainline/bin/gcc -save-temps -Wno-unused -Wno-write-strings
-Wdisabled-optimization -g -ftrapv -fno-strict-aliasing -fno-trapping-math
-fno-math-errno -fschedule-insns2 -foptimize-sibling-:qcalls -fipa-ra
-fmove-loop-invariants -fPIC -fno-common -mpc64 -I"../include" -c -o _parms.o
-I. -DHAVE_CONFIG_H _parms.c -D___LIBRARY
So I have the option -foptimize-sibling-calls
Many of these files end with a tail call, I'll include _parms.i because it's
one of the smaller one.
I compiled two versions of _parms.i with the only difference being the addition
of __attribute__((musttail)) on one call:
heine:~/programs/gambit/gambit/with-musttail> diff _parms.i
../without-musttail/
14082c14082
< ___ps->pc = ___pc; ___ps->hp=___hp; ___ps->fp=___fp; ___ps->r[0]=___r0;
___ps->r[1]=___r1; ___ps->r[2]=___r2; ___ps->r[3]=___r3; ___ps->r[4]=___r4; }
__attribute__((musttail)) return (*((___host*)((((long*)((___pc)-(1))) + (1
+((-2)))))))(___ps); }
---
> ___ps->pc = ___pc; ___ps->hp=___hp; ___ps->fp=___fp; ___ps->r[0]=___r0;
> ___ps->r[1]=___r1; ___ps->r[2]=___r2; ___ps->r[3]=___r3; ___ps->r[4]=___r4; }
> return (*((___host*)((((long*)((___pc)-(1))) + (1 +((-2)))))))(___ps); }
Without __attribute__((musttail)), the tail call is not optimized, and with the
same thing happening in the other files, eventually the call depth gets too big
for the stack and there's a segfault. I can see that in the backtrace from
gdb.
If I were to try to diagnose it by adding __attribute__((musttail)), I find
that the tail call is now optimized, there is no blowing up of the call stack,
there is no segfault.
If I hadn't looked at the backtrace I wouldn't have been able to know what was
going on.
I don't think __attribute__((musttail)) should change which tail calls get
optimized.
File to be added next.
