https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119718
--- Comment #12 from lucier at math dot purdue.edu ---
(In reply to Jakub Jelinek from comment #7)
> We can't just warn on all calls, most of them obviously aren't in tail call
> positions and such warning would have extreme amounts of false positives.
This statement got me wondering: when I added and tested the warning in August
2023, there were only about 300 warnings in building all of stage1 of gcc
mainline:
> These are the number of times this warning was triggered building stage1:
>
> grep warning: build.log | grep sibling | sed 's/^.*://' | sort | uniq -c
> 259 callee required more stack slots than the caller
> [-Wdisabled-optimization]
> 43 callee returns a structure [-Wdisabled-optimization]
So I added the warning to a recent mainline, and I now get many more warnings
when compiling my code, for example:
> _num.c:46380:13: warning: cannot apply sibling-call optimization: other
> reasons [-Wdisabled-optimization]
> 46380 | ___SET_F64(___F64V2,___F64FLOOR(___F64V1))
I don't understand this warning, because after macro expansion the code (plus
the next line) is
> ___F64V2=floor(___F64V1);
> ___r1=(({unsigned long ___u64_x = ((___F64_U64)(___F64V2)).u64; ___u64_x =
> ((58)==0 ? (___u64_x) : ((((unsigned long)(___u64_x))>>(58)) | (((unsigned
> long)(___u64_x))<<(64-(58)))))+(6*2 + 1); ___u64_x = ((1)==0 ? (___u64_x) :
> ((((unsigned long)(___u64_x))>>(1)) | (((unsigned
> long)(___u64_x))<<(64-(1))))); __builtin_expect(((((((unsigned
> int)(___u64_x))-(6))&(7))==0)),1) ? ((long)(___u64_x)) : (___hp += (1+1),
> ___hp[-(1+1)] = (((((1)<<3))<<(3 +5))+((30)<<3)+(0)), *((double*)(___hp-1)) =
> (___F64V2), (((long)(((long*)(___hp)) - (1 +(1))))+(7)));}));
followed by more code.
In other words, that call to floor is not in tail call position, which I
understand to mean only
return call();
In fact, while there is only one call in _num.c that is in tail call position,
I find
> grep '_num.c.*cannot' build-without-musttail-2 | wc
> 285 2565 30204
I have no idea about why maybe_complain_about_tail_call is being called when
processing calls that are not in tail call position. This did not happen in
August 2023.
