[cfe-dev] Clang optimizing an infinite loop

[David Chisnall via cfe-dev]

Hi,

'Clang' isn't actually doing anything here, clang emits LLVM IR, which 
is then optimised by LLVM.  In general, if you have questions about 
optimisations you're better off asking them on llvm-dev.  That said:

Infinite loops without side effects in C are undefined behaviour.  This 
is because proving that a loop terminate is probably the most famous 
example of a non-computable problem and so optimisers are left free to 
assume that loops terminate.  Specifically in N1570, Section 6.8.5, 
paragraph 6, states:

 > An iteration statement whose controlling expression is not a constant 
expression) that performs no input/output operations, does not access 
volatile objects, and performs no synchronization or atomic operations 
in its body, controlling expression, or (in the case of a for statement) 
its expression, may be assumed by the implementation to terminate.

This means that if the optimiser encounters an infinite loop it is free 
to assume that it is unreachable and elide it entirely (or replace it 
with a trap, or erase your hard disk).

David

On 19/10/2021 11:10, Gaier, Bjoern via cfe-dev wrote:
> Hello everyone,
> 
> for testing our code I wanted to create an infinite loop with Clang, so 
> I wrote the most basic infinite loop I came up with:
> 
> void Sample()
> 
> {
> 
>                  while(true);
> 
> }
> 
> To my surprise Clang generated the following assembly code:
> 
>                  .def       "?Sample@@YAXXZ";
> 
>                  .scl         2;
> 
>                  .type     32;
> 
>                  .endef
> 
>                  .section .text,"xr",one_only,"?Sample@@YAXXZ"
> 
>                  .globl     "?Sample@@YAXXZ"            # -- Begin 
> function ?Sample@@YAXXZ
> 
>                  .p2align                4, 0x90
> 
> "?Sample@@YAXXZ":                   # @"?Sample@@YAXXZ"
> 
> .Lfunc_begin4:
> 
>                  .cv_func_id 6
> 
> # %bb.0:
> 
>                  nop                                     # avoids 
> zero-length function
> 
> .Lfunc_end4:
> 
>                                          # -- End function
> 
> Which ended in an access violation D:
> 
> Why was Clang doing that? Is it because of some loop optimization since 
> the body of the loop is empty and nothing else happens in the loop? :0
> 
> But why was it missing the "ret" instruction then?
> 
> The full command line without include directories is:
> 1>clang version 13.0.0 (https://github.com/llvm/llvm-project.git 
> <https://github.com/llvm/llvm-project.git> 
> 5f500d73cd1aaff4c9ab2fd5c327c2d5ca9ae5c9)
> 
> 1>Target: x86_64-pc-windows-msvc
> 
> 1>Thread model: posix
> 
> 1> "clang-cl.exe" -cc1 -triple x86_64-pc-windows-msvc19.28.29912 -S 
> -disable-free -disable-llvm-verifier -discard-value-names 
> -main-file-name CMException.cpp -mrelocation-model pic -pic-level 2 
> -mframe-pointer=none -relaxed-aliasing -fmath-errno -fno-rounding-math 
> -mconstructor-aliases -munwind-tables -target-cpu skylake 
> -target-feature +fsgsbase -mllvm -x86-asm-syntax=intel -mllvm 
> -treat-scalable-fixed-error-as-warning -D_MT -flto-visibility-public-std 
> --dependent-lib=libcmt --dependent-lib=oldnames -stack-protector 2 
> -fcxx-exceptions -fexceptions -fexternc-nounwind -fms-volatile 
> -fdefault-calling-conv=cdecl -fdiagnostics-format msvc -gno-column-info 
> -gcodeview -debug-info-kind=limited -v -ffunction-sections -D WIN32 -D 
> NTSYSAD_CM -D CODEMODULE -D _CONSOLE -D _UNICODE -D NDEBUG -D 
> _CRT_SECURE_NO_WARNINGS -D _CRT_USE_BUILTIN_OFFSETOF -D NOMINMAX -D 
> _UNICODE -D UNICODE -O2 -Wall -Wno-error 
> -Wno-unused-command-line-argument -Wno-microsoft-cast 
> -Wno-writable-strings -Wno-microsoft-enum-forward-reference 
> -Wno-invalid-token-paste -fdeprecated-macro -ferror-limit 3000 
> -fno-use-cxa-atexit -fms-extensions -fms-compatibility 
> -fms-compatibility-version=19.28.29912 -std=c++17 
> -fdelayed-template-parsing -fno-caret-diagnostics -vectorize-loops 
> -vectorize-slp -faddrsig -o "x64\\Release\\CMException.asm" -x c++ 
> CMException.cpp
> 
> Thank you for any help in advance!
> 
> Kind regards
> 
> Björn
> 
> Als GmbH eingetragen im Handelsregister Bad Homburg v.d.H. HRB 9816, 
> USt.ID-Nr. DE 114 165 789 Geschäftsführer: Dr. Hiroshi Nakamura, Dr. 
> Robert Plank, Markus Bode, Takashi Nagano, Tetsuhiro Habe, Ergin Cansiz.
> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>