[Bug c/105886] New: GCC optimizes structs with FP registers but ignores alignment in this case

keven.kloeckner at siemens dot com gcc-bugzilla@gcc.gnu.org
Wed Jun 8 11:30:59 GMT 2022


            Bug ID: 105886
           Summary: GCC optimizes structs with FP registers but ignores
                    alignment in this case
           Product: gcc
           Version: 11.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: keven.kloeckner at siemens dot com
  Target Milestone: ---

Compiler output -v:

Using built-in specs.
Target: aarch64-elf
Configured with: ../../gcc-11.2.0//configure --host=x86_64-w64-mingw32
--prefix=/build/aarch64-elf_11.2.0-1/cross-gcc/aarch64-elf --target=aarch64-elf
--disable-nls --enable-multilib --with-multilib-list=lp64,ilp32
--enable-languages=c,c++ --disable-decimal-float
--without-headers --disable-shared --disable-threads --disable-lto
--disable-libmudflap --disable-libssp --disable-libgomp --disable-libffi
--disable-libstdcxx-pch --disable-win32-registry
--with-host-libstdcxx='-static-libgcc -Wl,-Bstatic,-lstdc++,-Bdynamic -lm'
--with-newlib --with-gcc --with-gnu-as --with-gnu-ld
--with-zstd=/build/aarch64-elf_11.2.0-1/host : (reconfigured)
../../gcc-11.2.0//configure --host=x86_64-w64-mingw32
--build=x86_64-w64-mingw32 --enable-languages=c,c++ --enable-multilib
--with-multilib-list=lp64,ilp32 --disable-lto --disable-libmudflap
--disable-libssp --disable-libgomp --disable-libffi --with-newlib --with-gcc
--with-gnu-ld --with-gnu-as --with-stabs --disable-shared --disable-threads
--disable-win32-registry --disable-nls --disable-libstdcxx-pch
--with-host-libstdcxx='-static-libgcc -Wl,-Bstatic,-lstdc++,-Bdynamic -lm'
--target=aarch64-elf --prefix=/build/aarch64-elf_11.2.0-1/cross-gcc/aarch64-elf
Thread model: single
Supported LTO compression algorithms: zlib zstd
gcc version 11.2.0 (GCC)

The problem occurs on our Windows to Arm Cortex A53 Cross compile.
No errors or warnings during compilation. We have enabled alignment check for
some purposes and we cannot disable it. This problem occurs also in C++, not
only in C (iam sorry if i put the bug wrong).

(for readability i removed all defines and link inputs)
Command Line: 
../../../../aarch64_gcc_elf_11.2.0/bin/aarch64-elf-gcc           -fno-builtin  
 -c    -std=gnu11    -Wno-pointer-to-int-cast    -Wno-int-to-pointer-cast   
-Wno-maybe-uninitialized       -Werror                
-mabi=ilp32 -march=armv8-a+crypto+crc -mtune=cortex-a53 -mstrict-align
-mno-outline-atomics      -Werror=return-type   -Werror=format  
-Werror=address   -Werror=array-bounds   -Werror=format-extra-args  
-Werror=init-self   -Wno-error=array-bounds        -Wno-error=stringop-overflow
    -Wno-error=free-nonheap-object       -mlittle-endian   -fno-common  
-fno-merge-constants   -ffunction-sections   -fdata-sections  
-fno-toplevel-reorder   -O0      -Werror      -nostdinc   -ggdb

For some reason the GCC 11.2 is using floating point register instructions to
optimize initialization of structs if a lot of members are initialized with the
same constant successively. 
The problem is that the GCC generates the following instructions:
stp   q0, q0, [x0]  //128 bit alignment required problem -> usually we use 64
str   q0, [x0, #32] // 64 bit alignment -> ok
stur  q0, [x0, #44] // offset with 32 bit alignment but intruction with 64 bit 
                    //required alignment -> hard problem

My assumption is the compiler somehow thinks the FP registers are 4 byte or the
instructions require 4 byte alignment. I expected the compiler to ensure the
alignment if he is optimizing, which i can clearly see he is ignoring or doing
it wrong. If more informations are needed please contact me.

With kind regards
Keven Kloeckner

More information about the Gcc-bugs mailing list