Re: head -r356426 for 32-bit powerpc vs. powerpc-unknown-freebsd13.0-g++9 and g++9: not (fully) clang++-ABI compatible (using system-headers and libraries, not gcc's)

On Sun, 12 Jan 2020 01:57:44 -0800
Mark Millard <[email protected]> wrote:

> [So far I have not checked if there is a
> somewhat analogous C (not C++) issue or
> not for gcc9 . For C++, when registers are
> used vs. when stack space is used does not
> always match system-clang++ for g++9
> targeting 32-bit powerpc.]
> 
> Building on a head -r356426 32-bit powerpc
> the following program:
> 
> # more just_now.cpp 
> #include <chrono>
> #include <vector>
> int main(void)
> {
>     auto now0{std::chrono::steady_clock::now()};
>     auto now1{std::chrono::steady_clock::now()};
>     volatile std::vector<decltype(now0)> ta{ {now0,now1} };
> 
>     return 0;
> }
> 
> via:
> 
> # g++9 -std=c++17 -nostdinc++ -I/usr/include/c++/v1 \
> -pedantic -g -O2 -nodefaultlibs -lc++ -lc -lgcc_s \
> just_now.cpp
> 
> produces an a.out that SIGSEGV's. Note: lack of -O?
> still fails, the code is just shorter for my presentation
> purposes when I use something like -O2 .
> 
> # ldd a.out
> a.out:
> 	libc++.so.1 => /usr/lib/libc++.so.1 (0x41861000)
> 	libcxxrt.so.1 => /lib/libcxxrt.so.1 (0x4193d000)
> 	libc.so.7 => /lib/libc.so.7 (0x41969000)
> 	libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x41b74000)
> 
> The same is true for powerpc-unknown-freebsd13.0-g++9 ,
> including when used via just:
> 
> # /usr/local/bin/powerpc-unknown-freebsd13.0-g++9 \
> -std=c++17 -pedantic -g -O2 just_now.cpp
> 
> # ldd a.out
> a.out:
> 	libc++.so.1 => /usr/lib/libc++.so.1 (0x41861000)
> 	libcxxrt.so.1 => /lib/libcxxrt.so.1 (0x4193d000)
> 	libm.so.5 => /lib/libm.so.5 (0x41969000)
> 	libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x419a4000)
> 	libc.so.7 => /lib/libc.so.7 (0x419cc000)
> 
> ( The two g++9 variants use /usr/local/bin/ld vs.
> /usr/local/bin/powerpc-unknown-freebsd13.0-ld . )
> 
> Here is the beginning of main's code from the g++9
> variants:
> 
> Dump of assembler code for function main():
>    0x01800530 <+0>:	stwu    r1,-48(r1)
>    0x01800534 <+4>:	mflr    r0
>    0x01800538 <+8>:	stw     r0,52(r1)
>    0x0180053c <+12>:	stw     r30,40(r1)
>    0x01800540 <+16>:	stw     r31,44(r1)
>    0x01800544 <+20>:	bl      0x1810d3c
> <[email protected]> 0x01800548
> <+24>:	mr      r30,r3 0x0180054c <+28>:	mr      r31,r4
>    0x01800550 <+32>:	bl      0x1810d3c
> <[email protected]> . . .
> 
> Note the last 2 mr instructions: the code is expecting
> now()'s time_point to be returned in registers, not
> in stack space provided by main. It is not expecting
> now() to require an address for the time_point (in r3).
> 
> (That does not match the system-libraries
> implementation: that code will try to use r3
> as pointing to where to put the time_point.)
> 
> 
> 
> By contrast, via system clang:
> 
> # c++ -std=c++17 -nostdinc++ -I/usr/include/c++/v1 \
> -pedantic -g -O2 -nodefaultlibs -lc++ -lc -lgcc_s \
> just_now.cpp
> 
> # ldd a.out
> a.out:
> 	libc++.so.1 => /usr/lib/libc++.so.1 (0x41861000)
> 	libcxxrt.so.1 => /lib/libcxxrt.so.1 (0x4193d000)
> 	libc.so.7 => /lib/libc.so.7 (0x41969000)
> 	libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x41b74000)
> 
> Dump of assembler code for function main():
>    0x01800768 <+0>:	mflr    r0
>    0x0180076c <+4>:	stw     r0,4(r1)
>    0x01800770 <+8>:	stwu    r1,-32(r1)
>    0x01800774 <+12>:	addi    r3,r1,24
>    0x01800778 <+16>:	bl      0x1810aa0
> <[email protected]> 0x0180077c
> <+20>:	addi    r3,r1,16 0x01800780 <+24>:	bl
> 0x1810aa0 <[email protected]> . . .
> 
> clang++ creates stack space and passes the address
> of that stack space, with distinct values for
> each call. It does not expect to get the
> time_point from a now() from registers after the
> calls.
> 
> (That does match the system-libraries
> implementation: that code will try to use r3
> as pointing to where to put the time_point.)
> 
> 
> This may mean that building gcc9 without
> the full bootstrap vs. with the full
> bootstrap would have consequences for the
> gcc library code and if it matched the
> FreeBSD system or not: when built by
> clang++ the code would have clang++'s
> code generation ABI properties.
> 
> But even without a full bootstrap, use
> of g++9 variants (as they are) may be
> problematical for building ports and
> other code.
> 
> 
> For reference:
> 
> # g++9 -v
> Using built-in specs.
> COLLECT_GCC=g++9
> COLLECT_LTO_WRAPPER=/usr/local/libexec/gcc9/gcc/powerpc-portbld-freebsd13.0/9.2.0/lto-wrapper
> Target: powerpc-portbld-freebsd13.0
> Configured with:
> /wrkdirs/usr/ports/lang/gcc9/work/gcc-9.2.0/configure
> --disable-multilib --disable-plugin --disable-bootstrap --disable-nls
> --enable-gnu-indirect-function --libdir=/usr/local/lib/gcc9
> --libexecdir=/usr/local/libexec/gcc9 --program-suffix=9
> --with-as=/usr/local/bin/as --with-gmp=/usr/local
> --with-gxx-include-dir=/usr/local/lib/gcc9/include/c++/
> --with-ld=/usr/local/bin/ld --with-pkgversion='FreeBSD Ports
> Collection' --with-system-zlib --enable-languages=c,c++,objc,fortran
> --prefix=/usr/local --localstatedir=/var --mandir=/usr/local/man
> --infodir=/usr/local/share/info/gcc9
> --build=powerpc-portbld-freebsd13.0 Thread model: posix gcc version
> 9.2.0 (FreeBSD Ports Collection) 
> 
> # powerpc-unknown-freebsd13.0-g++9 -v
> Using built-in specs.
> COLLECT_GCC=powerpc-unknown-freebsd13.0-g++9
> COLLECT_LTO_WRAPPER=/usr/local/libexec/gcc/powerpc-unknown-freebsd13.0/9.2.0/lto-wrapper
> Target: powerpc-unknown-freebsd13.0
> Configured with:
> /wrkdirs/usr/ports/devel/freebsd-gcc9/work-powerpc/gcc-9.2.0/configure
> --target=powerpc-unknown-freebsd13.0 --disable-nls
> --enable-languages=c,c++ --enable-gnu-indirect-function
> --enable-initfini-array --program-prefix=powerpc-unknown-freebsd13.0-
> --program-suffix=9 --without-headers --with-gmp=/usr/local
> --with-pkgversion='FreeBSD Ports Collection for powerpc'
> --with-system-zlib --with-gxx-include-dir=/usr/include/c++/v1/
> --with-sysroot=/
> --with-as=/usr/local/bin/powerpc-unknown-freebsd13.0-as
> --with-ld=/usr/local/bin/powerpc-unknown-freebsd13.0-ld
> --prefix=/usr/local --localstatedir=/var --mandir=/usr/local/man
> --infodir=/usr/local/share/info/ --build=powerpc-unknown-freebsd13.0
> Thread model: posix gcc version 9.2.0 (FreeBSD Ports Collection for
> powerpc) 
> 
> # c++ -v
> FreeBSD clang version 9.0.1 ([email protected]:llvm/llvm-project.git
> c1a0a213378a458fbea1a5c77b315c7dce08fd05) (based on LLVM 9.0.1)
> Target: powerpc-unknown-freebsd13.0 Thread model: posix
> InstalledDir: /usr/bin
> 
> 
> ===
> Mark Millard
> marklmi at yahoo.com
> ( dsl-only.net went
> away in early 2018-Mar)
> 

Hi Mark,

I think this is filed as a bug against LLVM already:
https://bugs.llvm.org/show_bug.cgi?id=40736 .  You can add to that PR.

- Justin
_______________________________________________
[email protected] mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-ppc
To unsubscribe, send any mail to "[email protected]"