Bug#860823: gcc-6: sqrt x64 intrinsic broken with -O3 and -ffast-math
Package: gcc-6
Version: 6.3.0-12
Severity: normal
After being "hit" by unexplained division by 0 related crashes of an
application, the
following simple test code shows the issue:
#include
#include
void TestSqrt(const float *Src,float *Dest,size_t Size,float &Gain)
{
while (Size--)
{
float G,X;
G=sqrt(X=*Src++);
if (G)
{
Gain=Gain*0.9f+1/G*0.1f;
X*=Gain;
}
*Dest++=X;
}
}
The compile and listing generating where done using:
gcc -O3 -m64 -ffast-math -c -o sqrt.o sqrt.cpp
objdump -M intel -S -r -C sqrt.o > sqrt.lst
The "sqrt" coresponding piece of code looks like:
30: f3 0f 10 0c 87 movss xmm1,DWORD PTR [rdi+rax*4]
35: f3 0f 52 d1 rsqrtss xmm2,xmm1
39: 0f 28 c2movaps xmm0,xmm2
3c: 0f 2f cbcomiss xmm1,xmm3
3f: f3 0f 59 c1 mulss xmm0,xmm1
43: f3 0f 59 c2 mulss xmm0,xmm2
47: f3 0f 59 d4 mulss xmm2,xmm4
4b: f3 0f 58 c5 addss xmm0,xmm5
4f: f3 0f 59 c2 mulss xmm0,xmm2
53: 74 18 je 6d
55: f3 0f 10 11 movss xmm2,DWORD PTR [rcx]
59: f3 0f 59 c6 mulss xmm0,xmm6
5d: f3 0f 59 d7 mulss xmm2,xmm7
61: f3 0f 58 c2 addss xmm0,xmm2
65: f3 0f 59 c8 mulss xmm1,xmm0
69: f3 0f 11 01 movss DWORD PTR [rcx],xmm0
6d: f3 0f 11 0c 86 movss DWORD PTR [rsi+rax*4],xmm1
72: 48 83 c0 01 addrax,0x1
76: 48 39 c2cmprdx,rax
79: 75 b5 jne30
The generated code is not handling in any way the case of 0 as input value. As
per Intel spec,
rsqrtss has an +/-INF result when the input is 0 and the final results of the
test code is
unpredictable, while 0 is a perfect valid input value for sqrt.
Regards,
Catalin
-- System Information:
Debian Release: 9.0
APT prefers testing
APT policy: (500, 'testing')
Architecture: amd64 (x86_64)
Kernel: Linux 4.9.0-2-amd64 (SMP w/2 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Init: systemd (via /run/systemd/system)
Versions of packages gcc-6 depends on:
ii binutils 2.28-3
ii cpp-6 6.3.0-12
ii gcc-6-base6.3.0-12
ii libc6 2.24-9
ii libcc1-0 6.3.0-12
ii libgcc-6-dev 6.3.0-12
ii libgcc1 1:6.3.0-12
ii libgmp10 2:6.1.2+dfsg-1
ii libisl15 0.18-1
ii libmpc3 1.0.3-1+b2
ii libmpfr4 3.1.5-1
ii libstdc++66.3.0-12
ii zlib1g1:1.2.8.dfsg-5
Versions of packages gcc-6 recommends:
ii libc6-dev 2.24-9
Versions of packages gcc-6 suggests:
pn gcc-6-doc
pn gcc-6-locales
pn gcc-6-multilib
pn libasan3-dbg
pn libatomic1-dbg
pn libcilkrts5-dbg
pn libgcc1-dbg
pn libgomp1-dbg
pn libitm1-dbg
pn liblsan0-dbg
pn libmpx2-dbg
pn libquadmath0-dbg
pn libtsan0-dbg
pn libubsan0-dbg
-- no debconf information
Bug#560812: gcc-4.4: Incorrect 3DNow! code generated
Package: gcc-4.4
Version: 4.4.2-4
Severity: important
*** Please type your report below this line ***
With the following simple test file, the 3DNow! generated code is
incorrect, with final _m_pfsub generated as pfsubr %mm0,%mm0, thus
producing completely incorrect result.
*** main.c
#include
void Butterfly_3(__m64 *D,__m64 SC,unsigned int IStep,unsigned int Off)
{ __m64 T,T1,T2;
T=_m_pfmul(D[Off+1],SC);
T1=D[Off+0];D[Off+0]=_m_pfadd(T1,T);D[Off+1]=_m_pfsub(T1,T);
}
***
The command line is:
gcc -march=athlon -S -O3 main.c
The generated code is:
*** main.s
.file"main.c"
.text
.p2align 4,,15
.globl Butterfly_3
.typeButterfly_3, @function
Butterfly_3:
pushl%ebp
movl%esp, %ebp
movl8(%ebp), %edx
movl16(%ebp), %ecx
leal8(%edx,%ecx,8), %eax
leal(%edx,%ecx,8), %edx
pfmul(%eax), %mm0
movq(%edx), %mm1
movq%mm1, %mm2
pfadd%mm0, %mm2
pfsubr%mm0, %mm0
movq%mm2, (%edx)
movq%mm0, (%eax)
leave
ret
.sizeButterfly_3, .-Butterfly_3
.ident"GCC: (Debian 4.4.2-4) 4.4.2"
.section.note.GNU-stack,"",@progbits
***
The problem exists in all current Debian releases (stable, testing,
unstable).
Without optimizations the problem is still there, but it's a lot more
difficult to follow.
This is the simplest test code that I could extract. The basic principle
is used in a large
cross-platform library and all GCC variants for x86 that I have used
(Debian, Cygwin, MinGW, TDM)
generate essentially the same code.
-- System Information:
Debian Release: squeeze/sid
APT prefers unstable
APT policy: (500, 'unstable')
Architecture: i386 (i686)
Kernel: Linux 2.6.31-1-686 (SMP w/1 CPU core)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Versions of packages gcc-4.4 depends on:
ii binutils 2.20-4 The GNU assembler, linker
and bina
ii cpp-4.4 4.4.2-4The GNU C preprocessor
ii gcc-4.4-base 4.4.2-4The GNU Compiler Collection
(base
ii libc6 2.10.2-2 GNU C Library: Shared libraries
ii libgcc1 1:4.4.2-4 GCC support library
ii libgomp1 4.4.2-4GCC OpenMP (GOMP) support
library
Versions of packages gcc-4.4 recommends:
ii libc6-dev 2.10.2-2 GNU C Library: Development
Librari
Versions of packages gcc-4.4 suggests:
pn gcc-4.4-doc(no description available)
pn gcc-4.4-locales(no description available)
pn gcc-4.4-multilib (no description available)
pn libcloog-ppl0 (no description available)
pn libgcc1-dbg(no description available)
pn libgomp1-dbg (no description available)
pn libmudflap0-4.4-dev(no description available)
pn libmudflap0-dbg(no description available)
pn libppl-c2 (no description available)
pn libppl7(no description available)
-- no debconf information
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#560282: gcc-4.4: cc1 crashes with simple test file
Please consider this one closed! In the end it proved to be a CIFS
related problem. By copying the test file on a local file system
everything works as far as this bug is concerned. Even GCC 4.3.4 that
previously worked is now failing with the same rather strange error.
Matthias Klose wrote:
> severity 560282 normal
> thanks
>
> builds without errors. closing. please reopen if you can reproduce
> this on another machine, and provide maybe a complete example which
> shows the error.
>
> On 10.12.2009 09:47, Catalin Ionescu wrote:
>> Package: gcc-4.4
>> Version: 4.4.2-3
>> Severity: critical
>>
>>
>> When trying to compile the following very simple test file:
>>
>> *** main.c
>> #include
>>
>> void Butterfly_3(__m64 *D,__m64 SC,unsigned int IStep,unsigned int Off)
>> { __m64 T,T1,T2;
>>
>>T=_m_pfmul(D[Off+1],SC);
>>T1=D[Off+0];D[Off+0]=_m_pfadd(T1,T);D[Off+1]=_m_pfsub(T1,T);
>> }
>> ***
>>
>> The following error appears:
>>
>> cc1: error: main.c: Value too large for defined data type
>>
>> The command line is:
>>
>> gcc -march=athlon -fverbose-asm -S -O3 main.c
>>
>> The same source file compiles with GCC 4.3.2 and GCC 4.3.4 as they
>> are in fully
>> updated Debian stable and testing releases.
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#560282: gcc-4.4: cc1 crashes with simple test file
With fresh install of Debian Unstable the error is still present. By using: gcc -march=athlon -fverbose-asm -S -O3 -save-temps main.c all I see is an empty (0 bytes) main.i file and compile process still ends with "cc1: error: main.c: Value too large for defined data type" -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#560282: gcc-4.4: cc1 crashes with simple test file
Right now I see gcc 4.4.2-4 available for Debian unstable. I have upgraded the compilers but I still get: "cc1: error: main.c: Value too large for defined data type" This is a Virtual Box installed Debian unstable. With older GCC versions, 4.3.2 (Debian stable) and 4.3.4 (Debian testing) the same code produced incorrect output, still on Virtual Box machines. I will try a new, clean install of Debian unstable and see if I can still reproduce this bug. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#560282: gcc-4.4: cc1 crashes with simple test file
Package: gcc-4.4
Version: 4.4.2-3
Severity: critical
When trying to compile the following very simple test file:
*** main.c
#include
void Butterfly_3(__m64 *D,__m64 SC,unsigned int IStep,unsigned int Off)
{ __m64 T,T1,T2;
T=_m_pfmul(D[Off+1],SC);
T1=D[Off+0];D[Off+0]=_m_pfadd(T1,T);D[Off+1]=_m_pfsub(T1,T);
}
***
The following error appears:
cc1: error: main.c: Value too large for defined data type
The command line is:
gcc -march=athlon -fverbose-asm -S -O3 main.c
The same source file compiles with GCC 4.3.2 and GCC 4.3.4 as they are in fully
updated Debian stable and testing releases.
-- System Information:
Debian Release: squeeze/sid
APT prefers unstable
APT policy: (500, 'unstable')
Architecture: i386 (i686)
Kernel: Linux 2.6.31-1-686 (SMP w/1 CPU core)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Versions of packages gcc-4.4 depends on:
ii binutils 2.20-4 The GNU assembler, linker and bina
ii cpp-4.4 4.4.2-3The GNU C preprocessor
ii gcc-4.4-base 4.4.2-3The GNU Compiler Collection (base
ii libc6 2.10.2-2 GNU C Library: Shared libraries
ii libgcc1 1:4.4.2-3 GCC support library
ii libgomp1 4.4.2-3GCC OpenMP (GOMP) support library
Versions of packages gcc-4.4 recommends:
ii libc6-dev 2.10.2-2 GNU C Library: Development Librari
Versions of packages gcc-4.4 suggests:
pn gcc-4.4-doc(no description available)
pn gcc-4.4-locales(no description available)
pn gcc-4.4-multilib (no description available)
pn libcloog-ppl0 (no description available)
pn libgcc1-dbg(no description available)
pn libgomp1-dbg (no description available)
pn libmudflap0-4.4-dev(no description available)
pn libmudflap0-dbg(no description available)
pn libppl-c2 (no description available)
pn libppl7(no description available)
-- no debconf information
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#558245: gcc-4.3: Broken 3DNow! built-ins with optimization turned on
Package: gcc-4.3
Version: 4.3.2-1.1
Severity: important
With the attached source file, the code generated is incorrect for the final
_m_pfsub. The compiler generates things like PFSUBR %MM0,%MM0.
The command line showing the problem is:
gcc -march=athlon -fverbose-asm -S -O3 main.c
Removing '-O3' from the command line generates correct, but not optimized,
code.
The problem is visible in both GCC 4.3.2 (Debian Lenny) and GCC 4.3.4-6
(Debian Squeeze).
*** main.c
#include
void Butterfly_3(__m64 *D,__m64 SC,unsigned int IStep,unsigned int Off)
{ __m64 T,T1,T2;
T=_m_pfmul(D[Off+1],SC);
T1=D[Off+0];D[Off+0]=_m_pfadd(T1,T);D[Off+1]=_m_pfsub(T1,T);
}
-- System Information:
Debian Release: 5.0.3
APT prefers stable
APT policy: (500, 'stable')
Architecture: i386 (i686)
Kernel: Linux 2.6.26-2-686 (SMP w/1 CPU core)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash
Versions of packages gcc-4.3 depends on:
ii binutils2.18.1~cvs20080103-7 The GNU assembler, linker and bina
ii cpp-4.3 4.3.2-1.1The GNU C preprocessor
ii gcc-4.3-base4.3.2-1.1The GNU Compiler Collection (base
ii libc6 2.7-18 GNU C Library: Shared libraries
ii libgcc1 1:4.3.2-1.1 GCC support library
ii libgomp14.3.2-1.1GCC OpenMP (GOMP) support library
Versions of packages gcc-4.3 recommends:
ii libc6-dev 2.7-18 GNU C Library: Development Librari
Versions of packages gcc-4.3 suggests:
pn gcc-4.3-doc(no description available)
pn gcc-4.3-locales(no description available)
pn gcc-4.3-multilib (no description available)
pn libgcc1-dbg(no description available)
pn libgomp1-dbg (no description available)
pn libmudflap0-4.3-dev(no description available)
pn libmudflap0-dbg(no description available)
-- no debconf information
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
