[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122 SergeyN nenakhov.sergey at gmail dot com changed: What|Removed |Added CC||nenakhov.sergey at gmail ||dot com --- Comment #6 from SergeyN nenakhov.sergey at gmail dot com 2012-09-05 07:42:49 UTC --- well, I get similiar error when I try to overload operators , =, ==, etc error: ‘__m256d operator(__m256d, __m256d)’ must have an argument of class or enumerated type.
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122 --- Comment #7 from Marc Glisse glisse at gcc dot gnu.org 2012-09-05 09:32:11 UTC --- http://gcc.gnu.org/ml/gcc-patches/2012-08/msg02098.html
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122 --- Comment #8 from SergeyN nenakhov.sergey at gmail dot com 2012-09-05 09:42:58 UTC --- (In reply to comment #7) http://gcc.gnu.org/ml/gcc-patches/2012-08/msg02098.html That's nice, but I would really prefer to define my own comparison operator, because the actual comparison instruction for __m256 and __m256d types has a way to control handling of nans, which wouldn't be allowed by a built-in comparison operator.
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added CC||jakub at gcc dot gnu.org --- Comment #9 from Jakub Jelinek jakub at gcc dot gnu.org 2012-09-05 12:13:55 UTC --- Then put it into a class and add overloaded comparison operators for the wrapper class. That is the same thing as with float/double, you can't overload float/double comparison operator either.
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122 --- Comment #10 from SergeyN nenakhov.sergey at gmail dot com 2012-09-05 12:24:49 UTC --- (In reply to comment #9) Then put it into a class and add overloaded comparison operators for the wrapper class. That is the same thing as with float/double, you can't overload float/double comparison operator either. Well, I'll of course find a way to workaround this issue by either using a class, or using a separate function like cmp_less or something like that. The point of this thread is that operator overloading for vector types doesn't not work in gcc, and not how to work around that (which I'm sure people have no problem doing given that the problem was reported 2 years ago). Sergey.
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
-- paolo dot carlini at oracle dot com changed: What|Removed |Added Severity|blocker |normal http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
--- Comment #1 from pinskia at gmail dot com 2010-02-19 16:05 ---
Subject: Re: New: g++ does not allow overloading operators for sse types
(__m128, __m128d)
Well there is already a builtin operator+ for vector types with the
generic vector support.
Sent from my iPhone
On Feb 19, 2010, at 7:50 AM, mr dot nuke dot me at gmail dot com
gcc-bugzi...@gcc.gnu.org
wrote:
When trying to overload operators for __m128 and __m128d types. For
example,
the following code:
#include xmmintrin.h
#include stddef.h
inline __m128 operator + (const __m128 A, const __m128 B)
{
return _mm_add_ps(A, B);
}
int main()
{
float a[4];
float b[4];
for(size_t i = 0; i 4; i++)
{
a[i] = 2 * i;
b[i] = 2 * i;
}
__m128 sseA, sseB;
sseA = _mm_loadu_ps(a);
sseB = _mm_loadu_ps(b);
__m128 sseSum = sseA + sseB;
return 0;
}
Generates
error: 'float __vector__ operator+(float __vector__, float
__vector__)' must
have an argument of class or enumerated type
Compiler options: -O3 -msse
Both ICL (windows and Linux) and MSVC accept this code and compile it
correctly.
--
Summary: g++ does not allow overloading operators for sse
types
(__m128, __m128d)
Product: gcc
Version: 4.4.3
Status: UNCONFIRMED
Severity: blocker
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: mr dot nuke dot me at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
--- Comment #2 from mr dot nuke dot me at gmail dot com 2010-02-19 18:34 --- I'm assuming you are referring to fvec.h and F32vec4. While I will apply this workaround (one of many many workarounds to get fancy C++ code to work with g++) it still doesn't solve the underlying problem, and affects portability performance-wise, something lethal for HPC code. Alex -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
--- Comment #4 from rguenth at gcc dot gnu dot org 2010-02-19 18:51 --- FYI, this is 15.3/6. __m128 isn't a class nor an enumeration. Maybe it is for the other compilers as implementation detail. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Component|target |c++ http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122
[Bug c++/43122] g++ does not allow overloading operators for sse types (__m128, __m128d)
--- Comment #5 from mr dot nuke dot me at gmail dot com 2010-02-19 19:15 --- Which means there is already an operator+ defined which is done using the generic GNU vector support. If you remove your definition of operator+, it works without it. I could have sworn I saw error: no operator + matches these operands. when I removed the operator definition (in a larger code base though). The error I saw also has the message: operand types are: __m128 + const __m128 I cannot reproduce this on the small test case I posted. I'm working on a repro case. Alex -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43122
