[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #1 from Iain Buclaw --- To be fair, C++ does the same as well. The problem is the inliner (though no inlining occurs when using immintrin.h)
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #2 from Guillaume Piolat --- It seems it is the same as https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55752 ?
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 Jakub Jelinek changed: What|Removed |Added CC||jakub at gcc dot gnu.org --- Comment #3 from Jakub Jelinek --- The problem isn't inlining, but the fact that the rounding mode isn't modeled in the IL, with dependencies between changing of that (or function calls that could have changed it) and all the floating point arithmetic operations that can be affected by that. So, one needs to use explicit optimization barriers (inline asm, separate functions with disabled inlining, etc.).
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #4 from Guillaume Piolat --- So, I need to disable inlining for _mm_setcsr and that will be enough?
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #5 from Jakub Jelinek --- No. You need to put the code that is done with different rounding modes into separate functions (and ensure those functions aren't inlined into their caller).
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #6 from Guillaume Piolat --- I provide intel intrinsics API for D, these intrinsics (like in C++) allows to change the rounding mode, and about 200 other functions that depend on the rounding mode can be called. I'm not the one in control of who calls what. I don't see a workaround for C++ headers at https://github.com/gcc-mirror/gcc/blob/16e2427f50c208dfe07d07f18009969502c25dc8/gcc/config/i386/xmmintrin.h#L872
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #7 from Guillaume Piolat --- https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47617 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=34678
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607
--- Comment #8 from Iain Buclaw ---
(In reply to Guillaume Piolat from comment #6)
> I provide intel intrinsics API for D, these intrinsics (like in C++) allows
> to change the rounding mode, and about 200 other functions that depend on
> the rounding mode can be called. I'm not the one in control of who calls
> what.
>
> I don't see a workaround for C++ headers at
> https://github.com/gcc-mirror/gcc/blob/
> 16e2427f50c208dfe07d07f18009969502c25dc8/gcc/config/i386/xmmintrin.h#L872
Ah, I was mistaken last night. Probably my quick conversion was wrong in some
way that prevented inlining. On a second look, you'll find nothing in
xmmintrin.h that'll help you.
---
#include
int main()
{
unsigned savedRounding = _MM_GET_ROUNDING_MODE();
_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);
union {
__m128i vec;
int array[4];
} A;
A.vec = _mm_cvtps_epi32(_mm_setr_ps(1.4f, -2.1f, 53.5f, -2.9f));
assert(A.array[0] == 1);
assert(A.array[1] == -2);
assert(A.array[2] == 54);
assert(A.array[3] == -3);
// GCC might merge this branch with above! Despite _MM_SET_ROUNDING_MODE
// not being pure.
_MM_SET_ROUNDING_MODE(_MM_ROUND_DOWN);
A.vec = _mm_cvtps_epi32(_mm_setr_ps(1.4f, -2.1f, 53.5f, -2.9f));
assert(A.array[0] == 1);
assert(A.array[1] == -3); // fails
assert(A.array[2] == 53);
assert(A.array[3] == -3);
return 0;
}
---
One workaround for the intel-intrinsics D library might be to insert a barrier
in _MM_SET_ROUNDING_MODE.
---
void _MM_SET_ROUNDING_MODE(int _MM_ROUND_)
{
asm { "" : : : "memory"; }
_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | _MM_ROUND_);
}
---
[Bug d/98607] GDC merging computations but rounding mode has changed
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98607 --- Comment #9 from Marc Glisse --- Since I doubt gdc handles rounding modes correctly for scalars, I think you can ignore this issue in the implementation of the vector intrinsics for now (same as we do in C and C++). Note that gcc isn't alone here, llvm doesn't implement pragma fenv_access either, and even visual studio, which does implement it for scalars, fails for vectors. I did not test with Intel's compiler.
