Fix rint SSE/NEON again, using optimization barrier.

This is a new version of !423, which failed for MSVC.

Defined `EIGEN_OPTIMIZATION_BARRIER(X)` that uses inline assembly to
prevent operations involving `X` from crossing that barrier. Should
work on most `GNUC` compatible compilers (MSVC doesn't seem to need
this). This is a modified version adapted from what was used in
`psincos_float` and tested on more platforms
(see #1674, https://godbolt.org/z/73ezTG).

Modified `rint` to use the barrier to prevent the add/subtract rounding
trick from being optimized away.

Also fixed an edge case for large inputs that get bumped up a power of two
and ends up rounding away more than just the fractional part.  If we are
over `2^digits` then just return the input.  This edge case was missed in
the test since the test was comparing approximate equality, which was still
satisfied.  Adding a strict equality option catches it.
This commit is contained in:
Antonio Sanchez
2021-03-03 19:22:15 -08:00
parent 5f0b4a4010
commit 82d61af3a4
6 changed files with 172 additions and 50 deletions

View File

@@ -630,14 +630,6 @@ __attribute__((optimize("-fno-unsafe-math-optimizations")))
#endif
Packet psincos_float(const Packet& _x)
{
// Workaround -ffast-math aggressive optimizations
// See bug 1674
#if EIGEN_COMP_CLANG && defined(EIGEN_VECTORIZE_SSE)
#define EIGEN_SINCOS_DONT_OPT(X) __asm__ ("" : "+x" (X));
#else
#define EIGEN_SINCOS_DONT_OPT(X)
#endif
typedef typename unpacket_traits<Packet>::integer_packet PacketI;
const Packet cst_2oPI = pset1<Packet>(0.636619746685028076171875f); // 2/PI
@@ -652,7 +644,7 @@ Packet psincos_float(const Packet& _x)
// Rounding trick:
Packet y_round = padd(y, cst_rounding_magic);
EIGEN_SINCOS_DONT_OPT(y_round)
EIGEN_OPTIMIZATION_BARRIER(y_round)
PacketI y_int = preinterpret<PacketI>(y_round); // last 23 digits represent integer (if abs(x)<2^24)
y = psub(y_round, cst_rounding_magic); // nearest integer to x*4/pi
@@ -674,9 +666,9 @@ Packet psincos_float(const Packet& _x)
// and 2 ULP up to:
const float huge_th = ComputeSine ? 25966.f : 18838.f;
x = pmadd(y, pset1<Packet>(-1.5703125), x); // = 0xbfc90000
EIGEN_SINCOS_DONT_OPT(x)
EIGEN_OPTIMIZATION_BARRIER(x)
x = pmadd(y, pset1<Packet>(-0.000483989715576171875), x); // = 0xb9fdc000
EIGEN_SINCOS_DONT_OPT(x)
EIGEN_OPTIMIZATION_BARRIER(x)
x = pmadd(y, pset1<Packet>(1.62865035235881805419921875e-07), x); // = 0x342ee000
x = pmadd(y, pset1<Packet>(5.5644315544167710640977020375430583953857421875e-11), x); // = 0x2e74b9ee
@@ -753,8 +745,6 @@ Packet psincos_float(const Packet& _x)
// Update the sign and filter huge inputs
return pxor(y, sign_bit);
#undef EIGEN_SINCOS_DONT_OPT
}
template<typename Packet>