mirror of
https://gitlab.com/libeigen/eigen.git
synced 2026-04-10 11:34:33 +08:00
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:
@@ -78,13 +78,18 @@ bool isApproxAbs(const Scalar& a, const Scalar& b, const typename NumTraits<Scal
|
||||
return internal::isMuchSmallerThan(a-b, refvalue);
|
||||
}
|
||||
|
||||
template<typename Scalar>
|
||||
inline void print_mismatch(const Scalar* ref, const Scalar* vec, int size) {
|
||||
std::cout << "ref: [" << Map<const Matrix<Scalar,1,Dynamic> >(ref,size) << "]" << " != vec: [" << Map<const Matrix<Scalar,1,Dynamic> >(vec,size) << "]\n";
|
||||
}
|
||||
|
||||
template<typename Scalar> bool areApproxAbs(const Scalar* a, const Scalar* b, int size, const typename NumTraits<Scalar>::Real& refvalue)
|
||||
{
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
if (!isApproxAbs(a[i],b[i],refvalue))
|
||||
{
|
||||
std::cout << "ref: [" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != vec: [" << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "]\n";
|
||||
print_mismatch(a, b, size);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
@@ -95,13 +100,23 @@ template<typename Scalar> bool areApprox(const Scalar* a, const Scalar* b, int s
|
||||
{
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
if (a[i]!=b[i] && !internal::isApprox(a[i],b[i]))
|
||||
if ( a[i]!=b[i] && !internal::isApprox(a[i],b[i])
|
||||
&& !((numext::isnan)(a[i]) && (numext::isnan)(b[i])) )
|
||||
{
|
||||
if((numext::isnan)(a[i]) && (numext::isnan)(b[i]))
|
||||
{
|
||||
continue;
|
||||
}
|
||||
std::cout << "ref: [" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != vec: [" << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "]\n";
|
||||
print_mismatch(a, b, size);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
template<typename Scalar> bool areEqual(const Scalar* a, const Scalar* b, int size)
|
||||
{
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
if ( (a[i] != b[i]) && !((numext::isnan)(a[i]) && (numext::isnan)(b[i])) )
|
||||
{
|
||||
print_mismatch(a, b, size);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
@@ -178,6 +193,14 @@ struct packet_helper<false,Packet>
|
||||
VERIFY(test::areApprox(ref, data2, PacketSize) && #POP); \
|
||||
}
|
||||
|
||||
#define CHECK_CWISE1_EXACT_IF(COND, REFOP, POP) if(COND) { \
|
||||
test::packet_helper<COND,Packet> h; \
|
||||
for (int i=0; i<PacketSize; ++i) \
|
||||
ref[i] = Scalar(REFOP(data1[i])); \
|
||||
h.store(data2, POP(h.load(data1))); \
|
||||
VERIFY(test::areEqual(ref, data2, PacketSize) && #POP); \
|
||||
}
|
||||
|
||||
#define CHECK_CWISE2_IF(COND, REFOP, POP) if(COND) { \
|
||||
test::packet_helper<COND,Packet> h; \
|
||||
for (int i=0; i<PacketSize; ++i) \
|
||||
|
||||
Reference in New Issue
Block a user