devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Replace memset with fill to work for non-trivial scalars.

Browse Source 
For custom scalars, zero is not necessarily represented by
a zeroed-out memory block (e.g. gnu MPFR). We therefore
cannot rely on `memset` if we want to fill a matrix or tensor
with zeroes. Instead, we should rely on `fill`, which for trivial
types does end up getting converted to a `memset` under-the-hood
(at least with gcc/clang).

Requires adding a `fill(begin, end, v)` to `TensorDevice`.

Replaced all potentially bad instances of memset with fill.

Fixes #2245.
This commit is contained in:
Antonio Sanchez
2021-05-11 09:52:00 -07:00
committed by Rasmus Munk Larsen
parent e9c9a3130b
commit 1e6c6c1576
18 changed files with 229 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
unsupported/Eigen/CXX11/src/Tensor/TensorContraction.h
View File

@@ -762,7 +762,7 @@ struct TensorContractionEvaluatorBase : internal::no_assignment_operator
const Index resIncr(1);
// zero out the result buffer (which must be of size at least rows * sizeof(Scalar)
m_device.memset(buffer, 0, rows * sizeof(Scalar));
m_device.fill(buffer, buffer + rows, Scalar(0));
internal::general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,false,RhsScalar,RhsMapper,false>::run(
rows, cols, lhs, rhs,
@@ -869,7 +869,7 @@ struct TensorContractionEvaluatorBase : internal::no_assignment_operator
// If a contraction kernel does not support beta, explicitly initialize
// output buffer with zeroes.
if (!TensorContractionKernel::HasBeta) {
this->m_device.memset(buffer, 0, m * n * sizeof(Scalar));
this->m_device.fill(buffer, buffer + m * n, Scalar(0));
}
for(Index i2=0; i2<m; i2+=mc)