eigen/doc/HiPerformance.dox

129 lines
5.3 KiB
Plaintext
Raw Normal View History

namespace Eigen {
2010-07-04 16:37:32 +08:00
/** \page TopicWritingEfficientProductExpression Writing efficient matrix product expressions
2009-07-28 18:08:26 +08:00
In general achieving good performance with Eigen does no require any special effort:
simply write your expressions in the most high level way. This is especially true
2010-07-04 16:37:32 +08:00
for small fixed size matrices. For large matrices, however, it might be useful to
2009-07-28 18:08:26 +08:00
take some care when writing your expressions in order to minimize useless evaluations
and optimize the performance.
In this page we will give a brief overview of the Eigen's internal mechanism to simplify
and evaluate complex product expressions, and discuss the current limitations.
2009-07-28 18:08:26 +08:00
In particular we will focus on expressions matching level 2 and 3 BLAS routines, i.e,
all kind of matrix products and triangular solvers.
Indeed, in Eigen we have implemented a set of highly optimized routines which are very similar
to BLAS's ones. Unlike BLAS, those routines are made available to user via a high level and
natural API. Each of these routines can compute in a single evaluation a wide variety of expressions.
2010-07-04 16:37:32 +08:00
Given an expression, the challenge is then to map it to a minimal set of routines.
2009-07-28 18:08:26 +08:00
As explained latter, this mechanism has some limitations, and knowing them will allow
you to write faster code by making your expressions more Eigen friendly.
\section GEMM General Matrix-Matrix product (GEMM)
Let's start with the most common primitive: the matrix product of general dense matrices.
In the BLAS world this corresponds to the GEMM routine. Our equivalent primitive can
perform the following operation:
\f$ C.noalias() += \alpha op1(A) op2(B) \f$
2009-07-28 18:08:26 +08:00
where A, B, and C are column and/or row major matrices (or sub-matrices),
alpha is a scalar value, and op1, op2 can be transpose, adjoint, conjugate, or the identity.
When Eigen detects a matrix product, it analyzes both sides of the product to extract a
2010-07-04 16:37:32 +08:00
unique scalar factor alpha, and for each side, its effective storage order, shape, and conjugation states.
2009-07-28 18:08:26 +08:00
More precisely each side is simplified by iteratively removing trivial expressions such as scalar multiple,
2010-07-04 16:37:32 +08:00
negation and conjugation. Transpose and Block expressions are not evaluated and they only modify the storage order
2009-07-28 18:08:26 +08:00
and shape. All other expressions are immediately evaluated.
For instance, the following expression:
2010-07-04 16:37:32 +08:00
\code m1.noalias() -= s4 * (s1 * m2.adjoint() * (-(s3*m3).conjugate()*s2)) \endcode
2009-07-28 18:08:26 +08:00
is automatically simplified to:
2010-07-04 16:37:32 +08:00
\code m1.noalias() += (s1*s2*conj(s3)*s4) * m2.adjoint() * m3.conjugate() \endcode
2009-07-28 18:08:26 +08:00
which exactly matches our GEMM routine.
\subsection GEMM_Limitations Limitations
Unfortunately, this simplification mechanism is not perfect yet and not all expressions which could be
handled by a single GEMM-like call are correctly detected.
<table class="manual" style="width:100%">
2009-07-28 18:08:26 +08:00
<tr>
<th>Not optimal expression</th>
<th>Evaluated as</th>
<th>Optimal version (single evaluation)</th>
<th>Comments</th>
2009-07-28 18:08:26 +08:00
</tr>
<tr>
2010-07-04 16:37:32 +08:00
<td>\code
m1 += m2 * m3; \endcode</td>
<td>\code
temp = m2 * m3;
m1 += temp; \endcode</td>
<td>\code
m1.noalias() += m2 * m3; \endcode</td>
<td>Use .noalias() to tell Eigen the result and right-hand-sides do not alias.
Otherwise the product m2 * m3 is evaluated into a temporary.</td>
2009-07-28 18:08:26 +08:00
</tr>
<tr class="alt">
<td></td>
<td></td>
2010-07-04 16:37:32 +08:00
<td>\code
m1.noalias() += s1 * (m2 * m3); \endcode</td>
<td>This is a special feature of Eigen. Here the product between a scalar
and a matrix product does not evaluate the matrix product but instead it
returns a matrix product expression tracking the scalar scaling factor. <br>
Without this optimization, the matrix product would be evaluated into a
temporary as in the next example.</td>
2009-07-28 18:08:26 +08:00
</tr>
<tr>
2010-07-04 16:37:32 +08:00
<td>\code
m1.noalias() += (m2 * m3).adjoint(); \endcode</td>
<td>\code
temp = m2 * m3;
m1 += temp.adjoint(); \endcode</td>
<td>\code
m1.noalias() += m3.adjoint()
* * m2.adjoint(); \endcode</td>
<td>This is because the product expression has the EvalBeforeNesting bit which
enforces the evaluation of the product by the Tranpose expression.</td>
2009-07-28 23:11:15 +08:00
</tr>
<tr class="alt">
2010-07-04 16:37:32 +08:00
<td>\code
m1 = m1 + m2 * m3; \endcode</td>
<td>\code
temp = m2 * m3;
m1 = m1 + temp; \endcode</td>
<td>\code m1.noalias() += m2 * m3; \endcode</td>
2009-07-28 18:08:26 +08:00
<td>Here there is no way to detect at compile time that the two m1 are the same,
and so the matrix product will be immediately evaluated.</td>
</tr>
<tr>
2010-07-04 16:37:32 +08:00
<td>\code
m1.noalias() = m4 + m2 * m3; \endcode</td>
<td>\code
temp = m2 * m3;
m1 = m4 + temp; \endcode</td>
<td>\code
m1 = m4;
m1.noalias() += m2 * m3; \endcode</td>
<td>First of all, here the .noalias() in the first expression is useless because
m2*m3 will be evaluated anyway. However, note how this expression can be rewritten
2010-07-04 16:37:32 +08:00
so that no temporary is required. (tip: for very small fixed size matrix
it is slighlty better to rewrite it like this: m1.noalias() = m2 * m3; m1 += m4;</td>
</tr>
<tr class="alt">
2010-07-04 16:37:32 +08:00
<td>\code
m1.noalias() += (s1*m2).block(..) * m3; \endcode</td>
<td>\code
temp = (s1*m2).block(..);
m1 += temp * m3; \endcode</td>
<td>\code
m1.noalias() += s1 * m2.block(..) * m3; \endcode</td>
2009-07-28 23:11:15 +08:00
<td>This is because our expression analyzer is currently not able to extract trivial
expressions nested in a Block expression. Therefore the nested scalar
multiple cannot be properly extracted.</td>
2009-07-28 18:08:26 +08:00
</tr>
</table>
Of course all these remarks hold for all other kind of products involving triangular or selfadjoint matrices.
*/
}