|
|
@@ -1228,16 +1228,16 @@ Partitioning can be applied several times, see
|
|
|
@section Performance model example
|
|
|
|
|
|
To achieve good scheduling, StarPU scheduling policies need to be able to
|
|
|
-estimate in advance the duration of a task. This is done by giving to codelets a
|
|
|
-performance model. There are several kinds of performance models.
|
|
|
+estimate in advance the duration of a task. This is done by giving to codelets
|
|
|
+a performance model, by defining a @code{starpu_perfmodel_t} structure and
|
|
|
+providing its address in the @code{model} field of the @code{starpu_codelet}
|
|
|
+structure. The @code{symbol} and @code{type} fields of @code{starpu_perfmodel_t}
|
|
|
+are mandatory, to give a name to the model, and the type of the model, since
|
|
|
+there are several kinds of performance models.
|
|
|
|
|
|
@itemize
|
|
|
@item
|
|
|
-Providing an estimation from the application itself (@code{STARPU_COMMON} model type and @code{cost_model} field),
|
|
|
-see for instance
|
|
|
-@code{examples/common/blas_model.h} and @code{examples/common/blas_model.c}. It can also be provided for each architecture (@code{STARPU_PER_ARCH} model type and @code{per_arch} field)
|
|
|
-@item
|
|
|
-Measured at runtime (STARPU_HISTORY_BASED model type). This assumes that for a
|
|
|
+Measured at runtime (@code{STARPU_HISTORY_BASED} model type). This assumes that for a
|
|
|
given set of data input/output sizes, the performance will always be about the
|
|
|
same. This is very true for regular kernels on GPUs for instance (<0.1% error),
|
|
|
and just a bit less true on CPUs (~=1% error). This also assumes that there are
|
|
|
@@ -1277,7 +1277,7 @@ starpu_codelet cl = @{
|
|
|
@end cartouche
|
|
|
|
|
|
@item
|
|
|
-Measured at runtime and refined by regression (STARPU_REGRESSION_*_BASED
|
|
|
+Measured at runtime and refined by regression (@code{STARPU_REGRESSION_*_BASED}
|
|
|
model type). This still assumes performance regularity, but can work
|
|
|
with various data input sizes, by applying regression over observed
|
|
|
execution times. STARPU_REGRESSION_BASED uses an a*n^b regression
|
|
|
@@ -1287,7 +1287,12 @@ STARPU_REGRESSION_BASED, but costs a lot more to compute). For instance,
|
|
|
model for the @code{memset} operation.
|
|
|
|
|
|
@item
|
|
|
-Provided explicitly by the application (STARPU_PER_ARCH model type): the
|
|
|
+Provided as an estimation from the application itself (@code{STARPU_COMMON} model type and @code{cost_model} field),
|
|
|
+see for instance
|
|
|
+@code{examples/common/blas_model.h} and @code{examples/common/blas_model.c}.
|
|
|
+
|
|
|
+@item
|
|
|
+Provided explicitly by the application (@code{STARPU_PER_ARCH} model type): the
|
|
|
@code{.per_arch[i].cost_model} fields have to be filled with pointers to
|
|
|
functions which return the expected duration of the task in micro-seconds, one
|
|
|
per architecture.
|
