 Samuel Thibault
|
aeff50217e
Accept going outside of benchmarked nl regression by 10%. Force calibration otherwise
|
14 anni fa |
|
Samuel Thibault
|
d963e59e8d
starpu_perfmodel_display can now access to history using the public interface
|
14 anni fa |
|
Samuel Thibault
|
2b4e61f351
Always make sure that starpu.h gets included first, since that one knows which order has to be used for the rest. Expose history entries to applications.
|
14 anni fa |
 Nathalie Furmento
|
e482ecdf0a
doc: use @var instead of @code when appropriate
|
14 anni fa |
 Cyril Roelandt
|
4c1d6fa28d
Avoid mixing signed and unsigned (cf. r4584)
|
14 anni fa |
 Olivier Aumage
|
fc07bb4782
- avoid mixing signed and unsigned
|
14 anni fa |
|
Cyril Roelandt
|
dc3ce715e4
Fix the SSE example.
|
14 anni fa |
|
Nathalie Furmento
|
3d931891f4
configure.ac: when CUDA is enabled, enable ICC only if it is compatible with CUDA
|
14 anni fa |
|
Samuel Thibault
|
b0bbe73041
include source headers before system headers, to fix build when older starpu is already installed
|
14 anni fa |
|
Samuel Thibault
|
6c4cb74536
Start plot at 1
|
14 anni fa |
|
Samuel Thibault
|
bf9bc759c4
Document that non-linear regression reverts to history-based on first run
|
14 anni fa |
|
Samuel Thibault
|
4b2f334cc5
fix plot when calibrations were not in the same order
|
14 anni fa |
|
Samuel Thibault
|
089b5aca13
Make nl regression revert to history-based when it is not valid yet, but enough history is available. This permits nice auto-calibration
|
14 anni fa |
|
Samuel Thibault
|
e5c40b85fd
drop spurious debugging print
|
14 anni fa |
|
Samuel Thibault
|
c02c566746
Fix plotting measures for non-linear regression
|
14 anni fa |
|
Samuel Thibault
|
b0b2e42d5f
Only use non-linear regression if the requested size falls inside the covered measures
|
14 anni fa |
|
Samuel Thibault
|
ad7869523c
Add missing piece of commit
|
14 anni fa |
|
Samuel Thibault
|
0a06263c45
Only validate a regression model if we have at least 10% difference in input size. Do not validate the non-linear regression model just due to number of samples.
|
14 anni fa |
|
Samuel Thibault
|
2d8c3bd434
mention which schedulers support choosing the implementation
|
14 anni fa |
|
Samuel Thibault
|
7a0f65790f
Also build vector_scal with icc when available
|
14 anni fa |
|
Samuel Thibault
|
f1cb2d818f
replace tabs with spaces
|
14 anni fa |
 Ludovic Courtès
|
863adbed76
doc: Clarify the wording in "Partitioning Vector Data".
|
14 anni fa |
|
Ludovic Courtès
|
eb918faec0
Remove `doc/version.texi' from the repository.
|
14 anni fa |
|
Ludovic Courtès
|
bc7cb0d9cf
test_interfaces: Remove malloc from the `vector' test; use a descriptive name.
|
14 anni fa |
|
Ludovic Courtès
|
33a2088388
test_interfaces: Change `register_func_t' to return a `starpu_data_handle'.
|
14 anni fa |
|
Ludovic Courtès
|
6d314b81aa
Properly parenthesize `STARPU_MULTIPLE_*_IMPLEMENTATIONS' macros.
|
14 anni fa |
|
Samuel Thibault
|
de83330e6a
We have to also dump empty models, for implementations which aren't calibrated yet
|
14 anni fa |
|
Samuel Thibault
|
c163358f96
Handle non-calibrated implementations
|
14 anni fa |
|
Nathalie Furmento
|
b78b99afa3
doc/chapters/perf-feedback.texi: typo
|
14 anni fa |
|
Nathalie Furmento
|
235b99337f
tools/dev: update w.r.t coccinelle version 1.0.0-rc7
|
14 anni fa |
