 Nathalie Furmento
|
92dddf41f5
core/perfmodel/perfmodel_bus.c: skip OpenCL CPU devices
|
il y a 13 ans |
 Samuel Thibault
|
915f4f78dc
Use hwloc to save/restore cpubind
|
il y a 13 ans |
|
Nathalie Furmento
|
f9a39d6fcb
src/core/perfmodel/perfmodel_bus.c: initialise to calm down valgrind
|
il y a 13 ans |
 Cyril Roelandt
|
43fc35e2a5
load_bus_affinity_file_content: do not try to open the affinity file if neither CUDA or OpenCL are defined.
|
il y a 13 ans |
|
Cyril Roelandt
|
f00cc9ea86
write_bus_affinity_file_content: Do not open the file if there is nothing to write.
|
il y a 13 ans |
|
Nathalie Furmento
|
e5f68ad9d6
src: update double values usage
|
il y a 13 ans |
|
Nathalie Furmento
|
5908a48866
src: let's make icc happy
|
il y a 13 ans |
|
Cyril Roelandt
|
b6df6beea3
Renaming public functions :
|
il y a 13 ans |
|
Cyril Roelandt
|
99465346a7
src/ : repalce calls to assert() by calls to STARPU_ASSERT().
|
il y a 14 ans |
|
Nathalie Furmento
|
3ad31951c8
src: code cleaning
|
il y a 14 ans |
|
Nathalie Furmento
|
0bd2fe9020
src: rename internal types following coding conventions
|
il y a 14 ans |
|
Nathalie Furmento
|
87ad636148
Type renaming:
|
il y a 14 ans |
|
Nathalie Furmento
|
96e3246e1c
src/core/perfmodel/perfmodel_bus.c: enable code only when hwloc is enabled
|
il y a 14 ans |
|
Cyril Roelandt
|
dcdb5343ce
measure_bandwidth_between_cpus_and_dev() : dynamically allocating arrays, so that we do not statically declare an array of size 0.
|
il y a 14 ans |
|
Samuel Thibault
|
5dc8c4d367
Fix warning
|
il y a 14 ans |
|
Nathalie Furmento
|
8afb82815b
src/core/perfmodel/perfmodel_bus.c: improve error checking for cudaHostAlloc
|
il y a 14 ans |
|
Samuel Thibault
|
907bfdd621
avoid measuring transfer inside GPU
|
il y a 14 ans |
|
Samuel Thibault
|
14f2b383d8
Fix bandwidth measurement when gpu sizes are differing and small: make timing µs per byte instead of just µs
|
il y a 14 ans |
|
Samuel Thibault
|
2eba67b128
Add GPU-GPU transfer measurement
|
il y a 14 ans |
|
Samuel Thibault
|
2b6c0e9af5
Make sure the topology is initialized when benchmarking the bus performance
|
il y a 14 ans |
|
Cyril Roelandt
|
2b60aedf3b
Replacing _starpu_topology_get_nhwcpu(config) by config->topology.ncpus
|
il y a 14 ans |
|
Cyril Roelandt
|
b29709b352
src/core/perfmodel/perfmodel_bus.c:replacing cudaGetDeviceCount by _starpu_get_cuda_device_count
|
il y a 14 ans |
|
Samuel Thibault
|
4a705e980b
fix format breach issue
|
il y a 14 ans |
|
Samuel Thibault
|
45b4b54f8e
Give user more patience while calibrating the bus.
|
il y a 14 ans |
|
Samuel Thibault
|
319454e08d
Fix CUDA/OpenCL bandwidth computation
|
il y a 14 ans |
|
Samuel Thibault
|
f351cf2d3f
StarPU does not yet measure GPU-GPU transfers
|
il y a 14 ans |
|
Samuel Thibault
|
47b21020da
print bandwidths too in starpu_machine_display
|
il y a 14 ans |
|
Samuel Thibault
|
d8aa7f19fa
use %f and %e instead of %lf and %le, as the latter are not defined by the ANSI standard, and have varying behavior depending on GNU or MSVC runtime
|
il y a 14 ans |
|
Samuel Thibault
|
7ff23acdf9
drop unrelated changes
|
il y a 14 ans |
|
Samuel Thibault
|
2668a39263
Fix gotoblas detection
|
il y a 14 ans |
