 Lucas Schnorr
|
2c31927bd4
back port stream id creation to non-poti trace export
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
f6e037da6c
If CUDA, workers might be streams, so create an unique name for each of them
|
%!s(int64=8) %!d(string=hai) anos |
 Corentin Salingue
|
4448408f1f
unregister pointer when reusing memory chunk
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
8472efdd87
there is no need to set these variables to zero when we are using the PajeSetVariable event only
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
1c1abe5128
simplify the code again, using a local variable to contain the current timestamp (now)
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
5dd23dc290
the "user event type" type is child of P, not T (as far as the rest of related tracing code is correct)
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
5fa0f17ad2
when using poti, output the detailed state when issuing a PajeSetState
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
64692de8eb
declare the extended PajeSetState with all 11 extra parameters, save the event identifier
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
063aaf82dd
only set (again?) the gf variable if the last_end_codelet_time is different of zero
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
29d658eeb9
the gf variable should be set to zero immediately after its container creation
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
89cd9ecdf1
simplify code with a local variable
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
bbd1cdddf1
use now to timestamp these events, since their containers have just been created
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
a0104c0146
use a variable to keep the current time of the event
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
54c6f96a61
when creating the mpi communication thread, adopt alias as name in poti
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
47ac7b3d9c
set to NULL
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
5a4ed50582
fix MPI link trace generation with poti
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
8b72dacf88
fix order of parameters ("MPIL" is the container)
|
%!s(int64=8) %!d(string=hai) anos |
 Samuel Thibault
|
1704a40942
Add missing documentation
|
%!s(int64=8) %!d(string=hai) anos |
|
Samuel Thibault
|
2905b3eb1c
Release the mc spinlock while freeing data, otherwise cudaFree might cause the whole StarPU to get stuck
|
%!s(int64=8) %!d(string=hai) anos |
|
Samuel Thibault
|
96fbf01691
Make the reclaim test trigger flushing the allocation cache by using different sizes and unregistering some data early
|
%!s(int64=8) %!d(string=hai) anos |
|
Samuel Thibault
|
4db3d8aa3e
Fix spurious change
|
%!s(int64=8) %!d(string=hai) anos |
 Nathalie Furmento
|
f9e5b1da3a
tools: (really) add script file for sonarqube
|
%!s(int64=8) %!d(string=hai) anos |
|
Nathalie Furmento
|
f70fbb13f1
tools: add script file for sonarqube
|
%!s(int64=8) %!d(string=hai) anos |
|
Nathalie Furmento
|
4dd7eb7df0
src/drivers/scc/driver_scc_sink.c: fix assignment call
|
%!s(int64=8) %!d(string=hai) anos |
|
Samuel Thibault
|
91cdd8d90e
Avoid undefined void* arithmetic
|
%!s(int64=8) %!d(string=hai) anos |
|
Samuel Thibault
|
7c9982e990
Fix build against older poti versions
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
7c2881f8e3
Revert "register thread might also return 0 if the tid has already been declared"
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
992c749394
Revert "Comment codelet details for now when using libpoti, until emitting them is implemented"
|
%!s(int64=8) %!d(string=hai) anos |
|
Lucas Schnorr
|
1167fb3a7e
register thread might also return 0 if the tid has already been declared
|
%!s(int64=8) %!d(string=hai) anos |
|
Nathalie Furmento
|
4b7f8373b8
tools: re-organize directories
|
%!s(int64=8) %!d(string=hai) anos |
