From 250c261e9e502cf179a1c468ba10e46d53404e9d Mon Sep 17 00:00:00 2001
From: PRUVOST Florent <florent.pruvost@inria.fr>
Date: Fri, 13 Dec 2019 15:10:42 +0100
Subject: [PATCH] Remove the FXT starpu option and the enable tracaing. It is
now done by eztrace, or directly with STARPU_USE_FXT
---
CMakeLists.txt | 37 ---
compute/CMakeLists.txt | 5 -
control/context.c | 2 +
coreblas/eztrace_module/CMakeLists.txt | 61 ++---
...ce_module => coreblas_core_eztrace_module} | 4 +-
.../coreblas_tcore_eztrace_module | 224 ++++++++++++++++++
doc/orgmode/chapters/installing.org | 30 ++-
doc/orgmode/chapters/using.org | 79 +++++-
example/link_chameleon/CMakeLists.txt | 2 +-
include/chameleon/config.h.in | 3 -
new-testing/testing_zauxiliary.c | 46 +++-
new-testing/testing_zauxiliary.h | 58 -----
new-testing/testing_zgetrf.c | 2 -
runtime/openmp/control/runtime_profiling.c | 2 -
runtime/parsec/control/runtime_profiling.c | 2 -
runtime/quark/control/runtime_profiling.c | 2 -
runtime/starpu/control/runtime_control.c | 4 +-
runtime/starpu/control/runtime_profiling.c | 10 +-
runtime/starpu/include/chameleon_starpu.h.in | 1 -
19 files changed, 386 insertions(+), 188 deletions(-)
rename coreblas/eztrace_module/{coreblas_eztrace_module => coreblas_core_eztrace_module} (99%)
create mode 100644 coreblas/eztrace_module/coreblas_tcore_eztrace_module
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 33bce25ca..27a355f8b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -184,13 +184,6 @@ if (CHAMELEON_ENABLE_CUDA AND NOT CHAMELEON_USE_CUDA)
message("-- ${BoldGreen}CHAMELEON_USE_CUDA is set to OFF, turn it ON to use CUDA (unsupported by Quark)${ColourReset}")
endif()
-# Enable FXT if StarPU
-option(CHAMELEON_ENABLE_TRACING "Enable tracing support" OFF)
-if (NOT CHAMELEON_ENABLE_TRACING)
- message("-- ${BoldGreen}CHAMELEON_ENABLE_TRACING is set to OFF, turn it ON to use FxT (with StarPU)${ColourReset}")
-endif()
-#option(CHAMELEON_USE_EZTRACE "Enable EZTRACE to build modules" OFF)
-
option(CHAMELEON_RUNTIME_SYNC "Enable synchronous task submission when available to debug the code without parallelism" OFF)
if (CHAMELEON_RUNTIME_SYNC)
message("-- ${BoldGreen}CHAMELEON_RUNTIME_SYNC is set to ON, turn it OFF to avoid synchronisation in the tasks submission${ColourReset}")
@@ -605,9 +598,6 @@ if( CHAMELEON_SCHED_STARPU )
if(CHAMELEON_USE_MPI)
list(APPEND STARPU_COMPONENT_LIST "MPI")
endif()
- if(CHAMELEON_ENABLE_TRACING)
- list(APPEND STARPU_COMPONENT_LIST "FXT")
- endif()
find_package(STARPU ${CHAMELEON_STARPU_VERSION} REQUIRED
COMPONENTS ${STARPU_COMPONENT_LIST})
@@ -673,18 +663,6 @@ if( CHAMELEON_SCHED_STARPU )
set(CHAMELEON_USE_MIGRATE "OFF")
message("-- ${Blue}CHAMELEON_USE_MIGRATE is turned OFF because starpu_mpi_data_migrate not found${ColourReset}")
endif()
- if(CHAMELEON_ENABLE_TRACING)
- # check if fxt profiling is accessible in starpu and activate it in chameleon
- check_function_exists(starpu_fxt_start_profiling HAVE_STARPU_FXT_PROFILING)
- if ( HAVE_STARPU_FXT_PROFILING )
- message("-- ${Blue}Add definition HAVE_STARPU_FXT_PROFILING"
- " - Activate FxT profiling through StarPU${ColourReset}")
- else()
- message("-- ${Red}Looking for starpu with fxt"
- " - starpu_fxt_start_profiling() test fails in StarPU${ColourReset}")
- message("-- ${Red}Check in CMakeFiles/CMakeError.log to figure out why it fails${ColourReset}")
- endif()
- endif()
if (CHAMELEON_USE_MPI)
# Check if a specific function exist
check_function_exists(starpu_mpi_data_register_comm HAVE_STARPU_MPI_DATA_REGISTER)
@@ -716,14 +694,6 @@ if( CHAMELEON_SCHED_STARPU )
Print_Find_Library_Status(hwloc libhwloc)
endif ()
endif()
- if(CHAMELEON_ENABLE_TRACING AND (NOT FXT_FOUND OR NOT FXT_LIBRARIES))
- if (NOT FXT_fxt.h_DIRS)
- Print_Find_Header_Status(fxt fxt.h)
- endif ()
- if (NOT FXT_fxt_LIBRARY)
- Print_Find_Library_Status(fxt libfxt)
- endif ()
- endif()
if(CHAMELEON_SIMULATION AND (NOT SIMGRID_FOUND OR NOT SIMGRID_LIBRARIES))
if (NOT SIMGRID_simgrid.h_DIRS)
Print_Find_Header_Status(simgrid simgrid.h)
@@ -758,9 +728,6 @@ if( CHAMELEON_SCHED_STARPU )
if(CHAMELEON_SIMULATION AND (NOT SIMGRID_FOUND OR NOT SIMGRID_LIBRARIES))
message(FATAL_ERROR "SimGrid library is required but has not been found")
endif()
- if(CHAMELEON_ENABLE_TRACING AND (NOT FXT_FOUND OR NOT FXT_LIBRARIES))
- message(FATAL_ERROR "FxT library is required but has not been found")
- endif()
if( (NOT STARPU_SHM_FOUND) OR (NOT STARPU_SHM_LIBRARIES) OR
( STARPU_LOOK_FOR_MPI AND (NOT STARPU_MPI_FOUND OR NOT STARPU_MPI_LIBRARIES) )
)
@@ -780,10 +747,6 @@ if( CHAMELEON_SCHED_PARSEC )
if(CHAMELEON_USE_CUDA)
list(APPEND PARSEC_COMPONENT_LIST "CUDA")
endif()
- # TODO: Add a CHAMELEON_WITH_PROFILING option that enables Fxt for StarPU, or PAPI for PaRSEC
- #if(CHAMELEON_WITH_PROFILING)
- # list(APPEND PARSEC_COMPONENT_LIST "PAPI")
- #endif()
find_package(PARSEC COMPONENTS ${PARSEC_COMPONENT_LIST})
if(PARSEC_FOUND)
message("-- ${Blue}Add definition CHAMELEON_SCHED_PARSEC"
diff --git a/compute/CMakeLists.txt b/compute/CMakeLists.txt
index 3ca2aaaa1..b49870887 100644
--- a/compute/CMakeLists.txt
+++ b/compute/CMakeLists.txt
@@ -304,11 +304,6 @@ elseif(CHAMELEON_SCHED_QUARK)
elseif(CHAMELEON_SCHED_OPENMP)
target_link_libraries(chameleon chameleon_openmp)
endif()
-if (NOT CHAMELEON_SIMULATION)
- # Depends on coreblas only for set_coreblas_gemm3m_enabled() (Maybe we should change that)
- add_dependencies(chameleon coreblas_include)
- target_link_libraries(chameleon coreblas)
-endif()
target_link_libraries(chameleon hqr)
add_dependencies(chameleon
diff --git a/control/context.c b/control/context.c
index 4e0a6672b..759716782 100644
--- a/control/context.c
+++ b/control/context.c
@@ -148,6 +148,7 @@ int CHAMELEON_Enable(int option)
break;
case CHAMELEON_PROFILING_MODE:
chamctxt->profiling_enabled = CHAMELEON_TRUE;
+ RUNTIME_start_profiling();
break;
case CHAMELEON_PROGRESS:
chamctxt->progress_enabled = CHAMELEON_TRUE;
@@ -218,6 +219,7 @@ int CHAMELEON_Disable(int option)
break;
case CHAMELEON_PROFILING_MODE:
chamctxt->profiling_enabled = CHAMELEON_FALSE;
+ RUNTIME_stop_profiling();
break;
case CHAMELEON_PROGRESS:
chamctxt->progress_enabled = CHAMELEON_FALSE;
diff --git a/coreblas/eztrace_module/CMakeLists.txt b/coreblas/eztrace_module/CMakeLists.txt
index 810bab667..2850ca003 100644
--- a/coreblas/eztrace_module/CMakeLists.txt
+++ b/coreblas/eztrace_module/CMakeLists.txt
@@ -39,36 +39,39 @@ if (EZTRACE_FOUND AND EZTRACE_DIR_FOUND)
set(EZTRACE_CREATE_PLUGIN "${EZTRACE_eztrace_create_plugin_DIR}/eztrace_create_plugin")
- add_custom_command(
- OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/output
- COMMAND ${EZTRACE_CREATE_PLUGIN}
- ARGS ${CMAKE_CURRENT_SOURCE_DIR}/coreblas_eztrace_module
- DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/coreblas_eztrace_module
- )
- add_custom_target(
- eztrace-module-chameleon_core-dir ALL
- DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/output
- )
- add_custom_command(
- OUTPUT libeztrace-convert-chameleon_core.so
- COMMAND make
- WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/output
- DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/output
- )
- add_custom_target(
- eztrace-module-chameleon_core-libs ALL
- DEPENDS libeztrace-convert-chameleon_core.so
- )
- # installation
- # ------------
- install(
- FILES
- ${CMAKE_CURRENT_BINARY_DIR}/output/libeztrace-autostart-chameleon_core.so
- ${CMAKE_CURRENT_BINARY_DIR}/output/libeztrace-chameleon_core.so
- ${CMAKE_CURRENT_BINARY_DIR}/output/libeztrace-convert-chameleon_core.so
- DESTINATION ${EZTRACE_LIBRARY_DIRS}
- )
+ set( COREBLAS_EZTRACE_MODULES core tcore )
+ foreach( _module ${COREBLAS_EZTRACE_MODULES} )
+ add_custom_command(
+ OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${_module}
+ COMMAND ${EZTRACE_CREATE_PLUGIN} -o ${CMAKE_CURRENT_BINARY_DIR}/${_module} > /dev/null
+ ARGS ${CMAKE_CURRENT_SOURCE_DIR}/coreblas_${_module}_eztrace_module
+ DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/coreblas_${_module}_eztrace_module
+ )
+ add_custom_target(
+ eztrace-module-chameleon_${_module}-dir ALL
+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${_module}
+ )
+ add_custom_command(
+ OUTPUT libeztrace-convert-chameleon_${_module}.so
+ COMMAND make
+ WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${_module}
+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${_module}
+ )
+ add_custom_target(
+ eztrace-module-chameleon_${_module}-libs ALL
+ DEPENDS libeztrace-convert-chameleon_${_module}.so
+ )
+ # installation
+ # ------------
+ install(
+ FILES
+ ${CMAKE_CURRENT_BINARY_DIR}/${_module}/libeztrace-autostart-chameleon_${_module}.so
+ ${CMAKE_CURRENT_BINARY_DIR}/${_module}/libeztrace-chameleon_${_module}.so
+ ${CMAKE_CURRENT_BINARY_DIR}/${_module}/libeztrace-convert-chameleon_${_module}.so
+ DESTINATION ${EZTRACE_LIBRARY_DIRS}
+ )
+ endforeach()
endif (EZTRACE_eztrace_create_plugin_DIR)
endif (EZTRACE_FOUND AND EZTRACE_DIR_FOUND)
diff --git a/coreblas/eztrace_module/coreblas_eztrace_module b/coreblas/eztrace_module/coreblas_core_eztrace_module
similarity index 99%
rename from coreblas/eztrace_module/coreblas_eztrace_module
rename to coreblas/eztrace_module/coreblas_core_eztrace_module
index c93f7a3a4..a9e517548 100644
--- a/coreblas/eztrace_module/coreblas_eztrace_module
+++ b/coreblas/eztrace_module/coreblas_core_eztrace_module
@@ -1,9 +1,7 @@
BEGIN_MODULE
NAME chameleon_core
DESC "Module for Chameleon CORE functions"
-ID 7770
-
-#include <chameleon.h>
+ID 0
void CORE_scasum(int storev, int uplo, int M, int N,
void *A, int lda, float *work);
diff --git a/coreblas/eztrace_module/coreblas_tcore_eztrace_module b/coreblas/eztrace_module/coreblas_tcore_eztrace_module
new file mode 100644
index 000000000..18698b8a1
--- /dev/null
+++ b/coreblas/eztrace_module/coreblas_tcore_eztrace_module
@@ -0,0 +1,224 @@
+BEGIN_MODULE
+NAME chameleon_tcore
+DESC "Module for Chameleon TCORE functions"
+ID 1
+
+
+void TCORE_sasum( int storev, int uplo, int M, int N, const void *A, float *work );
+int TCORE_saxpy( int M, void *alpha, const void *A, int incA, void *B, int incB );
+int TCORE_sgeadd( int trans, int M, int N, void *alpha, const void *A, float beta, void *B );
+int TCORE_sgelqt( int M, int N, int IB, void *A, void *T, float *TAU, float *WORK );
+void TCORE_sgemm( int transA, int transB, int M, int N, int K, void *alpha, const void *A, const void *B, float beta, void *C );
+int TCORE_sgeqrt( int M, int N, int IB, void *A, void *T, float *TAU, float *WORK );
+int TCORE_sgessm( int M, int N, int K, int IB, const int *IPIV, const void *L, void *A );
+int TCORE_sgessq( int storev, int M, int N, const void *A, void *sclssq );
+int TCORE_sgetrf( int M, int N, void *A, int *IPIV, int *INFO );
+int TCORE_sgetrf_incpiv( int M, int N, int IB, void *A, int *IPIV, int *INFO );
+int TCORE_sgetrf_nopiv( int M, int N, int IB, void *A, int *INFO );
+void TCORE_she2ge( int uplo, int M, int N, const void *A, void *B );
+int TCORE_ssyrfb( int uplo, int N, int K, int IB, int NB, const void *A, const void *T, void *C, float *WORK, int ldwork );
+void TCORE_slacpy( int uplo, int M, int N, const void *A, void *B );
+void TCORE_slange( int norm, int M, int N, const void *A, float *work, float *normA );
+void TCORE_slansy( int norm, int uplo, int N, const void *A, float *work, float *normA );
+void TCORE_slantr( int norm, int uplo, int diag, int M, int N, const void *A, float *work, float *normA );
+int TCORE_slascal( int uplo, int m, int n, void *alpha, void *A );
+void TCORE_slaset( int uplo, int n1, int n2, void *alpha, float beta, void *A );
+void TCORE_slaset2( int uplo, int n1, int n2, void *alpha, void *A );
+int TCORE_slatro( int uplo, int trans, int M, int N, const void *A, void *B );
+void TCORE_slauum( int uplo, int N, void *A );
+void TCORE_splgsy( float bump, int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_splrnt( int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_spotrf( int uplo, int n, void *A, int *INFO );
+int TCORE_sssssm( int M1, int N1, int M2, int N2, int K, int IB, void *A1, void *A2, const void *L1, const void *L2, const int *IPIV );
+void TCORE_ssymm( int side, int uplo, int M, int N, void *alpha, const void *A, const void *B, float beta, void *C );
+void TCORE_ssyrk( int uplo, int trans, int N, int K, void *alpha, const void *A, float beta, void *C );
+void TCORE_ssyr2k( int uplo, int trans, int N, int K, void *alpha, const void *A, const void *B, float beta, void *C );
+int TCORE_ssyssq( int storev, int uplo, int N, const void *A, void *sclssq );
+int TCORE_ssytf2_nopiv( int uplo, int n, void *A );
+int TCORE_stplqt( int M, int N, int L, int IB, void *A, void *B, void *T, float *WORK );
+int TCORE_stpmlqt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, float *WORK );
+int TCORE_stpmqrt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, float *WORK );
+int TCORE_stpqrt( int M, int N, int L, int IB, void *A, void *B, void *T, float *WORK );
+int TCORE_stradd( int uplo, int trans, int M, int N, void *alpha, const void *A, float beta, void *B );
+void TCORE_strasm( int storev, int uplo, int diag, int M, int N, const void *A, float *work );
+void TCORE_strmm( int side, int uplo, int transA, int diag, int M, int N, void *alpha, const void *A, void *B );
+void TCORE_strsm( int side, int uplo, int transA, int diag, int M, int N, void *alpha, const void *A, void *B );
+int TCORE_strssq( int uplo, int diag, int M, int N, const void *A, void *sclssq );
+void TCORE_strtri( int uplo, int diag, int N, void *A, int *info );
+int TCORE_stsmlq_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, float *WORK, int ldwork );
+int TCORE_stsmqr_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, float *WORK, int ldwork );
+int TCORE_ststrf( int M, int N, int IB, int NB, void *U, void *A, void *L, int *IPIV, float *WORK, int LDWORK, int *INFO );
+int TCORE_sormlq( int side, int trans, int M, int N, int IB, int K, const void *V, const void *T, void *C, float *WORK, int LDWORK );
+int TCORE_sormqr( int side, int trans, int M, int N, int K, int IB, const void *V, const void *T, void *C, float *WORK, int LDWORK );
+int TCORE_sgram( int uplo, int M, int N, int Mt, int Nt, const void *Di, const void *Dj, const void *D, void *A );
+
+void TCORE_dasum( int storev, int uplo, int M, int N, const void *A, double *work );
+int TCORE_daxpy( int M, void *alpha, const void *A, int incA, void *B, int incB );
+int TCORE_dgeadd( int trans, int M, int N, void *alpha, const void *A, double beta, void *B );
+int TCORE_dgelqt( int M, int N, int IB, void *A, void *T, double *TAU, double *WORK );
+void TCORE_dgemm( int transA, int transB, int M, int N, int K, void *alpha, const void *A, const void *B, double beta, void *C );
+int TCORE_dgeqrt( int M, int N, int IB, void *A, void *T, double *TAU, double *WORK );
+int TCORE_dgessm( int M, int N, int K, int IB, const int *IPIV, const void *L, void *A );
+int TCORE_dgessq( int storev, int M, int N, const void *A, void *sclssq );
+int TCORE_dgetrf( int M, int N, void *A, int *IPIV, int *INFO );
+int TCORE_dgetrf_incpiv( int M, int N, int IB, void *A, int *IPIV, int *INFO );
+int TCORE_dgetrf_nopiv( int M, int N, int IB, void *A, int *INFO );
+void TCORE_dhe2ge( int uplo, int M, int N, const void *A, void *B );
+int TCORE_dsyrfb( int uplo, int N, int K, int IB, int NB, const void *A, const void *T, void *C, double *WORK, int ldwork );
+void TCORE_dlacpy( int uplo, int M, int N, const void *A, void *B );
+void TCORE_dlange( int norm, int M, int N, const void *A, double *work, double *normA );
+void TCORE_dlansy( int norm, int uplo, int N, const void *A, double *work, double *normA );
+void TCORE_dlantr( int norm, int uplo, int diag, int M, int N, const void *A, double *work, double *normA );
+int TCORE_dlascal( int uplo, int m, int n, void *alpha, void *A );
+void TCORE_dlaset( int uplo, int n1, int n2, void *alpha, double beta, void *A );
+void TCORE_dlaset2( int uplo, int n1, int n2, void *alpha, void *A );
+int TCORE_dlatro( int uplo, int trans, int M, int N, const void *A, void *B );
+void TCORE_dlauum( int uplo, int N, void *A );
+void TCORE_dplgsy( double bump, int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_dplrnt( int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_dpotrf( int uplo, int n, void *A, int *INFO );
+int TCORE_dssssm( int M1, int N1, int M2, int N2, int K, int IB, void *A1, void *A2, const void *L1, const void *L2, const int *IPIV );
+void TCORE_dsymm( int side, int uplo, int M, int N, void *alpha, const void *A, const void *B, double beta, void *C );
+void TCORE_dsyrk( int uplo, int trans, int N, int K, void *alpha, const void *A, double beta, void *C );
+void TCORE_dsyr2k( int uplo, int trans, int N, int K, void *alpha, const void *A, const void *B, double beta, void *C );
+int TCORE_dsyssq( int storev, int uplo, int N, const void *A, void *sclssq );
+int TCORE_dsytf2_nopiv( int uplo, int n, void *A );
+int TCORE_dtplqt( int M, int N, int L, int IB, void *A, void *B, void *T, double *WORK );
+int TCORE_dtpmlqt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, double *WORK );
+int TCORE_dtpmqrt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, double *WORK );
+int TCORE_dtpqrt( int M, int N, int L, int IB, void *A, void *B, void *T, double *WORK );
+int TCORE_dtradd( int uplo, int trans, int M, int N, void *alpha, const void *A, double beta, void *B );
+void TCORE_dtrasm( int storev, int uplo, int diag, int M, int N, const void *A, double *work );
+void TCORE_dtrmm( int side, int uplo, int transA, int diag, int M, int N, void *alpha, const void *A, void *B );
+void TCORE_dtrsm( int side, int uplo, int transA, int diag, int M, int N, void *alpha, const void *A, void *B );
+int TCORE_dtrssq( int uplo, int diag, int M, int N, const void *A, void *sclssq );
+void TCORE_dtrtri( int uplo, int diag, int N, void *A, int *info );
+int TCORE_dtsmlq_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, double *WORK, int ldwork );
+int TCORE_dtsmqr_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, double *WORK, int ldwork );
+int TCORE_dtstrf( int M, int N, int IB, int NB, void *U, void *A, void *L, int *IPIV, double *WORK, int LDWORK, int *INFO );
+int TCORE_dormlq( int side, int trans, int M, int N, int IB, int K, const void *V, const void *T, void *C, double *WORK, int LDWORK );
+int TCORE_dormqr( int side, int trans, int M, int N, int K, int IB, const void *V, const void *T, void *C, double *WORK, int LDWORK );
+int TCORE_dgram( int uplo, int M, int N, int Mt, int Nt, const void *Di, const void *Dj, const void *D, void *A );
+
+void TCORE_scasum( int storev, int uplo, int M, int N, const void *A, float *work );
+int TCORE_caxpy( int M, void * alpha, const void *A, int incA, void *B, int incB );
+int TCORE_cgeadd( int trans, int M, int N, void * alpha, const void *A, void * beta, void *B );
+int TCORE_cgelqt( int M, int N, int IB, void *A, void *T, void *TAU, void *WORK );
+void TCORE_cgemm( int transA, int transB, int M, int N, int K, void * alpha, const void *A, const void *B, void * beta, void *C );
+int TCORE_cgeqrt( int M, int N, int IB, void *A, void *T, void *TAU, void *WORK );
+int TCORE_cgessm( int M, int N, int K, int IB, const int *IPIV, const void *L, void *A );
+int TCORE_cgessq( int storev, int M, int N, const void *A, void *sclssq );
+int TCORE_cgetrf( int M, int N, void *A, int *IPIV, int *INFO );
+int TCORE_cgetrf_incpiv( int M, int N, int IB, void *A, int *IPIV, int *INFO );
+int TCORE_cgetrf_nopiv( int M, int N, int IB, void *A, int *INFO );
+void TCORE_che2ge( int uplo, int M, int N, const void *A, void *B );
+void TCORE_chemm( int side, int uplo, int M, int N, void * alpha, const void *A, const void *B, void * beta, void *C );
+void TCORE_cherk( int uplo, int trans, int N, int K, void *alpha, const void *A, float beta, void *C );
+void TCORE_cher2k( int uplo, int trans, int N, int K, void * alpha, const void *A, const void *B, float beta, void *C );
+int TCORE_cherfb( int uplo, int N, int K, int IB, int NB, const void *A, const void *T, void *C, void *WORK, int ldwork );
+int TCORE_chessq( int storev, int uplo, int N, const void *A, void *sclssq );
+void TCORE_clacpy( int uplo, int M, int N, const void *A, void *B );
+void TCORE_clange( int norm, int M, int N, const void *A, float *work, float *normA );
+void TCORE_clanhe( int norm, int uplo, int N, const void *A, float *work, float *normA );
+void TCORE_clansy( int norm, int uplo, int N, const void *A, float *work, float *normA );
+void TCORE_clantr( int norm, int uplo, int diag, int M, int N, const void *A, float *work, float *normA );
+int TCORE_clascal( int uplo, int m, int n, void * alpha, void *A );
+void TCORE_claset( int uplo, int n1, int n2, void * alpha, void * beta, void *A );
+void TCORE_claset2( int uplo, int n1, int n2, void * alpha, void *A );
+int TCORE_clatro( int uplo, int trans, int M, int N, const void *A, void *B );
+void TCORE_clauum( int uplo, int N, void *A );
+void TCORE_cplghe( float bump, int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_cplgsy( void * bump, int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_cplrnt( int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_cpotrf( int uplo, int n, void *A, int *INFO );
+int TCORE_cssssm( int M1, int N1, int M2, int N2, int K, int IB, void *A1, void *A2, const void *L1, const void *L2, const int *IPIV );
+void TCORE_csymm( int side, int uplo, int M, int N, void * alpha, const void *A, const void *B, void * beta, void *C );
+void TCORE_csyrk( int uplo, int trans, int N, int K, void * alpha, const void *A, void * beta, void *C );
+void TCORE_csyr2k( int uplo, int trans, int N, int K, void * alpha, const void *A, const void *B, void * beta, void *C );
+int TCORE_csyssq( int storev, int uplo, int N, const void *A, void *sclssq );
+int TCORE_csytf2_nopiv( int uplo, int n, void *A );
+int TCORE_ctplqt( int M, int N, int L, int IB, void *A, void *B, void *T, void *WORK );
+int TCORE_ctpmlqt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, void *WORK );
+int TCORE_ctpmqrt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, void *WORK );
+int TCORE_ctpqrt( int M, int N, int L, int IB, void *A, void *B, void *T, void *WORK );
+int TCORE_ctradd( int uplo, int trans, int M, int N, void * alpha, const void *A, void * beta, void *B );
+void TCORE_ctrasm( int storev, int uplo, int diag, int M, int N, const void *A, float *work );
+void TCORE_ctrmm( int side, int uplo, int transA, int diag, int M, int N, void * alpha, const void *A, void *B );
+void TCORE_ctrsm( int side, int uplo, int transA, int diag, int M, int N, void * alpha, const void *A, void *B );
+int TCORE_ctrssq( int uplo, int diag, int M, int N, const void *A, void *sclssq );
+void TCORE_ctrtri( int uplo, int diag, int N, void *A, int *info );
+int TCORE_ctsmlq_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, void *WORK, int ldwork );
+int TCORE_ctsmqr_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, void *WORK, int ldwork );
+int TCORE_ctstrf( int M, int N, int IB, int NB, void *U, void *A, void *L, int *IPIV, void *WORK, int LDWORK, int *INFO );
+int TCORE_cunmlq( int side, int trans, int M, int N, int IB, int K, const void *V, const void *T, void *C, void *WORK, int LDWORK );
+int TCORE_cunmqr( int side, int trans, int M, int N, int K, int IB, const void *V, const void *T, void *C, void *WORK, int LDWORK );
+int TCORE_cgram( int uplo, int M, int N, int Mt, int Nt, const void *Di, const void *Dj, const void *D, void *A );
+
+void TCORE_dzasum( int storev, int uplo, int M, int N, const void *A, double *work );
+int TCORE_zaxpy( int M, void *alpha, const void *A, int incA, void *B, int incB );
+int TCORE_zgeadd( int trans, int M, int N, void *alpha, const void *A, void *beta, void *B );
+int TCORE_zgelqt( int M, int N, int IB, void *A, void *T, void *TAU, void *WORK );
+void TCORE_zgemm( int transA, int transB, int M, int N, int K, void *alpha, const void *A, const void *B, void *beta, void *C );
+int TCORE_zgeqrt( int M, int N, int IB, void *A, void *T, void *TAU, void *WORK );
+int TCORE_zgessm( int M, int N, int K, int IB, const int *IPIV, const void *L, void *A );
+int TCORE_zgessq( int storev, int M, int N, const void *A, void *sclssq );
+int TCORE_zgetrf( int M, int N, void *A, int *IPIV, int *INFO );
+int TCORE_zgetrf_incpiv( int M, int N, int IB, void *A, int *IPIV, int *INFO );
+int TCORE_zgetrf_nopiv( int M, int N, int IB, void *A, int *INFO );
+void TCORE_zhe2ge( int uplo, int M, int N, const void *A, void *B );
+void TCORE_zhemm( int side, int uplo, int M, int N, void *alpha, const void *A, const void *B, void *beta, void *C );
+void TCORE_zherk( int uplo, int trans, int N, int K, void *alpha, const void *A, double beta, void *C );
+void TCORE_zher2k( int uplo, int trans, int N, int K, void *alpha, const void *A, const void *B, double beta, void *C );
+int TCORE_zherfb( int uplo, int N, int K, int IB, int NB, const void *A, const void *T, void *C, void *WORK, int ldwork );
+int TCORE_zhessq( int storev, int uplo, int N, const void *A, void *sclssq );
+void TCORE_zlacpy( int uplo, int M, int N, const void *A, void *B );
+void TCORE_zlange( int norm, int M, int N, const void *A, double *work, double *normA );
+void TCORE_zlanhe( int norm, int uplo, int N, const void *A, double *work, double *normA );
+void TCORE_zlansy( int norm, int uplo, int N, const void *A, double *work, double *normA );
+void TCORE_zlantr( int norm, int uplo, int diag, int M, int N, const void *A, double *work, double *normA );
+int TCORE_zlascal( int uplo, int m, int n, void *alpha, void *A );
+void TCORE_zlaset( int uplo, int n1, int n2, void *alpha, void *beta, void *A );
+void TCORE_zlaset2( int uplo, int n1, int n2, void *alpha, void *A );
+int TCORE_zlatro( int uplo, int trans, int M, int N, const void *A, void *B );
+void TCORE_zlauum( int uplo, int N, void *A );
+void TCORE_zplghe( double bump, int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_zplgsy( void *bump, int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_zplrnt( int m, int n, void *tileA, int bigM, int m0, int n0, unsigned long long int seed );
+void TCORE_zpotrf( int uplo, int n, void *A, int *INFO );
+int TCORE_zssssm( int M1, int N1, int M2, int N2, int K, int IB, void *A1, void *A2, const void *L1, const void *L2, const int *IPIV );
+void TCORE_zsymm( int side, int uplo, int M, int N, void *alpha, const void *A, const void *B, void *beta, void *C );
+void TCORE_zsyrk( int uplo, int trans, int N, int K, void *alpha, const void *A, void *beta, void *C );
+void TCORE_zsyr2k( int uplo, int trans, int N, int K, void *alpha, const void *A, const void *B, void *beta, void *C );
+int TCORE_zsyssq( int storev, int uplo, int N, const void *A, void *sclssq );
+int TCORE_zsytf2_nopiv( int uplo, int n, void *A );
+int TCORE_ztplqt( int M, int N, int L, int IB, void *A, void *B, void *T, void *WORK );
+int TCORE_ztpmlqt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, void *WORK );
+int TCORE_ztpmqrt( int side, int trans, int M, int N, int K, int L, int IB, const void *V, const void *T, void *A, void *B, void *WORK );
+int TCORE_ztpqrt( int M, int N, int L, int IB, void *A, void *B, void *T, void *WORK );
+int TCORE_ztradd( int uplo, int trans, int M, int N, void *alpha, const void *A, void *beta, void *B );
+void TCORE_ztrasm( int storev, int uplo, int diag, int M, int N, const void *A, double *work );
+void TCORE_ztrmm( int side, int uplo, int transA, int diag, int M, int N, void *alpha, const void *A, void *B );
+void TCORE_ztrsm( int side, int uplo, int transA, int diag, int M, int N, void *alpha, const void *A, void *B );
+int TCORE_ztrssq( int uplo, int diag, int M, int N, const void *A, void *sclssq );
+void TCORE_ztrtri( int uplo, int diag, int N, void *A, int *info );
+int TCORE_ztsmlq_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, void *WORK, int ldwork );
+int TCORE_ztsmqr_hetra1( int side, int trans, int m1, int n1, int m2, int n2, int k, int ib, void *A1, void *A2, const void *V, const void *T, void *WORK, int ldwork );
+int TCORE_ztstrf( int M, int N, int IB, int NB, void *U, void *A, void *L, int *IPIV, void *WORK, int LDWORK, int *INFO );
+int TCORE_zunmlq( int side, int trans, int M, int N, int IB, int K, const void *V, const void *T, void *C, void *WORK, int LDWORK );
+int TCORE_zunmqr( int side, int trans, int M, int N, int K, int IB, const void *V, const void *T, void *C, void *WORK, int LDWORK );
+int TCORE_zgram( int uplo, int M, int N, int Mt, int Nt, const void *Di, const void *Dj, const void *D, void *A );
+
+void CORE_slacpy(int uplo, int M, int N, void *A, int LDA, void *B, int LDB);
+void CORE_dlacpy(int uplo, int M, int N, void *A, int LDA, void *B, int LDB);
+void CORE_clacpy(int uplo, int M, int N, void *A, int LDA, void *B, int LDB);
+void CORE_zlacpy(int uplo, int M, int N, void *A, int LDA, void *B, int LDB);
+int CORE_splssq( int storev, int M, int N, void *sclssqin, void *sclssqout );
+int CORE_dplssq( int storev, int M, int N, void *sclssqin, void *sclssqout );
+int CORE_cplssq( int storev, int M, int N, void *sclssqin, void *sclssqout );
+int CORE_zplssq( int storev, int M, int N, void *sclssqin, void *sclssqout );
+int CORE_splssq2( int N, void *sclssq );
+int CORE_dplssq2( int N, void *sclssq );
+int CORE_cplssq2( int N, void *sclssq );
+int CORE_zplssq2( int N, void *sclssq );
+
+END_MODULE
diff --git a/doc/orgmode/chapters/installing.org b/doc/orgmode/chapters/installing.org
index 0193d3dad..7675856d6 100644
--- a/doc/orgmode/chapters/installing.org
+++ b/doc/orgmode/chapters/installing.org
@@ -61,6 +61,9 @@ we encourage users to use [[sec:spack][Spack]].
sudo apt-get install -y libopenmpi-dev
# Install hwloc (used by StarPU or QUARK, already a dependency of OpenMPI)
sudo apt-get install -y libhwloc-dev
+ # install EZTrace, usefull to export some nice execution traces
+ with all runtimes
+ sudo apt-get install -y libeztrace-dev
# install FxT, usefull to export some nice execution traces with StarPU
sudo apt-get install -y libfxt-dev
# Install cuda and cuBLAS: only if you have a GPU cuda compatible
@@ -209,17 +212,17 @@ we encourage users to use [[sec:spack][Spack]].
*Caution about the compatibility:* Chameleon has been mainly tested
with the QUARK library coming from https://github.com/ecrc/quark.
-**** FxT
- [[http://download.savannah.gnu.org/releases/fkt/][FxT]] stands for both FKT (Fast Kernel Tracing) and FUT (Fast User
- Tracing). This library provides efficient support for recording
- traces. Chameleon can trace kernels execution on the different
- workers and produce .paje files if FxT is enabled. FxT can only
- be used through StarPU and StarPU must be compiled with FxT
- enabled, see how to use this feature here [[sec:trace][Execution trace using
- StarPU]].
-
- *Caution about the compatibility:* FxT should be compatible with
- the version of StarPU used.
+**** EZTrace
+ This library provides efficient modules for recording
+ traces. Chameleon can trace kernels execution on CPU workers
+ thanks to EZTrace and produce .paje files. EZTrace also provides
+ integrated modules to trace MPI calls and/or memory usage. See
+ how to use this feature here [[sec:trace_ezt][Execution trace
+ using EZTrace]]. To trace kernels execution on all kind of
+ workers, such as CUDA, We recommend to use the internal tracing
+ support of the runtime system used done by the underlying
+ runtime. See how to use this feature here [[sec:trace_fxt][Execution trace
+ using StarPU/FxT]].
**** hwloc
[[http://www.open-mpi.org/projects/hwloc/][hwloc]] (Portable Hardware Locality) is a software package for
@@ -394,11 +397,6 @@ we encourage users to use [[sec:spack][Spack]].
* *CHAMELEON_ENABLE_TIMING=ON|OFF* (default ON): to control build
of timing executables (performances check) contained in timing/
sub-directory
- * *CHAMELEON_ENABLE_TRACING=ON|OFF* (default OFF): to enable trace
- generation during execution of timing drivers. It requires
- StarPU to be linked with FxT library (trace execution of
- kernels on workers), see also [[sec:trace][Execution tracing
- with StarPU]].
* *CHAMELEON_SIMULATION=ON|OFF* (default OFF): to enable
simulation mode, means Chameleon will not really execute tasks,
see details in section [[sec:simu][Use simulation mode with
diff --git a/doc/orgmode/chapters/using.org b/doc/orgmode/chapters/using.org
index c00695dc2..9798c5831 100644
--- a/doc/orgmode/chapters/using.org
+++ b/doc/orgmode/chapters/using.org
@@ -129,28 +129,91 @@
* GELS: solves overdetermined or underdetermined linear systems involving a general matrix using the QR or the LQ factorization
* GESVD: general matrix singular value decomposition
-*** Execution trace using StarPU
- <<sec:trace>>
+*** Execution trace using EZTrace
+ <<sec:trace_ezt>>
+
+ [[http://eztrace.gforge.inria.fr/support.html][EZTrace]] can be used by chameleon to generate traces. Two modules
+ are automatically generated as soon as EZTrace is detected on the
+ system. The first one (which is recommended) is the
+ ~chameleon_tcore~ module. It traces all the ~TCORE_...()~ functions
+ that are called by the codelets of all the runtime but PaRSEC. The
+ second one is the ~chameleon_core~ module which traces the lower
+ level ~CORE_...()~ functions. If using PaRSEC, you need to use this
+ module to generate the traces.
+
+ To generate traces with EZTrace, you need first to compile with
+ *-DBUILD_SHARED_LIBS=ON*. EZTrace is using weak symbols to overload
+ function calls with ld_preload and enable trace generation. Then,
+ either you install the ~libeztrace-*.so~ files into the EZTrace
+ install directory, or you can add the path of the modules to your
+ environement
+ #+begin_src
+ export EZTRACE_LIBRARY_PATH=/path/to/your/modules
+ #+end_src
+
+ To check if the modules are available you should have
+ #+begin_src
+ $ eztrace_avail
+ 1 omp Module for OpenMP parallel regions
+ 2 pthread Module for PThread synchronization functions (mutex, semaphore, spinlock, etc.)
+ 3 stdio Module for stdio functions (read, write, select, poll, etc.)
+ 4 mpi Module for MPI functions
+ 5 memory Module for memory functions (malloc, free, etc.)
+ 6 papi Module for PAPI Performance counters
+ 128 chameleon_core Module for Chameleon CORE functions
+ 129 chameleon_tcore Module for Chameleon TCORE functions
+ #+end_src
+
+ Then, you can restrict the modules used during the execution
+ #+begin_src
+ export EZTRACE_TRACE="mpi chameleon_tcore"
+ #+end_src
+
+ _The module ~mpi~ is required if you want to run in distributed._
+
+ The setup can be checked with ~eztrace_loaded~
+ #+begin_src
+ $ eztrace_loaded
+ 4 mpi Module for MPI functions
+ 129 chameleon_tcore Module for Chameleon TCORE functions
+ #+end_src
+
+ To generate the traces, you need to run your binary through
+ eztrace:
+ #+begin_src
+ eztrace ./dnew-testing -o gemm -n 1000 -b 200
+ mpirun -np 4 eztrace ./dnew-testing -o gemm -n 1000 -b 200 -P 2
+ #+end_src
+
+ Convert the binary files into a ~.trace~ file, and visualize it.
+ #+begin_src
+ eztrace_convert <username>_eztrace_log_rank_<[0-9]*>
+ vite eztrace_output.trace
+ #+end_src
+
+ For more information on EZTrace, you can follow the [[http://eztrace.gforge.inria.fr/support.html][support page]].
+
+*** Execution trace using StarPU/FxT
+ <<sec:trace_fxt>>
StarPU can generate its own trace log files by compiling it with
the ~--with-fxt~ option at the configure step (you can have to
specify the directory where you installed FxT by giving
~--with-fxt=...~ instead of ~--with-fxt~ alone). By doing so, traces
are generated after each execution of a program which uses StarPU
- in the directory pointed by the STARPU_FXT_PREFIX environment
+ in the directory pointed by the [[http://starpu.gforge.inria.fr/doc/html/ExecutionConfigurationThroughEnvironmentVariables.html][STARPU_FXT_PREFIX]] environment
variable.
#+begin_example
export STARPU_FXT_PREFIX=/home/jdoe/fxt_files/
#+end_example
When executing a ~./timing/...~ Chameleon program, if it has been
- enabled (StarPU compiled with FxT and
- *-DCHAMELEON_ENABLE_TRACING=ON*), you can give the option ~--trace~ to
- tell the program to generate trace log files.
+ enabled (StarPU compiled with FxT), the program will generate
+ trace files in the directory $STARPU_FXT_PREFIX.
Finally, to generate the trace file which can be opened with [[http://vite.gforge.inria.fr/][Vite]]
program, you can use the *starpu_fxt_tool* executable of StarPU.
- This tool should be in ~$STARPU_INSTALL_REPOSITORY/bin~. You can
- use it to generate the trace file like this:
+ This tool should be in the bin directory of StarPU's installation.
+ You can use it to generate the trace file like this:
#+begin_src
path/to/your/install/starpu/bin/starpu_fxt_tool -i prof_filename
#+end_src
diff --git a/example/link_chameleon/CMakeLists.txt b/example/link_chameleon/CMakeLists.txt
index 6e0dac9e9..89a16531d 100644
--- a/example/link_chameleon/CMakeLists.txt
+++ b/example/link_chameleon/CMakeLists.txt
@@ -49,7 +49,7 @@ if (CHAMELEON_DISTRIB_DIR)
if (CHAMELEON_CHAMELEON_USE_QUARK)
find_package(CHAMELEON COMPONENTS QUARK)
else()
- find_package(CHAMELEON COMPONENTS STARPU MPI CUDA FXT)
+ find_package(CHAMELEON COMPONENTS STARPU MPI CUDA)
endif()
if (CHAMELEON_FOUND)
link_directories(${CHAMELEON_LIBRARY_DIRS_DEP})
diff --git a/include/chameleon/config.h.in b/include/chameleon/config.h.in
index 1596315d8..8432d7a93 100644
--- a/include/chameleon/config.h.in
+++ b/include/chameleon/config.h.in
@@ -47,9 +47,6 @@
/* Simulation */
#cmakedefine CHAMELEON_SIMULATION
-/* Tracing support */
-#cmakedefine CHAMELEON_ENABLE_TRACING
-
/* getopt */
#cmakedefine CHAMELEON_HAVE_GETOPT_H
#cmakedefine CHAMELEON_HAVE_GETOPT_LONG
diff --git a/new-testing/testing_zauxiliary.c b/new-testing/testing_zauxiliary.c
index da33718fa..16da205d6 100644
--- a/new-testing/testing_zauxiliary.c
+++ b/new-testing/testing_zauxiliary.c
@@ -30,13 +30,16 @@ struct option;
* @brief Defines all the parameters of the testings
*/
static parameter_t parameters[] = {
+ /* Name, helper, shname, flags, has_arg, psize, valtype, value, vallist, read, sprint */
{ "id", "Id of the run", 0, PARAM_OUTPUT, 0, 3, TestValInt, {0}, NULL, NULL, sprint_int },
{ NULL, "Options", 0, PARAM_OPTION, 0, 0, 0, {0}, NULL, NULL, NULL },
- { "help", "Show this help", 'h', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
- { "check", "Enable checking of the result", 'c', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
- { "human", "Enable human readable mode", 'H', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
- { "niter", "Perform multiple iteration per test", 'l', PARAM_OPTION, 1, 0, TestValInt, {1}, NULL, pread_int, sprint_int },
+ { "help", "Show this help", 'h', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
+ { "check", "Enable checking of the result", 'c', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
+ { "human", "Enable human readable mode", 'H', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
+ { "niter", "Perform multiple iteration per test", 'l', PARAM_OPTION, 1, 0, TestValInt, {1}, NULL, pread_int, sprint_int },
+ { "trace", "Enable the trace generation", -30, PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
+ { "nowarmup", "Disable the warmup run to load libraries", -31, PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
{ NULL, "Machine parameters", 0, PARAM_OPTION, 0, 0, 0, {0}, NULL, NULL, NULL },
{ "threads", "Number of CPU workers per node", 't', PARAM_OPTION | PARAM_OUTPUT, 1, 7, TestValInt, {1}, NULL, pread_int, sprint_int },
@@ -476,6 +479,7 @@ parameters_destroy()
int main (int argc, char **argv) {
int ncores, ngpus, human, check, i, niter;
+ int trace, nowarmup;
int rc, info = 0;
int run_id = 0;
char *func_name;
@@ -491,12 +495,14 @@ int main (int argc, char **argv) {
parameters_read_file( input_file );
free(input_file);
}
- ncores = parameters_getvalue_int( "threads" );
- ngpus = parameters_getvalue_int( "gpus" );
- check = parameters_getvalue_int( "check" );
- human = parameters_getvalue_int( "human" );
- func_name = parameters_getvalue_str( "op" );
- niter = parameters_getvalue_int( "niter" );
+ ncores = parameters_getvalue_int( "threads" );
+ ngpus = parameters_getvalue_int( "gpus" );
+ check = parameters_getvalue_int( "check" );
+ human = parameters_getvalue_int( "human" );
+ func_name = parameters_getvalue_str( "op" );
+ niter = parameters_getvalue_int( "niter" );
+ trace = parameters_getvalue_int( "trace" );
+ nowarmup = parameters_getvalue_int( "nowarmup" );
CHAMELEON_Init( ncores, ngpus );
@@ -510,6 +516,20 @@ int main (int argc, char **argv) {
/* Executes the tests */
run_print_header( test, check, human );
run = runlist->head;
+
+ /* Warmup */
+ if ( !nowarmup ) {
+ run_arg_list_t copy = run_arg_list_copy( &(run->args) );
+ rc = test->fptr( ©, check );
+ run_arg_list_destroy( © );
+ }
+
+ /* Start tracing */
+ if ( trace ) {
+ CHAMELEON_Enable( CHAMELEON_PROFILING_MODE );
+ }
+
+ /* Perform all runs */
while ( run != NULL ) {
for(i=0; i<niter; i++) {
run_arg_list_t copy = run_arg_list_copy( &(run->args) );
@@ -530,6 +550,12 @@ int main (int argc, char **argv) {
run_list_destroy( run );
run = next;
}
+
+ /* Stop tracing */
+ if ( trace ) {
+ CHAMELEON_Disable( CHAMELEON_PROFILING_MODE );
+ }
+
free( runlist );
CHAMELEON_Finalize();
diff --git a/new-testing/testing_zauxiliary.h b/new-testing/testing_zauxiliary.h
index 2be0d88c5..7b35505f8 100644
--- a/new-testing/testing_zauxiliary.h
+++ b/new-testing/testing_zauxiliary.h
@@ -23,47 +23,6 @@
#include "testings.h"
-/**
- *
- * Macro for trace generation
- *
- */
-#define START_TRACING() \
- RUNTIME_start_stats(); \
- if(iparam[IPARAM_TRACE] == 2) { \
- RUNTIME_start_profiling(); \
- } \
- if(iparam[IPARAM_BOUND]) { \
- CHAMELEON_Enable(CHAMELEON_BOUND); \
- }
-
-#define STOP_TRACING() \
- RUNTIME_stop_stats(); \
- if(iparam[IPARAM_TRACE] == 2) { \
- RUNTIME_stop_profiling(); \
- } \
- if(iparam[IPARAM_BOUND]) { \
- CHAMELEON_Disable(CHAMELEON_BOUND); \
- }
-
-/**
- *
- * Macro for DAG generation
- *
- */
-#if 0
-#define START_DAG() \
- if ( iparam[IPARAM_DAG] == 2 ) \
- CHAMELEON_Enable(CHAMELEON_DAG);
-
-#define STOP_DAG() \
- if ( iparam[IPARAM_DAG] == 2 ) \
- CHAMELEON_Disable(CHAMELEON_DAG);
-#else
-#define START_DAG() do {} while(0);
-#define STOP_DAG() do {} while(0);
-#endif
-
/**
*
* Synchro for distributed computations
@@ -82,23 +41,6 @@
* General Macros for timing
*
*/
-/* #define START_TIMING() \ */
-/* START_DAG(); \ */
-/* START_TRACING(); \ */
-/* START_DISTRIBUTED(); \ */
-/* t = -RUNTIME_get_time(); */
-
-/* #define STOP_TIMING() \ */
-/* STOP_DISTRIBUTED(); \ */
-/* t += RUNTIME_get_time(); \ */
-/* STOP_TRACING(); \ */
-/* STOP_DAG(); \ */
-/* if (iparam[IPARAM_PROFILE] == 2) { \ */
-/* RUNTIME_kernelprofile_display(); \ */
-/* RUNTIME_schedprofile_display(); \ */
-/* } \ */
-/* *t_ = t; */
-
#define START_TIMING( _t_ ) \
START_DISTRIBUTED(); \
(_t_) = RUNTIME_get_time();
diff --git a/new-testing/testing_zgetrf.c b/new-testing/testing_zgetrf.c
index 68491ba58..ebfea504e 100644
--- a/new-testing/testing_zgetrf.c
+++ b/new-testing/testing_zgetrf.c
@@ -55,8 +55,6 @@ testing_zgetrf( run_arg_list_t *args, int check )
run_arg_add_fixdbl( args, "time", t );
run_arg_add_fixdbl( args, "gflops", ( hres == CHAMELEON_SUCCESS ) ? gflops : -1. );
- fprintf( stdout, "hres = %d\n", hres );
-
/* Checks the factorisation and residue */
if ( check ) {
CHAM_desc_t *descA0 = CHAMELEON_Desc_Copy( descA, NULL );
diff --git a/runtime/openmp/control/runtime_profiling.c b/runtime/openmp/control/runtime_profiling.c
index dfb9e2579..b243c9d4d 100644
--- a/runtime/openmp/control/runtime_profiling.c
+++ b/runtime/openmp/control/runtime_profiling.c
@@ -27,12 +27,10 @@ double RUNTIME_get_time(){
void RUNTIME_start_profiling()
{
- chameleon_warning("RUNTIME_start_profiling()", "FxT profiling is not available with OpenMP\n");
}
void RUNTIME_stop_profiling()
{
- chameleon_warning("RUNTIME_stop_profiling()", "FxT profiling is not available with OpenMP\n");
}
void RUNTIME_start_stats()
diff --git a/runtime/parsec/control/runtime_profiling.c b/runtime/parsec/control/runtime_profiling.c
index 870233220..c8f970005 100644
--- a/runtime/parsec/control/runtime_profiling.c
+++ b/runtime/parsec/control/runtime_profiling.c
@@ -26,12 +26,10 @@ double RUNTIME_get_time(){
void RUNTIME_start_profiling()
{
- chameleon_warning("RUNTIME_start_profiling()", "FxT profiling is not available with PaRSEC\n");
}
void RUNTIME_stop_profiling()
{
- chameleon_warning("RUNTIME_stop_profiling()", "FxT profiling is not available with PaRSEC\n");
}
void RUNTIME_start_stats()
diff --git a/runtime/quark/control/runtime_profiling.c b/runtime/quark/control/runtime_profiling.c
index 685a81c40..10da952b0 100644
--- a/runtime/quark/control/runtime_profiling.c
+++ b/runtime/quark/control/runtime_profiling.c
@@ -26,12 +26,10 @@ double RUNTIME_get_time(){
void RUNTIME_start_profiling()
{
- chameleon_warning("RUNTIME_start_profiling()", "FxT profiling is not available with Quark\n");
}
void RUNTIME_stop_profiling()
{
- chameleon_warning("RUNTIME_stop_profiling()", "FxT profiling is not available with Quark\n");
}
void RUNTIME_start_stats()
diff --git a/runtime/starpu/control/runtime_control.c b/runtime/starpu/control/runtime_control.c
index 14659f785..a0f05c045 100644
--- a/runtime/starpu/control/runtime_control.c
+++ b/runtime/starpu/control/runtime_control.c
@@ -21,7 +21,7 @@
#include <stdio.h>
#include <stdlib.h>
#include "chameleon_starpu.h"
-#if defined(HAVE_STARPU_FXT_PROFILING)
+#if defined(STARPU_USE_FXT)
#include <starpu_fxt.h>
#endif
@@ -40,7 +40,7 @@ static int chameleon_starpu_init( starpu_conf_t *conf )
MPI_Initialized( &flag );
# endif
-#if defined(HAVE_STARPU_FXT_PROFILING)
+#if defined(STARPU_USE_FXT)
starpu_fxt_autostart_profiling(0);
#endif
diff --git a/runtime/starpu/control/runtime_profiling.c b/runtime/starpu/control/runtime_profiling.c
index 78d2394ef..3eb14c216 100644
--- a/runtime/starpu/control/runtime_profiling.c
+++ b/runtime/starpu/control/runtime_profiling.c
@@ -20,7 +20,7 @@
*/
#include <math.h>
#include "chameleon_starpu.h"
-#if defined(HAVE_STARPU_FXT_PROFILING)
+#if defined(STARPU_USE_FXT)
#include <starpu_fxt.h>
#endif
@@ -55,18 +55,14 @@ void RUNTIME_iteration_pop( CHAM_context_t *chamctxt )
}
void RUNTIME_start_profiling(){
-#if defined(HAVE_STARPU_FXT_PROFILING)
+#if defined(STARPU_USE_FXT)
starpu_fxt_start_profiling();
-#else
- fprintf(stderr, "Profiling throught FxT has not been enabled in StarPU runtime (configure StarPU with --with-fxt)\n");
#endif
}
void RUNTIME_stop_profiling(){
-#if defined(HAVE_STARPU_FXT_PROFILING)
+#if defined(STARPU_USE_FXT)
starpu_fxt_stop_profiling();
-#else
- fprintf(stderr, "Profiling throught FxT has not been enabled in StarPU runtime (configure StarPU with --with-fxt)\n");
#endif
}
diff --git a/runtime/starpu/include/chameleon_starpu.h.in b/runtime/starpu/include/chameleon_starpu.h.in
index 6cf49261c..223e5d045 100644
--- a/runtime/starpu/include/chameleon_starpu.h.in
+++ b/runtime/starpu/include/chameleon_starpu.h.in
@@ -24,7 +24,6 @@
#include "chameleon/config.h"
/* StarPU options */
-#cmakedefine HAVE_STARPU_FXT_PROFILING
#cmakedefine HAVE_STARPU_IDLE_PREFETCH
#cmakedefine HAVE_STARPU_ITERATION_PUSH
#cmakedefine HAVE_STARPU_DATA_WONT_USE
--
GitLab