From 2037e9be40edbc07a1417b20cf0ca164adf0d9d7 Mon Sep 17 00:00:00 2001
From: Raphael Boucherie <raphael.boucherie@inria.fr>
Date: Wed, 2 Aug 2017 14:32:47 +0200
Subject: [PATCH] Add lapack timing and fix --n_range option
---
timing/CMakeLists.txt | 27 ++++++++
timing/time_zgeqrf_lapacke.c | 127 +++++++++++++++++++++++++++++++++++
timing/timing.c | 51 +++++++++-----
3 files changed, 189 insertions(+), 16 deletions(-)
create mode 100644 timing/time_zgeqrf_lapacke.c
diff --git a/timing/CMakeLists.txt b/timing/CMakeLists.txt
index 19f3a16f7..e6621eb6d 100644
--- a/timing/CMakeLists.txt
+++ b/timing/CMakeLists.txt
@@ -229,6 +229,33 @@ endforeach()
#-------- Tests ---------
include(CTestLists.cmake)
+# Add BLAS/LAPACK timings
+set(ZSRC_LAPACKE
+ time_zgeqrf_lapacke.c
+ )
+
+precisions_rules_py(TIMINGS_LAPACKE "${ZSRC_LAPACKE}"
+ PRECISIONS "${CHAMELEON_PRECISION}")
+
+foreach(_timing ${TIMINGS_LAPACKE})
+ get_filename_component(_name_exe ${_timing} NAME_WE)
+ add_executable(${_name_exe} ${_timing})
+ add_dependencies(${_name_exe} timing_include)
+ set_property(TARGET ${_name_exe} PROPERTY LINKER_LANGUAGE Fortran)
+ target_compile_definitions(${_name_exe} PRIVATE TIMING_LAPACKE)
+ target_link_libraries(${_name_exe}
+ coreblas
+ ${LAPACKE_LIBRARIES}
+ ${TMG_LIBRARIES}
+ ${CBLAS_LIBRARIES}
+ ${LAPACK_MT_LIBRARIES}
+ ${BLAS_MT_LIBRARIES}
+ ${EXTRA_LIBRARIES}
+ )
+ install(TARGETS ${_name_exe}
+ DESTINATION ${CMAKE_INSTALL_PREFIX}/lib/chameleon/timing)
+endforeach()
+
###
### END CMakeLists.txt
###
diff --git a/timing/time_zgeqrf_lapacke.c b/timing/time_zgeqrf_lapacke.c
new file mode 100644
index 000000000..1753a6958
--- /dev/null
+++ b/timing/time_zgeqrf_lapacke.c
@@ -0,0 +1,127 @@
+/**
+ *
+ * @copyright (c) 2009-2014 The University of Tennessee and The University
+ * of Tennessee Research Foundation.
+ * All rights reserved.
+ * @copyright (c) 2012-2014 Inria. All rights reserved.
+ * @copyright (c) 2012-2014 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux. All rights reserved.
+ *
+ **/
+
+/**
+ *
+ * @precisions normal z -> c d s
+ *
+ **/
+#define _TYPE MORSE_Complex64_t
+#define _PREC double
+#define _LAMCH LAPACKE_dlamch_work
+
+#define _NAME "MORSE_zgeqrf"
+/* See Lawn 41 page 120 */
+#define _FMULS FMULS_GEQRF(M, N)
+#define _FADDS FADDS_GEQRF(M, N)
+
+#include "./timing.c"
+#include "timing_zauxiliary.h"
+
+#if defined( _WIN32 ) || defined( _WIN64 )
+#include <windows.h>
+#include <time.h>
+#include <sys/timeb.h>
+#if defined(_MSC_VER) || defined(_MSC_EXTENSIONS)
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000Ui64
+#else
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+struct timezone
+{
+ int tz_minuteswest; /* minutes W of Greenwich */
+ int tz_dsttime; /* type of dst correction */
+};
+
+int gettimeofday(struct timeval* tv, struct timezone* tz)
+{
+ FILETIME ft;
+ unsigned __int64 tmpres = 0;
+ static int tzflag;
+
+ if (NULL != tv)
+ {
+ GetSystemTimeAsFileTime(&ft);
+ tmpres |= ft.dwHighDateTime;
+ tmpres <<= 32;
+ tmpres |= ft.dwLowDateTime;
+
+ /*converting file time to unix epoch*/
+ tmpres /= 10; /*convert into microseconds*/
+ tmpres -= DELTA_EPOCH_IN_MICROSECS;
+
+ tv->tv_sec = (long)(tmpres / 1000000UL);
+ tv->tv_usec = (long)(tmpres % 1000000UL);
+ }
+ if (NULL != tz)
+ {
+ if (!tzflag)
+ {
+ _tzset();
+ tzflag++;
+ }
+ tz->tz_minuteswest = _timezone / 60;
+ tz->tz_dsttime = _daylight;
+ }
+ return 0;
+}
+
+#else /* Non-Windows */
+#include <sys/time.h>
+#endif
+
+/*
+ * struct timeval {time_t tv_sec; suseconds_t tv_usec;};
+ */
+double cWtime(void)
+{
+ struct timeval tp;
+ gettimeofday( &tp, NULL );
+ return tp.tv_sec + 1e-6 * tp.tv_usec;
+}
+
+static int
+RunTest(int *iparam, double *dparam, morse_time_t *t_)
+{
+ MORSE_Complex64_t *tau;
+ PASTE_CODE_IPARAM_LOCALS( iparam );
+
+ if ( M != N && check ) {
+ fprintf(stderr, "Check cannot be perfomed with M != N\n");
+ check = 0;
+ }
+
+ /* Allocate Data */
+ PASTE_CODE_ALLOCATE_MATRIX( A, 1, MORSE_Complex64_t, LDA, N );
+
+ /* Initialize Data */
+ CORE_zplrnt(M, N, A, LDA,
+ M, 0, 0, 3456);
+
+ /* Allocate Workspace */
+ tau = malloc( chameleon_min( M, N ) * sizeof(MORSE_Complex64_t) );
+
+ t = - cWtime();
+ LAPACKE_zgeqrf( LAPACK_COL_MAJOR, M, N, A, LDA, tau );
+ t += cWtime();
+ *t_ = t;
+
+ /* Check the solution */
+ if ( check )
+ {
+ }
+
+ /* Free Workspace */
+ free( tau );
+ free( A );
+
+ return 0;
+}
diff --git a/timing/timing.c b/timing/timing.c
index ca5f3d1b7..33c8ac131 100644
--- a/timing/timing.c
+++ b/timing/timing.c
@@ -63,6 +63,10 @@
#include <getopt.h>
#endif /* defined(CHAMELEON_HAVE_GETOPT_H) */
+#ifdef TIMING_LAPACKE
+int MORSE_My_Mpi_Rank() { return 0; }
+#endif
+
static int RunTest(int *iparam, _PREC *dparam, double *t_);
static inline void* morse_getaddr_null(const MORSE_desc_t *A, int m, int n)
{
@@ -198,6 +202,7 @@ Test(int64_t n, int *iparam) {
}
gflops = flops / t[iter];
+#ifndef TIMING_LAPACKE
#if defined (CHAMELEON_SCHED_STARPU)
/* TODO: create chameleon interface encapsulating this instead */
if (iparam[IPARAM_BOUND])
@@ -209,6 +214,7 @@ Test(int64_t n, int *iparam) {
upper_gflops = (flops / (tmin / 1000.0));
sumgf_upper += upper_gflops;
}
+#endif
#endif
sumt += t[iter];
sumgf += gflops;
@@ -496,7 +502,7 @@ set_iparam_default(int *iparam){
iparam[IPARAM_THRDNBR ] = -1;
iparam[IPARAM_THRDNBR_SUBGRP] = 1;
iparam[IPARAM_M ] = -1;
- iparam[IPARAM_N ] = 500;
+ iparam[IPARAM_N ] = -1;
iparam[IPARAM_K ] = 1;
iparam[IPARAM_LDA ] = -1;
iparam[IPARAM_LDB ] = -1;
@@ -624,7 +630,8 @@ parse_arguments(int *_argc, char ***_argv, int *iparam, int *start, int *stop, i
int
main(int argc, char *argv[]) {
- int i, m, mx, nx;
+ int i, m, n, mx, nx;
+ int status;
int nbnode = 1;
int start = 500;
int stop = 5000;
@@ -644,10 +651,12 @@ main(int argc, char *argv[]) {
}
#endif
+ n = iparam[IPARAM_N];
m = iparam[IPARAM_M];
mx = iparam[IPARAM_MX];
nx = iparam[IPARAM_NX];
+#ifndef TIMING_LAPACKE
/* Initialize MORSE */
MORSE_Init( iparam[IPARAM_THRDNBR],
iparam[IPARAM_NCUDAS] );
@@ -709,27 +718,37 @@ main(int argc, char *argv[]) {
if (step < 1) step = 1;
- int status = Test( -1, iparam ); /* print header */
+ status = Test( -1, iparam ); /* print header */
if (status != MORSE_SUCCESS) return status;
- for (i = start; i <= stop; i += step)
- {
- if ( nx > 0 ) {
- iparam[IPARAM_M] = i;
- iparam[IPARAM_N] = chameleon_max(1, i/nx);
- } else if ( mx > 0 ) {
- iparam[IPARAM_M] = chameleon_max(1, i/mx);
- iparam[IPARAM_N] = i;
- } else {
- if ( m == -1 )
+ if ( n == -1 ){
+ for (i = start; i <= stop; i += step)
+ {
+ if ( nx > 0 ) {
iparam[IPARAM_M] = i;
- iparam[IPARAM_N] = i;
+ iparam[IPARAM_N] = chameleon_max(1, i/nx);
+ } else if ( mx > 0 ) {
+ iparam[IPARAM_M] = chameleon_max(1, i/mx);
+ iparam[IPARAM_N] = i;
+ } else {
+ if ( m == -1 )
+ iparam[IPARAM_M] = i;
+ iparam[IPARAM_N] = i;
+ }
+ status = Test( iparam[IPARAM_N], iparam );
+ if (status != MORSE_SUCCESS) return status;
+ success += status;
}
- int status = Test( iparam[IPARAM_N], iparam );
+ } else{
+ if ( m == -1 )
+ iparam[IPARAM_M] = n;
+ iparam[IPARAM_N] = n;
+ status = Test( iparam[IPARAM_N], iparam );
if (status != MORSE_SUCCESS) return status;
success += status;
}
-
+#ifndef TIMING_LAPACKE
MORSE_Finalize();
+#endif
return success;
}
--
GitLab