Re: [Starpu-devel] Building StarPU from SVN on OSX [patch]

[George Russell <george@codeplay.com>]

Hi,

I have attached a patch to get StarPU OpenCL working (partly!) on OSX 10.6

This patch primarily does two things; it fixed the include paths for the CL headers in C code (these are, for some reason, different on OSX), and it ifdefs out the cpu_set_t code on OSX (as is done on Windows) as that type seems not to be supported on OSX. (it also fixed some compilation errors re INT_MAX by including limits.h)

Note that the changes I have made to configure.ac are essentially broken; It needs to conditionally handle OSX (which I don't know how to do, especially). However, there are ~3 things to note: The header test needed to be changed, which I have done; but it should test the CL/cl.h on non OSX, and OpenCL/opencl.h on OSX. I did not get the library check to work on OSX; I just forced the enable_opencl to be on. At least for OSX 10.6, OpenCL is supplied with the OS, and in a fixed location. So, OpenCL could be enabled on OSX 10.6 by default, unless explicitly disabled. I did however, add the "-framework OpenCL" to the LD_FLAGS to get it to link. Again, this should be OSX specific.

I forced the target to i686 to get the pthread configure check to pass; I believe this results in the compilation of a 32bit executable. I guess StarPU will simply not build for X86_64?

Now, the bad part: It seems broken, in the same way as the ATI CPU OpenCL was e.g. it appears to hang/deadlock in the same place.

./configure --target=i686-apple-darwin10 --with-opencl-include-dir=/System/Library/Frameworks/OpenCL.framework/Versions/A/Headers/

imac:StarPU grrussel$ examples/basic_examples/vector_scal
BEFORE : First element was 1.000000
[starpu][_starpu_opencl_init] Initialising OpenCL
[starpu][_starpu_opencl_init] Platforms detected: 1
[starpu][_starpu_opencl_init] Platform: Apple - Apple
[starpu][_starpu_opencl_init]   1 devices detected
Current configuration does not match the performance model (OpenCL: (stored) 0 != (current) 1), recalibrating...[starpu][benchmark_all_gpu_devices] Benchmarking the speed of the bus
[starpu][_starpu_opencl_init_context] Initialising context for dev 0
[starpu][_starpu_opencl_deinit_context] De-initialising context for dev 0
[starpu][_starpu_opencl_init_context] Initialising context for dev 0
[starpu][_starpu_opencl_deinit_context] De-initialising context for dev 0
[starpu][measure_bandwidth_between_host_and_dev] BANDWIDTH GPU 0 CPU 0 - htod 16782.898438 - dtoh 15493.679688 - 22841.186270
[starpu][measure_bandwidth_between_host_and_dev] BANDWIDTH GPU 0 CPU 1 - htod 15700.906250 - dtoh 15363.750000 - 21967.322803
[starpu][measure_bandwidth_between_host_and_dev] BANDWIDTH GPU 0 BEST CPU 0
[starpu][benchmark_all_gpu_devices] Benchmarking the speed of the bus is done.
done
[starpu][load_sched_policy] Use eager scheduler (greedy policy)
[starpu][_starpu_launch_drivers] initialising worker 0
[starpu][_starpu_opencl_init_context] Initialising context for dev 0
[starpu][_starpu_opencl_get_device_name] Device 0 : [GeForce 9400]
[starpu][_starpu_opencl_worker] OpenCL (GeForce 9400) dev id 0 thread is ready to run on CPU 0 !
[starpu][_starpu_opencl_locate_file] Trying to locate <examples/basic_examples/vector_scal_opencl_kernel.cl>
[starpu][starpu_opencl_load_opencl_from_file] Source file name : <examples/basic_examples/vector_scal_opencl_kernel.cl>
[starpu][_starpu_opencl_load_program_source] OpenCL kernel </* StarPU --- Runtime system for heterogeneous multicore architectures.
 *
 * Copyright (C) 2010  Centre National de la Recherche Scientifique
 *
 * StarPU is free software; you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published by
 * the Free Software Foundation; either version 2.1 of the License, or (at
 * your option) any later version.
 *
 * StarPU is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
 *
 * See the GNU Lesser General Public License in COPYING.LGPL for more details.
 */

__kernel void vector_mult_opencl(__global float* val, int nx, float factor)
{
        const int i = get_global_id(0);
        if (i < nx) {
                val[i] *= factor;
        }
}
>
^C


Cheers,
George

On 23/02/2011 21:47, George Russell wrote:

4D657251.3070402@codeplay.com"> Hi,

as a follow up to this;

The autogen.sh script invokes libtool on OSX, but that should (on OSX) be glibtool

The OpenCL checks don't work on OSX, for two reasons; the includes are different, and the library linking system is also different.

The includes in C code to use OpenCL should be

#ifdef __APPLE__
#include <OpenCL/opencl.h>
#else
#include <CL/cl.h>
#endif

The command to compile (and link) the code uses OSX's framework switch to gcc e.g. gcc -o foo foo.c -framework OpenCL

Knowing that, however, is not much help to me in getting a working configure.ac ;-)

Cheers,
George

On 23/02/2011 17:31, George Russell wrote:

4D653678.7080007@codeplay.com"> Hi,

I was also attempting to build StarPU from SVN on OSX.

However, on running the autogen.sh, I get the following output:

imac:StarPU grrussel$ sh autogen.sh
libtool: unknown option character `-' in: --version
Usage: libtool -static [-] file [...] [-filelist listfile[,dirname]] [-arch_only arch] [-sacLT]
Usage: libtool -dynamic [-] file [...] [-filelist listfile[,dirname]] [-arch_only arch] [-o output] [-install_name name] [-compatibility_version #] [-current_version #] [-seg1addr 0x#] [-segs_read_only_addr 0x#] [-segs_read_write_addr 0x#] [-seg_addr_table <filename>] [-seg_addr_table_filename <file_system_path>] [-all_load] [-noall_load]
Libtool is missing, please install it.


It looks like I have the wrong libtool installed? It would be the default from the OSX developer tools.

Is there any special care needed in compiling on OSX?

Cheers,
George

_______________________________________________
Starpu-devel mailing list
Starpu-devel@lists.gforge.inria.fr
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel

_______________________________________________
Starpu-devel mailing list
Starpu-devel@lists.gforge.inria.fr
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel

Index: include/starpu_opencl.h
===================================================================
--- include/starpu_opencl.h     (revision 3238)
+++ include/starpu_opencl.h     (working copy)
@@ -19,7 +19,11 @@
 #define __STARPU_OPENCL_H__
 
 #ifdef STARPU_USE_OPENCL
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
+#endif
 
 #ifdef __cplusplus
 extern "C" {
Index: src/core/workers.h
===================================================================
--- src/core/workers.h  (revision 3238)
+++ src/core/workers.h  (working copy)
@@ -79,8 +79,10 @@
        char name[32];
 
 #ifndef STARPU_HAVE_WINDOWS
+#ifndef __APPLE__
        cpu_set_t initial_cpu_set;
        cpu_set_t current_cpu_set;
+#endif /* __APPLE__ */
 #endif /* STARPU_HAVE_WINDOWS */
 #ifdef STARPU_HAVE_HWLOC
        hwloc_cpuset_t initial_hwloc_cpu_set;
@@ -96,7 +98,9 @@
        int combined_workerid[STARPU_NMAXWORKERS];
 
 #ifndef STARPU_HAVE_WINDOWS
+#ifndef __APPLE__
        cpu_set_t cpu_set;
+#endif /* __APPLE__ */
 #endif /* STARPU_HAVE_WINDOWS */
 #ifdef STARPU_HAVE_HWLOC
        hwloc_cpuset_t hwloc_cpu_set;
Index: src/core/topology.c
===================================================================
--- src/core/topology.c (revision 3238)
+++ src/core/topology.c (working copy)
@@ -723,11 +723,13 @@
                workerarg->memory_node = memory_node;
 
 #ifndef STARPU_HAVE_WINDOWS
+#ifndef __APPLE__
                /* Save the initial cpuset */
                CPU_ZERO(&workerarg->initial_cpu_set);
                CPU_SET(workerarg->bindid, &workerarg->initial_cpu_set);
                CPU_ZERO(&workerarg->current_cpu_set);
                CPU_SET(workerarg->bindid, &workerarg->current_cpu_set);
+#endif /* __APPLE__ */
 #endif /* STARPU_HAVE_WINDOWS */
 
 #ifdef STARPU_HAVE_HWLOC
Index: src/core/combined_workers.c
===================================================================
--- src/core/combined_workers.c (revision 3238)
+++ src/core/combined_workers.c (working copy)
@@ -114,7 +114,9 @@
        /* Note that we maintain both the cpu_set and the hwloc_cpu_set so 
that
         * the application is not forced to use hwloc when it is available. */
 #ifndef STARPU_HAVE_WINDOWS
+#ifndef __APPLE__
        CPU_ZERO(&combined_worker->cpu_set);
+#endif
 #endif /* STARPU_HAVE_WINDOWS */
 
 #ifdef STARPU_HAVE_HWLOC
@@ -125,6 +127,7 @@
        {
                int id = workerid_array[i];
 #ifndef STARPU_HAVE_WINDOWS
+#ifndef __APPLE__
 #ifdef CPU_OR
                CPU_OR(&combined_worker->cpu_set,
                        &combined_worker->cpu_set,
@@ -136,6 +139,7 @@
                                CPU_SET(j, &combined_worker->cpu_set);
                }
 #endif
+#endif /* __APPLE__ */
 #endif /* STARPU_HAVE_WINDOWS */
 
 #ifdef STARPU_HAVE_HWLOC
Index: src/datawizard/copy_driver.h
===================================================================
--- src/datawizard/copy_driver.h        (revision 3238)
+++ src/datawizard/copy_driver.h        (working copy)
@@ -30,8 +30,12 @@
 #endif
 
 #ifdef STARPU_USE_OPENCL
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
 #endif
+#endif
 
 struct starpu_data_request_s;
 struct starpu_data_replicate_s;
Index: src/drivers/opencl/driver_opencl.h
===================================================================
--- src/drivers/opencl/driver_opencl.h  (revision 3238)
+++ src/drivers/opencl/driver_opencl.h  (working copy)
@@ -24,7 +24,11 @@
 
 #ifdef STARPU_USE_OPENCL
 
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
+#endif
 
 extern
 int _starpu_opencl_init_context(int devid);
Index: tests/datawizard/increment_redux_v2.c
===================================================================
--- tests/datawizard/increment_redux_v2.c       (revision 3238)
+++ tests/datawizard/increment_redux_v2.c       (working copy)
@@ -20,8 +20,12 @@
 #include <cuda.h>
 #endif
 #ifdef STARPU_USE_OPENCL
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
 #endif
+#endif
 
 
 static unsigned var = 0;
Index: tests/datawizard/write_only_tmp_buffer.c
===================================================================
--- tests/datawizard/write_only_tmp_buffer.c    (revision 3238)
+++ tests/datawizard/write_only_tmp_buffer.c    (working copy)
@@ -26,7 +26,13 @@
 starpu_data_handle v_handle;
 
 #ifdef STARPU_USE_OPENCL
+
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
+#endif
+
 #include <starpu_opencl.h>
 
 static void opencl_codelet_null(void *descr[], __attribute__ ((unused)) void 
*_args)
Index: tests/datawizard/sync_and_notify_data_opencl.c
===================================================================
--- tests/datawizard/sync_and_notify_data_opencl.c      (revision 3238)
+++ tests/datawizard/sync_and_notify_data_opencl.c      (working copy)
@@ -17,7 +17,12 @@
 
 #include <starpu.h>
 #include <starpu_opencl.h>
+
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
+#endif
 
 extern struct starpu_opencl_program opencl_code;
 
Index: tests/datawizard/manual_reduction.c
===================================================================
--- tests/datawizard/manual_reduction.c (revision 3238)
+++ tests/datawizard/manual_reduction.c (working copy)
@@ -19,8 +19,12 @@
 #include <cuda.h>
 #endif
 #ifdef STARPU_USE_OPENCL
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
 #endif
+#endif
 
 #define INIT_VALUE     42
 #define NTASKS         10000
Index: tests/datawizard/increment_redux.c
===================================================================
--- tests/datawizard/increment_redux.c  (revision 3238)
+++ tests/datawizard/increment_redux.c  (working copy)
@@ -21,8 +21,12 @@
 #include <cuda.h>
 #endif
 #ifdef STARPU_USE_OPENCL
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
 #endif
+#endif
 
 
 static unsigned var = 0;
Index: configure.ac
===================================================================
--- configure.ac        (revision 3238)
+++ configure.ac        (working copy)
@@ -258,6 +258,7 @@
 
        AC_HAVE_LIBRARY([cuda],[],[have_valid_cuda=no])
 
+
        # To detect if we have a recent enough CUDA lib
        AC_HAVE_LIBRARY([cudart],[],[have_valid_cuda=no])
 
@@ -443,7 +444,7 @@
 
        # do we have a valid OpenCL setup ?
        have_valid_opencl=yes
-       AC_CHECK_HEADER([CL/cl.h],,[have_valid_opencl=no])
+       AC_CHECK_HEADER([OpenCL/opencl.h],,[have_valid_opencl=no])
 
        # we are looking for the proper option in LDFLAGS, so we save the
        # current value of LDFLAGS so that we can add new things in it and
@@ -456,7 +457,7 @@
        fi
 
        if test -d "$opencl_lib_dir"; then
-               LDFLAGS="${SAVED_LDFLAGS} -L$opencl_lib_dir "
+               LDFLAGS="${SAVED_LDFLAGS} -L$opencl_lib_dir -framework OpenCL"
        fi
        AC_HAVE_LIBRARY([OpenCL],[],[have_valid_opencl=no])
 
@@ -466,7 +467,8 @@
        fi
 
        # now we enable OpenCL if and only if a proper setup is available
-       enable_opencl=$have_valid_opencl
+       #enable_opencl=$have_valid_opencl
+       enable_opencl=yes
 fi
 
 AC_MSG_CHECKING(whether OpenCL should be used)
Index: autogen.sh
===================================================================
--- autogen.sh  (revision 3238)
+++ autogen.sh  (working copy)
@@ -16,7 +16,7 @@
 # 
 # See the GNU Lesser General Public License in COPYING.LGPL for more details.
 
-if ! libtool --version > /dev/null
+if ! glibtool --version > /dev/null
 then
        echo "Libtool is missing, please install it."
        exit 1
Index: examples/cholesky/cholesky_implicit.c
===================================================================
--- examples/cholesky/cholesky_implicit.c       (revision 3238)
+++ examples/cholesky/cholesky_implicit.c       (working copy)
@@ -17,7 +17,7 @@
  */
 
 #include "cholesky.h"
-
+#include <limits.h>
 /*
  *     Create the codelets
  */
Index: examples/mult/sgemm.c
===================================================================
--- examples/mult/sgemm.c       (revision 3238)
+++ examples/mult/sgemm.c       (working copy)
@@ -14,6 +14,6 @@
  *
  * See the GNU Lesser General Public License in COPYING.LGPL for more 
details.
  */
-
+#include <limits.h>
 #include "simple.h"
 #include "xgemm.c" 
Index: examples/mult/xgemm.c
===================================================================
--- examples/mult/xgemm.c       (revision 3238)
+++ examples/mult/xgemm.c       (working copy)
@@ -15,7 +15,7 @@
  *
  * See the GNU Lesser General Public License in COPYING.LGPL for more 
details.
  */
-
+#include <limits.h>
 #include <string.h>
 #include <math.h>
 #include <sys/types.h>
Index: examples/mandelbrot/mandelbrot.c
===================================================================
--- examples/mandelbrot/mandelbrot.c    (revision 3238)
+++ examples/mandelbrot/mandelbrot.c    (working copy)
@@ -14,7 +14,7 @@
  *
  * See the GNU Lesser General Public License in COPYING.LGPL for more 
details.
  */
-
+#include <limits.h>
 #include <starpu.h>
 #ifdef STARPU_USE_OPENCL
 #include <starpu_opencl.h>
Index: examples/incrementer/incrementer_kernels_opencl.c
===================================================================
--- examples/incrementer/incrementer_kernels_opencl.c   (revision 3238)
+++ examples/incrementer/incrementer_kernels_opencl.c   (working copy)
@@ -17,8 +17,11 @@
 
 #include <starpu.h>
 #include <starpu_opencl.h>
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
 #include <CL/cl.h>
-
+#endif
 extern struct starpu_opencl_program opencl_program;
 void opencl_codelet(void *descr[], void *_args)
 {