do not use typedef for starpu_codelet, use it directly as a struct

doc/chapters/advanced-api.texi

@@ -84,14 +84,14 @@ starpu_data_handle_t handle;
  */
 #ifdef STARPU_USE_OPENCL
 void cpu_to_opencl_opencl_func(void *buffers[], void *args);
-starpu_codelet cpu_to_opencl_cl = @{
+struct starpu_codelet cpu_to_opencl_cl = @{
     .where = STARPU_OPENCL,
     .opencl_func = cpu_to_opencl_opencl_func,
     .nbuffers = 1
 @};
 
 void opencl_to_cpu_func(void *buffers[], void *args);
-starpu_codelet opencl_to_cpu_cl = @{
+struct starpu_codelet opencl_to_cpu_cl = @{
     .where = STARPU_CPU,
     .cpu_func = opencl_to_cpu_func,
     .nbuffers = 1

doc/chapters/advanced-examples.texi

@@ -49,14 +49,14 @@ void scal_sse_func(void *buffers[], void *cl_arg)
 @end smallexample
 @end cartouche
 
-The @code{cpu_func} field of the @code{starpu_codelet} structure has to be set
+The @code{cpu_func} field of the @code{struct starpu_codelet} structure has to be set
 to the special value @code{STARPU_MULTIPLE_CPU_IMPLEMENTATIONS}. Note that
 @code{STARPU_MULTIPLE_CUDA_IMPLEMENTATIONS} and
 @code{STARPU_MULTIPLE_OPENCL_IMPLEMENTATIONS} are also available.
 
 @cartouche
 @smallexample
-starpu_codelet cl = @{
+struct starpu_codelet cl = @{
     .where = STARPU_CPU,
     .cpu_func = STARPU_MULTIPLE_CPU_IMPLEMENTATIONS,
     .cpu_funcs = @{ scal_cpu_func, scal_sse_func @},
@@ -181,7 +181,7 @@ Partitioning can be applied several times, see
 To achieve good scheduling, StarPU scheduling policies need to be able to
 estimate in advance the duration of a task. This is done by giving to codelets
 a performance model, by defining a @code{starpu_perfmodel} structure and
-providing its address in the @code{model} field of the @code{starpu_codelet}
+providing its address in the @code{model} field of the @code{struct starpu_codelet}
 structure. The @code{symbol} and @code{type} fields of @code{starpu_perfmodel}
 are mandatory, to give a name to the model, and the type of the model, since
 there are several kinds of performance models.
@@ -218,7 +218,7 @@ static struct starpu_perfmodel mult_perf_model = @{
     .symbol = "mult_perf_model"
 @};
 
-starpu_codelet cl = @{
+struct starpu_codelet cl = @{
     .where = STARPU_CPU,
     .cpu_func = cpu_mult,
     .nbuffers = 3,
@@ -324,7 +324,7 @@ transfers, which are assumed to be completely overlapped.
 StarPU provides the wrapper function @code{starpu_insert_task} to ease
 the creation and submission of tasks.
 
-@deftypefun int starpu_insert_task (starpu_codelet *@var{cl}, ...)
+@deftypefun int starpu_insert_task (struct starpu_codelet *@var{cl}, ...)
 Create and submit a task corresponding to @var{cl} with the following
 arguments.  The argument list must be zero-terminated.
 
@@ -371,7 +371,7 @@ void func_cpu(void *descr[], void *_args)
         *x1 = *x1 * ffactor;
 @}
 
-starpu_codelet mycodelet = @{
+struct starpu_codelet mycodelet = @{
         .where = STARPU_CPU,
         .cpu_func = func_cpu,
         .nbuffers = 2

doc/chapters/basic-api.texi

@@ -372,7 +372,7 @@ This function allows to specify that a piece of data can be discarded
 without impacting the application.
 @end deftypefun
 
-@deftypefun void starpu_data_set_reduction_methods (starpu_data_handle_t @var{handle}, {struct starpu_codelet_t *}@var{redux_cl}, {struct starpu_codelet_t *}@var{init_cl})
+@deftypefun void starpu_data_set_reduction_methods (starpu_data_handle_t @var{handle}, {struct starpu_codelet *}@var{redux_cl}, {struct starpu_codelet *}@var{init_cl})
 todo
 @end deftypefun
 
@@ -1095,7 +1095,7 @@ with @code{starpu_task_create}.
 
 @table @asis
 @item @code{cl}
-Is a pointer to the corresponding @code{starpu_codelet} data structure. This
+Is a pointer to the corresponding @code{struct starpu_codelet} data structure. This
 describes where the kernel should be executed, and supplies the appropriate
 implementations. When set to @code{NULL}, no code is executed during the tasks,
 such empty tasks can be useful for synchronization purposes.
@@ -1107,7 +1107,7 @@ The @code{struct starpu_buffer_descr} structure is composed of two fields, the
 @code{handle} field specifies the handle of the piece of data, and the
 @code{mode} field is the required access mode (eg @code{STARPU_RW}). The number
 of entries in this array must be specified in the @code{nbuffers} field of the
-@code{starpu_codelet} structure, and should not excede @code{STARPU_NMAXBUFS}.
+@code{struct starpu_codelet} structure, and should not excede @code{STARPU_NMAXBUFS}.
 If unsufficient, this value can be set with the @code{--enable-maxbuffers}
 option when configuring StarPU.
 
@@ -1260,7 +1260,7 @@ NULL if it is called either from a thread that is not a task or simply
 because there is no task being executed at the moment.
 @end deftypefun
 
-@deftypefun void starpu_display_codelet_stats ({struct starpu_codelet_t} *@var{cl})
+@deftypefun void starpu_display_codelet_stats ({struct starpu_codelet} *@var{cl})
 Output on @code{stderr} some statistics on the codelet @var{cl}.
 @end deftypefun
 
@@ -1709,7 +1709,7 @@ executed by every StarPU worker that may execute the function.
 The second argument is passed to the offloaded function.
 The last argument specifies on which types of processing units the function
 should be executed. Similarly to the @var{where} field of the
-@code{starpu_codelet} structure, it is possible to specify that the function
+@code{struct starpu_codelet} structure, it is possible to specify that the function
 should be executed on every CUDA device and every CPU by passing
 @code{STARPU_CPU|STARPU_CUDA}.
 This function blocks until the function has been executed on every appropriate

doc/chapters/basic-examples.texi

@@ -80,7 +80,7 @@ void cpu_func(void *buffers[], void *cl_arg)
     printf("Hello world (params = @{%i, %f@} )\n", params->i, params->f);
 @}
 
-starpu_codelet cl =
+struct starpu_codelet cl =
 @{
     .where = STARPU_CPU,
     .cpu_func = cpu_func,
@@ -316,7 +316,7 @@ void scal_cpu_func(void *buffers[], void *cl_arg)
         val[i] *= *factor;
 @}
 
-starpu_codelet cl = @{
+struct starpu_codelet cl = @{
     .where = STARPU_CPU,
     .cpu_func = scal_cpu_func,
     .nbuffers = 1
@@ -492,7 +492,7 @@ extern void scal_cpu_func(void *buffers[], void *_args);
 extern void scal_opencl_func(void *buffers[], void *_args);
 
 /* @b{Definition of the codelet} */
-static starpu_codelet cl = @{
+static struct starpu_codelet cl = @{
     .where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL; /* @b{It can be executed on a CPU,} */
                                      /* @b{on a CUDA device, or on an OpenCL device} */
     .cuda_func = scal_cuda_func,

doc/chapters/mpi-support.texi

@@ -226,7 +226,7 @@ to it.
 Returns the last value set by @code{starpu_data_set_rank}.
 @end deftypefun
 
-@deftypefun void starpu_mpi_insert_task (MPI_Comm @var{comm}, starpu_codelet *@var{cl}, ...)
+@deftypefun void starpu_mpi_insert_task (MPI_Comm @var{comm}, struct starpu_codelet *@var{cl}, ...)
 Create and submit a task corresponding to @var{cl} with the following
 arguments.  The argument list must be zero-terminated.
 

doc/chapters/perf-feedback.texi

@@ -72,7 +72,7 @@ the @code{starpu_get_current_task()} function.
 @node Codelet feedback
 @subsection Per-codelet feedback
 
-The @code{per_worker_stats} field of the @code{starpu_codelet_t} structure is
+The @code{per_worker_stats} field of the @code{struct starpu_codelet} structure is
 an array of counters. The i-th entry of the array is incremented every time a
 task implementing the codelet is executed on the i-th worker.
 This array is not reinitialized when profiling is enabled or disabled.

doc/chapters/vector_scal_c.texi

@@ -23,7 +23,7 @@ extern void scal_sse_func(void *buffers[], void *_args);
 extern void scal_cuda_func(void *buffers[], void *_args);
 extern void scal_opencl_func(void *buffers[], void *_args);
 
-static starpu_codelet cl = @{
+static struct starpu_codelet cl = @{
     .where = STARPU_CPU | STARPU_CUDA | STARPU_OPENCL,
     /* CPU implementation of the codelet */
     .cpu_func = STARPU_MULTIPLE_CPU_IMPLEMENTATIONS,

doc/tutorial/hello_world.c

@@ -42,7 +42,7 @@ void cpu_func(void *buffers[], void *cl_arg)
     printf("Hello world (params = {%i, %f} )\n", params->i, params->f);
 }
 
-starpu_codelet cl =
+struct starpu_codelet cl =
 {
     .where = STARPU_CPU,
     .cpu_func = cpu_func,

doc/tutorial/vector_scal.c

@@ -44,7 +44,7 @@ extern void scal_cpu_func(void *buffers[], void *_args);
 extern void scal_cuda_func(void *buffers[], void *_args);
 extern void scal_opencl_func(void *buffers[], void *_args);
 
-static starpu_codelet cl = {
+static struct starpu_codelet cl = {
     .where = STARPU_CPU | STARPU_CUDA | STARPU_OPENCL,
     /* CPU implementation of the codelet */
     .cpu_func = scal_cpu_func,

examples/audio/starpu_audio_processing.c

@@ -273,7 +273,7 @@ struct starpu_perfmodel band_filter_model = {
 	.symbol = "FFT_band_filter"
 };
 
-static starpu_codelet band_filter_cl = {
+static struct starpu_codelet band_filter_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 #ifdef STARPU_USE_CUDA
 	.cuda_func = band_filter_kernel_gpu,

examples/axpy/axpy.c

@@ -70,7 +70,7 @@ void axpy_gpu(void *descr[], __attribute__((unused)) void *arg)
 }
 #endif
 
-static starpu_codelet axpy_cl = {
+static struct starpu_codelet axpy_cl = {
         .where =
 #ifdef STARPU_USE_CUDA
                 STARPU_CUDA|

examples/basic_examples/block.c

@@ -35,7 +35,7 @@ typedef void (*device_func)(void **, void *);
 
 int execute_on(uint32_t where, device_func func, float *block, int pnx, int pny, int pnz, float multiplier)
 {
-	starpu_codelet cl = {};
+	struct starpu_codelet cl = {};
 	starpu_data_handle_t block_handle;
         int i;
 

examples/basic_examples/hello_world.c

@@ -57,7 +57,7 @@ void cpu_func(void *buffers[], void *cl_arg)
 	FPRINTF(stdout, "Hello world (params = {%i, %f} )\n", params->i, params->f);
 }
 
-starpu_codelet cl = {};
+struct starpu_codelet cl = {};
 
 int main(int argc, char **argv)
 {

examples/basic_examples/mult.c

@@ -251,7 +251,7 @@ static struct starpu_perfmodel mult_perf_model = {
 	.symbol = "mult_perf_model"
 };
 
-static starpu_codelet cl = {
+static struct starpu_codelet cl = {
         /* we can only execute that kernel on a CPU yet */
         .where = STARPU_CPU,
         /* CPU implementation of the codelet */

examples/basic_examples/multiformat.c

@@ -37,13 +37,13 @@ multiformat_scal_cpu_func(void *buffers[], void *args)
 }
 
 #ifdef STARPU_USE_CUDA
-extern starpu_codelet cpu_to_cuda_cl;
-extern starpu_codelet cuda_to_cpu_cl;
+extern struct starpu_codelet cpu_to_cuda_cl;
+extern struct starpu_codelet cuda_to_cpu_cl;
 #endif
 
 #ifdef STARPU_USE_OPENCL
-extern starpu_codelet cpu_to_opencl_cl;
-extern starpu_codelet opencl_to_cpu_cl;
+extern struct starpu_codelet cpu_to_opencl_cl;
+extern struct starpu_codelet opencl_to_cpu_cl;
 #endif
 
 static struct starpu_multiformat_data_interface_ops format_ops = {
@@ -73,7 +73,7 @@ static struct starpu_perfmodel conversion_model = {
 	.symbol = "multiformat_conversion_model"
 };
 
-static starpu_codelet  cl = {
+static struct starpu_codelet  cl = {
 	.where = STARPU_CPU | STARPU_CUDA | STARPU_OPENCL,
 	.cpu_func = multiformat_scal_cpu_func,
 #ifdef STARPU_USE_CUDA

examples/basic_examples/multiformat_conversion_codelets.c

@@ -42,14 +42,14 @@ void cuda_to_cpu(void *buffers[], void *arg)
 }
 
 extern void cpu_to_cuda_cuda_func(void *buffers[], void *args);
-starpu_codelet cpu_to_cuda_cl = {
+struct starpu_codelet cpu_to_cuda_cl = {
 	.where = STARPU_CUDA,
 	.cpu_func = cpu_to_cuda,
 	.cuda_func = cpu_to_cuda_cuda_func,
 	.nbuffers = 1
 };
 
-starpu_codelet cuda_to_cpu_cl = {
+struct starpu_codelet cuda_to_cpu_cl = {
 	.where = STARPU_CPU,
 	.cpu_func = cuda_to_cpu,
 	.nbuffers = 1
@@ -84,14 +84,14 @@ void opencl_to_cpu(void *buffers[], void *arg)
 }
 
 extern void cpu_to_opencl_opencl_func(void *buffers[], void *args);
-starpu_codelet cpu_to_opencl_cl = {
+struct starpu_codelet cpu_to_opencl_cl = {
 	.where = STARPU_OPENCL,
 	.cpu_func = cpu_to_opencl,
 	.opencl_func = cpu_to_opencl_opencl_func,
 	.nbuffers = 1
 };
 
-starpu_codelet opencl_to_cpu_cl = {
+struct starpu_codelet opencl_to_cpu_cl = {
 	.where = STARPU_CPU,
 	.cpu_func = opencl_to_cpu,
 	.nbuffers = 1

examples/basic_examples/variable.c

@@ -39,7 +39,7 @@ int main(int argc, char **argv)
 	unsigned i;
         float foo;
 	starpu_data_handle_t float_array_handle;
-	starpu_codelet cl = {};
+	struct starpu_codelet cl = {};
 
 	starpu_init(NULL);
 

examples/basic_examples/vector_scal.c

@@ -47,7 +47,7 @@ static struct starpu_perfmodel vector_scal_power_model = {
 	.symbol = "vector_scale_power"
 };
 
-static starpu_codelet cl = {
+static struct starpu_codelet cl = {
 	.where = STARPU_CPU | STARPU_CUDA | STARPU_OPENCL,
 	/* CPU implementation of the codelet */
 	.cpu_func = STARPU_MULTIPLE_CPU_IMPLEMENTATIONS,

examples/basic_examples/vector_scal_c.c

@@ -38,7 +38,7 @@ static struct starpu_perfmodel vector_scal_model = {
 	.symbol = "vector_scale_model"
 };
 
-static starpu_codelet cl = {
+static struct starpu_codelet cl = {
   .where = STARPU_CPU | STARPU_CUDA,
 	/* CPU implementation of the codelet */
 	.cpu_func = scal_cpu_func,

examples/callback/callback.c

@@ -30,7 +30,7 @@ void cpu_codelet(void *descr[], __attribute__ ((unused)) void *_args)
 	*val += 1;
 }
 
-starpu_codelet cl =
+struct starpu_codelet cl =
 {
 	.where = STARPU_CPU,
 	.cpu_func = cpu_codelet,

examples/cg/cg.c

@@ -84,10 +84,10 @@ static TYPE *r, *d, *q;
 static starpu_data_handle_t dtq_handle, rtr_handle;
 static TYPE dtq, rtr;
 
-extern starpu_codelet accumulate_variable_cl;
-extern starpu_codelet accumulate_vector_cl;
-extern starpu_codelet bzero_variable_cl;
-extern starpu_codelet bzero_vector_cl;
+extern struct starpu_codelet accumulate_variable_cl;
+extern struct starpu_codelet accumulate_vector_cl;
+extern struct starpu_codelet bzero_variable_cl;
+extern struct starpu_codelet bzero_vector_cl;
 
 /*
  *	Generate Input data

examples/cg/cg_kernels.c

@@ -73,7 +73,7 @@ static struct starpu_perfmodel accumulate_variable_model = {
 	.symbol = "accumulate_variable"
 };
 
-starpu_codelet accumulate_variable_cl = {
+struct starpu_codelet accumulate_variable_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = accumulate_variable_cpu,
 #ifdef STARPU_USE_CUDA
@@ -109,7 +109,7 @@ static struct starpu_perfmodel accumulate_vector_model = {
 	.symbol = "accumulate_vector"
 };
 
-starpu_codelet accumulate_vector_cl = {
+struct starpu_codelet accumulate_vector_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = accumulate_vector_cpu,
 #ifdef STARPU_USE_CUDA
@@ -147,7 +147,7 @@ static struct starpu_perfmodel bzero_variable_model = {
 	.symbol = "bzero_variable"
 };
 
-starpu_codelet bzero_variable_cl = {
+struct starpu_codelet bzero_variable_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = bzero_variable_cpu,
 #ifdef STARPU_USE_CUDA
@@ -182,7 +182,7 @@ static struct starpu_perfmodel bzero_vector_model = {
 	.symbol = "bzero_vector"
 };
 
-starpu_codelet bzero_vector_cl = {
+struct starpu_codelet bzero_vector_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = bzero_vector_cpu,
 #ifdef STARPU_USE_CUDA
@@ -235,7 +235,7 @@ static struct starpu_perfmodel dot_kernel_model = {
 	.symbol = "dot_kernel"
 };
 
-static starpu_codelet dot_kernel_cl = {
+static struct starpu_codelet dot_kernel_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dot_kernel_cpu,
 #ifdef STARPU_USE_CUDA
@@ -302,7 +302,7 @@ static struct starpu_perfmodel scal_kernel_model = {
 	.symbol = "scal_kernel"
 };
 
-static starpu_codelet scal_kernel_cl = {
+static struct starpu_codelet scal_kernel_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = scal_kernel_cpu,
 #ifdef STARPU_USE_CUDA
@@ -373,7 +373,7 @@ static struct starpu_perfmodel gemv_kernel_model = {
 	.symbol = "gemv_kernel"
 };
 
-static starpu_codelet gemv_kernel_cl = {
+static struct starpu_codelet gemv_kernel_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.type = STARPU_SPMD,
 	.max_parallelism = INT_MAX,
@@ -465,7 +465,7 @@ static struct starpu_perfmodel scal_axpy_kernel_model = {
 	.symbol = "scal_axpy_kernel"
 };
 
-static starpu_codelet scal_axpy_kernel_cl = {
+static struct starpu_codelet scal_axpy_kernel_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = scal_axpy_kernel_cpu,
 #ifdef STARPU_USE_CUDA
@@ -533,7 +533,7 @@ static struct starpu_perfmodel axpy_kernel_model = {
 	.symbol = "axpy_kernel"
 };
 
-static starpu_codelet axpy_kernel_cl = {
+static struct starpu_codelet axpy_kernel_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = axpy_kernel_cpu,
 #ifdef STARPU_USE_CUDA
@@ -593,7 +593,7 @@ static struct starpu_perfmodel copy_handle_model = {
 	.symbol = "copy_handle"
 };
 
-static starpu_codelet copy_handle_cl = {
+static struct starpu_codelet copy_handle_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = copy_handle_cpu,
 #ifdef STARPU_USE_CUDA

examples/cholesky/cholesky_grain_tag.c

@@ -36,7 +36,7 @@ static struct starpu_task *create_task(starpu_tag_t id)
  *	Create the codelets
  */
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u11,
@@ -70,7 +70,7 @@ static struct starpu_task * create_task_11(starpu_data_handle_t dataA, unsigned
 	return task;
 }
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u21,
@@ -108,7 +108,7 @@ static void create_task_21(starpu_data_handle_t dataA, unsigned k, unsigned j, u
 	starpu_task_submit(task);
 }
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u22,

examples/cholesky/cholesky_implicit.c

@@ -22,7 +22,7 @@
  *	Create the codelets
  */
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.type = STARPU_SEQ,
@@ -34,7 +34,7 @@ static starpu_codelet cl11 =
 	.model = &chol_model_11
 };
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.type = STARPU_SEQ,
@@ -46,7 +46,7 @@ static starpu_codelet cl21 =
 	.model = &chol_model_21
 };
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.type = STARPU_SEQ,

examples/cholesky/cholesky_tag.c

@@ -36,7 +36,7 @@ static struct starpu_task *create_task(starpu_tag_t id)
  *	Create the codelets
  */
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u11,
@@ -71,7 +71,7 @@ static struct starpu_task * create_task_11(starpu_data_handle_t dataA, unsigned
 	return task;
 }
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u21,
@@ -114,7 +114,7 @@ static void create_task_21(starpu_data_handle_t dataA, unsigned k, unsigned j)
 
 }
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u22,

examples/cholesky/cholesky_tile_tag.c

@@ -39,7 +39,7 @@ static struct starpu_task *create_task(starpu_tag_t id)
  *	Create the codelets
  */
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA|STARPU_GORDON,
 	.cpu_func = chol_cpu_codelet_update_u11,
@@ -80,7 +80,7 @@ static struct starpu_task * create_task_11(unsigned k, unsigned nblocks)
 	return task;
 }
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA|STARPU_GORDON,
 	.cpu_func = chol_cpu_codelet_update_u21,
@@ -125,7 +125,7 @@ static void create_task_21(unsigned k, unsigned j)
 	starpu_task_submit(task);
 }
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA|STARPU_GORDON,
 	.cpu_func = chol_cpu_codelet_update_u22,

examples/filters/fblock.c

@@ -83,7 +83,7 @@ int main(int argc, char **argv)
         }
 
 	starpu_data_handle_t handle;
-	starpu_codelet cl =
+	struct starpu_codelet cl =
 	{
                 .where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
                 .cpu_func = cpu_func,

examples/filters/fmatrix.c

@@ -56,7 +56,7 @@ int main(int argc, char **argv)
         FPRINTF(stderr,"\n");
 
         starpu_data_handle_t handle;
-        starpu_codelet cl = {
+        struct starpu_codelet cl = {
                 .where = STARPU_CPU,
                 .cpu_func = cpu_func,
                 .nbuffers = 1

examples/filters/fvector.c

@@ -42,7 +42,7 @@ int main(int argc, char **argv)
         starpu_data_handle_t handle;
         int factor=1;
 
-        starpu_codelet cl = {
+        struct starpu_codelet cl = {
                 .where = STARPU_CPU,
                 .cpu_func = cpu_func,
                 .nbuffers = 1

examples/heat/dw_factolu.c

@@ -32,7 +32,7 @@ static unsigned finished = 0;
 
 static unsigned no_prio = 0;
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u11,
@@ -43,7 +43,7 @@ static starpu_codelet cl11 =
 	.model = &model_11
 };
 
-static starpu_codelet cl12 =
+static struct starpu_codelet cl12 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u12,
@@ -54,7 +54,7 @@ static starpu_codelet cl12 =
 	.model = &model_12
 }; 
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u21,
@@ -65,7 +65,7 @@ static starpu_codelet cl21 =
 	.model = &model_21
 }; 
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u22,

examples/heat/dw_factolu_grain.c

@@ -42,7 +42,7 @@ static struct starpu_task *create_task(starpu_tag_t id)
 	return task;
 }
 
-static starpu_codelet cl11 = {
+static struct starpu_codelet cl11 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u11,
 #ifdef STARPU_USE_CUDA
@@ -75,7 +75,7 @@ static struct starpu_task *create_task_11(starpu_data_handle_t dataA, unsigned k
 	return task;
 }
 
-static starpu_codelet cl12 = {
+static struct starpu_codelet cl12 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u12,
 #ifdef STARPU_USE_CUDA
@@ -114,7 +114,7 @@ static void create_task_12(starpu_data_handle_t dataA, unsigned k, unsigned i, u
 	starpu_task_submit(task);
 }
 
-static starpu_codelet cl21 = {
+static struct starpu_codelet cl21 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u21,
 #ifdef STARPU_USE_CUDA
@@ -151,7 +151,7 @@ static void create_task_21(starpu_data_handle_t dataA, unsigned k, unsigned j, u
 	starpu_task_submit(task);
 }
 
-static starpu_codelet cl22 = {
+static struct starpu_codelet cl22 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u22,
 #ifdef STARPU_USE_CUDA

examples/heat/dw_factolu_tag.c

@@ -44,7 +44,7 @@ static struct starpu_task *create_task(starpu_tag_t id)
 	return task;
 }
 
-static starpu_codelet cl11 = {
+static struct starpu_codelet cl11 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u11,
 #ifdef STARPU_USE_CUDA
@@ -78,7 +78,7 @@ static struct starpu_task *create_task_11(starpu_data_handle_t dataA, unsigned k
 	return task;
 }
 
-static starpu_codelet cl12 = {
+static struct starpu_codelet cl12 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u12,
 #ifdef STARPU_USE_CUDA
@@ -117,7 +117,7 @@ static void create_task_12(starpu_data_handle_t dataA, unsigned k, unsigned i)
 	starpu_task_submit(task);
 }
 
-static starpu_codelet cl21 = {
+static struct starpu_codelet cl21 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u21,
 #ifdef STARPU_USE_CUDA
@@ -154,7 +154,7 @@ static void create_task_21(starpu_data_handle_t dataA, unsigned k, unsigned j)
 	starpu_task_submit(task);
 }
 
-static starpu_codelet cl22 = {
+static struct starpu_codelet cl22 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dw_cpu_codelet_update_u22,
 #ifdef STARPU_USE_CUDA

examples/heat/dw_sparse_cg.c

@@ -24,7 +24,7 @@
 
 static struct starpu_task *create_task(starpu_tag_t id)
 {
-	starpu_codelet *cl = calloc(1,sizeof(starpu_codelet));
+	struct starpu_codelet *cl = calloc(1,sizeof(struct starpu_codelet));
 
 	struct starpu_task *task = starpu_task_create();
 		task->cl = cl;

examples/incrementer/incrementer.c

@@ -61,7 +61,7 @@ int main(int argc, char **argv)
         starpu_opencl_load_opencl_from_file("examples/incrementer/incrementer_kernels_opencl_kernel.cl", &opencl_program, NULL);
 #endif
 
-	starpu_codelet cl =
+	struct starpu_codelet cl =
 	{
 		.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
 		.cpu_func = cpu_codelet,

examples/lu/xlu_kernels.c

@@ -103,7 +103,7 @@ static struct starpu_perfmodel STARPU_LU(model_22) = {
 #endif
 };
 
-starpu_codelet cl22 = {
+struct starpu_codelet cl22 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_LU(cpu_u22),
 #ifdef STARPU_USE_CUDA
@@ -186,7 +186,7 @@ static struct starpu_perfmodel STARPU_LU(model_12) = {
 #endif
 };
 
-starpu_codelet cl12 = {
+struct starpu_codelet cl12 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_LU(cpu_u12),
 #ifdef STARPU_USE_CUDA
@@ -266,7 +266,7 @@ static struct starpu_perfmodel STARPU_LU(model_21) = {
 #endif
 };
 
-starpu_codelet cl21 = {
+struct starpu_codelet cl21 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_LU(cpu_u21),
 #ifdef STARPU_USE_CUDA
@@ -361,7 +361,7 @@ static struct starpu_perfmodel STARPU_LU(model_11) = {
 #endif
 };
 
-starpu_codelet cl11 = {
+struct starpu_codelet cl11 = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_LU(cpu_u11),
 #ifdef STARPU_USE_CUDA
@@ -497,7 +497,7 @@ static struct starpu_perfmodel STARPU_LU(model_11_pivot) = {
 #endif
 };
 
-starpu_codelet cl11_pivot = {
+struct starpu_codelet cl11_pivot = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_LU(cpu_u11_pivot),
 #ifdef STARPU_USE_CUDA
@@ -582,7 +582,7 @@ static struct starpu_perfmodel STARPU_LU(model_pivot) = {
 #endif
 };
 
-starpu_codelet cl_pivot = {
+struct starpu_codelet cl_pivot = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_LU(cpu_pivot),
 #ifdef STARPU_USE_CUDA

examples/lu/xlu_kernels.h

@@ -36,11 +36,11 @@ void STARPU_LU(cublas_u21)(void *descr[], void *_args);
 void STARPU_LU(cublas_u22)(void *descr[], void *_args);
 #endif
 
-extern starpu_codelet cl11;
-extern starpu_codelet cl11_pivot;
-extern starpu_codelet cl12;
-extern starpu_codelet cl21;
-extern starpu_codelet cl22;
-extern starpu_codelet cl_pivot;
+extern struct starpu_codelet cl11;
+extern struct starpu_codelet cl11_pivot;
+extern struct starpu_codelet cl12;
+extern struct starpu_codelet cl21;
+extern struct starpu_codelet cl22;
+extern struct starpu_codelet cl_pivot;
 
 #endif /* __XLU_KERNELS_H__ */

examples/mandelbrot/mandelbrot.c

@@ -371,7 +371,7 @@ static void compute_block_spmd(void *descr[], void *cl_arg)
 
 
 
-static starpu_codelet spmd_mandelbrot_cl = {
+static struct starpu_codelet spmd_mandelbrot_cl = {
 	.where = STARPU_CPU|STARPU_OPENCL,
 	.type = STARPU_SPMD,
 	.max_parallelism = INT_MAX,
@@ -382,7 +382,7 @@ static starpu_codelet spmd_mandelbrot_cl = {
 	.nbuffers = 1
 };
 
-static starpu_codelet mandelbrot_cl = {
+static struct starpu_codelet mandelbrot_cl = {
 	.where = STARPU_CPU|STARPU_OPENCL,
 	.type = STARPU_SEQ,
 	.cpu_func = compute_block,

examples/matvecmult/matvecmult.c

@@ -115,7 +115,7 @@ int compareL2fe(const float* reference, const float* data, const unsigned int le
 
 int main(int argc, char **argv)
 {
-	starpu_codelet cl = {};
+	struct starpu_codelet cl = {};
 
 	struct starpu_conf conf = {
 		.ncpus = 0,

examples/mult/xgemm.c

@@ -181,7 +181,7 @@ static struct starpu_perfmodel starpu_gemm_model = {
 	.symbol = STARPU_GEMM_STR(gemm)
 };
 
-static starpu_codelet cl = {
+static struct starpu_codelet cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.type = STARPU_SEQ, /* changed to STARPU_SPMD if -spmd is passed */
 	.max_parallelism = INT_MAX,

examples/openmp/vector_scal.c

@@ -47,7 +47,7 @@ static struct starpu_perfmodel vector_scal_model = {
 	.symbol = "vector_scale_parallel"
 };
 
-static starpu_codelet cl = {
+static struct starpu_codelet cl = {
 	.where = STARPU_CPU,
 	.type = STARPU_FORKJOIN,
 	.max_parallelism = INT_MAX,

examples/opt/pi/pi.c

@@ -113,7 +113,7 @@ int main(int argc, char **argv)
 		.symbol = "monte_carlo_pi"
 	};
 
-	struct starpu_codelet_t cl = {
+	struct starpu_codelet cl = {
 		.where = STARPU_CPU|STARPU_CUDA,
 		.cpu_func = cpu_kernel,
 #ifdef STARPU_USE_CUDA

examples/opt/pi/pi_redux.c

@@ -183,7 +183,7 @@ static void pi_func_cuda(void *descr[], void *cl_arg __attribute__ ((unused)))
 }
 #endif
 
-static struct starpu_codelet_t pi_cl = {
+static struct starpu_codelet pi_cl = {
 	.where =
 #ifdef STARPU_HAVE_CURAND
 		STARPU_CUDA|
@@ -216,7 +216,7 @@ static void init_cuda_func(void *descr[], void *cl_arg)
 }
 #endif
 
-static struct starpu_codelet_t init_codelet = {
+static struct starpu_codelet init_codelet = {
 	.where =
 #ifdef STARPU_HAVE_CURAND
 		STARPU_CUDA|
@@ -255,7 +255,7 @@ static void redux_cpu_func(void *descr[], void *cl_arg)
 	*a = *a + *b;
 };
 
-static struct starpu_codelet_t redux_codelet = {
+static struct starpu_codelet redux_codelet = {
 	.where =
 #ifdef STARPU_HAVE_CURAND
 		STARPU_CUDA|

examples/ppm_downscaler/yuv_downscaler.c

@@ -82,7 +82,7 @@ static void ds_kernel_cpu(void *descr[], __attribute__((unused)) void *arg)
 	}
 }
 
-static struct starpu_codelet_t ds_codelet = {
+static struct starpu_codelet ds_codelet = {
 	.where = STARPU_CPU,
 	.cpu_func = ds_kernel_cpu,
 	.nbuffers = 2, /* input -> output */

examples/profiling/profiling.c

@@ -44,7 +44,7 @@ int main(int argc, char **argv)
 	 * worker. */
 	usleep(500000);
 
-	starpu_codelet cl =
+	struct starpu_codelet cl =
 	{
 		.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
 		.cpu_func = sleep_codelet,

examples/reductions/dot_product.c

@@ -56,7 +56,7 @@ void init_cuda_func(void *descr[], void *cl_arg)
 }
 #endif
 
-static struct starpu_codelet_t init_codelet = {
+static struct starpu_codelet init_codelet = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = init_cpu_func,
 #ifdef STARPU_USE_CUDA
@@ -81,7 +81,7 @@ void redux_cpu_func(void *descr[], void *cl_arg)
 extern void redux_cuda_func(void *descr[], void *_args);
 #endif
 
-static struct starpu_codelet_t redux_codelet = {
+static struct starpu_codelet redux_codelet = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = redux_cpu_func,
 #ifdef STARPU_USE_CUDA
@@ -142,7 +142,7 @@ void dot_cuda_func(void *descr[], void *cl_arg)
 }
 #endif
 
-static struct starpu_codelet_t dot_codelet = {
+static struct starpu_codelet dot_codelet = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = dot_cpu_func,
 #ifdef STARPU_USE_CUDA

examples/reductions/minmax_reduction.c

@@ -50,7 +50,7 @@ static void minmax_neutral_cpu_func(void *descr[], void *cl_arg)
 	array[1] = TYPE_MIN;
 }
 
-static struct starpu_codelet_t minmax_init_codelet = {
+static struct starpu_codelet minmax_init_codelet = {
 	.where = STARPU_CPU,
 	.cpu_func = minmax_neutral_cpu_func,
 	.nbuffers = 1
@@ -76,7 +76,7 @@ void minmax_redux_cpu_func(void *descr[], void *cl_arg)
 	array_dst[1] = STARPU_MAX(max_dst, max_src);
 }
 
-static struct starpu_codelet_t minmax_redux_codelet = {
+static struct starpu_codelet minmax_redux_codelet = {
 	.where = STARPU_CPU,
 	.cpu_func = minmax_redux_cpu_func,
 	.nbuffers = 2
@@ -110,7 +110,7 @@ void minmax_cpu_func(void *descr[], void *cl_arg)
 	minmax[1] = local_max;
 }
 
-static struct starpu_codelet_t minmax_codelet = {
+static struct starpu_codelet minmax_codelet = {
 	.where = STARPU_CPU,
 	.cpu_func = minmax_cpu_func,
 	.nbuffers = 2

examples/scheduler/dummy_sched.c

@@ -105,7 +105,7 @@ static void dummy_func(void *descr[] __attribute__ ((unused)), void *arg __attri
 {
 }
 
-static starpu_codelet dummy_codelet = 
+static struct starpu_codelet dummy_codelet = 
 {
 	.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
 	.cpu_func = dummy_func,

examples/spmv/dw_block_spmv.c

@@ -140,7 +140,7 @@ void call_filters(void)
 #define NSPMV	32
 unsigned totaltasks;
 
-starpu_codelet cl = {
+struct starpu_codelet cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func =  cpu_block_spmv,
 #ifdef STARPU_USE_CUDA

examples/spmv/spmv.c

@@ -87,7 +87,7 @@ static struct starpu_data_filter vector_f = {
 	.nchildren = -1,
 };
 
-static starpu_codelet spmv_cl = {
+static struct starpu_codelet spmv_cl = {
 	.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
 	.cpu_func = spmv_kernel_cpu,
 #ifdef STARPU_USE_CUDA

examples/starpufft/starpufftx1d.c

@@ -314,7 +314,7 @@ static struct starpu_perfmodel STARPUFFT(twist3_1d_model) = {
 };
 
 /* codelet pointers for the 5 kinds of tasks */
-static starpu_codelet STARPUFFT(twist1_1d_codelet) = {
+static struct starpu_codelet STARPUFFT(twist1_1d_codelet) = {
 	.where =
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -328,7 +328,7 @@ static starpu_codelet STARPUFFT(twist1_1d_codelet) = {
 	.nbuffers = 2
 };
 
-static starpu_codelet STARPUFFT(fft1_1d_codelet) = {
+static struct starpu_codelet STARPUFFT(fft1_1d_codelet) = {
 	.where =
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -347,14 +347,14 @@ static starpu_codelet STARPUFFT(fft1_1d_codelet) = {
 	.nbuffers = 3
 };
 
-static starpu_codelet STARPUFFT(twist2_1d_codelet) = {
+static struct starpu_codelet STARPUFFT(twist2_1d_codelet) = {
 	.where = STARPU_CPU,
 	.cpu_func = STARPUFFT(twist2_1d_kernel_cpu),
 	.model = &STARPUFFT(twist2_1d_model),
 	.nbuffers = 1
 };
 
-static starpu_codelet STARPUFFT(fft2_1d_codelet) = {
+static struct starpu_codelet STARPUFFT(fft2_1d_codelet) = {
 	.where =
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -373,7 +373,7 @@ static starpu_codelet STARPUFFT(fft2_1d_codelet) = {
 	.nbuffers = 2
 };
 
-static starpu_codelet STARPUFFT(twist3_1d_codelet) = {
+static struct starpu_codelet STARPUFFT(twist3_1d_codelet) = {
 	.where = STARPU_CPU,
 	.cpu_func = STARPUFFT(twist3_1d_kernel_cpu),
 	.model = &STARPUFFT(twist3_1d_model),

examples/starpufft/starpufftx2d.c

@@ -297,7 +297,7 @@ struct starpu_perfmodel STARPUFFT(twist3_2d_model) = {
 	.symbol = TYPE"twist3_2d"
 };
 
-static starpu_codelet STARPUFFT(twist1_2d_codelet) = {
+static struct starpu_codelet STARPUFFT(twist1_2d_codelet) = {
 	.where =
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -311,7 +311,7 @@ static starpu_codelet STARPUFFT(twist1_2d_codelet) = {
 	.nbuffers = 2
 };
 
-static starpu_codelet STARPUFFT(fft1_2d_codelet) = {
+static struct starpu_codelet STARPUFFT(fft1_2d_codelet) = {
 	.where =
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -330,14 +330,14 @@ static starpu_codelet STARPUFFT(fft1_2d_codelet) = {
 	.nbuffers = 4
 };
 
-static starpu_codelet STARPUFFT(twist2_2d_codelet) = {
+static struct starpu_codelet STARPUFFT(twist2_2d_codelet) = {
 	.where = STARPU_CPU,
 	.cpu_func = STARPUFFT(twist2_2d_kernel_cpu),
 	.model = &STARPUFFT(twist2_2d_model),
 	.nbuffers = 1
 };
 
-static starpu_codelet STARPUFFT(fft2_2d_codelet) = {
+static struct starpu_codelet STARPUFFT(fft2_2d_codelet) = {
 	.where =
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -356,7 +356,7 @@ static starpu_codelet STARPUFFT(fft2_2d_codelet) = {
 	.nbuffers = 2
 };
 
-static starpu_codelet STARPUFFT(twist3_2d_codelet) = {
+static struct starpu_codelet STARPUFFT(twist3_2d_codelet) = {
 	.where = STARPU_CPU,
 	.cpu_func = STARPUFFT(twist3_2d_kernel_cpu),
 	.model = &STARPUFFT(twist3_2d_model),

examples/stencil/stencil-kernels.c

@@ -444,7 +444,7 @@ static struct starpu_perfmodel cl_update_model = {
 	.symbol = "cl_update" 
 };
 
-starpu_codelet cl_update = {
+struct starpu_codelet cl_update = {
 	.where = 0 |
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -644,7 +644,7 @@ static struct starpu_perfmodel save_cl_top_model = {
 	.symbol = "save_cl_top" 
 };
 
-starpu_codelet save_cl_bottom = {
+struct starpu_codelet save_cl_bottom = {
 	.where = 0 |
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|
@@ -664,7 +664,7 @@ starpu_codelet save_cl_bottom = {
 	.nbuffers = 4
 };
 
-starpu_codelet save_cl_top = {
+struct starpu_codelet save_cl_top = {
 	.where = 0|
 #ifdef STARPU_USE_CUDA
 		STARPU_CUDA|

examples/stencil/stencil-tasks.c

@@ -218,7 +218,7 @@ void create_task_update(unsigned iter, unsigned z, unsigned local_rank)
 
 /* Dummy empty codelet taking one buffer */
 static void null_func(void *descr[] __attribute__((unused)), void *arg __attribute__((unused))) { }
-static starpu_codelet null = {
+static struct starpu_codelet null = {
 	.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
 	.cpu_func = null_func,
 	.cuda_func = null_func,

examples/stencil/stencil.h

@@ -120,9 +120,9 @@ extern int starpu_mpi_initialize(void);
 extern int starpu_mpi_shutdown(void);
 
 /* kernels */
-extern starpu_codelet cl_update;
-extern starpu_codelet save_cl_bottom;
-extern starpu_codelet save_cl_top;
+extern struct starpu_codelet cl_update;
+extern struct starpu_codelet save_cl_bottom;
+extern struct starpu_codelet save_cl_top;
 
 extern unsigned update_per_worker[STARPU_NMAXWORKERS];
 extern unsigned top_per_worker[STARPU_NMAXWORKERS];

examples/tag_example/tag_example.c

@@ -30,7 +30,7 @@
 #define FPRINTF(ofile, fmt, args ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ##args); }} while(0)
 #define TAG(i, j, iter)	((starpu_tag_t) ( ((uint64_t)(iter)<<48) |  ((uint64_t)(j)<<24) | (i)) )
 
-starpu_codelet cl = {};
+struct starpu_codelet cl = {};
 
 #define Ni	64
 #define Nj	32

examples/tag_example/tag_example2.c

@@ -30,7 +30,7 @@
 #define FPRINTF(ofile, fmt, args ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ##args); }} while(0)
 #define TAG(i, iter)	((starpu_tag_t)  (((uint64_t)iter)<<32 | (i)) )
 
-starpu_codelet cl = {};
+struct starpu_codelet cl = {};
 
 #define Ni	64
 #define Nk	256

examples/tag_example/tag_example3.c

@@ -30,7 +30,7 @@
 #define FPRINTF(ofile, fmt, args ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ##args); }} while(0)
 #define TAG(i, iter)	((starpu_tag_t)  (((uint64_t)iter)<<32 | (i)) )
 
-starpu_codelet cl = {};
+struct starpu_codelet cl = {};
 
 #define Ni	64
 #define Nk	256

examples/tag_example/tag_restartable.c

@@ -35,7 +35,7 @@
 #define FPRINTF(ofile, fmt, args ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ##args); }} while(0)
 #define TAG(i, iter)	((starpu_tag_t)  (((uint64_t)((iter)%Nrolls))<<32 | (i)) )
 
-starpu_codelet cl = {};
+struct starpu_codelet cl = {};
 
 #define Ni	64
 #define Nk	256

examples/top/hello_world_top.c

@@ -95,7 +95,7 @@ void callback_number_addition_changed(starpu_top_param* param)
 	starpu_top_debug_log(message);
 }
 
-starpu_codelet cl =
+struct starpu_codelet cl =
 {
 	/* this codelet may only be executed on a CPU, and its cpu
  	 * implementation is function "cpu_func" */

gcc-plugin/src/starpu.c

@@ -77,7 +77,7 @@ static const char task_implementation_wrapper_attribute_name[] =
   ".task_implementation_wrapper";
 
 /* Names of data structures defined in <starpu.h>.  */
-static const char codelet_struct_name[] = "starpu_codelet";
+static const char codelet_struct_name[] = "struct starpu_codelet";
 static const char task_struct_name[] = "starpu_task";
 
 /* Cached function declarations.  */
@@ -712,7 +712,7 @@ handle_task_attribute (tree *node, tree name, tree args,
 		   NULL_TREE,
 		   NULL_TREE);
 
-      /* Push a declaration for the corresponding `starpu_codelet' object and
+      /* Push a declaration for the corresponding `struct starpu_codelet' object and
 	 add it as an attribute of FN.  */
       tree cl = build_codelet_declaration (fn);
       DECL_ATTRIBUTES (fn) =
@@ -860,7 +860,7 @@ handle_heap_allocated_attribute (tree *node, tree name, tree args,
 }
 
 
-/* Return the declaration of the `starpu_codelet' variable associated with
+/* Return the declaration of the `struct starpu_codelet' variable associated with
    TASK_DECL.  */
 
 static tree
@@ -1258,7 +1258,7 @@ define_codelet_wrappers (tree task)
   for_each (define, task_implementation_list (task));
 }
 
-/* Return a NODE_IDENTIFIER for the variable holding the `starpu_codelet'
+/* Return a NODE_IDENTIFIER for the variable holding the `struct starpu_codelet'
    structure associated with TASK_DECL.  */
 
 static tree
@@ -1289,7 +1289,7 @@ codelet_type (void)
 
   if (type_decl == NULL_TREE)
     {
-      /* Lookup the `starpu_codelet' struct type.  This should succeed since
+      /* Lookup the `struct starpu_codelet' struct type.  This should succeed since
 	 we push <starpu.h> early on.  */
 
       type_decl = lookup_name (get_identifier (codelet_struct_name));
@@ -1299,7 +1299,7 @@ codelet_type (void)
   return TREE_TYPE (type_decl);
 }
 
-/* Return a VAR_DECL that declares a `starpu_codelet' structure for
+/* Return a VAR_DECL that declares a `struct starpu_codelet' structure for
    TASK_DECL.  */
 
 static tree
@@ -1324,7 +1324,7 @@ build_codelet_declaration (tree task_decl)
   return cl_decl;
 }
 
-/* Return a `starpu_codelet' initializer for TASK_DECL.  */
+/* Return a `struct starpu_codelet' initializer for TASK_DECL.  */
 
 static tree
 build_codelet_initializer (tree task_decl)
@@ -1441,14 +1441,14 @@ build_codelet_initializer (tree task_decl)
   return build_constructor_from_unsorted_list (codelet_type (), inits);
 }
 
-/* Return the VAR_DECL that defines a `starpu_codelet' structure for
+/* Return the VAR_DECL that defines a `struct starpu_codelet' structure for
    TASK_DECL.  The VAR_DECL is assumed to already exists, so it must not be
    pushed again.  */
 
 static tree
 declare_codelet (tree task_decl)
 {
-  /* Retrieve the declaration of the `starpu_codelet' object.  */
+  /* Retrieve the declaration of the `struct starpu_codelet' object.  */
   tree cl_decl;
   cl_decl = lookup_name (build_codelet_identifier (task_decl));
   gcc_assert (cl_decl != NULL_TREE && TREE_CODE (cl_decl) == VAR_DECL);
@@ -1625,7 +1625,7 @@ lower_starpu (void)
 
   if (task_p (fndecl))
     {
-      /* Generate a `starpu_codelet' structure and a wrapper function for
+      /* Generate a `struct starpu_codelet' structure and a wrapper function for
 	 each implementation of TASK_DECL.  This cannot be done earlier
 	 because we need to have a complete list of task implementations.  */
 

gcc-plugin/tests/mocks.h

@@ -54,7 +54,7 @@ struct insert_task_argument
 const struct insert_task_argument *expected_insert_task_arguments;
 
 int
-starpu_insert_task (starpu_codelet *cl, ...)
+starpu_insert_task (struct starpu_codelet *cl, ...)
 {
   assert (cl->where == (STARPU_CPU | STARPU_OPENCL));
 

include/starpu_data.h

@@ -105,11 +105,11 @@ void starpu_data_set_default_sequential_consistency_flag(unsigned flag);
 /* Query the status of the handle on the specified memory node. */
 void starpu_data_query_status(starpu_data_handle_t handle, int memory_node, int *is_allocated, int *is_valid, int *is_requested);
 
-struct starpu_codelet_t;
+struct starpu_codelet;
 
 void starpu_data_set_reduction_methods(starpu_data_handle_t handle,
-					struct starpu_codelet_t *redux_cl,
-					struct starpu_codelet_t *init_cl);
+					struct starpu_codelet *redux_cl,
+					struct starpu_codelet *init_cl);
 
 int starpu_data_set_rank(starpu_data_handle_t handle, int rank);
 int starpu_data_get_rank(starpu_data_handle_t handle);

include/starpu_data_interfaces.h

@@ -332,13 +332,13 @@ struct starpu_multiformat_data_interface_ops {
 	size_t cpu_elemsize;
 #ifdef STARPU_USE_OPENCL
 	size_t opencl_elemsize;
-	struct starpu_codelet_t *cpu_to_opencl_cl;
-	struct starpu_codelet_t *opencl_to_cpu_cl;
+	struct starpu_codelet *cpu_to_opencl_cl;
+	struct starpu_codelet *opencl_to_cpu_cl;
 #endif
 #ifdef STARPU_USE_CUDA
 	size_t cuda_elemsize;
-	struct starpu_codelet_t *cpu_to_cuda_cl;
-	struct starpu_codelet_t *cuda_to_cpu_cl;
+	struct starpu_codelet *cpu_to_cuda_cl;
+	struct starpu_codelet *cuda_to_cpu_cl;
 #endif
 };
 

include/starpu_deprecated_api.h

@@ -20,6 +20,7 @@
 
 #warning deprecated types. Please update your code to use the latest API.
 
+#define starpu_codelet			struct starpu_codelet
 #define starpu_data_handle		starpu_data_handle_t
 #define starpu_block_interface_t	struct starpu_block_interface
 #define starpu_matrix_interface_t	struct starpu_matrix_interface

include/starpu_task.h

@@ -74,7 +74,7 @@ typedef uint8_t starpu_gordon_func_t; /* Cell SPU */
  * A codelet describes the various function 
  * that may be called from a worker
  */
-typedef struct starpu_codelet_t {
+struct starpu_codelet {
 	/* where can it be performed ? */
 	uint32_t where;
 	unsigned type;
@@ -107,10 +107,10 @@ typedef struct starpu_codelet_t {
 	 * not be accessed directly (use the starpu_display_codelet_stats
 	 * function instead for instance). */
 	unsigned long per_worker_stats[STARPU_NMAXWORKERS];
-} starpu_codelet;
+};
 
 struct starpu_task {
-	struct starpu_codelet_t *cl;
+	struct starpu_codelet *cl;
 
 	/* arguments managed by the DSM */
 	struct starpu_buffer_descr buffers[STARPU_NMAXBUFS];
@@ -284,7 +284,7 @@ int starpu_task_wait_for_all(void);
 /* This function waits until there is no more ready task. */
 int starpu_task_wait_for_no_ready(void);
 
-void starpu_display_codelet_stats(struct starpu_codelet_t *cl);
+void starpu_display_codelet_stats(struct starpu_codelet *cl);
 
 /* Return the task currently executed by the worker, or NULL if this is called
  * either from a thread that is not a task or simply because there is no task

include/starpu_util.h

@@ -241,7 +241,7 @@ int starpu_data_cpy(starpu_data_handle_t dst_handle, starpu_data_handle_t src_ha
 #define STARPU_EXECUTE_ON_DATA	(1<<10)	/* Used by MPI to define which task is going to execute the codelet */
 
 /* Wrapper to create a task. */
-int starpu_insert_task(starpu_codelet *cl, ...);
+int starpu_insert_task(struct starpu_codelet *cl, ...);
 
 /* Retrieve the arguments of type STARPU_VALUE associated to a task
  * automatically created using starpu_insert_task. */

mpi/examples/cholesky/mpi_cholesky.c

@@ -24,7 +24,7 @@
  *	Create the codelets
  */
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u11,
@@ -35,7 +35,7 @@ static starpu_codelet cl11 =
 	.model = &chol_model_11
 };
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u21,
@@ -46,7 +46,7 @@ static starpu_codelet cl21 =
 	.model = &chol_model_21
 };
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u22,

mpi/examples/cholesky/mpi_cholesky_distributed.c

@@ -24,7 +24,7 @@
  *	Create the codelets
  */
 
-static starpu_codelet cl11 =
+static struct starpu_codelet cl11 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u11,
@@ -35,7 +35,7 @@ static starpu_codelet cl11 =
 	.model = &chol_model_11
 };
 
-static starpu_codelet cl21 =
+static struct starpu_codelet cl21 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u21,
@@ -46,7 +46,7 @@ static starpu_codelet cl21 =
 	.model = &chol_model_21
 };
 
-static starpu_codelet cl22 =
+static struct starpu_codelet cl22 =
 {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = chol_cpu_codelet_update_u22,

mpi/examples/mpi_lu/pxlu_kernels.c

@@ -107,7 +107,7 @@ static struct starpu_perfmodel STARPU_PLU(model_22) = {
 #endif
 };
 
-starpu_codelet STARPU_PLU(cl22) = {
+struct starpu_codelet STARPU_PLU(cl22) = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_PLU(cpu_u22),
 #ifdef STARPU_USE_CUDA
@@ -214,7 +214,7 @@ static struct starpu_perfmodel STARPU_PLU(model_12) = {
 #endif
 };
 
-starpu_codelet STARPU_PLU(cl12) = {
+struct starpu_codelet STARPU_PLU(cl12) = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_PLU(cpu_u12),
 #ifdef STARPU_USE_CUDA
@@ -322,7 +322,7 @@ static struct starpu_perfmodel STARPU_PLU(model_21) = {
 #endif
 };
 
-starpu_codelet STARPU_PLU(cl21) = {
+struct starpu_codelet STARPU_PLU(cl21) = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_PLU(cpu_u21),
 #ifdef STARPU_USE_CUDA
@@ -427,7 +427,7 @@ static struct starpu_perfmodel STARPU_PLU(model_11) = {
 #endif
 };
 
-starpu_codelet STARPU_PLU(cl11) = {
+struct starpu_codelet STARPU_PLU(cl11) = {
 	.where = STARPU_CPU|STARPU_CUDA,
 	.cpu_func = STARPU_PLU(cpu_u11),
 #ifdef STARPU_USE_CUDA

mpi/examples/mpi_lu/pxlu_kernels.h

@@ -24,9 +24,9 @@
 #define xstr(s)        str(s)
 #define STARPU_PLU_STR(name)  xstr(STARPU_PLU(name))
 
-starpu_codelet STARPU_PLU(cl11);
-starpu_codelet STARPU_PLU(cl12);
-starpu_codelet STARPU_PLU(cl21);
-starpu_codelet STARPU_PLU(cl22);
+struct starpu_codelet STARPU_PLU(cl11);
+struct starpu_codelet STARPU_PLU(cl12);
+struct starpu_codelet STARPU_PLU(cl21);
+struct starpu_codelet STARPU_PLU(cl22);
 
 #endif // __PXLU_KERNELS_H__

mpi/examples/scatter_gather/mpi_scatter_gather.c

@@ -45,7 +45,7 @@ void cpu_codelet(void *descr[], void *_args)
 	}
 }
 
-static starpu_codelet cl =
+static struct starpu_codelet cl =
 {
 	.where = STARPU_CPU,
 	.cpu_func = cpu_codelet,

mpi/examples/stencil/stencil5.c

@@ -29,7 +29,7 @@ void stencil5_cpu(void *descr[], __attribute__ ((unused)) void *_args)
         *xy = (*xy + *xm1y + *xp1y + *xym1 + *xyp1) / 5;
 }
 
-starpu_codelet stencil5_cl = {
+struct starpu_codelet stencil5_cl = {
 	.where = STARPU_CPU,
 	.cpu_func = stencil5_cpu,
         .nbuffers = 5

mpi/starpu_mpi.h

@@ -40,7 +40,7 @@ int starpu_mpi_initialize(void);
 int starpu_mpi_initialize_extended(int *rank, int *world_size);
 int starpu_mpi_shutdown(void);
 
-int starpu_mpi_insert_task(MPI_Comm comm, starpu_codelet *codelet, ...);
+int starpu_mpi_insert_task(MPI_Comm comm, struct starpu_codelet *codelet, ...);
 void starpu_mpi_get_data_on_node(MPI_Comm comm, starpu_data_handle_t data_handle, int node);
 
 int starpu_mpi_scatter_detached(starpu_data_handle_t *data_handles, int count, int root, MPI_Comm comm);

mpi/starpu_mpi_insert_task.c

@@ -100,7 +100,7 @@ void _starpu_data_deallocate(starpu_data_handle_t data_handle)
 #endif
 }
 
-int starpu_mpi_insert_task(MPI_Comm comm, starpu_codelet *codelet, ...)
+int starpu_mpi_insert_task(MPI_Comm comm, struct starpu_codelet *codelet, ...)
 {
         int arg_type;
         va_list varg_list;

mpi/tests/insert_task.c

@@ -26,7 +26,7 @@ void func_cpu(void *descr[], __attribute__ ((unused)) void *_args)
         *x = (*x + *y) / 2;
 }
 
-starpu_codelet mycodelet = {
+struct starpu_codelet mycodelet = {
 	.where = STARPU_CPU,
 	.cpu_func = func_cpu,
         .nbuffers = 2

mpi/tests/insert_task_block.c

@@ -39,7 +39,7 @@ void func_cpu(void *descr[], __attribute__ ((unused)) void *_args)
         }
 }
 
-starpu_codelet mycodelet = {
+struct starpu_codelet mycodelet = {
 	.where = STARPU_CPU,
 	.cpu_func = func_cpu,
         .nbuffers = 1

mpi/tests/insert_task_cache.c

@@ -26,7 +26,7 @@ void func_cpu(void *descr[], __attribute__ ((unused)) void *_args)
         *x = (*x + *y) / 2;
 }
 
-starpu_codelet mycodelet = {
+struct starpu_codelet mycodelet = {
 	.where = STARPU_CPU,
 	.cpu_func = func_cpu,
         .nbuffers = 2

mpi/tests/insert_task_owner.c

@@ -26,7 +26,7 @@ void func_cpu(void *descr[], __attribute__ ((unused)) void *_args)
         *y = *y + 1;
 }
 
-starpu_codelet mycodelet = {
+struct starpu_codelet mycodelet = {
 	.where = STARPU_CPU,
 	.cpu_func = func_cpu,
         .nbuffers = 2

mpi/tests/insert_task_owner2.c

@@ -38,7 +38,7 @@ void func_cpu(void *descr[], __attribute__ ((unused)) void *_args)
         fprintf(stderr, "-------> CODELET VALUES: %d %d %d %d\n", *x0, *x1, *x2, *y);
 }
 
-starpu_codelet mycodelet = {
+struct starpu_codelet mycodelet = {
 	.where = STARPU_CPU,
 	.cpu_func = func_cpu,
         .nbuffers = 4

mpi/tests/insert_task_owner_data.c

@@ -27,7 +27,7 @@ void func_cpu(void *descr[], __attribute__ ((unused)) void *_args)
 	*x1 *= *x1;
 }
 
-starpu_codelet mycodelet = {
+struct starpu_codelet mycodelet = {
 	.where = STARPU_CPU,
 	.cpu_func = func_cpu,
         .nbuffers = 2

mpi/tests/ring.c

@@ -32,7 +32,7 @@ void increment_cpu(void *descr[], __attribute__ ((unused)) void *_args)
 	(*tokenptr)++;
 }
 
-static starpu_codelet increment_cl = {
+static struct starpu_codelet increment_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 #ifdef STARPU_USE_CUDA
 	.cuda_func = increment_cuda,

mpi/tests/ring_async.c

@@ -32,7 +32,7 @@ void increment_cpu(void *descr[], __attribute__ ((unused)) void *_args)
 	(*tokenptr)++;
 }
 
-static starpu_codelet increment_cl = {
+static struct starpu_codelet increment_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 #ifdef STARPU_USE_CUDA
 	.cuda_func = increment_cuda,

mpi/tests/ring_async_implicit.c

@@ -32,7 +32,7 @@ void increment_cpu(void *descr[], __attribute__ ((unused)) void *_args)
 	(*tokenptr)++;
 }
 
-static starpu_codelet increment_cl = {
+static struct starpu_codelet increment_cl = {
 	.where = STARPU_CPU|STARPU_CUDA,
 #ifdef STARPU_USE_CUDA
 	.cuda_func = increment_cuda,

socl/src/cl_enqueuecopybuffer.c

@@ -58,7 +58,7 @@ static void soclEnqueueCopyBuffer_cpu_task(void *descr[], void *args) {
    free(arg);
 }
 
-static starpu_codelet codelet_copybuffer = {
+static struct starpu_codelet codelet_copybuffer = {
    .where = STARPU_CPU | STARPU_OPENCL,
    .model = NULL,
    .cpu_func = &soclEnqueueCopyBuffer_cpu_task,

socl/src/cl_enqueuendrangekernel.c

@@ -117,7 +117,7 @@ cl_int command_ndrange_kernel_submit(command_ndrange_kernel cmd) {
 	task->cl_arg = cmd;
 	task->cl_arg_size = sizeof(cmd);
 
-	starpu_codelet * codelet = cmd->codelet;
+	struct starpu_codelet * codelet = cmd->codelet;
 
 	/* We need to detect which parameters are OpenCL's memory objects and
 	 * we retrieve their corresponding StarPU buffers */

socl/src/cl_enqueuereadbuffer.c

@@ -58,7 +58,7 @@ static void soclEnqueueReadBuffer_opencl_task(void *descr[], void *args) {
    free(args);
 }
 
-static starpu_codelet codelet_readbuffer = {
+static struct starpu_codelet codelet_readbuffer = {
    .where = STARPU_OPENCL,
    .model = NULL,
    .cpu_func = &soclEnqueueReadBuffer_cpu_task,

socl/src/cl_enqueuewritebuffer.c

@@ -60,7 +60,7 @@ static void soclEnqueueWriteBuffer_opencl_task(void *descr[], void *args) {
    free(args);
 }
 
-static starpu_codelet codelet_writebuffer = {
+static struct starpu_codelet codelet_writebuffer = {
    .where = STARPU_OPENCL,
    .model = NULL,
    .cpu_func = &soclEnqueueWriteBuffer_cpu_task,

socl/src/cl_getkernelworkgroupinfo.c

@@ -81,7 +81,7 @@ static void gkwgi_task2(void **UNUSED(desc), void *data) {
    }
 }
 
-static starpu_codelet gkwgi_codelet = {
+static struct starpu_codelet gkwgi_codelet = {
    .where = STARPU_OPENCL,
    .opencl_func = gkwgi_task2,
    .nbuffers = 0,

socl/src/command.c

@@ -106,8 +106,8 @@ command_ndrange_kernel command_ndrange_kernel_create (
 	nullOrDup(local_work_size, work_dim*sizeof(size_t));
 
    	/* Codelet */
-   	cmd->codelet = (starpu_codelet*)malloc(sizeof(starpu_codelet));
-	starpu_codelet * codelet = cmd->codelet;
+   	cmd->codelet = (struct starpu_codelet*)malloc(sizeof(struct starpu_codelet));
+	struct starpu_codelet * codelet = cmd->codelet;
 	codelet->where = STARPU_OPENCL;
 	codelet->power_model = NULL;
 	codelet->opencl_func = &soclEnqueueNDRangeKernel_task;

socl/src/command.h

@@ -75,7 +75,7 @@ typedef struct command_ndrange_kernel_t {
 	size_t *	 arg_sizes;
 	enum kernel_arg_type * arg_types;
 	void **		 args;
-	starpu_codelet * codelet;
+	struct starpu_codelet * codelet;
 	cl_uint		 num_buffers;
 	cl_mem *	 buffers;
 } * command_ndrange_kernel;

socl/src/task.c

@@ -126,7 +126,7 @@ static void cputask_task(__attribute__((unused)) void *descr[], void *args) {
   free(arg);
 }
 
-static starpu_codelet cputask_codelet = {
+static struct starpu_codelet cputask_codelet = {
    .where = STARPU_CPU,
    .model = NULL,
    .cpu_func = &cputask_task

src/core/dependencies/data_concurrency.c

@@ -218,7 +218,7 @@ static unsigned _submit_job_enforce_data_deps(starpu_job_t j, unsigned start_buf
    reading and another writing) */
 unsigned _starpu_submit_job_enforce_data_deps(starpu_job_t j)
 {
-	struct starpu_codelet_t *cl = j->task->cl;
+	struct starpu_codelet *cl = j->task->cl;
 
 	if ((cl == NULL) || (cl->nbuffers == 0))
 		return 0;

src/core/task.c

@@ -306,7 +306,7 @@ int starpu_task_submit(struct starpu_task *task)
 	return ret;
 }
 
-void starpu_display_codelet_stats(struct starpu_codelet_t *cl)
+void starpu_display_codelet_stats(struct starpu_codelet *cl)
 {
 	unsigned worker;
 	unsigned nworkers = starpu_worker_get_count();

src/core/workers.c

@@ -69,7 +69,7 @@ uint32_t _starpu_may_submit_opencl_task(void)
 	return (STARPU_OPENCL & config.worker_mask);
 }
 
-static int _starpu_may_use_nth_implementation(enum starpu_archtype arch, struct starpu_codelet_t *cl, unsigned nimpl)
+static int _starpu_may_use_nth_implementation(enum starpu_archtype arch, struct starpu_codelet *cl, unsigned nimpl)
 {
 	switch(arch) {
 	case STARPU_CPU_WORKER:
@@ -111,7 +111,7 @@ int starpu_combined_worker_may_execute_task(unsigned workerid, struct starpu_tas
 	/* TODO: call application-provided function for various cases like
 	 * double support, shared memory size limit, etc. */
 
-	struct starpu_codelet_t *cl = task->cl;
+	struct starpu_codelet *cl = task->cl;
 	unsigned nworkers = config.topology.nworkers;
 
 	/* Is this a parallel worker ? */

src/datawizard/coherency.h

@@ -184,8 +184,8 @@ struct _starpu_data_state {
 	 * the reduction of an interface into another one (eg. "+="), and init_func
 	 * initializes the data interface to a default value that is stable by
 	 * reduction (eg. 0 for +=). */
-	struct starpu_codelet_t *redux_cl;
-	struct starpu_codelet_t *init_cl;
+	struct starpu_codelet *redux_cl;
+	struct starpu_codelet *init_cl;
 
 	/* Are we currently performing a reduction on that handle ? If so the
 	 * reduction_refcnt should be non null until there are pending tasks

src/datawizard/interfaces/multiformat_interface.c

@@ -372,7 +372,7 @@ static int copy_cuda_common(void *src_interface, unsigned src_node,
 			double tmp = starpu_timing_now();
 			void *buffers[1];
 			buffers[0] = src_interface;
-			starpu_codelet *cl = src_multiformat->ops->cpu_to_cuda_cl;
+			struct starpu_codelet *cl = src_multiformat->ops->cpu_to_cuda_cl;
 			cl->cpu_func(buffers, NULL);
 			dst_multiformat->conversion_time = starpu_timing_now() - tmp;
 
@@ -390,7 +390,7 @@ static int copy_cuda_common(void *src_interface, unsigned src_node,
 				STARPU_CUDA_REPORT_ERROR(status);
 		
 			void *buffers[1];
-			starpu_codelet *cl = src_multiformat->ops->cuda_to_cpu_cl;
+			struct starpu_codelet *cl = src_multiformat->ops->cuda_to_cpu_cl;
 			buffers[0] = dst_interface;
 			cl->cpu_func(buffers, NULL);
 							  
@@ -446,7 +446,7 @@ static int copy_cuda_common_async(void *src_interface, unsigned src_node, void *
 			double tmp = starpu_timing_now();
 			void *buffers[1]; // XXX
 			buffers[0] = src_interface;
-			starpu_codelet *cl = src_multiformat->ops->cpu_to_cuda_cl;
+			struct starpu_codelet *cl = src_multiformat->ops->cpu_to_cuda_cl;
 			cl->cpu_func(buffers, NULL);
 			dst_multiformat->conversion_time = starpu_timing_now() - tmp;
 
@@ -466,7 +466,7 @@ static int copy_cuda_common_async(void *src_interface, unsigned src_node, void *
 
 			/* Converting data */
 			void *buffers[1];
-			starpu_codelet *cl = src_multiformat->ops->cuda_to_cpu_cl;
+			struct starpu_codelet *cl = src_multiformat->ops->cuda_to_cpu_cl;
 			buffers[0] = dst_interface;
 			cl->cpu_func(buffers, NULL);
 
@@ -611,7 +611,7 @@ static int copy_ram_to_opencl_async(void *src_interface, unsigned src_node,
 
 		double tmp = starpu_timing_now();
 		void *buffers[1];
-		starpu_codelet *cl = src_multiformat->ops->cpu_to_opencl_cl;
+		struct starpu_codelet *cl = src_multiformat->ops->cpu_to_opencl_cl;
 		buffers[0] = src_interface;
 		cl->cpu_func(buffers, NULL);
 		dst_multiformat->conversion_time = starpu_timing_now() - tmp;
@@ -670,7 +670,7 @@ static int copy_opencl_to_ram_async(void *src_interface, unsigned src_node,
 	/* XXX So much for asynchronicity */
 	clWaitForEvents(1, _event);
 	void *buffers[1];
-	starpu_codelet *cl = src_multiformat->ops->opencl_to_cpu_cl;
+	struct starpu_codelet *cl = src_multiformat->ops->opencl_to_cpu_cl;
 	buffers[0] = dst_interface;
 	cl->cpu_func(buffers, NULL);
 

src/datawizard/reduction.c

@@ -21,8 +21,8 @@
 #include <datawizard/datawizard.h>
 
 void starpu_data_set_reduction_methods(starpu_data_handle_t handle,
-					struct starpu_codelet_t *redux_cl,
-					struct starpu_codelet_t *init_cl)
+					struct starpu_codelet *redux_cl,
+					struct starpu_codelet *init_cl)
 {
 	_starpu_spin_lock(&handle->header_lock);
 
@@ -46,7 +46,7 @@ void _starpu_redux_init_data_replicate(starpu_data_handle_t handle, struct starp
 	STARPU_ASSERT(replicate);
 	STARPU_ASSERT(replicate->allocated);
 
-	struct starpu_codelet_t *init_cl = handle->init_cl;
+	struct starpu_codelet *init_cl = handle->init_cl;
 	STARPU_ASSERT(init_cl);
 
 	cl_func init_func = NULL;

src/drivers/cpu/driver_cpu.c

@@ -31,7 +31,7 @@ static int execute_job_on_cpu(starpu_job_t j, struct starpu_worker_s *cpu_args,
 	struct timespec codelet_start, codelet_end;
 
 	struct starpu_task *task = j->task;
-	struct starpu_codelet_t *cl = task->cl;
+	struct starpu_codelet *cl = task->cl;
 
 	STARPU_ASSERT(cl);
 	STARPU_ASSERT(cl->cpu_func);

src/drivers/cuda/driver_cuda.c

@@ -187,7 +187,7 @@ static int execute_job_on_cuda(starpu_job_t j, struct starpu_worker_s *args)
 	unsigned calibrate_model = 0;
 
 	STARPU_ASSERT(task);
-	struct starpu_codelet_t *cl = task->cl;
+	struct starpu_codelet *cl = task->cl;
 	STARPU_ASSERT(cl);
 
 	if (cl->model && cl->model->benchmarking)