bool
GOMP_OFFLOAD_unload_image (int ord, unsigned version, const void *target_data)
{
struct ptx_image_data *image, **prev_p;
struct ptx_device *dev = ptx_devices[ord];
if (GOMP_VERSION_DEV (version) > GOMP_VERSION_NVIDIA_PTX)
{
GOMP_PLUGIN_error ("Offload data incompatible with PTX plugin"
" (expected %u, received %u)",
GOMP_VERSION_NVIDIA_PTX, GOMP_VERSION_DEV (version));
return false;
}
bool ret = true;
pthread_mutex_lock (&dev->image_lock);
for (prev_p = &dev->images; (image = *prev_p) != 0; prev_p = &image->next)
if (image->target_data == target_data)
{
*prev_p = image->next;
if (cuModuleUnload (image->module) != CUDA_SUCCESS)
ret = false;
free (image->fns);
free (image);
break;
}
pthread_mutex_unlock (&dev->image_lock);
return ret;
}
void
GOMP_OFFLOAD_unload_image (int ord, unsigned version, const void *target_data)
{
struct ptx_image_data *image, **prev_p;
struct ptx_device *dev = ptx_devices[ord];
if (GOMP_VERSION_DEV (version) > GOMP_VERSION_NVIDIA_PTX)
return;
pthread_mutex_lock (&dev->image_lock);
for (prev_p = &dev->images; (image = *prev_p) != 0; prev_p = &image->next)
if (image->target_data == target_data)
{
*prev_p = image->next;
cuModuleUnload (image->module);
free (image->fns);
free (image);
break;
}
pthread_mutex_unlock (&dev->image_lock);
}
int
GOMP_OFFLOAD_load_image (int ord, unsigned version, const void *target_data,
struct addr_pair **target_table)
{
CUmodule module;
const char *const *var_names;
const struct targ_fn_launch *fn_descs;
unsigned int fn_entries, var_entries, i, j;
CUresult r;
struct targ_fn_descriptor *targ_fns;
struct addr_pair *targ_tbl;
const nvptx_tdata_t *img_header = (const nvptx_tdata_t *) target_data;
struct ptx_image_data *new_image;
struct ptx_device *dev;
if (GOMP_VERSION_DEV (version) > GOMP_VERSION_NVIDIA_PTX)
GOMP_PLUGIN_fatal ("Offload data incompatible with PTX plugin"
" (expected %u, received %u)",
GOMP_VERSION_NVIDIA_PTX, GOMP_VERSION_DEV (version));
GOMP_OFFLOAD_init_device (ord);
dev = ptx_devices[ord];
nvptx_attach_host_thread_to_device (ord);
link_ptx (&module, img_header->ptx_objs, img_header->ptx_num);
/* The mkoffload utility emits a struct of pointers/integers at the
start of each offload image. The array of kernel names and the
functions addresses form a one-to-one correspondence. */
var_entries = img_header->var_num;
var_names = img_header->var_names;
fn_entries = img_header->fn_num;
fn_descs = img_header->fn_descs;
targ_tbl = GOMP_PLUGIN_malloc (sizeof (struct addr_pair)
* (fn_entries + var_entries));
targ_fns = GOMP_PLUGIN_malloc (sizeof (struct targ_fn_descriptor)
* fn_entries);
*target_table = targ_tbl;
new_image = GOMP_PLUGIN_malloc (sizeof (struct ptx_image_data));
new_image->target_data = target_data;
new_image->module = module;
new_image->fns = targ_fns;
pthread_mutex_lock (&dev->image_lock);
new_image->next = dev->images;
dev->images = new_image;
pthread_mutex_unlock (&dev->image_lock);
for (i = 0; i < fn_entries; i++, targ_fns++, targ_tbl++)
{
CUfunction function;
r = cuModuleGetFunction (&function, module, fn_descs[i].fn);
if (r != CUDA_SUCCESS)
GOMP_PLUGIN_fatal ("cuModuleGetFunction error: %s", cuda_error (r));
targ_fns->fn = function;
targ_fns->launch = &fn_descs[i];
targ_tbl->start = (uintptr_t) targ_fns;
targ_tbl->end = targ_tbl->start + 1;
}
for (j = 0; j < var_entries; j++, targ_tbl++)
{
CUdeviceptr var;
size_t bytes;
r = cuModuleGetGlobal (&var, &bytes, module, var_names[j]);
if (r != CUDA_SUCCESS)
GOMP_PLUGIN_fatal ("cuModuleGetGlobal error: %s", cuda_error (r));
targ_tbl->start = (uintptr_t) var;
targ_tbl->end = targ_tbl->start + bytes;
}
return fn_entries + var_entries;
}
