Hello.
Following patch has been installed to HSA branch.
Thanks,
Martin
>From 64f7f75a8b16dee2071fef7f546fbf6b06f82c4f Mon Sep 17 00:00:00 2001
From: mliska <mli...@suse.cz>
Date: Fri, 31 Jul 2015 17:53:57 +0200
Subject: [PATCH 1/3] HSA: introduce more pure kernel information emission.
libgomp/ChangeLog:
2015-07-31 Martin Liska <mli...@suse.cz>
* plugin/plugin-hsa.c (struct hsa_kernel_description): New structure.
(struct brig_image_desc): Use it.
(struct kernel_info): Add omp_data_size member.
(struct module_info): Remove unused field.
(GOMP_OFFLOAD_load_image): Load the newly added format.
(create_kernel_dispatch): Allocate exactly ammount of memory needed
for OMP data passing.
(init_single_kernel): Print the OMP memory.
hsa.c (init_hsa_image): Use new structure.
gcc/ChangeLog:
2015-07-31 Martin Liska <mli...@suse.cz>
* hsa-brig.c (hsa_output_kernel_mapping): Serialize new data structure.
* hsa-gen.c (hsa_function_representation::hsa_function_representation):
Add new field.
(gen_hsa_insns_for_kernel_call): Calculate maximum amount of memory
needed for OMP data.
(generate_hsa): Likewise.
* hsa.c (hsa_add_kern_decl_mapping): New function.
(hsa_get_decl_kernel_mapping_omp_size): Likewise.
* hsa.h: Add declaration of these functions.
---
gcc/hsa-brig.c | 231 ++++++++++++++++++++++++++------------------
gcc/hsa-gen.c | 16 ++-
gcc/hsa.c | 13 ++-
gcc/hsa.h | 8 +-
libgomp/hsa.c | 35 +++----
libgomp/plugin/plugin-hsa.c | 107 +++++---------------
6 files changed, 209 insertions(+), 201 deletions(-)
diff --git a/gcc/hsa-brig.c b/gcc/hsa-brig.c
index 27c41a5..1cc0b22 100644
--- a/gcc/hsa-brig.c
+++ b/gcc/hsa-brig.c
@@ -1753,7 +1753,7 @@ hsa_output_kernel_mapping (tree brig_decl)
unsigned map_count = hsa_get_number_decl_kernel_mappings ();
tree int_num_of_kernels;
- int_num_of_kernels = build_int_cst (integer_type_node, (int) map_count);
+ int_num_of_kernels = build_int_cst (uint32_type_node, map_count);
tree kernel_num_index_type = build_index_type (int_num_of_kernels);
tree host_functions_array_type = build_array_type (ptr_type_node,
kernel_num_index_type);
@@ -1768,7 +1768,7 @@ hsa_output_kernel_mapping (tree brig_decl)
tree host_functions_ctor = build_constructor (host_functions_array_type,
host_functions_vec);
char tmp_name[64];
- ASM_GENERATE_INTERNAL_LABEL (tmp_name, "hsa_host_functions", 1);
+ ASM_GENERATE_INTERNAL_LABEL (tmp_name, "__hsa_host_functions", 1);
tree hsa_host_func_table = build_decl (UNKNOWN_LOCATION, VAR_DECL,
get_identifier (tmp_name),
host_functions_array_type);
@@ -1782,115 +1782,161 @@ hsa_output_kernel_mapping (tree brig_decl)
DECL_INITIAL (hsa_host_func_table) = host_functions_ctor;
varpool_node::finalize_decl (hsa_host_func_table);
- int len = 0;
- for (unsigned i = 0; i < map_count; ++i)
- {
- char *name = hsa_get_decl_kernel_mapping_name (i);
- /* We add 1 for the terminating zero and 1 for an ampersand prefix. */
- len = len + strlen (name) + 2;
- }
- len++;
+ /* Following code emits list of kernel_info structures. */
- /* Kernel mapping is a list of string names terminated by '\0'. */
- char *buf = XNEWVEC (char, len);
- char *p = buf;
- for (unsigned i = 0; i < map_count; ++i)
- {
- char *name = hsa_get_decl_kernel_mapping_name (i);
- int ll = strlen (name);
- gcc_assert (ll > 0);
- *p = '&';
- p++;
- memcpy (p, name, ll);
- p += ll;
- *p = '\0';
- p++;
- }
- *p = '\0';
- tree kern_names = build_string (len, buf);
- TREE_TYPE (kern_names) = build_array_type (char_type_node,
- build_index_type (size_int (len)));
- free (buf);
+ tree kernel_info_type = make_node (RECORD_TYPE);
+ tree id_f1 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ get_identifier ("name"), ptr_type_node);
+ DECL_CHAIN (id_f1) = NULL_TREE;
+ tree id_f2 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ get_identifier ("omp_data_size"),
+ unsigned_type_node);
+ DECL_CHAIN (id_f2) = id_f1;
+ tree id_f3 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ get_identifier ("kernel_dependencies_count"),
+ unsigned_type_node);
+ DECL_CHAIN (id_f3) = id_f2;
+ tree id_f4 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ get_identifier ("kernel_dependencies"),
+ build_pointer_type (build_pointer_type
+ (char_type_node)));
+ DECL_CHAIN (id_f4) = id_f3;
+ finish_builtin_struct (kernel_info_type, "__hsa_kernel_info", id_f4,
+ NULL_TREE);
- /* Kernel dependencies is a list of lists, where for a given kernel A all
- direct kernel dispatches to A1, A2, .., An are organized are joined by '.'
- character and each entry in the list is separated by '\0'. */
+ int_num_of_kernels = build_int_cstu (uint32_type_node, map_count);
+ tree kernel_info_vector_type = build_array_type
+ (kernel_info_type, build_index_type (int_num_of_kernels));
+
+ vec<constructor_elt, va_gc> *kernel_info_vector_vec = NULL;
+ tree kernel_dependencies_vector_type = NULL;
- len = 0;
for (unsigned i = 0; i < map_count; ++i)
{
- tree caller = hsa_get_decl_kernel_mapping_decl (i);
-
+ tree kernel = hsa_get_decl_kernel_mapping_decl (i);
+ char *name = hsa_get_decl_kernel_mapping_name (i);
+ unsigned len = strlen (name);
+ char *copy = XNEWVEC (char, len + 2);
+ copy[0] = '&';
+ memcpy (copy + 1, name, len);
+ copy[len + 1] = '\0';
+ len++;
+
+ tree kern_name = build_string (len, copy);
+ TREE_TYPE (kern_name) = build_array_type
+ (char_type_node, build_index_type (size_int (len)));
+ free (copy);
+
+ unsigned omp_size = hsa_get_decl_kernel_mapping_omp_size (i);
+ tree omp_data_size = build_int_cstu (uint32_type_node, omp_size);
+ unsigned count = 0;
+
+ kernel_dependencies_vector_type = build_array_type
+ (build_pointer_type (char_type_node),
+ build_index_type (size_int (0)));
+
+ vec<constructor_elt, va_gc> *kernel_dependencies_vec = NULL;
if (hsa_decl_kernel_dependencies)
{
- vec<char *> **slot = hsa_decl_kernel_dependencies->get (caller);
+ vec<char *> **slot = hsa_decl_kernel_dependencies->get (kernel);
if (slot)
{
- vec<char *> *s = *slot;
- for (unsigned i = 0; i < s->length (); i++)
- len += (strlen ((*s)[i]) + 1);
+ vec <char *> *dependencies = *slot;
+ count = dependencies->length ();
+
+ kernel_dependencies_vector_type = build_array_type
+ (build_pointer_type (char_type_node),
+ build_index_type (size_int (count)));
- /* Add N-1 dot characters. */
- len += 2 * (s->length () - 1);
+ for (unsigned j = 0; j < count; j++)
+ {
+ char *d = (*dependencies)[j];
+ len = strlen (d);
+ tree dependency_name = build_string (len, d);
+ TREE_TYPE (dependency_name) = build_array_type
+ (char_type_node, build_index_type (size_int (len)));
+
+ CONSTRUCTOR_APPEND_ELT
+ (kernel_dependencies_vec, NULL_TREE,
+ build1 (ADDR_EXPR,
+ build_pointer_type (TREE_TYPE (dependency_name)),
+ dependency_name));
+ }
}
}
- /* Zero termination character. */
- ++len;
- }
+ tree dependencies_count = build_int_cstu (uint32_type_node, count);
- buf = XNEWVEC (char, len);
- p = buf;
- for (unsigned i = 0; i < map_count; ++i)
- {
- tree caller = hsa_get_decl_kernel_mapping_decl (i);
+ vec<constructor_elt, va_gc> *kernel_info_vec = NULL;
+ CONSTRUCTOR_APPEND_ELT (kernel_info_vec, NULL_TREE,
+ build1 (ADDR_EXPR,
+ build_pointer_type (TREE_TYPE
+ (kern_name)),
+ kern_name));
+ CONSTRUCTOR_APPEND_ELT (kernel_info_vec, NULL_TREE, omp_data_size);
+ CONSTRUCTOR_APPEND_ELT (kernel_info_vec, NULL_TREE, dependencies_count);
- if (hsa_decl_kernel_dependencies)
+ tree kernel_info_ctor = build_constructor (kernel_info_type,
+ kernel_info_vec);
+
+ if (count > 0)
{
- vec<char *> **slot = hsa_decl_kernel_dependencies->get (caller);
- if (slot)
- {
- vec<char *> *s = *slot;
- unsigned k, j = 0;
- for (k = 0; k < s->length (); k++)
- {
- unsigned ll = strlen ((*s)[k]);
- gcc_assert (ll > 0);
- memcpy (p, (*s)[k], ll);
- p += ll;
- *p = '\0';
- p++;
-
- /* If it is not a last elements, append '.' string. */
- if (j != s->length () - 1)
- {
- *p++ = '.';
- *p++ = '\0';
- }
-
- j++;
- }
- }
+ ASM_GENERATE_INTERNAL_LABEL (tmp_name, "__hsa_dependencies_list", i);
+ tree dependencies_list = build_decl (UNKNOWN_LOCATION, VAR_DECL,
+ get_identifier (tmp_name),
+ kernel_dependencies_vector_type);
+
+ TREE_STATIC (dependencies_list) = 1;
+ TREE_READONLY (dependencies_list) = 1;
+ TREE_PUBLIC (dependencies_list) = 0;
+ DECL_ARTIFICIAL (dependencies_list) = 1;
+ DECL_IGNORED_P (dependencies_list) = 1;
+ DECL_EXTERNAL (dependencies_list) = 0;
+ TREE_CONSTANT (dependencies_list) = 1;
+ DECL_INITIAL (dependencies_list) = build_constructor
+ (kernel_dependencies_vector_type, kernel_dependencies_vec);
+ varpool_node::finalize_decl (dependencies_list);
+
+ CONSTRUCTOR_APPEND_ELT (kernel_info_vec, NULL_TREE,
+ build1 (ADDR_EXPR,
+ build_pointer_type
+ (TREE_TYPE (dependencies_list)),
+ dependencies_list));
}
+ else
+ CONSTRUCTOR_APPEND_ELT (kernel_info_vec, NULL_TREE, null_pointer_node);
- /* Zero termination character. */
- *p++ = '\0';
+ CONSTRUCTOR_APPEND_ELT (kernel_info_vector_vec, NULL_TREE,
+ kernel_info_ctor);
}
- tree kern_dependencies = build_string (len, buf);
- TREE_TYPE (kern_dependencies) = build_array_type
- (char_type_node, build_index_type (size_int (len)));
- free (buf);
+ ASM_GENERATE_INTERNAL_LABEL (tmp_name, "__hsa_kernels", 1);
+ tree hsa_kernels = build_decl (UNKNOWN_LOCATION, VAR_DECL,
+ get_identifier (tmp_name),
+ kernel_info_vector_type);
+
+ TREE_STATIC (hsa_kernels) = 1;
+ TREE_READONLY (hsa_kernels) = 1;
+ TREE_PUBLIC (hsa_kernels) = 0;
+ DECL_ARTIFICIAL (hsa_kernels) = 1;
+ DECL_IGNORED_P (hsa_kernels) = 1;
+ DECL_EXTERNAL (hsa_kernels) = 0;
+ TREE_CONSTANT (hsa_kernels) = 1;
+ DECL_INITIAL (hsa_kernels) = build_constructor (kernel_info_vector_type,
+ kernel_info_vector_vec);
+ varpool_node::finalize_decl (hsa_kernels);
tree hsa_image_desc_type = make_node (RECORD_TYPE);
- tree id_f1 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ id_f1 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
get_identifier ("brig_module"), ptr_type_node);
DECL_CHAIN (id_f1) = NULL_TREE;
- tree id_f2 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
- get_identifier ("kern_names"), ptr_type_node);
+ id_f2 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ get_identifier ("kernel_count"),
+ unsigned_type_node);
+
DECL_CHAIN (id_f2) = id_f1;
- tree id_f3 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
- get_identifier ("kern_dependencies"),
+ id_f3 = build_decl (BUILTINS_LOCATION, FIELD_DECL,
+ get_identifier ("hsa_kernel_infos"),
ptr_type_node);
DECL_CHAIN (id_f3) = id_f2;
finish_builtin_struct (hsa_image_desc_type, "__hsa_image_desc", id_f3,
@@ -1900,19 +1946,16 @@ hsa_output_kernel_mapping (tree brig_decl)
CONSTRUCTOR_APPEND_ELT (img_desc_vec, NULL_TREE,
build_fold_addr_expr (brig_decl));
CONSTRUCTOR_APPEND_ELT (img_desc_vec, NULL_TREE,
- build1 (ADDR_EXPR,
- build_pointer_type (TREE_TYPE (kern_names)),
- kern_names));
+ build_int_cstu (unsigned_type_node, map_count));
CONSTRUCTOR_APPEND_ELT (img_desc_vec, NULL_TREE,
build1 (ADDR_EXPR,
build_pointer_type (TREE_TYPE
- (kern_dependencies)),
- kern_dependencies));
-
+ (hsa_kernels)),
+ hsa_kernels));
tree img_desc_ctor = build_constructor (hsa_image_desc_type, img_desc_vec);
- ASM_GENERATE_INTERNAL_LABEL (tmp_name, "hsa_img_descriptor", 1);
+ ASM_GENERATE_INTERNAL_LABEL (tmp_name, "__hsa_img_descriptor", 1);
tree hsa_img_descriptor = build_decl (UNKNOWN_LOCATION, VAR_DECL,
get_identifier (tmp_name),
hsa_image_desc_type);
@@ -1948,7 +1991,7 @@ hsa_output_kernel_mapping (tree brig_decl)
CONSTRUCTOR_APPEND_ELT (libgomp_host_table_vec, NULL_TREE, null_pointer_node);
tree libgomp_host_table_ctor = build_constructor (libgomp_host_table_type,
libgomp_host_table_vec);
- ASM_GENERATE_INTERNAL_LABEL (tmp_name, "hsa_libgomp_host_table", 1);
+ ASM_GENERATE_INTERNAL_LABEL (tmp_name, "__hsa_libgomp_host_table", 1);
tree hsa_libgomp_host_table = build_decl (UNKNOWN_LOCATION, VAR_DECL,
get_identifier (tmp_name),
libgomp_host_table_type);
diff --git a/gcc/hsa-gen.c b/gcc/hsa-gen.c
index 9784cb8..58ca26c 100644
--- a/gcc/hsa-gen.c
+++ b/gcc/hsa-gen.c
@@ -186,6 +186,7 @@ hsa_function_representation::hsa_function_representation ()
called_functions = vNULL;
shadow_reg = NULL;
kernel_dispatch_count = 0;
+ maximum_omp_data_size = 0;
}
/* Destructor of class holding function/kernel-wide informaton and state. */
@@ -2572,6 +2573,14 @@ gen_hsa_insns_for_kernel_call (hsa_bb *hbb, gcall *call)
tree argument = gimple_call_arg (call, 1);
gcc_assert (TREE_CODE (argument) == ADDR_EXPR);
+ tree d = TREE_TYPE (TREE_OPERAND (argument, 0));
+ unsigned omp_data_size = tree_to_uhwi
+ (TYPE_SIZE_UNIT (d));
+ gcc_checking_assert (omp_data_size > 0);
+
+ if (omp_data_size > hsa_cfun->maximum_omp_data_size)
+ hsa_cfun->maximum_omp_data_size = omp_data_size;
+
hsa_symbol *var_decl = get_symbol_for_decl (TREE_OPERAND (argument, 0));
hbb->append_insn (new (hsa_allocp_inst_comment)
@@ -3201,9 +3210,6 @@ generate_hsa (bool kernel)
= xstrdup (get_declaration_name (current_function_decl));
hsa_sanitize_name (hsa_cfun->name);
- if (hsa_cfun->kern_p)
- hsa_add_kern_decl_mapping (current_function_decl, hsa_cfun->name);
-
gen_function_def_parameters (hsa_cfun, &ssa_map);
if (seen_error ())
goto fail;
@@ -3211,6 +3217,10 @@ generate_hsa (bool kernel)
if (seen_error ())
goto fail;
+ if (hsa_cfun->kern_p)
+ hsa_add_kern_decl_mapping (current_function_decl, hsa_cfun->name,
+ hsa_cfun->maximum_omp_data_size);
+
#ifdef ENABLE_CHECKING
for (unsigned i = 0; i < ssa_map.length (); i++)
if (ssa_map[i])
diff --git a/gcc/hsa.c b/gcc/hsa.c
index 2c02a82..b857342 100644
--- a/gcc/hsa.c
+++ b/gcc/hsa.c
@@ -84,6 +84,8 @@ struct GTY(()) hsa_decl_kernel_map_element
tree decl;
/* Name of the HSA kernel in BRIG. */
char * GTY((skip)) name;
+ /* Size of OMP data, if the kernel contains a kernel dispatch. */
+ unsigned omp_data_size;
};
/* Mapping between decls and corresponding HSA kernels in this compilation
@@ -319,11 +321,12 @@ hsa_destroy_insn (hsa_insn_basic *insn)
/* Create a mapping between the original function DECL and kernel name NAME. */
void
-hsa_add_kern_decl_mapping (tree decl, char *name)
+hsa_add_kern_decl_mapping (tree decl, char *name, unsigned omp_data_size)
{
hsa_decl_kernel_map_element dkm;
dkm.decl = decl;
dkm.name = name;
+ dkm.omp_data_size = omp_data_size;
vec_safe_push (hsa_decl_kernel_mapping, dkm);
}
@@ -351,6 +354,14 @@ hsa_get_decl_kernel_mapping_name (unsigned i)
return (*hsa_decl_kernel_mapping)[i].name;
}
+/* Return maximum OMP size for kernel decl name mapping. */
+
+unsigned
+hsa_get_decl_kernel_mapping_omp_size (unsigned i)
+{
+ return (*hsa_decl_kernel_mapping)[i].omp_data_size;
+}
+
/* Free the mapping between original decls and kernel names. */
void
diff --git a/gcc/hsa.h b/gcc/hsa.h
index c9102f6..eec5f17 100644
--- a/gcc/hsa.h
+++ b/gcc/hsa.h
@@ -821,6 +821,11 @@ public:
/* Number of kernel dispatched which take place in the function. */
unsigned kernel_dispatch_count;
+
+ /* If the function representation contains a kernel dispatch,
+ OMP data size is necessary memory that is used for copying before
+ a kernel dispatch. */
+ unsigned maximum_omp_data_size;
};
/* in hsa.c */
@@ -837,10 +842,11 @@ unsigned hsa_type_bit_size (BrigType16_t t);
BrigType16_t hsa_bittype_for_type (BrigType16_t t);
bool hsa_type_float_p (BrigType16_t type);
void hsa_destroy_insn (hsa_insn_basic *insn);
-void hsa_add_kern_decl_mapping (tree decl, char *name);
+void hsa_add_kern_decl_mapping (tree decl, char *name, unsigned);
unsigned hsa_get_number_decl_kernel_mappings (void);
tree hsa_get_decl_kernel_mapping_decl (unsigned i);
char *hsa_get_decl_kernel_mapping_name (unsigned i);
+unsigned hsa_get_decl_kernel_mapping_omp_size (unsigned i);
void hsa_free_decl_kernel_mapping (void);
void hsa_add_kernel_dependency (tree caller, char *called_function);
void hsa_sanitize_name (char *p);
diff --git a/libgomp/hsa.c b/libgomp/hsa.c
index bda822b..b7c1a71 100644
--- a/libgomp/hsa.c
+++ b/libgomp/hsa.c
@@ -39,10 +39,19 @@ struct hsa_program_info
hsa_executable_t executable;
};
+struct hsa_kernel_description
+{
+ const char *name;
+ uint32_t omp_data_size;
+ uint32_t kernel_dependencies_count;
+ const char **kernel_dependencies;
+};
+
struct hsa_image_desc
{
hsa_ext_module_t module;
- const char *names;
+ const uint32_t kernel_count;
+ struct hsa_kernel_description *kernel_infos;
};
struct hsa_kernel_info
@@ -238,27 +247,17 @@ init_hsa_image (struct hsa_image_info *ii)
struct hsa_one_image *img = ii->first_image;
while (img)
{
- const char *p;
- int count = 0;
+ int count = img->image_descriptor->kernel_count;
struct hsa_kernel_info *kernel;
status = hsa_ext_program_add_module (hsa_program.handle,
img->image_descriptor->module);
+
if (status != HSA_STATUS_SUCCESS)
gomp_fatal ("Could not add a module to the HSA program");
if (debug)
fprintf (stderr, "Added a module to the HSA program\n");
- p = img->image_descriptor->names;
- while (*p)
- {
- count++;
- do
- p++;
- while (*p);
- p++;
- }
-
if (debug)
fprintf (stderr, "Encountered %d kernels an image\n", count);
img->kernels = gomp_malloc_cleared (sizeof (struct hsa_kernel_info)
@@ -266,17 +265,13 @@ init_hsa_image (struct hsa_image_info *ii)
if (!img->kernels)
gomp_fatal ("Could not allocate memory for HSA kertnel descriptors");
- p = img->image_descriptor->names;
kernel = img->kernels;
- while (*p)
+ for (unsigned i = 0; i < count; i++)
{
- kernel->name = p;
+ kernel->name = img->image_descriptor->kernel_infos[i].name;
kernel++;
- do
- p++;
- while (*p);
- p++;
}
+
img = img->next;
}
diff --git a/libgomp/plugin/plugin-hsa.c b/libgomp/plugin/plugin-hsa.c
index c5af705..144548a 100644
--- a/libgomp/plugin/plugin-hsa.c
+++ b/libgomp/plugin/plugin-hsa.c
@@ -60,14 +60,22 @@ hsa_fatal (const char *str, hsa_status_t status)
GOMP_PLUGIN_fatal ("HSA fatal error: %s (%s)", str, hsa_error);
}
+struct hsa_kernel_description
+{
+ const char *name;
+ unsigned omp_data_size;
+ unsigned kernel_dependencies_count;
+ const char **kernel_dependencies;
+};
+
/* Data passed by the static initializer of a compilation unit containing BRIG
to GOMP_offload_register. */
struct brig_image_desc
{
hsa_ext_module_t brig_module;
- const char *names;
- const char *kernel_dependencies;
+ const unsigned kernel_count;
+ struct hsa_kernel_description *kernel_infos;
};
struct agent_info;
@@ -78,6 +86,8 @@ struct kernel_info
{
/* Name of the kernel, required to locate it within the brig module. */
const char *name;
+ /* Size of memory space for OMP data. */
+ unsigned omp_data_size;
/* The specific agent the kernel has been or will be finalized for and run
on. */
struct agent_info *agent;
@@ -115,8 +125,6 @@ struct module_info
/* Number of kernels in this module. */
int kernel_count;
- /* Number of kernel from kernel dispatches. */
- unsigned kernel_dispatch_count;
/* An array of kernel_info structures describing each kernel in this
module. */
struct kernel_info kernels[];
@@ -442,8 +450,7 @@ GOMP_OFFLOAD_load_image (int ord, void *target_data,
struct addr_pair *pair;
struct module_info *module;
struct kernel_info *kernel;
- int kernel_count = 0;
- const char *p;
+ int kernel_count = image_desc->kernel_count;
agent = get_agent_info (ord);
if (pthread_rwlock_wrlock (&agent->modules_rwlock))
@@ -451,15 +458,6 @@ GOMP_OFFLOAD_load_image (int ord, void *target_data,
if (agent->prog_finalized)
destroy_hsa_program (agent);
- p = image_desc->names;
- while (*p)
- {
- kernel_count++;
- do
- p++;
- while (*p);
- p++;
- }
if (kernel_count == 0)
GOMP_PLUGIN_fatal ("No kernels encountered in a brig module description");
if (debug)
@@ -472,84 +470,27 @@ GOMP_OFFLOAD_load_image (int ord, void *target_data,
module->image_desc = image_desc;
module->kernel_count = kernel_count;
- p = image_desc->names;
kernel = &module->kernels[0];
- /* Parse all kernels. */
- while (*p)
+ /* Allocate memory for kernel dependencies. */
+ for (unsigned i = 0; i < kernel_count; i++)
{
pair->start = (uintptr_t) kernel;
pair->end = (uintptr_t) (kernel + 1);
- kernel->name = p;
+
+ struct hsa_kernel_description *d = &image_desc->kernel_infos[i];
kernel->agent = agent;
kernel->module = module;
+ kernel->name = d->name;
+ kernel->omp_data_size = d->omp_data_size;
+ kernel->dependencies_count = d->kernel_dependencies_count;
+ kernel->dependencies = d->kernel_dependencies;
if (pthread_mutex_init (&kernel->init_mutex, NULL))
GOMP_PLUGIN_fatal ("Failed to initialize an HSA kernel mutex");
+
kernel++;
pair++;
- do
- p++;
- while (*p);
- p++;
- }
-
- /* Load length of kernel dependencies. */
- p = image_desc->kernel_dependencies;
- unsigned i = 0;
- do
- {
- while (*p != 0)
- {
- if (*p == '.')
- p += 2;
- else
- {
- module->kernels[i].dependencies_count++;
- module->kernel_dispatch_count++;
-
- do
- p++;
- while (*p);
- p++;
- }
- }
-
- i++;
- p++;
- }
- while (i <= kernel_count - 1);
-
- /* Allocate memory for kernel dependencies. */
- for (unsigned i = 0; i < kernel_count; i++)
- module->kernels[i].dependencies = GOMP_PLUGIN_malloc
- (sizeof (char *) * module->kernels[i].dependencies_count);
-
- /* Parse all kernel dependencies. */
- p = image_desc->kernel_dependencies;
- i = 0;
-
- do
- {
- int j = 0;
- while (*p != 0)
- {
- if (*p == '.')
- p += 2;
- else
- {
- module->kernels[i].dependencies[j++] = p;
-
- do
- p++;
- while (*p);
- p++;
- }
- }
-
- i++;
- p++;
}
- while (i <= kernel_count - 1);
add_module_to_agent (agent, module);
if (pthread_rwlock_unlock (&agent->modules_rwlock))
@@ -641,7 +582,7 @@ create_kernel_dispatch (struct kernel_info *kernel)
shadow->queue = agent->command_q;
/* Compute right size needed for memory allocation. */
- shadow->omp_data_memory = GOMP_PLUGIN_malloc (100);
+ shadow->omp_data_memory = GOMP_PLUGIN_malloc (kernel->omp_data_size);
unsigned dispatch_count = kernel->dependencies_count;
shadow->kernel_dispatch_count = dispatch_count;
@@ -738,6 +679,8 @@ init_single_kernel (struct kernel_info *kernel)
(unsigned) kernel->private_segment_size);
fprintf (stderr, " kernarg_segment_size: %u\n",
(unsigned) kernel->kernarg_segment_size);
+ fprintf (stderr, " omp_data_size: %u\n",
+ kernel->omp_data_size);
}
}
--
2.4.6
