aboutsummaryrefslogtreecommitdiff
path: root/nerv/matrix/cumatrix.c
diff options
context:
space:
mode:
Diffstat (limited to 'nerv/matrix/cumatrix.c')
-rw-r--r--nerv/matrix/cumatrix.c44
1 files changed, 10 insertions, 34 deletions
diff --git a/nerv/matrix/cumatrix.c b/nerv/matrix/cumatrix.c
index af34fb4..1bcb0f1 100644
--- a/nerv/matrix/cumatrix.c
+++ b/nerv/matrix/cumatrix.c
@@ -1,6 +1,7 @@
#define NERV_GENERIC_CUMATRIX
#include "../common.h"
-#include "cuda_helper.h"
+#include "../lib/matrix/cumatrix.h"
+#include "../lib/matrix/cuda_helper.h"
#include <string.h>
#define PROFILE_HASHMAP_SIZE 123457
static cublasHandle_t cublas_handle;
@@ -8,54 +9,29 @@ static cudaEvent_t profile_start, profile_stop;
static HashMap *profile;
static int print_profile(lua_State *L) {
- (void)L;
- size_t i;
- fprintf(stderr, "*** [nerv cumatrix profile] **\n");
- for (i = 0; i < profile->size; i++)
- {
- HashNode *ptr;
- for (ptr = profile->bucket[i]; ptr; ptr = ptr->next)
- {
- fprintf(stderr, "%s:\t%.6f\n", ptr->key, *(float *)ptr->val);
- }
- }
+ nerv_cumatrix_print_profile();
return 0;
}
static int clear_profile(lua_State *L) {
- (void)L;
- hashmap_clear(profile);
+ nerv_cumatrix_clear_profile();
return 0;
}
-void accu_profile(const char *name, float delta) {
- float *val = hashmap_getval(profile, name);
- if (!val)
- {
- val = malloc(sizeof(float));
- *val = 0;
- hashmap_setval(profile, name, val);
- }
- *val += delta;
-}
-
static const luaL_Reg cumatrix_methods[] = {
{"print_profile", print_profile},
{"clear_profile", clear_profile},
{NULL, NULL}
};
-extern void nerv_matrix_cuda_float_init(lua_State *L);
-extern void nerv_matrix_cuda_double_init(lua_State *L);
+extern void nerv_matrix_cuda_float_lua_init(lua_State *L);
+extern void nerv_matrix_cuda_double_lua_init(lua_State *L);
-void nerv_cumatrix_init(lua_State *L) {
+void nerv_lua_cumatrix_init(lua_State *L) {
luaL_register(L, NULL, cumatrix_methods);
- cublasCreate(&cublas_handle);
- cudaEventCreate(&profile_start);
- cudaEventCreate(&profile_stop);
- profile = hashmap_create(PROFILE_HASHMAP_SIZE, bkdr_hash, strcmp);
- nerv_matrix_cuda_float_init(L);
- nerv_matrix_cuda_double_init(L);
+ nerv_cumatrix_init();
+ nerv_matrix_cuda_float_lua_init(L);
+ nerv_matrix_cuda_double_lua_init(L);
}
#define MATRIX_USE_FLOAT