gpudata *cuda_make_buf(void *c, CUdeviceptr p, size_t sz) {
cuda_context *ctx = (cuda_context *)c;
gpudata *res;
int flags = CU_EVENT_DISABLE_TIMING;
res = malloc(sizeof(*res));
if (res == NULL) return NULL;
res->refcnt = 1;
cuda_enter(ctx);
if (ctx->err != CUDA_SUCCESS) {
free(res);
return NULL;
}
res->ptr = p;
if (ctx->flags & GA_CTX_MULTI_THREAD)
flags |= CU_EVENT_BLOCKING_SYNC;
ctx->err = cuEventCreate(&res->ev, flags);
if (ctx->err != CUDA_SUCCESS) {
free(res);
cuda_exit(ctx);
return NULL;
}
res->sz = sz;
res->flags = DONTFREE;
res->ctx = ctx;
ctx->refcnt++;
cuda_exit(ctx);
TAG_BUF(res);
return res;
}
static gpudata *new_gpudata(cuda_context *ctx, CUdeviceptr ptr, size_t size) {
gpudata *res;
int fl = CU_EVENT_DISABLE_TIMING;
res = malloc(sizeof(*res));
if (res == NULL) return NULL;
res->refcnt = 0;
res->sz = size;
res->flags = 0;
cuda_enter(ctx);
if (ctx->flags & GA_CTX_MULTI_THREAD)
fl |= CU_EVENT_BLOCKING_SYNC;
ctx->err = cuEventCreate(&res->rev, fl);
if (ctx->err != CUDA_SUCCESS) {
cuda_exit(ctx);
free(res);
return NULL;
}
ctx->err = cuEventCreate(&res->wev, fl);
if (ctx->err != CUDA_SUCCESS) {
cuEventDestroy(res->rev);
cuda_exit(ctx);
free(res);
return NULL;
}
cuda_exit(ctx);
res->ptr = ptr;
res->next = NULL;
res->ctx = ctx;
TAG_BUF(res);
return res;
}