All the embedding APIs supported by the runtime are defined under folder core/iwasm/include. The API details are available in the header files.
Embed WAMR into developer's project
WAMR is designed to be easy embeddable in any project, a typical way of building WAMR is to use cmake, developer can configure features by setting cmake variables and then include the script runtime_lib.cmake under folder build-scripts in his CMakeList.txt, for example:
char *buffer, error_buf[128];
wasm_module_t module;
wasm_module_inst_t module_inst;
wasm_function_inst_t func;
wasm_exec_env_t exec_env;
uint32 size, stack_size = 8092, heap_size = 8092;
/* initialize the wasm runtime by default configurations */
wasm_runtime_init();
/* read WASM file into a memory buffer */
buffer = read_wasm_binary_to_buffer(…, &size);
/* add line below if we want to export native functions to WASM app */
wasm_runtime_register_natives(...);
/* parse the WASM file from buffer and create a WASM module */
module = wasm_runtime_load(buffer, size, error_buf, sizeof(error_buf));
/* create an instance of the WASM module (WASM linear memory is ready) */
module_inst = wasm_runtime_instantiate(module, stack_size, heap_size,
error_buf, sizeof(error_buf));
The wasm_runtime_init() uses the default memory allocator os_malloc/os_free function from the core/shared/platform for the runtime memory management.
WAMR supports to restrict its all memory allocations in a raw buffer. It ensures the dynamic memories used by the WASM applications won't harm the system availability, which is extremely important for embedded systems. This can be done by using wasm_runtime_full_init(). This function also allows you to configure the native API's for exporting to WASM app, and set the maximum thread number when multi-thread feature is enabled.
Refer to the following sample:
/* the native functions that will be exported to WASM app */static NativeSymbol native_symbols[]= {EXPORT_WASM_API_WITH_SIG(display_input_read,"(*)i"),EXPORT_WASM_API_WITH_SIG(display_flush,"(iiii*)")};/* all the runtime memory allocations are retricted in the global_heap_buf array */staticchar global_heap_buf[512*1024];RuntimeInitArgs init_args;memset(&init_args,0,sizeof(RuntimeInitArgs));/* configure the memory allocator for the runtime */init_args.mem_alloc_type = Alloc_With_Pool;init_args.mem_alloc_option.pool.heap_buf = global_heap_buf;init_args.mem_alloc_option.pool.heap_size =sizeof(global_heap_buf);/* configure the native functions being exported to WASM app */init_args.native_module_name ="env";init_args.n_native_symbols =sizeof(native_symbols) /sizeof(NativeSymbol);init_args.native_symbols = native_symbols;/* set maximum thread number if needed when multi-thread is enabled, the default value is 4 */init_args.max_thread_num = max_thread_num;/* initialize runtime environment with user configurations*/if (!wasm_runtime_full_init(&init_args)) {return-1;}
Native calls WASM functions and passes parameters
After a module is instantiated, the runtime embedder can lookup the target WASM function by name, and create execution environment to call the function.
/* lookup a WASM function by its name The function signature can NULL here */ func =wasm_runtime_lookup_function(module_inst,"fib"); /* creat an execution environment to execute the WASM functions */ exec_env =wasm_runtime_create_exec_env(module_inst, stack_size);
There are several ways to call WASM function:
Function call with parameters in an array of 32 bits elements and size:
uint32 argv[2]; /* arguments are always transferred in 32-bit element */ argv[0] =8; /* call the WASM function */if (wasm_runtime_call_wasm(exec_env, func,1, argv) ) { /* the return value is stored in argv[0] */printf("fib function return: %d\n", argv[0]); }else { /* exception is thrown if call fails */printf("%s\n", wasm_runtime_get_exception(module_inst)); }
The parameters are transferred in an array of 32 bits elements. For parameters that occupy 4 or fewer bytes, each parameter can be a single array element. For parameters in types like double or int64, each parameter will take two array elements. The function return value will be sent back in the first one or two elements of the array according to the value type. See the sample code below:
uint32 argv[6];char arg1 ='a';int arg2 =10;double arg3 =1.0; int64 arg4 =100;double ret; argv[0] = arg1; argv[1] = arg2; /** * use memory copy for 8-byte parameters rather than * *(double*)(&argv[2]) = arg3 here because some archs * like ARM, MIPS require the address must be 8-byte aligned. * Or use the aligned malloc or compiler align attribute * to ensure the array address is 8-byte aligned */memcpy(&argv[2],&arg3,sizeof(arg3));memcpy(&argv[4],&arg4,sizeof(arg4)); /* attention: the arg number is 6 here since both arg3 and arg4 each takes 2 elements */wasm_runtime_call_wasm(exec_env, func,6, argv); /* if the return value is type of 8 bytes, it takes the first two array elements */memcpy(&ret,&argv[0],sizeof(ret));
Function call with results and arguments both in wasm_val_t struct and size:
uint32 num_args =1, num_results =1;wasm_val_t args[1], results[1]; /* set the argument type and value */ args[0].kind = WASM_I32; args[0].of.i32 =8; /* call the WASM function */if (wasm_runtime_call_wasm_a(exec_env, func, num_results, results, num_args, args)) { /* the return value is stored in results */printf("fib function return: %d\n", results[0].of.i32); }else { /* exception is thrown if call fails */printf("%s\n", wasm_runtime_get_exception(module_inst)); }
Function call with variant argument support:
uint32 num_args =1, num_results =1;wasm_val_t results[1]; /* call the WASM function */if (wasm_runtime_call_wasm_v(exec_env, func,1, results,1,8)) { /* the return value is stored in results */printf("fib function return: %d\n", results[0].of.i32); }else { /* exception is thrown if call fails */printf("%s\n", wasm_runtime_get_exception(module_inst)); }
Pass buffer to WASM function
If we need to transfer a buffer to WASM function, we can pass the buffer address through a parameter. Attention: The sandbox will forbid the WASM code to access outside memory, we must allocate the buffer from WASM instance's own memory space and pass the buffer address in instance's space (not the runtime native address).
There are two runtime APIs available for this purpose.
/** * malloc a buffer from instance's private memory space. * * return: the buffer address in instance's memory space (pass to the WASM function) * p_native_addr: return the native address of allocated memory * size: the buffer size to allocate */uint64_twasm_runtime_module_malloc(wasm_module_inst_t module_inst,uint64_t size,void**p_native_addr);/** * malloc a buffer from instance's private memory space, * and copy the data from another native buffer to it. * * return: the buffer address in instance's memory space (pass to the WASM function) * src: the native buffer address * size: the size of buffer to be allocated and copy data */uint64_twasm_runtime_module_dup_data(wasm_module_inst_t module_inst,constchar*src,uint64_t size);/* free the memory allocated from module memory space */voidwasm_runtime_module_free(wasm_module_inst_t module_inst,uint64_t ptr);
Usage sample:
char* buffer =NULL;uint64_t buffer_for_wasm;buffer_for_wasm =wasm_runtime_module_malloc(module_inst,100,&buffer);if (buffer_for_wasm !=0) { uint32 argv[3];strncpy(buffer,"hello",100); /* use native address for accessing in runtime */ argv[0] = buffer_for_wasm; /* pass the buffer address for WASM space */ argv[2] =100; /* the size of buffer */wasm_runtime_call_wasm(exec_env, func,3, argv); /* it is runtime embedder's responsibility to release the memory, unless the WASM app will free the passed pointer in its code */wasm_runtime_module_free(module_inst, buffer_for_wasm);}
Pass structured data to WASM function
We can't pass structure data or class objects through the pointer since the memory layout can different in two worlds. The way to do it is serialization. Refer to export_native_api.md for the details.
Execute wasm functions in multiple threads
It isn't safe to use an exec_env object in multiple threads concurrently. To run a multi-threaded application, you basically need a separate exec_env for each threads.
Approaches to manage exec_env objects and threads
WAMR supports two approaches to manage exec_env and threads as described below. While they are not exclusive, you usually only need to use one of them.
Make your WASM application manage threads
You can make your WASM application spawn threads by itself, typically using pthread APIs like pthread_create. See pthread library and pthread implementations for more details. In this case, WAMR manages exec_env for the spawned threads.
Make your embedder manage threads
The spawn exec_env and spawn thread APIs are available for the embedder. You can use these APIs to manage the threads. See Thread related embedder API for details.
Other notes about threads
You can manage the maximum number of threads
init_args.max_thread_num = THREAD_NUM;
/* If this init argument is not set, the default maximum thread number is 4 */
To share memory among threads, you need to build your WASM application with shared memory
For example, it can be done with --shared-memory and -pthread.
