optimize bytecode

This commit is contained in:
Sky Johnson 2025-03-27 21:41:06 -05:00
parent faab0a2d08
commit fed0c2ad34

View File

@ -12,7 +12,7 @@ typedef struct {
const char *name;
} BytecodeReader;
static const char *bytecode_reader(lua_State *L, void *ud, size_t *size) {
const char *bytecode_reader(lua_State *L, void *ud, size_t *size) {
BytecodeReader *r = (BytecodeReader *)ud;
(void)L; // unused
if (r->size == 0) return NULL;
@ -21,45 +21,23 @@ static const char *bytecode_reader(lua_State *L, void *ud, size_t *size) {
return (const char *)r->buf;
}
static int load_bytecode(lua_State *L, const unsigned char *buf, size_t len, const char *name) {
int load_bytecode(lua_State *L, const unsigned char *buf, size_t len, const char *name) {
BytecodeReader reader = {buf, len, name};
return lua_load(L, bytecode_reader, &reader, name);
}
typedef struct {
unsigned char *buf;
size_t len;
size_t capacity;
} BytecodeWriter;
static int bytecode_writer(lua_State *L, const void *p, size_t sz, void *ud) {
BytecodeWriter *w = (BytecodeWriter *)ud;
unsigned char *newbuf;
(void)L; // unused
// Check if we need to reallocate
if (w->len + sz > w->capacity) {
size_t new_capacity = w->capacity * 2;
if (new_capacity < w->len + sz) {
new_capacity = w->len + sz;
}
newbuf = (unsigned char *)realloc(w->buf, new_capacity);
// Direct bytecode dumping without intermediate buffer - more efficient
int direct_bytecode_writer(lua_State *L, const void *p, size_t sz, void *ud) {
void **data = (void **)ud;
size_t current_size = (size_t)data[1];
void *newbuf = realloc(data[0], current_size + sz);
if (newbuf == NULL) return 1;
w->buf = newbuf;
w->capacity = new_capacity;
}
memcpy(w->buf + w->len, p, sz);
w->len += sz;
memcpy((unsigned char*)newbuf + current_size, p, sz);
data[0] = newbuf;
data[1] = (void*)(current_size + sz);
return 0;
}
// Wrapper function that calls lua_dump with bytecode_writer
static int dump_lua_function(lua_State *L, BytecodeWriter *w) {
return lua_dump(L, bytecode_writer, w);
}
*/
import "C"
import (
@ -73,31 +51,23 @@ func (s *State) CompileBytecode(code string, name string) ([]byte, error) {
return nil, fmt.Errorf("failed to load string: %w", err)
}
// Set up writer with initial capacity
var writer C.BytecodeWriter
writer.buf = nil
writer.len = 0
writer.capacity = 0
// Initial allocation with a reasonable size
const initialSize = 4096
writer.buf = (*C.uchar)(C.malloc(initialSize))
if writer.buf == nil {
s.Pop(1) // Remove the loaded function
return nil, fmt.Errorf("failed to allocate memory for bytecode")
}
writer.capacity = initialSize
// Use a simpler direct writer with just two pointers
data := [2]unsafe.Pointer{nil, nil}
// Dump the function to bytecode
err := s.safeCall(func() C.int {
return C.dump_lua_function(s.L, (*C.BytecodeWriter)(unsafe.Pointer(&writer)))
return C.lua_dump(s.L, (*[0]byte)(unsafe.Pointer(C.direct_bytecode_writer)), unsafe.Pointer(&data))
})
// Copy bytecode to Go slice regardless of the result
bytecode := C.GoBytes(unsafe.Pointer(writer.buf), C.int(writer.len))
// Get result
var bytecode []byte
if data[0] != nil {
// Create Go slice that references the C memory
length := uintptr(data[1])
bytecode = C.GoBytes(data[0], C.int(length))
C.free(data[0])
}
// Clean up
C.free(unsafe.Pointer(writer.buf))
s.Pop(1) // Remove the function from stack
if err != nil {
@ -164,6 +134,11 @@ func (s *State) LoadAndRunBytecodeWithResults(bytecode []byte, name string, nres
// CompileAndRun compiles and immediately executes Lua code
func (s *State) CompileAndRun(code string, name string) error {
// Skip bytecode step for small scripts - direct execution is faster
if len(code) < 1024 {
return s.DoString(code)
}
bytecode, err := s.CompileBytecode(code, name)
if err != nil {
return fmt.Errorf("compile error: %w", err)