feat(ir): 重构指针操作API并优化数组访问逻辑

- 将scc_ir_builder_get_ptr重命名为scc_ir_builder_get_elem_ptr以更好地反映其功能
- 移除旧的GET_PTR相关枚举和结构体定义
- 更新IR构建器中指针类型的处理逻辑,支持数组元素指针计算
- 在代码生成阶段正确处理元素指针操作的索引计算

fix(ast2ir): 修正数组下标表达式的指针获取操作

- 将数组访问中的get_ptr调用替换为get_elem_ptr
- 确保数组元素访问使用正确的IR指令

perf(runtime): 优化fprintf函数的大字符串处理性能

- 实现动态缓冲区分配策略,小字符串使用栈缓冲区,大字符串使用堆分配
- 避免固定大小缓冲区可能导致的截断问题
- 添加适当的内存清理机制
This commit is contained in:
zzy
2026-04-11 10:12:22 +08:00
parent eeb4c4fc3c
commit 053c6abf51
9 changed files with 106 additions and 63 deletions

View File

@@ -1,5 +1,7 @@
#include <scc_ast2ir.h> #include <scc_ast2ir.h>
static inline void parse_struct_union_layout(scc_ast_type_t *type) {}
static inline void parse_lexme2const_int(const char *lexme, static inline void parse_lexme2const_int(const char *lexme,
scc_ir_const_int_t *value) { scc_ir_const_int_t *value) {
// FIXME // FIXME
@@ -430,7 +432,7 @@ scc_ir_value_ref_t scc_ast2ir_expr(scc_ast2ir_ctx_t *ctx, scc_ast_expr_t *expr,
scc_ast2ir_expr(ctx, expr->subscript.index, false); scc_ast2ir_expr(ctx, expr->subscript.index, false);
// 3. 生成 getptrGEP // 3. 生成 getptrGEP
scc_ir_value_ref_t elem_ptr = scc_ir_value_ref_t elem_ptr =
scc_ir_builder_get_ptr(&ctx->builder, base_ptr, index); scc_ir_builder_get_elem_ptr(&ctx->builder, base_ptr, index);
// 4. 根据左值/右值返回 // 4. 根据左值/右值返回
if (is_lvalue) { if (is_lvalue) {
return elem_ptr; // 作为左值:返回地址 return elem_ptr; // 作为左值:返回地址
@@ -523,8 +525,8 @@ scc_ir_value_ref_t scc_ast2ir_expr(scc_ast2ir_ctx_t *ctx, scc_ast_expr_t *expr,
ctx->builder.ctx.module, ir_type->data.pointer.base); ctx->builder.ctx.module, ir_type->data.pointer.base);
if (target_type->tag == SCC_IR_TYPE_ARRAY) { if (target_type->tag == SCC_IR_TYPE_ARRAY) {
// 生成 getptr 获取数组首地址 // 生成 getptr 获取数组首地址
return scc_ir_builder_get_ptr(&ctx->builder, in, return scc_ir_builder_get_elem_ptr(&ctx->builder, in,
SCC_IR_REF_nullptr); SCC_IR_REF_nullptr);
} else { } else {
// 标量类型:加载值 // 标量类型:加载值
return scc_ir_builder_load(&ctx->builder, in); return scc_ir_builder_load(&ctx->builder, in);

View File

@@ -144,11 +144,12 @@ scc_ir_builder_const_string(scc_ir_builder_t *builder, const char *str,
// scc_hashtable_insert(builder); // scc_hashtable_insert(builder);
scc_ir_value_ref_t pointer_to_global_value = scc_ir_module_add_value( scc_ir_value_ref_t pointer_to_global_value = scc_ir_module_add_value(
builder->ctx.module, &(scc_ir_value_t){ builder->ctx.module,
.tag = SCC_IR_VALUE_TAG_GET_PTR, &(scc_ir_value_t){
.data.get_ptr.src_addr = global_value_ref, .tag = SCC_IR_VALUE_TAG_GET_ELEM_PTR,
.data.get_ptr.index = SCC_IR_VALUE_TAG_NULLPTR, .data.get_elem_ptr.src_addr = global_value_ref,
}); .data.get_elem_ptr.index = SCC_IR_VALUE_TAG_NULLPTR,
});
scc_ir_builder_add_instr(builder, pointer_to_global_value); scc_ir_builder_add_instr(builder, pointer_to_global_value);
return pointer_to_global_value; return pointer_to_global_value;
} }
@@ -236,9 +237,9 @@ scc_ir_value_ref_t scc_ir_builder_store(scc_ir_builder_t *builder,
* @param ptr 基础指针 * @param ptr 基础指针
* @param index 索引值 * @param index 索引值
*/ */
scc_ir_value_ref_t scc_ir_builder_get_ptr(scc_ir_builder_t *builder, scc_ir_value_ref_t scc_ir_builder_get_elem_ptr(scc_ir_builder_t *builder,
scc_ir_value_ref_t ptr, scc_ir_value_ref_t ptr,
scc_ir_value_ref_t index); scc_ir_value_ref_t index);
/** /**
* @brief 创建二元运算指令 * @brief 创建二元运算指令

View File

@@ -100,7 +100,6 @@ typedef enum scc_ir_value_tag {
SCC_IR_VALUE_TAG_GLOBAL_ALLOC, ///< 全局分配 SCC_IR_VALUE_TAG_GLOBAL_ALLOC, ///< 全局分配
SCC_IR_VALUE_TAG_LOAD, ///< 加载数据 SCC_IR_VALUE_TAG_LOAD, ///< 加载数据
SCC_IR_VALUE_TAG_STORE, ///< 存储数据 SCC_IR_VALUE_TAG_STORE, ///< 存储数据
SCC_IR_VALUE_TAG_GET_PTR, ///< 获取指针
SCC_IR_VALUE_TAG_GET_ELEM_PTR, ///< 获取元素指针 SCC_IR_VALUE_TAG_GET_ELEM_PTR, ///< 获取元素指针
SCC_IR_VALUE_TAG_OP, ///< 二元运算 SCC_IR_VALUE_TAG_OP, ///< 二元运算
SCC_IR_VALUE_TAG_BRANCH, ///< 有条件分支 SCC_IR_VALUE_TAG_BRANCH, ///< 有条件分支
@@ -258,10 +257,6 @@ struct scc_ir_value {
scc_ir_value_ref_t target; scc_ir_value_ref_t target;
scc_ir_value_ref_t value; scc_ir_value_ref_t value;
} store; } store;
struct {
scc_ir_value_ref_t src_addr;
scc_ir_value_ref_t index;
} get_ptr;
struct { struct {
scc_ir_value_ref_t src_addr; scc_ir_value_ref_t src_addr;
scc_ir_value_ref_t index; scc_ir_value_ref_t index;

View File

@@ -227,19 +227,33 @@ scc_ir_value_ref_t scc_ir_builder_store(scc_ir_builder_t *builder,
return value_ref; return value_ref;
} }
scc_ir_value_ref_t scc_ir_builder_get_ptr(scc_ir_builder_t *builder, scc_ir_value_ref_t scc_ir_builder_get_elem_ptr(scc_ir_builder_t *builder,
scc_ir_value_ref_t target, scc_ir_value_ref_t target,
scc_ir_value_ref_t index) { scc_ir_value_ref_t index) {
scc_ir_value_t get_ptr_node = {0}; scc_ir_value_t get_ptr_node = {0};
get_ptr_node.tag = SCC_IR_VALUE_TAG_GET_PTR; get_ptr_node.tag = SCC_IR_VALUE_TAG_GET_ELEM_PTR;
get_ptr_node.data.get_ptr.src_addr = target; get_ptr_node.data.get_elem_ptr.src_addr = target;
get_ptr_node.data.get_ptr.index = index; get_ptr_node.data.get_elem_ptr.index = index;
// 类型应与源地址相同(都是指针) // 类型应与源地址相同(都是指针)
scc_ir_value_t *src_node = scc_ir_type_t *type =
scc_ir_module_get_value(GET_MODULE(builder), target); scc_ir_module_get_type_by_value(GET_MODULE(builder), target);
if (src_node) { Assert(type != nullptr);
get_ptr_node.type = src_node->type; if (type->tag == SCC_IR_TYPE_PTR) {
scc_ir_type_t *base_type = scc_ir_module_get_type(
GET_MODULE(builder), type->data.pointer.base);
if (base_type->tag == SCC_IR_TYPE_ARRAY) {
// FIXME GEP maybe multiple levels
get_ptr_node.type = scc_ir_builder_type(
builder, &(scc_ir_type_t){
.tag = SCC_IR_TYPE_PTR,
.data.pointer.base = base_type->data.array.base,
});
} else {
get_ptr_node.type = scc_ir_builder_type(builder, type);
}
} else {
get_ptr_node.type = scc_ir_builder_type(builder, type);
} }
scc_ir_value_ref_t value_ref = scc_ir_value_ref_t value_ref =

View File

@@ -20,7 +20,6 @@ static const char *get_node_type_str(scc_ir_value_tag_t tag) {
[SCC_IR_VALUE_TAG_GLOBAL_ALLOC] = "GlobalAlloc", [SCC_IR_VALUE_TAG_GLOBAL_ALLOC] = "GlobalAlloc",
[SCC_IR_VALUE_TAG_LOAD] = "Load", [SCC_IR_VALUE_TAG_LOAD] = "Load",
[SCC_IR_VALUE_TAG_STORE] = "Store", [SCC_IR_VALUE_TAG_STORE] = "Store",
[SCC_IR_VALUE_TAG_GET_PTR] = "GetPtr",
[SCC_IR_VALUE_TAG_GET_ELEM_PTR] = "GetElemPtr", [SCC_IR_VALUE_TAG_GET_ELEM_PTR] = "GetElemPtr",
[SCC_IR_VALUE_TAG_OP] = "Op", [SCC_IR_VALUE_TAG_OP] = "Op",
[SCC_IR_VALUE_TAG_BRANCH] = "Branch", [SCC_IR_VALUE_TAG_BRANCH] = "Branch",
@@ -117,12 +116,12 @@ static void dump_store_node(scc_ir_dump_ctx_t *ctx,
dump_child_node_ref(ctx, value->data.store.value, true); dump_child_node_ref(ctx, value->data.store.value, true);
} }
static void dump_get_ptr_node(scc_ir_dump_ctx_t *ctx, static void dump_get_elem_ptr_node(scc_ir_dump_ctx_t *ctx,
const scc_ir_value_t *value) { const scc_ir_value_t *value) {
if (value->data.get_ptr.src_addr) if (value->data.get_elem_ptr.src_addr)
dump_child_node_ref(ctx, value->data.get_ptr.src_addr, false); dump_child_node_ref(ctx, value->data.get_elem_ptr.src_addr, false);
if (value->data.get_ptr.index) if (value->data.get_elem_ptr.index)
dump_child_node_ref(ctx, value->data.get_ptr.index, true); dump_child_node_ref(ctx, value->data.get_elem_ptr.index, true);
} }
static void dump_branch_node(scc_ir_dump_ctx_t *ctx, static void dump_branch_node(scc_ir_dump_ctx_t *ctx,
@@ -236,8 +235,8 @@ void scc_ir_dump_value(scc_ir_dump_ctx_t *ctx, scc_ir_value_ref_t value_ref) {
case SCC_IR_VALUE_TAG_STORE: case SCC_IR_VALUE_TAG_STORE:
dump_store_node(ctx, value); dump_store_node(ctx, value);
break; break;
case SCC_IR_VALUE_TAG_GET_PTR: case SCC_IR_VALUE_TAG_GET_ELEM_PTR:
dump_get_ptr_node(ctx, value); dump_get_elem_ptr_node(ctx, value);
break; break;
case SCC_IR_VALUE_TAG_OP: case SCC_IR_VALUE_TAG_OP:
dump_op_node(ctx, value); dump_op_node(ctx, value);
@@ -556,13 +555,10 @@ void scc_ir_dump_value_linear(scc_ir_dump_ctx_t *ctx,
format_ref_or_value(ctx, value->data.store.target); format_ref_or_value(ctx, value->data.store.target);
break; break;
case SCC_IR_VALUE_TAG_GET_ELEM_PTR: case SCC_IR_VALUE_TAG_GET_ELEM_PTR:
case SCC_IR_VALUE_TAG_GET_PTR: scc_tree_dump_append(ctx->dump_ctx, "getelemptr ");
scc_tree_dump_append( format_ref_or_value(ctx, value->data.get_elem_ptr.src_addr);
ctx->dump_ctx,
value->tag == SCC_IR_VALUE_TAG_GET_PTR ? "getptr " : "getelemptr ");
format_ref_or_value(ctx, value->data.get_ptr.src_addr);
scc_tree_dump_append(ctx->dump_ctx, ", "); scc_tree_dump_append(ctx->dump_ctx, ", ");
format_ref_or_value(ctx, value->data.get_ptr.index); format_ref_or_value(ctx, value->data.get_elem_ptr.index);
break; break;
case SCC_IR_VALUE_TAG_OP: case SCC_IR_VALUE_TAG_OP:
format_ref_or_value(ctx, value->data.op.lhs); format_ref_or_value(ctx, value->data.op.lhs);

View File

@@ -82,9 +82,9 @@ void scc_ir_value_init(scc_ir_value_t *in, const char *name,
in->data.store.target = 0; in->data.store.target = 0;
in->data.store.value = 0; in->data.store.value = 0;
break; break;
case SCC_IR_VALUE_TAG_GET_PTR: case SCC_IR_VALUE_TAG_GET_ELEM_PTR:
in->data.get_ptr.src_addr = 0; in->data.get_elem_ptr.src_addr = 0;
in->data.get_ptr.index = 0; in->data.get_elem_ptr.index = 0;
break; break;
case SCC_IR_VALUE_TAG_OP: case SCC_IR_VALUE_TAG_OP:
in->data.op.op = SCC_IR_OP_EMPTY; in->data.op.op = SCC_IR_OP_EMPTY;

View File

@@ -287,14 +287,14 @@ static void parse_value(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
break; break;
} }
///< 获取指针 ///< 获取指针
case SCC_IR_VALUE_TAG_GET_PTR: { case SCC_IR_VALUE_TAG_GET_ELEM_PTR: {
scc_reg_loc_t loc_res; scc_reg_loc_t loc_res;
parse_location(ctx, &loc_res, node_ref); parse_location(ctx, &loc_res, node_ref);
scc_ir_value_t *src_addr = scc_ir_module_get_value( scc_ir_value_t *src_addr = scc_ir_module_get_value(
GET_MODULE(ctx), value->data.get_ptr.src_addr); GET_MODULE(ctx), value->data.get_elem_ptr.src_addr);
Assert(src_addr != nullptr); Assert(src_addr != nullptr);
if (value->data.get_ptr.index == 0) { if (value->data.get_elem_ptr.index == 0) {
if (src_addr->tag == SCC_IR_VALUE_TAG_GLOBAL_ALLOC) { if (src_addr->tag == SCC_IR_VALUE_TAG_GLOBAL_ALLOC) {
// 全局变量RIP相对寻址 // 全局变量RIP相对寻址
scc_mcode_amd64_lea_r64_rip_rel32(&ctx->sect_mcode, scc_mcode_amd64_lea_r64_rip_rel32(&ctx->sect_mcode,
@@ -314,30 +314,35 @@ static void parse_value(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
} else if (src_addr->tag == SCC_IR_VALUE_TAG_ALLOC) { } else if (src_addr->tag == SCC_IR_VALUE_TAG_ALLOC) {
// 栈上变量:地址为 rbp - offset // 栈上变量:地址为 rbp - offset
scc_reg_loc_t src_loc; scc_reg_loc_t src_loc;
parse_location(ctx, &src_loc, value->data.get_ptr.src_addr); parse_location(ctx, &src_loc,
value->data.get_elem_ptr.src_addr);
src_loc.kind = SCC_REG_KIND_STACK_ADDR; src_loc.kind = SCC_REG_KIND_STACK_ADDR;
load_value_to_reg(ctx, &src_loc, SCC_AMD64_RAX); load_value_to_reg(ctx, &src_loc, SCC_AMD64_RAX);
} else { } else {
// 其他情况(如链式 // 其他情况(如链式getptr源地址值已经存储在某个位置直接加载到
// getptr源地址值已经存储在某个位置直接加载到 RAX // RAX
scc_reg_loc_t src_loc; scc_reg_loc_t src_loc;
parse_location(ctx, &src_loc, value->data.get_ptr.src_addr); parse_location(ctx, &src_loc,
value->data.get_elem_ptr.src_addr);
load_value_to_reg(ctx, &src_loc, SCC_AMD64_RAX); load_value_to_reg(ctx, &src_loc, SCC_AMD64_RAX);
} }
store_value_from_reg(ctx, &loc_res, SCC_AMD64_RAX);
} else { } else {
// TODO // TODO
scc_ir_value_t *index = scc_ir_module_get_value( parse_value(ctx, value->data.get_elem_ptr.index, patches);
GET_MODULE(ctx), value->data.get_ptr.index);
Assert(index != nullptr &&
index->tag == SCC_IR_VALUE_TAG_CONST_INT);
}
scc_reg_loc_t src_loc;
scc_reg_loc_t idx_loc;
parse_location(ctx, &src_loc, value->data.get_elem_ptr.src_addr);
parse_location(ctx, &idx_loc, value->data.get_elem_ptr.index);
src_loc.kind = SCC_REG_KIND_STACK_ADDR;
load_value_to_reg(ctx, &src_loc, SCC_AMD64_RAX);
load_value_to_reg(ctx, &idx_loc, SCC_AMD64_RDX);
scc_mcode_amd64_lea_r64_m64_sib(&ctx->sect_mcode, SCC_AMD64_RAX,
SCC_AMD64_RAX, SCC_AMD64_RDX, 1, 0);
}
store_value_from_reg(ctx, &loc_res, SCC_AMD64_RAX);
break; break;
} }
case SCC_IR_VALUE_TAG_GET_ELEM_PTR: ///< 获取元素指针(used by array)
TODO();
///< 二元运算 ///< 二元运算
case SCC_IR_VALUE_TAG_OP: { case SCC_IR_VALUE_TAG_OP: {
scc_reg_loc_t loc_lhs; scc_reg_loc_t loc_lhs;

View File

@@ -49,7 +49,6 @@ void scc_reg_alloc_strategy_pure_stack(scc_reg_alloc_t *ctx,
switch (node->tag) { switch (node->tag) {
case SCC_IR_VALUE_TAG_LOAD: case SCC_IR_VALUE_TAG_LOAD:
case SCC_IR_VALUE_TAG_OP: case SCC_IR_VALUE_TAG_OP:
case SCC_IR_VALUE_TAG_GET_PTR:
case SCC_IR_VALUE_TAG_GET_ELEM_PTR: case SCC_IR_VALUE_TAG_GET_ELEM_PTR:
case SCC_IR_VALUE_TAG_CALL: // 返回值 case SCC_IR_VALUE_TAG_CALL: // 返回值
loc.kind = SCC_REG_KIND_STACK; loc.kind = SCC_REG_KIND_STACK;

View File

@@ -79,8 +79,33 @@ int scc_fprintf(scc_file_t file, const char *format, ...) {
} }
int scc_vfprintf(scc_file_t file, const char *format, va_list args) { int scc_vfprintf(scc_file_t file, const char *format, va_list args) {
char buf[4096] = {0}; va_list args_copy;
int size = vsnprintf_(buf, sizeof(buf), format, args); va_copy(args_copy, args);
// 先计算所需长度
int required_size = vsnprintf_(nullptr, 0, format, args_copy);
va_end(args_copy);
if (required_size < 0)
return -1;
char *buf = nullptr;
int size = 0;
if (required_size < 4096) {
// 小输出使用栈上缓冲区
char stack_buf[4096];
size = vsnprintf_(stack_buf, sizeof(stack_buf), format, args);
buf = stack_buf;
} else {
// 大输出使用堆分配
buf = scc_pal_malloc(required_size + 1);
if (!buf)
return -1;
size = vsnprintf_(buf, required_size + 1, format, args);
}
// 输出处理...
if (file == scc_stdout) { if (file == scc_stdout) {
scc_pal_write(buf, size); scc_pal_write(buf, size);
} else if (file == scc_stderr) { } else if (file == scc_stderr) {
@@ -88,6 +113,12 @@ int scc_vfprintf(scc_file_t file, const char *format, va_list args) {
} else { } else {
scc_pal_fwrite(file, buf, size); scc_pal_fwrite(file, buf, size);
} }
// 如果使用了堆分配,释放内存
if (buf && required_size >= 4096) {
scc_pal_free(buf);
}
return size; return size;
} }