feat(ir): 实现函数调用和参数处理功能

- 在AST定义中移除函数调用结构体中的冗余name字段
- 实现完整的函数声明和定义处理流程,支持符号表查找
- 添加函数参数引用节点类型,支持参数传递和访问
- 实现函数调用的IR生成,包括参数处理和符号解析
- 添加断言确保节点有效性,提升代码健壮性

fix(ast2ir): 优化类型转换处理逻辑

- 移除多余的注释说明
- 简化参数为空检查逻辑,提高代码简洁性
- 修复函数调用时的符号表查找机制

refactor(ir): 改进IR构建器接口设计

- 修改函数构建相关API,使接口更加清晰
- 添加函数声明集合管理
- 重构内置类型缓存机制

feat(ir2mcode): 完善AMD64代码生成

- 实现函数参数到寄存器的映射
- 添加函数调用约定支持(最多4个参数)
- 实现函数符号和重定位处理
- 添加栈帧管理机制
- 修正栈偏移计算

chore(ir): 清理和优化IR dump输出

- 更新节点类型描述信息
- 改进函数声明和定义的输出格式
- 修正格式化输出中的符号显示问题

style: 代码格式化和命名规范化

- 统一重定位类型枚举命名
- 优化函数参数验证和错误处理
This commit is contained in:
zzy
2026-03-23 16:02:23 +08:00
parent 097dbdcc2a
commit 741171dbba
17 changed files with 356 additions and 152 deletions

View File

@@ -7,6 +7,7 @@
typedef enum {
SCC_REG_KIND_UNDEF,
SCC_REG_KIND_FUNC_ARG,
SCC_REG_KIND_GPR, ///< 通用寄存器(整数)
SCC_REG_KIND_FPR, ///< 浮点数寄存器
SCC_REG_KIND_STACK, ///< 栈

View File

@@ -12,6 +12,8 @@ typedef struct {
scc_mcode_t mcode;
sccf_builder_t builder;
// FIXME
usize stack_size;
scc_reg_alloc_t reg_alloc;
scc_hashtable_t *noderef2regloc;
} scc_ir2mcode_ctx_t;

View File

@@ -4,7 +4,7 @@
#include <scc_ir2mcode.h>
static void parse_location(scc_ir2mcode_ctx_t *ctx, scc_reg_loc_t *loc,
scc_ir_bblock_ref_t node_ref) {
scc_ir_node_ref_t node_ref) {
Assert(ctx != null && loc != null);
scc_ir_node_t *node = scc_ir_ctx_get_node(ctx->ir_ctx, node_ref);
if (node == null) {
@@ -27,6 +27,17 @@ static void parse_location(scc_ir2mcode_ctx_t *ctx, scc_reg_loc_t *loc,
case SCC_IR_NODE_CONST_FLOAT:
TODO();
break;
case SCC_IR_NODE_FUNC_ARG_REF: {
scc_ir_type_t *type = scc_ir_ctx_get_type(ctx->ir_ctx, node->type);
Assert(type != 0);
scc_reg_loc_t arg_loc;
// arg_loc.kind = SCC_REG_KIND_FUNC_ARG;
// arg_loc.idx = node->data.arg_ref.idx;
arg_loc.kind = SCC_REG_KIND_STACK;
arg_loc.idx = 8 * node->data.arg_ref.idx;
*loc = arg_loc;
return;
}
default:
idx = (usize)scc_hashtable_get(ctx->noderef2regloc,
(void *)(usize)node_ref);
@@ -47,8 +58,9 @@ static void load_value_to_reg(scc_mcode_t *mcode, scc_reg_loc_t *loc, int reg) {
}
break;
case SCC_REG_KIND_STACK:
// FIXME -8 for rdp
scc_mcode_amd64_mov_r64_m64_disp32(mcode, reg, SCC_AMD64_RBP,
-loc->idx);
-loc->idx - 8);
break;
case SCC_REG_KIND_IMM:
scc_mcode_amd64_mov_r64_imm64(mcode, reg, loc->idx); // 或 imm32
@@ -67,7 +79,8 @@ static void store_value_from_reg(scc_mcode_t *mcode, scc_reg_loc_t *loc,
}
break;
case SCC_REG_KIND_STACK:
scc_mcode_amd64_mov_m64_disp32_r64(mcode, SCC_AMD64_RBP, -loc->idx,
// FIXME -8 for rdp
scc_mcode_amd64_mov_m64_disp32_r64(mcode, SCC_AMD64_RBP, -loc->idx - 8,
reg);
break;
case SCC_REG_KIND_IMM:
@@ -95,13 +108,16 @@ static void parse_node(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
}
switch (node->tag) {
case SCC_IR_NODE_CONV: ///< 类型转换
case SCC_IR_NODE_FUNC_ARG_REF: ///< 函数参数引用
case SCC_IR_NODE_BLOCK_ARG_REF: ///< 基本块参数引用
case SCC_IR_NODE_CONV: ///< 类型转换
LOG_FATAL("Unsupported node type: %d", node->tag);
break;
case SCC_IR_NODE_ALLOC: ///< 分配内存(stack)
case SCC_IR_NODE_GLOBAL_ALLOC: ///< 全局分配(bss)
///< 函数参数引用
case SCC_IR_NODE_FUNC_ARG_REF:
///< ABI
break;
case SCC_IR_NODE_BLOCK_ARG_REF: ///< 基本块参数引用
case SCC_IR_NODE_ALLOC: ///< 分配内存(stack)
case SCC_IR_NODE_GLOBAL_ALLOC: ///< 全局分配(bss)
break;
case SCC_IR_NODE_LOAD: ///< 加载数据
{
@@ -127,7 +143,8 @@ static void parse_node(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
case SCC_IR_NODE_GET_PTR: ///< 获取指针
case SCC_IR_NODE_GET_ELEM_PTR: ///< 获取元素指针(used by array)
TODO();
case SCC_IR_NODE_OP: ///< 二元运算
///< 二元运算
case SCC_IR_NODE_OP: {
scc_reg_loc_t loc_lhs;
parse_location(ctx, &loc_lhs, node->data.op.lhs);
scc_reg_loc_t loc_rhs;
@@ -216,6 +233,7 @@ static void parse_node(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
// 将 RAX 中的结果存储到 res 位置
store_value_from_reg(&ctx->mcode, &loc_res, SCC_AMD64_RAX);
break;
}
///< 有条件分支
case SCC_IR_NODE_BRANCH: {
scc_reg_loc_t loc;
@@ -257,7 +275,7 @@ static void parse_node(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
R9 不稳定的 第四个整型自变量
*/
scc_vec_foreach(node->data.call.args, i) {
parse_location(ctx, &loc, node->data.branch.cond);
parse_location(ctx, &loc, scc_vec_at(node->data.call.args, i));
if (i == 0) {
load_value_to_reg(&ctx->mcode, &loc, SCC_AMD64_RCX);
} else if (i == 1) {
@@ -280,9 +298,22 @@ static void parse_node(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
}
scc_mcode_amd64_call_rel32(&ctx->mcode, 0);
usize sym_idx = sccf_builder_get_symbol_idx(&ctx->builder, func->name);
Assert(sym_idx != 0);
sccf_builder_add_reloc(&ctx->builder,
(sccf_reloc_t){
.reloc_type = SCCF_RELOC_TYPE_REL,
.offset = scc_vec_size(ctx->mcode.mcode) - 4,
.addend = 4,
.sect_type = SCCF_SECT_CODE,
.sym_idx = sym_idx,
});
// 处理返回值
scc_ir_type_t *ret_type = scc_ir_ctx_get_type(ctx->ir_ctx, node->type);
scc_ir_type_t *func_type = scc_ir_ctx_get_type(ctx->ir_ctx, func->type);
Assert(func_type);
scc_ir_type_t *ret_type =
scc_ir_ctx_get_type(ctx->ir_ctx, func_type->data.function.ret_type);
if (ret_type && ret_type->tag != SCC_IR_TYPE_VOID) {
parse_location(ctx, &loc, node_ref);
store_value_from_reg(&ctx->mcode, &loc, SCC_AMD64_RAX);
@@ -296,12 +327,13 @@ static void parse_node(scc_ir2mcode_ctx_t *ctx, scc_ir_bblock_ref_t node_ref,
parse_location(ctx, &loc, node->data.ret.ret_val);
load_value_to_reg(&ctx->mcode, &loc, SCC_AMD64_RAX);
}
scc_mcode_amd64_mov_r64_r64(&ctx->mcode, SCC_AMD64_RSP, SCC_AMD64_RBP);
scc_mcode_amd64_add_rsp_imm32(&ctx->mcode, ctx->stack_size);
scc_mcode_amd64_pop_r64(&ctx->mcode, SCC_AMD64_RBP);
scc_mcode_amd64_ret(&ctx->mcode);
break;
}
default:
LOG_FATAL("unknown node type: %d", node->tag);
UNREACHABLE();
break;
}
@@ -325,7 +357,9 @@ static int equal_func(const void *key1, const void *key2) {
static void parse_function(scc_ir2mcode_ctx_t *ctx, scc_ir_func_t *func) {
ctx->noderef2regloc = scc_reg_alloc(&ctx->reg_alloc, func);
// 对齐到 16 字节
usize stack_size = (ctx->reg_alloc.alloc_stack_size + 15) & ~15;
// FIXME
ctx->stack_size += 8; ///< for rbp
ctx->stack_size = (ctx->reg_alloc.alloc_stack_size + 15) & ~15;
usize bblock_cnt = scc_vec_size(func->bblocks);
usize *bblock_offsets = scc_calloc(bblock_cnt, sizeof(usize));
@@ -341,8 +375,28 @@ static void parse_function(scc_ir2mcode_ctx_t *ctx, scc_ir_func_t *func) {
scc_vec_init(patches);
scc_mcode_amd64_push_r64(&ctx->mcode, SCC_AMD64_RBP);
scc_mcode_amd64_mov_r64_r64(&ctx->mcode, SCC_AMD64_RBP, SCC_AMD64_RSP);
scc_mcode_amd64_sub_rsp_imm32(&ctx->mcode, stack_size);
scc_mcode_amd64_sub_rsp_imm32(&ctx->mcode, ctx->stack_size);
scc_mcode_amd64_lea_r64_m64_disp32(&ctx->mcode, SCC_AMD64_RBP,
SCC_AMD64_RSP, ctx->stack_size);
scc_reg_loc_t loc;
scc_vec_foreach(func->params, i) {
// scc_ir_node_t *param =
// scc_ir_ctx_get_node(ctx->ir_ctx, );
scc_ir_node_ref_t node_ref = scc_vec_at(func->params, i);
parse_location(ctx, &loc, node_ref);
if (i == 0) {
store_value_from_reg(&ctx->mcode, &loc, SCC_AMD64_RCX);
} else if (i == 1) {
store_value_from_reg(&ctx->mcode, &loc, SCC_AMD64_RDX);
} else if (i == 2) {
store_value_from_reg(&ctx->mcode, &loc, SCC_AMD64_R8);
} else if (i == 3) {
store_value_from_reg(&ctx->mcode, &loc, SCC_AMD64_R9);
} else {
LOG_FATAL("not support more than 4 args");
}
// scc_mcode_amd64_push_r64();
}
for (usize i = 0; i < scc_vec_size(func->bblocks); i++) {
scc_ir_bblock_ref_t bblock_ref = scc_vec_at(func->bblocks, i);
@@ -367,12 +421,8 @@ static void parse_function(scc_ir2mcode_ctx_t *ctx, scc_ir_func_t *func) {
usize target_off = bblock_offsets[target_id];
usize next_off = p->pos;
i32 rel = (i32)(target_off - next_off);
// 写入到指令的偏移字段(小端)
// FIXME 写入到指令的偏移字段(小端)
*(u32 *)(&buf[p->pos - 4]) = rel;
// buf[p->pos + 1] = (u8)(rel >> 0);
// buf[p->pos + 2] = (u8)(rel >> 8);
// buf[p->pos + 3] = (u8)(rel >> 16);
// buf[p->pos + 4] = (u8)(rel >> 24);
}
scc_free(bblock_offsets);
scc_vec_free(patches);
@@ -382,6 +432,39 @@ static void parse_function(scc_ir2mcode_ctx_t *ctx, scc_ir_func_t *func) {
void scc_ir2amd64(scc_ir2mcode_ctx_t *ctx) {
scc_reg_alloc_init(&ctx->reg_alloc, scc_reg_alloc_with_stack, ctx->ir_ctx);
scc_vec_foreach(ctx->cprog->func_decls, i) {
scc_ir_node_ref_t func_ref = scc_vec_at(ctx->cprog->func_decls, i);
scc_ir_func_t *func = scc_ir_ctx_get_func(ctx->ir_ctx, func_ref);
if (!func) {
LOG_ERROR("invalid function reference");
return;
}
sccf_sym_t sym = {0};
if (scc_vec_size(func->bblocks)) {
sym = (sccf_sym_t){
.sccf_sect_offset = 0,
.sccf_sect_type = SCCF_SECT_CODE,
.sccf_sym_bind = SCCF_SYM_BIND_GLOBAL,
.sccf_sym_size = 0,
.sccf_sym_type = SCCF_SYM_TYPE_FUNC,
.sccf_sym_vis = SCCF_SYM_VIS_DEFAULT,
};
} else {
sym = (sccf_sym_t){
.sccf_sect_offset = 0,
.sccf_sect_type = SCCF_SECT_NONE,
.sccf_sym_bind = SCCF_SYM_BIND_GLOBAL,
.sccf_sym_size = 0,
.sccf_sym_type = SCCF_SYM_TYPE_EXTERN,
.sccf_sym_vis = SCCF_SYM_VIS_DEFAULT,
};
}
usize sym_idx =
sccf_builder_add_symbol(&ctx->builder, func->name, &sym);
Assert(sym_idx != 0);
}
scc_vec_foreach(ctx->cprog->func_defs, i) {
scc_ir_node_ref_t func_ref = scc_vec_at(ctx->cprog->func_defs, i);
scc_ir_func_t *func = scc_ir_ctx_get_func(ctx->ir_ctx, func_ref);
@@ -389,16 +472,10 @@ void scc_ir2amd64(scc_ir2mcode_ctx_t *ctx) {
LOG_ERROR("invalid function reference");
return;
}
sccf_sym_t sym = {
.sccf_sect_offset = scc_vec_size(ctx->mcode.mcode),
.sccf_sect_type = SCCF_SECT_CODE,
.sccf_sym_bind = SCCF_SYM_BIND_GLOBAL,
.sccf_sym_size = 0,
.sccf_sym_type = SCCF_SYM_TYPE_FUNC,
.sccf_sym_vis = SCCF_SYM_VIS_DEFAULT,
};
sccf_builder_add_symbol(&ctx->builder, func->name, &sym);
sccf_sym_t *sym =
sccf_builder_get_symbol_unsafe(&ctx->builder, func->name);
Assert(sym != null);
sym->sccf_sect_offset = scc_vec_size(ctx->mcode.mcode);
parse_function(ctx, func);
}
sccf_sect_data_t text_section;
@@ -409,4 +486,28 @@ void scc_ir2amd64(scc_ir2mcode_ctx_t *ctx) {
sccf_sect_data_t data_section;
scc_vec_init(data_section);
sccf_builder_add_data_section(&ctx->builder, &data_section);
u8 *buf = scc_vec_unsafe_get_data(ctx->mcode.mcode);
scc_vec_foreach(ctx->builder.relocs, i) {
sccf_reloc_t *reloc = &scc_vec_at(ctx->builder.relocs, i);
if (reloc->sym_idx == 0) {
Panic("relocate to an invalid symbol");
}
sccf_sym_t *sym = &scc_vec_at(ctx->builder.symtab, reloc->sym_idx);
if (sym->sccf_sym_type != SCCF_SYM_TYPE_EXTERN) {
Assert(reloc->reloc_type == SCCF_RELOC_TYPE_REL);
Assert(sym->sccf_sect_type == SCCF_SECT_CODE);
Assert(sym->sccf_sym_type == SCCF_SYM_TYPE_FUNC);
i64 target_off = sym->sccf_sect_offset;
i64 next_off = reloc->offset + reloc->addend;
i32 rel = (i32)(target_off - next_off);
// FIXME 写入到指令的偏移字段(小端)
*(i32 *)(&buf[reloc->offset]) = rel;
reloc->reloc_type = SCCF_RELOC_TYPE_EMPTY;
}
}
// FIXME
ctx->builder.entry_symbol_name = "main";
}

View File

@@ -23,7 +23,17 @@ scc_hashtable_t *scc_reg_alloc_with_stack(scc_reg_alloc_t *ctx,
scc_hashtable_drop(&ctx->node_ref2reg_loc);
scc_vec_free(ctx->reg_loc_vec);
scc_vec_init(ctx->reg_loc_vec);
scc_reg_loc_t loc;
scc_vec_foreach(func->params, i) {
scc_ir_node_ref_t node_ref = scc_vec_at(func->params, i);
loc.kind = SCC_REG_KIND_FUNC_ARG;
loc.idx = i;
ctx->alloc_stack_size += 8;
scc_hashtable_set(&ctx->node_ref2reg_loc, (void *)(usize)node_ref,
(void *)scc_vec_size(ctx->reg_loc_vec));
break;
}
scc_vec_foreach(func->bblocks, i) {
scc_ir_bblock_ref_t bblock_ref = scc_vec_at(func->bblocks, i);
scc_ir_bblock_t *bblock =
@@ -33,8 +43,6 @@ scc_hashtable_t *scc_reg_alloc_with_stack(scc_reg_alloc_t *ctx,
scc_ir_node_ref_t node_ref = scc_vec_at(bblock->instrs, j);
scc_ir_node_t *node = scc_ir_ctx_get_node(ctx->ir_ctx, node_ref);
Assert(node != null);
scc_reg_loc_t loc;
loc.kind = SCC_REG_KIND_UNDEF;
switch (node->tag) {
case SCC_IR_NODE_LOAD:
@@ -51,8 +59,12 @@ scc_hashtable_t *scc_reg_alloc_with_stack(scc_reg_alloc_t *ctx,
break;
}
case SCC_IR_NODE_CALL: {
scc_ir_type_t *ret_type =
scc_ir_ctx_get_type(ctx->ir_ctx, node->type);
// 处理返回值
scc_ir_type_t *func_type =
scc_ir_ctx_get_type(ctx->ir_ctx, func->type);
Assert(func_type);
scc_ir_type_t *ret_type = scc_ir_ctx_get_type(
ctx->ir_ctx, func_type->data.function.ret_type);
if (ret_type && ret_type->tag != SCC_IR_TYPE_VOID) {
loc.kind = SCC_REG_KIND_STACK;
loc.idx = ctx->alloc_stack_size;