zzy/scc
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat add func call and rewrite codes

Browse Source
This commit is contained in:
ZZY
2025-03-07 12:29:53 +08:00
parent 09299e339c
commit 95bf44eb3f
37 changed files with 3369 additions and 1063 deletions
Download Patch File Download Diff File Expand all files Collapse all files

413
ccompiler/backend/riscv32/rv32ima_codegen.c Normal file
View File

@@ -0,0 +1,413 @@
#define RISCV_VM_BUILDIN_ECALL
#include "rv32gen.h"
#include <stdio.h>
#include <assert.h>
// 指令编码联合体（自动处理小端序）
typedef union rv32code {
uint32_t code;
uint8_t bytes[4];
} rv32code_t;
#define CRT_CODE_SIZE 16
// 使用示例
rv32code_t gcodes[] = {
LI(REG_SP, 0x1000),
LI(REG_RA, 0x0),
CALL_ABS(CRT_CODE_SIZE << 2),
// Exit
ECALL_EXIT2(),
};
void test_raw_gen(FILE* out) {
fwrite(gcodes, sizeof(rv32code_t), sizeof(gcodes)/sizeof(gcodes[0]), out);
}
#include "../../frontend/frontend.h"
#include "../../middleend/ir.h"
typedef struct {
int code_pos;
int to_idx;
int cur_idx;
int base_offset;
enum {
JMP_BRANCH,
JMP_JUMP,
JMP_CALL,
} type;
} jmp_t;
static struct {
vector_header(codes, rv32code_t);
int stack_offset;
int stack_base;
int tmp_reg;
ir_bblock_t* cur_block;
ir_func_t* cur_func;
ir_prog_t* prog;
vector_header(jmp, jmp_t*);
vector_header(call, jmp_t*);
int cur_func_offset;
int cur_block_offset;
} ctx;
int write_inst(union rv32code ins, FILE* fp) {
return fwrite(&ins, sizeof(union rv32code), 1, fp);
}
#define GENCODE(code) vector_push(ctx.codes, (rv32code_t)(code)); len += 4
#define GENCODES(code) do { \
rv32code_t codes[] = { \
code \
}; \
for (int i = 0; i < sizeof(codes) / sizeof(codes[0]); i ++) { \
GENCODE(codes[i]); \
} \
} while (0)
static int stack_offset(ir_node_t* ptr) {
int offset = ctx.stack_base;
for (int i = 0; i < ctx.cur_func->bblocks.size; i ++) {
ir_bblock_t* block = vector_at(ctx.cur_func->bblocks, i);
for (int i = 0; i < block->instrs.size; i++) {
if (vector_at(block->instrs, i) == ptr) {
offset += i * 4;
assert(offset >= 0 && offset < ctx.stack_offset);
return offset;
}
}
offset += block->instrs.size * 4;
}
assert(0);
}
static int block_idx(ir_bblock_t* toblock) {
for (int i = 0; i < ctx.cur_func->bblocks.size; i ++) {
ir_bblock_t* block = vector_at(ctx.cur_func->bblocks, i);
if (toblock == block) {
return i;
}
}
assert(0);
}
static int func_idx(ir_func_t* tofunc) {
for (int i = 0; i < ctx.prog->funcs.size; i ++) {
ir_func_t* func = vector_at(ctx.prog->funcs, i);
if (tofunc == func) {
return i;
}
}
assert(0);
}
static int system_func(const char* name) {
static const char defined_func[][16] = {
"ecall_pnt_int",
};
for (int j = 0; j < sizeof(defined_func)/sizeof(defined_func[0]); j++) {
if (strcmp(name, defined_func[j]) == 0) {
return j;
}
}
return -1;
}
static int get_node_val(ir_node_t* ptr, int reg) {
int len = 0;
if (ptr->tag == IR_NODE_CONST_INT) {
GENCODES(LI(reg, ptr->data.const_int.val));
} else {
int offset = stack_offset(ptr);
GENCODE(LW(reg, REG_SP, offset));
}
return len;
}
static int gen_instr(ir_bblock_t* block, ir_node_t* instr) {
int len = 0;
int offset;
switch (instr->tag) {
case IR_NODE_ALLOC: {
break;
}
case IR_NODE_LOAD: {
// S1 = *(S0 + imm)
offset = stack_offset(instr->data.load.target);
GENCODE(LW(REG_T0, REG_SP, offset));
// offset = STACK_OFFSET(instr);
// GENCODE(SW(REG_T0, REG_SP, offset));
break;
}
case IR_NODE_STORE: {
// *(S0 + imm) = S1
len += get_node_val(instr->data.store.value, REG_T0);
offset = stack_offset(instr->data.store.target);
GENCODE(SW(REG_T0, REG_SP, offset));
break;
}
case IR_NODE_RET: {
// A0 = S0
if (instr->data.ret.ret_val != NULL) {
len += get_node_val(instr->data.ret.ret_val, REG_A0);
}
GENCODE(LW(REG_RA, REG_SP, 0));
GENCODE(ADDI(REG_SP, REG_SP, ctx.stack_offset));
GENCODE(RET());
break;
}
case IR_NODE_OP: {
len += get_node_val(instr->data.op.lhs, REG_T1);
len += get_node_val(instr->data.op.rhs, REG_T2);
switch (instr->data.op.op) {
case IR_OP_ADD:
GENCODE(ADD(REG_T0, REG_T1, REG_T2));
break;
case IR_OP_SUB:
GENCODE(SUB(REG_T0, REG_T1, REG_T2));
break;
case IR_OP_MUL:
GENCODE(MUL(REG_T0, REG_T1, REG_T2));
break;
case IR_OP_DIV:
GENCODE(DIV(REG_T0, REG_T1, REG_T2));
break;
case IR_OP_MOD:
GENCODE(REM(REG_T0, REG_T1, REG_T2));
break;
default:
error("ERROR gen_instr op in riscv");
break;
}
offset = stack_offset(instr);
GENCODE(SW(REG_T0, REG_SP, offset));
break;
}
case IR_NODE_BRANCH: {
len += get_node_val(instr->data.branch.cond, REG_T0);
int tidx = block_idx(instr->data.branch.true_bblock);
int fidx = block_idx(instr->data.branch.false_bblock);
int cidx = block_idx(ctx.cur_block);
jmp_t* jmp;
jmp = xmalloc(sizeof(jmp_t));
*jmp = (jmp_t) {
.base_offset = 8,
.code_pos = ctx.codes.size,
.type = JMP_BRANCH,
.to_idx = tidx,
.cur_idx=cidx,
};
vector_push(ctx.jmp, jmp);
GENCODE(BNEZ(REG_T0, 0));
jmp = xmalloc(sizeof(jmp_t));
*jmp = (jmp_t) {
.base_offset = 4,
.code_pos = ctx.codes.size,
.type = JMP_JUMP,
.to_idx = fidx,
.cur_idx=cidx,
};
vector_push(ctx.jmp, jmp);
GENCODE(J(0));
break;
}
case IR_NODE_JUMP: {
int idx = block_idx(instr->data.jump.target_bblock);
jmp_t* jmp = xmalloc(sizeof(jmp_t));
*jmp = (jmp_t) {
.base_offset = 4,
.code_pos = ctx.codes.size,
.type = JMP_JUMP,
.to_idx = idx,
.cur_idx=block_idx(ctx.cur_block),
};
vector_push(ctx.jmp, jmp);
GENCODE(J(0));
break;
}
case IR_NODE_CALL: {
if (instr->data.call.args.size > 8) {
error("can't add so much params");
}
int param_regs[8] = {
REG_A0, REG_A1, REG_A2, REG_A3,
REG_A4, REG_A5, REG_A6, REG_A7
};
for (int i = 0; i < instr->data.call.args.size; i++) {
ir_node_t* param = vector_at(instr->data.call.args, i);
len += get_node_val(param, param_regs[i]);
}
int system_func_idx = system_func(instr->data.call.callee->name);
if (system_func_idx == 0) {
// ecall_pnt_int
GENCODE(ADDI(REG_A7, REG_X0, 0x1));
GENCODE(ECALL());
break;
}
jmp_t* jmp = xmalloc(sizeof(jmp_t));
*jmp = (jmp_t) {
.base_offset = ctx.cur_func_offset + ctx.cur_block_offset + len,
.code_pos = ctx.codes.size,
.type = JMP_CALL,
.to_idx = func_idx(instr->data.call.callee),
.cur_idx = func_idx(ctx.cur_func),
};
vector_push(ctx.call, jmp);
GENCODES((
CALL(0)
));
break;
}
default:
error("ERROR gen_instr in riscv");
}
return len;
}
static int gen_block(ir_bblock_t* block) {
int len = 0;
ctx.cur_block = block;
for (int i = 0; i < block->instrs.size; i ++) {
ctx.cur_block_offset = len;
len += gen_instr(block, vector_at(block->instrs, i));
}
return len;
}
static int gen_func(ir_func_t* func) {
int len = 0;
ctx.cur_func = func;
ctx.stack_base = 16;
ctx.stack_offset = ctx.stack_base;
for (int i = 0; i < func->bblocks.size; i++) {
ctx.stack_offset += 4 * (*vector_at(func->bblocks, i)).instrs.size;
}
GENCODE(ADDI(REG_SP, REG_SP, -ctx.stack_offset));
GENCODE(SW(REG_RA, REG_SP, 0));
int param_regs[8] = {
REG_A0, REG_A1, REG_A2, REG_A3,
REG_A4, REG_A5, REG_A6, REG_A7
};
if (func->params.size > 8) {
error("can't add so much params");
}
for (int i = 0; i < func->params.size; i++) {
int offset = stack_offset(vector_at(func->params, i));
GENCODE(SW(param_regs[i], REG_SP, offset));
}
int jmp_cache[func->bblocks.size + 1];
if (ctx.jmp.data != NULL) vector_free(ctx.jmp);
vector_init(ctx.jmp);
jmp_cache[0] = 0;
for(int i = 0; i < func->bblocks.size; i ++) {
ctx.cur_func_offset = len;
jmp_cache[i + 1] = jmp_cache[i];
int ret = gen_block(vector_at(func->bblocks, i));
jmp_cache[i + 1] += ret;
len += ret;
}
for (int i = 0; i < ctx.jmp.size; i++) {
jmp_t* jmp = vector_at(ctx.jmp, i);
int32_t code = 0;
int offset = jmp_cache[jmp->to_idx] - (jmp_cache[jmp->cur_idx + 1] - jmp->base_offset);
if (jmp->type == JMP_JUMP) {
code = J(offset);
} else {
code = BNEZ(REG_T0, offset);
}
ctx.codes.data[jmp->code_pos] = (rv32code_t) {
.code = code,
};
}
return len;
}
static void gen_code(ir_prog_t* prog) {
ctx.prog = prog;
for (int i = 0; i < prog->extern_funcs.size; i++) {
if (system_func(prog->extern_funcs.data[i]->name) == -1) {
error("func %s not defined and not a system func", prog->extern_funcs.data[i]->name);
}
}
int len = 0;
int jmp_cache[prog->funcs.size + 1];
for(int i = 0; i < prog->funcs.size; i ++) {
jmp_cache[i + 1] = jmp_cache[i];
int ret = gen_func(vector_at(prog->funcs, i));
jmp_cache[i + 1] += ret;
len += ret;
}
for (int i = 0; i < ctx.call.size; i++) {
jmp_t* jmp = vector_at(ctx.call, i);
int32_t code = 0;
// FIXME ERROR
int offset = jmp_cache[jmp->to_idx] - (jmp_cache[jmp->cur_idx] + jmp->base_offset);
int32_t codes[2] = {
CALL(offset)
};
for (int i = 0; i < 2; i++) {
ctx.codes.data[jmp->code_pos + i] = (rv32code_t) {
.code = codes[i],
};
}
}
}
int main(int argc, char** argv) {
// gcc rv32ima_codegen.c -o rv32gen.exe
const char* infilename = "test.c";
const char* outfilename = "flat.bin";
if (argc >= 2) {
infilename = argv[1];
}
if (argc >= 3) {
outfilename = argv[2];
}
FILE* in = fopen(infilename, "r");
FILE* out = fopen(outfilename, "wb");
if (in == NULL || out == NULL) {
printf("Failed to open file\n");
return 1;
}
struct ASTNode* root = frontend(infilename, in, (sread_fn)fread_s);
gen_ir_from_ast(root);
gen_code(&prog);
for (int i = 0; i < CRT_CODE_SIZE; i++) {
write_inst((union rv32code) {
.code = NOP(),
}, out);
}
fflush(out);
assert(CRT_CODE_SIZE >= sizeof(gcodes) / sizeof(gcodes[0]));
fseek(out, 0, SEEK_SET);
fwrite(gcodes, sizeof(gcodes), 1, out);
fflush(out);
fseek(out, CRT_CODE_SIZE * 4, SEEK_SET);
fwrite(ctx.codes.data, sizeof(ctx.codes.data[0]), ctx.codes.size, out);
fflush(out);
fclose(in);
fclose(out);
// printf("comiler end out: %s\n", outfilename);
return 0;
}