const char *asmfn;
static int asmln;
+#define anote(fmt, ...) \
+ printf("%s:%d: note: " fmt, asmfn, asmln, ##__VA_ARGS__)
#define awarn(fmt, ...) \
- printf("warning:%s:%d: " fmt, asmfn, asmln, ##__VA_ARGS__)
+ printf("%s:%d: warning: " fmt, asmfn, asmln, ##__VA_ARGS__)
#define aerr(fmt, ...) do { \
- printf("error:%s:%d: " fmt, asmfn, asmln, ##__VA_ARGS__); \
+ printf("%s:%d: error: " fmt, asmfn, asmln, ##__VA_ARGS__); \
exit(1); \
} while (0)
-enum op_class {
- OPC_UNSPEC,
- OPC_RMD, /* removed or optimized out */
- OPC_DATA, /* data processing */
- OPC_DATA_FLAGS, /* data processing + sets flags */
- OPC_JMP, /* .. and call */
- OPC_JCC, /* conditional jump */
- OPC_SCC, /* conditionel set */
+enum op_flags {
+ OPF_RMD = (1 << 0), /* removed or optimized out */
+ OPF_DATA = (1 << 1), /* data processing - writes to dst opr */
+ OPF_FLAGS = (1 << 2), /* sets flags */
+ OPF_JMP = (1 << 3), /* branches, ret and call */
+ OPF_CC = (1 << 4), /* uses flags */
+ OPF_TAIL = (1 << 5), /* ret or tail call */
};
enum op_op {
OP_MOVZX,
OP_MOVSX,
OP_NOT,
+ OP_CDQ,
OP_RET,
OP_ADD,
OP_SUB,
OP_SHL,
OP_SHR,
OP_SAR,
+ OP_ADC,
OP_SBB,
OP_INC,
OP_DEC,
+ OP_NEG,
OP_MUL,
OP_IMUL,
+ OP_DIV,
+ OP_IDIV,
OP_TEST,
OP_CMP,
OP_CALL,
};
enum opr_type {
- OPT_UNSPEC,
- OPT_REG,
- OPT_REGMEM,
- OPT_LABEL,
+ OPT_UNSPEC,
+ OPT_REG,
+ OPT_REGMEM,
+ OPT_LABEL,
OPT_OFFSET,
- OPT_CONST,
+ OPT_CONST,
};
enum opr_lenmod {
};
struct parsed_op {
- enum op_class cls;
enum op_op op;
struct parsed_opr operand[MAX_OPERANDS];
+ unsigned int flags;
int operand_cnt;
- int regmask; // all referensed regs
+ int regmask_src; // all referensed regs
+ int regmask_dst;
+ int pfomask; // flagop: parsed_flag_op that can't be delayed
+ int argmask; // push: args that are altered before call
+ int cc_scratch; // scratch storage during analysis
+ int bt_i; // branch target (for branches)
+ struct parsed_op *lrl; // label reference list entry
void *datap;
};
+// datap:
+// OP_CALL - ptr to parsed_proto
+// (OPF_CC) - point to corresponding (OPF_FLAGS)
+
struct parsed_equ {
char name[64];
enum opr_lenmod lmod;
static struct parsed_equ *g_eqs;
static int g_eqcnt;
static char g_labels[MAX_OPS][32];
+static struct parsed_op *g_label_refs[MAX_OPS];
static struct parsed_proto g_func_pp;
static char g_func[256];
static char g_comment[256];
enum x86_regs { xUNSPEC = -1, xAX, xBX, xCX, xDX, xSI, xDI, xBP, xSP };
+// possible basic comparison types (without inversion)
+enum parsed_flag_op {
+ PFO_O, // 0 OF=1
+ PFO_C, // 2 CF=1
+ PFO_Z, // 4 ZF=1
+ PFO_BE, // 6 CF=1||ZF=1
+ PFO_S, // 8 SF=1
+ PFO_P, // a PF=1
+ PFO_L, // c SF!=OF
+ PFO_LE, // e ZF=1||SF!=OF
+};
+
+static const char *parsed_flag_op_names[] = {
+ "o", "c", "z", "be", "s", "p", "l", "le"
+};
+
static int char_array_i(const char *array[], size_t len, const char *s)
{
int i;
return -1;
}
-static int parse_reg(int *reg_out, enum opr_lenmod *reg_lmod,
- int *regmask, char *s)
+static void printf_number(char *buf, size_t buf_size, long number)
{
- char w[16];
- int reg = xUNSPEC;
- int c = 0;
+ // output in C-friendly form
+ snprintf(buf, buf_size, number < 10 ? "%lu" : "0x%02lx", number);
+}
- while (*s != 0) {
- while (my_isblank(*s) || my_issep(*s))
- s++;
- s = next_idt(w, sizeof(w), s);
- if (w[0] == 0)
- break;
- c++;
- reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), w);
- if (reg >= 0) {
- *reg_lmod = OPLM_DWORD;
- *regmask |= 1 << reg;
- continue;
- }
- reg = char_array_i(regs_r16, ARRAY_SIZE(regs_r16), w);
- if (reg >= 0) {
- *reg_lmod = OPLM_WORD;
- *regmask |= 1 << reg;
- continue;
- }
- reg = char_array_i(regs_r8h, ARRAY_SIZE(regs_r8h), w);
- if (reg >= 0) {
- *reg_lmod = OPLM_BYTE;
- *regmask |= 1 << reg;
- continue;
- }
- reg = char_array_i(regs_r8l, ARRAY_SIZE(regs_r8l), w);
- if (reg >= 0) {
- *reg_lmod = OPLM_BYTE;
- *regmask |= 1 << reg;
- continue;
- }
+static int parse_reg(enum opr_lenmod *reg_lmod, const char *s)
+{
+ int reg;
- return -1;
+ reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), s);
+ if (reg >= 0) {
+ *reg_lmod = OPLM_DWORD;
+ return reg;
}
-
- if (c == 1) {
- *reg_out = reg;
- return 0;
+ reg = char_array_i(regs_r16, ARRAY_SIZE(regs_r16), s);
+ if (reg >= 0) {
+ *reg_lmod = OPLM_WORD;
+ return reg;
+ }
+ reg = char_array_i(regs_r8h, ARRAY_SIZE(regs_r8h), s);
+ if (reg >= 0) {
+ *reg_lmod = OPLM_BYTE;
+ return reg;
+ }
+ reg = char_array_i(regs_r8l, ARRAY_SIZE(regs_r8l), s);
+ if (reg >= 0) {
+ *reg_lmod = OPLM_BYTE;
+ return reg;
}
return -1;
return ret;
}
+static int parse_indmode(char *name, int *regmask, int need_c_cvt)
+{
+ enum opr_lenmod lmod;
+ char cvtbuf[256];
+ char *d = cvtbuf;
+ char *s = name;
+ char w[64];
+ long number;
+ int reg;
+ int c = 0;
+
+ *d = 0;
+
+ while (*s != 0) {
+ d += strlen(d);
+ while (my_isblank(*s))
+ s++;
+ for (; my_issep(*s); d++, s++)
+ *d = *s;
+ while (my_isblank(*s))
+ s++;
+ *d = 0;
+
+ // skip 'ds:' prefix
+ if (!strncmp(s, "ds:", 3))
+ s += 3;
+
+ s = next_idt(w, sizeof(w), s);
+ if (w[0] == 0)
+ break;
+ c++;
+
+ reg = parse_reg(&lmod, w);
+ if (reg >= 0) {
+ *regmask |= 1 << reg;
+ goto pass;
+ }
+
+ if ('0' <= w[0] && w[0] <= '9') {
+ number = parse_number(w);
+ printf_number(d, sizeof(cvtbuf) - (d - cvtbuf), number);
+ continue;
+ }
+
+ // probably some label/identifier - pass
+
+pass:
+ snprintf(d, sizeof(cvtbuf) - (d - cvtbuf), "%s", w);
+ }
+
+ if (need_c_cvt)
+ strcpy(name, cvtbuf);
+
+ return c;
+}
+
static int guess_lmod_from_name(struct parsed_opr *opr)
{
if (!strncmp(opr->name, "dword_", 6)) {
return 0;
}
-static int parse_operand(struct parsed_opr *opr, int *regmask,
- char words[16][256], int wordc, int w, enum op_class cls)
+static void setup_reg_opr(struct parsed_opr *opr, int reg, enum opr_lenmod lmod,
+ int *regmask)
+{
+ opr->type = OPT_REG;
+ opr->reg = reg;
+ opr->lmod = lmod;
+ *regmask |= 1 << reg;
+}
+
+static struct parsed_equ *equ_find(struct parsed_op *po, const char *name);
+
+static int parse_operand(struct parsed_opr *opr,
+ int *regmask, int *regmask_indirect,
+ char words[16][256], int wordc, int w, unsigned int op_flags)
{
enum opr_lenmod tmplmod;
- int tmpreg;
int ret, len;
+ long number;
int i;
if (w >= wordc)
}
}
- if (cls == OPC_JMP || cls == OPC_JCC) {
+ if (op_flags & OPF_JMP) {
const char *label;
if (wordc - w == 3 && IS(words[w + 1], "ptr"))
ret = sscanf(words[w], "[%[^]]]", opr->name);
if (ret != 1)
aerr("[] parse failure\n");
- // only need the regmask
- parse_reg(&tmpreg, &tmplmod, regmask, opr->name);
+
+ parse_indmode(opr->name, regmask_indirect, 1);
+ if (opr->lmod == OPLM_UNSPEC && !strncmp(opr->name, "ebp+", 4)) {
+ // might be an equ
+ struct parsed_equ *eq = equ_find(NULL, opr->name + 4);
+ if (eq)
+ opr->lmod = eq->lmod;
+ }
return wordc;
}
else if (strchr(words[w], '[')) {
opr->type = OPT_REGMEM;
if (opr->lmod == OPLM_UNSPEC)
guess_lmod_from_name(opr);
- parse_reg(&tmpreg, &tmplmod, regmask, strchr(words[w], '['));
+ parse_indmode(strchr(words[w], '['), regmask_indirect, 0);
return wordc;
}
else if (('0' <= words[w][0] && words[w][0] <= '9')
|| words[w][0] == '-')
{
+ number = parse_number(words[w]);
opr->type = OPT_CONST;
- opr->val = (unsigned int)parse_number(words[w]);
+ opr->val = number;
+ printf_number(opr->name, sizeof(opr->name), number);
return wordc;
}
- ret = parse_reg(&opr->reg, &tmplmod, regmask, opr->name);
- if (ret == 0) {
- opr->type = OPT_REG;
- opr->lmod = tmplmod;
+ ret = parse_reg(&tmplmod, opr->name);
+ if (ret >= 0) {
+ setup_reg_opr(opr, ret, tmplmod, regmask);
return wordc;
}
static const struct {
const char *name;
enum op_op op;
- enum op_class cls;
- int minopr;
- int maxopr;
+ unsigned int minopr;
+ unsigned int maxopr;
+ unsigned int flags;
} op_table[] = {
- { "push", OP_PUSH, OPC_DATA, 1, 1 },
- { "pop", OP_POP, OPC_DATA, 1, 1 },
- { "mov" , OP_MOV, OPC_DATA, 2, 2 },
- { "lea", OP_LEA, OPC_DATA, 2, 2 },
- { "movzx",OP_MOVZX, OPC_DATA, 2, 2 },
- { "movsx",OP_MOVSX, OPC_DATA, 2, 2 },
- { "not", OP_NOT, OPC_DATA, 1, 1 },
- { "add", OP_ADD, OPC_DATA_FLAGS, 2, 2 },
- { "sub", OP_SUB, OPC_DATA_FLAGS, 2, 2 },
- { "and", OP_AND, OPC_DATA_FLAGS, 2, 2 },
- { "or", OP_OR, OPC_DATA_FLAGS, 2, 2 },
- { "xor", OP_XOR, OPC_DATA_FLAGS, 2, 2 },
- { "shl", OP_SHL, OPC_DATA_FLAGS, 2, 2 },
- { "shr", OP_SHR, OPC_DATA_FLAGS, 2, 2 },
- { "sal", OP_SHL, OPC_DATA_FLAGS, 2, 2 },
- { "sar", OP_SAR, OPC_DATA_FLAGS, 2, 2 },
- { "sbb", OP_SBB, OPC_DATA_FLAGS, 2, 2 },
- { "inc", OP_INC, OPC_DATA_FLAGS, 1, 1 },
- { "dec", OP_DEC, OPC_DATA_FLAGS, 1, 1 },
- { "mul", OP_MUL, OPC_DATA_FLAGS, 1, 1 },
- { "imul", OP_IMUL, OPC_DATA_FLAGS, 1, 3 },
- { "test", OP_TEST, OPC_DATA_FLAGS, 2, 2 },
- { "cmp", OP_CMP, OPC_DATA_FLAGS, 2, 2 },
- { "retn", OP_RET, OPC_JMP, 0, 1 },
- { "call", OP_CALL, OPC_JMP, 1, 1 },
- { "jmp", OP_JMP, OPC_JMP, 1, 1 },
- { "jo", OP_JO, OPC_JCC, 1, 1 }, // 70 OF=1
- { "jno", OP_JNO, OPC_JCC, 1, 1 }, // 71 OF=0
- { "jc", OP_JC, OPC_JCC, 1, 1 }, // 72 CF=1
- { "jb", OP_JC, OPC_JCC, 1, 1 }, // 72
- { "jnc", OP_JNC, OPC_JCC, 1, 1 }, // 73 CF=0
- { "jae", OP_JNC, OPC_JCC, 1, 1 }, // 73
- { "jz", OP_JZ, OPC_JCC, 1, 1 }, // 74 ZF=1
- { "je", OP_JZ, OPC_JCC, 1, 1 }, // 74
- { "jnz", OP_JNZ, OPC_JCC, 1, 1 }, // 75 ZF=0
- { "jne", OP_JNZ, OPC_JCC, 1, 1 }, // 75
- { "jbe", OP_JBE, OPC_JCC, 1, 1 }, // 76 CF=1 || ZF=1
- { "jna", OP_JBE, OPC_JCC, 1, 1 }, // 76
- { "ja", OP_JA, OPC_JCC, 1, 1 }, // 77 CF=0 && ZF=0
- { "jnbe", OP_JA, OPC_JCC, 1, 1 }, // 77
- { "js", OP_JS, OPC_JCC, 1, 1 }, // 78 SF=1
- { "jns", OP_JNS, OPC_JCC, 1, 1 }, // 79 SF=0
- { "jp", OP_JP, OPC_JCC, 1, 1 }, // 7a PF=1
- { "jpe", OP_JP, OPC_JCC, 1, 1 }, // 7a
- { "jnp", OP_JNP, OPC_JCC, 1, 1 }, // 7b PF=0
- { "jpo", OP_JNP, OPC_JCC, 1, 1 }, // 7b
- { "jl", OP_JL, OPC_JCC, 1, 1 }, // 7c SF!=OF
- { "jnge", OP_JL, OPC_JCC, 1, 1 }, // 7c
- { "jge", OP_JGE, OPC_JCC, 1, 1 }, // 7d SF=OF
- { "jnl", OP_JGE, OPC_JCC, 1, 1 }, // 7d
- { "jle", OP_JLE, OPC_JCC, 1, 1 }, // 7e ZF=1 || SF!=OF
- { "jng", OP_JLE, OPC_JCC, 1, 1 }, // 7e
- { "jg", OP_JG, OPC_JCC, 1, 1 }, // 7f ZF=0 && SF=OF
- { "jnle", OP_JG, OPC_JCC, 1, 1 }, // 7f
- };
+ { "push", OP_PUSH, 1, 1, 0 },
+ { "pop", OP_POP, 1, 1, OPF_DATA },
+ { "mov" , OP_MOV, 2, 2, OPF_DATA },
+ { "lea", OP_LEA, 2, 2, OPF_DATA },
+ { "movzx",OP_MOVZX, 2, 2, OPF_DATA },
+ { "movsx",OP_MOVSX, 2, 2, OPF_DATA },
+ { "not", OP_NOT, 1, 1, OPF_DATA },
+ { "cdq", OP_CDQ, 0, 0, OPF_DATA },
+ { "add", OP_ADD, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "sub", OP_SUB, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "and", OP_AND, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "or", OP_OR, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "xor", OP_XOR, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "shl", OP_SHL, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "shr", OP_SHR, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "sal", OP_SHL, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "sar", OP_SAR, 2, 2, OPF_DATA|OPF_FLAGS },
+ { "adc", OP_ADC, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC },
+ { "sbb", OP_SBB, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC },
+ { "inc", OP_INC, 1, 1, OPF_DATA|OPF_FLAGS },
+ { "dec", OP_DEC, 1, 1, OPF_DATA|OPF_FLAGS },
+ { "neg", OP_NEG, 1, 1, OPF_DATA|OPF_FLAGS },
+ { "mul", OP_MUL, 1, 1, OPF_DATA|OPF_FLAGS },
+ { "imul", OP_IMUL, 1, 3, OPF_DATA|OPF_FLAGS },
+ { "div", OP_DIV, 1, 1, OPF_DATA|OPF_FLAGS },
+ { "idiv", OP_IDIV, 1, 1, OPF_DATA|OPF_FLAGS },
+ { "test", OP_TEST, 2, 2, OPF_FLAGS },
+ { "cmp", OP_CMP, 2, 2, OPF_FLAGS },
+ { "retn", OP_RET, 0, 1, OPF_JMP|OPF_TAIL },
+ { "call", OP_CALL, 1, 1, OPF_JMP|OPF_FLAGS },
+ { "jmp", OP_JMP, 1, 1, OPF_JMP },
+ { "jo", OP_JO, 1, 1, OPF_JMP|OPF_CC }, // 70 OF=1
+ { "jno", OP_JNO, 1, 1, OPF_JMP|OPF_CC }, // 71 OF=0
+ { "jc", OP_JC, 1, 1, OPF_JMP|OPF_CC }, // 72 CF=1
+ { "jb", OP_JC, 1, 1, OPF_JMP|OPF_CC }, // 72
+ { "jnc", OP_JNC, 1, 1, OPF_JMP|OPF_CC }, // 73 CF=0
+ { "jae", OP_JNC, 1, 1, OPF_JMP|OPF_CC }, // 73
+ { "jz", OP_JZ, 1, 1, OPF_JMP|OPF_CC }, // 74 ZF=1
+ { "je", OP_JZ, 1, 1, OPF_JMP|OPF_CC }, // 74
+ { "jnz", OP_JNZ, 1, 1, OPF_JMP|OPF_CC }, // 75 ZF=0
+ { "jne", OP_JNZ, 1, 1, OPF_JMP|OPF_CC }, // 75
+ { "jbe", OP_JBE, 1, 1, OPF_JMP|OPF_CC }, // 76 CF=1 || ZF=1
+ { "jna", OP_JBE, 1, 1, OPF_JMP|OPF_CC }, // 76
+ { "ja", OP_JA, 1, 1, OPF_JMP|OPF_CC }, // 77 CF=0 && ZF=0
+ { "jnbe", OP_JA, 1, 1, OPF_JMP|OPF_CC }, // 77
+ { "js", OP_JS, 1, 1, OPF_JMP|OPF_CC }, // 78 SF=1
+ { "jns", OP_JNS, 1, 1, OPF_JMP|OPF_CC }, // 79 SF=0
+ { "jp", OP_JP, 1, 1, OPF_JMP|OPF_CC }, // 7a PF=1
+ { "jpe", OP_JP, 1, 1, OPF_JMP|OPF_CC }, // 7a
+ { "jnp", OP_JNP, 1, 1, OPF_JMP|OPF_CC }, // 7b PF=0
+ { "jpo", OP_JNP, 1, 1, OPF_JMP|OPF_CC }, // 7b
+ { "jl", OP_JL, 1, 1, OPF_JMP|OPF_CC }, // 7c SF!=OF
+ { "jnge", OP_JL, 1, 1, OPF_JMP|OPF_CC }, // 7c
+ { "jge", OP_JGE, 1, 1, OPF_JMP|OPF_CC }, // 7d SF=OF
+ { "jnl", OP_JGE, 1, 1, OPF_JMP|OPF_CC }, // 7d
+ { "jle", OP_JLE, 1, 1, OPF_JMP|OPF_CC }, // 7e ZF=1 || SF!=OF
+ { "jng", OP_JLE, 1, 1, OPF_JMP|OPF_CC }, // 7e
+ { "jg", OP_JG, 1, 1, OPF_JMP|OPF_CC }, // 7f ZF=0 && SF=OF
+ { "jnle", OP_JG, 1, 1, OPF_JMP|OPF_CC }, // 7f
+ { "seto", OP_JO, 1, 1, OPF_DATA|OPF_CC },
+ { "setno", OP_JNO, 1, 1, OPF_DATA|OPF_CC },
+ { "setc", OP_JC, 1, 1, OPF_DATA|OPF_CC },
+ { "setb", OP_JC, 1, 1, OPF_DATA|OPF_CC },
+ { "setnc", OP_JNC, 1, 1, OPF_DATA|OPF_CC },
+ { "setae", OP_JNC, 1, 1, OPF_DATA|OPF_CC },
+ { "setz", OP_JZ, 1, 1, OPF_DATA|OPF_CC },
+ { "sete", OP_JZ, 1, 1, OPF_DATA|OPF_CC },
+ { "setnz", OP_JNZ, 1, 1, OPF_DATA|OPF_CC },
+ { "setne", OP_JNZ, 1, 1, OPF_DATA|OPF_CC },
+ { "setbe", OP_JBE, 1, 1, OPF_DATA|OPF_CC },
+ { "setna", OP_JBE, 1, 1, OPF_DATA|OPF_CC },
+ { "seta", OP_JA, 1, 1, OPF_DATA|OPF_CC },
+ { "setnbe", OP_JA, 1, 1, OPF_DATA|OPF_CC },
+ { "sets", OP_JS, 1, 1, OPF_DATA|OPF_CC },
+ { "setns", OP_JNS, 1, 1, OPF_DATA|OPF_CC },
+ { "setp", OP_JP, 1, 1, OPF_DATA|OPF_CC },
+ { "setpe", OP_JP, 1, 1, OPF_DATA|OPF_CC },
+ { "setnp", OP_JNP, 1, 1, OPF_DATA|OPF_CC },
+ { "setpo", OP_JNP, 1, 1, OPF_DATA|OPF_CC },
+ { "setl", OP_JL, 1, 1, OPF_DATA|OPF_CC },
+ { "setnge", OP_JL, 1, 1, OPF_DATA|OPF_CC },
+ { "setge", OP_JGE, 1, 1, OPF_DATA|OPF_CC },
+ { "setnl", OP_JGE, 1, 1, OPF_DATA|OPF_CC },
+ { "setle", OP_JLE, 1, 1, OPF_DATA|OPF_CC },
+ { "setng", OP_JLE, 1, 1, OPF_DATA|OPF_CC },
+ { "setg", OP_JG, 1, 1, OPF_DATA|OPF_CC },
+ { "setnle", OP_JG, 1, 1, OPF_DATA|OPF_CC },
+};
static void parse_op(struct parsed_op *op, char words[16][256], int wordc)
{
+ int regmask_ind;
+ int regmask;
int opr = 0;
int w = 1;
int i;
for (i = 0; i < ARRAY_SIZE(op_table); i++) {
- if (!IS(words[0], op_table[i].name))
- continue;
-
- op->regmask = 0;
-
- for (opr = 0; opr < op_table[i].minopr; opr++) {
- w = parse_operand(&op->operand[opr], &op->regmask,
- words, wordc, w, op_table[i].cls);
- }
+ if (IS(words[0], op_table[i].name))
+ break;
+ }
- for (; w < wordc && opr < op_table[i].maxopr; opr++) {
- w = parse_operand(&op->operand[opr], &op->regmask,
- words, wordc, w, op_table[i].cls);
- }
+ if (i == ARRAY_SIZE(op_table))
+ aerr("unhandled op: '%s'\n", words[0]);
- goto done;
+ op->op = op_table[i].op;
+ op->flags = op_table[i].flags;
+ op->regmask_src = op->regmask_dst = 0;
+
+ for (opr = 0; opr < op_table[i].minopr; opr++) {
+ regmask = regmask_ind = 0;
+ w = parse_operand(&op->operand[opr], ®mask, ®mask_ind,
+ words, wordc, w, op->flags);
+
+ if (opr == 0 && (op->flags & OPF_DATA))
+ op->regmask_dst = regmask;
+ // for now, mark dst as src too
+ op->regmask_src |= regmask | regmask_ind;
}
- aerr("unhandled op: '%s'\n", words[0]);
+ for (; w < wordc && opr < op_table[i].maxopr; opr++) {
+ w = parse_operand(&op->operand[opr],
+ &op->regmask_src, &op->regmask_src,
+ words, wordc, w, op->flags);
+ }
-done:
if (w < wordc)
aerr("parse_op %s incomplete: %d/%d\n",
words[0], w, wordc);
- op->cls = op_table[i].cls;
- op->op = op_table[i].op;
+ // special cases
op->operand_cnt = opr;
- return;
+ if (!strncmp(op_table[i].name, "set", 3))
+ op->operand[0].lmod = OPLM_BYTE;
+
+ // ops with implicit argumets
+ switch (op->op) {
+ case OP_CDQ:
+ op->operand_cnt = 2;
+ setup_reg_opr(&op->operand[0], xDX, OPLM_DWORD, &op->regmask_dst);
+ setup_reg_opr(&op->operand[1], xAX, OPLM_DWORD, &op->regmask_src);
+ break;
+
+ case OP_IMUL:
+ if (op->operand_cnt != 1)
+ break;
+ // fallthrough
+ case OP_MUL:
+ // singleop mul
+ op->regmask_dst = (1 << xDX) | (1 << xAX);
+ op->regmask_src |= (1 << xAX);
+ if (op->operand[0].lmod == OPLM_UNSPEC)
+ op->operand[0].lmod = OPLM_DWORD;
+ break;
+
+ case OP_DIV:
+ case OP_IDIV:
+ // we could set up operands for edx:eax, but there is no real need to
+ // (see is_opr_modified())
+ regmask = (1 << xDX) | (1 << xAX);
+ op->regmask_dst = regmask;
+ op->regmask_src |= regmask;
+ if (op->operand[0].lmod == OPLM_UNSPEC)
+ op->operand[0].lmod = OPLM_DWORD;
+ break;
+
+ case OP_SHL:
+ case OP_SHR:
+ case OP_SAR:
+ if (op->operand[1].lmod == OPLM_UNSPEC)
+ op->operand[1].lmod = OPLM_BYTE;
+ break;
+
+ default:
+ break;
+ }
}
static const char *op_name(enum op_op op)
for (i = 0; i < g_eqcnt; i++)
if (IS(g_eqs[i].name, name))
break;
- if (i >= g_eqcnt)
- ferr(po, "unresolved equ name: '%s'\n", name);
+ if (i >= g_eqcnt) {
+ if (po != NULL)
+ ferr(po, "unresolved equ name: '%s'\n", name);
+ return NULL;
+ }
return &g_eqs[i];
}
snprintf(buf, buf_size, "(u16)%s", opr_reg_p(po, popr));
break;
case OPLM_BYTE:
- snprintf(buf, buf_size, "(u8)%s", opr_reg_p(po, popr));
+ if (popr->name[1] == 'h') // XXX..
+ snprintf(buf, buf_size, "(u8)(%s >> 8)", opr_reg_p(po, popr));
+ else
+ snprintf(buf, buf_size, "(u8)%s", opr_reg_p(po, popr));
break;
default:
ferr(po, "invalid src lmod: %d\n", popr->lmod);
if (is_lea)
snprintf(buf, buf_size, "(u32)&%s", popr->name);
else
- snprintf(buf, buf_size, "%s", popr->name);
+ snprintf(buf, buf_size, "(u32)%s", popr->name);
break;
case OPT_OFFSET:
if (is_lea)
ferr(po, "lea from const?\n");
- snprintf(buf, buf_size, popr->val < 10 ? "%u" : "0x%02x", popr->val);
+ printf_number(buf, buf_size, popr->val);
break;
default:
break;
case OPLM_BYTE:
// ugh..
- snprintf(buf, buf_size, "LOBYTE(%s)", opr_reg_p(po, popr));
+ if (popr->name[1] == 'h') // XXX..
+ snprintf(buf, buf_size, "BYTE1(%s)", opr_reg_p(po, popr));
+ else
+ snprintf(buf, buf_size, "LOBYTE(%s)", opr_reg_p(po, popr));
break;
default:
ferr(po, "invalid dst lmod: %d\n", popr->lmod);
return buf;
}
-static void split_cond(struct parsed_op *po, enum op_op *op, int *is_neg)
+static const char *lmod_cast_u(struct parsed_op *po,
+ enum opr_lenmod lmod)
+{
+ switch (lmod) {
+ case OPLM_DWORD:
+ return "";
+ case OPLM_WORD:
+ return "(u16)";
+ case OPLM_BYTE:
+ return "(u8)";
+ default:
+ ferr(po, "invalid lmod: %d\n", lmod);
+ return "(_invalid_)";
+ }
+}
+
+static const char *lmod_cast_s(struct parsed_op *po,
+ enum opr_lenmod lmod)
+{
+ switch (lmod) {
+ case OPLM_DWORD:
+ return "(s32)";
+ case OPLM_WORD:
+ return "(s16)";
+ case OPLM_BYTE:
+ return "(s8)";
+ default:
+ ferr(po, "invalid lmod: %d\n", lmod);
+ return "(_invalid_)";
+ }
+}
+
+static const char *lmod_cast(struct parsed_op *po,
+ enum opr_lenmod lmod, int is_signed)
+{
+ return is_signed ?
+ lmod_cast_s(po, lmod) :
+ lmod_cast_u(po, lmod);
+}
+
+static enum parsed_flag_op split_cond(struct parsed_op *po,
+ enum op_op op, int *is_inv)
{
- *is_neg = 0;
+ *is_inv = 0;
+
+ switch (op) {
+ case OP_JO:
+ return PFO_O;
+ case OP_JC:
+ return PFO_C;
+ case OP_JZ:
+ return PFO_Z;
+ case OP_JBE:
+ return PFO_BE;
+ case OP_JS:
+ return PFO_S;
+ case OP_JP:
+ return PFO_P;
+ case OP_JL:
+ return PFO_L;
+ case OP_JLE:
+ return PFO_LE;
- switch (*op) {
case OP_JNO:
- *op = OP_JO;
- *is_neg = 1;
- break;
+ *is_inv = 1;
+ return PFO_O;
case OP_JNC:
- *op = OP_JC;
- *is_neg = 1;
- break;
+ *is_inv = 1;
+ return PFO_C;
case OP_JNZ:
- *op = OP_JZ;
- *is_neg = 1;
- break;
+ *is_inv = 1;
+ return PFO_Z;
+ case OP_JA:
+ *is_inv = 1;
+ return PFO_BE;
case OP_JNS:
- *op = OP_JS;
- *is_neg = 1;
- break;
+ *is_inv = 1;
+ return PFO_S;
case OP_JNP:
- *op = OP_JP;
- *is_neg = 1;
- break;
- case OP_JLE:
- *op = OP_JG;
- *is_neg = 1;
- break;
+ *is_inv = 1;
+ return PFO_P;
case OP_JGE:
- *op = OP_JL;
- *is_neg = 1;
- break;
- case OP_JA:
- *op = OP_JBE;
- *is_neg = 1;
- break;
- case OP_JO:
- case OP_JC:
- case OP_JZ:
- case OP_JS:
- case OP_JP:
+ *is_inv = 1;
+ return PFO_L;
case OP_JG:
- case OP_JL:
- case OP_JBE:
- break;
- //break;
+ *is_inv = 1;
+ return PFO_LE;
+
+ case OP_ADC:
+ case OP_SBB:
+ return PFO_C;
+
default:
- ferr(po, "split_cond: bad op %d\n", *op);
- break;
+ ferr(po, "split_cond: bad op %d\n", op);
+ return -1;
}
}
static void out_test_for_cc(char *buf, size_t buf_size,
- struct parsed_op *po, enum op_op op, enum opr_lenmod lmod,
- const char *expr)
+ struct parsed_op *po, enum parsed_flag_op pfo, int is_inv,
+ enum opr_lenmod lmod, const char *expr)
{
- const char *cast = "";
- int is_neg = 0;
+ const char *cast, *scast;
- split_cond(po, &op, &is_neg);
- switch (op) {
- case OP_JZ:
- switch (lmod) {
- case OPLM_DWORD:
- break;
- case OPLM_WORD:
- cast = "(u16)";
- break;
- case OPLM_BYTE:
- cast = "(u8)";
- break;
- default:
- ferr(po, "%s: invalid lmod for JZ: %d\n", __func__, lmod);
- }
+ cast = lmod_cast_u(po, lmod);
+ scast = lmod_cast_s(po, lmod);
+
+ switch (pfo) {
+ case PFO_Z:
+ case PFO_BE: // CF=1||ZF=1; CF=0
snprintf(buf, buf_size, "(%s%s %s 0)",
- cast, expr, is_neg ? "!=" : "==");
+ cast, expr, is_inv ? "!=" : "==");
break;
- case OP_JG: // ZF=0 && SF=OF; OF=0 after test
- switch (lmod) {
- case OPLM_DWORD:
- snprintf(buf, buf_size, "((int)%s %s 0)", expr, is_neg ? "<=" : ">");
- break;
- default:
- ferr(po, "%s: unhandled lmod for JG: %d\n", __func__, lmod);
- }
+ case PFO_S:
+ case PFO_L: // SF!=OF; OF=0
+ snprintf(buf, buf_size, "(%s%s %s 0)",
+ scast, expr, is_inv ? ">=" : "<");
+ break;
+
+ case PFO_LE: // ZF=1||SF!=OF; OF=0
+ snprintf(buf, buf_size, "(%s%s %s 0)",
+ scast, expr, is_inv ? ">" : "<=");
break;
default:
- ferr(po, "%s: unhandled op: %d\n", __func__, op);
+ ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo);
}
}
static void out_cmp_for_cc(char *buf, size_t buf_size,
- struct parsed_op *po, enum op_op op, enum opr_lenmod lmod,
- const char *expr1, const char *expr2)
+ struct parsed_op *po, enum parsed_flag_op pfo, int is_inv,
+ enum opr_lenmod lmod, const char *expr1, const char *expr2)
{
- const char *cast = "";
- const char *scast = "";
- int is_neg = 0;
+ const char *cast, *scast;
- switch (lmod) {
- case OPLM_DWORD:
- scast = "(s32)";
- break;
- case OPLM_WORD:
- cast = "(u16)";
- scast = "(s16)";
- break;
- case OPLM_BYTE:
- cast = "(u8)";
- scast = "(s8)";
+ cast = lmod_cast_u(po, lmod);
+ scast = lmod_cast_s(po, lmod);
+
+ switch (pfo) {
+ case PFO_C:
+ // note: must be unsigned compare
+ snprintf(buf, buf_size, "(%s%s %s %s%s)",
+ cast, expr1, is_inv ? ">=" : "<", cast, expr2);
break;
- default:
- ferr(po, "%s: invalid lmod: %d\n", __func__, lmod);
- }
- split_cond(po, &op, &is_neg);
- switch (op) {
- case OP_JZ:
+ case PFO_Z:
snprintf(buf, buf_size, "(%s%s %s %s%s)",
- cast, expr1, is_neg ? "!=" : "==", cast, expr2);
+ cast, expr1, is_inv ? "!=" : "==", cast, expr2);
break;
- case OP_JC:
+ case PFO_BE: // !a
// note: must be unsigned compare
snprintf(buf, buf_size, "(%s%s %s %s%s)",
- cast, expr1, is_neg ? ">=" : "<", cast, expr2);
+ cast, expr1, is_inv ? ">" : "<=", cast, expr2);
break;
- case OP_JL:
- // note: must be unsigned compare
+ // note: must be signed compare
+ case PFO_S:
+ snprintf(buf, buf_size, "(%s(%s - %s) %s 0)",
+ scast, expr1, expr2, is_inv ? ">=" : "<");
+ break;
+
+ case PFO_L: // !ge
+ snprintf(buf, buf_size, "(%s%s %s %s%s)",
+ scast, expr1, is_inv ? ">=" : "<", scast, expr2);
+ break;
+
+ case PFO_LE:
snprintf(buf, buf_size, "(%s%s %s %s%s)",
- scast, expr1, is_neg ? ">=" : "<", scast, expr2);
+ scast, expr1, is_inv ? ">" : "<=", scast, expr2);
break;
default:
- ferr(po, "%s: unhandled op: %d\n", __func__, op);
+ ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo);
}
}
-static void propagate_lmod(struct parsed_op *po, struct parsed_opr *popr1,
- struct parsed_opr *popr2)
+static void out_cmp_test(char *buf, size_t buf_size,
+ struct parsed_op *po, enum parsed_flag_op pfo, int is_inv)
{
- struct parsed_equ *eq;
+ char buf1[256], buf2[256], buf3[256];
- if (popr1->lmod == OPLM_UNSPEC && popr2->lmod == OPLM_UNSPEC) {
- // lmod could be specified by equ..
- if (!strncmp(popr1->name, "ebp+", 4)) {
- eq = equ_find(po, popr1->name + 4);
- popr1->lmod = eq->lmod;
+ if (po->op == OP_TEST) {
+ if (IS(opr_name(po, 0), opr_name(po, 1))) {
+ out_src_opr(buf3, sizeof(buf3), po, &po->operand[0], 0);
}
- if (!strncmp(popr2->name, "ebp+", 4)) {
- eq = equ_find(po, popr2->name + 4);
- popr2->lmod = eq->lmod;
+ else {
+ out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], 0);
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 0);
+ snprintf(buf3, sizeof(buf3), "(%s & %s)", buf1, buf2);
}
+ out_test_for_cc(buf, buf_size, po, pfo, is_inv,
+ po->operand[0].lmod, buf3);
}
+ else if (po->op == OP_CMP) {
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[0], 0);
+ out_src_opr(buf3, sizeof(buf3), po, &po->operand[1], 0);
+ out_cmp_for_cc(buf, buf_size, po, pfo, is_inv,
+ po->operand[0].lmod, buf2, buf3);
+ }
+ else
+ ferr(po, "%s: unhandled op: %d\n", __func__, po->op);
+}
+static void propagate_lmod(struct parsed_op *po, struct parsed_opr *popr1,
+ struct parsed_opr *popr2)
+{
if (popr1->lmod == OPLM_UNSPEC && popr2->lmod == OPLM_UNSPEC)
ferr(po, "missing lmod for both operands\n");
switch (po->op)
{
case OP_ADD:
+ case OP_ADC:
return "+";
case OP_SUB:
+ case OP_SBB:
return "-";
case OP_AND:
return "&";
}
}
-static int scan_for_pop(int i, int opcnt, const char *reg)
+static int scan_for_pop(int i, int opcnt, const char *reg,
+ int magic, int do_patch)
{
+ struct parsed_op *po;
+ int ret = 0;
+
for (; i < opcnt; i++) {
- if (ops[i].cls == OPC_RMD)
+ po = &ops[i];
+ if (po->cc_scratch == magic)
+ break; // already checked
+ po->cc_scratch = magic;
+
+ if (po->flags & OPF_TAIL)
+ return -1; // deadend
+
+ if (po->flags & OPF_RMD)
continue;
- if (ops[i].cls == OPC_JMP || ops[i].cls == OPC_JCC
- || g_labels[i][0] != 0)
- return -1;
+ if ((po->flags & OPF_JMP) && po->op != OP_CALL) {
+ if (po->bt_i < 0) {
+ ferr(po, "dead branch\n");
+ return -1;
+ }
- if (ops[i].op == OP_POP && ops[i].operand[0].type == OPT_REG
- && IS(ops[i].operand[0].name, reg))
- return i;
+ if (po->flags & OPF_CC) {
+ ret |= scan_for_pop(po->bt_i, opcnt, reg, magic, do_patch);
+ if (ret < 0)
+ return ret; // dead end
+ }
+ else {
+ i = po->bt_i - 1;
+ }
+ continue;
+ }
+
+ if (po->op == OP_POP && po->operand[0].type == OPT_REG
+ && IS(po->operand[0].name, reg))
+ {
+ if (do_patch)
+ po->flags |= OPF_RMD;
+ return 1;
+ }
}
- return -1;
+ return ret;
}
-// scan for pop starting from ret
+// scan for pop starting from 'ret' op (all paths)
static int scan_for_pop_ret(int i, int opcnt, const char *reg, int do_patch)
{
int found = 0;
int j;
for (; i < opcnt; i++) {
- if (ops[i].op != OP_RET)
+ if (!(ops[i].flags & OPF_TAIL))
continue;
for (j = i - 1; j >= 0; j--) {
- if (ops[j].cls == OPC_JMP || ops[j].cls == OPC_JCC)
+ if (ops[j].flags & OPF_RMD)
+ continue;
+ if (ops[j].flags & OPF_JMP)
return -1;
if (ops[j].op == OP_POP && ops[j].operand[0].type == OPT_REG
{
found = 1;
if (do_patch)
- ops[j].cls = OPC_RMD;
+ ops[j].flags |= OPF_RMD;
break;
}
return found ? 0 : -1;
}
+// is operand 'opr modified' by parsed_op 'po'?
+static int is_opr_modified(const struct parsed_opr *opr,
+ const struct parsed_op *po)
+{
+ if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA))
+ return 0;
+
+ if (opr->type == OPT_REG && po->operand[0].type == OPT_REG) {
+ if (po->regmask_dst & (1 << opr->reg))
+ return 1;
+ else
+ return 0;
+ }
+
+ return IS(po->operand[0].name, opr->name);
+}
+
+// is any operand of parsed_op 'po_test' modified by parsed_op 'po'?
+static int is_any_opr_modified(const struct parsed_op *po_test,
+ const struct parsed_op *po)
+{
+ int i;
+
+ if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA))
+ return 0;
+
+ if (po_test->regmask_src & po->regmask_dst)
+ return 1;
+
+ for (i = 0; i < po_test->operand_cnt; i++)
+ if (IS(po_test->operand[i].name, po->operand[0].name))
+ return 1;
+
+ return 0;
+}
+
+// scan for any po_test operand modification in range given
+static int scan_for_mod(struct parsed_op *po_test, int i, int opcnt)
+{
+ for (; i < opcnt; i++) {
+ if (is_any_opr_modified(po_test, &ops[i]))
+ return i;
+ }
+
+ return -1;
+}
+
+// scan for po_test operand[0] modification in range given
+static int scan_for_mod_opr0(struct parsed_op *po_test,
+ int i, int opcnt)
+{
+ for (; i < opcnt; i++) {
+ if (is_opr_modified(&po_test->operand[0], &ops[i]))
+ return i;
+ }
+
+ return -1;
+}
+
+static int scan_for_flag_set(int i)
+{
+ for (; i >= 0; i--) {
+ if (ops[i].flags & OPF_FLAGS)
+ return i;
+
+ if ((ops[i].flags & OPF_JMP) && !(ops[i].flags & OPF_CC))
+ return -1;
+ if (g_labels[i][0] != 0)
+ return -1;
+ }
+
+ return -1;
+}
+
+// scan back for cdq, if anything modifies edx, fail
+static int scan_for_cdq_edx(int i)
+{
+ for (; i >= 0; i--) {
+ if (ops[i].op == OP_CDQ)
+ return i;
+
+ if (ops[i].regmask_dst & (1 << xDX))
+ return -1;
+ if (g_labels[i][0] != 0)
+ return -1;
+ }
+
+ return -1;
+}
+
static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
{
- struct parsed_op *po, *delayed_op = NULL, *tmp_op;
+ struct parsed_op *po, *delayed_flag_op = NULL, *tmp_op;
struct parsed_opr *last_arith_dst = NULL;
char buf1[256], buf2[256], buf3[256];
- struct parsed_proto *pp;
+ struct parsed_proto *pp, *pp_tmp;
const char *tmpname;
+ enum parsed_flag_op pfo;
+ int save_arg_vars = 0;
+ int cmp_result_vars = 0;
+ int need_mul_var = 0;
int had_decl = 0;
int regmask_arg = 0;
int regmask = 0;
- int special_sbb;
+ int pfomask = 0;
int no_output;
+ int dummy;
int arg;
int i, j;
int reg;
&& IS(opr_name(&ops[1], 0), "ebp")
&& IS(opr_name(&ops[1], 1), "esp"))
{
+ int ecx_push = 0;
+
g_bp_frame = 1;
- ops[0].cls = OPC_RMD;
- ops[1].cls = OPC_RMD;
+ ops[0].flags |= OPF_RMD;
+ ops[1].flags |= OPF_RMD;
if (ops[2].op == OP_SUB && IS(opr_name(&ops[2], 0), "esp")) {
g_bp_stack = opr_const(&ops[2], 1);
- ops[2].cls = OPC_RMD;
+ ops[2].flags |= OPF_RMD;
+ }
+ else {
+ // another way msvc builds stack frame..
+ i = 2;
+ while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
+ g_bp_stack += 4;
+ ops[i].flags |= OPF_RMD;
+ ecx_push++;
+ i++;
+ }
}
i = 2;
break;
if (ops[i - 1].op != OP_POP || !IS(opr_name(&ops[i - 1], 0), "ebp"))
ferr(&ops[i - 1], "'pop ebp' expected\n");
- ops[i - 1].cls = OPC_RMD;
+ ops[i - 1].flags |= OPF_RMD;
if (g_bp_stack != 0) {
if (ops[i - 2].op != OP_MOV
{
ferr(&ops[i - 2], "esp restore expected\n");
}
- ops[i - 2].cls = OPC_RMD;
+ ops[i - 2].flags |= OPF_RMD;
+
+ if (ecx_push && ops[i - 3].op == OP_POP
+ && IS(opr_name(&ops[i - 3], 0), "ecx"))
+ {
+ ferr(&ops[i - 3], "unexpected ecx pop\n");
+ }
}
+
i++;
} while (i < opcnt);
}
// pass2:
+ // - resolve all branches
+ for (i = 0; i < opcnt; i++) {
+ po = &ops[i];
+ po->bt_i = -1;
+
+ if ((po->flags & OPF_RMD) || !(po->flags & OPF_JMP)
+ || po->op == OP_CALL || po->op == OP_RET)
+ continue;
+
+ for (j = 0; j < opcnt; j++) {
+ if (g_labels[j][0] && IS(po->operand[0].name, g_labels[j])) {
+ po->bt_i = j;
+ po->lrl = g_label_refs[j];
+ g_label_refs[j] = po;
+ break;
+ }
+ }
+
+ if (po->bt_i == -1) {
+ // assume tail call
+ po->op = OP_CALL;
+ po->flags |= OPF_TAIL;
+ }
+ }
+
+ // pass3:
+ // - find POPs for PUSHes, rm both
// - scan for all used registers
+ // - find flag set ops for their users
// - process calls
- // - find POPs for PUSHes, rm both
for (i = 0; i < opcnt; i++) {
- if (ops[i].cls == OPC_RMD)
+ po = &ops[i];
+ if (po->flags & OPF_RMD)
continue;
- if (ops[i].op == OP_PUSH && ops[i].operand[0].type == OPT_REG) {
- if (ops[i].operand[0].reg < 0)
- ferr(&ops[i], "reg not set for push?\n");
- if (!(regmask & (1 << ops[i].operand[0].reg))) { // reg save
- ret = scan_for_pop(i + 1, opcnt, ops[i].operand[0].name);
- if (ret >= 0) {
- ops[i].cls = ops[ret].cls = OPC_RMD;
+ if (po->op == OP_PUSH && po->operand[0].type == OPT_REG) {
+ if (po->operand[0].reg < 0)
+ ferr(po, "reg not set for push?\n");
+ if (!(regmask & (1 << po->operand[0].reg))) { // reg save
+ ret = scan_for_pop(i + 1, opcnt, po->operand[0].name, i + 1, 0);
+ if (ret == 1) {
+ po->flags |= OPF_RMD;
+ scan_for_pop(i + 1, opcnt, po->operand[0].name, i + 2, 1);
continue;
}
- ret = scan_for_pop_ret(i + 1, opcnt, ops[i].operand[0].name, 0);
+ ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, 0);
if (ret == 0) {
- ops[i].cls = OPC_RMD;
- scan_for_pop_ret(i + 1, opcnt, ops[i].operand[0].name, 1);
+ po->flags |= OPF_RMD;
+ scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, 1);
continue;
}
}
}
- regmask |= ops[i].regmask;
+ regmask |= po->regmask_src | po->regmask_dst;
- if (ops[i].op == OP_CALL) {
+ if (po->flags & OPF_CC)
+ {
+ ret = scan_for_flag_set(i - 1);
+ if (ret < 0)
+ ferr(po, "unable to trace flag setter\n");
+
+ tmp_op = &ops[ret]; // flag setter
+ pfo = split_cond(po, po->op, &dummy);
+ pfomask = 0;
+
+ // to get nicer code, we try to delay test and cmp;
+ // if we can't because of operand modification, or if we
+ // have math op, make it calculate flags explicitly
+ if (tmp_op->op == OP_TEST || tmp_op->op == OP_CMP) {
+ if (scan_for_mod(tmp_op, ret + 1, i) >= 0)
+ pfomask = 1 << pfo;
+ }
+ else {
+ if ((pfo != PFO_Z && pfo != PFO_S && pfo != PFO_P)
+ || scan_for_mod_opr0(tmp_op, ret + 1, i) >= 0)
+ pfomask = 1 << pfo;
+ }
+ if (pfomask) {
+ tmp_op->pfomask |= pfomask;
+ cmp_result_vars |= pfomask;
+ po->datap = tmp_op;
+ }
+
+ if (po->op == OP_ADC || po->op == OP_SBB)
+ cmp_result_vars |= 1 << PFO_C;
+ }
+ else if (po->op == OP_CALL)
+ {
pp = malloc(sizeof(*pp));
my_assert_not(pp, NULL);
tmpname = opr_name(&ops[i], 0);
ret = proto_parse(fhdr, tmpname, pp);
if (ret)
- ferr(&ops[i], "proto_parse failed for '%s'\n", tmpname);
+ ferr(po, "proto_parse failed for '%s'\n", tmpname);
for (arg = 0; arg < pp->argc; arg++)
if (pp->arg[arg].reg == NULL)
break;
for (j = i - 1; j >= 0 && arg < pp->argc; j--) {
- if (ops[j].cls == OPC_RMD)
- continue;
- if (ops[j].op != OP_PUSH)
- continue;
+ if (ops[j].op == OP_CALL) {
+ pp_tmp = ops[j].datap;
+ if (pp_tmp == NULL)
+ ferr(po, "arg collect hit unparsed call\n");
+ if (pp_tmp->argc_stack > 0)
+ ferr(po, "arg collect hit '%s' with %d stack args\n",
+ opr_name(&ops[j], 0), pp_tmp->argc_stack);
+ }
+ else if ((ops[j].flags & OPF_TAIL)
+ || (ops[j].flags & (OPF_JMP|OPF_CC)) == OPF_JMP)
+ {
+ break;
+ }
+
+ if (ops[j].op == OP_PUSH) {
+ pp->arg[arg].datap = &ops[j];
+ ret = scan_for_mod(&ops[j], j + 1, i);
+ if (ret >= 0) {
+ // mark this push as one that needs operand saving
+ ops[j].argmask |= 1 << arg;
+ save_arg_vars |= 1 << arg;
+ }
+ else
+ ops[j].flags |= OPF_RMD;
- pp->arg[arg].datap = &ops[j];
- ops[j].cls = OPC_RMD;
- for (arg++; arg < pp->argc; arg++)
- if (pp->arg[arg].reg == NULL)
- break;
+ // next arg
+ for (arg++; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+ }
+
+ if (g_labels[j][0] != 0) {
+ if (j > 0 && ((ops[j - 1].flags & OPF_TAIL)
+ || (ops[j - 1].flags & (OPF_JMP|OPF_CC)) == OPF_JMP))
+ {
+ // follow the branch in reverse
+ if (g_label_refs[j] == NULL)
+ ferr(po, "no refs for '%s'?\n", g_labels[j]);
+ if (g_label_refs[j]->lrl != NULL)
+ ferr(po, "unhandled multiple fefs to '%s'\n", g_labels[j]);
+ j = (g_label_refs[j] - ops) + 1;
+ continue;
+ }
+ break;
+ }
}
if (arg < pp->argc)
- ferr(&ops[i], "arg collect failed for '%s'\n", tmpname);
- ops[i].datap = pp;
+ ferr(po, "arg collect failed for '%s'\n", tmpname);
+ po->datap = pp;
+ }
+ else if (po->op == OP_MUL
+ || (po->op == OP_IMUL && po->operand_cnt == 1))
+ {
+ need_mul_var = 1;
}
}
fprintf(fout, " union { u32 d[%d]; u16 w[%d]; u8 b[%d]; } sf;\n",
(g_bp_stack + 3) / 4, (g_bp_stack + 1) / 2, g_bp_stack);
- // instantiate arg-registers
+ // declare arg-registers
for (i = 0; i < g_func_pp.argc; i++) {
if (g_func_pp.arg[i].reg != NULL) {
reg = char_array_i(regs_r32,
}
}
- // instantiate other regs - special case for eax
+ // declare other regs - special case for eax
if (!((regmask | regmask_arg) & 1) && !IS(g_func_pp.ret_type, "void")) {
fprintf(fout, " u32 eax = 0;\n");
had_decl = 1;
}
}
+ if (save_arg_vars) {
+ for (reg = 0; reg < 32; reg++) {
+ if (save_arg_vars & (1 << reg)) {
+ fprintf(fout, " u32 s_a%d;\n", reg + 1);
+ had_decl = 1;
+ }
+ }
+ }
+
+ if (cmp_result_vars) {
+ for (i = 0; i < 8; i++) {
+ if (cmp_result_vars & (1 << i)) {
+ fprintf(fout, " u32 cond_%s;\n", parsed_flag_op_names[i]);
+ had_decl = 1;
+ }
+ }
+ }
+
+ if (need_mul_var) {
+ fprintf(fout, " u64 mul_tmp;\n");
+ had_decl = 1;
+ }
+
if (had_decl)
fprintf(fout, "\n");
// output ops
- for (i = 0; i < opcnt; i++) {
+ for (i = 0; i < opcnt; i++)
+ {
if (g_labels[i][0] != 0)
fprintf(fout, "\n%s:\n", g_labels[i]);
- if (ops[i].cls == OPC_RMD)
+ po = &ops[i];
+ if (po->flags & OPF_RMD)
continue;
- po = &ops[i];
no_output = 0;
#define assert_operand_cnt(n_) \
if (po->operand_cnt != n_) \
ferr(po, "operand_cnt is %d/%d\n", po->operand_cnt, n_)
- if (last_arith_dst != NULL && po->operand_cnt > 0
- && po->operand[0].type == OPT_REG)
+ // conditional/flag using op?
+ if (po->flags & OPF_CC)
{
- if (IS(last_arith_dst->name, po->operand[0].name))
- last_arith_dst = NULL;
- }
+ int is_delayed = 0;
+ int is_inv = 0;
- // conditional op?
- special_sbb = 0;
- if (po->op == OP_SBB && IS(opr_name(po, 0), opr_name(po, 1)))
- special_sbb = 1;
+ pfo = split_cond(po, po->op, &is_inv);
- if (po->cls == OPC_JCC || po->cls == OPC_SCC || special_sbb)
- {
- enum op_op flag_op = po->op;
- if (special_sbb)
- flag_op = OP_JC;
-
- if (delayed_op != NULL)
+ // we go through all this trouble to avoid using parsed_flag_op,
+ // which makes generated code much nicer
+ if (delayed_flag_op != NULL)
{
- if (delayed_op->op == OP_TEST) {
- if (IS(opr_name(delayed_op, 0), opr_name(delayed_op, 1))) {
- out_src_opr(buf3, sizeof(buf3), delayed_op,
- &delayed_op->operand[0], 0);
- }
- else {
- out_src_opr(buf1, sizeof(buf1), delayed_op,
- &delayed_op->operand[0], 0);
- out_src_opr(buf2, sizeof(buf2), delayed_op,
- &delayed_op->operand[1], 0);
- snprintf(buf3, sizeof(buf3), "(%s & %s)", buf1, buf2);
- }
- out_test_for_cc(buf1, sizeof(buf1), po, flag_op,
- delayed_op->operand[0].lmod, buf3);
- }
- else if (delayed_op->op == OP_CMP) {
- out_src_opr(buf2, sizeof(buf2), delayed_op,
- &delayed_op->operand[0], 0);
- out_src_opr(buf3, sizeof(buf3), delayed_op,
- &delayed_op->operand[1], 0);
- out_cmp_for_cc(buf1, sizeof(buf1), po, flag_op,
- delayed_op->operand[0].lmod, buf2, buf3);
- }
+ out_cmp_test(buf1, sizeof(buf1), delayed_flag_op, pfo, is_inv);
+ is_delayed = 1;
}
else if (last_arith_dst != NULL
- && (flag_op == OP_JZ || flag_op == OP_JNZ))
+ && (pfo == PFO_Z || pfo == PFO_S || pfo == PFO_P))
{
out_src_opr(buf3, sizeof(buf3), po, last_arith_dst, 0);
- out_test_for_cc(buf1, sizeof(buf1), po, flag_op,
+ out_test_for_cc(buf1, sizeof(buf1), po, pfo, is_inv,
last_arith_dst->lmod, buf3);
+ is_delayed = 1;
+ }
+ else if (po->datap != NULL) {
+ // use preprocessed results
+ tmp_op = po->datap;
+ if (!tmp_op || !(tmp_op->pfomask & (1 << pfo)))
+ ferr(po, "not prepared for pfo %d\n", pfo);
+
+ // note: is_inv was not yet applied
+ snprintf(buf1, sizeof(buf1), "(%scond_%s)",
+ is_inv ? "!" : "", parsed_flag_op_names[pfo]);
+ }
+ else {
+ ferr(po, "all methods of finding comparison failed\n");
}
- else
- ferr(po, "no delayed_op or last arith result before cond op\n");
- if (po->cls == OPC_JCC) {
+ if (po->flags & OPF_JMP) {
fprintf(fout, " if %s\n", buf1);
}
- else if (special_sbb) {
- out_dst_opr(buf2, sizeof(buf2), po, &po->operand[0]);
- fprintf(fout, " %s = %s * -1;", buf2, buf1);
+ else if (po->op == OP_ADC || po->op == OP_SBB) {
+ if (is_delayed)
+ fprintf(fout, " cond_%s = %s;\n",
+ parsed_flag_op_names[pfo], buf1);
}
- else {
+ else if (po->flags & OPF_DATA) { // SETcc
out_dst_opr(buf2, sizeof(buf2), po, &po->operand[0]);
fprintf(fout, " %s = %s;", buf2, buf1);
}
+ else {
+ ferr(po, "unhandled conditional op\n");
+ }
}
+ pfomask = po->pfomask;
+
switch (po->op)
{
case OP_MOV:
case OP_LEA:
assert_operand_cnt(2);
- propagate_lmod(po, &po->operand[0], &po->operand[1]);
+ po->operand[1].lmod = OPLM_DWORD; // always
fprintf(fout, " %s = %s;",
out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 1));
fprintf(fout, " %s = ~%s;", buf1, buf1);
break;
+ case OP_CDQ:
+ assert_operand_cnt(2);
+ fprintf(fout, " %s = (s32)%s >> 31;",
+ out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 0));
+ strcpy(g_comment, "cdq");
+ break;
+
// arithmetic w/flags
case OP_ADD:
case OP_SUB:
case OP_AND:
case OP_OR:
- case OP_XOR:
+ propagate_lmod(po, &po->operand[0], &po->operand[1]);
+ // fallthrough
case OP_SHL:
case OP_SHR:
dualop_arith:
assert_operand_cnt(2);
- propagate_lmod(po, &po->operand[0], &po->operand[1]);
fprintf(fout, " %s %s= %s;",
out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
op_to_c(po),
out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 0));
last_arith_dst = &po->operand[0];
- delayed_op = NULL;
+ delayed_flag_op = NULL;
break;
- case OP_SAR:
+ case OP_XOR:
assert_operand_cnt(2);
propagate_lmod(po, &po->operand[0], &po->operand[1]);
- switch (po->operand[0].lmod) {
- case OPLM_BYTE:
- strcpy(buf3, "(s8)");
- break;
- case OPLM_WORD:
- strcpy(buf3, "(s16)");
+ if (IS(opr_name(po, 0), opr_name(po, 1))) {
+ // special case for XOR
+ fprintf(fout, " %s = 0;",
+ out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]));
+ last_arith_dst = &po->operand[0];
+ delayed_flag_op = NULL;
break;
- case OPLM_DWORD:
- strcpy(buf3, "(s32)");
- break;
- default:
- ferr(po, "invalid dst lmod: %d\n", po->operand[0].lmod);
}
+ goto dualop_arith;
+
+ case OP_SAR:
+ assert_operand_cnt(2);
out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
- fprintf(fout, " %s = %s%s >> %s;", buf1, buf3, buf1,
- out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 0));
+ fprintf(fout, " %s = %s%s >> %s;", buf1,
+ lmod_cast_s(po, po->operand[0].lmod), buf1,
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 0));
last_arith_dst = &po->operand[0];
- delayed_op = NULL;
+ delayed_flag_op = NULL;
break;
+ case OP_ADC:
case OP_SBB:
- if (!special_sbb)
- ferr(po, "TODO\n");
+ assert_operand_cnt(2);
+ propagate_lmod(po, &po->operand[0], &po->operand[1]);
+ fprintf(fout, " %s %s= %s + cond_c;",
+ out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
+ op_to_c(po),
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], 0));
+ last_arith_dst = &po->operand[0];
+ delayed_flag_op = NULL;
break;
case OP_INC:
case OP_DEC:
out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
strcpy(buf2, po->op == OP_INC ? "++" : "--");
- switch (po->operand[0].lmod) {
- case OPLM_BYTE:
- fprintf(fout, " LOBYTE(%s)%s;", buf1, buf2);
- break;
- case OPLM_WORD:
- fprintf(fout, " LOWORD(%s)%s;", buf1, buf2);
- break;
- case OPLM_DWORD:
- fprintf(fout, " %s%s;", buf1, buf2);
- break;
- default:
- ferr(po, "invalid dst lmod: %d\n", po->operand[0].lmod);
- }
+ fprintf(fout, " %s%s;", buf1, buf2);
+ last_arith_dst = &po->operand[0];
+ delayed_flag_op = NULL;
+ break;
+
+ case OP_NEG:
+ out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[0], 0);
+ fprintf(fout, " %s = -%s%s;", buf1,
+ lmod_cast_s(po, po->operand[0].lmod), buf2);
last_arith_dst = &po->operand[0];
- delayed_op = NULL;
+ delayed_flag_op = NULL;
+ if (pfomask & (1 << PFO_C)) {
+ fprintf(fout, "\n cond_c = (%s != 0);", buf1);
+ pfomask &= ~(1 << PFO_C);
+ }
break;
case OP_IMUL:
if (po->operand_cnt == 2)
goto dualop_arith;
- ferr(po, "TODO imul\n");
- last_arith_dst = &po->operand[0];
- delayed_op = NULL;
+ if (po->operand_cnt == 3)
+ ferr(po, "TODO imul3\n");
+ // fallthrough
+ case OP_MUL:
+ assert_operand_cnt(1);
+ strcpy(buf1, po->op == OP_IMUL ? "(s64)(s32)" : "(u64)");
+ fprintf(fout, " mul_tmp = %seax * %s%s;\n", buf1, buf1,
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[0], 0));
+ fprintf(fout, " edx = mul_tmp >> 32;\n");
+ fprintf(fout, " eax = mul_tmp;");
+ last_arith_dst = NULL;
+ delayed_flag_op = NULL;
+ break;
+
+ case OP_DIV:
+ case OP_IDIV:
+ assert_operand_cnt(1);
+ if (po->operand[0].lmod != OPLM_DWORD)
+ ferr(po, "unhandled lmod %d\n", po->operand[0].lmod);
+
+ // 32bit division is common, look for it
+ if (scan_for_cdq_edx(i - 1) >= 0) {
+ out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], 0);
+ strcpy(buf2, lmod_cast(po, po->operand[0].lmod,
+ po->op == OP_IDIV));
+ fprintf(fout, " edx = %seax %% %s%s;\n", buf2, buf2, buf1);
+ fprintf(fout, " eax = %seax / %s%s;", buf2, buf2, buf1);
+ }
+ else
+ ferr(po, "TODO 64bit divident\n");
+ last_arith_dst = NULL;
+ delayed_flag_op = NULL;
break;
case OP_TEST:
case OP_CMP:
propagate_lmod(po, &po->operand[0], &po->operand[1]);
- delayed_op = po;
- no_output = 1;
+ if (pfomask != 0) {
+ for (j = 0; j < 8; j++) {
+ if (pfomask & (1 << j)) {
+ out_cmp_test(buf1, sizeof(buf1), po, j, 0);
+ fprintf(fout, " cond_%s = %s;",
+ parsed_flag_op_names[j], buf1);
+ }
+ }
+ pfomask = 0;
+ }
+ else
+ no_output = 1;
+ delayed_flag_op = po;
break;
- // note: we reuse OP_Jcc for SETcc, only cls differs
+ // note: we reuse OP_Jcc for SETcc, only flags differ
case OP_JO ... OP_JG:
- if (po->cls == OPC_JCC)
+ if (po->flags & OPF_JMP)
fprintf(fout, " goto %s;", po->operand[0].name);
+ // else SETcc - should already be handled
break;
case OP_JMP:
+ assert_operand_cnt(1);
+ if (po->operand[0].type != OPT_LABEL)
+ ferr(po, "unhandled call type\n");
+
fprintf(fout, " goto %s;", po->operand[0].name);
break;
case OP_CALL:
+ assert_operand_cnt(1);
+ if (po->operand[0].type != OPT_LABEL)
+ ferr(po, "unhandled call type\n");
+
pp = po->datap;
if (pp == NULL)
ferr(po, "NULL pp\n");
fprintf(fout, " ");
if (!IS(pp->ret_type, "void")) {
- fprintf(fout, "eax = ");
+ if (po->flags & OPF_TAIL)
+ fprintf(fout, "return ");
+ else
+ fprintf(fout, "eax = ");
if (strchr(pp->ret_type, '*'))
fprintf(fout, "(u32)");
}
+
fprintf(fout, "%s(", opr_name(po, 0));
for (arg = 0; arg < pp->argc; arg++) {
if (arg > 0)
fprintf(fout, ", ");
+
+ if (strchr(pp->arg[arg].type, '*'))
+ fprintf(fout, "(%s)", pp->arg[arg].type);
+
if (pp->arg[arg].reg != NULL) {
fprintf(fout, "%s", pp->arg[arg].reg);
continue;
tmp_op = pp->arg[arg].datap;
if (tmp_op == NULL)
ferr(po, "parsed_op missing for arg%d\n", arg);
- fprintf(fout, "%s",
- out_src_opr(buf1, sizeof(buf1), tmp_op, &tmp_op->operand[0], 0));
+ if (tmp_op->argmask) {
+ fprintf(fout, "s_a%d", arg + 1);
+ }
+ else {
+ fprintf(fout, "%s",
+ out_src_opr(buf1, sizeof(buf1),
+ tmp_op, &tmp_op->operand[0], 0));
+ }
}
fprintf(fout, ");");
+
+ if (po->flags & OPF_TAIL) {
+ strcpy(g_comment, "tailcall");
+ if (IS(pp->ret_type, "void"))
+ fprintf(fout, "\n return;");
+ }
+ delayed_flag_op = NULL;
+ last_arith_dst = NULL;
break;
case OP_RET:
break;
case OP_PUSH:
+ if (po->argmask) {
+ // special case - saved func arg
+ for (j = 0; j < 32; j++) {
+ if (po->argmask & (1 << j)) {
+ fprintf(fout, " s_a%d = %s;", j + 1,
+ out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], 0));
+ }
+ }
+ break;
+ }
ferr(po, "push encountered\n");
break;
default:
no_output = 1;
- ferr(po, "unhandled op type %d, cls %d\n",
- po->op, po->cls);
+ ferr(po, "unhandled op type %d, flags %x\n",
+ po->op, po->flags);
break;
}
}
if (!no_output)
fprintf(fout, "\n");
+
+ if (pfomask != 0)
+ ferr(po, "missed flag calc, pfomask=%x\n", pfomask);
+
+ // see is delayed flag stuff is still valid
+ if (delayed_flag_op != NULL && delayed_flag_op != po) {
+ if (is_any_opr_modified(delayed_flag_op, po))
+ delayed_flag_op = NULL;
+ }
+
+ if (last_arith_dst != NULL && last_arith_dst != &po->operand[0]) {
+ if (is_opr_modified(last_arith_dst, po))
+ last_arith_dst = NULL;
+ }
}
fprintf(fout, "}\n\n");
char line[256];
char words[16][256];
int in_func = 0;
+ int skip_warned = 0;
int eq_alloc;
int pi = 0;
int len;
words[0], g_func);
gen_func(fout, fhdr, g_func, pi);
in_func = 0;
+ skip_warned = 0;
g_func[0] = 0;
if (pi != 0) {
memset(&ops, 0, pi * sizeof(ops[0]));
memset(g_labels, 0, pi * sizeof(g_labels[0]));
+ memset(g_label_refs, 0, pi * sizeof(g_label_refs[0]));
pi = 0;
}
g_eqcnt = 0;
continue;
}
+ if (!in_func) {
+ if (!skip_warned && g_labels[pi][0] != 0) {
+ anote("skipping from '%s'\n", g_labels[pi]);
+ skip_warned = 1;
+ }
+ g_labels[pi][0] = 0;
+ continue;
+ }
+
parse_op(&ops[pi], words, wordc);
pi++;
-
- (void)proto_parse;
}
fclose(fout);
notaz.gp2x.de / ia32rtools.git / blobdiff