[ASMPP] Improve handling of rip relative addressing

This commit is contained in:
Timo Kreuzer 2023-10-28 18:52:48 +03:00
parent a34f1ed03c
commit 2245dd78f5

View file

@ -296,27 +296,54 @@ vector<TOKEN_DEF> g_TokenList =
// FIXME: use context?
unsigned int g_label_number = 0;
vector<string> g_identifiers;
bool g_processing_jmp = false;
enum class IDTYPE
{
Memory,
Register,
Label,
Constant,
Macro,
Instruction,
String,
Unknown
};
struct IDENTIFIER
{
string Name;
IDTYPE Type;
};
vector<IDENTIFIER> g_identifiers;
static
void
add_identifier(Token& tok, IDTYPE type)
{
g_identifiers.push_back(IDENTIFIER{ tok.str(), type });
//fprintf(stderr, "Added id: '%s'\n", tok.str().c_str());
}
void
add_mem_id(Token& tok)
{
g_identifiers.push_back(tok.str());
//fprintf(stderr, "Added mem id: '%s'\n", tok.str().c_str());
add_identifier(tok, IDTYPE::Memory);
}
bool
is_mem_id(Token& tok)
{
for (auto id : g_identifiers)
for (IDENTIFIER& identifier : g_identifiers)
{
if (id == tok.str())
if (identifier.Name == tok.str())
{
return true;
return identifier.Type == IDTYPE::Memory;
}
}
return false;
return true;
}
bool
@ -587,10 +614,13 @@ size_t translate_instruction_param(TokenList& tokens, size_t index, const vector
case TOKEN_TYPE::Operator:
if (tok.str() == ",")
return index;
return translate_token(tokens, index, macro_params);
case TOKEN_TYPE::Identifier:
index = translate_token(tokens, index, macro_params);
if (is_mem_id(tok))
if (is_mem_id(tok) &&
!is_string_in_list(macro_params, tok.str()) &&
!g_processing_jmp)
{
printf("[rip]");
}
@ -604,8 +634,36 @@ size_t translate_instruction_param(TokenList& tokens, size_t index, const vector
return index;
}
static
bool
is_jmp_or_call(const Token& tok)
{
const char* inst_list[] = {
"jmp", "call", "ja", "jae", "jb", "jbe", "jc", "jcxz", "je", "jecxz", "jg", "jge",
"jl", "jle", "jna", "jnae", "jnb", "jnbe", "jnc", "jne", "jng", "jnge", "jnl", "jnle",
"jno", "jnp", "jns", "jnz", "jo", "jp", "jpe", "jpo", "jrcxz", "js", "jz", "loop", "loope",
"loopne", "loopnz", "loopz"
};
for (const char* inst : inst_list)
{
if (iequals(tok.str(), inst))
{
return true;
}
}
return false;
}
size_t translate_instruction(TokenList& tokens, size_t index, const vector<string>& macro_params)
{
// Check for jump/call instructions
if (is_jmp_or_call(tokens[index]))
{
g_processing_jmp = true;
}
// Translate the instruction itself
index = translate_token(tokens, index, macro_params);
@ -624,6 +682,7 @@ size_t translate_instruction(TokenList& tokens, size_t index, const vector<strin
{
case TOKEN_TYPE::Comment:
case TOKEN_TYPE::NewLine:
g_processing_jmp = false;
return index;
case TOKEN_TYPE::WhiteSpace:
@ -637,6 +696,7 @@ size_t translate_instruction(TokenList& tokens, size_t index, const vector<strin
}
}
g_processing_jmp = false;
return index;
}
@ -891,6 +951,7 @@ translate_identifier_construct(TokenList& tokens, size_t index, const vector<str
{
printf("%s:", tok.str().c_str());
}
add_identifier(tok, IDTYPE::Label);
return index + 2;
}
@ -912,6 +973,7 @@ translate_identifier_construct(TokenList& tokens, size_t index, const vector<str
case TOKEN_TYPE::KW_EQU:
//printf("%s%s", tok.str().c_str(), tok1.str().c_str());
printf("#define %s ", tok.str().c_str());
add_identifier(tok, IDTYPE::Constant);
return translate_expression(tokens, index + 3, macro_params);
case TOKEN_TYPE::KW_TEXTEQU:
@ -921,6 +983,7 @@ translate_identifier_construct(TokenList& tokens, size_t index, const vector<str
string textdef = tok4.str();
printf("#define %s %s", tok.str().c_str(), textdef.substr(1, textdef.size() - 2).c_str());
add_identifier(tok, IDTYPE::Constant);
return index + 5;
}
@ -940,6 +1003,7 @@ translate_identifier_construct(TokenList& tokens, size_t index, const vector<str
#endif
index += 2;
}
add_identifier(tok, IDTYPE::Label);
break;
}