gmqcc/ir.cpp
Dale Weiler d9127bf28a
Merge pull request #180 from xonotic/mem_leak_fix_on_failure_paths
two small memory leak fixes on failure paths
2017-11-26 17:30:30 -05:00

4093 lines
127 KiB
C++

#include <stdlib.h>
#include <string.h>
#include "gmqcc.h"
#include "ir.h"
/***********************************************************************
* Type sizes used at multiple points in the IR codegen
*/
const char *type_name[TYPE_COUNT] = {
"void",
"string",
"float",
"vector",
"entity",
"field",
"function",
"pointer",
"integer",
"variant",
"struct",
"union",
"array",
"nil",
"<no-expression>"
};
static size_t type_sizeof_[TYPE_COUNT] = {
1, /* TYPE_VOID */
1, /* TYPE_STRING */
1, /* TYPE_FLOAT */
3, /* TYPE_VECTOR */
1, /* TYPE_ENTITY */
1, /* TYPE_FIELD */
1, /* TYPE_FUNCTION */
1, /* TYPE_POINTER */
1, /* TYPE_INTEGER */
3, /* TYPE_VARIANT */
0, /* TYPE_STRUCT */
0, /* TYPE_UNION */
0, /* TYPE_ARRAY */
0, /* TYPE_NIL */
0, /* TYPE_NOESPR */
};
const uint16_t type_store_instr[TYPE_COUNT] = {
INSTR_STORE_F, /* should use I when having integer support */
INSTR_STORE_S,
INSTR_STORE_F,
INSTR_STORE_V,
INSTR_STORE_ENT,
INSTR_STORE_FLD,
INSTR_STORE_FNC,
INSTR_STORE_ENT, /* should use I */
#if 0
INSTR_STORE_I, /* integer type */
#else
INSTR_STORE_F,
#endif
INSTR_STORE_V, /* variant, should never be accessed */
VINSTR_END, /* struct */
VINSTR_END, /* union */
VINSTR_END, /* array */
VINSTR_END, /* nil */
VINSTR_END, /* noexpr */
};
const uint16_t field_store_instr[TYPE_COUNT] = {
INSTR_STORE_FLD,
INSTR_STORE_FLD,
INSTR_STORE_FLD,
INSTR_STORE_V,
INSTR_STORE_FLD,
INSTR_STORE_FLD,
INSTR_STORE_FLD,
INSTR_STORE_FLD,
#if 0
INSTR_STORE_FLD, /* integer type */
#else
INSTR_STORE_FLD,
#endif
INSTR_STORE_V, /* variant, should never be accessed */
VINSTR_END, /* struct */
VINSTR_END, /* union */
VINSTR_END, /* array */
VINSTR_END, /* nil */
VINSTR_END, /* noexpr */
};
const uint16_t type_storep_instr[TYPE_COUNT] = {
INSTR_STOREP_F, /* should use I when having integer support */
INSTR_STOREP_S,
INSTR_STOREP_F,
INSTR_STOREP_V,
INSTR_STOREP_ENT,
INSTR_STOREP_FLD,
INSTR_STOREP_FNC,
INSTR_STOREP_ENT, /* should use I */
#if 0
INSTR_STOREP_ENT, /* integer type */
#else
INSTR_STOREP_F,
#endif
INSTR_STOREP_V, /* variant, should never be accessed */
VINSTR_END, /* struct */
VINSTR_END, /* union */
VINSTR_END, /* array */
VINSTR_END, /* nil */
VINSTR_END, /* noexpr */
};
const uint16_t type_eq_instr[TYPE_COUNT] = {
INSTR_EQ_F, /* should use I when having integer support */
INSTR_EQ_S,
INSTR_EQ_F,
INSTR_EQ_V,
INSTR_EQ_E,
INSTR_EQ_E, /* FLD has no comparison */
INSTR_EQ_FNC,
INSTR_EQ_E, /* should use I */
#if 0
INSTR_EQ_I,
#else
INSTR_EQ_F,
#endif
INSTR_EQ_V, /* variant, should never be accessed */
VINSTR_END, /* struct */
VINSTR_END, /* union */
VINSTR_END, /* array */
VINSTR_END, /* nil */
VINSTR_END, /* noexpr */
};
const uint16_t type_ne_instr[TYPE_COUNT] = {
INSTR_NE_F, /* should use I when having integer support */
INSTR_NE_S,
INSTR_NE_F,
INSTR_NE_V,
INSTR_NE_E,
INSTR_NE_E, /* FLD has no comparison */
INSTR_NE_FNC,
INSTR_NE_E, /* should use I */
#if 0
INSTR_NE_I,
#else
INSTR_NE_F,
#endif
INSTR_NE_V, /* variant, should never be accessed */
VINSTR_END, /* struct */
VINSTR_END, /* union */
VINSTR_END, /* array */
VINSTR_END, /* nil */
VINSTR_END, /* noexpr */
};
const uint16_t type_not_instr[TYPE_COUNT] = {
INSTR_NOT_F, /* should use I when having integer support */
VINSTR_END, /* not to be used, depends on string related -f flags */
INSTR_NOT_F,
INSTR_NOT_V,
INSTR_NOT_ENT,
INSTR_NOT_ENT,
INSTR_NOT_FNC,
INSTR_NOT_ENT, /* should use I */
#if 0
INSTR_NOT_I, /* integer type */
#else
INSTR_NOT_F,
#endif
INSTR_NOT_V, /* variant, should never be accessed */
VINSTR_END, /* struct */
VINSTR_END, /* union */
VINSTR_END, /* array */
VINSTR_END, /* nil */
VINSTR_END, /* noexpr */
};
/* protos */
static void ir_function_dump(ir_function*, char *ind, int (*oprintf)(const char*,...));
static ir_value* ir_block_create_general_instr(ir_block *self, lex_ctx_t, const char *label,
int op, ir_value *a, ir_value *b, qc_type outype);
static bool GMQCC_WARN ir_block_create_store(ir_block*, lex_ctx_t, ir_value *target, ir_value *what);
static void ir_block_dump(ir_block*, char *ind, int (*oprintf)(const char*,...));
static bool ir_instr_op(ir_instr*, int op, ir_value *value, bool writing);
static void ir_instr_dump(ir_instr* in, char *ind, int (*oprintf)(const char*,...));
/* error functions */
static void irerror(lex_ctx_t ctx, const char *msg, ...)
{
va_list ap;
va_start(ap, msg);
con_cvprintmsg(ctx, LVL_ERROR, "internal error", msg, ap);
va_end(ap);
}
static bool GMQCC_WARN irwarning(lex_ctx_t ctx, int warntype, const char *fmt, ...)
{
bool r;
va_list ap;
va_start(ap, fmt);
r = vcompile_warning(ctx, warntype, fmt, ap);
va_end(ap);
return r;
}
/***********************************************************************
* Vector utility functions
*/
static bool GMQCC_WARN vec_ir_value_find(std::vector<ir_value *> &vec, const ir_value *what, size_t *idx)
{
for (auto &it : vec) {
if (it != what)
continue;
if (idx)
*idx = &it - &vec[0];
return true;
}
return false;
}
static bool GMQCC_WARN vec_ir_block_find(std::vector<ir_block *> &vec, ir_block *what, size_t *idx)
{
for (auto &it : vec) {
if (it != what)
continue;
if (idx)
*idx = &it - &vec[0];
return true;
}
return false;
}
static bool GMQCC_WARN vec_ir_instr_find(std::vector<ir_instr *> &vec, ir_instr *what, size_t *idx)
{
for (auto &it : vec) {
if (it != what)
continue;
if (idx)
*idx = &it - &vec[0];
return true;
}
return false;
}
/***********************************************************************
* IR Builder
*/
static void ir_block_delete_quick(ir_block* self);
static void ir_instr_delete_quick(ir_instr *self);
static void ir_function_delete_quick(ir_function *self);
ir_builder::ir_builder(const std::string& modulename)
: m_name(modulename),
m_code(new code_t)
{
m_htglobals = util_htnew(IR_HT_SIZE);
m_htfields = util_htnew(IR_HT_SIZE);
m_htfunctions = util_htnew(IR_HT_SIZE);
m_nil = new ir_value("nil", store_value, TYPE_NIL);
m_nil->m_cvq = CV_CONST;
for (size_t i = 0; i != IR_MAX_VINSTR_TEMPS; ++i) {
/* we write to them, but they're not supposed to be used outside the IR, so
* let's not allow the generation of ir_instrs which use these.
* So it's a constant noexpr.
*/
m_vinstr_temp[i] = new ir_value("vinstr_temp", store_value, TYPE_NOEXPR);
m_vinstr_temp[i]->m_cvq = CV_CONST;
}
}
ir_builder::~ir_builder()
{
util_htdel(m_htglobals);
util_htdel(m_htfields);
util_htdel(m_htfunctions);
for (auto& f : m_functions)
ir_function_delete_quick(f.release());
m_functions.clear(); // delete them now before deleting the rest:
delete m_nil;
for (size_t i = 0; i != IR_MAX_VINSTR_TEMPS; ++i) {
delete m_vinstr_temp[i];
}
m_extparams.clear();
m_extparam_protos.clear();
}
ir_function* ir_builder::createFunction(const std::string& name, qc_type outtype)
{
ir_function *fn = (ir_function*)util_htget(m_htfunctions, name.c_str());
if (fn)
return nullptr;
fn = new ir_function(this, outtype);
fn->m_name = name;
m_functions.emplace_back(fn);
util_htset(m_htfunctions, name.c_str(), fn);
fn->m_value = createGlobal(fn->m_name, TYPE_FUNCTION);
if (!fn->m_value) {
delete fn;
return nullptr;
}
fn->m_value->m_hasvalue = true;
fn->m_value->m_outtype = outtype;
fn->m_value->m_constval.vfunc = fn;
fn->m_value->m_context = fn->m_context;
return fn;
}
ir_value* ir_builder::createGlobal(const std::string& name, qc_type vtype)
{
ir_value *ve;
if (name[0] != '#')
{
ve = (ir_value*)util_htget(m_htglobals, name.c_str());
if (ve) {
return nullptr;
}
}
ve = new ir_value(std::string(name), store_global, vtype);
m_globals.emplace_back(ve);
util_htset(m_htglobals, name.c_str(), ve);
return ve;
}
ir_value* ir_builder::get_va_count()
{
if (m_reserved_va_count)
return m_reserved_va_count;
return (m_reserved_va_count = createGlobal("reserved:va_count", TYPE_FLOAT));
}
ir_value* ir_builder::createField(const std::string& name, qc_type vtype)
{
ir_value *ve = (ir_value*)util_htget(m_htfields, name.c_str());
if (ve) {
return nullptr;
}
ve = new ir_value(std::string(name), store_global, TYPE_FIELD);
ve->m_fieldtype = vtype;
m_fields.emplace_back(ve);
util_htset(m_htfields, name.c_str(), ve);
return ve;
}
/***********************************************************************
*IR Function
*/
static bool ir_function_naive_phi(ir_function*);
static void ir_function_enumerate(ir_function*);
static bool ir_function_calculate_liferanges(ir_function*);
static bool ir_function_allocate_locals(ir_function*);
ir_function::ir_function(ir_builder* owner_, qc_type outtype_)
: m_owner(owner_),
m_name("<@unnamed>"),
m_outtype(outtype_)
{
m_context.file = "<@no context>";
m_context.line = 0;
}
ir_function::~ir_function()
{
}
static void ir_function_delete_quick(ir_function *self)
{
for (auto& b : self->m_blocks)
ir_block_delete_quick(b.release());
delete self;
}
static void ir_function_collect_value(ir_function *self, ir_value *v)
{
self->m_values.emplace_back(v);
}
ir_block* ir_function_create_block(lex_ctx_t ctx, ir_function *self, const char *label)
{
ir_block* bn = new ir_block(self, label ? std::string(label) : std::string());
bn->m_context = ctx;
self->m_blocks.emplace_back(bn);
if ((self->m_flags & IR_FLAG_BLOCK_COVERAGE) && self->m_owner->m_coverage_func)
(void)ir_block_create_call(bn, ctx, nullptr, self->m_owner->m_coverage_func, false);
return bn;
}
static bool instr_is_operation(uint16_t op)
{
return ( (op >= INSTR_MUL_F && op <= INSTR_GT) ||
(op >= INSTR_LOAD_F && op <= INSTR_LOAD_FNC) ||
(op == INSTR_ADDRESS) ||
(op >= INSTR_NOT_F && op <= INSTR_NOT_FNC) ||
(op >= INSTR_AND && op <= INSTR_BITOR) ||
(op >= INSTR_CALL0 && op <= INSTR_CALL8) ||
(op >= VINSTR_BITAND_V && op <= VINSTR_NEG_V) );
}
static bool ir_function_pass_peephole(ir_function *self)
{
for (auto& bp : self->m_blocks) {
ir_block *block = bp.get();
for (size_t i = 0; i < block->m_instr.size(); ++i) {
ir_instr *inst;
inst = block->m_instr[i];
if (i >= 1 &&
(inst->m_opcode >= INSTR_STORE_F &&
inst->m_opcode <= INSTR_STORE_FNC))
{
ir_instr *store;
ir_instr *oper;
ir_value *value;
store = inst;
oper = block->m_instr[i-1];
if (!instr_is_operation(oper->m_opcode))
continue;
/* Don't change semantics of MUL_VF in engines where these may not alias. */
if (OPTS_FLAG(LEGACY_VECTOR_MATHS)) {
if (oper->m_opcode == INSTR_MUL_VF && oper->_m_ops[2]->m_memberof == oper->_m_ops[1])
continue;
if (oper->m_opcode == INSTR_MUL_FV && oper->_m_ops[1]->m_memberof == oper->_m_ops[2])
continue;
}
value = oper->_m_ops[0];
/* only do it for SSA values */
if (value->m_store != store_value)
continue;
/* don't optimize out the temp if it's used later again */
if (value->m_reads.size() != 1)
continue;
/* The very next store must use this value */
if (value->m_reads[0] != store)
continue;
/* And of course the store must _read_ from it, so it's in
* OP 1 */
if (store->_m_ops[1] != value)
continue;
++opts_optimizationcount[OPTIM_PEEPHOLE];
(void)!ir_instr_op(oper, 0, store->_m_ops[0], true);
block->m_instr.erase(block->m_instr.begin() + i);
delete store;
}
else if (inst->m_opcode == VINSTR_COND)
{
/* COND on a value resulting from a NOT could
* remove the NOT and swap its operands
*/
while (true) {
ir_block *tmp;
size_t inotid;
ir_instr *inot;
ir_value *value;
value = inst->_m_ops[0];
if (value->m_store != store_value || value->m_reads.size() != 1 || value->m_reads[0] != inst)
break;
inot = value->m_writes[0];
if (inot->_m_ops[0] != value ||
inot->m_opcode < INSTR_NOT_F ||
inot->m_opcode > INSTR_NOT_FNC ||
inot->m_opcode == INSTR_NOT_V || /* can't do these */
inot->m_opcode == INSTR_NOT_S)
{
break;
}
/* count */
++opts_optimizationcount[OPTIM_PEEPHOLE];
/* change operand */
(void)!ir_instr_op(inst, 0, inot->_m_ops[1], false);
/* remove NOT */
tmp = inot->m_owner;
for (inotid = 0; inotid < tmp->m_instr.size(); ++inotid) {
if (tmp->m_instr[inotid] == inot)
break;
}
if (inotid >= tmp->m_instr.size()) {
compile_error(inst->m_context, "sanity-check failed: failed to find instruction to optimize out");
return false;
}
tmp->m_instr.erase(tmp->m_instr.begin() + inotid);
delete inot;
/* swap ontrue/onfalse */
tmp = inst->m_bops[0];
inst->m_bops[0] = inst->m_bops[1];
inst->m_bops[1] = tmp;
}
continue;
}
}
}
return true;
}
static bool ir_function_pass_tailrecursion(ir_function *self)
{
size_t p;
for (auto& bp : self->m_blocks) {
ir_block *block = bp.get();
ir_value *funcval;
ir_instr *ret, *call, *store = nullptr;
if (!block->m_final || block->m_instr.size() < 2)
continue;
ret = block->m_instr.back();
if (ret->m_opcode != INSTR_DONE && ret->m_opcode != INSTR_RETURN)
continue;
call = block->m_instr[block->m_instr.size()-2];
if (call->m_opcode >= INSTR_STORE_F && call->m_opcode <= INSTR_STORE_FNC) {
/* account for the unoptimized
* CALL
* STORE %return, %tmp
* RETURN %tmp
* version
*/
if (block->m_instr.size() < 3)
continue;
store = call;
call = block->m_instr[block->m_instr.size()-3];
}
if (call->m_opcode < INSTR_CALL0 || call->m_opcode > INSTR_CALL8)
continue;
if (store) {
/* optimize out the STORE */
if (ret->_m_ops[0] &&
ret->_m_ops[0] == store->_m_ops[0] &&
store->_m_ops[1] == call->_m_ops[0])
{
++opts_optimizationcount[OPTIM_PEEPHOLE];
call->_m_ops[0] = store->_m_ops[0];
block->m_instr.erase(block->m_instr.end()-2);
delete store;
}
else
continue;
}
if (!call->_m_ops[0])
continue;
funcval = call->_m_ops[1];
if (!funcval)
continue;
if (funcval->m_vtype != TYPE_FUNCTION || funcval->m_constval.vfunc != self)
continue;
/* now we have a CALL and a RET, check if it's a tailcall */
if (ret->_m_ops[0] && call->_m_ops[0] != ret->_m_ops[0])
continue;
++opts_optimizationcount[OPTIM_TAIL_RECURSION];
block->m_instr.erase(block->m_instr.end()-2, block->m_instr.end());
block->m_final = false; /* open it back up */
/* emite parameter-stores */
for (p = 0; p < call->m_params.size(); ++p) {
/* assert(call->params_count <= self->locals_count); */
if (!ir_block_create_store(block, call->m_context, self->m_locals[p].get(), call->m_params[p])) {
irerror(call->m_context, "failed to create tailcall store instruction for parameter %i", (int)p);
return false;
}
}
if (!ir_block_create_jump(block, call->m_context, self->m_blocks[0].get())) {
irerror(call->m_context, "failed to create tailcall jump");
return false;
}
delete call;
delete ret;
}
return true;
}
bool ir_function_finalize(ir_function *self)
{
if (self->m_builtin)
return true;
for (auto& lp : self->m_locals) {
ir_value *v = lp.get();
if (v->m_reads.empty() && v->m_writes.size() && !(v->m_flags & IR_FLAG_NOREF)) {
// if it's a vector check to ensure all it's members are unused before
// claiming it's unused, otherwise skip the vector entierly
if (v->m_vtype == TYPE_VECTOR)
{
size_t mask = (1 << 3) - 1, bits = 0;
for (size_t i = 0; i < 3; i++)
if (!v->m_members[i] || (v->m_members[i]->m_reads.empty()
&& v->m_members[i]->m_writes.size()))
bits |= (1 << i);
// all components are unused so just report the vector
if (bits == mask && irwarning(v->m_context, WARN_UNUSED_VARIABLE,
"unused variable: `%s`", v->m_name.c_str()))
return false;
else if (bits != mask)
// individual components are unused so mention them
for (size_t i = 0; i < 3; i++)
if ((bits & (1 << i))
&& irwarning(v->m_context, WARN_UNUSED_COMPONENT,
"unused vector component: `%s.%c`", v->m_name.c_str(), "xyz"[i]))
return false;
}
// just a standard variable
else if (irwarning(v->m_context, WARN_UNUSED_VARIABLE,
"unused variable: `%s`", v->m_name.c_str())) return false;
}
}
if (OPTS_OPTIMIZATION(OPTIM_PEEPHOLE)) {
if (!ir_function_pass_peephole(self)) {
irerror(self->m_context, "generic optimization pass broke something in `%s`", self->m_name.c_str());
return false;
}
}
if (OPTS_OPTIMIZATION(OPTIM_TAIL_RECURSION)) {
if (!ir_function_pass_tailrecursion(self)) {
irerror(self->m_context, "tail-recursion optimization pass broke something in `%s`", self->m_name.c_str());
return false;
}
}
if (!ir_function_naive_phi(self)) {
irerror(self->m_context, "internal error: ir_function_naive_phi failed");
return false;
}
for (auto& lp : self->m_locals) {
ir_value *v = lp.get();
if (v->m_vtype == TYPE_VECTOR ||
(v->m_vtype == TYPE_FIELD && v->m_outtype == TYPE_VECTOR))
{
v->vectorMember(0);
v->vectorMember(1);
v->vectorMember(2);
}
}
for (auto& vp : self->m_values) {
ir_value *v = vp.get();
if (v->m_vtype == TYPE_VECTOR ||
(v->m_vtype == TYPE_FIELD && v->m_outtype == TYPE_VECTOR))
{
v->vectorMember(0);
v->vectorMember(1);
v->vectorMember(2);
}
}
ir_function_enumerate(self);
if (!ir_function_calculate_liferanges(self))
return false;
if (!ir_function_allocate_locals(self))
return false;
return true;
}
ir_value* ir_function_create_local(ir_function *self, const std::string& name, qc_type vtype, bool param)
{
ir_value *ve;
if (param &&
!self->m_locals.empty() &&
self->m_locals.back()->m_store != store_param)
{
irerror(self->m_context, "cannot add parameters after adding locals");
return nullptr;
}
ve = new ir_value(std::string(name), (param ? store_param : store_local), vtype);
if (param)
ve->m_locked = true;
self->m_locals.emplace_back(ve);
return ve;
}
/***********************************************************************
*IR Block
*/
ir_block::ir_block(ir_function* owner, const std::string& name)
: m_owner(owner),
m_label(name)
{
m_context.file = "<@no context>";
m_context.line = 0;
}
ir_block::~ir_block()
{
for (auto &i : m_instr)
delete i;
}
static void ir_block_delete_quick(ir_block* self)
{
for (auto &i : self->m_instr)
ir_instr_delete_quick(i);
self->m_instr.clear();
delete self;
}
/***********************************************************************
*IR Instructions
*/
ir_instr::ir_instr(lex_ctx_t ctx, ir_block* owner_, int op)
: m_opcode(op),
m_context(ctx),
m_owner(owner_)
{
}
ir_instr::~ir_instr()
{
// The following calls can only delete from
// vectors, we still want to delete this instruction
// so ignore the return value. Since with the warn_unused_result attribute
// gcc doesn't care about an explicit: (void)foo(); to ignore the result,
// I have to improvise here and use if(foo());
for (auto &it : m_phi) {
size_t idx;
if (vec_ir_instr_find(it.value->m_writes, this, &idx))
it.value->m_writes.erase(it.value->m_writes.begin() + idx);
if (vec_ir_instr_find(it.value->m_reads, this, &idx))
it.value->m_reads.erase(it.value->m_reads.begin() + idx);
}
for (auto &it : m_params) {
size_t idx;
if (vec_ir_instr_find(it->m_writes, this, &idx))
it->m_writes.erase(it->m_writes.begin() + idx);
if (vec_ir_instr_find(it->m_reads, this, &idx))
it->m_reads.erase(it->m_reads.begin() + idx);
}
(void)!ir_instr_op(this, 0, nullptr, false);
(void)!ir_instr_op(this, 1, nullptr, false);
(void)!ir_instr_op(this, 2, nullptr, false);
}
static void ir_instr_delete_quick(ir_instr *self)
{
self->m_phi.clear();
self->m_params.clear();
self->_m_ops[0] = nullptr;
self->_m_ops[1] = nullptr;
self->_m_ops[2] = nullptr;
delete self;
}
static bool ir_instr_op(ir_instr *self, int op, ir_value *v, bool writing)
{
if (v && v->m_vtype == TYPE_NOEXPR) {
irerror(self->m_context, "tried to use a NOEXPR value");
return false;
}
if (self->_m_ops[op]) {
size_t idx;
if (writing && vec_ir_instr_find(self->_m_ops[op]->m_writes, self, &idx))
self->_m_ops[op]->m_writes.erase(self->_m_ops[op]->m_writes.begin() + idx);
else if (vec_ir_instr_find(self->_m_ops[op]->m_reads, self, &idx))
self->_m_ops[op]->m_reads.erase(self->_m_ops[op]->m_reads.begin() + idx);
}
if (v) {
if (writing)
v->m_writes.push_back(self);
else
v->m_reads.push_back(self);
}
self->_m_ops[op] = v;
return true;
}
/***********************************************************************
*IR Value
*/
void ir_value::setCodeAddress(int32_t gaddr)
{
m_code.globaladdr = gaddr;
if (m_members[0]) m_members[0]->m_code.globaladdr = gaddr;
if (m_members[1]) m_members[1]->m_code.globaladdr = gaddr;
if (m_members[2]) m_members[2]->m_code.globaladdr = gaddr;
}
int32_t ir_value::codeAddress() const
{
if (m_store == store_return)
return OFS_RETURN + m_code.addroffset;
return m_code.globaladdr + m_code.addroffset;
}
ir_value::ir_value(std::string&& name_, store_type store_, qc_type vtype_)
: m_name(move(name_))
, m_vtype(vtype_)
, m_store(store_)
{
m_fieldtype = TYPE_VOID;
m_outtype = TYPE_VOID;
m_flags = 0;
m_cvq = CV_NONE;
m_hasvalue = false;
m_context.file = "<@no context>";
m_context.line = 0;
memset(&m_constval, 0, sizeof(m_constval));
memset(&m_code, 0, sizeof(m_code));
m_members[0] = nullptr;
m_members[1] = nullptr;
m_members[2] = nullptr;
m_memberof = nullptr;
m_unique_life = false;
m_locked = false;
m_callparam = false;
}
ir_value::ir_value(ir_function *owner, std::string&& name, store_type storetype, qc_type vtype)
: ir_value(move(name), storetype, vtype)
{
ir_function_collect_value(owner, this);
}
ir_value::~ir_value()
{
size_t i;
if (m_hasvalue) {
if (m_vtype == TYPE_STRING)
mem_d((void*)m_constval.vstring);
}
if (!(m_flags & IR_FLAG_SPLIT_VECTOR)) {
for (i = 0; i < 3; ++i) {
if (m_members[i])
delete m_members[i];
}
}
}
/* helper function */
ir_value* ir_builder::literalFloat(float value, bool add_to_list) {
ir_value *v = new ir_value("#IMMEDIATE", store_global, TYPE_FLOAT);
v->m_flags |= IR_FLAG_ERASABLE;
v->m_hasvalue = true;
v->m_cvq = CV_CONST;
v->m_constval.vfloat = value;
m_globals.emplace_back(v);
if (add_to_list)
m_const_floats.emplace_back(v);
return v;
}
ir_value* ir_value::vectorMember(unsigned int member)
{
std::string name;
ir_value *m;
if (member >= 3)
return nullptr;
if (m_members[member])
return m_members[member];
if (!m_name.empty()) {
char member_name[3] = { '_', char('x' + member), 0 };
name = m_name + member_name;
}
if (m_vtype == TYPE_VECTOR)
{
m = new ir_value(move(name), m_store, TYPE_FLOAT);
if (!m)
return nullptr;
m->m_context = m_context;
m_members[member] = m;
m->m_code.addroffset = member;
}
else if (m_vtype == TYPE_FIELD)
{
if (m_fieldtype != TYPE_VECTOR)
return nullptr;
m = new ir_value(move(name), m_store, TYPE_FIELD);
if (!m)
return nullptr;
m->m_fieldtype = TYPE_FLOAT;
m->m_context = m_context;
m_members[member] = m;
m->m_code.addroffset = member;
}
else
{
irerror(m_context, "invalid member access on %s", m_name.c_str());
return nullptr;
}
m->m_memberof = this;
return m;
}
size_t ir_value::size() const {
if (m_vtype == TYPE_FIELD && m_fieldtype == TYPE_VECTOR)
return type_sizeof_[TYPE_VECTOR];
return type_sizeof_[m_vtype];
}
bool ir_value::setFloat(float f)
{
if (m_vtype != TYPE_FLOAT)
return false;
m_constval.vfloat = f;
m_hasvalue = true;
return true;
}
bool ir_value::setFunc(int f)
{
if (m_vtype != TYPE_FUNCTION)
return false;
m_constval.vint = f;
m_hasvalue = true;
return true;
}
bool ir_value::setVector(vec3_t v)
{
if (m_vtype != TYPE_VECTOR)
return false;
m_constval.vvec = v;
m_hasvalue = true;
return true;
}
bool ir_value::setField(ir_value *fld)
{
if (m_vtype != TYPE_FIELD)
return false;
m_constval.vpointer = fld;
m_hasvalue = true;
return true;
}
bool ir_value::setString(const char *str)
{
if (m_vtype != TYPE_STRING)
return false;
m_constval.vstring = util_strdupe(str);
m_hasvalue = true;
return true;
}
#if 0
bool ir_value::setInt(int i)
{
if (m_vtype != TYPE_INTEGER)
return false;
m_constval.vint = i;
m_hasvalue = true;
return true;
}
#endif
bool ir_value::lives(size_t at)
{
for (auto& l : m_life) {
if (l.start <= at && at <= l.end)
return true;
if (l.start > at) /* since it's ordered */
return false;
}
return false;
}
bool ir_value::insertLife(size_t idx, ir_life_entry_t e)
{
m_life.insert(m_life.begin() + idx, e);
return true;
}
bool ir_value::setAlive(size_t s)
{
size_t i;
const size_t vs = m_life.size();
ir_life_entry_t *life_found = nullptr;
ir_life_entry_t *before = nullptr;
ir_life_entry_t new_entry;
/* Find the first range >= s */
for (i = 0; i < vs; ++i)
{
before = life_found;
life_found = &m_life[i];
if (life_found->start > s)
break;
}
/* nothing found? append */
if (i == vs) {
ir_life_entry_t e;
if (life_found && life_found->end+1 == s)
{
/* previous life range can be merged in */
life_found->end++;
return true;
}
if (life_found && life_found->end >= s)
return false;
e.start = e.end = s;
m_life.emplace_back(e);
return true;
}
/* found */
if (before)
{
if (before->end + 1 == s &&
life_found->start - 1 == s)
{
/* merge */
before->end = life_found->end;
m_life.erase(m_life.begin()+i);
return true;
}
if (before->end + 1 == s)
{
/* extend before */
before->end++;
return true;
}
/* already contained */
if (before->end >= s)
return false;
}
/* extend */
if (life_found->start - 1 == s)
{
life_found->start--;
return true;
}
/* insert a new entry */
new_entry.start = new_entry.end = s;
return insertLife(i, new_entry);
}
bool ir_value::mergeLife(const ir_value *other)
{
size_t i, myi;
if (other->m_life.empty())
return true;
if (m_life.empty()) {
m_life = other->m_life;
return true;
}
myi = 0;
for (i = 0; i < other->m_life.size(); ++i)
{
const ir_life_entry_t &otherlife = other->m_life[i];
while (true)
{
ir_life_entry_t *entry = &m_life[myi];
if (otherlife.end+1 < entry->start)
{
/* adding an interval before entry */
if (!insertLife(myi, otherlife))
return false;
++myi;
break;
}
if (otherlife.start < entry->start &&
otherlife.end+1 >= entry->start)
{
/* starts earlier and overlaps */
entry->start = otherlife.start;
}
if (otherlife.end > entry->end &&
otherlife.start <= entry->end+1)
{
/* ends later and overlaps */
entry->end = otherlife.end;
}
/* see if our change combines it with the next ranges */
while (myi+1 < m_life.size() &&
entry->end+1 >= m_life[1+myi].start)
{
/* overlaps with (myi+1) */
if (entry->end < m_life[1+myi].end)
entry->end = m_life[1+myi].end;
m_life.erase(m_life.begin() + (myi + 1));
entry = &m_life[myi];
}
/* see if we're after the entry */
if (otherlife.start > entry->end)
{
++myi;
/* append if we're at the end */
if (myi >= m_life.size()) {
m_life.emplace_back(otherlife);
break;
}
/* otherweise check the next range */
continue;
}
break;
}
}
return true;
}
static bool ir_values_overlap(const ir_value *a, const ir_value *b)
{
/* For any life entry in A see if it overlaps with
* any life entry in B.
* Note that the life entries are orderes, so we can make a
* more efficient algorithm there than naively translating the
* statement above.
*/
const ir_life_entry_t *la, *lb, *enda, *endb;
/* first of all, if either has no life range, they cannot clash */
if (a->m_life.empty() || b->m_life.empty())
return false;
la = &a->m_life.front();
lb = &b->m_life.front();
enda = &a->m_life.back() + 1;
endb = &b->m_life.back() + 1;
while (true)
{
/* check if the entries overlap, for that,
* both must start before the other one ends.
*/
if (la->start < lb->end &&
lb->start < la->end)
{
return true;
}
/* entries are ordered
* one entry is earlier than the other
* that earlier entry will be moved forward
*/
if (la->start < lb->start)
{
/* order: A B, move A forward
* check if we hit the end with A
*/
if (++la == enda)
break;
}
else /* if (lb->start < la->start) actually <= */
{
/* order: B A, move B forward
* check if we hit the end with B
*/
if (++lb == endb)
break;
}
}
return false;
}
/***********************************************************************
*IR main operations
*/
static bool ir_check_unreachable(ir_block *self)
{
/* The IR should never have to deal with unreachable code */
if (!self->m_final/* || OPTS_FLAG(ALLOW_UNREACHABLE_CODE)*/)
return true;
irerror(self->m_context, "unreachable statement (%s)", self->m_label.c_str());
return false;
}
bool ir_block_create_store_op(ir_block *self, lex_ctx_t ctx, int op, ir_value *target, ir_value *what)
{
ir_instr *in;
if (!ir_check_unreachable(self))
return false;
if (target->m_store == store_value &&
(op < INSTR_STOREP_F || op > INSTR_STOREP_FNC))
{
irerror(self->m_context, "cannot store to an SSA value");
irerror(self->m_context, "trying to store: %s <- %s", target->m_name.c_str(), what->m_name.c_str());
irerror(self->m_context, "instruction: %s", util_instr_str[op]);
return false;
}
in = new ir_instr(ctx, self, op);
if (!in)
return false;
if (!ir_instr_op(in, 0, target, (op < INSTR_STOREP_F || op > INSTR_STOREP_FNC)) ||
!ir_instr_op(in, 1, what, false))
{
delete in;
return false;
}
self->m_instr.push_back(in);
return true;
}
bool ir_block_create_state_op(ir_block *self, lex_ctx_t ctx, ir_value *frame, ir_value *think)
{
ir_instr *in;
if (!ir_check_unreachable(self))
return false;
in = new ir_instr(ctx, self, INSTR_STATE);
if (!in)
return false;
if (!ir_instr_op(in, 0, frame, false) ||
!ir_instr_op(in, 1, think, false))
{
delete in;
return false;
}
self->m_instr.push_back(in);
return true;
}
static bool ir_block_create_store(ir_block *self, lex_ctx_t ctx, ir_value *target, ir_value *what)
{
int op = 0;
qc_type vtype;
if (target->m_vtype == TYPE_VARIANT)
vtype = what->m_vtype;
else
vtype = target->m_vtype;
#if 0
if (vtype == TYPE_FLOAT && what->m_vtype == TYPE_INTEGER)
op = INSTR_CONV_ITOF;
else if (vtype == TYPE_INTEGER && what->m_vtype == TYPE_FLOAT)
op = INSTR_CONV_FTOI;
#endif
op = type_store_instr[vtype];
if (OPTS_FLAG(ADJUST_VECTOR_FIELDS)) {
if (op == INSTR_STORE_FLD && what->m_fieldtype == TYPE_VECTOR)
op = INSTR_STORE_V;
}
return ir_block_create_store_op(self, ctx, op, target, what);
}
bool ir_block_create_storep(ir_block *self, lex_ctx_t ctx, ir_value *target, ir_value *what)
{
int op = 0;
qc_type vtype;
if (target->m_vtype != TYPE_POINTER)
return false;
/* storing using pointer - target is a pointer, type must be
* inferred from source
*/
vtype = what->m_vtype;
op = type_storep_instr[vtype];
if (OPTS_FLAG(ADJUST_VECTOR_FIELDS)) {
if (op == INSTR_STOREP_FLD && what->m_fieldtype == TYPE_VECTOR)
op = INSTR_STOREP_V;
}
return ir_block_create_store_op(self, ctx, op, target, what);
}
bool ir_block_create_return(ir_block *self, lex_ctx_t ctx, ir_value *v)
{
ir_instr *in;
if (!ir_check_unreachable(self))
return false;
self->m_final = true;
self->m_is_return = true;
in = new ir_instr(ctx, self, INSTR_RETURN);
if (!in)
return false;
if (v && !ir_instr_op(in, 0, v, false)) {
delete in;
return false;
}
self->m_instr.push_back(in);
return true;
}
bool ir_block_create_if(ir_block *self, lex_ctx_t ctx, ir_value *v,
ir_block *ontrue, ir_block *onfalse)
{
ir_instr *in;
if (!ir_check_unreachable(self))
return false;
self->m_final = true;
/*in = new ir_instr(ctx, self, (v->m_vtype == TYPE_STRING ? INSTR_IF_S : INSTR_IF_F));*/
in = new ir_instr(ctx, self, VINSTR_COND);
if (!in)
return false;
if (!ir_instr_op(in, 0, v, false)) {
delete in;
return false;
}
in->m_bops[0] = ontrue;
in->m_bops[1] = onfalse;
self->m_instr.push_back(in);
self->m_exits.push_back(ontrue);
self->m_exits.push_back(onfalse);
ontrue->m_entries.push_back(self);
onfalse->m_entries.push_back(self);
return true;
}
bool ir_block_create_jump(ir_block *self, lex_ctx_t ctx, ir_block *to)
{
ir_instr *in;
if (!ir_check_unreachable(self))
return false;
self->m_final = true;
in = new ir_instr(ctx, self, VINSTR_JUMP);
if (!in)
return false;
in->m_bops[0] = to;
self->m_instr.push_back(in);
self->m_exits.push_back(to);
to->m_entries.push_back(self);
return true;
}
bool ir_block_create_goto(ir_block *self, lex_ctx_t ctx, ir_block *to)
{
self->m_owner->m_flags |= IR_FLAG_HAS_GOTO;
return ir_block_create_jump(self, ctx, to);
}
ir_instr* ir_block_create_phi(ir_block *self, lex_ctx_t ctx, const char *label, qc_type ot)
{
ir_value *out;
ir_instr *in;
if (!ir_check_unreachable(self))
return nullptr;
in = new ir_instr(ctx, self, VINSTR_PHI);
if (!in)
return nullptr;
out = new ir_value(self->m_owner, label ? label : "", store_value, ot);
if (!out) {
delete in;
return nullptr;
}
if (!ir_instr_op(in, 0, out, true)) {
delete in;
return nullptr;
}
self->m_instr.push_back(in);
return in;
}
ir_value* ir_phi_value(ir_instr *self)
{
return self->_m_ops[0];
}
void ir_phi_add(ir_instr* self, ir_block *b, ir_value *v)
{
ir_phi_entry_t pe;
if (!vec_ir_block_find(self->m_owner->m_entries, b, nullptr)) {
// Must not be possible to cause this, otherwise the AST
// is doing something wrong.
irerror(self->m_context, "Invalid entry block for PHI");
exit(EXIT_FAILURE);
}
pe.value = v;
pe.from = b;
v->m_reads.push_back(self);
self->m_phi.push_back(pe);
}
/* call related code */
ir_instr* ir_block_create_call(ir_block *self, lex_ctx_t ctx, const char *label, ir_value *func, bool noreturn)
{
ir_value *out;
ir_instr *in;
if (!ir_check_unreachable(self))
return nullptr;
in = new ir_instr(ctx, self, (noreturn ? VINSTR_NRCALL : INSTR_CALL0));
if (!in)
return nullptr;
if (noreturn) {
self->m_final = true;
self->m_is_return = true;
}
out = new ir_value(self->m_owner, label ? label : "", (func->m_outtype == TYPE_VOID) ? store_return : store_value, func->m_outtype);
if (!out) {
delete in;
return nullptr;
}
if (!ir_instr_op(in, 0, out, true) ||
!ir_instr_op(in, 1, func, false))
{
delete in;
delete out;
return nullptr;
}
self->m_instr.push_back(in);
/*
if (noreturn) {
if (!ir_block_create_return(self, ctx, nullptr)) {
compile_error(ctx, "internal error: failed to generate dummy-return instruction");
delete in;
return nullptr;
}
}
*/
return in;
}
ir_value* ir_call_value(ir_instr *self)
{
return self->_m_ops[0];
}
void ir_call_param(ir_instr* self, ir_value *v)
{
self->m_params.push_back(v);
v->m_reads.push_back(self);
}
/* binary op related code */
ir_value* ir_block_create_binop(ir_block *self, lex_ctx_t ctx,
const char *label, int opcode,
ir_value *left, ir_value *right)
{
qc_type ot = TYPE_VOID;
switch (opcode) {
case INSTR_ADD_F:
case INSTR_SUB_F:
case INSTR_DIV_F:
case INSTR_MUL_F:
case INSTR_MUL_V:
case INSTR_AND:
case INSTR_OR:
#if 0
case INSTR_AND_I:
case INSTR_AND_IF:
case INSTR_AND_FI:
case INSTR_OR_I:
case INSTR_OR_IF:
case INSTR_OR_FI:
#endif
case INSTR_BITAND:
case INSTR_BITOR:
case VINSTR_BITXOR:
#if 0
case INSTR_SUB_S: /* -- offset of string as float */
case INSTR_MUL_IF:
case INSTR_MUL_FI:
case INSTR_DIV_IF:
case INSTR_DIV_FI:
case INSTR_BITOR_IF:
case INSTR_BITOR_FI:
case INSTR_BITAND_FI:
case INSTR_BITAND_IF:
case INSTR_EQ_I:
case INSTR_NE_I:
#endif
ot = TYPE_FLOAT;
break;
#if 0
case INSTR_ADD_I:
case INSTR_ADD_IF:
case INSTR_ADD_FI:
case INSTR_SUB_I:
case INSTR_SUB_FI:
case INSTR_SUB_IF:
case INSTR_MUL_I:
case INSTR_DIV_I:
case INSTR_BITAND_I:
case INSTR_BITOR_I:
case INSTR_XOR_I:
case INSTR_RSHIFT_I:
case INSTR_LSHIFT_I:
ot = TYPE_INTEGER;
break;
#endif
case INSTR_ADD_V:
case INSTR_SUB_V:
case INSTR_MUL_VF:
case INSTR_MUL_FV:
case VINSTR_BITAND_V:
case VINSTR_BITOR_V:
case VINSTR_BITXOR_V:
case VINSTR_BITAND_VF:
case VINSTR_BITOR_VF:
case VINSTR_BITXOR_VF:
case VINSTR_CROSS:
#if 0
case INSTR_DIV_VF:
case INSTR_MUL_IV:
case INSTR_MUL_VI:
#endif
ot = TYPE_VECTOR;
break;
#if 0
case INSTR_ADD_SF:
ot = TYPE_POINTER;
break;
#endif
/*
* after the following default case, the value of opcode can never
* be 1, 2, 3, 4, 5, 6, 7, 8, 9, 62, 63, 64, 65
*/
default:
/* ranges: */
/* boolean operations result in floats */
/*
* opcode >= 10 takes true branch opcode is at least 10
* opcode <= 23 takes false branch opcode is at least 24
*/
if (opcode >= INSTR_EQ_F && opcode <= INSTR_GT)
ot = TYPE_FLOAT;
/*
* At condition "opcode <= 23", the value of "opcode" must be
* at least 24.
* At condition "opcode <= 23", the value of "opcode" cannot be
* equal to any of {1, 2, 3, 4, 5, 6, 7, 8, 9, 62, 63, 64, 65}.
* The condition "opcode <= 23" cannot be true.
*
* Thus ot=2 (TYPE_FLOAT) can never be true
*/
#if 0
else if (opcode >= INSTR_LE && opcode <= INSTR_GT)
ot = TYPE_FLOAT;
else if (opcode >= INSTR_LE_I && opcode <= INSTR_EQ_FI)
ot = TYPE_FLOAT;
#endif
break;
};
if (ot == TYPE_VOID) {
/* The AST or parser were supposed to check this! */
return nullptr;
}
return ir_block_create_general_instr(self, ctx, label, opcode, left, right, ot);
}
ir_value* ir_block_create_unary(ir_block *self, lex_ctx_t ctx,
const char *label, int opcode,
ir_value *operand)
{
qc_type ot = TYPE_FLOAT;
switch (opcode) {
case INSTR_NOT_F:
case INSTR_NOT_V:
case INSTR_NOT_S:
case INSTR_NOT_ENT:
case INSTR_NOT_FNC: /*
case INSTR_NOT_I: */
ot = TYPE_FLOAT;
break;
/*
* Negation for virtual instructions is emulated with 0-value. Thankfully
* the operand for 0 already exists so we just source it from here.
*/
case VINSTR_NEG_F:
return ir_block_create_general_instr(self, ctx, label, INSTR_SUB_F, nullptr, operand, ot);
case VINSTR_NEG_V:
return ir_block_create_general_instr(self, ctx, label, INSTR_SUB_V, self->m_owner->m_owner->m_nil, operand, TYPE_VECTOR);
default:
ot = operand->m_vtype;
break;
};
if (ot == TYPE_VOID) {
/* The AST or parser were supposed to check this! */
return nullptr;
}
/* let's use the general instruction creator and pass nullptr for OPB */
return ir_block_create_general_instr(self, ctx, label, opcode, operand, nullptr, ot);
}
static ir_value* ir_block_create_general_instr(ir_block *self, lex_ctx_t ctx, const char *label,
int op, ir_value *a, ir_value *b, qc_type outype)
{
ir_instr *instr;
ir_value *out;
out = new ir_value(self->m_owner, label ? label : "", store_value, outype);
if (!out)
return nullptr;
instr = new ir_instr(ctx, self, op);
if (!instr) {
return nullptr;
}
if (!ir_instr_op(instr, 0, out, true) ||
!ir_instr_op(instr, 1, a, false) ||
!ir_instr_op(instr, 2, b, false) )
{
goto on_error;
}
self->m_instr.push_back(instr);
return out;
on_error:
delete instr;
return nullptr;
}
ir_value* ir_block_create_fieldaddress(ir_block *self, lex_ctx_t ctx, const char *label, ir_value *ent, ir_value *field)
{
ir_value *v;
/* Support for various pointer types todo if so desired */
if (ent->m_vtype != TYPE_ENTITY)
return nullptr;
if (field->m_vtype != TYPE_FIELD)
return nullptr;
v = ir_block_create_general_instr(self, ctx, label, INSTR_ADDRESS, ent, field, TYPE_POINTER);
v->m_fieldtype = field->m_fieldtype;
return v;
}
ir_value* ir_block_create_load_from_ent(ir_block *self, lex_ctx_t ctx, const char *label, ir_value *ent, ir_value *field, qc_type outype)
{
int op;
if (ent->m_vtype != TYPE_ENTITY)
return nullptr;
/* at some point we could redirect for TYPE_POINTER... but that could lead to carelessness */
if (field->m_vtype != TYPE_FIELD)
return nullptr;
switch (outype)
{
case TYPE_FLOAT: op = INSTR_LOAD_F; break;
case TYPE_VECTOR: op = INSTR_LOAD_V; break;
case TYPE_STRING: op = INSTR_LOAD_S; break;
case TYPE_FIELD: op = INSTR_LOAD_FLD; break;
case TYPE_ENTITY: op = INSTR_LOAD_ENT; break;
case TYPE_FUNCTION: op = INSTR_LOAD_FNC; break;
#if 0
case TYPE_POINTER: op = INSTR_LOAD_I; break;
case TYPE_INTEGER: op = INSTR_LOAD_I; break;
#endif
default:
irerror(self->m_context, "invalid type for ir_block_create_load_from_ent: %s", type_name[outype]);
return nullptr;
}
return ir_block_create_general_instr(self, ctx, label, op, ent, field, outype);
}
/* PHI resolving breaks the SSA, and must thus be the last
* step before life-range calculation.
*/
static bool ir_block_naive_phi(ir_block *self);
bool ir_function_naive_phi(ir_function *self)
{
for (auto& b : self->m_blocks)
if (!ir_block_naive_phi(b.get()))
return false;
return true;
}
static bool ir_block_naive_phi(ir_block *self)
{
size_t i;
/* FIXME: optionally, create_phi can add the phis
* to a list so we don't need to loop through blocks
* - anyway: "don't optimize YET"
*/
for (i = 0; i < self->m_instr.size(); ++i)
{
ir_instr *instr = self->m_instr[i];
if (instr->m_opcode != VINSTR_PHI)
continue;
self->m_instr.erase(self->m_instr.begin()+i);
--i; /* NOTE: i+1 below */
for (auto &it : instr->m_phi) {
ir_value *v = it.value;
ir_block *b = it.from;
if (v->m_store == store_value && v->m_reads.size() == 1 && v->m_writes.size() == 1) {
/* replace the value */
if (!ir_instr_op(v->m_writes[0], 0, instr->_m_ops[0], true))
return false;
} else {
/* force a move instruction */
ir_instr *prevjump = b->m_instr.back();
b->m_instr.pop_back();
b->m_final = false;
instr->_m_ops[0]->m_store = store_global;
if (!ir_block_create_store(b, instr->m_context, instr->_m_ops[0], v))
return false;
instr->_m_ops[0]->m_store = store_value;
b->m_instr.push_back(prevjump);
b->m_final = true;
}
}
delete instr;
}
return true;
}
/***********************************************************************
*IR Temp allocation code
* Propagating value life ranges by walking through the function backwards
* until no more changes are made.
* In theory this should happen once more than once for every nested loop
* level.
* Though this implementation might run an additional time for if nests.
*/
/* Enumerate instructions used by value's life-ranges
*/
static void ir_block_enumerate(ir_block *self, size_t *_eid)
{
size_t eid = *_eid;
for (auto &i : self->m_instr)
i->m_eid = eid++;
*_eid = eid;
}
/* Enumerate blocks and instructions.
* The block-enumeration is unordered!
* We do not really use the block enumreation, however
* the instruction enumeration is important for life-ranges.
*/
void ir_function_enumerate(ir_function *self)
{
size_t instruction_id = 0;
size_t block_eid = 0;
for (auto& block : self->m_blocks)
{
/* each block now gets an additional "entry" instruction id
* we can use to avoid point-life issues
*/
block->m_entry_id = instruction_id;
block->m_eid = block_eid;
++instruction_id;
++block_eid;
ir_block_enumerate(block.get(), &instruction_id);
}
}
/* Local-value allocator
* After finishing creating the liferange of all values used in a function
* we can allocate their global-positions.
* This is the counterpart to register-allocation in register machines.
*/
struct function_allocator {
std::vector<std::unique_ptr<ir_value>> locals;
std::vector<size_t> sizes;
std::vector<size_t> positions;
std::vector<bool> unique;
};
static bool function_allocator_alloc(function_allocator *alloc, ir_value *var)
{
ir_value *slot;
size_t vsize = var->size();
var->m_code.local = alloc->locals.size();
slot = new ir_value("reg", store_global, var->m_vtype);
if (!slot)
return false;
if (!slot->mergeLife(var))
goto localerror;
alloc->locals.emplace_back(slot);
alloc->sizes.push_back(vsize);
alloc->unique.push_back(var->m_unique_life);
return true;
localerror:
delete slot;
return false;
}
static bool ir_function_allocator_assign(ir_function *self, function_allocator *alloc, ir_value *v)
{
size_t a;
if (v->m_unique_life)
return function_allocator_alloc(alloc, v);
for (a = 0; a < alloc->locals.size(); ++a)
{
/* if it's reserved for a unique liferange: skip */
if (alloc->unique[a])
continue;
ir_value *slot = alloc->locals[a].get();
/* never resize parameters
* will be required later when overlapping temps + locals
*/
if (a < self->m_params.size() &&
alloc->sizes[a] < v->size())
{
continue;
}
if (ir_values_overlap(v, slot))
continue;
if (!slot->mergeLife(v))
return false;
/* adjust size for this slot */
if (alloc->sizes[a] < v->size())
alloc->sizes[a] = v->size();
v->m_code.local = a;
return true;
}
if (a >= alloc->locals.size()) {
if (!function_allocator_alloc(alloc, v))
return false;
}
return true;
}
bool ir_function_allocate_locals(ir_function *self)
{
size_t pos;
bool opt_gt = OPTS_OPTIMIZATION(OPTIM_GLOBAL_TEMPS);
function_allocator lockalloc, globalloc;
if (self->m_locals.empty() && self->m_values.empty())
return true;
size_t i;
for (i = 0; i < self->m_locals.size(); ++i)
{
ir_value *v = self->m_locals[i].get();
if ((self->m_flags & IR_FLAG_MASK_NO_LOCAL_TEMPS) || !OPTS_OPTIMIZATION(OPTIM_LOCAL_TEMPS)) {
v->m_locked = true;
v->m_unique_life = true;
}
else if (i >= self->m_params.size())
break;
else
v->m_locked = true; /* lock parameters locals */
if (!function_allocator_alloc((v->m_locked || !opt_gt ? &lockalloc : &globalloc), v))
return false;
}
for (; i < self->m_locals.size(); ++i)
{
ir_value *v = self->m_locals[i].get();
if (v->m_life.empty())
continue;
if (!ir_function_allocator_assign(self, (v->m_locked || !opt_gt ? &lockalloc : &globalloc), v))
return false;
}
/* Allocate a slot for any value that still exists */
for (i = 0; i < self->m_values.size(); ++i)
{
ir_value *v = self->m_values[i].get();
if (v->m_life.empty())
continue;
/* CALL optimization:
* If the value is a parameter-temp: 1 write, 1 read from a CALL
* and it's not "locked", write it to the OFS_PARM directly.
*/
if (OPTS_OPTIMIZATION(OPTIM_CALL_STORES) && !v->m_locked && !v->m_unique_life) {
if (v->m_reads.size() == 1 && v->m_writes.size() == 1 &&
(v->m_reads[0]->m_opcode == VINSTR_NRCALL ||
(v->m_reads[0]->m_opcode >= INSTR_CALL0 && v->m_reads[0]->m_opcode <= INSTR_CALL8)
)
)
{
size_t param;
ir_instr *call = v->m_reads[0];
if (!vec_ir_value_find(call->m_params, v, &param)) {
irerror(call->m_context, "internal error: unlocked parameter %s not found", v->m_name.c_str());
return false;
}
++opts_optimizationcount[OPTIM_CALL_STORES];
v->m_callparam = true;
if (param < 8)
v->setCodeAddress(OFS_PARM0 + 3*param);
else {
size_t nprotos = self->m_owner->m_extparam_protos.size();
ir_value *ep;
param -= 8;
if (nprotos > param)
ep = self->m_owner->m_extparam_protos[param].get();
else
{
ep = self->m_owner->generateExtparamProto();
while (++nprotos <= param)
ep = self->m_owner->generateExtparamProto();
}
ir_instr_op(v->m_writes[0], 0, ep, true);
call->m_params[param+8] = ep;
}
continue;
}
if (v->m_writes.size() == 1 && v->m_writes[0]->m_opcode == INSTR_CALL0) {
v->m_store = store_return;
if (v->m_members[0]) v->m_members[0]->m_store = store_return;
if (v->m_members[1]) v->m_members[1]->m_store = store_return;
if (v->m_members[2]) v->m_members[2]->m_store = store_return;
++opts_optimizationcount[OPTIM_CALL_STORES];
continue;
}
}
if (!ir_function_allocator_assign(self, (v->m_locked || !opt_gt ? &lockalloc : &globalloc), v))
return false;
}
if (lockalloc.sizes.empty() && globalloc.sizes.empty())
return true;
lockalloc.positions.push_back(0);
globalloc.positions.push_back(0);
/* Adjust slot positions based on sizes */
if (!lockalloc.sizes.empty()) {
pos = (lockalloc.sizes.size() ? lockalloc.positions[0] : 0);
for (i = 1; i < lockalloc.sizes.size(); ++i)
{
pos = lockalloc.positions[i-1] + lockalloc.sizes[i-1];
lockalloc.positions.push_back(pos);
}
self->m_allocated_locals = pos + lockalloc.sizes.back();
}
if (!globalloc.sizes.empty()) {
pos = (globalloc.sizes.size() ? globalloc.positions[0] : 0);
for (i = 1; i < globalloc.sizes.size(); ++i)
{
pos = globalloc.positions[i-1] + globalloc.sizes[i-1];
globalloc.positions.push_back(pos);
}
self->m_globaltemps = pos + globalloc.sizes.back();
}
/* Locals need to know their new position */
for (auto& local : self->m_locals) {
if (local->m_locked || !opt_gt)
local->m_code.local = lockalloc.positions[local->m_code.local];
else
local->m_code.local = globalloc.positions[local->m_code.local];
}
/* Take over the actual slot positions on values */
for (auto& value : self->m_values) {
if (value->m_locked || !opt_gt)
value->m_code.local = lockalloc.positions[value->m_code.local];
else
value->m_code.local = globalloc.positions[value->m_code.local];
}
return true;
}
/* Get information about which operand
* is read from, or written to.
*/
static void ir_op_read_write(int op, size_t *read, size_t *write)
{
switch (op)
{
case VINSTR_JUMP:
case INSTR_GOTO:
*write = 0;
*read = 0;
break;
case INSTR_IF:
case INSTR_IFNOT:
#if 0
case INSTR_IF_S:
case INSTR_IFNOT_S:
#endif
case INSTR_RETURN:
case VINSTR_COND:
*write = 0;
*read = 1;
break;
case INSTR_STOREP_F:
case INSTR_STOREP_V:
case INSTR_STOREP_S:
case INSTR_STOREP_ENT:
case INSTR_STOREP_FLD:
case INSTR_STOREP_FNC:
*write = 0;
*read = 7;
break;
default:
*write = 1;
*read = 6;
break;
};
}
static bool ir_block_living_add_instr(ir_block *self, size_t eid) {
bool changed = false;
for (auto &it : self->m_living)
if (it->setAlive(eid))
changed = true;
return changed;
}
static bool ir_block_living_lock(ir_block *self) {
bool changed = false;
for (auto &it : self->m_living) {
if (it->m_locked)
continue;
it->m_locked = true;
changed = true;
}
return changed;
}
static bool ir_block_life_propagate(ir_block *self, bool *changed)
{
ir_instr *instr;
ir_value *value;
size_t i, o, mem;
// bitmasks which operands are read from or written to
size_t read, write;
self->m_living.clear();
for (auto &prev : self->m_exits) {
for (auto &it : prev->m_living)
if (!vec_ir_value_find(self->m_living, it, nullptr))
self->m_living.push_back(it);
}
i = self->m_instr.size();
while (i)
{ --i;
instr = self->m_instr[i];
/* See which operands are read and write operands */
ir_op_read_write(instr->m_opcode, &read, &write);
/* Go through the 3 main operands
* writes first, then reads
*/
for (o = 0; o < 3; ++o)
{
if (!instr->_m_ops[o]) /* no such operand */
continue;
value = instr->_m_ops[o];
/* We only care about locals */
/* we also calculate parameter liferanges so that locals
* can take up parameter slots */
if (value->m_store != store_value &&
value->m_store != store_local &&
value->m_store != store_param)
continue;
/* write operands */
/* When we write to a local, we consider it "dead" for the
* remaining upper part of the function, since in SSA a value
* can only be written once (== created)
*/
if (write & (1<<o))
{
size_t idx;
bool in_living = vec_ir_value_find(self->m_living, value, &idx);
if (!in_living)
{
/* If the value isn't alive it hasn't been read before... */
/* TODO: See if the warning can be emitted during parsing or AST processing
* otherwise have warning printed here.
* IF printing a warning here: include filecontext_t,
* and make sure it's only printed once
* since this function is run multiple times.
*/
/* con_err( "Value only written %s\n", value->m_name); */
if (value->setAlive(instr->m_eid))
*changed = true;
} else {
/* since 'living' won't contain it
* anymore, merge the value, since
* (A) doesn't.
*/
if (value->setAlive(instr->m_eid))
*changed = true;
// Then remove
self->m_living.erase(self->m_living.begin() + idx);
}
/* Removing a vector removes all members */
for (mem = 0; mem < 3; ++mem) {
if (value->m_members[mem] && vec_ir_value_find(self->m_living, value->m_members[mem], &idx)) {
if (value->m_members[mem]->setAlive(instr->m_eid))
*changed = true;
self->m_living.erase(self->m_living.begin() + idx);
}
}
/* Removing the last member removes the vector */
if (value->m_memberof) {
value = value->m_memberof;
for (mem = 0; mem < 3; ++mem) {
if (value->m_members[mem] && vec_ir_value_find(self->m_living, value->m_members[mem], nullptr))
break;
}
if (mem == 3 && vec_ir_value_find(self->m_living, value, &idx)) {
if (value->setAlive(instr->m_eid))
*changed = true;
self->m_living.erase(self->m_living.begin() + idx);
}
}
}
}
/* These operations need a special case as they can break when using
* same source and destination operand otherwise, as the engine may
* read the source multiple times. */
if (instr->m_opcode == INSTR_MUL_VF ||
instr->m_opcode == VINSTR_BITAND_VF ||
instr->m_opcode == VINSTR_BITOR_VF ||
instr->m_opcode == VINSTR_BITXOR ||
instr->m_opcode == VINSTR_BITXOR_VF ||
instr->m_opcode == VINSTR_BITXOR_V ||
instr->m_opcode == VINSTR_CROSS)
{
value = instr->_m_ops[2];
/* the float source will get an additional lifetime */
if (value->setAlive(instr->m_eid+1))
*changed = true;
if (value->m_memberof && value->m_memberof->setAlive(instr->m_eid+1))
*changed = true;
}
if (instr->m_opcode == INSTR_MUL_FV ||
instr->m_opcode == INSTR_LOAD_V ||
instr->m_opcode == VINSTR_BITXOR ||
instr->m_opcode == VINSTR_BITXOR_VF ||
instr->m_opcode == VINSTR_BITXOR_V ||
instr->m_opcode == VINSTR_CROSS)
{
value = instr->_m_ops[1];
/* the float source will get an additional lifetime */
if (value->setAlive(instr->m_eid+1))
*changed = true;
if (value->m_memberof && value->m_memberof->setAlive(instr->m_eid+1))
*changed = true;
}
for (o = 0; o < 3; ++o)
{
if (!instr->_m_ops[o]) /* no such operand */
continue;
value = instr->_m_ops[o];
/* We only care about locals */
/* we also calculate parameter liferanges so that locals
* can take up parameter slots */
if (value->m_store != store_value &&
value->m_store != store_local &&
value->m_store != store_param)
continue;
/* read operands */
if (read & (1<<o))
{
if (!vec_ir_value_find(self->m_living, value, nullptr))
self->m_living.push_back(value);
/* reading adds the full vector */
if (value->m_memberof && !vec_ir_value_find(self->m_living, value->m_memberof, nullptr))
self->m_living.push_back(value->m_memberof);
for (mem = 0; mem < 3; ++mem) {
if (value->m_members[mem] && !vec_ir_value_find(self->m_living, value->m_members[mem], nullptr))
self->m_living.push_back(value->m_members[mem]);
}
}
}
/* PHI operands are always read operands */
for (auto &it : instr->m_phi) {
value = it.value;
if (!vec_ir_value_find(self->m_living, value, nullptr))
self->m_living.push_back(value);
/* reading adds the full vector */
if (value->m_memberof && !vec_ir_value_find(self->m_living, value->m_memberof, nullptr))
self->m_living.push_back(value->m_memberof);
for (mem = 0; mem < 3; ++mem) {
if (value->m_members[mem] && !vec_ir_value_find(self->m_living, value->m_members[mem], nullptr))
self->m_living.push_back(value->m_members[mem]);
}
}
/* on a call, all these values must be "locked" */
if (instr->m_opcode >= INSTR_CALL0 && instr->m_opcode <= INSTR_CALL8) {
if (ir_block_living_lock(self))
*changed = true;
}
/* call params are read operands too */
for (auto &it : instr->m_params) {
value = it;
if (!vec_ir_value_find(self->m_living, value, nullptr))
self->m_living.push_back(value);
/* reading adds the full vector */
if (value->m_memberof && !vec_ir_value_find(self->m_living, value->m_memberof, nullptr))
self->m_living.push_back(value->m_memberof);
for (mem = 0; mem < 3; ++mem) {
if (value->m_members[mem] && !vec_ir_value_find(self->m_living, value->m_members[mem], nullptr))
self->m_living.push_back(value->m_members[mem]);
}
}
/* (A) */
if (ir_block_living_add_instr(self, instr->m_eid))
*changed = true;
}
/* the "entry" instruction ID */
if (ir_block_living_add_instr(self, self->m_entry_id))
*changed = true;
return true;
}
bool ir_function_calculate_liferanges(ir_function *self)
{
/* parameters live at 0 */
for (size_t i = 0; i < self->m_params.size(); ++i)
if (!self->m_locals[i].get()->setAlive(0))
compile_error(self->m_context, "internal error: failed value-life merging");
bool changed;
do {
self->m_run_id++;
changed = false;
for (auto i = self->m_blocks.rbegin(); i != self->m_blocks.rend(); ++i)
ir_block_life_propagate(i->get(), &changed);
} while (changed);
if (self->m_blocks.size()) {
ir_block *block = self->m_blocks[0].get();
for (auto &it : block->m_living) {
ir_value *v = it;
if (v->m_store != store_local)
continue;
if (v->m_vtype == TYPE_VECTOR)
continue;
self->m_flags |= IR_FLAG_HAS_UNINITIALIZED;
/* find the instruction reading from it */
size_t s = 0;
for (; s < v->m_reads.size(); ++s) {
if (v->m_reads[s]->m_eid == v->m_life[0].end)
break;
}
if (s < v->m_reads.size()) {
if (irwarning(v->m_context, WARN_USED_UNINITIALIZED,
"variable `%s` may be used uninitialized in this function\n"
" -> %s:%i",
v->m_name.c_str(),
v->m_reads[s]->m_context.file, v->m_reads[s]->m_context.line)
)
{
return false;
}
continue;
}
if (v->m_memberof) {
ir_value *vec = v->m_memberof;
for (s = 0; s < vec->m_reads.size(); ++s) {
if (vec->m_reads[s]->m_eid == v->m_life[0].end)
break;
}
if (s < vec->m_reads.size()) {
if (irwarning(v->m_context, WARN_USED_UNINITIALIZED,
"variable `%s` may be used uninitialized in this function\n"
" -> %s:%i",
v->m_name.c_str(),
vec->m_reads[s]->m_context.file, vec->m_reads[s]->m_context.line)
)
{
return false;
}
continue;
}
}
if (irwarning(v->m_context, WARN_USED_UNINITIALIZED,
"variable `%s` may be used uninitialized in this function", v->m_name.c_str()))
{
return false;
}
}
}
return true;
}
/***********************************************************************
*IR Code-Generation
*
* Since the IR has the convention of putting 'write' operands
* at the beginning, we have to rotate the operands of instructions
* properly in order to generate valid QCVM code.
*
* Having destinations at a fixed position is more convenient. In QC
* this is *mostly* OPC, but FTE adds at least 2 instructions which
* read from from OPA, and store to OPB rather than OPC. Which is
* partially the reason why the implementation of these instructions
* in darkplaces has been delayed for so long.
*
* Breaking conventions is annoying...
*/
static bool gen_global_field(code_t *code, ir_value *global)
{
if (global->m_hasvalue)
{
ir_value *fld = global->m_constval.vpointer;
if (!fld) {
irerror(global->m_context, "Invalid field constant with no field: %s", global->m_name.c_str());
return false;
}
/* copy the field's value */
global->setCodeAddress(code->globals.size());
code->globals.push_back(fld->m_code.fieldaddr);
if (global->m_fieldtype == TYPE_VECTOR) {
code->globals.push_back(fld->m_code.fieldaddr+1);
code->globals.push_back(fld->m_code.fieldaddr+2);
}
}
else
{
global->setCodeAddress(code->globals.size());
code->globals.push_back(0);
if (global->m_fieldtype == TYPE_VECTOR) {
code->globals.push_back(0);
code->globals.push_back(0);
}
}
if (global->m_code.globaladdr < 0)
return false;
return true;
}
static bool gen_global_pointer(code_t *code, ir_value *global)
{
if (global->m_hasvalue)
{
ir_value *target = global->m_constval.vpointer;
if (!target) {
irerror(global->m_context, "Invalid pointer constant: %s", global->m_name.c_str());
/* nullptr pointers are pointing to the nullptr constant, which also
* sits at address 0, but still has an ir_value for itself.
*/
return false;
}
/* Here, relocations ARE possible - in fteqcc-enhanced-qc:
* void() foo; <- proto
* void() *fooptr = &foo;
* void() foo = { code }
*/
if (!target->m_code.globaladdr) {
/* FIXME: Check for the constant nullptr ir_value!
* because then code.globaladdr being 0 is valid.
*/
irerror(global->m_context, "FIXME: Relocation support");
return false;
}
global->setCodeAddress(code->globals.size());
code->globals.push_back(target->m_code.globaladdr);
}
else
{
global->setCodeAddress(code->globals.size());
code->globals.push_back(0);
}
if (global->m_code.globaladdr < 0)
return false;
return true;
}
static bool gen_blocks_recursive(code_t *code, ir_function *func, ir_block *block)
{
prog_section_statement_t stmt;
ir_instr *instr;
ir_block *target;
ir_block *ontrue;
ir_block *onfalse;
size_t stidx;
size_t i;
int j;
block->m_generated = true;
block->m_code_start = code->statements.size();
for (i = 0; i < block->m_instr.size(); ++i)
{
instr = block->m_instr[i];
if (instr->m_opcode == VINSTR_PHI) {
irerror(block->m_context, "cannot generate virtual instruction (phi)");
return false;
}
if (instr->m_opcode == VINSTR_JUMP) {
target = instr->m_bops[0];
/* for uncoditional jumps, if the target hasn't been generated
* yet, we generate them right here.
*/
if (!target->m_generated)
return gen_blocks_recursive(code, func, target);
/* otherwise we generate a jump instruction */
stmt.opcode = INSTR_GOTO;
stmt.o1.s1 = target->m_code_start - code->statements.size();
stmt.o2.s1 = 0;
stmt.o3.s1 = 0;
if (stmt.o1.s1 != 1)
code_push_statement(code, &stmt, instr->m_context);
/* no further instructions can be in this block */
return true;
}
if (instr->m_opcode == VINSTR_BITXOR) {
stmt.opcode = INSTR_BITOR;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress();
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
stmt.opcode = INSTR_BITAND;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress();
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
stmt.opcode = INSTR_SUB_F;
stmt.o1.s1 = instr->_m_ops[0]->codeAddress();
stmt.o2.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_BITAND_V) {
stmt.opcode = INSTR_BITAND;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress();
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o2.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o2.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_BITOR_V) {
stmt.opcode = INSTR_BITOR;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress();
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o2.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o2.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_BITXOR_V) {
for (j = 0; j < 3; ++j) {
stmt.opcode = INSTR_BITOR;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress() + j;
stmt.o2.s1 = instr->_m_ops[2]->codeAddress() + j;
stmt.o3.s1 = instr->_m_ops[0]->codeAddress() + j;
code_push_statement(code, &stmt, instr->m_context);
stmt.opcode = INSTR_BITAND;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress() + j;
stmt.o2.s1 = instr->_m_ops[2]->codeAddress() + j;
stmt.o3.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress() + j;
code_push_statement(code, &stmt, instr->m_context);
}
stmt.opcode = INSTR_SUB_V;
stmt.o1.s1 = instr->_m_ops[0]->codeAddress();
stmt.o2.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_BITAND_VF) {
stmt.opcode = INSTR_BITAND;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress();
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_BITOR_VF) {
stmt.opcode = INSTR_BITOR;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress();
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
++stmt.o1.s1;
++stmt.o3.s1;
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_BITXOR_VF) {
for (j = 0; j < 3; ++j) {
stmt.opcode = INSTR_BITOR;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress() + j;
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress() + j;
code_push_statement(code, &stmt, instr->m_context);
stmt.opcode = INSTR_BITAND;
stmt.o1.s1 = instr->_m_ops[1]->codeAddress() + j;
stmt.o2.s1 = instr->_m_ops[2]->codeAddress();
stmt.o3.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress() + j;
code_push_statement(code, &stmt, instr->m_context);
}
stmt.opcode = INSTR_SUB_V;
stmt.o1.s1 = instr->_m_ops[0]->codeAddress();
stmt.o2.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_CROSS) {
stmt.opcode = INSTR_MUL_F;
for (j = 0; j < 3; ++j) {
stmt.o1.s1 = instr->_m_ops[1]->codeAddress() + (j + 1) % 3;
stmt.o2.s1 = instr->_m_ops[2]->codeAddress() + (j + 2) % 3;
stmt.o3.s1 = instr->_m_ops[0]->codeAddress() + j;
code_push_statement(code, &stmt, instr->m_context);
stmt.o1.s1 = instr->_m_ops[1]->codeAddress() + (j + 2) % 3;
stmt.o2.s1 = instr->_m_ops[2]->codeAddress() + (j + 1) % 3;
stmt.o3.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress() + j;
code_push_statement(code, &stmt, instr->m_context);
}
stmt.opcode = INSTR_SUB_V;
stmt.o1.s1 = instr->_m_ops[0]->codeAddress();
stmt.o2.s1 = func->m_owner->m_vinstr_temp[0]->codeAddress();
stmt.o3.s1 = instr->_m_ops[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
/* instruction generated */
continue;
}
if (instr->m_opcode == VINSTR_COND) {
ontrue = instr->m_bops[0];
onfalse = instr->m_bops[1];
/* TODO: have the AST signal which block should
* come first: eg. optimize IFs without ELSE...
*/
stmt.o1.u1 = instr->_m_ops[0]->codeAddress();
stmt.o2.u1 = 0;
stmt.o3.s1 = 0;
if (ontrue->m_generated) {
stmt.opcode = INSTR_IF;
stmt.o2.s1 = ontrue->m_code_start - code->statements.size();
if (stmt.o2.s1 != 1)
code_push_statement(code, &stmt, instr->m_context);
}
if (onfalse->m_generated) {
stmt.opcode = INSTR_IFNOT;
stmt.o2.s1 = onfalse->m_code_start - code->statements.size();
if (stmt.o2.s1 != 1)
code_push_statement(code, &stmt, instr->m_context);
}
if (!ontrue->m_generated) {
if (onfalse->m_generated)
return gen_blocks_recursive(code, func, ontrue);
}
if (!onfalse->m_generated) {
if (ontrue->m_generated)
return gen_blocks_recursive(code, func, onfalse);
}
/* neither ontrue nor onfalse exist */
stmt.opcode = INSTR_IFNOT;
if (!instr->m_likely) {
/* Honor the likelyhood hint */
ir_block *tmp = onfalse;
stmt.opcode = INSTR_IF;
onfalse = ontrue;
ontrue = tmp;
}
stidx = code->statements.size();
code_push_statement(code, &stmt, instr->m_context);
/* on false we jump, so add ontrue-path */
if (!gen_blocks_recursive(code, func, ontrue))
return false;
/* fixup the jump address */
code->statements[stidx].o2.s1 = code->statements.size() - stidx;
/* generate onfalse path */
if (onfalse->m_generated) {
/* fixup the jump address */
code->statements[stidx].o2.s1 = onfalse->m_code_start - stidx;
if (stidx+2 == code->statements.size() && code->statements[stidx].o2.s1 == 1) {
code->statements[stidx] = code->statements[stidx+1];
if (code->statements[stidx].o1.s1 < 0)
code->statements[stidx].o1.s1++;
code_pop_statement(code);
}
stmt.opcode = code->statements.back().opcode;
if (stmt.opcode == INSTR_GOTO ||
stmt.opcode == INSTR_IF ||
stmt.opcode == INSTR_IFNOT ||
stmt.opcode == INSTR_RETURN ||
stmt.opcode == INSTR_DONE)
{
/* no use jumping from here */
return true;
}
/* may have been generated in the previous recursive call */
stmt.opcode = INSTR_GOTO;
stmt.o1.s1 = onfalse->m_code_start - code->statements.size();
stmt.o2.s1 = 0;
stmt.o3.s1 = 0;
if (stmt.o1.s1 != 1)
code_push_statement(code, &stmt, instr->m_context);
return true;
}
else if (stidx+2 == code->statements.size() && code->statements[stidx].o2.s1 == 1) {
code->statements[stidx] = code->statements[stidx+1];
if (code->statements[stidx].o1.s1 < 0)
code->statements[stidx].o1.s1++;
code_pop_statement(code);
}
/* if not, generate now */
return gen_blocks_recursive(code, func, onfalse);
}
if ( (instr->m_opcode >= INSTR_CALL0 && instr->m_opcode <= INSTR_CALL8)
|| instr->m_opcode == VINSTR_NRCALL)
{
size_t p, first;
ir_value *retvalue;
first = instr->m_params.size();
if (first > 8)
first = 8;
for (p = 0; p < first; ++p)
{
ir_value *param = instr->m_params[p];
if (param->m_callparam)
continue;
stmt.opcode = INSTR_STORE_F;
stmt.o3.u1 = 0;
if (param->m_vtype == TYPE_FIELD)
stmt.opcode = field_store_instr[param->m_fieldtype];
else if (param->m_vtype == TYPE_NIL)
stmt.opcode = INSTR_STORE_V;
else
stmt.opcode = type_store_instr[param->m_vtype];
stmt.o1.u1 = param->codeAddress();
stmt.o2.u1 = OFS_PARM0 + 3 * p;
if (param->m_vtype == TYPE_VECTOR && (param->m_flags & IR_FLAG_SPLIT_VECTOR)) {
/* fetch 3 separate floats */
stmt.opcode = INSTR_STORE_F;
stmt.o1.u1 = param->m_members[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
stmt.o2.u1++;
stmt.o1.u1 = param->m_members[1]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
stmt.o2.u1++;
stmt.o1.u1 = param->m_members[2]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
}
else
code_push_statement(code, &stmt, instr->m_context);
}
/* Now handle extparams */
first = instr->m_params.size();
for (; p < first; ++p)
{
ir_builder *ir = func->m_owner;
ir_value *param = instr->m_params[p];
ir_value *targetparam;
if (param->m_callparam)
continue;
if (p-8 >= ir->m_extparams.size())
ir->generateExtparam();
targetparam = ir->m_extparams[p-8];
stmt.opcode = INSTR_STORE_F;
stmt.o3.u1 = 0;
if (param->m_vtype == TYPE_FIELD)
stmt.opcode = field_store_instr[param->m_fieldtype];
else if (param->m_vtype == TYPE_NIL)
stmt.opcode = INSTR_STORE_V;
else
stmt.opcode = type_store_instr[param->m_vtype];
stmt.o1.u1 = param->codeAddress();
stmt.o2.u1 = targetparam->codeAddress();
if (param->m_vtype == TYPE_VECTOR && (param->m_flags & IR_FLAG_SPLIT_VECTOR)) {
/* fetch 3 separate floats */
stmt.opcode = INSTR_STORE_F;
stmt.o1.u1 = param->m_members[0]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
stmt.o2.u1++;
stmt.o1.u1 = param->m_members[1]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
stmt.o2.u1++;
stmt.o1.u1 = param->m_members[2]->codeAddress();
code_push_statement(code, &stmt, instr->m_context);
}
else
code_push_statement(code, &stmt, instr->m_context);
}
stmt.opcode = INSTR_CALL0 + instr->m_params.size();
if (stmt.opcode > INSTR_CALL8)
stmt.opcode = INSTR_CALL8;
stmt.o1.u1 = instr->_m_ops[1]->codeAddress();
stmt.o2.u1 = 0;
stmt.o3.u1 = 0;
code_push_statement(code, &stmt, instr->m_context);
retvalue = instr->_m_ops[0];
if (retvalue && retvalue->m_store != store_return &&
(retvalue->m_store == store_global || retvalue->m_life.size()))
{
/* not to be kept in OFS_RETURN */
if (retvalue->m_vtype == TYPE_FIELD && OPTS_FLAG(ADJUST_VECTOR_FIELDS))
stmt.opcode = field_store_instr[retvalue->m_fieldtype];
else
stmt.opcode = type_store_instr[retvalue->m_vtype];
stmt.o1.u1 = OFS_RETURN;
stmt.o2.u1 = retvalue->codeAddress();
stmt.o3.u1 = 0;
code_push_statement(code, &stmt, instr->m_context);
}
continue;
}
if (instr->m_opcode == INSTR_STATE) {
stmt.opcode = instr->m_opcode;
if (instr->_m_ops[0])
stmt.o1.u1 = instr->_m_ops[0]->codeAddress();
if (instr->_m_ops[1])
stmt.o2.u1 = instr->_m_ops[1]->codeAddress();
stmt.o3.u1 = 0;
code_push_statement(code, &stmt, instr->m_context);
continue;
}
stmt.opcode = instr->m_opcode;
stmt.o1.u1 = 0;
stmt.o2.u1 = 0;
stmt.o3.u1 = 0;
/* This is the general order of operands */
if (instr->_m_ops[0])
stmt.o3.u1 = instr->_m_ops[0]->codeAddress();
if (instr->_m_ops[1])
stmt.o1.u1 = instr->_m_ops[1]->codeAddress();
if (instr->_m_ops[2])
stmt.o2.u1 = instr->_m_ops[2]->codeAddress();
if (stmt.opcode == INSTR_RETURN || stmt.opcode == INSTR_DONE)
{
stmt.o1.u1 = stmt.o3.u1;
stmt.o3.u1 = 0;
}
else if ((stmt.opcode >= INSTR_STORE_F &&
stmt.opcode <= INSTR_STORE_FNC) ||
(stmt.opcode >= INSTR_STOREP_F &&
stmt.opcode <= INSTR_STOREP_FNC))
{
/* 2-operand instructions with A -> B */
stmt.o2.u1 = stmt.o3.u1;
stmt.o3.u1 = 0;
/* tiny optimization, don't output
* STORE a, a
*/
if (stmt.o2.u1 == stmt.o1.u1 &&
OPTS_OPTIMIZATION(OPTIM_PEEPHOLE))
{
++opts_optimizationcount[OPTIM_PEEPHOLE];
continue;
}
}
code_push_statement(code, &stmt, instr->m_context);
}
return true;
}
static bool gen_function_code(code_t *code, ir_function *self)
{
ir_block *block;
prog_section_statement_t stmt, *retst;
/* Starting from entry point, we generate blocks "as they come"
* for now. Dead blocks will not be translated obviously.
*/
if (self->m_blocks.empty()) {
irerror(self->m_context, "Function '%s' declared without body.", self->m_name.c_str());
return false;
}
block = self->m_blocks[0].get();
if (block->m_generated)
return true;
if (!gen_blocks_recursive(code, self, block)) {
irerror(self->m_context, "failed to generate blocks for '%s'", self->m_name.c_str());
return false;
}
/* code_write and qcvm -disasm need to know that the function ends here */
retst = &code->statements.back();
if (OPTS_OPTIMIZATION(OPTIM_VOID_RETURN) &&
self->m_outtype == TYPE_VOID &&
retst->opcode == INSTR_RETURN &&
!retst->o1.u1 && !retst->o2.u1 && !retst->o3.u1)
{
retst->opcode = INSTR_DONE;
++opts_optimizationcount[OPTIM_VOID_RETURN];
} else {
lex_ctx_t last;
stmt.opcode = INSTR_DONE;
stmt.o1.u1 = 0;
stmt.o2.u1 = 0;
stmt.o3.u1 = 0;
last.line = code->linenums.back();
last.column = code->columnnums.back();
code_push_statement(code, &stmt, last);
}
return true;
}
qcint_t ir_builder::filestring(const char *filename)
{
/* NOTE: filename pointers are copied, we never strdup them,
* thus we can use pointer-comparison to find the string.
*/
qcint_t str;
for (size_t i = 0; i != m_filenames.size(); ++i) {
if (!strcmp(m_filenames[i], filename))
return i;
}
str = code_genstring(m_code.get(), filename);
m_filenames.push_back(filename);
m_filestrings.push_back(str);
return str;
}
bool ir_builder::generateGlobalFunction(ir_value *global)
{
prog_section_function_t fun;
ir_function *irfun;
size_t i;
if (!global->m_hasvalue || (!global->m_constval.vfunc)) {
irerror(global->m_context, "Invalid state of function-global: not constant: %s", global->m_name.c_str());
return false;
}
irfun = global->m_constval.vfunc;
fun.name = global->m_code.name;
fun.file = filestring(global->m_context.file);
fun.profile = 0; /* always 0 */
fun.nargs = irfun->m_params.size();
if (fun.nargs > 8)
fun.nargs = 8;
for (i = 0; i < 8; ++i) {
if ((int32_t)i >= fun.nargs)
fun.argsize[i] = 0;
else
fun.argsize[i] = type_sizeof_[irfun->m_params[i]];
}
fun.firstlocal = 0;
fun.locals = irfun->m_allocated_locals;
if (irfun->m_builtin)
fun.entry = irfun->m_builtin+1;
else {
irfun->m_code_function_def = m_code->functions.size();
fun.entry = m_code->statements.size();
}
m_code->functions.push_back(fun);
return true;
}
ir_value* ir_builder::generateExtparamProto()
{
char name[128];
util_snprintf(name, sizeof(name), "EXTPARM#%i", (int)(m_extparam_protos.size()));
ir_value *global = new ir_value(name, store_global, TYPE_VECTOR);
m_extparam_protos.emplace_back(global);
return global;
}
void ir_builder::generateExtparam()
{
prog_section_def_t def;
ir_value *global;
if (m_extparam_protos.size() < m_extparams.size()+1)
global = generateExtparamProto();
else
global = m_extparam_protos[m_extparams.size()].get();
def.name = code_genstring(m_code.get(), global->m_name.c_str());
def.type = TYPE_VECTOR;
def.offset = m_code->globals.size();
m_code->defs.push_back(def);
global->setCodeAddress(def.offset);
m_code->globals.push_back(0);
m_code->globals.push_back(0);
m_code->globals.push_back(0);
m_extparams.emplace_back(global);
}
static bool gen_function_extparam_copy(code_t *code, ir_function *self)
{
ir_builder *ir = self->m_owner;
size_t numparams = self->m_params.size();
if (!numparams)
return true;
prog_section_statement_t stmt;
stmt.opcode = INSTR_STORE_F;
stmt.o3.s1 = 0;
for (size_t i = 8; i < numparams; ++i) {
size_t ext = i - 8;
if (ext >= ir->m_extparams.size())
ir->generateExtparam();
ir_value *ep = ir->m_extparams[ext];
stmt.opcode = type_store_instr[self->m_locals[i]->m_vtype];
if (self->m_locals[i]->m_vtype == TYPE_FIELD &&
self->m_locals[i]->m_fieldtype == TYPE_VECTOR)
{
stmt.opcode = INSTR_STORE_V;
}
stmt.o1.u1 = ep->codeAddress();
stmt.o2.u1 = self->m_locals[i].get()->codeAddress();
code_push_statement(code, &stmt, self->m_context);
}
return true;
}
static bool gen_function_varargs_copy(code_t *code, ir_function *self)
{
size_t i, ext, numparams, maxparams;
ir_builder *ir = self->m_owner;
ir_value *ep;
prog_section_statement_t stmt;
numparams = self->m_params.size();
if (!numparams)
return true;
stmt.opcode = INSTR_STORE_V;
stmt.o3.s1 = 0;
maxparams = numparams + self->m_max_varargs;
for (i = numparams; i < maxparams; ++i) {
if (i < 8) {
stmt.o1.u1 = OFS_PARM0 + 3*i;
stmt.o2.u1 = self->m_locals[i].get()->codeAddress();
code_push_statement(code, &stmt, self->m_context);
continue;
}
ext = i - 8;
while (ext >= ir->m_extparams.size())
ir->generateExtparam();
ep = ir->m_extparams[ext];
stmt.o1.u1 = ep->codeAddress();
stmt.o2.u1 = self->m_locals[i].get()->codeAddress();
code_push_statement(code, &stmt, self->m_context);
}
return true;
}
bool ir_builder::generateFunctionLocals(ir_value *global)
{
prog_section_function_t *def;
ir_function *irfun;
uint32_t firstlocal, firstglobal;
irfun = global->m_constval.vfunc;
def = &m_code->functions[0] + irfun->m_code_function_def;
if (OPTS_OPTION_BOOL(OPTION_G) ||
!OPTS_OPTIMIZATION(OPTIM_OVERLAP_LOCALS) ||
(irfun->m_flags & IR_FLAG_MASK_NO_OVERLAP))
{
firstlocal = def->firstlocal = m_code->globals.size();
} else {
firstlocal = def->firstlocal = m_first_common_local;
++opts_optimizationcount[OPTIM_OVERLAP_LOCALS];
}
firstglobal = (OPTS_OPTIMIZATION(OPTIM_GLOBAL_TEMPS) ? m_first_common_globaltemp : firstlocal);
for (size_t i = m_code->globals.size(); i < firstlocal + irfun->m_allocated_locals; ++i)
m_code->globals.push_back(0);
for (auto& lp : irfun->m_locals) {
ir_value *v = lp.get();
if (v->m_locked || !OPTS_OPTIMIZATION(OPTIM_GLOBAL_TEMPS)) {
v->setCodeAddress(firstlocal + v->m_code.local);
if (!generateGlobal(v, true)) {
irerror(v->m_context, "failed to generate local %s", v->m_name.c_str());
return false;
}
}
else
v->setCodeAddress(firstglobal + v->m_code.local);
}
for (auto& vp : irfun->m_values) {
ir_value *v = vp.get();
if (v->m_callparam)
continue;
if (v->m_locked)
v->setCodeAddress(firstlocal + v->m_code.local);
else
v->setCodeAddress(firstglobal + v->m_code.local);
}
return true;
}
bool ir_builder::generateGlobalFunctionCode(ir_value *global)
{
prog_section_function_t *fundef;
ir_function *irfun;
irfun = global->m_constval.vfunc;
if (!irfun) {
if (global->m_cvq == CV_NONE) {
if (irwarning(global->m_context, WARN_IMPLICIT_FUNCTION_POINTER,
"function `%s` has no body and in QC implicitly becomes a function-pointer",
global->m_name.c_str()))
{
/* Not bailing out just now. If this happens a lot you don't want to have
* to rerun gmqcc for each such function.
*/
/* return false; */
}
}
/* this was a function pointer, don't generate code for those */
return true;
}
if (irfun->m_builtin)
return true;
/*
* If there is no definition and the thing is eraseable, we can ignore
* outputting the function to begin with.
*/
if (global->m_flags & IR_FLAG_ERASABLE && irfun->m_code_function_def < 0) {
return true;
}
if (irfun->m_code_function_def < 0) {
irerror(irfun->m_context, "`%s`: IR global wasn't generated, failed to access function-def", irfun->m_name.c_str());
return false;
}
fundef = &m_code->functions[irfun->m_code_function_def];
fundef->entry = m_code->statements.size();
if (!generateFunctionLocals(global)) {
irerror(irfun->m_context, "Failed to generate locals for function %s", irfun->m_name.c_str());
return false;
}
if (!gen_function_extparam_copy(m_code.get(), irfun)) {
irerror(irfun->m_context, "Failed to generate extparam-copy code for function %s", irfun->m_name.c_str());
return false;
}
if (irfun->m_max_varargs && !gen_function_varargs_copy(m_code.get(), irfun)) {
irerror(irfun->m_context, "Failed to generate vararg-copy code for function %s", irfun->m_name.c_str());
return false;
}
if (!gen_function_code(m_code.get(), irfun)) {
irerror(irfun->m_context, "Failed to generate code for function %s", irfun->m_name.c_str());
return false;
}
return true;
}
static void gen_vector_defs(code_t *code, prog_section_def_t def, const char *name, int type)
{
char *component;
size_t len, i;
if (!name || name[0] == '#' || OPTS_FLAG(SINGLE_VECTOR_DEFS))
return;
def.type = type;
len = strlen(name);
component = (char*)mem_a(len+3);
memcpy(component, name, len);
len += 2;
component[len-0] = 0;
component[len-2] = '_';
component[len-1] = 'x';
for (i = 0; i < 3; ++i) {
def.name = code_genstring(code, component);
code->defs.push_back(def);
def.offset++;
component[len-1]++;
}
mem_d(component);
}
static void gen_vector_fields(code_t *code, prog_section_field_t fld, const char *name)
{
char *component;
size_t len, i;
if (!name || OPTS_FLAG(SINGLE_VECTOR_DEFS))
return;
fld.type = TYPE_FLOAT;
len = strlen(name);
component = (char*)mem_a(len+3);
memcpy(component, name, len);
len += 2;
component[len-0] = 0;
component[len-2] = '_';
component[len-1] = 'x';
for (i = 0; i < 3; ++i) {
fld.name = code_genstring(code, component);
code->fields.push_back(fld);
fld.offset++;
component[len-1]++;
}
mem_d(component);
}
bool ir_builder::generateGlobal(ir_value *global, bool islocal)
{
size_t i;
int32_t *iptr;
prog_section_def_t def;
bool pushdef = opts.optimizeoff;
/* we don't generate split-vectors */
if (global->m_vtype == TYPE_VECTOR && (global->m_flags & IR_FLAG_SPLIT_VECTOR))
return true;
def.type = global->m_vtype;
def.offset = m_code->globals.size();
def.name = 0;
if (OPTS_OPTION_BOOL(OPTION_G) || !islocal)
{
pushdef = true;
/*
* if we're eraseable and the function isn't referenced ignore outputting
* the function.
*/
if (global->m_flags & IR_FLAG_ERASABLE && global->m_reads.empty()) {
return true;
}
if (OPTS_OPTIMIZATION(OPTIM_STRIP_CONSTANT_NAMES) &&
!(global->m_flags & IR_FLAG_INCLUDE_DEF) &&
(global->m_name[0] == '#' || global->m_cvq == CV_CONST))
{
pushdef = false;
}
if (pushdef) {
if (global->m_name[0] == '#') {
if (!m_str_immediate)
m_str_immediate = code_genstring(m_code.get(), "IMMEDIATE");
def.name = global->m_code.name = m_str_immediate;
}
else
def.name = global->m_code.name = code_genstring(m_code.get(), global->m_name.c_str());
}
else
def.name = 0;
if (islocal) {
def.offset = global->codeAddress();
m_code->defs.push_back(def);
if (global->m_vtype == TYPE_VECTOR)
gen_vector_defs(m_code.get(), def, global->m_name.c_str(), TYPE_FLOAT);
else if (global->m_vtype == TYPE_FIELD && global->m_fieldtype == TYPE_VECTOR)
gen_vector_defs(m_code.get(), def, global->m_name.c_str(), TYPE_FIELD);
return true;
}
}
if (islocal)
return true;
switch (global->m_vtype)
{
case TYPE_VOID:
if (0 == global->m_name.compare("end_sys_globals")) {
// TODO: remember this point... all the defs before this one
// should be checksummed and added to progdefs.h when we generate it.
}
else if (0 == global->m_name.compare("end_sys_fields")) {
// TODO: same as above but for entity-fields rather than globsl
}
else if(irwarning(global->m_context, WARN_VOID_VARIABLES, "unrecognized variable of type void `%s`",
global->m_name.c_str()))
{
/* Not bailing out */
/* return false; */
}
/* I'd argue setting it to 0 is sufficient, but maybe some depend on knowing how far
* the system fields actually go? Though the engine knows this anyway...
* Maybe this could be an -foption
* fteqcc creates data for end_sys_* - of size 1, so let's do the same
*/
global->setCodeAddress(m_code->globals.size());
m_code->globals.push_back(0);
/* Add the def */
if (pushdef)
m_code->defs.push_back(def);
return true;
case TYPE_POINTER:
if (pushdef)
m_code->defs.push_back(def);
return gen_global_pointer(m_code.get(), global);
case TYPE_FIELD:
if (pushdef) {
m_code->defs.push_back(def);
if (global->m_fieldtype == TYPE_VECTOR)
gen_vector_defs(m_code.get(), def, global->m_name.c_str(), TYPE_FIELD);
}
return gen_global_field(m_code.get(), global);
case TYPE_ENTITY:
/* fall through */
case TYPE_FLOAT:
{
global->setCodeAddress(m_code->globals.size());
if (global->m_hasvalue) {
if (global->m_cvq == CV_CONST && global->m_reads.empty())
return true;
iptr = (int32_t*)&global->m_constval.ivec[0];
m_code->globals.push_back(*iptr);
} else {
m_code->globals.push_back(0);
}
if (!islocal && global->m_cvq != CV_CONST)
def.type |= DEF_SAVEGLOBAL;
if (pushdef)
m_code->defs.push_back(def);
return global->m_code.globaladdr >= 0;
}
case TYPE_STRING:
{
global->setCodeAddress(m_code->globals.size());
if (global->m_hasvalue) {
if (global->m_cvq == CV_CONST && global->m_reads.empty())
return true;
uint32_t load = code_genstring(m_code.get(), global->m_constval.vstring);
m_code->globals.push_back(load);
} else {
m_code->globals.push_back(0);
}
if (!islocal && global->m_cvq != CV_CONST)
def.type |= DEF_SAVEGLOBAL;
if (pushdef)
m_code->defs.push_back(def);
return global->m_code.globaladdr >= 0;
}
case TYPE_VECTOR:
{
size_t d;
global->setCodeAddress(m_code->globals.size());
if (global->m_hasvalue) {
iptr = (int32_t*)&global->m_constval.ivec[0];
m_code->globals.push_back(iptr[0]);
if (global->m_code.globaladdr < 0)
return false;
for (d = 1; d < type_sizeof_[global->m_vtype]; ++d) {
m_code->globals.push_back(iptr[d]);
}
} else {
m_code->globals.push_back(0);
if (global->m_code.globaladdr < 0)
return false;
for (d = 1; d < type_sizeof_[global->m_vtype]; ++d) {
m_code->globals.push_back(0);
}
}
if (!islocal && global->m_cvq != CV_CONST)
def.type |= DEF_SAVEGLOBAL;
if (pushdef) {
m_code->defs.push_back(def);
def.type &= ~DEF_SAVEGLOBAL;
gen_vector_defs(m_code.get(), def, global->m_name.c_str(), TYPE_FLOAT);
}
return global->m_code.globaladdr >= 0;
}
case TYPE_FUNCTION:
global->setCodeAddress(m_code->globals.size());
if (!global->m_hasvalue) {
m_code->globals.push_back(0);
if (global->m_code.globaladdr < 0)
return false;
} else {
m_code->globals.push_back(m_code->functions.size());
if (!generateGlobalFunction(global))
return false;
}
if (!islocal && global->m_cvq != CV_CONST)
def.type |= DEF_SAVEGLOBAL;
if (pushdef)
m_code->defs.push_back(def);
return true;
case TYPE_VARIANT:
/* assume biggest type */
global->setCodeAddress(m_code->globals.size());
m_code->globals.push_back(0);
for (i = 1; i < type_sizeof_[TYPE_VARIANT]; ++i)
m_code->globals.push_back(0);
return true;
default:
/* refuse to create 'void' type or any other fancy business. */
irerror(global->m_context, "Invalid type for global variable `%s`: %s",
global->m_name.c_str(), type_name[global->m_vtype]);
return false;
}
}
static GMQCC_INLINE void ir_builder_prepare_field(code_t *code, ir_value *field)
{
field->m_code.fieldaddr = code_alloc_field(code, type_sizeof_[field->m_fieldtype]);
}
static bool ir_builder_gen_field(ir_builder *self, ir_value *field)
{
prog_section_def_t def;
prog_section_field_t fld;
(void)self;
def.type = (uint16_t)field->m_vtype;
def.offset = (uint16_t)self->m_code->globals.size();
/* create a global named the same as the field */
if (OPTS_OPTION_U32(OPTION_STANDARD) == COMPILER_GMQCC) {
/* in our standard, the global gets a dot prefix */
size_t len = field->m_name.length();
char name[1024];
/* we really don't want to have to allocate this, and 1024
* bytes is more than enough for a variable/field name
*/
if (len+2 >= sizeof(name)) {
irerror(field->m_context, "invalid field name size: %u", (unsigned int)len);
return false;
}
name[0] = '.';
memcpy(name+1, field->m_name.c_str(), len); // no strncpy - we used strlen above
name[len+1] = 0;
def.name = code_genstring(self->m_code.get(), name);
fld.name = def.name + 1; /* we reuse that string table entry */
} else {
/* in plain QC, there cannot be a global with the same name,
* and so we also name the global the same.
* FIXME: fteqcc should create a global as well
* check if it actually uses the same name. Probably does
*/
def.name = code_genstring(self->m_code.get(), field->m_name.c_str());
fld.name = def.name;
}
field->m_code.name = def.name;
self->m_code->defs.push_back(def);
fld.type = field->m_fieldtype;
if (fld.type == TYPE_VOID) {
irerror(field->m_context, "field is missing a type: %s - don't know its size", field->m_name.c_str());
return false;
}
fld.offset = field->m_code.fieldaddr;
self->m_code->fields.push_back(fld);
field->setCodeAddress(self->m_code->globals.size());
self->m_code->globals.push_back(fld.offset);
if (fld.type == TYPE_VECTOR) {
self->m_code->globals.push_back(fld.offset+1);
self->m_code->globals.push_back(fld.offset+2);
}
if (field->m_fieldtype == TYPE_VECTOR) {
gen_vector_defs (self->m_code.get(), def, field->m_name.c_str(), TYPE_FIELD);
gen_vector_fields(self->m_code.get(), fld, field->m_name.c_str());
}
return field->m_code.globaladdr >= 0;
}
static void ir_builder_collect_reusables(ir_builder *builder) {
std::vector<ir_value*> reusables;
for (auto& gp : builder->m_globals) {
ir_value *value = gp.get();
if (value->m_vtype != TYPE_FLOAT || !value->m_hasvalue)
continue;
if (value->m_cvq == CV_CONST || (value->m_name.length() >= 1 && value->m_name[0] == '#'))
reusables.emplace_back(value);
}
builder->m_const_floats = move(reusables);
}
static void ir_builder_split_vector(ir_builder *self, ir_value *vec) {
ir_value* found[3] = { nullptr, nullptr, nullptr };
// must not be written to
if (vec->m_writes.size())
return;
// must not be trying to access individual members
if (vec->m_members[0] || vec->m_members[1] || vec->m_members[2])
return;
// should be actually used otherwise it won't be generated anyway
if (vec->m_reads.empty())
return;
//size_t count = vec->m_reads.size();
//if (!count)
// return;
// may only be used directly as function parameters, so if we find some other instruction cancel
for (ir_instr *user : vec->m_reads) {
// we only split vectors if they're used directly as parameter to a call only!
if ((user->m_opcode < INSTR_CALL0 || user->m_opcode > INSTR_CALL8) && user->m_opcode != VINSTR_NRCALL)
return;
}
vec->m_flags |= IR_FLAG_SPLIT_VECTOR;
// find existing floats making up the split
for (ir_value *c : self->m_const_floats) {
if (!found[0] && c->m_constval.vfloat == vec->m_constval.vvec.x)
found[0] = c;
if (!found[1] && c->m_constval.vfloat == vec->m_constval.vvec.y)
found[1] = c;
if (!found[2] && c->m_constval.vfloat == vec->m_constval.vvec.z)
found[2] = c;
if (found[0] && found[1] && found[2])
break;
}
// generate floats for not yet found components
if (!found[0])
found[0] = self->literalFloat(vec->m_constval.vvec.x, true);
if (!found[1]) {
if (vec->m_constval.vvec.y == vec->m_constval.vvec.x)
found[1] = found[0];
else
found[1] = self->literalFloat(vec->m_constval.vvec.y, true);
}
if (!found[2]) {
if (vec->m_constval.vvec.z == vec->m_constval.vvec.x)
found[2] = found[0];
else if (vec->m_constval.vvec.z == vec->m_constval.vvec.y)
found[2] = found[1];
else
found[2] = self->literalFloat(vec->m_constval.vvec.z, true);
}
// the .members array should be safe to use here
vec->m_members[0] = found[0];
vec->m_members[1] = found[1];
vec->m_members[2] = found[2];
// register the readers for these floats
found[0]->m_reads.insert(found[0]->m_reads.end(), vec->m_reads.begin(), vec->m_reads.end());
found[1]->m_reads.insert(found[1]->m_reads.end(), vec->m_reads.begin(), vec->m_reads.end());
found[2]->m_reads.insert(found[2]->m_reads.end(), vec->m_reads.begin(), vec->m_reads.end());
}
static void ir_builder_split_vectors(ir_builder *self) {
// member values may be added to self->m_globals during this operation, but
// no new vectors will be added, we need to iterate via an index as
// c++ iterators would be invalidated
const size_t count = self->m_globals.size();
for (size_t i = 0; i != count; ++i) {
ir_value *v = self->m_globals[i].get();
if (v->m_vtype != TYPE_VECTOR || !v->m_name.length() || v->m_name[0] != '#')
continue;
ir_builder_split_vector(self, v);
}
}
bool ir_builder::generate(const char *filename)
{
prog_section_statement_t stmt;
char *lnofile = nullptr;
if (OPTS_FLAG(SPLIT_VECTOR_PARAMETERS)) {
ir_builder_collect_reusables(this);
if (!m_const_floats.empty())
ir_builder_split_vectors(this);
}
for (auto& fp : m_fields)
ir_builder_prepare_field(m_code.get(), fp.get());
for (auto& gp : m_globals) {
ir_value *global = gp.get();
if (!generateGlobal(global, false)) {
return false;
}
if (global->m_vtype == TYPE_FUNCTION) {
ir_function *func = global->m_constval.vfunc;
if (func && m_max_locals < func->m_allocated_locals &&
!(func->m_flags & IR_FLAG_MASK_NO_OVERLAP))
{
m_max_locals = func->m_allocated_locals;
}
if (func && m_max_globaltemps < func->m_globaltemps)
m_max_globaltemps = func->m_globaltemps;
}
}
for (auto& fp : m_fields) {
if (!ir_builder_gen_field(this, fp.get()))
return false;
}
// generate nil
m_nil->setCodeAddress(m_code->globals.size());
m_code->globals.push_back(0);
m_code->globals.push_back(0);
m_code->globals.push_back(0);
// generate virtual-instruction temps
for (size_t i = 0; i < IR_MAX_VINSTR_TEMPS; ++i) {
m_vinstr_temp[i]->setCodeAddress(m_code->globals.size());
m_code->globals.push_back(0);
m_code->globals.push_back(0);
m_code->globals.push_back(0);
}
// generate global temps
m_first_common_globaltemp = m_code->globals.size();
m_code->globals.insert(m_code->globals.end(), m_max_globaltemps, 0);
// FIXME:DELME:
//for (size_t i = 0; i < m_max_globaltemps; ++i) {
// m_code->globals.push_back(0);
//}
// generate common locals
m_first_common_local = m_code->globals.size();
m_code->globals.insert(m_code->globals.end(), m_max_locals, 0);
// FIXME:DELME:
//for (i = 0; i < m_max_locals; ++i) {
// m_code->globals.push_back(0);
//}
// generate function code
for (auto& gp : m_globals) {
ir_value *global = gp.get();
if (global->m_vtype == TYPE_FUNCTION) {
if (!this->generateGlobalFunctionCode(global))
return false;
}
}
if (m_code->globals.size() >= 65536) {
irerror(m_globals.back()->m_context,
"This progs file would require more globals than the metadata can handle (%zu). Bailing out.",
m_code->globals.size());
return false;
}
/* DP errors if the last instruction is not an INSTR_DONE. */
if (m_code->statements.back().opcode != INSTR_DONE)
{
lex_ctx_t last;
stmt.opcode = INSTR_DONE;
stmt.o1.u1 = 0;
stmt.o2.u1 = 0;
stmt.o3.u1 = 0;
last.line = m_code->linenums.back();
last.column = m_code->columnnums.back();
code_push_statement(m_code.get(), &stmt, last);
}
if (OPTS_OPTION_BOOL(OPTION_PP_ONLY))
return true;
if (m_code->statements.size() != m_code->linenums.size()) {
con_err("Linecounter wrong: %lu != %lu\n",
m_code->statements.size(),
m_code->linenums.size());
} else if (OPTS_FLAG(LNO)) {
char *dot;
size_t filelen = strlen(filename);
memcpy(vec_add(lnofile, filelen+1), filename, filelen+1);
dot = strrchr(lnofile, '.');
if (!dot) {
vec_pop(lnofile);
} else {
vec_shrinkto(lnofile, dot - lnofile);
}
memcpy(vec_add(lnofile, 5), ".lno", 5);
}
if (!code_write(m_code.get(), filename, lnofile)) {
vec_free(lnofile);
return false;
}
vec_free(lnofile);
return true;
}
/***********************************************************************
*IR DEBUG Dump functions...
*/
#define IND_BUFSZ 1024
static const char *qc_opname(int op)
{
if (op < 0) return "<INVALID>";
if (op < VINSTR_END)
return util_instr_str[op];
switch (op) {
case VINSTR_END: return "END";
case VINSTR_PHI: return "PHI";
case VINSTR_JUMP: return "JUMP";
case VINSTR_COND: return "COND";
case VINSTR_BITXOR: return "BITXOR";
case VINSTR_BITAND_V: return "BITAND_V";
case VINSTR_BITOR_V: return "BITOR_V";
case VINSTR_BITXOR_V: return "BITXOR_V";
case VINSTR_BITAND_VF: return "BITAND_VF";
case VINSTR_BITOR_VF: return "BITOR_VF";
case VINSTR_BITXOR_VF: return "BITXOR_VF";
case VINSTR_CROSS: return "CROSS";
case VINSTR_NEG_F: return "NEG_F";
case VINSTR_NEG_V: return "NEG_V";
default: return "<UNK>";
}
}
void ir_builder::dump(int (*oprintf)(const char*, ...)) const
{
size_t i;
char indent[IND_BUFSZ];
indent[0] = '\t';
indent[1] = 0;
oprintf("module %s\n", m_name.c_str());
for (i = 0; i < m_globals.size(); ++i)
{
oprintf("global ");
if (m_globals[i]->m_hasvalue)
oprintf("%s = ", m_globals[i]->m_name.c_str());
m_globals[i].get()->dump(oprintf);
oprintf("\n");
}
for (i = 0; i < m_functions.size(); ++i)
ir_function_dump(m_functions[i].get(), indent, oprintf);
oprintf("endmodule %s\n", m_name.c_str());
}
static const char *storenames[] = {
"[global]", "[local]", "[param]", "[value]", "[return]"
};
void ir_function_dump(ir_function *f, char *ind,
int (*oprintf)(const char*, ...))
{
size_t i;
if (f->m_builtin != 0) {
oprintf("%sfunction %s = builtin %i\n", ind, f->m_name.c_str(), -f->m_builtin);
return;
}
oprintf("%sfunction %s\n", ind, f->m_name.c_str());
util_strncat(ind, "\t", IND_BUFSZ-1);
if (f->m_locals.size())
{
oprintf("%s%i locals:\n", ind, (int)f->m_locals.size());
for (i = 0; i < f->m_locals.size(); ++i) {
oprintf("%s\t", ind);
f->m_locals[i].get()->dump(oprintf);
oprintf("\n");
}
}
oprintf("%sliferanges:\n", ind);
for (i = 0; i < f->m_locals.size(); ++i) {
const char *attr = "";
size_t l, m;
ir_value *v = f->m_locals[i].get();
if (v->m_unique_life && v->m_locked)
attr = "unique,locked ";
else if (v->m_unique_life)
attr = "unique ";
else if (v->m_locked)
attr = "locked ";
oprintf("%s\t%s: %s %s %s%s@%i ", ind, v->m_name.c_str(), type_name[v->m_vtype],
storenames[v->m_store],
attr, (v->m_callparam ? "callparam " : ""),
(int)v->m_code.local);
if (v->m_life.empty())
oprintf("[null]");
for (l = 0; l < v->m_life.size(); ++l) {
oprintf("[%i,%i] ", v->m_life[l].start, v->m_life[l].end);
}
oprintf("\n");
for (m = 0; m < 3; ++m) {
ir_value *vm = v->m_members[m];
if (!vm)
continue;
oprintf("%s\t%s: @%i ", ind, vm->m_name.c_str(), (int)vm->m_code.local);
for (l = 0; l < vm->m_life.size(); ++l) {
oprintf("[%i,%i] ", vm->m_life[l].start, vm->m_life[l].end);
}
oprintf("\n");
}
}
for (i = 0; i < f->m_values.size(); ++i) {
const char *attr = "";
size_t l, m;
ir_value *v = f->m_values[i].get();
if (v->m_unique_life && v->m_locked)
attr = "unique,locked ";
else if (v->m_unique_life)
attr = "unique ";
else if (v->m_locked)
attr = "locked ";
oprintf("%s\t%s: %s %s %s%s@%i ", ind, v->m_name.c_str(), type_name[v->m_vtype],
storenames[v->m_store],
attr, (v->m_callparam ? "callparam " : ""),
(int)v->m_code.local);
if (v->m_life.empty())
oprintf("[null]");
for (l = 0; l < v->m_life.size(); ++l) {
oprintf("[%i,%i] ", v->m_life[l].start, v->m_life[l].end);
}
oprintf("\n");
for (m = 0; m < 3; ++m) {
ir_value *vm = v->m_members[m];
if (!vm)
continue;
if (vm->m_unique_life && vm->m_locked)
attr = "unique,locked ";
else if (vm->m_unique_life)
attr = "unique ";
else if (vm->m_locked)
attr = "locked ";
oprintf("%s\t%s: %s@%i ", ind, vm->m_name.c_str(), attr, (int)vm->m_code.local);
for (l = 0; l < vm->m_life.size(); ++l) {
oprintf("[%i,%i] ", vm->m_life[l].start, vm->m_life[l].end);
}
oprintf("\n");
}
}
if (f->m_blocks.size())
{
oprintf("%slife passes: %i\n", ind, (int)f->m_run_id);
for (i = 0; i < f->m_blocks.size(); ++i) {
ir_block_dump(f->m_blocks[i].get(), ind, oprintf);
}
}
ind[strlen(ind)-1] = 0;
oprintf("%sendfunction %s\n", ind, f->m_name.c_str());
}
void ir_block_dump(ir_block* b, char *ind,
int (*oprintf)(const char*, ...))
{
oprintf("%s:%s\n", ind, b->m_label.c_str());
util_strncat(ind, "\t", IND_BUFSZ-1);
if (!b->m_instr.empty() && b->m_instr[0])
oprintf("%s (%i) [entry]\n", ind, (int)(b->m_instr[0]->m_eid-1));
for (auto &i : b->m_instr)
ir_instr_dump(i, ind, oprintf);
ind[strlen(ind)-1] = 0;
}
static void dump_phi(ir_instr *in, int (*oprintf)(const char*, ...))
{
oprintf("%s <- phi ", in->_m_ops[0]->m_name.c_str());
for (auto &it : in->m_phi) {
oprintf("([%s] : %s) ", it.from->m_label.c_str(),
it.value->m_name.c_str());
}
oprintf("\n");
}
void ir_instr_dump(ir_instr *in, char *ind,
int (*oprintf)(const char*, ...))
{
size_t i;
const char *comma = nullptr;
oprintf("%s (%i) ", ind, (int)in->m_eid);
if (in->m_opcode == VINSTR_PHI) {
dump_phi(in, oprintf);
return;
}
util_strncat(ind, "\t", IND_BUFSZ-1);
if (in->_m_ops[0] && (in->_m_ops[1] || in->_m_ops[2])) {
in->_m_ops[0]->dump(oprintf);
if (in->_m_ops[1] || in->_m_ops[2])
oprintf(" <- ");
}
if (in->m_opcode == INSTR_CALL0 || in->m_opcode == VINSTR_NRCALL) {
oprintf("CALL%i\t", in->m_params.size());
} else
oprintf("%s\t", qc_opname(in->m_opcode));
if (in->_m_ops[0] && !(in->_m_ops[1] || in->_m_ops[2])) {
in->_m_ops[0]->dump(oprintf);
comma = ",\t";
}
else
{
for (i = 1; i != 3; ++i) {
if (in->_m_ops[i]) {
if (comma)
oprintf(comma);
in->_m_ops[i]->dump(oprintf);
comma = ",\t";
}
}
}
if (in->m_bops[0]) {
if (comma)
oprintf(comma);
oprintf("[%s]", in->m_bops[0]->m_label.c_str());
comma = ",\t";
}
if (in->m_bops[1])
oprintf("%s[%s]", comma, in->m_bops[1]->m_label.c_str());
if (in->m_params.size()) {
oprintf("\tparams: ");
for (auto &it : in->m_params)
oprintf("%s, ", it->m_name.c_str());
}
oprintf("\n");
ind[strlen(ind)-1] = 0;
}
static void ir_value_dump_string(const char *str, int (*oprintf)(const char*, ...))
{
oprintf("\"");
for (; *str; ++str) {
switch (*str) {
case '\n': oprintf("\\n"); break;
case '\r': oprintf("\\r"); break;
case '\t': oprintf("\\t"); break;
case '\v': oprintf("\\v"); break;
case '\f': oprintf("\\f"); break;
case '\b': oprintf("\\b"); break;
case '\a': oprintf("\\a"); break;
case '\\': oprintf("\\\\"); break;
case '"': oprintf("\\\""); break;
default: oprintf("%c", *str); break;
}
}
oprintf("\"");
}
void ir_value::dump(int (*oprintf)(const char*, ...)) const
{
if (m_hasvalue) {
switch (m_vtype) {
default:
case TYPE_VOID:
oprintf("(void)");
break;
case TYPE_FUNCTION:
oprintf("fn:%s", m_name.c_str());
break;
case TYPE_FLOAT:
oprintf("%g", m_constval.vfloat);
break;
case TYPE_VECTOR:
oprintf("'%g %g %g'",
m_constval.vvec.x,
m_constval.vvec.y,
m_constval.vvec.z);
break;
case TYPE_ENTITY:
oprintf("(entity)");
break;
case TYPE_STRING:
ir_value_dump_string(m_constval.vstring, oprintf);
break;
#if 0
case TYPE_INTEGER:
oprintf("%i", m_constval.vint);
break;
#endif
case TYPE_POINTER:
oprintf("&%s",
m_constval.vpointer->m_name.c_str());
break;
}
} else {
oprintf("%s", m_name.c_str());
}
}
void ir_value::dumpLife(int (*oprintf)(const char*,...)) const
{
oprintf("Life of %12s:", m_name.c_str());
for (size_t i = 0; i < m_life.size(); ++i)
{
oprintf(" + [%i, %i]\n", m_life[i].start, m_life[i].end);
}
}