Files
Odin/src/ir_opt.cpp

496 lines
11 KiB
C++

// Optimizations for the IR code
void ir_opt_add_operands(Array<irValue *> *ops, irInstr *i) {
switch (i->kind) {
case irInstr_Comment:
break;
case irInstr_Local:
break;
case irInstr_ZeroInit:
array_add(ops, i->ZeroInit.address);
break;
case irInstr_Store:
array_add(ops, i->Store.address);
array_add(ops, i->Store.value);
break;
case irInstr_Load:
array_add(ops, i->Load.address);
break;
case irInstr_ArrayElementPtr:
array_add(ops, i->ArrayElementPtr.address);
array_add(ops, i->ArrayElementPtr.elem_index);
break;
case irInstr_StructElementPtr:
array_add(ops, i->StructElementPtr.address);
break;
case irInstr_PtrOffset:
array_add(ops, i->PtrOffset.address);
array_add(ops, i->PtrOffset.offset);
break;
case irInstr_StructExtractValue:
array_add(ops, i->StructExtractValue.address);
break;
case irInstr_Conv:
array_add(ops, i->Conv.value);
break;
case irInstr_Jump:
break;
case irInstr_If:
array_add(ops, i->If.cond);
break;
case irInstr_Return:
if (i->Return.value != nullptr) {
array_add(ops, i->Return.value);
}
break;
case irInstr_Select:
array_add(ops, i->Select.cond);
break;
case irInstr_Phi:
for_array(j, i->Phi.edges) {
array_add(ops, i->Phi.edges[j]);
}
break;
case irInstr_Unreachable:
break;
case irInstr_UnaryOp:
array_add(ops, i->UnaryOp.expr);
break;
case irInstr_BinaryOp:
array_add(ops, i->BinaryOp.left);
array_add(ops, i->BinaryOp.right);
break;
case irInstr_Call:
array_add(ops, i->Call.value);
for_array(j, i->Call.args) {
array_add(ops, i->Call.args[j]);
}
break;
// case irInstr_VectorExtractElement:
// array_add(ops, i->VectorExtractElement.vector);
// array_add(ops, i->VectorExtractElement.index);
// break;
// case irInstr_VectorInsertElement:
// array_add(ops, i->VectorInsertElement.vector);
// array_add(ops, i->VectorInsertElement.elem);
// array_add(ops, i->VectorInsertElement.index);
// break;
// case irInstr_VectorShuffle:
// array_add(ops, i->VectorShuffle.vector);
// break;
case irInstr_StartupRuntime:
break;
#if 0
case irInstr_BoundsCheck:
array_add(ops, i->BoundsCheck.index);
array_add(ops, i->BoundsCheck.len);
break;
case irInstr_SliceBoundsCheck:
array_add(ops, i->SliceBoundsCheck.low);
array_add(ops, i->SliceBoundsCheck.high);
break;
#endif
}
}
void ir_opt_block_replace_pred(irBlock *b, irBlock *from, irBlock *to) {
for_array(i, b->preds) {
irBlock *pred = b->preds[i];
if (pred == from) {
b->preds[i] = to;
}
}
}
void ir_opt_block_replace_succ(irBlock *b, irBlock *from, irBlock *to) {
for_array(i, b->succs) {
irBlock *succ = b->succs[i];
if (succ == from) {
b->succs[i] = to;
}
}
}
bool ir_opt_block_has_phi(irBlock *b) {
return b->instrs[0]->Instr.kind == irInstr_Phi;
}
Array<irValue *> ir_get_block_phi_nodes(irBlock *b) {
Array<irValue *> phis = {0};
for_array(i, b->instrs) {
irInstr *instr = &b->instrs[i]->Instr;
if (instr->kind != irInstr_Phi) {
phis = b->instrs;
phis.count = i;
return phis;
}
}
return phis;
}
void ir_remove_pred(irBlock *b, irBlock *p) {
Array<irValue *> phis = ir_get_block_phi_nodes(b);
isize i = 0;
for_array(j, b->preds) {
irBlock *pred = b->preds[j];
if (pred != p) {
b->preds[i] = b->preds[j];
for_array(k, phis) {
irInstrPhi *phi = &phis[k]->Instr.Phi;
phi->edges[i] = phi->edges[j];
}
i++;
}
}
b->preds.count = i;
for_array(k, phis) {
irInstrPhi *phi = &phis[k]->Instr.Phi;
phi->edges.count = i;
}
}
void ir_remove_dead_blocks(irProcedure *proc) {
isize j = 0;
for_array(i, proc->blocks) {
irBlock *b = proc->blocks[i];
if (b == nullptr) {
continue;
}
// NOTE(bill): Swap order
b->index = cast(i32)j;
proc->blocks[j++] = b;
}
proc->blocks.count = j;
}
void ir_mark_reachable(irBlock *b) {
isize const WHITE = 0;
isize const BLACK = -1;
b->index = BLACK;
for_array(i, b->succs) {
irBlock *succ = b->succs[i];
if (succ->index == WHITE) {
ir_mark_reachable(succ);
}
}
}
void ir_remove_unreachable_blocks(irProcedure *proc) {
isize const WHITE = 0;
isize const BLACK = -1;
for_array(i, proc->blocks) {
proc->blocks[i]->index = WHITE;
}
ir_mark_reachable(proc->blocks[0]);
for_array(i, proc->blocks) {
irBlock *b = proc->blocks[i];
if (b->index == WHITE) {
for_array(j, b->succs) {
irBlock *c = b->succs[j];
if (c->index == BLACK) {
ir_remove_pred(c, b);
}
}
// NOTE(bill): Mark as empty but don't actually free it
// As it's been allocated with an arena
proc->blocks[i] = nullptr;
}
}
ir_remove_dead_blocks(proc);
}
bool ir_opt_block_fusion(irProcedure *proc, irBlock *a) {
if (a->succs.count != 1) {
return false;
}
irBlock *b = a->succs[0];
if (b->preds.count != 1) {
return false;
}
if (ir_opt_block_has_phi(b)) {
return false;
}
array_pop(&a->instrs); // Remove branch at end
for_array(i, b->instrs) {
array_add(&a->instrs, b->instrs[i]);
ir_set_instr_block(b->instrs[i], a);
}
array_clear(&a->succs);
for_array(i, b->succs) {
array_add(&a->succs, b->succs[i]);
}
// Fix preds links
for_array(i, b->succs) {
ir_opt_block_replace_pred(b->succs[i], b, a);
}
proc->blocks[b->index] = nullptr;
return true;
}
void ir_opt_blocks(irProcedure *proc) {
ir_remove_unreachable_blocks(proc);
#if 1
bool changed = true;
while (changed) {
changed = false;
for_array(i, proc->blocks) {
irBlock *b = proc->blocks[i];
if (b == nullptr) {
continue;
}
GB_ASSERT_MSG(b->index == i, "%d, %td", b->index, i);
if (ir_opt_block_fusion(proc, b)) {
changed = true;
}
// TODO(bill): other simple block optimizations
}
}
#endif
ir_remove_dead_blocks(proc);
}
void ir_opt_build_referrers(irProcedure *proc) {
gbTempArenaMemory tmp = gb_temp_arena_memory_begin(&proc->module->tmp_arena);
// NOTE(bill): Acta as a buffer
auto ops = array_make<irValue *>(proc->module->tmp_allocator, 0, 64); // TODO HACK(bill): This _could_ overflow the temp arena
for_array(i, proc->blocks) {
irBlock *b = proc->blocks[i];
for_array(j, b->instrs) {
irValue *instr = b->instrs[j];
array_clear(&ops);
ir_opt_add_operands(&ops, &instr->Instr);
for_array(k, ops) {
irValue *op = ops[k];
if (op == nullptr) {
continue;
}
Array<irValue *> *refs = ir_value_referrers(op);
if (refs != nullptr) {
array_add(refs, instr);
}
}
}
}
gb_temp_arena_memory_end(tmp);
}
// State of Lengauer-Tarjan algorithm
// Based on this paper: http://jgaa.info/accepted/2006/GeorgiadisTarjanWerneck2006.10.1.pdf
typedef struct irLTState {
isize count;
// NOTE(bill): These are arrays
irBlock **sdom; // Semidominator
irBlock **parent; // Parent in DFS traversal of CFG
irBlock **ancestor;
} irLTState;
// §2.2 - bottom of page
void ir_lt_link(irLTState *lt, irBlock *p, irBlock *q) {
lt->ancestor[q->index] = p;
}
i32 ir_lt_depth_first_search(irLTState *lt, irBlock *p, i32 i, irBlock **preorder) {
preorder[i] = p;
p->dom.pre = i++;
lt->sdom[p->index] = p;
ir_lt_link(lt, nullptr, p);
for_array(index, p->succs) {
irBlock *q = p->succs[index];
if (lt->sdom[q->index] == nullptr) {
lt->parent[q->index] = p;
i = ir_lt_depth_first_search(lt, q, i, preorder);
}
}
return i;
}
irBlock *ir_lt_eval(irLTState *lt, irBlock *v) {
irBlock *u = v;
for (;
lt->ancestor[v->index] != nullptr;
v = lt->ancestor[v->index]) {
if (lt->sdom[v->index]->dom.pre < lt->sdom[u->index]->dom.pre) {
u = v;
}
}
return u;
}
typedef struct irDomPrePost {
i32 pre, post;
} irDomPrePost;
irDomPrePost ir_opt_number_dom_tree(irBlock *v, i32 pre, i32 post) {
irDomPrePost result = {pre, post};
v->dom.pre = pre++;
for_array(i, v->dom.children) {
result = ir_opt_number_dom_tree(v->dom.children[i], result.pre, result.post);
}
v->dom.post = post++;
result.pre = pre;
result.post = post;
return result;
}
// NOTE(bill): Requires `ir_opt_blocks` to be called before this
void ir_opt_build_dom_tree(irProcedure *proc) {
// Based on this paper: http://jgaa.info/accepted/2006/GeorgiadisTarjanWerneck2006.10.1.pdf
gbTempArenaMemory tmp = gb_temp_arena_memory_begin(&proc->module->tmp_arena);
i32 n = cast(i32)proc->blocks.count;
irBlock **buf = gb_alloc_array(proc->module->tmp_allocator, irBlock *, 5*n);
irLTState lt = {0};
lt.count = n;
lt.sdom = &buf[0*n];
lt.parent = &buf[1*n];
lt.ancestor = &buf[2*n];
irBlock **preorder = &buf[3*n];
irBlock **buckets = &buf[4*n];
irBlock *root = proc->blocks[0];
// Step 1 - number vertices
i32 pre_num = ir_lt_depth_first_search(&lt, root, 0, preorder);
gb_memmove(buckets, preorder, n*gb_size_of(preorder[0]));
for (i32 i = n-1; i > 0; i--) {
irBlock *w = preorder[i];
// Step 3 - Implicitly define idom for nodes
for (irBlock *v = buckets[i]; v != w; v = buckets[v->dom.pre]) {
irBlock *u = ir_lt_eval(&lt, v);
if (lt.sdom[u->index]->dom.pre < i) {
v->dom.idom = u;
} else {
v->dom.idom = w;
}
}
// Step 2 - Compute all sdoms
lt.sdom[w->index] = lt.parent[w->index];
for_array(pred_index, w->preds) {
irBlock *v = w->preds[pred_index];
irBlock *u = ir_lt_eval(&lt, v);
if (lt.sdom[u->index]->dom.pre < lt.sdom[w->index]->dom.pre) {
lt.sdom[w->index] = lt.sdom[u->index];
}
}
ir_lt_link(&lt, lt.parent[w->index], w);
if (lt.parent[w->index] == lt.sdom[w->index]) {
w->dom.idom = lt.parent[w->index];
} else {
buckets[i] = buckets[lt.sdom[w->index]->dom.pre];
buckets[lt.sdom[w->index]->dom.pre] = w;
}
}
// The rest of Step 3
for (irBlock *v = buckets[0]; v != root; v = buckets[v->dom.pre]) {
v->dom.idom = root;
}
// Step 4 - Explicitly define idom for nodes (in preorder)
for (isize i = 1; i < n; i++) {
irBlock *w = preorder[i];
if (w == root) {
w->dom.idom = nullptr;
} else {
// Weird tree relationships here!
if (w->dom.idom != lt.sdom[w->index]) {
w->dom.idom = w->dom.idom->dom.idom;
}
// Calculate children relation as inverse of idom
if (w->dom.idom->dom.children.data == nullptr) {
// TODO(bill): Is this good enough for memory allocations?
array_init(&w->dom.idom->dom.children, heap_allocator());
}
array_add(&w->dom.idom->dom.children, w);
}
}
ir_opt_number_dom_tree(root, 0, 0);
gb_temp_arena_memory_end(tmp);
}
void ir_opt_mem2reg(irProcedure *proc) {
// TODO(bill): ir_opt_mem2reg
}
void ir_opt_tree(irGen *s) {
s->opt_called = true;
for_array(member_index, s->module.procs) {
irProcedure *proc = s->module.procs[member_index];
if (proc->blocks.count == 0) { // Prototype/external procedure
continue;
}
ir_opt_blocks(proc);
#if 0
ir_opt_build_referrers(proc);
ir_opt_build_dom_tree(proc);
// TODO(bill): ir optimization
// [ ] cse (common-subexpression) elim
// [ ] copy elim
// [ ] dead code elim
// [ ] dead store/load elim
// [ ] phi elim
// [ ] short circuit elim
// [ ] bounds check elim
// [ ] lift/mem2reg
// [ ] lift/mem2reg
ir_opt_mem2reg(proc);
#endif
GB_ASSERT(proc->blocks.count > 0);
ir_number_proc_registers(proc);
}
}