2
* Copyright 2013 Vadim Girlin <vadimgirlin@gmail.com>
4
* Permission is hereby granted, free of charge, to any person obtaining a
5
* copy of this software and associated documentation files (the "Software"),
6
* to deal in the Software without restriction, including without limitation
7
* on the rights to use, copy, modify, merge, publish, distribute, sub
8
* license, and/or sell copies of the Software, and to permit persons to whom
9
* the Software is furnished to do so, subject to the following conditions:
11
* The above copyright notice and this permission notice (including the next
12
* paragraph) shall be included in all copies or substantial portions of the
15
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
* FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18
* THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21
* USE OR OTHER DEALINGS IN THE SOFTWARE.
30
#include "sb_shader.h"
35
container_node* ssa_prepare::create_phi_nodes(int count) {
36
container_node *p = sh.create_container();
37
val_set &vars = cur_set();
40
for (val_set::iterator I = vars.begin(sh), E = vars.end(sh); I != E; ++I) {
41
nn = sh.create_node(NT_OP, NST_PHI);
42
nn->dst.assign(1, *I);
43
nn->src.assign(count, *I);
49
void ssa_prepare::add_defs(node &n) {
50
val_set &s = cur_set();
51
for (vvec::iterator I = n.dst.begin(), E = n.dst.end(); I != E; ++I) {
63
bool ssa_prepare::visit(cf_node& n, bool enter) {
73
bool ssa_prepare::visit(alu_node& n, bool enter) {
81
bool ssa_prepare::visit(fetch_node& n, bool enter) {
89
bool ssa_prepare::visit(region_node& n, bool enter) {
94
cur_set().add_set(n.vars_defined);
95
if (n.dep_count() > 0)
96
n.phi = create_phi_nodes(n.dep_count());
97
if (n.rep_count() > 1) {
98
n.loop_phi = create_phi_nodes(n.rep_count());
99
n.loop_phi->subtype = NST_LOOP_PHI_CONTAINER;
101
n.vars_defined.clear();
107
bool ssa_prepare::visit(repeat_node& n, bool enter) {
112
n.target->vars_defined.add_set(cur_set());
119
bool ssa_prepare::visit(depart_node& n, bool enter) {
124
n.target->vars_defined.add_set(cur_set());
131
// ===============================
133
int ssa_rename::init() {
134
rename_stack.push(def_map());
138
bool ssa_rename::visit(alu_group_node& n, bool enter) {
139
// taking into account parallel execution of the alu group
141
for (node_iterator I = n.begin(), E = n.end(); I != E; ++I) {
142
I->accept(*this, true);
145
for (node_iterator I = n.begin(), E = n.end(); I != E; ++I) {
146
I->accept(*this, false);
152
bool ssa_rename::visit(cf_node& n, bool enter) {
161
bool ssa_rename::visit(alu_node& n, bool enter) {
168
if (n.pred && n.dst[0]) {
171
unsigned index = get_index(rename_stack.top(), d);
172
value *p = sh.get_value_version(d, index);
174
psi = sh.create_node(NT_OP, NST_PSI);
176
container_node *parent;
177
if (n.parent->subtype == NST_ALU_GROUP)
180
assert (n.parent->parent->subtype == NST_ALU_GROUP);
181
parent = n.parent->parent;
183
parent->insert_after(psi);
185
assert(n.bc.pred_sel);
189
psi->src[3] = n.pred;
190
psi->src[4] = sh.get_pred_sel(n.bc.pred_sel - PRED_SEL_0);
192
psi->dst.push_back(d);
202
if (!n.dst.empty() && n.dst[0]) {
203
// FIXME probably use separate pass for such things
204
if ((n.bc.op_ptr->flags & AF_INTERP) || n.bc.op == ALU_OP2_CUBE)
205
n.dst[0]->flags |= VLF_PIN_CHAN;
211
bool ssa_rename::visit(alu_packed_node& n, bool enter) {
213
for (node_iterator I = n.begin(), E = n.end(); I != E; ++I) {
214
I->accept(*this, true);
217
for (node_iterator I = n.begin(), E = n.end(); I != E; ++I) {
218
I->accept(*this, false);
221
bool repl = (n.op_ptr()->flags & AF_REPL) ||
222
(ctx.is_cayman() && (n.first->alu_op_slot_flags() & AF_S));
229
bool ssa_rename::visit(fetch_node& n, bool enter) {
238
bool ssa_rename::visit(region_node& n, bool enter) {
241
rename_phi_args(n.loop_phi, 0, true);
244
rename_phi_args(n.phi, ~0u, true);
249
bool ssa_rename::visit(repeat_node& n, bool enter) {
251
push(n.target->loop_phi);
253
if (n.target->loop_phi)
254
rename_phi_args(n.target->loop_phi, n.rep_id, false);
260
bool ssa_rename::visit(depart_node& n, bool enter) {
265
rename_phi_args(n.target->phi, n.dep_id, false);
271
bool ssa_rename::visit(if_node& n, bool enter) {
274
n.cond = rename_use(&n, n.cond);
279
void ssa_rename::push(node* phi) {
280
rename_stack.push(rename_stack.top());
283
void ssa_rename::pop() {
287
value* ssa_rename::rename_use(node *n, value* v) {
291
unsigned index = get_index(rename_stack.top(), v);
292
v = sh.get_value_version(v, index);
294
// if (alu) instruction is predicated and source arg comes from psi node
295
// (that is, from another predicated instruction through its psi node),
296
// we can try to select the corresponding source value directly
297
if (n->pred && v->def && v->def->subtype == NST_PSI) {
298
assert(n->subtype == NST_ALU_INST);
299
alu_node *an = static_cast<alu_node*>(n);
301
// FIXME make it more generic ???
302
if (pn->src.size() == 6) {
303
if (pn->src[3] == n->pred) {
304
value* ps = sh.get_pred_sel(an->bc.pred_sel - PRED_SEL_0);
305
if (pn->src[4] == ps)
315
value* ssa_rename::rename_def(node *n, value* v) {
316
unsigned index = new_index(def_count, v);
317
set_index(rename_stack.top(), v, index);
318
value *r = sh.get_value_version(v, index);
322
void ssa_rename::rename_src_vec(node *n, vvec &vv, bool src) {
323
for(vvec::iterator I = vv.begin(), E = vv.end(); I != E; ++I) {
325
if (!v || v->is_readonly())
329
if (!v->rel->is_readonly())
330
v->rel = rename_use(n, v->rel);
331
rename_src_vec(n, v->muse, true);
333
v = rename_use(n, v);
337
void ssa_rename::rename_src(node* n) {
339
n->pred = rename_use(n, n->pred);
341
rename_src_vec(n, n->src, true);
342
rename_src_vec(n, n->dst, false);
346
void ssa_rename::rename_dst_vec(node *n, vvec &vv, bool set_def) {
348
for(vvec::iterator I = vv.begin(), E = vv.end(); I != E; ++I) {
354
rename_dst_vec(n, v->mdef, false);
356
v = rename_def(n, v);
363
void ssa_rename::rename_dst(node* n) {
364
rename_dst_vec(n, n->dst, true);
367
unsigned ssa_rename::get_index(def_map& m, value* v) {
368
def_map::iterator I = m.find(v);
374
void ssa_rename::set_index(def_map& m, value* v, unsigned index) {
375
std::pair<def_map::iterator,bool> r = m.insert(std::make_pair(v, index));
377
r.first->second = index;
380
unsigned ssa_rename::new_index(def_map& m, value* v) {
382
def_map::iterator I = m.find(v);
386
m.insert(std::make_pair(v, index));
390
bool ssa_rename::visit(node& n, bool enter) {
392
assert(n.subtype == NST_PSI);
399
bool ssa_rename::visit(container_node& n, bool enter) {
402
// should be root container node
403
assert(n.parent == NULL);
404
rename_src_vec(&n, n.src, true);
409
void ssa_rename::rename_phi_args(container_node* phi, unsigned op, bool def) {
410
for (node_iterator I = phi->begin(), E = phi->end(); I != E; ++I) {
413
o->src[op] = rename_use(o, o->src[op]);
415
o->dst[0] = rename_def(o, o->dst[0]);
421
} // namespace r600_sb