1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
|
/*
* Copyright © 2013 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*/
/**
* \file opt_vectorize.cpp
*
* Combines scalar assignments of the same expression (modulo swizzle) to
* multiple channels of the same variable into a single vectorized expression
* and assignment.
*
* Many generated shaders contain scalarized code. That is, they contain
*
* r1.x = log2(v0.x);
* r1.y = log2(v0.y);
* r1.z = log2(v0.z);
*
* rather than
*
* r1.xyz = log2(v0.xyz);
*
* We look for consecutive assignments of the same expression (modulo swizzle)
* to each channel of the same variable.
*
* For instance, we want to convert these three scalar operations
*
* (assign (x) (var_ref r1) (expression float log2 (swiz x (var_ref v0))))
* (assign (y) (var_ref r1) (expression float log2 (swiz y (var_ref v0))))
* (assign (z) (var_ref r1) (expression float log2 (swiz z (var_ref v0))))
*
* into a single vector operation
*
* (assign (xyz) (var_ref r1) (expression vec3 log2 (swiz xyz (var_ref v0))))
*/
#include "ir.h"
#include "ir_visitor.h"
#include "ir_optimization.h"
#include "glsl_types.h"
#include "program/prog_instruction.h"
namespace {
class ir_vectorize_visitor : public ir_hierarchical_visitor {
public:
void clear()
{
assignment[0] = NULL;
assignment[1] = NULL;
assignment[2] = NULL;
assignment[3] = NULL;
current_assignment = NULL;
last_assignment = NULL;
channels = 0;
has_swizzle = false;
}
ir_vectorize_visitor()
{
clear();
progress = false;
}
virtual ir_visitor_status visit_enter(ir_assignment *);
virtual ir_visitor_status visit_enter(ir_swizzle *);
virtual ir_visitor_status visit_enter(ir_dereference_array *);
virtual ir_visitor_status visit_enter(ir_expression *);
virtual ir_visitor_status visit_enter(ir_if *);
virtual ir_visitor_status visit_enter(ir_loop *);
virtual ir_visitor_status visit_leave(ir_assignment *);
void try_vectorize();
ir_assignment *assignment[4];
ir_assignment *current_assignment, *last_assignment;
unsigned channels;
bool has_swizzle;
bool progress;
};
} /* unnamed namespace */
/**
* Rewrites the swizzles and types of a right-hand side of an assignment.
*
* From the example above, this function would be called (by visit_tree()) on
* the nodes of the tree (expression float log2 (swiz z (var_ref v0))),
* rewriting it into (expression vec3 log2 (swiz xyz (var_ref v0))).
*
* The function operates on ir_expressions (and its operands) and ir_swizzles.
* For expressions it sets a new type and swizzles any non-expression and non-
* swizzle scalar operands into appropriately sized vector arguments. For
* example, if combining
*
* (assign (x) (var_ref r1) (expression float + (swiz x (var_ref v0) (var_ref v1))))
* (assign (y) (var_ref r1) (expression float + (swiz y (var_ref v0) (var_ref v1))))
*
* where v1 is a scalar, rewrite_swizzle() would insert a swizzle on
* (var_ref v1) such that the final result was
*
* (assign (xy) (var_ref r1) (expression vec2 + (swiz xy (var_ref v0))
* (swiz xx (var_ref v1))))
*
* For swizzles, it sets a new type, and if the variable being swizzled is a
* vector it overwrites the swizzle mask with the ir_swizzle_mask passed as the
* data parameter. If the swizzled variable is scalar, then the swizzle was
* added by an earlier call to rewrite_swizzle() on an expression, so the
* mask should not be modified.
*/
static void
rewrite_swizzle(ir_instruction *ir, void *data)
{
ir_swizzle_mask *mask = (ir_swizzle_mask *)data;
switch (ir->ir_type) {
case ir_type_swizzle: {
ir_swizzle *swz = (ir_swizzle *)ir;
if (swz->val->type->is_vector()) {
swz->mask = *mask;
}
swz->type = glsl_type::get_instance(swz->type->base_type,
mask->num_components, 1);
break;
}
case ir_type_expression: {
ir_expression *expr = (ir_expression *)ir;
expr->type = glsl_type::get_instance(expr->type->base_type,
mask->num_components, 1);
for (unsigned i = 0; i < 4; i++) {
if (expr->operands[i]) {
ir_rvalue *rval = expr->operands[i]->as_rvalue();
if (rval && rval->type->is_scalar() &&
!rval->as_expression() && !rval->as_swizzle()) {
expr->operands[i] = new(ir) ir_swizzle(rval, 0, 0, 0, 0,
mask->num_components);
}
}
}
break;
}
default:
break;
}
}
/**
* Attempt to vectorize the previously saved assignments, and clear them from
* consideration.
*
* If the assignments are able to be combined, it modifies in-place the last
* assignment seen to be an equivalent vector form of the scalar assignments.
* It then removes the other now obsolete scalar assignments.
*/
void
ir_vectorize_visitor::try_vectorize()
{
if (this->last_assignment && this->channels > 1) {
ir_swizzle_mask mask = {0, 0, 0, 0, channels, 0};
this->last_assignment->write_mask = 0;
for (unsigned i = 0, j = 0; i < 4; i++) {
if (this->assignment[i]) {
this->last_assignment->write_mask |= 1 << i;
if (this->assignment[i] != this->last_assignment) {
this->assignment[i]->remove();
}
switch (j) {
case 0: mask.x = i; break;
case 1: mask.y = i; break;
case 2: mask.z = i; break;
case 3: mask.w = i; break;
}
j++;
}
}
visit_tree(this->last_assignment->rhs, rewrite_swizzle, &mask);
this->progress = true;
}
clear();
}
/**
* Returns whether the write mask is a single channel.
*/
static bool
single_channel_write_mask(unsigned write_mask)
{
return write_mask != 0 && (write_mask & (write_mask - 1)) == 0;
}
/**
* Translates single-channeled write mask to single-channeled swizzle.
*/
static unsigned
write_mask_to_swizzle(unsigned write_mask)
{
switch (write_mask) {
case WRITEMASK_X: return SWIZZLE_X;
case WRITEMASK_Y: return SWIZZLE_Y;
case WRITEMASK_Z: return SWIZZLE_Z;
case WRITEMASK_W: return SWIZZLE_W;
}
assert(!"not reached");
unreachable();
}
/**
* Returns whether a single-channeled write mask matches a swizzle.
*/
static bool
write_mask_matches_swizzle(unsigned write_mask,
const ir_swizzle *swz)
{
return ((write_mask == WRITEMASK_X && swz->mask.x == SWIZZLE_X) ||
(write_mask == WRITEMASK_Y && swz->mask.x == SWIZZLE_Y) ||
(write_mask == WRITEMASK_Z && swz->mask.x == SWIZZLE_Z) ||
(write_mask == WRITEMASK_W && swz->mask.x == SWIZZLE_W));
}
/**
* Upon entering an ir_assignment, attempt to vectorize the currently tracked
* assignments if the current assignment is not suitable. Keep a pointer to
* the current assignment.
*/
ir_visitor_status
ir_vectorize_visitor::visit_enter(ir_assignment *ir)
{
ir_dereference *lhs = this->last_assignment != NULL ?
this->last_assignment->lhs : NULL;
ir_rvalue *rhs = this->last_assignment != NULL ?
this->last_assignment->rhs : NULL;
if (ir->condition ||
this->channels >= 4 ||
!single_channel_write_mask(ir->write_mask) ||
this->assignment[write_mask_to_swizzle(ir->write_mask)] != NULL ||
(lhs && !ir->lhs->equals(lhs)) ||
(rhs && !ir->rhs->equals(rhs, ir_type_swizzle))) {
try_vectorize();
}
this->current_assignment = ir;
return visit_continue;
}
/**
* Upon entering an ir_swizzle, set ::has_swizzle if we're visiting from an
* ir_assignment (i.e., that ::current_assignment is set) and the swizzle mask
* matches the current assignment's write mask.
*
* If the write mask doesn't match the swizzle mask, remove the current
* assignment from further consideration.
*/
ir_visitor_status
ir_vectorize_visitor::visit_enter(ir_swizzle *ir)
{
if (this->current_assignment) {
if (write_mask_matches_swizzle(this->current_assignment->write_mask, ir)) {
this->has_swizzle = true;
} else {
this->current_assignment = NULL;
}
}
return visit_continue;
}
/* Upon entering an ir_array_dereference, remove the current assignment from
* further consideration. Since the index of an array dereference must scalar,
* we are not able to vectorize it.
*
* FINISHME: If all of scalar indices are identical we could vectorize.
*/
ir_visitor_status
ir_vectorize_visitor::visit_enter(ir_dereference_array *)
{
this->current_assignment = NULL;
return visit_continue_with_parent;
}
/**
* Upon entering an ir_expression, remove the current assignment from further
* consideration if the expression operates horizontally on vectors.
*/
ir_visitor_status
ir_vectorize_visitor::visit_enter(ir_expression *ir)
{
if (ir->is_horizontal()) {
this->current_assignment = NULL;
return visit_continue_with_parent;
}
return visit_continue;
}
/* Since there is no statement to visit between the "then" and "else"
* instructions try to vectorize before, in between, and after them to avoid
* combining statements from different basic blocks.
*/
ir_visitor_status
ir_vectorize_visitor::visit_enter(ir_if *ir)
{
try_vectorize();
visit_list_elements(this, &ir->then_instructions);
try_vectorize();
visit_list_elements(this, &ir->else_instructions);
try_vectorize();
return visit_continue_with_parent;
}
/* Since there is no statement to visit between the instructions in the body of
* the loop and the instructions after it try to vectorize before and after the
* body to avoid combining statements from different basic blocks.
*/
ir_visitor_status
ir_vectorize_visitor::visit_enter(ir_loop *ir)
{
try_vectorize();
visit_list_elements(this, &ir->body_instructions);
try_vectorize();
return visit_continue_with_parent;
}
/**
* Upon leaving an ir_assignment, save a pointer to it in ::assignment[] if
* the swizzle mask(s) found were appropriate. Also save a pointer in
* ::last_assignment so that we can compare future assignments with it.
*
* Finally, clear ::current_assignment and ::has_swizzle.
*/
ir_visitor_status
ir_vectorize_visitor::visit_leave(ir_assignment *ir)
{
if (this->has_swizzle && this->current_assignment) {
assert(this->current_assignment == ir);
unsigned channel = write_mask_to_swizzle(this->current_assignment->write_mask);
this->assignment[channel] = ir;
this->channels++;
this->last_assignment = this->current_assignment;
}
this->current_assignment = NULL;
this->has_swizzle = false;
return visit_continue;
}
/**
* Combines scalar assignments of the same expression (modulo swizzle) to
* multiple channels of the same variable into a single vectorized expression
* and assignment.
*/
bool
do_vectorize(exec_list *instructions)
{
ir_vectorize_visitor v;
v.run(instructions);
/* Try to vectorize the last assignments seen. */
v.try_vectorize();
return v.progress;
}
|