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
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
|
/*
* Copyright (c) 2013 Rob Clark <robdclark@gmail.com>
*
* 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.
*/
#ifndef IR3_H_
#define IR3_H_
#include <stdint.h>
#include <stdbool.h>
#include "instr-a3xx.h"
#include "disasm.h" /* TODO move 'enum shader_t' somewhere else.. */
/* low level intermediate representation of an adreno shader program */
struct ir3_shader;
struct ir3_instruction;
struct ir3_block;
struct ir3_shader * fd_asm_parse(const char *src);
struct ir3_shader_info {
uint16_t sizedwords;
uint16_t instrs_count; /* expanded to account for rpt's */
/* NOTE: max_reg, etc, does not include registers not touched
* by the shader (ie. vertex fetched via VFD_DECODE but not
* touched by shader)
*/
int8_t max_reg; /* highest GPR # used by shader */
int8_t max_half_reg;
int8_t max_const;
};
struct ir3_register {
enum {
IR3_REG_CONST = 0x001,
IR3_REG_IMMED = 0x002,
IR3_REG_HALF = 0x004,
IR3_REG_RELATIV= 0x008,
IR3_REG_R = 0x010,
IR3_REG_NEGATE = 0x020,
IR3_REG_ABS = 0x040,
IR3_REG_EVEN = 0x080,
IR3_REG_POS_INF= 0x100,
/* (ei) flag, end-input? Set on last bary, presumably to signal
* that the shader needs no more input:
*/
IR3_REG_EI = 0x200,
/* meta-flags, for intermediate stages of IR, ie.
* before register assignment is done:
*/
IR3_REG_SSA = 0x1000, /* 'instr' is ptr to assigning instr */
IR3_REG_IA = 0x2000, /* meta-input dst is "assigned" */
} flags;
union {
/* normal registers:
* the component is in the low two bits of the reg #, so
* rN.x becomes: (N << 2) | x
*/
int num;
/* immediate: */
int iim_val;
float fim_val;
/* relative: */
int offset;
/* for IR3_REG_SSA, src registers contain ptr back to
* assigning instruction.
*/
struct ir3_instruction *instr;
};
/* used for cat5 instructions, but also for internal/IR level
* tracking of what registers are read/written by an instruction.
* wrmask may be a bad name since it is used to represent both
* src and dst that touch multiple adjacent registers.
*/
int wrmask;
};
struct ir3_instruction {
struct ir3_block *block;
int category;
opc_t opc;
enum {
/* (sy) flag is set on first instruction, and after sample
* instructions (probably just on RAW hazard).
*/
IR3_INSTR_SY = 0x001,
/* (ss) flag is set on first instruction, and first instruction
* to depend on the result of "long" instructions (RAW hazard):
*
* rcp, rsq, log2, exp2, sin, cos, sqrt
*
* It seems to synchronize until all in-flight instructions are
* completed, for example:
*
* rsq hr1.w, hr1.w
* add.f hr2.z, (neg)hr2.z, hc0.y
* mul.f hr2.w, (neg)hr2.y, (neg)hr2.y
* rsq hr2.x, hr2.x
* (rpt1)nop
* mad.f16 hr2.w, hr2.z, hr2.z, hr2.w
* nop
* mad.f16 hr2.w, (neg)hr0.w, (neg)hr0.w, hr2.w
* (ss)(rpt2)mul.f hr1.x, (r)hr1.x, hr1.w
* (rpt2)mul.f hr0.x, (neg)(r)hr0.x, hr2.x
*
* The last mul.f does not have (ss) set, presumably because the
* (ss) on the previous instruction does the job.
*
* The blob driver also seems to set it on WAR hazards, although
* not really clear if this is needed or just blob compiler being
* sloppy. So far I haven't found a case where removing the (ss)
* causes problems for WAR hazard, but I could just be getting
* lucky:
*
* rcp r1.y, r3.y
* (ss)(rpt2)mad.f32 r3.y, (r)c9.x, r1.x, (r)r3.z
*
*/
IR3_INSTR_SS = 0x002,
/* (jp) flag is set on jump targets:
*/
IR3_INSTR_JP = 0x004,
IR3_INSTR_UL = 0x008,
IR3_INSTR_3D = 0x010,
IR3_INSTR_A = 0x020,
IR3_INSTR_O = 0x040,
IR3_INSTR_P = 0x080,
IR3_INSTR_S = 0x100,
IR3_INSTR_S2EN = 0x200,
/* meta-flags, for intermediate stages of IR, ie.
* before register assignment is done:
*/
IR3_INSTR_MARK = 0x1000,
} flags;
int repeat;
unsigned regs_count;
struct ir3_register *regs[5];
union {
struct {
char inv;
char comp;
int immed;
} cat0;
struct {
type_t src_type, dst_type;
} cat1;
struct {
enum {
IR3_COND_LT = 0,
IR3_COND_LE = 1,
IR3_COND_GT = 2,
IR3_COND_GE = 3,
IR3_COND_EQ = 4,
IR3_COND_NE = 5,
} condition;
} cat2;
struct {
unsigned samp, tex;
type_t type;
} cat5;
struct {
type_t type;
int offset;
int iim_val;
} cat6;
/* for meta-instructions, just used to hold extra data
* before instruction scheduling, etc
*/
struct {
int off; /* component/offset */
} fo;
struct {
struct ir3_block *if_block, *else_block;
} flow;
struct {
struct ir3_block *block;
} inout;
};
/* transient values used during various algorithms: */
union {
/* The instruction depth is the max dependency distance to output.
*
* You can also think of it as the "cost", if we did any sort of
* optimization for register footprint. Ie. a value that is just
* result of moving a const to a reg would have a low cost, so to
* it could make sense to duplicate the instruction at various
* points where the result is needed to reduce register footprint.
*/
unsigned depth;
};
struct ir3_instruction *next;
#ifdef DEBUG
uint32_t serialno;
#endif
};
#define MAX_INSTRS 1024
struct ir3_shader {
unsigned instrs_count;
struct ir3_instruction *instrs[MAX_INSTRS];
uint32_t heap[128 * MAX_INSTRS];
unsigned heap_idx;
};
struct ir3_block {
struct ir3_shader *shader;
unsigned ntemporaries, ninputs, noutputs;
/* maps TGSI_FILE_TEMPORARY index back to the assigning instruction: */
struct ir3_instruction **temporaries;
struct ir3_instruction **inputs;
struct ir3_instruction **outputs;
struct ir3_block *parent;
struct ir3_instruction *head;
};
struct ir3_shader * ir3_shader_create(void);
void ir3_shader_destroy(struct ir3_shader *shader);
void * ir3_shader_assemble(struct ir3_shader *shader,
struct ir3_shader_info *info);
void * ir3_alloc(struct ir3_shader *shader, int sz);
struct ir3_block * ir3_block_create(struct ir3_shader *shader,
unsigned ntmp, unsigned nin, unsigned nout);
struct ir3_instruction * ir3_instr_create(struct ir3_block *block,
int category, opc_t opc);
struct ir3_instruction * ir3_instr_clone(struct ir3_instruction *instr);
const char *ir3_instr_name(struct ir3_instruction *instr);
struct ir3_register * ir3_reg_create(struct ir3_instruction *instr,
int num, int flags);
static inline bool ir3_instr_check_mark(struct ir3_instruction *instr)
{
if (instr->flags & IR3_INSTR_MARK)
return true; /* already visited */
instr->flags ^= IR3_INSTR_MARK;
return false;
}
static inline void ir3_shader_clear_mark(struct ir3_shader *shader)
{
/* TODO would be nice to drop the instruction array.. for
* new compiler, _clear_mark() is all we use it for, and
* we could probably manage a linked list instead..
*/
unsigned i;
for (i = 0; i < shader->instrs_count; i++) {
struct ir3_instruction *instr = shader->instrs[i];
instr->flags &= ~IR3_INSTR_MARK;
}
}
static inline int ir3_instr_regno(struct ir3_instruction *instr,
struct ir3_register *reg)
{
unsigned i;
for (i = 0; i < instr->regs_count; i++)
if (reg == instr->regs[i])
return i;
return -1;
}
/* comp:
* 0 - x
* 1 - y
* 2 - z
* 3 - w
*/
static inline uint32_t regid(int num, int comp)
{
return (num << 2) | (comp & 0x3);
}
static inline uint32_t reg_num(struct ir3_register *reg)
{
return reg->num >> 2;
}
static inline uint32_t reg_comp(struct ir3_register *reg)
{
return reg->num & 0x3;
}
static inline bool is_flow(struct ir3_instruction *instr)
{
return (instr->category == 0);
}
static inline bool is_kill(struct ir3_instruction *instr)
{
return is_flow(instr) && (instr->opc == OPC_KILL);
}
static inline bool is_nop(struct ir3_instruction *instr)
{
return is_flow(instr) && (instr->opc == OPC_NOP);
}
static inline bool is_alu(struct ir3_instruction *instr)
{
return (1 <= instr->category) && (instr->category <= 3);
}
static inline bool is_sfu(struct ir3_instruction *instr)
{
return (instr->category == 4);
}
static inline bool is_tex(struct ir3_instruction *instr)
{
return (instr->category == 5);
}
static inline bool is_input(struct ir3_instruction *instr)
{
return (instr->category == 2) && (instr->opc == OPC_BARY_F);
}
static inline bool is_meta(struct ir3_instruction *instr)
{
/* TODO how should we count PHI (and maybe fan-in/out) which
* might actually contribute some instructions to the final
* result?
*/
return (instr->category == -1);
}
/* TODO combine is_gpr()/reg_gpr().. */
static inline bool reg_gpr(struct ir3_register *r)
{
if (r->flags & (IR3_REG_CONST | IR3_REG_IMMED | IR3_REG_RELATIV | IR3_REG_SSA))
return false;
if ((reg_num(r) == REG_A0) || (reg_num(r) == REG_P0))
return false;
return true;
}
/* dump: */
#include <stdio.h>
void ir3_shader_dump(struct ir3_shader *shader, const char *name,
struct ir3_block *block /* XXX maybe 'block' ptr should move to ir3_shader? */,
FILE *f);
void ir3_dump_instr_single(struct ir3_instruction *instr);
void ir3_dump_instr_list(struct ir3_instruction *instr);
/* flatten if/else: */
int ir3_block_flatten(struct ir3_block *block);
/* depth calculation: */
int ir3_delayslots(struct ir3_instruction *assigner,
struct ir3_instruction *consumer, unsigned n);
void ir3_block_depth(struct ir3_block *block);
/* copy-propagate: */
void ir3_block_cp(struct ir3_block *block);
/* scheduling: */
void ir3_block_sched(struct ir3_block *block);
/* register assignment: */
int ir3_block_ra(struct ir3_block *block, enum shader_t type,
bool half_precision, bool frag_coord, bool frag_face,
bool *has_samp);
#ifndef ARRAY_SIZE
# define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
#endif
/* ************************************************************************* */
/* split this out or find some helper to use.. like main/bitset.h.. */
#include <string.h>
#define MAX_REG 256
typedef uint8_t regmask_t[2 * MAX_REG / 8];
static inline unsigned regmask_idx(struct ir3_register *reg)
{
unsigned num = reg->num;
assert(num < MAX_REG);
if (reg->flags & IR3_REG_HALF)
num += MAX_REG;
return num;
}
static inline void regmask_init(regmask_t *regmask)
{
memset(regmask, 0, sizeof(*regmask));
}
static inline void regmask_set(regmask_t *regmask, struct ir3_register *reg)
{
unsigned idx = regmask_idx(reg);
unsigned i;
for (i = 0; i < 4; i++, idx++)
if (reg->wrmask & (1 << i))
(*regmask)[idx / 8] |= 1 << (idx % 8);
}
/* set bits in a if not set in b, conceptually:
* a |= (reg & ~b)
*/
static inline void regmask_set_if_not(regmask_t *a,
struct ir3_register *reg, regmask_t *b)
{
unsigned idx = regmask_idx(reg);
unsigned i;
for (i = 0; i < 4; i++, idx++)
if (reg->wrmask & (1 << i))
if (!((*b)[idx / 8] & (1 << (idx % 8))))
(*a)[idx / 8] |= 1 << (idx % 8);
}
static inline unsigned regmask_get(regmask_t *regmask,
struct ir3_register *reg)
{
unsigned idx = regmask_idx(reg);
unsigned i;
for (i = 0; i < 4; i++, idx++)
if (reg->wrmask & (1 << i))
if ((*regmask)[idx / 8] & (1 << (idx % 8)))
return true;
return false;
}
/* ************************************************************************* */
#endif /* IR3_H_ */
|