si_state_binning.c 18.6 KB
Newer Older
1
2
/*
 * Copyright 2017 Advanced Micro Devices, Inc.
Marek Olšák's avatar
Marek Olšák committed
3
 * All Rights Reserved.
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
 *
 * 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
 * on the rights to use, copy, modify, merge, publish, distribute, sub
 * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS 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.
 */

/* This file handles register programming of primitive binning. */

27
#include "si_build_pm4.h"
28
#include "sid.h"
29
30

struct uvec2 {
31
   unsigned x, y;
32
33
34
};

struct si_bin_size_map {
35
36
37
   unsigned start;
   unsigned bin_size_x;
   unsigned bin_size_y;
38
39
};

40
typedef struct si_bin_size_map si_bin_size_subtable[3][10];
41
42

/* Find the bin size where sum is >= table[i].start and < table[i + 1].start. */
43
44
static struct uvec2 si_find_bin_size(struct si_screen *sscreen, const si_bin_size_subtable table[],
                                     unsigned sum)
45
{
46
   unsigned log_num_rb_per_se =
47
      util_logbase2_ceil(sscreen->info.max_render_backends / sscreen->info.max_se);
48
49
50
51
52
53
54
55
56
57
58
59
60
   unsigned log_num_se = util_logbase2_ceil(sscreen->info.max_se);
   unsigned i;

   /* Get the chip-specific subtable. */
   const struct si_bin_size_map *subtable = &table[log_num_rb_per_se][log_num_se][0];

   for (i = 0; subtable[i].bin_size_x != 0; i++) {
      if (sum >= subtable[i].start && sum < subtable[i + 1].start)
         break;
   }

   struct uvec2 size = {subtable[i].bin_size_x, subtable[i].bin_size_y};
   return size;
61
62
}

63
static struct uvec2 si_get_color_bin_size(struct si_context *sctx, unsigned cb_target_enabled_4bit)
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
   unsigned num_fragments = sctx->framebuffer.nr_color_samples;
   unsigned sum = 0;

   /* Compute the sum of all Bpp. */
   for (unsigned i = 0; i < sctx->framebuffer.state.nr_cbufs; i++) {
      if (!(cb_target_enabled_4bit & (0xf << (i * 4))))
         continue;

      struct si_texture *tex = (struct si_texture *)sctx->framebuffer.state.cbufs[i]->texture;
      sum += tex->surface.bpe;
   }

   /* Multiply the sum by some function of the number of samples. */
   if (num_fragments >= 2) {
      if (si_get_ps_iter_samples(sctx) >= 2)
         sum *= num_fragments;
      else
         sum *= 2;
   }

   static const si_bin_size_subtable table[] = {
      {
         /* One RB / SE */
         {
            /* One shader engine */
            {0, 128, 128},
            {1, 64, 128},
            {2, 32, 128},
            {3, 16, 128},
            {17, 0, 0},
         },
         {
            /* Two shader engines */
            {0, 128, 128},
            {2, 64, 128},
            {3, 32, 128},
            {5, 16, 128},
            {17, 0, 0},
         },
         {
            /* Four shader engines */
            {0, 128, 128},
            {3, 64, 128},
            {5, 16, 128},
            {17, 0, 0},
         },
      },
      {
         /* Two RB / SE */
         {
            /* One shader engine */
            {0, 128, 128},
            {2, 64, 128},
            {3, 32, 128},
            {9, 16, 128},
            {33, 0, 0},
         },
         {
            /* Two shader engines */
            {0, 128, 128},
            {3, 64, 128},
            {5, 32, 128},
            {9, 16, 128},
            {33, 0, 0},
         },
         {
            /* Four shader engines */
            {0, 256, 256},
            {2, 128, 256},
            {3, 128, 128},
            {5, 64, 128},
            {9, 16, 128},
            {33, 0, 0},
         },
      },
      {
         /* Four RB / SE */
         {
            /* One shader engine */
            {0, 128, 256},
            {2, 128, 128},
            {3, 64, 128},
            {5, 32, 128},
            {9, 16, 128},
            {17, 0, 0},
         },
         {
            /* Two shader engines */
            {0, 256, 256},
            {2, 128, 256},
            {3, 128, 128},
            {5, 64, 128},
            {9, 32, 128},
            {17, 16, 128},
            {33, 0, 0},
         },
         {
            /* Four shader engines */
            {0, 256, 512},
            {2, 128, 512},
            {3, 64, 512},
            {5, 32, 512},
            {9, 32, 256},
            {17, 32, 128},
            {33, 0, 0},
         },
      },
   };

   return si_find_bin_size(sctx->screen, table, sum);
175
176
177
178
}

static struct uvec2 si_get_depth_bin_size(struct si_context *sctx)
{
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
   struct si_state_dsa *dsa = sctx->queued.named.dsa;

   if (!sctx->framebuffer.state.zsbuf || (!dsa->depth_enabled && !dsa->stencil_enabled)) {
      /* Return the max size. */
      struct uvec2 size = {512, 512};
      return size;
   }

   struct si_texture *tex = (struct si_texture *)sctx->framebuffer.state.zsbuf->texture;
   unsigned depth_coeff = dsa->depth_enabled ? 5 : 0;
   unsigned stencil_coeff = tex->surface.has_stencil && dsa->stencil_enabled ? 1 : 0;
   unsigned sum = 4 * (depth_coeff + stencil_coeff) * MAX2(tex->buffer.b.b.nr_samples, 1);

   static const si_bin_size_subtable table[] = {
      {
         // One RB / SE
         {
            // One shader engine
            {0, 64, 512},
            {2, 64, 256},
            {4, 64, 128},
            {7, 32, 128},
            {13, 16, 128},
            {49, 0, 0},
         },
         {
            // Two shader engines
            {0, 128, 512},
            {2, 64, 512},
            {4, 64, 256},
            {7, 64, 128},
            {13, 32, 128},
            {25, 16, 128},
            {49, 0, 0},
         },
         {
            // Four shader engines
            {0, 256, 512},
            {2, 128, 512},
            {4, 64, 512},
            {7, 64, 256},
            {13, 64, 128},
            {25, 16, 128},
            {49, 0, 0},
         },
      },
      {
         // Two RB / SE
         {
            // One shader engine
            {0, 128, 512},
            {2, 64, 512},
            {4, 64, 256},
            {7, 64, 128},
            {13, 32, 128},
            {25, 16, 128},
            {97, 0, 0},
         },
         {
            // Two shader engines
            {0, 256, 512},
            {2, 128, 512},
            {4, 64, 512},
            {7, 64, 256},
            {13, 64, 128},
            {25, 32, 128},
            {49, 16, 128},
            {97, 0, 0},
         },
         {
            // Four shader engines
            {0, 512, 512},
            {2, 256, 512},
            {4, 128, 512},
            {7, 64, 512},
            {13, 64, 256},
            {25, 64, 128},
            {49, 16, 128},
            {97, 0, 0},
         },
      },
      {
         // Four RB / SE
         {
            // One shader engine
            {0, 256, 512},
            {2, 128, 512},
            {4, 64, 512},
            {7, 64, 256},
            {13, 64, 128},
            {25, 32, 128},
            {49, 16, 128},
            {193, 0, 0},
         },
         {
            // Two shader engines
            {0, 512, 512},
            {2, 256, 512},
            {4, 128, 512},
            {7, 64, 512},
            {13, 64, 256},
            {25, 64, 128},
            {49, 32, 128},
            {97, 16, 128},
            {193, 0, 0},
         },
         {
            // Four shader engines
            {0, 512, 512},
            {4, 256, 512},
            {7, 128, 512},
            {13, 64, 512},
            {25, 32, 512},
            {49, 32, 256},
            {97, 16, 128},
            {193, 0, 0},
         },
      },
   };

   return si_find_bin_size(sctx->screen, table, sum);
300
301
}

302
303
static void gfx10_get_bin_sizes(struct si_context *sctx, unsigned cb_target_enabled_4bit,
                                struct uvec2 *color_bin_size, struct uvec2 *depth_bin_size)
304
{
305
306
307
308
309
310
311
   const unsigned ZsTagSize = 64;
   const unsigned ZsNumTags = 312;
   const unsigned CcTagSize = 1024;
   const unsigned CcReadTags = 31;
   const unsigned FcTagSize = 256;
   const unsigned FcReadTags = 44;

312
   const unsigned num_rbs = sctx->screen->info.max_render_backends;
313
   const unsigned num_pipes = MAX2(num_rbs, sctx->screen->info.num_tcc_blocks);
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

   const unsigned depthBinSizeTagPart =
      ((ZsNumTags * num_rbs / num_pipes) * (ZsTagSize * num_pipes));
   const unsigned colorBinSizeTagPart =
      ((CcReadTags * num_rbs / num_pipes) * (CcTagSize * num_pipes));
   const unsigned fmaskBinSizeTagPart =
      ((FcReadTags * num_rbs / num_pipes) * (FcTagSize * num_pipes));

   const unsigned minBinSizeX = 128;
   const unsigned minBinSizeY = 64;

   const unsigned num_fragments = sctx->framebuffer.nr_color_samples;
   const unsigned num_samples = sctx->framebuffer.nr_samples;
   const bool ps_iter_sample = si_get_ps_iter_samples(sctx) >= 2;

   /* Calculate cColor and cFmask(if applicable) */
   unsigned cColor = 0;
   unsigned cFmask = 0;
   bool has_fmask = false;

   for (unsigned i = 0; i < sctx->framebuffer.state.nr_cbufs; i++) {
      if (!sctx->framebuffer.state.cbufs[i])
         continue;

      struct si_texture *tex = (struct si_texture *)sctx->framebuffer.state.cbufs[i]->texture;
      const unsigned mmrt = num_fragments == 1 ? 1 : (ps_iter_sample ? num_fragments : 2);

      cColor += tex->surface.bpe * mmrt;
      if (num_samples >= 2 /* if FMASK is bound */) {
         const unsigned fragmentsLog2 = util_logbase2(num_fragments);
         const unsigned samplesLog2 = util_logbase2(num_samples);

         static const unsigned cFmaskMrt[4 /* fragments */][5 /* samples */] = {
            {0, 1, 1, 1, 2}, /* fragments = 1 */
            {0, 1, 1, 2, 4}, /* fragments = 2 */
            {0, 1, 1, 4, 8}, /* fragments = 4 */
            {0, 1, 2, 4, 8}  /* fragments = 8 */
         };
         cFmask += cFmaskMrt[fragmentsLog2][samplesLog2];
         has_fmask = true;
      }
   }
   cColor = MAX2(cColor, 1u);

   const unsigned colorLog2Pixels = util_logbase2(colorBinSizeTagPart / cColor);
   const unsigned colorBinSizeX = 1 << ((colorLog2Pixels + 1) / 2); /* round up width */
   const unsigned colorBinSizeY = 1 << (colorLog2Pixels / 2);       /* round down height */

   unsigned binSizeX = colorBinSizeX;
   unsigned binSizeY = colorBinSizeY;

   if (has_fmask) {
      cFmask = MAX2(cFmask, 1u);

      const unsigned fmaskLog2Pixels = util_logbase2(fmaskBinSizeTagPart / cFmask);
      const unsigned fmaskBinSizeX = 1 << ((fmaskLog2Pixels + 1) / 2); /* round up width */
      const unsigned fmaskBinSizeY = 1 << (fmaskLog2Pixels / 2);       /* round down height */

      /* use the smaller of the Color vs. Fmask bin sizes */
      if (fmaskLog2Pixels < colorLog2Pixels) {
         binSizeX = fmaskBinSizeX;
         binSizeY = fmaskBinSizeY;
      }
   }

   /* Return size adjusted for minimum bin size */
   color_bin_size->x = MAX2(binSizeX, minBinSizeX);
   color_bin_size->y = MAX2(binSizeY, minBinSizeY);

   if (!sctx->framebuffer.state.zsbuf) {
      /* Set to max sizes when no depth buffer is bound. */
      depth_bin_size->x = 512;
      depth_bin_size->y = 512;
   } else {
      struct si_texture *zstex = (struct si_texture *)sctx->framebuffer.state.zsbuf->texture;
      struct si_state_dsa *dsa = sctx->queued.named.dsa;

      const unsigned cPerDepthSample = dsa->depth_enabled ? 5 : 0;
      const unsigned cPerStencilSample = dsa->stencil_enabled ? 1 : 0;
      const unsigned cDepth =
         (cPerDepthSample + cPerStencilSample) * MAX2(zstex->buffer.b.b.nr_samples, 1);

      const unsigned depthLog2Pixels = util_logbase2(depthBinSizeTagPart / MAX2(cDepth, 1u));
      unsigned depthBinSizeX = 1 << ((depthLog2Pixels + 1) / 2);
      unsigned depthBinSizeY = 1 << (depthLog2Pixels / 2);

      depth_bin_size->x = MAX2(depthBinSizeX, minBinSizeX);
      depth_bin_size->y = MAX2(depthBinSizeY, minBinSizeY);
   }
403
404
}

405
406
static void si_emit_dpbb_disable(struct si_context *sctx)
{
407
   radeon_begin(&sctx->gfx_cs);
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

   if (sctx->chip_class >= GFX10) {
      struct uvec2 bin_size = {};
      struct uvec2 bin_size_extend = {};

      bin_size.x = 128;
      bin_size.y = sctx->framebuffer.min_bytes_per_pixel <= 4 ? 128 : 64;

      if (bin_size.x >= 32)
         bin_size_extend.x = util_logbase2(bin_size.x) - 5;
      if (bin_size.y >= 32)
         bin_size_extend.y = util_logbase2(bin_size.y) - 5;

      radeon_opt_set_context_reg(
         sctx, R_028C44_PA_SC_BINNER_CNTL_0, SI_TRACKED_PA_SC_BINNER_CNTL_0,
         S_028C44_BINNING_MODE(V_028C44_DISABLE_BINNING_USE_NEW_SC) |
            S_028C44_BIN_SIZE_X(bin_size.x == 16) | S_028C44_BIN_SIZE_Y(bin_size.y == 16) |
            S_028C44_BIN_SIZE_X_EXTEND(bin_size_extend.x) |
            S_028C44_BIN_SIZE_Y_EXTEND(bin_size_extend.y) | S_028C44_DISABLE_START_OF_PRIM(1) |
            S_028C44_FLUSH_ON_BINNING_TRANSITION(sctx->last_binning_enabled != 0));
   } else {
      radeon_opt_set_context_reg(
         sctx, R_028C44_PA_SC_BINNER_CNTL_0, SI_TRACKED_PA_SC_BINNER_CNTL_0,
         S_028C44_BINNING_MODE(V_028C44_DISABLE_BINNING_USE_LEGACY_SC) |
            S_028C44_DISABLE_START_OF_PRIM(1) |
            S_028C44_FLUSH_ON_BINNING_TRANSITION((sctx->family == CHIP_VEGA12 ||
                                                  sctx->family == CHIP_VEGA20 ||
                                                  sctx->family >= CHIP_RAVEN2) &&
                                                 sctx->last_binning_enabled != 0));
   }

   unsigned db_dfsm_control =
      sctx->chip_class >= GFX10 ? R_028038_DB_DFSM_CONTROL : R_028060_DB_DFSM_CONTROL;
   radeon_opt_set_context_reg(
      sctx, db_dfsm_control, SI_TRACKED_DB_DFSM_CONTROL,
      S_028060_PUNCHOUT_MODE(V_028060_FORCE_OFF) | S_028060_POPS_DRAIN_PS_ON_OVERLAP(1));
444
   radeon_end_update_context_roll(sctx);
445
446

   sctx->last_binning_enabled = false;
447
448
}

449
void si_emit_dpbb_state(struct si_context *sctx)
450
{
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
   struct si_screen *sscreen = sctx->screen;
   struct si_state_blend *blend = sctx->queued.named.blend;
   struct si_state_dsa *dsa = sctx->queued.named.dsa;
   unsigned db_shader_control = sctx->ps_db_shader_control;

   assert(sctx->chip_class >= GFX9);

   if (!sscreen->dpbb_allowed || sctx->dpbb_force_off) {
      si_emit_dpbb_disable(sctx);
      return;
   }

   bool ps_can_kill =
      G_02880C_KILL_ENABLE(db_shader_control) || G_02880C_MASK_EXPORT_ENABLE(db_shader_control) ||
      G_02880C_COVERAGE_TO_MASK_ENABLE(db_shader_control) || blend->alpha_to_coverage;

   bool db_can_reject_z_trivially = !G_02880C_Z_EXPORT_ENABLE(db_shader_control) ||
                                    G_02880C_CONSERVATIVE_Z_EXPORT(db_shader_control) ||
                                    G_02880C_DEPTH_BEFORE_SHADER(db_shader_control);

   /* Disable DPBB when it's believed to be inefficient. */
472
   if (sscreen->info.max_render_backends > 4 && ps_can_kill && db_can_reject_z_trivially &&
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
       sctx->framebuffer.state.zsbuf && dsa->db_can_write) {
      si_emit_dpbb_disable(sctx);
      return;
   }

   /* Compute the bin size. */
   /* TODO: We could also look at enabled pixel shader outputs. */
   unsigned cb_target_enabled_4bit =
      sctx->framebuffer.colorbuf_enabled_4bit & blend->cb_target_enabled_4bit;
   struct uvec2 color_bin_size, depth_bin_size;

   if (sctx->chip_class >= GFX10) {
      gfx10_get_bin_sizes(sctx, cb_target_enabled_4bit, &color_bin_size, &depth_bin_size);
   } else {
      color_bin_size = si_get_color_bin_size(sctx, cb_target_enabled_4bit);
      depth_bin_size = si_get_depth_bin_size(sctx);
   }

   unsigned color_area = color_bin_size.x * color_bin_size.y;
   unsigned depth_area = depth_bin_size.x * depth_bin_size.y;

   struct uvec2 bin_size = color_area < depth_area ? color_bin_size : depth_bin_size;

   if (!bin_size.x || !bin_size.y) {
      si_emit_dpbb_disable(sctx);
      return;
   }

   /* Enable DFSM if it's preferred. */
   unsigned punchout_mode = V_028060_FORCE_OFF;
   bool disable_start_of_prim = true;
   bool zs_eqaa_dfsm_bug =
      sctx->chip_class == GFX9 && sctx->framebuffer.state.zsbuf &&
      sctx->framebuffer.nr_samples != MAX2(1, sctx->framebuffer.state.zsbuf->texture->nr_samples);

   if (sscreen->dfsm_allowed && !zs_eqaa_dfsm_bug && cb_target_enabled_4bit &&
       !G_02880C_KILL_ENABLE(db_shader_control) &&
       /* These two also imply that DFSM is disabled when PS writes to memory. */
       !G_02880C_EXEC_ON_HIER_FAIL(db_shader_control) &&
       !G_02880C_EXEC_ON_NOOP(db_shader_control) &&
       G_02880C_Z_ORDER(db_shader_control) == V_02880C_EARLY_Z_THEN_LATE_Z) {
      punchout_mode = V_028060_AUTO;
      disable_start_of_prim = (cb_target_enabled_4bit & blend->blend_enable_4bit) != 0;
   }

518
519
   /* Tunable parameters. */
   unsigned fpovs_per_batch = 63; /* allowed range: [0, 255], 0 = unlimited */
520
521
522
523
524
525
526
527

   /* Emit registers. */
   struct uvec2 bin_size_extend = {};
   if (bin_size.x >= 32)
      bin_size_extend.x = util_logbase2(bin_size.x) - 5;
   if (bin_size.y >= 32)
      bin_size_extend.y = util_logbase2(bin_size.y) - 5;

528
   radeon_begin(&sctx->gfx_cs);
529
530
531
532
533
   radeon_opt_set_context_reg(
      sctx, R_028C44_PA_SC_BINNER_CNTL_0, SI_TRACKED_PA_SC_BINNER_CNTL_0,
      S_028C44_BINNING_MODE(V_028C44_BINNING_ALLOWED) | S_028C44_BIN_SIZE_X(bin_size.x == 16) |
         S_028C44_BIN_SIZE_Y(bin_size.y == 16) | S_028C44_BIN_SIZE_X_EXTEND(bin_size_extend.x) |
         S_028C44_BIN_SIZE_Y_EXTEND(bin_size_extend.y) |
534
535
         S_028C44_CONTEXT_STATES_PER_BIN(sscreen->pbb_context_states_per_bin - 1) |
         S_028C44_PERSISTENT_STATES_PER_BIN(sscreen->pbb_persistent_states_per_bin - 1) |
536
537
538
539
540
541
542
543
544
545
546
547
         S_028C44_DISABLE_START_OF_PRIM(disable_start_of_prim) |
         S_028C44_FPOVS_PER_BATCH(fpovs_per_batch) | S_028C44_OPTIMAL_BIN_SELECTION(1) |
         S_028C44_FLUSH_ON_BINNING_TRANSITION((sctx->family == CHIP_VEGA12 ||
                                               sctx->family == CHIP_VEGA20 ||
                                               sctx->family >= CHIP_RAVEN2) &&
                                              sctx->last_binning_enabled != 1));

   unsigned db_dfsm_control =
      sctx->chip_class >= GFX10 ? R_028038_DB_DFSM_CONTROL : R_028060_DB_DFSM_CONTROL;
   radeon_opt_set_context_reg(
      sctx, db_dfsm_control, SI_TRACKED_DB_DFSM_CONTROL,
      S_028060_PUNCHOUT_MODE(punchout_mode) | S_028060_POPS_DRAIN_PS_ON_OVERLAP(1));
548
   radeon_end_update_context_roll(sctx);
549
550

   sctx->last_binning_enabled = true;
551
}