1 /*
2 * Copyright 2017 Advanced Micro Devices, Inc.
3 *
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:
10 *
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
13 * Software.
14 *
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.
22 */
23
24 /* This file handles register programming of primitive binning. */
25
26 #include "si_pipe.h"
27 #include "sid.h"
28 #include "gfx9d.h"
29 #include "radeon/r600_cs.h"
30
31 struct uvec2 {
32 unsigned x, y;
33 };
34
35 struct si_bin_size_map {
36 unsigned start;
37 unsigned bin_size_x;
38 unsigned bin_size_y;
39 };
40
41 typedef struct si_bin_size_map si_bin_size_subtable[3][9];
42
43 /* Find the bin size where sum is >= table[i].start and < table[i + 1].start. */
si_find_bin_size(struct si_screen * sscreen,const si_bin_size_subtable table[],unsigned sum)44 static struct uvec2 si_find_bin_size(struct si_screen *sscreen,
45 const si_bin_size_subtable table[],
46 unsigned sum)
47 {
48 unsigned log_num_rb_per_se =
49 util_logbase2_ceil(sscreen->info.num_render_backends /
50 sscreen->info.max_se);
51 unsigned log_num_se = util_logbase2_ceil(sscreen->info.max_se);
52 unsigned i;
53
54 /* Get the chip-specific subtable. */
55 const struct si_bin_size_map *subtable =
56 &table[log_num_rb_per_se][log_num_se][0];
57
58 for (i = 0; subtable[i].start != UINT_MAX; i++) {
59 if (sum >= subtable[i].start && sum < subtable[i + 1].start)
60 break;
61 }
62
63 struct uvec2 size = {subtable[i].bin_size_x, subtable[i].bin_size_y};
64 return size;
65 }
66
si_get_color_bin_size(struct si_context * sctx,unsigned cb_target_enabled_4bit)67 static struct uvec2 si_get_color_bin_size(struct si_context *sctx,
68 unsigned cb_target_enabled_4bit)
69 {
70 unsigned nr_samples = sctx->framebuffer.nr_samples;
71 unsigned sum = 0;
72
73 /* Compute the sum of all Bpp. */
74 for (unsigned i = 0; i < sctx->framebuffer.state.nr_cbufs; i++) {
75 if (!(cb_target_enabled_4bit & (0xf << (i * 4))))
76 continue;
77
78 struct r600_texture *rtex =
79 (struct r600_texture*)sctx->framebuffer.state.cbufs[i]->texture;
80 sum += rtex->surface.bpe;
81 }
82
83 /* Multiply the sum by some function of the number of samples. */
84 if (nr_samples >= 2) {
85 if (sctx->ps_iter_samples >= 2)
86 sum *= nr_samples;
87 else
88 sum *= 2;
89 }
90
91 static const si_bin_size_subtable table[] = {
92 {
93 /* One RB / SE */
94 {
95 /* One shader engine */
96 { 0, 128, 128 },
97 { 1, 64, 128 },
98 { 2, 32, 128 },
99 { 3, 16, 128 },
100 { 17, 0, 0 },
101 { UINT_MAX, 0, 0 },
102 },
103 {
104 /* Two shader engines */
105 { 0, 128, 128 },
106 { 2, 64, 128 },
107 { 3, 32, 128 },
108 { 5, 16, 128 },
109 { 17, 0, 0 },
110 { UINT_MAX, 0, 0 },
111 },
112 {
113 /* Four shader engines */
114 { 0, 128, 128 },
115 { 3, 64, 128 },
116 { 5, 16, 128 },
117 { 17, 0, 0 },
118 { UINT_MAX, 0, 0 },
119 },
120 },
121 {
122 /* Two RB / SE */
123 {
124 /* One shader engine */
125 { 0, 128, 128 },
126 { 2, 64, 128 },
127 { 3, 32, 128 },
128 { 5, 16, 128 },
129 { 33, 0, 0 },
130 { UINT_MAX, 0, 0 },
131 },
132 {
133 /* Two shader engines */
134 { 0, 128, 128 },
135 { 3, 64, 128 },
136 { 5, 32, 128 },
137 { 9, 16, 128 },
138 { 33, 0, 0 },
139 { UINT_MAX, 0, 0 },
140 },
141 {
142 /* Four shader engines */
143 { 0, 256, 256 },
144 { 2, 128, 256 },
145 { 3, 128, 128 },
146 { 5, 64, 128 },
147 { 9, 16, 128 },
148 { 33, 0, 0 },
149 { UINT_MAX, 0, 0 },
150 },
151 },
152 {
153 /* Four RB / SE */
154 {
155 /* One shader engine */
156 { 0, 128, 256 },
157 { 2, 128, 128 },
158 { 3, 64, 128 },
159 { 5, 32, 128 },
160 { 9, 16, 128 },
161 { 33, 0, 0 },
162 { UINT_MAX, 0, 0 },
163 },
164 {
165 /* Two shader engines */
166 { 0, 256, 256 },
167 { 2, 128, 256 },
168 { 3, 128, 128 },
169 { 5, 64, 128 },
170 { 9, 32, 128 },
171 { 17, 16, 128 },
172 { 33, 0, 0 },
173 { UINT_MAX, 0, 0 },
174 },
175 {
176 /* Four shader engines */
177 { 0, 256, 512 },
178 { 2, 256, 256 },
179 { 3, 128, 256 },
180 { 5, 128, 128 },
181 { 9, 64, 128 },
182 { 17, 16, 128 },
183 { 33, 0, 0 },
184 { UINT_MAX, 0, 0 },
185 },
186 },
187 };
188
189 return si_find_bin_size(sctx->screen, table, sum);
190 }
191
si_get_depth_bin_size(struct si_context * sctx)192 static struct uvec2 si_get_depth_bin_size(struct si_context *sctx)
193 {
194 struct si_state_dsa *dsa = sctx->queued.named.dsa;
195
196 if (!sctx->framebuffer.state.zsbuf ||
197 (!dsa->depth_enabled && !dsa->stencil_enabled)) {
198 /* Return the max size. */
199 struct uvec2 size = {512, 512};
200 return size;
201 }
202
203 struct r600_texture *rtex =
204 (struct r600_texture*)sctx->framebuffer.state.zsbuf->texture;
205 unsigned depth_coeff = dsa->depth_enabled ? 5 : 0;
206 unsigned stencil_coeff = rtex->surface.has_stencil &&
207 dsa->stencil_enabled ? 1 : 0;
208 unsigned sum = 4 * (depth_coeff + stencil_coeff) *
209 sctx->framebuffer.nr_samples;
210
211 static const si_bin_size_subtable table[] = {
212 {
213 // One RB / SE
214 {
215 // One shader engine
216 { 0, 128, 256 },
217 { 2, 128, 128 },
218 { 4, 64, 128 },
219 { 7, 32, 128 },
220 { 13, 16, 128 },
221 { 49, 0, 0 },
222 { UINT_MAX, 0, 0 },
223 },
224 {
225 // Two shader engines
226 { 0, 256, 256 },
227 { 2, 128, 256 },
228 { 4, 128, 128 },
229 { 7, 64, 128 },
230 { 13, 32, 128 },
231 { 25, 16, 128 },
232 { 49, 0, 0 },
233 { UINT_MAX, 0, 0 },
234 },
235 {
236 // Four shader engines
237 { 0, 256, 512 },
238 { 2, 256, 256 },
239 { 4, 128, 256 },
240 { 7, 128, 128 },
241 { 13, 64, 128 },
242 { 25, 16, 128 },
243 { 49, 0, 0 },
244 { UINT_MAX, 0, 0 },
245 },
246 },
247 {
248 // Two RB / SE
249 {
250 // One shader engine
251 { 0, 256, 256 },
252 { 2, 128, 256 },
253 { 4, 128, 128 },
254 { 7, 64, 128 },
255 { 13, 32, 128 },
256 { 25, 16, 128 },
257 { 97, 0, 0 },
258 { UINT_MAX, 0, 0 },
259 },
260 {
261 // Two shader engines
262 { 0, 256, 512 },
263 { 2, 256, 256 },
264 { 4, 128, 256 },
265 { 7, 128, 128 },
266 { 13, 64, 128 },
267 { 25, 32, 128 },
268 { 49, 16, 128 },
269 { 97, 0, 0 },
270 { UINT_MAX, 0, 0 },
271 },
272 {
273 // Four shader engines
274 { 0, 512, 512 },
275 { 2, 256, 512 },
276 { 4, 256, 256 },
277 { 7, 128, 256 },
278 { 13, 128, 128 },
279 { 25, 64, 128 },
280 { 49, 16, 128 },
281 { 97, 0, 0 },
282 { UINT_MAX, 0, 0 },
283 },
284 },
285 {
286 // Four RB / SE
287 {
288 // One shader engine
289 { 0, 256, 512 },
290 { 2, 256, 256 },
291 { 4, 128, 256 },
292 { 7, 128, 128 },
293 { 13, 64, 128 },
294 { 25, 32, 128 },
295 { 49, 16, 128 },
296 { UINT_MAX, 0, 0 },
297 },
298 {
299 // Two shader engines
300 { 0, 512, 512 },
301 { 2, 256, 512 },
302 { 4, 256, 256 },
303 { 7, 128, 256 },
304 { 13, 128, 128 },
305 { 25, 64, 128 },
306 { 49, 32, 128 },
307 { 97, 16, 128 },
308 { UINT_MAX, 0, 0 },
309 },
310 {
311 // Four shader engines
312 { 0, 512, 512 },
313 { 4, 256, 512 },
314 { 7, 256, 256 },
315 { 13, 128, 256 },
316 { 25, 128, 128 },
317 { 49, 64, 128 },
318 { 97, 16, 128 },
319 { UINT_MAX, 0, 0 },
320 },
321 },
322 };
323
324 return si_find_bin_size(sctx->screen, table, sum);
325 }
326
si_emit_dpbb_disable(struct si_context * sctx)327 static void si_emit_dpbb_disable(struct si_context *sctx)
328 {
329 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
330
331 radeon_set_context_reg(cs, R_028C44_PA_SC_BINNER_CNTL_0,
332 S_028C44_BINNING_MODE(V_028C44_DISABLE_BINNING_USE_LEGACY_SC) |
333 S_028C44_DISABLE_START_OF_PRIM(1));
334 radeon_set_context_reg(cs, R_028060_DB_DFSM_CONTROL,
335 S_028060_PUNCHOUT_MODE(V_028060_FORCE_OFF));
336 }
337
si_emit_dpbb_state(struct si_context * sctx,struct r600_atom * state)338 void si_emit_dpbb_state(struct si_context *sctx, struct r600_atom *state)
339 {
340 struct si_screen *sscreen = sctx->screen;
341 struct si_state_blend *blend = sctx->queued.named.blend;
342 struct si_state_dsa *dsa = sctx->queued.named.dsa;
343 unsigned db_shader_control = sctx->ps_db_shader_control;
344
345 assert(sctx->b.chip_class >= GFX9);
346
347 if (!sscreen->dpbb_allowed || !blend || !dsa) {
348 si_emit_dpbb_disable(sctx);
349 return;
350 }
351
352 bool ps_can_kill = G_02880C_KILL_ENABLE(db_shader_control) ||
353 G_02880C_MASK_EXPORT_ENABLE(db_shader_control) ||
354 G_02880C_COVERAGE_TO_MASK_ENABLE(db_shader_control) ||
355 blend->alpha_to_coverage;
356
357 /* This is ported from Vulkan, but it doesn't make much sense to me.
358 * Maybe it's for RE-Z? But Vulkan doesn't use RE-Z. TODO: Clarify this.
359 */
360 bool ps_can_reject_z_trivially =
361 !G_02880C_Z_EXPORT_ENABLE(db_shader_control) ||
362 G_02880C_CONSERVATIVE_Z_EXPORT(db_shader_control);
363
364 /* Disable binning if PS can kill trivially with DB writes.
365 * Ported from Vulkan. (heuristic?)
366 */
367 if (ps_can_kill &&
368 ps_can_reject_z_trivially &&
369 sctx->framebuffer.state.zsbuf &&
370 dsa->db_can_write) {
371 si_emit_dpbb_disable(sctx);
372 return;
373 }
374
375 /* Compute the bin size. */
376 /* TODO: We could also look at enabled pixel shader outputs. */
377 unsigned cb_target_enabled_4bit = sctx->framebuffer.colorbuf_enabled_4bit &
378 blend->cb_target_enabled_4bit;
379 struct uvec2 color_bin_size =
380 si_get_color_bin_size(sctx, cb_target_enabled_4bit);
381 struct uvec2 depth_bin_size = si_get_depth_bin_size(sctx);
382
383 unsigned color_area = color_bin_size.x * color_bin_size.y;
384 unsigned depth_area = depth_bin_size.x * depth_bin_size.y;
385
386 struct uvec2 bin_size = color_area < depth_area ? color_bin_size
387 : depth_bin_size;
388
389 if (!bin_size.x || !bin_size.y) {
390 si_emit_dpbb_disable(sctx);
391 return;
392 }
393
394 /* Enable DFSM if it's preferred. */
395 unsigned punchout_mode = V_028060_FORCE_OFF;
396 bool disable_start_of_prim = true;
397
398 if (sscreen->dfsm_allowed &&
399 cb_target_enabled_4bit &&
400 !G_02880C_KILL_ENABLE(db_shader_control) &&
401 /* These two also imply that DFSM is disabled when PS writes to memory. */
402 !G_02880C_EXEC_ON_HIER_FAIL(db_shader_control) &&
403 !G_02880C_EXEC_ON_NOOP(db_shader_control) &&
404 G_02880C_Z_ORDER(db_shader_control) == V_02880C_EARLY_Z_THEN_LATE_Z) {
405 punchout_mode = V_028060_AUTO;
406 disable_start_of_prim = (cb_target_enabled_4bit &
407 blend->blend_enable_4bit) != 0;
408 }
409
410 /* Tunable parameters. Also test with DFSM enabled/disabled. */
411 unsigned context_states_per_bin; /* allowed range: [0, 5] */
412 unsigned persistent_states_per_bin; /* allowed range: [0, 31] */
413 unsigned fpovs_per_batch; /* allowed range: [0, 255], 0 = unlimited */
414
415 switch (sctx->b.family) {
416 case CHIP_VEGA10:
417 case CHIP_RAVEN:
418 /* Tuned for Raven. Vega might need different values. */
419 context_states_per_bin = 5;
420 persistent_states_per_bin = 31;
421 fpovs_per_batch = 63;
422 break;
423 default:
424 assert(0);
425 }
426
427 /* Emit registers. */
428 struct uvec2 bin_size_extend = {};
429 if (bin_size.x >= 32)
430 bin_size_extend.x = util_logbase2(bin_size.x) - 5;
431 if (bin_size.y >= 32)
432 bin_size_extend.y = util_logbase2(bin_size.y) - 5;
433
434 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
435 radeon_set_context_reg(cs, R_028C44_PA_SC_BINNER_CNTL_0,
436 S_028C44_BINNING_MODE(V_028C44_BINNING_ALLOWED) |
437 S_028C44_BIN_SIZE_X(bin_size.x == 16) |
438 S_028C44_BIN_SIZE_Y(bin_size.y == 16) |
439 S_028C44_BIN_SIZE_X_EXTEND(bin_size_extend.x) |
440 S_028C44_BIN_SIZE_Y_EXTEND(bin_size_extend.y) |
441 S_028C44_CONTEXT_STATES_PER_BIN(context_states_per_bin) |
442 S_028C44_PERSISTENT_STATES_PER_BIN(persistent_states_per_bin) |
443 S_028C44_DISABLE_START_OF_PRIM(disable_start_of_prim) |
444 S_028C44_FPOVS_PER_BATCH(fpovs_per_batch) |
445 S_028C44_OPTIMAL_BIN_SELECTION(1));
446 radeon_set_context_reg(cs, R_028060_DB_DFSM_CONTROL,
447 S_028060_PUNCHOUT_MODE(punchout_mode));
448 }
449