projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radeonsi: set almost optimal settings in SC_MODE_CNTL_1
[mesa.git] / src / gallium / drivers / radeonsi / si_state.c
1 /*
2 * Copyright 2012 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 * Authors:
24 * Christian König <christian.koenig@amd.com>
25 */
26
27 #include "si_pipe.h"
28 #include "si_shader.h"
29 #include "sid.h"
30 #include "radeon/r600_cs.h"
31
32 #include "util/u_dual_blend.h"
33 #include "util/u_format.h"
34 #include "util/u_format_s3tc.h"
35 #include "util/u_memory.h"
36 #include "util/u_pstipple.h"
37 #include "util/u_resource.h"
38
39 /* Initialize an external atom (owned by ../radeon). */
40 static void
41 si_init_external_atom(struct si_context *sctx, struct r600_atom *atom,
42 struct r600_atom **list_elem)
43 {
44 atom->id = list_elem - sctx->atoms.array + 1;
45 *list_elem = atom;
46 }
47
48 /* Initialize an atom owned by radeonsi. */
49 void si_init_atom(struct si_context *sctx, struct r600_atom *atom,
50 struct r600_atom **list_elem,
51 void (*emit_func)(struct si_context *ctx, struct r600_atom *state))
52 {
53 atom->emit = (void*)emit_func;
54 atom->id = list_elem - sctx->atoms.array + 1; /* index+1 in the atom array */
55 *list_elem = atom;
56 }
57
58 static unsigned si_map_swizzle(unsigned swizzle)
59 {
60 switch (swizzle) {
61 case PIPE_SWIZZLE_Y:
62 return V_008F0C_SQ_SEL_Y;
63 case PIPE_SWIZZLE_Z:
64 return V_008F0C_SQ_SEL_Z;
65 case PIPE_SWIZZLE_W:
66 return V_008F0C_SQ_SEL_W;
67 case PIPE_SWIZZLE_0:
68 return V_008F0C_SQ_SEL_0;
69 case PIPE_SWIZZLE_1:
70 return V_008F0C_SQ_SEL_1;
71 default: /* PIPE_SWIZZLE_X */
72 return V_008F0C_SQ_SEL_X;
73 }
74 }
75
76 static uint32_t S_FIXED(float value, uint32_t frac_bits)
77 {
78 return value * (1 << frac_bits);
79 }
80
81 /* 12.4 fixed-point */
82 static unsigned si_pack_float_12p4(float x)
83 {
84 return x <= 0 ? 0 :
85 x >= 4096 ? 0xffff : x * 16;
86 }
87
88 /*
89 * Inferred framebuffer and blender state.
90 *
91 * One of the reasons CB_TARGET_MASK must be derived from the framebuffer state
92 * is that:
93 * - The blend state mask is 0xf most of the time.
94 * - The COLOR1 format isn't INVALID because of possible dual-source blending,
95 * so COLOR1 is enabled pretty much all the time.
96 * So CB_TARGET_MASK is the only register that can disable COLOR1.
97 *
98 * Another reason is to avoid a hang with dual source blending.
99 */
100 static void si_emit_cb_render_state(struct si_context *sctx, struct r600_atom *atom)
101 {
102 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
103 struct si_state_blend *blend = sctx->queued.named.blend;
104 uint32_t cb_target_mask = 0, i;
105
106 for (i = 0; i < sctx->framebuffer.state.nr_cbufs; i++)
107 if (sctx->framebuffer.state.cbufs[i])
108 cb_target_mask |= 0xf << (4*i);
109
110 if (blend)
111 cb_target_mask &= blend->cb_target_mask;
112
113 /* Avoid a hang that happens when dual source blending is enabled
114 * but there is not enough color outputs. This is undefined behavior,
115 * so disable color writes completely.
116 *
117 * Reproducible with Unigine Heaven 4.0 and drirc missing.
118 */
119 if (blend && blend->dual_src_blend &&
120 sctx->ps_shader.cso &&
121 (sctx->ps_shader.cso->info.colors_written & 0x3) != 0x3)
122 cb_target_mask = 0;
123
124 radeon_set_context_reg(cs, R_028238_CB_TARGET_MASK, cb_target_mask);
125
126 /* STONEY-specific register settings. */
127 if (sctx->b.family == CHIP_STONEY) {
128 unsigned spi_shader_col_format =
129 sctx->ps_shader.cso ?
130 sctx->ps_shader.current->key.ps.epilog.spi_shader_col_format : 0;
131 unsigned sx_ps_downconvert = 0;
132 unsigned sx_blend_opt_epsilon = 0;
133 unsigned sx_blend_opt_control = 0;
134
135 for (i = 0; i < sctx->framebuffer.state.nr_cbufs; i++) {
136 struct r600_surface *surf =
137 (struct r600_surface*)sctx->framebuffer.state.cbufs[i];
138 unsigned format, swap, spi_format, colormask;
139 bool has_alpha, has_rgb;
140
141 if (!surf)
142 continue;
143
144 format = G_028C70_FORMAT(surf->cb_color_info);
145 swap = G_028C70_COMP_SWAP(surf->cb_color_info);
146 spi_format = (spi_shader_col_format >> (i * 4)) & 0xf;
147 colormask = (cb_target_mask >> (i * 4)) & 0xf;
148
149 /* Set if RGB and A are present. */
150 has_alpha = !G_028C74_FORCE_DST_ALPHA_1(surf->cb_color_attrib);
151
152 if (format == V_028C70_COLOR_8 ||
153 format == V_028C70_COLOR_16 ||
154 format == V_028C70_COLOR_32)
155 has_rgb = !has_alpha;
156 else
157 has_rgb = true;
158
159 /* Check the colormask and export format. */
160 if (!(colormask & (PIPE_MASK_RGBA & ~PIPE_MASK_A)))
161 has_rgb = false;
162 if (!(colormask & PIPE_MASK_A))
163 has_alpha = false;
164
165 if (spi_format == V_028714_SPI_SHADER_ZERO) {
166 has_rgb = false;
167 has_alpha = false;
168 }
169
170 /* Disable value checking for disabled channels. */
171 if (!has_rgb)
172 sx_blend_opt_control |= S_02875C_MRT0_COLOR_OPT_DISABLE(1) << (i * 4);
173 if (!has_alpha)
174 sx_blend_opt_control |= S_02875C_MRT0_ALPHA_OPT_DISABLE(1) << (i * 4);
175
176 /* Enable down-conversion for 32bpp and smaller formats. */
177 switch (format) {
178 case V_028C70_COLOR_8:
179 case V_028C70_COLOR_8_8:
180 case V_028C70_COLOR_8_8_8_8:
181 /* For 1 and 2-channel formats, use the superset thereof. */
182 if (spi_format == V_028714_SPI_SHADER_FP16_ABGR ||
183 spi_format == V_028714_SPI_SHADER_UINT16_ABGR ||
184 spi_format == V_028714_SPI_SHADER_SINT16_ABGR) {
185 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_8_8_8_8 << (i * 4);
186 sx_blend_opt_epsilon |= V_028758_8BIT_FORMAT << (i * 4);
187 }
188 break;
189
190 case V_028C70_COLOR_5_6_5:
191 if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
192 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_5_6_5 << (i * 4);
193 sx_blend_opt_epsilon |= V_028758_6BIT_FORMAT << (i * 4);
194 }
195 break;
196
197 case V_028C70_COLOR_1_5_5_5:
198 if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
199 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_1_5_5_5 << (i * 4);
200 sx_blend_opt_epsilon |= V_028758_5BIT_FORMAT << (i * 4);
201 }
202 break;
203
204 case V_028C70_COLOR_4_4_4_4:
205 if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
206 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_4_4_4_4 << (i * 4);
207 sx_blend_opt_epsilon |= V_028758_4BIT_FORMAT << (i * 4);
208 }
209 break;
210
211 case V_028C70_COLOR_32:
212 if (swap == V_0280A0_SWAP_STD &&
213 spi_format == V_028714_SPI_SHADER_32_R)
214 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_32_R << (i * 4);
215 else if (swap == V_0280A0_SWAP_ALT_REV &&
216 spi_format == V_028714_SPI_SHADER_32_AR)
217 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_32_A << (i * 4);
218 break;
219
220 case V_028C70_COLOR_16:
221 case V_028C70_COLOR_16_16:
222 /* For 1-channel formats, use the superset thereof. */
223 if (spi_format == V_028714_SPI_SHADER_UNORM16_ABGR ||
224 spi_format == V_028714_SPI_SHADER_SNORM16_ABGR ||
225 spi_format == V_028714_SPI_SHADER_UINT16_ABGR ||
226 spi_format == V_028714_SPI_SHADER_SINT16_ABGR) {
227 if (swap == V_0280A0_SWAP_STD ||
228 swap == V_0280A0_SWAP_STD_REV)
229 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_16_16_GR << (i * 4);
230 else
231 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_16_16_AR << (i * 4);
232 }
233 break;
234
235 case V_028C70_COLOR_10_11_11:
236 if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
237 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_10_11_11 << (i * 4);
238 sx_blend_opt_epsilon |= V_028758_11BIT_FORMAT << (i * 4);
239 }
240 break;
241
242 case V_028C70_COLOR_2_10_10_10:
243 if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
244 sx_ps_downconvert |= V_028754_SX_RT_EXPORT_2_10_10_10 << (i * 4);
245 sx_blend_opt_epsilon |= V_028758_10BIT_FORMAT << (i * 4);
246 }
247 break;
248 }
249 }
250
251 if (sctx->screen->b.debug_flags & DBG_NO_RB_PLUS) {
252 sx_ps_downconvert = 0;
253 sx_blend_opt_epsilon = 0;
254 sx_blend_opt_control = 0;
255 }
256
257 radeon_set_context_reg_seq(cs, R_028754_SX_PS_DOWNCONVERT, 3);
258 radeon_emit(cs, sx_ps_downconvert); /* R_028754_SX_PS_DOWNCONVERT */
259 radeon_emit(cs, sx_blend_opt_epsilon); /* R_028758_SX_BLEND_OPT_EPSILON */
260 radeon_emit(cs, sx_blend_opt_control); /* R_02875C_SX_BLEND_OPT_CONTROL */
261 }
262 }
263
264 /*
265 * Blender functions
266 */
267
268 static uint32_t si_translate_blend_function(int blend_func)
269 {
270 switch (blend_func) {
271 case PIPE_BLEND_ADD:
272 return V_028780_COMB_DST_PLUS_SRC;
273 case PIPE_BLEND_SUBTRACT:
274 return V_028780_COMB_SRC_MINUS_DST;
275 case PIPE_BLEND_REVERSE_SUBTRACT:
276 return V_028780_COMB_DST_MINUS_SRC;
277 case PIPE_BLEND_MIN:
278 return V_028780_COMB_MIN_DST_SRC;
279 case PIPE_BLEND_MAX:
280 return V_028780_COMB_MAX_DST_SRC;
281 default:
282 R600_ERR("Unknown blend function %d\n", blend_func);
283 assert(0);
284 break;
285 }
286 return 0;
287 }
288
289 static uint32_t si_translate_blend_factor(int blend_fact)
290 {
291 switch (blend_fact) {
292 case PIPE_BLENDFACTOR_ONE:
293 return V_028780_BLEND_ONE;
294 case PIPE_BLENDFACTOR_SRC_COLOR:
295 return V_028780_BLEND_SRC_COLOR;
296 case PIPE_BLENDFACTOR_SRC_ALPHA:
297 return V_028780_BLEND_SRC_ALPHA;
298 case PIPE_BLENDFACTOR_DST_ALPHA:
299 return V_028780_BLEND_DST_ALPHA;
300 case PIPE_BLENDFACTOR_DST_COLOR:
301 return V_028780_BLEND_DST_COLOR;
302 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
303 return V_028780_BLEND_SRC_ALPHA_SATURATE;
304 case PIPE_BLENDFACTOR_CONST_COLOR:
305 return V_028780_BLEND_CONSTANT_COLOR;
306 case PIPE_BLENDFACTOR_CONST_ALPHA:
307 return V_028780_BLEND_CONSTANT_ALPHA;
308 case PIPE_BLENDFACTOR_ZERO:
309 return V_028780_BLEND_ZERO;
310 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
311 return V_028780_BLEND_ONE_MINUS_SRC_COLOR;
312 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
313 return V_028780_BLEND_ONE_MINUS_SRC_ALPHA;
314 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
315 return V_028780_BLEND_ONE_MINUS_DST_ALPHA;
316 case PIPE_BLENDFACTOR_INV_DST_COLOR:
317 return V_028780_BLEND_ONE_MINUS_DST_COLOR;
318 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
319 return V_028780_BLEND_ONE_MINUS_CONSTANT_COLOR;
320 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
321 return V_028780_BLEND_ONE_MINUS_CONSTANT_ALPHA;
322 case PIPE_BLENDFACTOR_SRC1_COLOR:
323 return V_028780_BLEND_SRC1_COLOR;
324 case PIPE_BLENDFACTOR_SRC1_ALPHA:
325 return V_028780_BLEND_SRC1_ALPHA;
326 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
327 return V_028780_BLEND_INV_SRC1_COLOR;
328 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
329 return V_028780_BLEND_INV_SRC1_ALPHA;
330 default:
331 R600_ERR("Bad blend factor %d not supported!\n", blend_fact);
332 assert(0);
333 break;
334 }
335 return 0;
336 }
337
338 static uint32_t si_translate_blend_opt_function(int blend_func)
339 {
340 switch (blend_func) {
341 case PIPE_BLEND_ADD:
342 return V_028760_OPT_COMB_ADD;
343 case PIPE_BLEND_SUBTRACT:
344 return V_028760_OPT_COMB_SUBTRACT;
345 case PIPE_BLEND_REVERSE_SUBTRACT:
346 return V_028760_OPT_COMB_REVSUBTRACT;
347 case PIPE_BLEND_MIN:
348 return V_028760_OPT_COMB_MIN;
349 case PIPE_BLEND_MAX:
350 return V_028760_OPT_COMB_MAX;
351 default:
352 return V_028760_OPT_COMB_BLEND_DISABLED;
353 }
354 }
355
356 static uint32_t si_translate_blend_opt_factor(int blend_fact, bool is_alpha)
357 {
358 switch (blend_fact) {
359 case PIPE_BLENDFACTOR_ZERO:
360 return V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_ALL;
361 case PIPE_BLENDFACTOR_ONE:
362 return V_028760_BLEND_OPT_PRESERVE_ALL_IGNORE_NONE;
363 case PIPE_BLENDFACTOR_SRC_COLOR:
364 return is_alpha ? V_028760_BLEND_OPT_PRESERVE_A1_IGNORE_A0
365 : V_028760_BLEND_OPT_PRESERVE_C1_IGNORE_C0;
366 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
367 return is_alpha ? V_028760_BLEND_OPT_PRESERVE_A0_IGNORE_A1
368 : V_028760_BLEND_OPT_PRESERVE_C0_IGNORE_C1;
369 case PIPE_BLENDFACTOR_SRC_ALPHA:
370 return V_028760_BLEND_OPT_PRESERVE_A1_IGNORE_A0;
371 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
372 return V_028760_BLEND_OPT_PRESERVE_A0_IGNORE_A1;
373 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
374 return is_alpha ? V_028760_BLEND_OPT_PRESERVE_ALL_IGNORE_NONE
375 : V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_A0;
376 default:
377 return V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
378 }
379 }
380
381 /**
382 * Get rid of DST in the blend factors by commuting the operands:
383 * func(src * DST, dst * 0) ---> func(src * 0, dst * SRC)
384 */
385 static void si_blend_remove_dst(unsigned *func, unsigned *src_factor,
386 unsigned *dst_factor, unsigned expected_dst,
387 unsigned replacement_src)
388 {
389 if (*src_factor == expected_dst &&
390 *dst_factor == PIPE_BLENDFACTOR_ZERO) {
391 *src_factor = PIPE_BLENDFACTOR_ZERO;
392 *dst_factor = replacement_src;
393
394 /* Commuting the operands requires reversing subtractions. */
395 if (*func == PIPE_BLEND_SUBTRACT)
396 *func = PIPE_BLEND_REVERSE_SUBTRACT;
397 else if (*func == PIPE_BLEND_REVERSE_SUBTRACT)
398 *func = PIPE_BLEND_SUBTRACT;
399 }
400 }
401
402 static bool si_blend_factor_uses_dst(unsigned factor)
403 {
404 return factor == PIPE_BLENDFACTOR_DST_COLOR ||
405 factor == PIPE_BLENDFACTOR_DST_ALPHA ||
406 factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
407 factor == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
408 factor == PIPE_BLENDFACTOR_INV_DST_COLOR;
409 }
410
411 static void *si_create_blend_state_mode(struct pipe_context *ctx,
412 const struct pipe_blend_state *state,
413 unsigned mode)
414 {
415 struct si_context *sctx = (struct si_context*)ctx;
416 struct si_state_blend *blend = CALLOC_STRUCT(si_state_blend);
417 struct si_pm4_state *pm4 = &blend->pm4;
418 uint32_t sx_mrt_blend_opt[8] = {0};
419 uint32_t color_control = 0;
420
421 if (!blend)
422 return NULL;
423
424 blend->alpha_to_coverage = state->alpha_to_coverage;
425 blend->alpha_to_one = state->alpha_to_one;
426 blend->dual_src_blend = util_blend_state_is_dual(state, 0);
427
428 if (state->logicop_enable) {
429 color_control |= S_028808_ROP3(state->logicop_func | (state->logicop_func << 4));
430 } else {
431 color_control |= S_028808_ROP3(0xcc);
432 }
433
434 si_pm4_set_reg(pm4, R_028B70_DB_ALPHA_TO_MASK,
435 S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
436 S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
437 S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
438 S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
439 S_028B70_ALPHA_TO_MASK_OFFSET3(2));
440
441 if (state->alpha_to_coverage)
442 blend->need_src_alpha_4bit |= 0xf;
443
444 blend->cb_target_mask = 0;
445 for (int i = 0; i < 8; i++) {
446 /* state->rt entries > 0 only written if independent blending */
447 const int j = state->independent_blend_enable ? i : 0;
448
449 unsigned eqRGB = state->rt[j].rgb_func;
450 unsigned srcRGB = state->rt[j].rgb_src_factor;
451 unsigned dstRGB = state->rt[j].rgb_dst_factor;
452 unsigned eqA = state->rt[j].alpha_func;
453 unsigned srcA = state->rt[j].alpha_src_factor;
454 unsigned dstA = state->rt[j].alpha_dst_factor;
455
456 unsigned srcRGB_opt, dstRGB_opt, srcA_opt, dstA_opt;
457 unsigned blend_cntl = 0;
458
459 sx_mrt_blend_opt[i] =
460 S_028760_COLOR_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED) |
461 S_028760_ALPHA_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED);
462
463 if (!state->rt[j].colormask)
464 continue;
465
466 /* cb_render_state will disable unused ones */
467 blend->cb_target_mask |= (unsigned)state->rt[j].colormask << (4 * i);
468
469 if (!state->rt[j].blend_enable) {
470 si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
471 continue;
472 }
473
474 /* Blending optimizations for Stoney.
475 * These transformations don't change the behavior.
476 *
477 * First, get rid of DST in the blend factors:
478 * func(src * DST, dst * 0) ---> func(src * 0, dst * SRC)
479 */
480 si_blend_remove_dst(&eqRGB, &srcRGB, &dstRGB,
481 PIPE_BLENDFACTOR_DST_COLOR,
482 PIPE_BLENDFACTOR_SRC_COLOR);
483 si_blend_remove_dst(&eqA, &srcA, &dstA,
484 PIPE_BLENDFACTOR_DST_COLOR,
485 PIPE_BLENDFACTOR_SRC_COLOR);
486 si_blend_remove_dst(&eqA, &srcA, &dstA,
487 PIPE_BLENDFACTOR_DST_ALPHA,
488 PIPE_BLENDFACTOR_SRC_ALPHA);
489
490 /* Look up the ideal settings from tables. */
491 srcRGB_opt = si_translate_blend_opt_factor(srcRGB, false);
492 dstRGB_opt = si_translate_blend_opt_factor(dstRGB, false);
493 srcA_opt = si_translate_blend_opt_factor(srcA, true);
494 dstA_opt = si_translate_blend_opt_factor(dstA, true);
495
496 /* Handle interdependencies. */
497 if (si_blend_factor_uses_dst(srcRGB))
498 dstRGB_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
499 if (si_blend_factor_uses_dst(srcA))
500 dstA_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
501
502 if (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE &&
503 (dstRGB == PIPE_BLENDFACTOR_ZERO ||
504 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
505 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE))
506 dstRGB_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_A0;
507
508 /* Set the final value. */
509 sx_mrt_blend_opt[i] =
510 S_028760_COLOR_SRC_OPT(srcRGB_opt) |
511 S_028760_COLOR_DST_OPT(dstRGB_opt) |
512 S_028760_COLOR_COMB_FCN(si_translate_blend_opt_function(eqRGB)) |
513 S_028760_ALPHA_SRC_OPT(srcA_opt) |
514 S_028760_ALPHA_DST_OPT(dstA_opt) |
515 S_028760_ALPHA_COMB_FCN(si_translate_blend_opt_function(eqA));
516
517 /* Set blend state. */
518 blend_cntl |= S_028780_ENABLE(1);
519 blend_cntl |= S_028780_COLOR_COMB_FCN(si_translate_blend_function(eqRGB));
520 blend_cntl |= S_028780_COLOR_SRCBLEND(si_translate_blend_factor(srcRGB));
521 blend_cntl |= S_028780_COLOR_DESTBLEND(si_translate_blend_factor(dstRGB));
522
523 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
524 blend_cntl |= S_028780_SEPARATE_ALPHA_BLEND(1);
525 blend_cntl |= S_028780_ALPHA_COMB_FCN(si_translate_blend_function(eqA));
526 blend_cntl |= S_028780_ALPHA_SRCBLEND(si_translate_blend_factor(srcA));
527 blend_cntl |= S_028780_ALPHA_DESTBLEND(si_translate_blend_factor(dstA));
528 }
529 si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
530
531 blend->blend_enable_4bit |= 0xfu << (i * 4);
532
533 /* This is only important for formats without alpha. */
534 if (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
535 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
536 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
537 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
538 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
539 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA)
540 blend->need_src_alpha_4bit |= 0xfu << (i * 4);
541 }
542
543 if (blend->cb_target_mask) {
544 color_control |= S_028808_MODE(mode);
545 } else {
546 color_control |= S_028808_MODE(V_028808_CB_DISABLE);
547 }
548
549 if (sctx->b.family == CHIP_STONEY) {
550 for (int i = 0; i < 8; i++)
551 si_pm4_set_reg(pm4, R_028760_SX_MRT0_BLEND_OPT + i * 4,
552 sx_mrt_blend_opt[i]);
553
554 /* RB+ doesn't work with dual source blending, logic op, and RESOLVE. */
555 if (blend->dual_src_blend || state->logicop_enable ||
556 mode == V_028808_CB_RESOLVE)
557 color_control |= S_028808_DISABLE_DUAL_QUAD(1);
558 }
559
560 si_pm4_set_reg(pm4, R_028808_CB_COLOR_CONTROL, color_control);
561 return blend;
562 }
563
564 static void *si_create_blend_state(struct pipe_context *ctx,
565 const struct pipe_blend_state *state)
566 {
567 return si_create_blend_state_mode(ctx, state, V_028808_CB_NORMAL);
568 }
569
570 static void si_bind_blend_state(struct pipe_context *ctx, void *state)
571 {
572 struct si_context *sctx = (struct si_context *)ctx;
573 si_pm4_bind_state(sctx, blend, (struct si_state_blend *)state);
574 si_mark_atom_dirty(sctx, &sctx->cb_render_state);
575 }
576
577 static void si_delete_blend_state(struct pipe_context *ctx, void *state)
578 {
579 struct si_context *sctx = (struct si_context *)ctx;
580 si_pm4_delete_state(sctx, blend, (struct si_state_blend *)state);
581 }
582
583 static void si_set_blend_color(struct pipe_context *ctx,
584 const struct pipe_blend_color *state)
585 {
586 struct si_context *sctx = (struct si_context *)ctx;
587
588 if (memcmp(&sctx->blend_color.state, state, sizeof(*state)) == 0)
589 return;
590
591 sctx->blend_color.state = *state;
592 si_mark_atom_dirty(sctx, &sctx->blend_color.atom);
593 }
594
595 static void si_emit_blend_color(struct si_context *sctx, struct r600_atom *atom)
596 {
597 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
598
599 radeon_set_context_reg_seq(cs, R_028414_CB_BLEND_RED, 4);
600 radeon_emit_array(cs, (uint32_t*)sctx->blend_color.state.color, 4);
601 }
602
603 /*
604 * Clipping
605 */
606
607 static void si_set_clip_state(struct pipe_context *ctx,
608 const struct pipe_clip_state *state)
609 {
610 struct si_context *sctx = (struct si_context *)ctx;
611 struct pipe_constant_buffer cb;
612
613 if (memcmp(&sctx->clip_state.state, state, sizeof(*state)) == 0)
614 return;
615
616 sctx->clip_state.state = *state;
617 si_mark_atom_dirty(sctx, &sctx->clip_state.atom);
618
619 cb.buffer = NULL;
620 cb.user_buffer = state->ucp;
621 cb.buffer_offset = 0;
622 cb.buffer_size = 4*4*8;
623 si_set_rw_buffer(sctx, SI_VS_CONST_CLIP_PLANES, &cb);
624 pipe_resource_reference(&cb.buffer, NULL);
625 }
626
627 static void si_emit_clip_state(struct si_context *sctx, struct r600_atom *atom)
628 {
629 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
630
631 radeon_set_context_reg_seq(cs, R_0285BC_PA_CL_UCP_0_X, 6*4);
632 radeon_emit_array(cs, (uint32_t*)sctx->clip_state.state.ucp, 6*4);
633 }
634
635 #define SIX_BITS 0x3F
636
637 static void si_emit_clip_regs(struct si_context *sctx, struct r600_atom *atom)
638 {
639 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
640 struct tgsi_shader_info *info = si_get_vs_info(sctx);
641 unsigned window_space =
642 info->properties[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION];
643 unsigned clipdist_mask =
644 info->writes_clipvertex ? SIX_BITS : info->clipdist_writemask;
645
646 radeon_set_context_reg(cs, R_02881C_PA_CL_VS_OUT_CNTL,
647 S_02881C_USE_VTX_POINT_SIZE(info->writes_psize) |
648 S_02881C_USE_VTX_EDGE_FLAG(info->writes_edgeflag) |
649 S_02881C_USE_VTX_RENDER_TARGET_INDX(info->writes_layer) |
650 S_02881C_USE_VTX_VIEWPORT_INDX(info->writes_viewport_index) |
651 S_02881C_VS_OUT_CCDIST0_VEC_ENA((clipdist_mask & 0x0F) != 0) |
652 S_02881C_VS_OUT_CCDIST1_VEC_ENA((clipdist_mask & 0xF0) != 0) |
653 S_02881C_VS_OUT_MISC_VEC_ENA(info->writes_psize ||
654 info->writes_edgeflag ||
655 info->writes_layer ||
656 info->writes_viewport_index) |
657 S_02881C_VS_OUT_MISC_SIDE_BUS_ENA(1) |
658 (sctx->queued.named.rasterizer->clip_plane_enable &
659 clipdist_mask));
660 radeon_set_context_reg(cs, R_028810_PA_CL_CLIP_CNTL,
661 sctx->queued.named.rasterizer->pa_cl_clip_cntl |
662 (clipdist_mask ? 0 :
663 sctx->queued.named.rasterizer->clip_plane_enable & SIX_BITS) |
664 S_028810_CLIP_DISABLE(window_space));
665
666 /* reuse needs to be set off if we write oViewport */
667 radeon_set_context_reg(cs, R_028AB4_VGT_REUSE_OFF,
668 S_028AB4_REUSE_OFF(info->writes_viewport_index));
669 }
670
671 /*
672 * inferred state between framebuffer and rasterizer
673 */
674 static void si_update_poly_offset_state(struct si_context *sctx)
675 {
676 struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
677
678 if (!rs || !rs->uses_poly_offset || !sctx->framebuffer.state.zsbuf)
679 return;
680
681 switch (sctx->framebuffer.state.zsbuf->texture->format) {
682 case PIPE_FORMAT_Z16_UNORM:
683 si_pm4_bind_state(sctx, poly_offset, &rs->pm4_poly_offset[0]);
684 break;
685 default: /* 24-bit */
686 si_pm4_bind_state(sctx, poly_offset, &rs->pm4_poly_offset[1]);
687 break;
688 case PIPE_FORMAT_Z32_FLOAT:
689 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
690 si_pm4_bind_state(sctx, poly_offset, &rs->pm4_poly_offset[2]);
691 break;
692 }
693 }
694
695 /*
696 * Rasterizer
697 */
698
699 static uint32_t si_translate_fill(uint32_t func)
700 {
701 switch(func) {
702 case PIPE_POLYGON_MODE_FILL:
703 return V_028814_X_DRAW_TRIANGLES;
704 case PIPE_POLYGON_MODE_LINE:
705 return V_028814_X_DRAW_LINES;
706 case PIPE_POLYGON_MODE_POINT:
707 return V_028814_X_DRAW_POINTS;
708 default:
709 assert(0);
710 return V_028814_X_DRAW_POINTS;
711 }
712 }
713
714 static void *si_create_rs_state(struct pipe_context *ctx,
715 const struct pipe_rasterizer_state *state)
716 {
717 struct si_state_rasterizer *rs = CALLOC_STRUCT(si_state_rasterizer);
718 struct si_pm4_state *pm4 = &rs->pm4;
719 unsigned tmp, i;
720 float psize_min, psize_max;
721
722 if (!rs) {
723 return NULL;
724 }
725
726 rs->scissor_enable = state->scissor;
727 rs->two_side = state->light_twoside;
728 rs->multisample_enable = state->multisample;
729 rs->force_persample_interp = state->force_persample_interp;
730 rs->clip_plane_enable = state->clip_plane_enable;
731 rs->line_stipple_enable = state->line_stipple_enable;
732 rs->poly_stipple_enable = state->poly_stipple_enable;
733 rs->line_smooth = state->line_smooth;
734 rs->poly_smooth = state->poly_smooth;
735 rs->uses_poly_offset = state->offset_point || state->offset_line ||
736 state->offset_tri;
737 rs->clamp_fragment_color = state->clamp_fragment_color;
738 rs->flatshade = state->flatshade;
739 rs->sprite_coord_enable = state->sprite_coord_enable;
740 rs->rasterizer_discard = state->rasterizer_discard;
741 rs->pa_sc_line_stipple = state->line_stipple_enable ?
742 S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
743 S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
744 rs->pa_cl_clip_cntl =
745 S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) |
746 S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
747 S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
748 S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard) |
749 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
750
751 si_pm4_set_reg(pm4, R_0286D4_SPI_INTERP_CONTROL_0,
752 S_0286D4_FLAT_SHADE_ENA(1) |
753 S_0286D4_PNT_SPRITE_ENA(1) |
754 S_0286D4_PNT_SPRITE_OVRD_X(V_0286D4_SPI_PNT_SPRITE_SEL_S) |
755 S_0286D4_PNT_SPRITE_OVRD_Y(V_0286D4_SPI_PNT_SPRITE_SEL_T) |
756 S_0286D4_PNT_SPRITE_OVRD_Z(V_0286D4_SPI_PNT_SPRITE_SEL_0) |
757 S_0286D4_PNT_SPRITE_OVRD_W(V_0286D4_SPI_PNT_SPRITE_SEL_1) |
758 S_0286D4_PNT_SPRITE_TOP_1(state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT));
759
760 /* point size 12.4 fixed point */
761 tmp = (unsigned)(state->point_size * 8.0);
762 si_pm4_set_reg(pm4, R_028A00_PA_SU_POINT_SIZE, S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
763
764 if (state->point_size_per_vertex) {
765 psize_min = util_get_min_point_size(state);
766 psize_max = 8192;
767 } else {
768 /* Force the point size to be as if the vertex output was disabled. */
769 psize_min = state->point_size;
770 psize_max = state->point_size;
771 }
772 /* Divide by two, because 0.5 = 1 pixel. */
773 si_pm4_set_reg(pm4, R_028A04_PA_SU_POINT_MINMAX,
774 S_028A04_MIN_SIZE(si_pack_float_12p4(psize_min/2)) |
775 S_028A04_MAX_SIZE(si_pack_float_12p4(psize_max/2)));
776
777 tmp = (unsigned)state->line_width * 8;
778 si_pm4_set_reg(pm4, R_028A08_PA_SU_LINE_CNTL, S_028A08_WIDTH(tmp));
779 si_pm4_set_reg(pm4, R_028A48_PA_SC_MODE_CNTL_0,
780 S_028A48_LINE_STIPPLE_ENABLE(state->line_stipple_enable) |
781 S_028A48_MSAA_ENABLE(state->multisample ||
782 state->poly_smooth ||
783 state->line_smooth) |
784 S_028A48_VPORT_SCISSOR_ENABLE(1));
785
786 si_pm4_set_reg(pm4, R_028BE4_PA_SU_VTX_CNTL,
787 S_028BE4_PIX_CENTER(state->half_pixel_center) |
788 S_028BE4_QUANT_MODE(V_028BE4_X_16_8_FIXED_POINT_1_256TH));
789
790 si_pm4_set_reg(pm4, R_028B7C_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
791 si_pm4_set_reg(pm4, R_028814_PA_SU_SC_MODE_CNTL,
792 S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) |
793 S_028814_CULL_FRONT((state->cull_face & PIPE_FACE_FRONT) ? 1 : 0) |
794 S_028814_CULL_BACK((state->cull_face & PIPE_FACE_BACK) ? 1 : 0) |
795 S_028814_FACE(!state->front_ccw) |
796 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state, state->fill_front)) |
797 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state, state->fill_back)) |
798 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_point || state->offset_line) |
799 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL ||
800 state->fill_back != PIPE_POLYGON_MODE_FILL) |
801 S_028814_POLYMODE_FRONT_PTYPE(si_translate_fill(state->fill_front)) |
802 S_028814_POLYMODE_BACK_PTYPE(si_translate_fill(state->fill_back)));
803 si_pm4_set_reg(pm4, R_00B130_SPI_SHADER_USER_DATA_VS_0 +
804 SI_SGPR_VS_STATE_BITS * 4, state->clamp_vertex_color);
805
806 /* Precalculate polygon offset states for 16-bit, 24-bit, and 32-bit zbuffers. */
807 for (i = 0; i < 3; i++) {
808 struct si_pm4_state *pm4 = &rs->pm4_poly_offset[i];
809 float offset_units = state->offset_units;
810 float offset_scale = state->offset_scale * 16.0f;
811
812 switch (i) {
813 case 0: /* 16-bit zbuffer */
814 offset_units *= 4.0f;
815 break;
816 case 1: /* 24-bit zbuffer */
817 offset_units *= 2.0f;
818 break;
819 case 2: /* 32-bit zbuffer */
820 offset_units *= 1.0f;
821 break;
822 }
823
824 si_pm4_set_reg(pm4, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE,
825 fui(offset_scale));
826 si_pm4_set_reg(pm4, R_028B84_PA_SU_POLY_OFFSET_FRONT_OFFSET,
827 fui(offset_units));
828 si_pm4_set_reg(pm4, R_028B88_PA_SU_POLY_OFFSET_BACK_SCALE,
829 fui(offset_scale));
830 si_pm4_set_reg(pm4, R_028B8C_PA_SU_POLY_OFFSET_BACK_OFFSET,
831 fui(offset_units));
832 }
833
834 return rs;
835 }
836
837 static void si_bind_rs_state(struct pipe_context *ctx, void *state)
838 {
839 struct si_context *sctx = (struct si_context *)ctx;
840 struct si_state_rasterizer *old_rs =
841 (struct si_state_rasterizer*)sctx->queued.named.rasterizer;
842 struct si_state_rasterizer *rs = (struct si_state_rasterizer *)state;
843
844 if (!state)
845 return;
846
847 if (sctx->framebuffer.nr_samples > 1 &&
848 (!old_rs || old_rs->multisample_enable != rs->multisample_enable))
849 si_mark_atom_dirty(sctx, &sctx->db_render_state);
850
851 r600_set_scissor_enable(&sctx->b, rs->scissor_enable);
852
853 si_pm4_bind_state(sctx, rasterizer, rs);
854 si_update_poly_offset_state(sctx);
855
856 si_mark_atom_dirty(sctx, &sctx->clip_regs);
857 }
858
859 static void si_delete_rs_state(struct pipe_context *ctx, void *state)
860 {
861 struct si_context *sctx = (struct si_context *)ctx;
862
863 if (sctx->queued.named.rasterizer == state)
864 si_pm4_bind_state(sctx, poly_offset, NULL);
865 si_pm4_delete_state(sctx, rasterizer, (struct si_state_rasterizer *)state);
866 }
867
868 /*
869 * infeered state between dsa and stencil ref
870 */
871 static void si_emit_stencil_ref(struct si_context *sctx, struct r600_atom *atom)
872 {
873 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
874 struct pipe_stencil_ref *ref = &sctx->stencil_ref.state;
875 struct si_dsa_stencil_ref_part *dsa = &sctx->stencil_ref.dsa_part;
876
877 radeon_set_context_reg_seq(cs, R_028430_DB_STENCILREFMASK, 2);
878 radeon_emit(cs, S_028430_STENCILTESTVAL(ref->ref_value[0]) |
879 S_028430_STENCILMASK(dsa->valuemask[0]) |
880 S_028430_STENCILWRITEMASK(dsa->writemask[0]) |
881 S_028430_STENCILOPVAL(1));
882 radeon_emit(cs, S_028434_STENCILTESTVAL_BF(ref->ref_value[1]) |
883 S_028434_STENCILMASK_BF(dsa->valuemask[1]) |
884 S_028434_STENCILWRITEMASK_BF(dsa->writemask[1]) |
885 S_028434_STENCILOPVAL_BF(1));
886 }
887
888 static void si_set_stencil_ref(struct pipe_context *ctx,
889 const struct pipe_stencil_ref *state)
890 {
891 struct si_context *sctx = (struct si_context *)ctx;
892
893 if (memcmp(&sctx->stencil_ref.state, state, sizeof(*state)) == 0)
894 return;
895
896 sctx->stencil_ref.state = *state;
897 si_mark_atom_dirty(sctx, &sctx->stencil_ref.atom);
898 }
899
900
901 /*
902 * DSA
903 */
904
905 static uint32_t si_translate_stencil_op(int s_op)
906 {
907 switch (s_op) {
908 case PIPE_STENCIL_OP_KEEP:
909 return V_02842C_STENCIL_KEEP;
910 case PIPE_STENCIL_OP_ZERO:
911 return V_02842C_STENCIL_ZERO;
912 case PIPE_STENCIL_OP_REPLACE:
913 return V_02842C_STENCIL_REPLACE_TEST;
914 case PIPE_STENCIL_OP_INCR:
915 return V_02842C_STENCIL_ADD_CLAMP;
916 case PIPE_STENCIL_OP_DECR:
917 return V_02842C_STENCIL_SUB_CLAMP;
918 case PIPE_STENCIL_OP_INCR_WRAP:
919 return V_02842C_STENCIL_ADD_WRAP;
920 case PIPE_STENCIL_OP_DECR_WRAP:
921 return V_02842C_STENCIL_SUB_WRAP;
922 case PIPE_STENCIL_OP_INVERT:
923 return V_02842C_STENCIL_INVERT;
924 default:
925 R600_ERR("Unknown stencil op %d", s_op);
926 assert(0);
927 break;
928 }
929 return 0;
930 }
931
932 static void *si_create_dsa_state(struct pipe_context *ctx,
933 const struct pipe_depth_stencil_alpha_state *state)
934 {
935 struct si_state_dsa *dsa = CALLOC_STRUCT(si_state_dsa);
936 struct si_pm4_state *pm4 = &dsa->pm4;
937 unsigned db_depth_control;
938 uint32_t db_stencil_control = 0;
939
940 if (!dsa) {
941 return NULL;
942 }
943
944 dsa->stencil_ref.valuemask[0] = state->stencil[0].valuemask;
945 dsa->stencil_ref.valuemask[1] = state->stencil[1].valuemask;
946 dsa->stencil_ref.writemask[0] = state->stencil[0].writemask;
947 dsa->stencil_ref.writemask[1] = state->stencil[1].writemask;
948
949 db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) |
950 S_028800_Z_WRITE_ENABLE(state->depth.writemask) |
951 S_028800_ZFUNC(state->depth.func) |
952 S_028800_DEPTH_BOUNDS_ENABLE(state->depth.bounds_test);
953
954 /* stencil */
955 if (state->stencil[0].enabled) {
956 db_depth_control |= S_028800_STENCIL_ENABLE(1);
957 db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func);
958 db_stencil_control |= S_02842C_STENCILFAIL(si_translate_stencil_op(state->stencil[0].fail_op));
959 db_stencil_control |= S_02842C_STENCILZPASS(si_translate_stencil_op(state->stencil[0].zpass_op));
960 db_stencil_control |= S_02842C_STENCILZFAIL(si_translate_stencil_op(state->stencil[0].zfail_op));
961
962 if (state->stencil[1].enabled) {
963 db_depth_control |= S_028800_BACKFACE_ENABLE(1);
964 db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func);
965 db_stencil_control |= S_02842C_STENCILFAIL_BF(si_translate_stencil_op(state->stencil[1].fail_op));
966 db_stencil_control |= S_02842C_STENCILZPASS_BF(si_translate_stencil_op(state->stencil[1].zpass_op));
967 db_stencil_control |= S_02842C_STENCILZFAIL_BF(si_translate_stencil_op(state->stencil[1].zfail_op));
968 }
969 }
970
971 /* alpha */
972 if (state->alpha.enabled) {
973 dsa->alpha_func = state->alpha.func;
974
975 si_pm4_set_reg(pm4, R_00B030_SPI_SHADER_USER_DATA_PS_0 +
976 SI_SGPR_ALPHA_REF * 4, fui(state->alpha.ref_value));
977 } else {
978 dsa->alpha_func = PIPE_FUNC_ALWAYS;
979 }
980
981 si_pm4_set_reg(pm4, R_028800_DB_DEPTH_CONTROL, db_depth_control);
982 si_pm4_set_reg(pm4, R_02842C_DB_STENCIL_CONTROL, db_stencil_control);
983 if (state->depth.bounds_test) {
984 si_pm4_set_reg(pm4, R_028020_DB_DEPTH_BOUNDS_MIN, fui(state->depth.bounds_min));
985 si_pm4_set_reg(pm4, R_028024_DB_DEPTH_BOUNDS_MAX, fui(state->depth.bounds_max));
986 }
987
988 return dsa;
989 }
990
991 static void si_bind_dsa_state(struct pipe_context *ctx, void *state)
992 {
993 struct si_context *sctx = (struct si_context *)ctx;
994 struct si_state_dsa *dsa = state;
995
996 if (!state)
997 return;
998
999 si_pm4_bind_state(sctx, dsa, dsa);
1000
1001 if (memcmp(&dsa->stencil_ref, &sctx->stencil_ref.dsa_part,
1002 sizeof(struct si_dsa_stencil_ref_part)) != 0) {
1003 sctx->stencil_ref.dsa_part = dsa->stencil_ref;
1004 si_mark_atom_dirty(sctx, &sctx->stencil_ref.atom);
1005 }
1006 }
1007
1008 static void si_delete_dsa_state(struct pipe_context *ctx, void *state)
1009 {
1010 struct si_context *sctx = (struct si_context *)ctx;
1011 si_pm4_delete_state(sctx, dsa, (struct si_state_dsa *)state);
1012 }
1013
1014 static void *si_create_db_flush_dsa(struct si_context *sctx)
1015 {
1016 struct pipe_depth_stencil_alpha_state dsa = {};
1017
1018 return sctx->b.b.create_depth_stencil_alpha_state(&sctx->b.b, &dsa);
1019 }
1020
1021 /* DB RENDER STATE */
1022
1023 static void si_set_active_query_state(struct pipe_context *ctx, boolean enable)
1024 {
1025 struct si_context *sctx = (struct si_context*)ctx;
1026
1027 /* Pipeline stat & streamout queries. */
1028 if (enable) {
1029 sctx->b.flags &= ~R600_CONTEXT_STOP_PIPELINE_STATS;
1030 sctx->b.flags |= R600_CONTEXT_START_PIPELINE_STATS;
1031 } else {
1032 sctx->b.flags &= ~R600_CONTEXT_START_PIPELINE_STATS;
1033 sctx->b.flags |= R600_CONTEXT_STOP_PIPELINE_STATS;
1034 }
1035
1036 /* Occlusion queries. */
1037 if (sctx->occlusion_queries_disabled != !enable) {
1038 sctx->occlusion_queries_disabled = !enable;
1039 si_mark_atom_dirty(sctx, &sctx->db_render_state);
1040 }
1041 }
1042
1043 static void si_set_occlusion_query_state(struct pipe_context *ctx, bool enable)
1044 {
1045 struct si_context *sctx = (struct si_context*)ctx;
1046
1047 si_mark_atom_dirty(sctx, &sctx->db_render_state);
1048 }
1049
1050 static void si_emit_db_render_state(struct si_context *sctx, struct r600_atom *state)
1051 {
1052 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
1053 struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
1054 unsigned db_shader_control;
1055
1056 radeon_set_context_reg_seq(cs, R_028000_DB_RENDER_CONTROL, 2);
1057
1058 /* DB_RENDER_CONTROL */
1059 if (sctx->dbcb_depth_copy_enabled ||
1060 sctx->dbcb_stencil_copy_enabled) {
1061 radeon_emit(cs,
1062 S_028000_DEPTH_COPY(sctx->dbcb_depth_copy_enabled) |
1063 S_028000_STENCIL_COPY(sctx->dbcb_stencil_copy_enabled) |
1064 S_028000_COPY_CENTROID(1) |
1065 S_028000_COPY_SAMPLE(sctx->dbcb_copy_sample));
1066 } else if (sctx->db_flush_depth_inplace || sctx->db_flush_stencil_inplace) {
1067 radeon_emit(cs,
1068 S_028000_DEPTH_COMPRESS_DISABLE(sctx->db_flush_depth_inplace) |
1069 S_028000_STENCIL_COMPRESS_DISABLE(sctx->db_flush_stencil_inplace));
1070 } else {
1071 radeon_emit(cs,
1072 S_028000_DEPTH_CLEAR_ENABLE(sctx->db_depth_clear) |
1073 S_028000_STENCIL_CLEAR_ENABLE(sctx->db_stencil_clear));
1074 }
1075
1076 /* DB_COUNT_CONTROL (occlusion queries) */
1077 if (sctx->b.num_occlusion_queries > 0 &&
1078 !sctx->occlusion_queries_disabled) {
1079 bool perfect = sctx->b.num_perfect_occlusion_queries > 0;
1080
1081 if (sctx->b.chip_class >= CIK) {
1082 radeon_emit(cs,
1083 S_028004_PERFECT_ZPASS_COUNTS(perfect) |
1084 S_028004_SAMPLE_RATE(sctx->framebuffer.log_samples) |
1085 S_028004_ZPASS_ENABLE(1) |
1086 S_028004_SLICE_EVEN_ENABLE(1) |
1087 S_028004_SLICE_ODD_ENABLE(1));
1088 } else {
1089 radeon_emit(cs,
1090 S_028004_PERFECT_ZPASS_COUNTS(perfect) |
1091 S_028004_SAMPLE_RATE(sctx->framebuffer.log_samples));
1092 }
1093 } else {
1094 /* Disable occlusion queries. */
1095 if (sctx->b.chip_class >= CIK) {
1096 radeon_emit(cs, 0);
1097 } else {
1098 radeon_emit(cs, S_028004_ZPASS_INCREMENT_DISABLE(1));
1099 }
1100 }
1101
1102 /* DB_RENDER_OVERRIDE2 */
1103 radeon_set_context_reg(cs, R_028010_DB_RENDER_OVERRIDE2,
1104 S_028010_DISABLE_ZMASK_EXPCLEAR_OPTIMIZATION(sctx->db_depth_disable_expclear) |
1105 S_028010_DISABLE_SMEM_EXPCLEAR_OPTIMIZATION(sctx->db_stencil_disable_expclear) |
1106 S_028010_DECOMPRESS_Z_ON_FLUSH(sctx->framebuffer.nr_samples >= 4));
1107
1108 db_shader_control = S_02880C_ALPHA_TO_MASK_DISABLE(sctx->framebuffer.cb0_is_integer) |
1109 sctx->ps_db_shader_control;
1110
1111 /* Bug workaround for smoothing (overrasterization) on SI. */
1112 if (sctx->b.chip_class == SI && sctx->smoothing_enabled) {
1113 db_shader_control &= C_02880C_Z_ORDER;
1114 db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
1115 }
1116
1117 /* Disable the gl_SampleMask fragment shader output if MSAA is disabled. */
1118 if (sctx->framebuffer.nr_samples <= 1 || (rs && !rs->multisample_enable))
1119 db_shader_control &= C_02880C_MASK_EXPORT_ENABLE;
1120
1121 if (sctx->b.family == CHIP_STONEY &&
1122 sctx->screen->b.debug_flags & DBG_NO_RB_PLUS)
1123 db_shader_control |= S_02880C_DUAL_QUAD_DISABLE(1);
1124
1125 radeon_set_context_reg(cs, R_02880C_DB_SHADER_CONTROL,
1126 db_shader_control);
1127 }
1128
1129 /*
1130 * format translation
1131 */
1132 static uint32_t si_translate_colorformat(enum pipe_format format)
1133 {
1134 const struct util_format_description *desc = util_format_description(format);
1135
1136 #define HAS_SIZE(x,y,z,w) \
1137 (desc->channel[0].size == (x) && desc->channel[1].size == (y) && \
1138 desc->channel[2].size == (z) && desc->channel[3].size == (w))
1139
1140 if (format == PIPE_FORMAT_R11G11B10_FLOAT) /* isn't plain */
1141 return V_028C70_COLOR_10_11_11;
1142
1143 if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
1144 return V_028C70_COLOR_INVALID;
1145
1146 /* hw cannot support mixed formats (except depth/stencil, since
1147 * stencil is not written to). */
1148 if (desc->is_mixed && desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
1149 return V_028C70_COLOR_INVALID;
1150
1151 switch (desc->nr_channels) {
1152 case 1:
1153 switch (desc->channel[0].size) {
1154 case 8:
1155 return V_028C70_COLOR_8;
1156 case 16:
1157 return V_028C70_COLOR_16;
1158 case 32:
1159 return V_028C70_COLOR_32;
1160 }
1161 break;
1162 case 2:
1163 if (desc->channel[0].size == desc->channel[1].size) {
1164 switch (desc->channel[0].size) {
1165 case 8:
1166 return V_028C70_COLOR_8_8;
1167 case 16:
1168 return V_028C70_COLOR_16_16;
1169 case 32:
1170 return V_028C70_COLOR_32_32;
1171 }
1172 } else if (HAS_SIZE(8,24,0,0)) {
1173 return V_028C70_COLOR_24_8;
1174 } else if (HAS_SIZE(24,8,0,0)) {
1175 return V_028C70_COLOR_8_24;
1176 }
1177 break;
1178 case 3:
1179 if (HAS_SIZE(5,6,5,0)) {
1180 return V_028C70_COLOR_5_6_5;
1181 } else if (HAS_SIZE(32,8,24,0)) {
1182 return V_028C70_COLOR_X24_8_32_FLOAT;
1183 }
1184 break;
1185 case 4:
1186 if (desc->channel[0].size == desc->channel[1].size &&
1187 desc->channel[0].size == desc->channel[2].size &&
1188 desc->channel[0].size == desc->channel[3].size) {
1189 switch (desc->channel[0].size) {
1190 case 4:
1191 return V_028C70_COLOR_4_4_4_4;
1192 case 8:
1193 return V_028C70_COLOR_8_8_8_8;
1194 case 16:
1195 return V_028C70_COLOR_16_16_16_16;
1196 case 32:
1197 return V_028C70_COLOR_32_32_32_32;
1198 }
1199 } else if (HAS_SIZE(5,5,5,1)) {
1200 return V_028C70_COLOR_1_5_5_5;
1201 } else if (HAS_SIZE(10,10,10,2)) {
1202 return V_028C70_COLOR_2_10_10_10;
1203 }
1204 break;
1205 }
1206 return V_028C70_COLOR_INVALID;
1207 }
1208
1209 static uint32_t si_colorformat_endian_swap(uint32_t colorformat)
1210 {
1211 if (SI_BIG_ENDIAN) {
1212 switch(colorformat) {
1213 /* 8-bit buffers. */
1214 case V_028C70_COLOR_8:
1215 return V_028C70_ENDIAN_NONE;
1216
1217 /* 16-bit buffers. */
1218 case V_028C70_COLOR_5_6_5:
1219 case V_028C70_COLOR_1_5_5_5:
1220 case V_028C70_COLOR_4_4_4_4:
1221 case V_028C70_COLOR_16:
1222 case V_028C70_COLOR_8_8:
1223 return V_028C70_ENDIAN_8IN16;
1224
1225 /* 32-bit buffers. */
1226 case V_028C70_COLOR_8_8_8_8:
1227 case V_028C70_COLOR_2_10_10_10:
1228 case V_028C70_COLOR_8_24:
1229 case V_028C70_COLOR_24_8:
1230 case V_028C70_COLOR_16_16:
1231 return V_028C70_ENDIAN_8IN32;
1232
1233 /* 64-bit buffers. */
1234 case V_028C70_COLOR_16_16_16_16:
1235 return V_028C70_ENDIAN_8IN16;
1236
1237 case V_028C70_COLOR_32_32:
1238 return V_028C70_ENDIAN_8IN32;
1239
1240 /* 128-bit buffers. */
1241 case V_028C70_COLOR_32_32_32_32:
1242 return V_028C70_ENDIAN_8IN32;
1243 default:
1244 return V_028C70_ENDIAN_NONE; /* Unsupported. */
1245 }
1246 } else {
1247 return V_028C70_ENDIAN_NONE;
1248 }
1249 }
1250
1251 static uint32_t si_translate_dbformat(enum pipe_format format)
1252 {
1253 switch (format) {
1254 case PIPE_FORMAT_Z16_UNORM:
1255 return V_028040_Z_16;
1256 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1257 case PIPE_FORMAT_X8Z24_UNORM:
1258 case PIPE_FORMAT_Z24X8_UNORM:
1259 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1260 return V_028040_Z_24; /* deprecated on SI */
1261 case PIPE_FORMAT_Z32_FLOAT:
1262 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1263 return V_028040_Z_32_FLOAT;
1264 default:
1265 return V_028040_Z_INVALID;
1266 }
1267 }
1268
1269 /*
1270 * Texture translation
1271 */
1272
1273 static uint32_t si_translate_texformat(struct pipe_screen *screen,
1274 enum pipe_format format,
1275 const struct util_format_description *desc,
1276 int first_non_void)
1277 {
1278 struct si_screen *sscreen = (struct si_screen*)screen;
1279 bool enable_compressed_formats = (sscreen->b.info.drm_major == 2 &&
1280 sscreen->b.info.drm_minor >= 31) ||
1281 sscreen->b.info.drm_major == 3;
1282 boolean uniform = TRUE;
1283 int i;
1284
1285 /* Colorspace (return non-RGB formats directly). */
1286 switch (desc->colorspace) {
1287 /* Depth stencil formats */
1288 case UTIL_FORMAT_COLORSPACE_ZS:
1289 switch (format) {
1290 case PIPE_FORMAT_Z16_UNORM:
1291 return V_008F14_IMG_DATA_FORMAT_16;
1292 case PIPE_FORMAT_X24S8_UINT:
1293 case PIPE_FORMAT_Z24X8_UNORM:
1294 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1295 return V_008F14_IMG_DATA_FORMAT_8_24;
1296 case PIPE_FORMAT_X8Z24_UNORM:
1297 case PIPE_FORMAT_S8X24_UINT:
1298 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1299 return V_008F14_IMG_DATA_FORMAT_24_8;
1300 case PIPE_FORMAT_S8_UINT:
1301 return V_008F14_IMG_DATA_FORMAT_8;
1302 case PIPE_FORMAT_Z32_FLOAT:
1303 return V_008F14_IMG_DATA_FORMAT_32;
1304 case PIPE_FORMAT_X32_S8X24_UINT:
1305 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1306 return V_008F14_IMG_DATA_FORMAT_X24_8_32;
1307 default:
1308 goto out_unknown;
1309 }
1310
1311 case UTIL_FORMAT_COLORSPACE_YUV:
1312 goto out_unknown; /* TODO */
1313
1314 case UTIL_FORMAT_COLORSPACE_SRGB:
1315 if (desc->nr_channels != 4 && desc->nr_channels != 1)
1316 goto out_unknown;
1317 break;
1318
1319 default:
1320 break;
1321 }
1322
1323 if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
1324 if (!enable_compressed_formats)
1325 goto out_unknown;
1326
1327 switch (format) {
1328 case PIPE_FORMAT_RGTC1_SNORM:
1329 case PIPE_FORMAT_LATC1_SNORM:
1330 case PIPE_FORMAT_RGTC1_UNORM:
1331 case PIPE_FORMAT_LATC1_UNORM:
1332 return V_008F14_IMG_DATA_FORMAT_BC4;
1333 case PIPE_FORMAT_RGTC2_SNORM:
1334 case PIPE_FORMAT_LATC2_SNORM:
1335 case PIPE_FORMAT_RGTC2_UNORM:
1336 case PIPE_FORMAT_LATC2_UNORM:
1337 return V_008F14_IMG_DATA_FORMAT_BC5;
1338 default:
1339 goto out_unknown;
1340 }
1341 }
1342
1343 if (desc->layout == UTIL_FORMAT_LAYOUT_ETC &&
1344 sscreen->b.family == CHIP_STONEY) {
1345 switch (format) {
1346 case PIPE_FORMAT_ETC1_RGB8:
1347 case PIPE_FORMAT_ETC2_RGB8:
1348 case PIPE_FORMAT_ETC2_SRGB8:
1349 return V_008F14_IMG_DATA_FORMAT_ETC2_RGB;
1350 case PIPE_FORMAT_ETC2_RGB8A1:
1351 case PIPE_FORMAT_ETC2_SRGB8A1:
1352 return V_008F14_IMG_DATA_FORMAT_ETC2_RGBA1;
1353 case PIPE_FORMAT_ETC2_RGBA8:
1354 case PIPE_FORMAT_ETC2_SRGBA8:
1355 return V_008F14_IMG_DATA_FORMAT_ETC2_RGBA;
1356 case PIPE_FORMAT_ETC2_R11_UNORM:
1357 case PIPE_FORMAT_ETC2_R11_SNORM:
1358 return V_008F14_IMG_DATA_FORMAT_ETC2_R;
1359 case PIPE_FORMAT_ETC2_RG11_UNORM:
1360 case PIPE_FORMAT_ETC2_RG11_SNORM:
1361 return V_008F14_IMG_DATA_FORMAT_ETC2_RG;
1362 default:
1363 goto out_unknown;
1364 }
1365 }
1366
1367 if (desc->layout == UTIL_FORMAT_LAYOUT_BPTC) {
1368 if (!enable_compressed_formats)
1369 goto out_unknown;
1370
1371 switch (format) {
1372 case PIPE_FORMAT_BPTC_RGBA_UNORM:
1373 case PIPE_FORMAT_BPTC_SRGBA:
1374 return V_008F14_IMG_DATA_FORMAT_BC7;
1375 case PIPE_FORMAT_BPTC_RGB_FLOAT:
1376 case PIPE_FORMAT_BPTC_RGB_UFLOAT:
1377 return V_008F14_IMG_DATA_FORMAT_BC6;
1378 default:
1379 goto out_unknown;
1380 }
1381 }
1382
1383 if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
1384 switch (format) {
1385 case PIPE_FORMAT_R8G8_B8G8_UNORM:
1386 case PIPE_FORMAT_G8R8_B8R8_UNORM:
1387 return V_008F14_IMG_DATA_FORMAT_GB_GR;
1388 case PIPE_FORMAT_G8R8_G8B8_UNORM:
1389 case PIPE_FORMAT_R8G8_R8B8_UNORM:
1390 return V_008F14_IMG_DATA_FORMAT_BG_RG;
1391 default:
1392 goto out_unknown;
1393 }
1394 }
1395
1396 if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
1397 if (!enable_compressed_formats)
1398 goto out_unknown;
1399
1400 if (!util_format_s3tc_enabled) {
1401 goto out_unknown;
1402 }
1403
1404 switch (format) {
1405 case PIPE_FORMAT_DXT1_RGB:
1406 case PIPE_FORMAT_DXT1_RGBA:
1407 case PIPE_FORMAT_DXT1_SRGB:
1408 case PIPE_FORMAT_DXT1_SRGBA:
1409 return V_008F14_IMG_DATA_FORMAT_BC1;
1410 case PIPE_FORMAT_DXT3_RGBA:
1411 case PIPE_FORMAT_DXT3_SRGBA:
1412 return V_008F14_IMG_DATA_FORMAT_BC2;
1413 case PIPE_FORMAT_DXT5_RGBA:
1414 case PIPE_FORMAT_DXT5_SRGBA:
1415 return V_008F14_IMG_DATA_FORMAT_BC3;
1416 default:
1417 goto out_unknown;
1418 }
1419 }
1420
1421 if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
1422 return V_008F14_IMG_DATA_FORMAT_5_9_9_9;
1423 } else if (format == PIPE_FORMAT_R11G11B10_FLOAT) {
1424 return V_008F14_IMG_DATA_FORMAT_10_11_11;
1425 }
1426
1427 /* R8G8Bx_SNORM - TODO CxV8U8 */
1428
1429 /* hw cannot support mixed formats (except depth/stencil, since only
1430 * depth is read).*/
1431 if (desc->is_mixed && desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
1432 goto out_unknown;
1433
1434 /* See whether the components are of the same size. */
1435 for (i = 1; i < desc->nr_channels; i++) {
1436 uniform = uniform && desc->channel[0].size == desc->channel[i].size;
1437 }
1438
1439 /* Non-uniform formats. */
1440 if (!uniform) {
1441 switch(desc->nr_channels) {
1442 case 3:
1443 if (desc->channel[0].size == 5 &&
1444 desc->channel[1].size == 6 &&
1445 desc->channel[2].size == 5) {
1446 return V_008F14_IMG_DATA_FORMAT_5_6_5;
1447 }
1448 goto out_unknown;
1449 case 4:
1450 if (desc->channel[0].size == 5 &&
1451 desc->channel[1].size == 5 &&
1452 desc->channel[2].size == 5 &&
1453 desc->channel[3].size == 1) {
1454 return V_008F14_IMG_DATA_FORMAT_1_5_5_5;
1455 }
1456 if (desc->channel[0].size == 10 &&
1457 desc->channel[1].size == 10 &&
1458 desc->channel[2].size == 10 &&
1459 desc->channel[3].size == 2) {
1460 return V_008F14_IMG_DATA_FORMAT_2_10_10_10;
1461 }
1462 goto out_unknown;
1463 }
1464 goto out_unknown;
1465 }
1466
1467 if (first_non_void < 0 || first_non_void > 3)
1468 goto out_unknown;
1469
1470 /* uniform formats */
1471 switch (desc->channel[first_non_void].size) {
1472 case 4:
1473 switch (desc->nr_channels) {
1474 #if 0 /* Not supported for render targets */
1475 case 2:
1476 return V_008F14_IMG_DATA_FORMAT_4_4;
1477 #endif
1478 case 4:
1479 return V_008F14_IMG_DATA_FORMAT_4_4_4_4;
1480 }
1481 break;
1482 case 8:
1483 switch (desc->nr_channels) {
1484 case 1:
1485 return V_008F14_IMG_DATA_FORMAT_8;
1486 case 2:
1487 return V_008F14_IMG_DATA_FORMAT_8_8;
1488 case 4:
1489 return V_008F14_IMG_DATA_FORMAT_8_8_8_8;
1490 }
1491 break;
1492 case 16:
1493 switch (desc->nr_channels) {
1494 case 1:
1495 return V_008F14_IMG_DATA_FORMAT_16;
1496 case 2:
1497 return V_008F14_IMG_DATA_FORMAT_16_16;
1498 case 4:
1499 return V_008F14_IMG_DATA_FORMAT_16_16_16_16;
1500 }
1501 break;
1502 case 32:
1503 switch (desc->nr_channels) {
1504 case 1:
1505 return V_008F14_IMG_DATA_FORMAT_32;
1506 case 2:
1507 return V_008F14_IMG_DATA_FORMAT_32_32;
1508 #if 0 /* Not supported for render targets */
1509 case 3:
1510 return V_008F14_IMG_DATA_FORMAT_32_32_32;
1511 #endif
1512 case 4:
1513 return V_008F14_IMG_DATA_FORMAT_32_32_32_32;
1514 }
1515 }
1516
1517 out_unknown:
1518 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1519 return ~0;
1520 }
1521
1522 static unsigned si_tex_wrap(unsigned wrap)
1523 {
1524 switch (wrap) {
1525 default:
1526 case PIPE_TEX_WRAP_REPEAT:
1527 return V_008F30_SQ_TEX_WRAP;
1528 case PIPE_TEX_WRAP_CLAMP:
1529 return V_008F30_SQ_TEX_CLAMP_HALF_BORDER;
1530 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1531 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL;
1532 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1533 return V_008F30_SQ_TEX_CLAMP_BORDER;
1534 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1535 return V_008F30_SQ_TEX_MIRROR;
1536 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1537 return V_008F30_SQ_TEX_MIRROR_ONCE_HALF_BORDER;
1538 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1539 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL;
1540 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1541 return V_008F30_SQ_TEX_MIRROR_ONCE_BORDER;
1542 }
1543 }
1544
1545 static unsigned si_tex_mipfilter(unsigned filter)
1546 {
1547 switch (filter) {
1548 case PIPE_TEX_MIPFILTER_NEAREST:
1549 return V_008F38_SQ_TEX_Z_FILTER_POINT;
1550 case PIPE_TEX_MIPFILTER_LINEAR:
1551 return V_008F38_SQ_TEX_Z_FILTER_LINEAR;
1552 default:
1553 case PIPE_TEX_MIPFILTER_NONE:
1554 return V_008F38_SQ_TEX_Z_FILTER_NONE;
1555 }
1556 }
1557
1558 static unsigned si_tex_compare(unsigned compare)
1559 {
1560 switch (compare) {
1561 default:
1562 case PIPE_FUNC_NEVER:
1563 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER;
1564 case PIPE_FUNC_LESS:
1565 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS;
1566 case PIPE_FUNC_EQUAL:
1567 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL;
1568 case PIPE_FUNC_LEQUAL:
1569 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL;
1570 case PIPE_FUNC_GREATER:
1571 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER;
1572 case PIPE_FUNC_NOTEQUAL:
1573 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL;
1574 case PIPE_FUNC_GEQUAL:
1575 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL;
1576 case PIPE_FUNC_ALWAYS:
1577 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS;
1578 }
1579 }
1580
1581 static unsigned si_tex_dim(unsigned res_target, unsigned view_target,
1582 unsigned nr_samples)
1583 {
1584 if (view_target == PIPE_TEXTURE_CUBE ||
1585 view_target == PIPE_TEXTURE_CUBE_ARRAY)
1586 res_target = view_target;
1587
1588 switch (res_target) {
1589 default:
1590 case PIPE_TEXTURE_1D:
1591 return V_008F1C_SQ_RSRC_IMG_1D;
1592 case PIPE_TEXTURE_1D_ARRAY:
1593 return V_008F1C_SQ_RSRC_IMG_1D_ARRAY;
1594 case PIPE_TEXTURE_2D:
1595 case PIPE_TEXTURE_RECT:
1596 return nr_samples > 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA :
1597 V_008F1C_SQ_RSRC_IMG_2D;
1598 case PIPE_TEXTURE_2D_ARRAY:
1599 return nr_samples > 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY :
1600 V_008F1C_SQ_RSRC_IMG_2D_ARRAY;
1601 case PIPE_TEXTURE_3D:
1602 return V_008F1C_SQ_RSRC_IMG_3D;
1603 case PIPE_TEXTURE_CUBE:
1604 case PIPE_TEXTURE_CUBE_ARRAY:
1605 return V_008F1C_SQ_RSRC_IMG_CUBE;
1606 }
1607 }
1608
1609 /*
1610 * Format support testing
1611 */
1612
1613 static bool si_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
1614 {
1615 return si_translate_texformat(screen, format, util_format_description(format),
1616 util_format_get_first_non_void_channel(format)) != ~0U;
1617 }
1618
1619 static uint32_t si_translate_buffer_dataformat(struct pipe_screen *screen,
1620 const struct util_format_description *desc,
1621 int first_non_void)
1622 {
1623 unsigned type;
1624 int i;
1625
1626 if (desc->format == PIPE_FORMAT_R11G11B10_FLOAT)
1627 return V_008F0C_BUF_DATA_FORMAT_10_11_11;
1628
1629 assert(first_non_void >= 0);
1630 type = desc->channel[first_non_void].type;
1631
1632 if (type == UTIL_FORMAT_TYPE_FIXED)
1633 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1634
1635 if (desc->nr_channels == 4 &&
1636 desc->channel[0].size == 10 &&
1637 desc->channel[1].size == 10 &&
1638 desc->channel[2].size == 10 &&
1639 desc->channel[3].size == 2)
1640 return V_008F0C_BUF_DATA_FORMAT_2_10_10_10;
1641
1642 /* See whether the components are of the same size. */
1643 for (i = 0; i < desc->nr_channels; i++) {
1644 if (desc->channel[first_non_void].size != desc->channel[i].size)
1645 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1646 }
1647
1648 switch (desc->channel[first_non_void].size) {
1649 case 8:
1650 switch (desc->nr_channels) {
1651 case 1:
1652 return V_008F0C_BUF_DATA_FORMAT_8;
1653 case 2:
1654 return V_008F0C_BUF_DATA_FORMAT_8_8;
1655 case 3:
1656 case 4:
1657 return V_008F0C_BUF_DATA_FORMAT_8_8_8_8;
1658 }
1659 break;
1660 case 16:
1661 switch (desc->nr_channels) {
1662 case 1:
1663 return V_008F0C_BUF_DATA_FORMAT_16;
1664 case 2:
1665 return V_008F0C_BUF_DATA_FORMAT_16_16;
1666 case 3:
1667 case 4:
1668 return V_008F0C_BUF_DATA_FORMAT_16_16_16_16;
1669 }
1670 break;
1671 case 32:
1672 /* From the Southern Islands ISA documentation about MTBUF:
1673 * 'Memory reads of data in memory that is 32 or 64 bits do not
1674 * undergo any format conversion.'
1675 */
1676 if (type != UTIL_FORMAT_TYPE_FLOAT &&
1677 !desc->channel[first_non_void].pure_integer)
1678 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1679
1680 switch (desc->nr_channels) {
1681 case 1:
1682 return V_008F0C_BUF_DATA_FORMAT_32;
1683 case 2:
1684 return V_008F0C_BUF_DATA_FORMAT_32_32;
1685 case 3:
1686 return V_008F0C_BUF_DATA_FORMAT_32_32_32;
1687 case 4:
1688 return V_008F0C_BUF_DATA_FORMAT_32_32_32_32;
1689 }
1690 break;
1691 }
1692
1693 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1694 }
1695
1696 static uint32_t si_translate_buffer_numformat(struct pipe_screen *screen,
1697 const struct util_format_description *desc,
1698 int first_non_void)
1699 {
1700 if (desc->format == PIPE_FORMAT_R11G11B10_FLOAT)
1701 return V_008F0C_BUF_NUM_FORMAT_FLOAT;
1702
1703 assert(first_non_void >= 0);
1704
1705 switch (desc->channel[first_non_void].type) {
1706 case UTIL_FORMAT_TYPE_SIGNED:
1707 if (desc->channel[first_non_void].normalized)
1708 return V_008F0C_BUF_NUM_FORMAT_SNORM;
1709 else if (desc->channel[first_non_void].pure_integer)
1710 return V_008F0C_BUF_NUM_FORMAT_SINT;
1711 else
1712 return V_008F0C_BUF_NUM_FORMAT_SSCALED;
1713 break;
1714 case UTIL_FORMAT_TYPE_UNSIGNED:
1715 if (desc->channel[first_non_void].normalized)
1716 return V_008F0C_BUF_NUM_FORMAT_UNORM;
1717 else if (desc->channel[first_non_void].pure_integer)
1718 return V_008F0C_BUF_NUM_FORMAT_UINT;
1719 else
1720 return V_008F0C_BUF_NUM_FORMAT_USCALED;
1721 break;
1722 case UTIL_FORMAT_TYPE_FLOAT:
1723 default:
1724 return V_008F0C_BUF_NUM_FORMAT_FLOAT;
1725 }
1726 }
1727
1728 static bool si_is_vertex_format_supported(struct pipe_screen *screen, enum pipe_format format)
1729 {
1730 const struct util_format_description *desc;
1731 int first_non_void;
1732 unsigned data_format;
1733
1734 desc = util_format_description(format);
1735 first_non_void = util_format_get_first_non_void_channel(format);
1736 data_format = si_translate_buffer_dataformat(screen, desc, first_non_void);
1737 return data_format != V_008F0C_BUF_DATA_FORMAT_INVALID;
1738 }
1739
1740 static bool si_is_colorbuffer_format_supported(enum pipe_format format)
1741 {
1742 return si_translate_colorformat(format) != V_028C70_COLOR_INVALID &&
1743 r600_translate_colorswap(format, FALSE) != ~0U;
1744 }
1745
1746 static bool si_is_zs_format_supported(enum pipe_format format)
1747 {
1748 return si_translate_dbformat(format) != V_028040_Z_INVALID;
1749 }
1750
1751 boolean si_is_format_supported(struct pipe_screen *screen,
1752 enum pipe_format format,
1753 enum pipe_texture_target target,
1754 unsigned sample_count,
1755 unsigned usage)
1756 {
1757 unsigned retval = 0;
1758
1759 if (target >= PIPE_MAX_TEXTURE_TYPES) {
1760 R600_ERR("r600: unsupported texture type %d\n", target);
1761 return FALSE;
1762 }
1763
1764 if (!util_format_is_supported(format, usage))
1765 return FALSE;
1766
1767 if (sample_count > 1) {
1768 if (!screen->get_param(screen, PIPE_CAP_TEXTURE_MULTISAMPLE))
1769 return FALSE;
1770
1771 switch (sample_count) {
1772 case 2:
1773 case 4:
1774 case 8:
1775 break;
1776 case 16:
1777 if (format == PIPE_FORMAT_NONE)
1778 return TRUE;
1779 else
1780 return FALSE;
1781 default:
1782 return FALSE;
1783 }
1784 }
1785
1786 if (usage & (PIPE_BIND_SAMPLER_VIEW |
1787 PIPE_BIND_SHADER_IMAGE)) {
1788 if (target == PIPE_BUFFER) {
1789 if (si_is_vertex_format_supported(screen, format))
1790 retval |= usage & (PIPE_BIND_SAMPLER_VIEW |
1791 PIPE_BIND_SHADER_IMAGE);
1792 } else {
1793 if (si_is_sampler_format_supported(screen, format))
1794 retval |= usage & (PIPE_BIND_SAMPLER_VIEW |
1795 PIPE_BIND_SHADER_IMAGE);
1796 }
1797 }
1798
1799 if ((usage & (PIPE_BIND_RENDER_TARGET |
1800 PIPE_BIND_DISPLAY_TARGET |
1801 PIPE_BIND_SCANOUT |
1802 PIPE_BIND_SHARED |
1803 PIPE_BIND_BLENDABLE)) &&
1804 si_is_colorbuffer_format_supported(format)) {
1805 retval |= usage &
1806 (PIPE_BIND_RENDER_TARGET |
1807 PIPE_BIND_DISPLAY_TARGET |
1808 PIPE_BIND_SCANOUT |
1809 PIPE_BIND_SHARED);
1810 if (!util_format_is_pure_integer(format) &&
1811 !util_format_is_depth_or_stencil(format))
1812 retval |= usage & PIPE_BIND_BLENDABLE;
1813 }
1814
1815 if ((usage & PIPE_BIND_DEPTH_STENCIL) &&
1816 si_is_zs_format_supported(format)) {
1817 retval |= PIPE_BIND_DEPTH_STENCIL;
1818 }
1819
1820 if ((usage & PIPE_BIND_VERTEX_BUFFER) &&
1821 si_is_vertex_format_supported(screen, format)) {
1822 retval |= PIPE_BIND_VERTEX_BUFFER;
1823 }
1824
1825 if (usage & PIPE_BIND_TRANSFER_READ)
1826 retval |= PIPE_BIND_TRANSFER_READ;
1827 if (usage & PIPE_BIND_TRANSFER_WRITE)
1828 retval |= PIPE_BIND_TRANSFER_WRITE;
1829
1830 if ((usage & PIPE_BIND_LINEAR) &&
1831 !util_format_is_compressed(format) &&
1832 !(usage & PIPE_BIND_DEPTH_STENCIL))
1833 retval |= PIPE_BIND_LINEAR;
1834
1835 return retval == usage;
1836 }
1837
1838 /*
1839 * framebuffer handling
1840 */
1841
1842 static void si_choose_spi_color_formats(struct r600_surface *surf,
1843 unsigned format, unsigned swap,
1844 unsigned ntype, bool is_depth)
1845 {
1846 /* Alpha is needed for alpha-to-coverage.
1847 * Blending may be with or without alpha.
1848 */
1849 unsigned normal = 0; /* most optimal, may not support blending or export alpha */
1850 unsigned alpha = 0; /* exports alpha, but may not support blending */
1851 unsigned blend = 0; /* supports blending, but may not export alpha */
1852 unsigned blend_alpha = 0; /* least optimal, supports blending and exports alpha */
1853
1854 /* Choose the SPI color formats. These are required values for Stoney/RB+.
1855 * Other chips have multiple choices, though they are not necessarily better.
1856 */
1857 switch (format) {
1858 case V_028C70_COLOR_5_6_5:
1859 case V_028C70_COLOR_1_5_5_5:
1860 case V_028C70_COLOR_5_5_5_1:
1861 case V_028C70_COLOR_4_4_4_4:
1862 case V_028C70_COLOR_10_11_11:
1863 case V_028C70_COLOR_11_11_10:
1864 case V_028C70_COLOR_8:
1865 case V_028C70_COLOR_8_8:
1866 case V_028C70_COLOR_8_8_8_8:
1867 case V_028C70_COLOR_10_10_10_2:
1868 case V_028C70_COLOR_2_10_10_10:
1869 if (ntype == V_028C70_NUMBER_UINT)
1870 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_UINT16_ABGR;
1871 else if (ntype == V_028C70_NUMBER_SINT)
1872 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_SINT16_ABGR;
1873 else
1874 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_FP16_ABGR;
1875 break;
1876
1877 case V_028C70_COLOR_16:
1878 case V_028C70_COLOR_16_16:
1879 case V_028C70_COLOR_16_16_16_16:
1880 if (ntype == V_028C70_NUMBER_UNORM ||
1881 ntype == V_028C70_NUMBER_SNORM) {
1882 /* UNORM16 and SNORM16 don't support blending */
1883 if (ntype == V_028C70_NUMBER_UNORM)
1884 normal = alpha = V_028714_SPI_SHADER_UNORM16_ABGR;
1885 else
1886 normal = alpha = V_028714_SPI_SHADER_SNORM16_ABGR;
1887
1888 /* Use 32 bits per channel for blending. */
1889 if (format == V_028C70_COLOR_16) {
1890 if (swap == V_028C70_SWAP_STD) { /* R */
1891 blend = V_028714_SPI_SHADER_32_R;
1892 blend_alpha = V_028714_SPI_SHADER_32_AR;
1893 } else if (swap == V_028C70_SWAP_ALT_REV) /* A */
1894 blend = blend_alpha = V_028714_SPI_SHADER_32_AR;
1895 else
1896 assert(0);
1897 } else if (format == V_028C70_COLOR_16_16) {
1898 if (swap == V_028C70_SWAP_STD) { /* RG */
1899 blend = V_028714_SPI_SHADER_32_GR;
1900 blend_alpha = V_028714_SPI_SHADER_32_ABGR;
1901 } else if (swap == V_028C70_SWAP_ALT) /* RA */
1902 blend = blend_alpha = V_028714_SPI_SHADER_32_AR;
1903 else
1904 assert(0);
1905 } else /* 16_16_16_16 */
1906 blend = blend_alpha = V_028714_SPI_SHADER_32_ABGR;
1907 } else if (ntype == V_028C70_NUMBER_UINT)
1908 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_UINT16_ABGR;
1909 else if (ntype == V_028C70_NUMBER_SINT)
1910 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_SINT16_ABGR;
1911 else if (ntype == V_028C70_NUMBER_FLOAT)
1912 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_FP16_ABGR;
1913 else
1914 assert(0);
1915 break;
1916
1917 case V_028C70_COLOR_32:
1918 if (swap == V_028C70_SWAP_STD) { /* R */
1919 blend = normal = V_028714_SPI_SHADER_32_R;
1920 alpha = blend_alpha = V_028714_SPI_SHADER_32_AR;
1921 } else if (swap == V_028C70_SWAP_ALT_REV) /* A */
1922 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_AR;
1923 else
1924 assert(0);
1925 break;
1926
1927 case V_028C70_COLOR_32_32:
1928 if (swap == V_028C70_SWAP_STD) { /* RG */
1929 blend = normal = V_028714_SPI_SHADER_32_GR;
1930 alpha = blend_alpha = V_028714_SPI_SHADER_32_ABGR;
1931 } else if (swap == V_028C70_SWAP_ALT) /* RA */
1932 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_AR;
1933 else
1934 assert(0);
1935 break;
1936
1937 case V_028C70_COLOR_32_32_32_32:
1938 case V_028C70_COLOR_8_24:
1939 case V_028C70_COLOR_24_8:
1940 case V_028C70_COLOR_X24_8_32_FLOAT:
1941 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_ABGR;
1942 break;
1943
1944 default:
1945 assert(0);
1946 return;
1947 }
1948
1949 /* The DB->CB copy needs 32_ABGR. */
1950 if (is_depth)
1951 alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_ABGR;
1952
1953 surf->spi_shader_col_format = normal;
1954 surf->spi_shader_col_format_alpha = alpha;
1955 surf->spi_shader_col_format_blend = blend;
1956 surf->spi_shader_col_format_blend_alpha = blend_alpha;
1957 }
1958
1959 static void si_initialize_color_surface(struct si_context *sctx,
1960 struct r600_surface *surf)
1961 {
1962 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1963 unsigned color_info, color_attrib, color_view;
1964 unsigned format, swap, ntype, endian;
1965 const struct util_format_description *desc;
1966 int i;
1967 unsigned blend_clamp = 0, blend_bypass = 0;
1968
1969 color_view = S_028C6C_SLICE_START(surf->base.u.tex.first_layer) |
1970 S_028C6C_SLICE_MAX(surf->base.u.tex.last_layer);
1971
1972 desc = util_format_description(surf->base.format);
1973 for (i = 0; i < 4; i++) {
1974 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
1975 break;
1976 }
1977 }
1978 if (i == 4 || desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT) {
1979 ntype = V_028C70_NUMBER_FLOAT;
1980 } else {
1981 ntype = V_028C70_NUMBER_UNORM;
1982 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
1983 ntype = V_028C70_NUMBER_SRGB;
1984 else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
1985 if (desc->channel[i].pure_integer) {
1986 ntype = V_028C70_NUMBER_SINT;
1987 } else {
1988 assert(desc->channel[i].normalized);
1989 ntype = V_028C70_NUMBER_SNORM;
1990 }
1991 } else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
1992 if (desc->channel[i].pure_integer) {
1993 ntype = V_028C70_NUMBER_UINT;
1994 } else {
1995 assert(desc->channel[i].normalized);
1996 ntype = V_028C70_NUMBER_UNORM;
1997 }
1998 }
1999 }
2000
2001 format = si_translate_colorformat(surf->base.format);
2002 if (format == V_028C70_COLOR_INVALID) {
2003 R600_ERR("Invalid CB format: %d, disabling CB.\n", surf->base.format);
2004 }
2005 assert(format != V_028C70_COLOR_INVALID);
2006 swap = r600_translate_colorswap(surf->base.format, FALSE);
2007 endian = si_colorformat_endian_swap(format);
2008
2009 /* blend clamp should be set for all NORM/SRGB types */
2010 if (ntype == V_028C70_NUMBER_UNORM ||
2011 ntype == V_028C70_NUMBER_SNORM ||
2012 ntype == V_028C70_NUMBER_SRGB)
2013 blend_clamp = 1;
2014
2015 /* set blend bypass according to docs if SINT/UINT or
2016 8/24 COLOR variants */
2017 if (ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT ||
2018 format == V_028C70_COLOR_8_24 || format == V_028C70_COLOR_24_8 ||
2019 format == V_028C70_COLOR_X24_8_32_FLOAT) {
2020 blend_clamp = 0;
2021 blend_bypass = 1;
2022 }
2023
2024 if ((ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT) &&
2025 (format == V_028C70_COLOR_8 ||
2026 format == V_028C70_COLOR_8_8 ||
2027 format == V_028C70_COLOR_8_8_8_8))
2028 surf->color_is_int8 = true;
2029
2030 color_info = S_028C70_FORMAT(format) |
2031 S_028C70_COMP_SWAP(swap) |
2032 S_028C70_BLEND_CLAMP(blend_clamp) |
2033 S_028C70_BLEND_BYPASS(blend_bypass) |
2034 S_028C70_NUMBER_TYPE(ntype) |
2035 S_028C70_ENDIAN(endian);
2036
2037 /* Intensity is implemented as Red, so treat it that way. */
2038 color_attrib = S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == PIPE_SWIZZLE_1 ||
2039 util_format_is_intensity(surf->base.format));
2040
2041 if (rtex->resource.b.b.nr_samples > 1) {
2042 unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples);
2043
2044 color_attrib |= S_028C74_NUM_SAMPLES(log_samples) |
2045 S_028C74_NUM_FRAGMENTS(log_samples);
2046
2047 if (rtex->fmask.size) {
2048 color_info |= S_028C70_COMPRESSION(1);
2049 unsigned fmask_bankh = util_logbase2(rtex->fmask.bank_height);
2050
2051 if (sctx->b.chip_class == SI) {
2052 /* due to a hw bug, FMASK_BANK_HEIGHT must be set on SI too */
2053 color_attrib |= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
2054 }
2055 }
2056 }
2057
2058 surf->cb_color_view = color_view;
2059 surf->cb_color_info = color_info;
2060 surf->cb_color_attrib = color_attrib;
2061
2062 if (sctx->b.chip_class >= VI) {
2063 unsigned max_uncompressed_block_size = 2;
2064
2065 if (rtex->surface.nsamples > 1) {
2066 if (rtex->surface.bpe == 1)
2067 max_uncompressed_block_size = 0;
2068 else if (rtex->surface.bpe == 2)
2069 max_uncompressed_block_size = 1;
2070 }
2071
2072 surf->cb_dcc_control = S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size) |
2073 S_028C78_INDEPENDENT_64B_BLOCKS(1);
2074 }
2075
2076 /* This must be set for fast clear to work without FMASK. */
2077 if (!rtex->fmask.size && sctx->b.chip_class == SI) {
2078 unsigned bankh = util_logbase2(rtex->surface.bankh);
2079 surf->cb_color_attrib |= S_028C74_FMASK_BANK_HEIGHT(bankh);
2080 }
2081
2082 /* Determine pixel shader export format */
2083 si_choose_spi_color_formats(surf, format, swap, ntype, rtex->is_depth);
2084
2085 surf->color_initialized = true;
2086 }
2087
2088 static void si_init_depth_surface(struct si_context *sctx,
2089 struct r600_surface *surf)
2090 {
2091 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
2092 unsigned level = surf->base.u.tex.level;
2093 struct radeon_surf_level *levelinfo = &rtex->surface.level[level];
2094 unsigned format;
2095 uint32_t z_info, s_info, db_depth_info;
2096 uint64_t z_offs, s_offs;
2097 uint32_t db_htile_data_base, db_htile_surface, pa_su_poly_offset_db_fmt_cntl = 0;
2098
2099 switch (sctx->framebuffer.state.zsbuf->texture->format) {
2100 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2101 case PIPE_FORMAT_X8Z24_UNORM:
2102 case PIPE_FORMAT_Z24X8_UNORM:
2103 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
2104 pa_su_poly_offset_db_fmt_cntl = S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
2105 break;
2106 case PIPE_FORMAT_Z32_FLOAT:
2107 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
2108 pa_su_poly_offset_db_fmt_cntl = S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
2109 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
2110 break;
2111 case PIPE_FORMAT_Z16_UNORM:
2112 pa_su_poly_offset_db_fmt_cntl = S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
2113 break;
2114 default:
2115 assert(0);
2116 }
2117
2118 format = si_translate_dbformat(rtex->resource.b.b.format);
2119
2120 if (format == V_028040_Z_INVALID) {
2121 R600_ERR("Invalid DB format: %d, disabling DB.\n", rtex->resource.b.b.format);
2122 }
2123 assert(format != V_028040_Z_INVALID);
2124
2125 s_offs = z_offs = rtex->resource.gpu_address;
2126 z_offs += rtex->surface.level[level].offset;
2127 s_offs += rtex->surface.stencil_level[level].offset;
2128
2129 db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
2130
2131 z_info = S_028040_FORMAT(format);
2132 if (rtex->resource.b.b.nr_samples > 1) {
2133 z_info |= S_028040_NUM_SAMPLES(util_logbase2(rtex->resource.b.b.nr_samples));
2134 }
2135
2136 if (rtex->surface.flags & RADEON_SURF_SBUFFER)
2137 s_info = S_028044_FORMAT(V_028044_STENCIL_8);
2138 else
2139 s_info = S_028044_FORMAT(V_028044_STENCIL_INVALID);
2140
2141 if (sctx->b.chip_class >= CIK) {
2142 struct radeon_info *info = &sctx->screen->b.info;
2143 unsigned index = rtex->surface.tiling_index[level];
2144 unsigned stencil_index = rtex->surface.stencil_tiling_index[level];
2145 unsigned macro_index = rtex->surface.macro_tile_index;
2146 unsigned tile_mode = info->si_tile_mode_array[index];
2147 unsigned stencil_tile_mode = info->si_tile_mode_array[stencil_index];
2148 unsigned macro_mode = info->cik_macrotile_mode_array[macro_index];
2149
2150 db_depth_info |=
2151 S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode)) |
2152 S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode)) |
2153 S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode)) |
2154 S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode)) |
2155 S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode)) |
2156 S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode));
2157 z_info |= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode));
2158 s_info |= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode));
2159 } else {
2160 unsigned tile_mode_index = si_tile_mode_index(rtex, level, false);
2161 z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
2162 tile_mode_index = si_tile_mode_index(rtex, level, true);
2163 s_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
2164 }
2165
2166 /* HiZ aka depth buffer htile */
2167 /* use htile only for first level */
2168 if (rtex->htile_buffer && !level) {
2169 z_info |= S_028040_TILE_SURFACE_ENABLE(1) |
2170 S_028040_ALLOW_EXPCLEAR(1);
2171
2172 if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
2173 /* Workaround: For a not yet understood reason, the
2174 * combination of MSAA, fast stencil clear and stencil
2175 * decompress messes with subsequent stencil buffer
2176 * uses. Problem was reproduced on Verde, Bonaire,
2177 * Tonga, and Carrizo.
2178 *
2179 * Disabling EXPCLEAR works around the problem.
2180 *
2181 * Check piglit's arb_texture_multisample-stencil-clear
2182 * test if you want to try changing this.
2183 */
2184 if (rtex->resource.b.b.nr_samples <= 1)
2185 s_info |= S_028044_ALLOW_EXPCLEAR(1);
2186 } else
2187 /* Use all of the htile_buffer for depth if there's no stencil. */
2188 s_info |= S_028044_TILE_STENCIL_DISABLE(1);
2189
2190 uint64_t va = rtex->htile_buffer->gpu_address;
2191 db_htile_data_base = va >> 8;
2192 db_htile_surface = S_028ABC_FULL_CACHE(1);
2193 } else {
2194 db_htile_data_base = 0;
2195 db_htile_surface = 0;
2196 }
2197
2198 assert(levelinfo->nblk_x % 8 == 0 && levelinfo->nblk_y % 8 == 0);
2199
2200 surf->db_depth_view = S_028008_SLICE_START(surf->base.u.tex.first_layer) |
2201 S_028008_SLICE_MAX(surf->base.u.tex.last_layer);
2202 surf->db_htile_data_base = db_htile_data_base;
2203 surf->db_depth_info = db_depth_info;
2204 surf->db_z_info = z_info;
2205 surf->db_stencil_info = s_info;
2206 surf->db_depth_base = z_offs >> 8;
2207 surf->db_stencil_base = s_offs >> 8;
2208 surf->db_depth_size = S_028058_PITCH_TILE_MAX((levelinfo->nblk_x / 8) - 1) |
2209 S_028058_HEIGHT_TILE_MAX((levelinfo->nblk_y / 8) - 1);
2210 surf->db_depth_slice = S_02805C_SLICE_TILE_MAX((levelinfo->nblk_x *
2211 levelinfo->nblk_y) / 64 - 1);
2212 surf->db_htile_surface = db_htile_surface;
2213 surf->pa_su_poly_offset_db_fmt_cntl = pa_su_poly_offset_db_fmt_cntl;
2214
2215 surf->depth_initialized = true;
2216 }
2217
2218 void si_dec_framebuffer_counters(const struct pipe_framebuffer_state *state)
2219 {
2220 for (int i = 0; i < state->nr_cbufs; ++i) {
2221 struct r600_surface *surf = NULL;
2222 struct r600_texture *rtex;
2223
2224 if (!state->cbufs[i])
2225 continue;
2226 surf = (struct r600_surface*)state->cbufs[i];
2227 rtex = (struct r600_texture*)surf->base.texture;
2228
2229 p_atomic_dec(&rtex->framebuffers_bound);
2230 }
2231 }
2232
2233 static void si_set_framebuffer_state(struct pipe_context *ctx,
2234 const struct pipe_framebuffer_state *state)
2235 {
2236 struct si_context *sctx = (struct si_context *)ctx;
2237 struct pipe_constant_buffer constbuf = {0};
2238 struct r600_surface *surf = NULL;
2239 struct r600_texture *rtex;
2240 bool old_cb0_is_integer = sctx->framebuffer.cb0_is_integer;
2241 unsigned old_nr_samples = sctx->framebuffer.nr_samples;
2242 int i;
2243
2244 /* Only flush TC when changing the framebuffer state, because
2245 * the only client not using TC that can change textures is
2246 * the framebuffer.
2247 *
2248 * Flush all CB and DB caches here because all buffers can be used
2249 * for write by both TC (with shader image stores) and CB/DB.
2250 */
2251 sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1 |
2252 SI_CONTEXT_INV_GLOBAL_L2 |
2253 SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER |
2254 SI_CONTEXT_CS_PARTIAL_FLUSH;
2255
2256 /* Take the maximum of the old and new count. If the new count is lower,
2257 * dirtying is needed to disable the unbound colorbuffers.
2258 */
2259 sctx->framebuffer.dirty_cbufs |=
2260 (1 << MAX2(sctx->framebuffer.state.nr_cbufs, state->nr_cbufs)) - 1;
2261 sctx->framebuffer.dirty_zsbuf |= sctx->framebuffer.state.zsbuf != state->zsbuf;
2262
2263 si_dec_framebuffer_counters(&sctx->framebuffer.state);
2264 util_copy_framebuffer_state(&sctx->framebuffer.state, state);
2265
2266 sctx->framebuffer.spi_shader_col_format = 0;
2267 sctx->framebuffer.spi_shader_col_format_alpha = 0;
2268 sctx->framebuffer.spi_shader_col_format_blend = 0;
2269 sctx->framebuffer.spi_shader_col_format_blend_alpha = 0;
2270 sctx->framebuffer.color_is_int8 = 0;
2271
2272 sctx->framebuffer.compressed_cb_mask = 0;
2273 sctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
2274 sctx->framebuffer.log_samples = util_logbase2(sctx->framebuffer.nr_samples);
2275 sctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
2276 util_format_is_pure_integer(state->cbufs[0]->format);
2277
2278 if (sctx->framebuffer.cb0_is_integer != old_cb0_is_integer)
2279 si_mark_atom_dirty(sctx, &sctx->db_render_state);
2280
2281 for (i = 0; i < state->nr_cbufs; i++) {
2282 if (!state->cbufs[i])
2283 continue;
2284
2285 surf = (struct r600_surface*)state->cbufs[i];
2286 rtex = (struct r600_texture*)surf->base.texture;
2287
2288 if (!surf->color_initialized) {
2289 si_initialize_color_surface(sctx, surf);
2290 }
2291
2292 sctx->framebuffer.spi_shader_col_format |=
2293 surf->spi_shader_col_format << (i * 4);
2294 sctx->framebuffer.spi_shader_col_format_alpha |=
2295 surf->spi_shader_col_format_alpha << (i * 4);
2296 sctx->framebuffer.spi_shader_col_format_blend |=
2297 surf->spi_shader_col_format_blend << (i * 4);
2298 sctx->framebuffer.spi_shader_col_format_blend_alpha |=
2299 surf->spi_shader_col_format_blend_alpha << (i * 4);
2300
2301 if (surf->color_is_int8)
2302 sctx->framebuffer.color_is_int8 |= 1 << i;
2303
2304 if (rtex->fmask.size && rtex->cmask.size) {
2305 sctx->framebuffer.compressed_cb_mask |= 1 << i;
2306 }
2307 r600_context_add_resource_size(ctx, surf->base.texture);
2308
2309 p_atomic_inc(&rtex->framebuffers_bound);
2310 }
2311 /* Set the second SPI format for possible dual-src blending. */
2312 if (i == 1 && surf) {
2313 sctx->framebuffer.spi_shader_col_format |=
2314 surf->spi_shader_col_format << (i * 4);
2315 sctx->framebuffer.spi_shader_col_format_alpha |=
2316 surf->spi_shader_col_format_alpha << (i * 4);
2317 sctx->framebuffer.spi_shader_col_format_blend |=
2318 surf->spi_shader_col_format_blend << (i * 4);
2319 sctx->framebuffer.spi_shader_col_format_blend_alpha |=
2320 surf->spi_shader_col_format_blend_alpha << (i * 4);
2321 }
2322
2323 if (state->zsbuf) {
2324 surf = (struct r600_surface*)state->zsbuf;
2325
2326 if (!surf->depth_initialized) {
2327 si_init_depth_surface(sctx, surf);
2328 }
2329 r600_context_add_resource_size(ctx, surf->base.texture);
2330 }
2331
2332 si_update_poly_offset_state(sctx);
2333 si_mark_atom_dirty(sctx, &sctx->cb_render_state);
2334 si_mark_atom_dirty(sctx, &sctx->framebuffer.atom);
2335
2336 if (sctx->framebuffer.nr_samples != old_nr_samples) {
2337 si_mark_atom_dirty(sctx, &sctx->msaa_config);
2338 si_mark_atom_dirty(sctx, &sctx->db_render_state);
2339
2340 /* Set sample locations as fragment shader constants. */
2341 switch (sctx->framebuffer.nr_samples) {
2342 case 1:
2343 constbuf.user_buffer = sctx->b.sample_locations_1x;
2344 break;
2345 case 2:
2346 constbuf.user_buffer = sctx->b.sample_locations_2x;
2347 break;
2348 case 4:
2349 constbuf.user_buffer = sctx->b.sample_locations_4x;
2350 break;
2351 case 8:
2352 constbuf.user_buffer = sctx->b.sample_locations_8x;
2353 break;
2354 case 16:
2355 constbuf.user_buffer = sctx->b.sample_locations_16x;
2356 break;
2357 default:
2358 R600_ERR("Requested an invalid number of samples %i.\n",
2359 sctx->framebuffer.nr_samples);
2360 assert(0);
2361 }
2362 constbuf.buffer_size = sctx->framebuffer.nr_samples * 2 * 4;
2363 si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS, &constbuf);
2364
2365 /* Smoothing (only possible with nr_samples == 1) uses the same
2366 * sample locations as the MSAA it simulates.
2367 *
2368 * Therefore, don't update the sample locations when
2369 * transitioning from no AA to smoothing-equivalent AA, and
2370 * vice versa.
2371 */
2372 if ((sctx->framebuffer.nr_samples != 1 ||
2373 old_nr_samples != SI_NUM_SMOOTH_AA_SAMPLES) &&
2374 (sctx->framebuffer.nr_samples != SI_NUM_SMOOTH_AA_SAMPLES ||
2375 old_nr_samples != 1))
2376 si_mark_atom_dirty(sctx, &sctx->msaa_sample_locs);
2377 }
2378
2379 sctx->need_check_render_feedback = true;
2380 }
2381
2382 static void si_emit_framebuffer_state(struct si_context *sctx, struct r600_atom *atom)
2383 {
2384 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
2385 struct pipe_framebuffer_state *state = &sctx->framebuffer.state;
2386 unsigned i, nr_cbufs = state->nr_cbufs;
2387 struct r600_texture *tex = NULL;
2388 struct r600_surface *cb = NULL;
2389 unsigned cb_color_info = 0;
2390
2391 /* Colorbuffers. */
2392 for (i = 0; i < nr_cbufs; i++) {
2393 unsigned pitch_tile_max, slice_tile_max, tile_mode_index;
2394 unsigned cb_color_base, cb_color_fmask, cb_color_attrib;
2395 unsigned cb_color_pitch, cb_color_slice, cb_color_fmask_slice;
2396
2397 if (!(sctx->framebuffer.dirty_cbufs & (1 << i)))
2398 continue;
2399
2400 cb = (struct r600_surface*)state->cbufs[i];
2401 if (!cb) {
2402 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C,
2403 S_028C70_FORMAT(V_028C70_COLOR_INVALID));
2404 continue;
2405 }
2406
2407 tex = (struct r600_texture *)cb->base.texture;
2408 radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
2409 &tex->resource, RADEON_USAGE_READWRITE,
2410 tex->surface.nsamples > 1 ?
2411 RADEON_PRIO_COLOR_BUFFER_MSAA :
2412 RADEON_PRIO_COLOR_BUFFER);
2413
2414 if (tex->cmask_buffer && tex->cmask_buffer != &tex->resource) {
2415 radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
2416 tex->cmask_buffer, RADEON_USAGE_READWRITE,
2417 RADEON_PRIO_CMASK);
2418 }
2419
2420 /* Compute mutable surface parameters. */
2421 pitch_tile_max = cb->level_info->nblk_x / 8 - 1;
2422 slice_tile_max = cb->level_info->nblk_x *
2423 cb->level_info->nblk_y / 64 - 1;
2424 tile_mode_index = si_tile_mode_index(tex, cb->base.u.tex.level, false);
2425
2426 cb_color_base = (tex->resource.gpu_address + cb->level_info->offset) >> 8;
2427 cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max);
2428 cb_color_slice = S_028C68_TILE_MAX(slice_tile_max);
2429 cb_color_attrib = cb->cb_color_attrib |
2430 S_028C74_TILE_MODE_INDEX(tile_mode_index);
2431
2432 if (tex->fmask.size) {
2433 if (sctx->b.chip_class >= CIK)
2434 cb_color_pitch |= S_028C64_FMASK_TILE_MAX(tex->fmask.pitch_in_pixels / 8 - 1);
2435 cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tex->fmask.tile_mode_index);
2436 cb_color_fmask = (tex->resource.gpu_address + tex->fmask.offset) >> 8;
2437 cb_color_fmask_slice = S_028C88_TILE_MAX(tex->fmask.slice_tile_max);
2438 } else {
2439 /* This must be set for fast clear to work without FMASK. */
2440 if (sctx->b.chip_class >= CIK)
2441 cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch_tile_max);
2442 cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
2443 cb_color_fmask = cb_color_base;
2444 cb_color_fmask_slice = S_028C88_TILE_MAX(slice_tile_max);
2445 }
2446
2447 cb_color_info = cb->cb_color_info | tex->cb_color_info;
2448
2449 if (tex->dcc_offset && cb->level_info->dcc_enabled) {
2450 bool is_msaa_resolve_dst = state->cbufs[0] &&
2451 state->cbufs[0]->texture->nr_samples > 1 &&
2452 state->cbufs[1] == &cb->base &&
2453 state->cbufs[1]->texture->nr_samples <= 1;
2454
2455 if (!is_msaa_resolve_dst)
2456 cb_color_info |= S_028C70_DCC_ENABLE(1);
2457 }
2458
2459 radeon_set_context_reg_seq(cs, R_028C60_CB_COLOR0_BASE + i * 0x3C,
2460 sctx->b.chip_class >= VI ? 14 : 13);
2461 radeon_emit(cs, cb_color_base); /* R_028C60_CB_COLOR0_BASE */
2462 radeon_emit(cs, cb_color_pitch); /* R_028C64_CB_COLOR0_PITCH */
2463 radeon_emit(cs, cb_color_slice); /* R_028C68_CB_COLOR0_SLICE */
2464 radeon_emit(cs, cb->cb_color_view); /* R_028C6C_CB_COLOR0_VIEW */
2465 radeon_emit(cs, cb_color_info); /* R_028C70_CB_COLOR0_INFO */
2466 radeon_emit(cs, cb_color_attrib); /* R_028C74_CB_COLOR0_ATTRIB */
2467 radeon_emit(cs, cb->cb_dcc_control); /* R_028C78_CB_COLOR0_DCC_CONTROL */
2468 radeon_emit(cs, tex->cmask.base_address_reg); /* R_028C7C_CB_COLOR0_CMASK */
2469 radeon_emit(cs, tex->cmask.slice_tile_max); /* R_028C80_CB_COLOR0_CMASK_SLICE */
2470 radeon_emit(cs, cb_color_fmask); /* R_028C84_CB_COLOR0_FMASK */
2471 radeon_emit(cs, cb_color_fmask_slice); /* R_028C88_CB_COLOR0_FMASK_SLICE */
2472 radeon_emit(cs, tex->color_clear_value[0]); /* R_028C8C_CB_COLOR0_CLEAR_WORD0 */
2473 radeon_emit(cs, tex->color_clear_value[1]); /* R_028C90_CB_COLOR0_CLEAR_WORD1 */
2474
2475 if (sctx->b.chip_class >= VI) /* R_028C94_CB_COLOR0_DCC_BASE */
2476 radeon_emit(cs, (tex->resource.gpu_address +
2477 tex->dcc_offset +
2478 tex->surface.level[cb->base.u.tex.level].dcc_offset) >> 8);
2479 }
2480 /* set CB_COLOR1_INFO for possible dual-src blending */
2481 if (i == 1 && state->cbufs[0] &&
2482 sctx->framebuffer.dirty_cbufs & (1 << 0)) {
2483 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + 1 * 0x3C,
2484 cb_color_info);
2485 i++;
2486 }
2487 for (; i < 8 ; i++)
2488 if (sctx->framebuffer.dirty_cbufs & (1 << i))
2489 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C, 0);
2490
2491 /* ZS buffer. */
2492 if (state->zsbuf && sctx->framebuffer.dirty_zsbuf) {
2493 struct r600_surface *zb = (struct r600_surface*)state->zsbuf;
2494 struct r600_texture *rtex = (struct r600_texture*)zb->base.texture;
2495
2496 radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
2497 &rtex->resource, RADEON_USAGE_READWRITE,
2498 zb->base.texture->nr_samples > 1 ?
2499 RADEON_PRIO_DEPTH_BUFFER_MSAA :
2500 RADEON_PRIO_DEPTH_BUFFER);
2501
2502 if (zb->db_htile_data_base) {
2503 radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
2504 rtex->htile_buffer, RADEON_USAGE_READWRITE,
2505 RADEON_PRIO_HTILE);
2506 }
2507
2508 radeon_set_context_reg(cs, R_028008_DB_DEPTH_VIEW, zb->db_depth_view);
2509 radeon_set_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, zb->db_htile_data_base);
2510
2511 radeon_set_context_reg_seq(cs, R_02803C_DB_DEPTH_INFO, 9);
2512 radeon_emit(cs, zb->db_depth_info); /* R_02803C_DB_DEPTH_INFO */
2513 radeon_emit(cs, zb->db_z_info | /* R_028040_DB_Z_INFO */
2514 S_028040_ZRANGE_PRECISION(rtex->depth_clear_value != 0));
2515 radeon_emit(cs, zb->db_stencil_info); /* R_028044_DB_STENCIL_INFO */
2516 radeon_emit(cs, zb->db_depth_base); /* R_028048_DB_Z_READ_BASE */
2517 radeon_emit(cs, zb->db_stencil_base); /* R_02804C_DB_STENCIL_READ_BASE */
2518 radeon_emit(cs, zb->db_depth_base); /* R_028050_DB_Z_WRITE_BASE */
2519 radeon_emit(cs, zb->db_stencil_base); /* R_028054_DB_STENCIL_WRITE_BASE */
2520 radeon_emit(cs, zb->db_depth_size); /* R_028058_DB_DEPTH_SIZE */
2521 radeon_emit(cs, zb->db_depth_slice); /* R_02805C_DB_DEPTH_SLICE */
2522
2523 radeon_set_context_reg_seq(cs, R_028028_DB_STENCIL_CLEAR, 2);
2524 radeon_emit(cs, rtex->stencil_clear_value); /* R_028028_DB_STENCIL_CLEAR */
2525 radeon_emit(cs, fui(rtex->depth_clear_value)); /* R_02802C_DB_DEPTH_CLEAR */
2526
2527 radeon_set_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, zb->db_htile_surface);
2528 radeon_set_context_reg(cs, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
2529 zb->pa_su_poly_offset_db_fmt_cntl);
2530 } else if (sctx->framebuffer.dirty_zsbuf) {
2531 radeon_set_context_reg_seq(cs, R_028040_DB_Z_INFO, 2);
2532 radeon_emit(cs, S_028040_FORMAT(V_028040_Z_INVALID)); /* R_028040_DB_Z_INFO */
2533 radeon_emit(cs, S_028044_FORMAT(V_028044_STENCIL_INVALID)); /* R_028044_DB_STENCIL_INFO */
2534 }
2535
2536 /* Framebuffer dimensions. */
2537 /* PA_SC_WINDOW_SCISSOR_TL is set in si_init_config() */
2538 radeon_set_context_reg(cs, R_028208_PA_SC_WINDOW_SCISSOR_BR,
2539 S_028208_BR_X(state->width) | S_028208_BR_Y(state->height));
2540
2541 sctx->framebuffer.dirty_cbufs = 0;
2542 sctx->framebuffer.dirty_zsbuf = false;
2543 }
2544
2545 static void si_emit_msaa_sample_locs(struct si_context *sctx,
2546 struct r600_atom *atom)
2547 {
2548 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
2549 unsigned nr_samples = sctx->framebuffer.nr_samples;
2550
2551 cayman_emit_msaa_sample_locs(cs, nr_samples > 1 ? nr_samples :
2552 SI_NUM_SMOOTH_AA_SAMPLES);
2553 }
2554
2555 static void si_emit_msaa_config(struct si_context *sctx, struct r600_atom *atom)
2556 {
2557 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
2558 unsigned num_tile_pipes = sctx->screen->b.info.num_tile_pipes;
2559 bool dst_is_linear = false; /* TODO */
2560 unsigned sc_mode_cntl_1 =
2561 S_028A4C_WALK_SIZE(dst_is_linear) |
2562 S_028A4C_WALK_FENCE_ENABLE(!dst_is_linear) |
2563 S_028A4C_WALK_FENCE_SIZE(num_tile_pipes == 2 ? 2 : 3) |
2564 /* always 1: */
2565 S_028A4C_WALK_ALIGN8_PRIM_FITS_ST(1) |
2566 S_028A4C_SUPERTILE_WALK_ORDER_ENABLE(1) |
2567 S_028A4C_TILE_WALK_ORDER_ENABLE(1) |
2568 S_028A4C_MULTI_SHADER_ENGINE_PRIM_DISCARD_ENABLE(1) |
2569 S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
2570 S_028A4C_FORCE_EOV_REZ_ENABLE(1);
2571
2572 cayman_emit_msaa_config(cs, sctx->framebuffer.nr_samples,
2573 sctx->ps_iter_samples,
2574 sctx->smoothing_enabled ? SI_NUM_SMOOTH_AA_SAMPLES : 0,
2575 sc_mode_cntl_1);
2576 }
2577
2578 static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
2579 {
2580 struct si_context *sctx = (struct si_context *)ctx;
2581
2582 if (sctx->ps_iter_samples == min_samples)
2583 return;
2584
2585 sctx->ps_iter_samples = min_samples;
2586
2587 if (sctx->framebuffer.nr_samples > 1)
2588 si_mark_atom_dirty(sctx, &sctx->msaa_config);
2589 }
2590
2591 /*
2592 * Samplers
2593 */
2594
2595 /**
2596 * Build the sampler view descriptor for a buffer texture.
2597 * @param state 256-bit descriptor; only the high 128 bits are filled in
2598 */
2599 void
2600 si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
2601 enum pipe_format format,
2602 unsigned first_element, unsigned last_element,
2603 uint32_t *state)
2604 {
2605 const struct util_format_description *desc;
2606 int first_non_void;
2607 uint64_t va;
2608 unsigned stride;
2609 unsigned num_records;
2610 unsigned num_format, data_format;
2611
2612 desc = util_format_description(format);
2613 first_non_void = util_format_get_first_non_void_channel(format);
2614 stride = desc->block.bits / 8;
2615 va = buf->gpu_address + first_element * stride;
2616 num_format = si_translate_buffer_numformat(&screen->b.b, desc, first_non_void);
2617 data_format = si_translate_buffer_dataformat(&screen->b.b, desc, first_non_void);
2618
2619 num_records = last_element + 1 - first_element;
2620 num_records = MIN2(num_records, buf->b.b.width0 / stride);
2621
2622 if (screen->b.chip_class >= VI)
2623 num_records *= stride;
2624
2625 state[4] = va;
2626 state[5] = S_008F04_BASE_ADDRESS_HI(va >> 32) |
2627 S_008F04_STRIDE(stride);
2628 state[6] = num_records;
2629 state[7] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
2630 S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
2631 S_008F0C_DST_SEL_Z(si_map_swizzle(desc->swizzle[2])) |
2632 S_008F0C_DST_SEL_W(si_map_swizzle(desc->swizzle[3])) |
2633 S_008F0C_NUM_FORMAT(num_format) |
2634 S_008F0C_DATA_FORMAT(data_format);
2635 }
2636
2637 /**
2638 * Build the sampler view descriptor for a texture.
2639 */
2640 void
2641 si_make_texture_descriptor(struct si_screen *screen,
2642 struct r600_texture *tex,
2643 bool sampler,
2644 enum pipe_texture_target target,
2645 enum pipe_format pipe_format,
2646 const unsigned char state_swizzle[4],
2647 unsigned first_level, unsigned last_level,
2648 unsigned first_layer, unsigned last_layer,
2649 unsigned width, unsigned height, unsigned depth,
2650 uint32_t *state,
2651 uint32_t *fmask_state)
2652 {
2653 struct pipe_resource *res = &tex->resource.b.b;
2654 const struct util_format_description *desc;
2655 unsigned char swizzle[4];
2656 int first_non_void;
2657 unsigned num_format, data_format, type;
2658 uint64_t va;
2659
2660 desc = util_format_description(pipe_format);
2661
2662 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
2663 const unsigned char swizzle_xxxx[4] = {0, 0, 0, 0};
2664 const unsigned char swizzle_yyyy[4] = {1, 1, 1, 1};
2665
2666 switch (pipe_format) {
2667 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2668 case PIPE_FORMAT_X24S8_UINT:
2669 case PIPE_FORMAT_X32_S8X24_UINT:
2670 case PIPE_FORMAT_X8Z24_UNORM:
2671 util_format_compose_swizzles(swizzle_yyyy, state_swizzle, swizzle);
2672 break;
2673 default:
2674 util_format_compose_swizzles(swizzle_xxxx, state_swizzle, swizzle);
2675 }
2676 } else {
2677 util_format_compose_swizzles(desc->swizzle, state_swizzle, swizzle);
2678 }
2679
2680 first_non_void = util_format_get_first_non_void_channel(pipe_format);
2681
2682 switch (pipe_format) {
2683 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2684 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2685 break;
2686 default:
2687 if (first_non_void < 0) {
2688 if (util_format_is_compressed(pipe_format)) {
2689 switch (pipe_format) {
2690 case PIPE_FORMAT_DXT1_SRGB:
2691 case PIPE_FORMAT_DXT1_SRGBA:
2692 case PIPE_FORMAT_DXT3_SRGBA:
2693 case PIPE_FORMAT_DXT5_SRGBA:
2694 case PIPE_FORMAT_BPTC_SRGBA:
2695 case PIPE_FORMAT_ETC2_SRGB8:
2696 case PIPE_FORMAT_ETC2_SRGB8A1:
2697 case PIPE_FORMAT_ETC2_SRGBA8:
2698 num_format = V_008F14_IMG_NUM_FORMAT_SRGB;
2699 break;
2700 case PIPE_FORMAT_RGTC1_SNORM:
2701 case PIPE_FORMAT_LATC1_SNORM:
2702 case PIPE_FORMAT_RGTC2_SNORM:
2703 case PIPE_FORMAT_LATC2_SNORM:
2704 case PIPE_FORMAT_ETC2_R11_SNORM:
2705 case PIPE_FORMAT_ETC2_RG11_SNORM:
2706 /* implies float, so use SNORM/UNORM to determine
2707 whether data is signed or not */
2708 case PIPE_FORMAT_BPTC_RGB_FLOAT:
2709 num_format = V_008F14_IMG_NUM_FORMAT_SNORM;
2710 break;
2711 default:
2712 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2713 break;
2714 }
2715 } else if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
2716 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2717 } else {
2718 num_format = V_008F14_IMG_NUM_FORMAT_FLOAT;
2719 }
2720 } else if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
2721 num_format = V_008F14_IMG_NUM_FORMAT_SRGB;
2722 } else {
2723 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2724
2725 switch (desc->channel[first_non_void].type) {
2726 case UTIL_FORMAT_TYPE_FLOAT:
2727 num_format = V_008F14_IMG_NUM_FORMAT_FLOAT;
2728 break;
2729 case UTIL_FORMAT_TYPE_SIGNED:
2730 if (desc->channel[first_non_void].normalized)
2731 num_format = V_008F14_IMG_NUM_FORMAT_SNORM;
2732 else if (desc->channel[first_non_void].pure_integer)
2733 num_format = V_008F14_IMG_NUM_FORMAT_SINT;
2734 else
2735 num_format = V_008F14_IMG_NUM_FORMAT_SSCALED;
2736 break;
2737 case UTIL_FORMAT_TYPE_UNSIGNED:
2738 if (desc->channel[first_non_void].normalized)
2739 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2740 else if (desc->channel[first_non_void].pure_integer)
2741 num_format = V_008F14_IMG_NUM_FORMAT_UINT;
2742 else
2743 num_format = V_008F14_IMG_NUM_FORMAT_USCALED;
2744 }
2745 }
2746 }
2747
2748 data_format = si_translate_texformat(&screen->b.b, pipe_format, desc, first_non_void);
2749 if (data_format == ~0) {
2750 data_format = 0;
2751 }
2752
2753 if (!sampler &&
2754 (res->target == PIPE_TEXTURE_CUBE ||
2755 res->target == PIPE_TEXTURE_CUBE_ARRAY ||
2756 res->target == PIPE_TEXTURE_3D)) {
2757 /* For the purpose of shader images, treat cube maps and 3D
2758 * textures as 2D arrays. For 3D textures, the address
2759 * calculations for mipmaps are different, so we rely on the
2760 * caller to effectively disable mipmaps.
2761 */
2762 type = V_008F1C_SQ_RSRC_IMG_2D_ARRAY;
2763
2764 assert(res->target != PIPE_TEXTURE_3D || (first_level == 0 && last_level == 0));
2765 } else {
2766 type = si_tex_dim(res->target, target, res->nr_samples);
2767 }
2768
2769 if (type == V_008F1C_SQ_RSRC_IMG_1D_ARRAY) {
2770 height = 1;
2771 depth = res->array_size;
2772 } else if (type == V_008F1C_SQ_RSRC_IMG_2D_ARRAY ||
2773 type == V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY) {
2774 if (sampler || res->target != PIPE_TEXTURE_3D)
2775 depth = res->array_size;
2776 } else if (type == V_008F1C_SQ_RSRC_IMG_CUBE)
2777 depth = res->array_size / 6;
2778
2779 state[0] = 0;
2780 state[1] = (S_008F14_DATA_FORMAT(data_format) |
2781 S_008F14_NUM_FORMAT(num_format));
2782 state[2] = (S_008F18_WIDTH(width - 1) |
2783 S_008F18_HEIGHT(height - 1));
2784 state[3] = (S_008F1C_DST_SEL_X(si_map_swizzle(swizzle[0])) |
2785 S_008F1C_DST_SEL_Y(si_map_swizzle(swizzle[1])) |
2786 S_008F1C_DST_SEL_Z(si_map_swizzle(swizzle[2])) |
2787 S_008F1C_DST_SEL_W(si_map_swizzle(swizzle[3])) |
2788 S_008F1C_BASE_LEVEL(res->nr_samples > 1 ?
2789 0 : first_level) |
2790 S_008F1C_LAST_LEVEL(res->nr_samples > 1 ?
2791 util_logbase2(res->nr_samples) :
2792 last_level) |
2793 S_008F1C_POW2_PAD(res->last_level > 0) |
2794 S_008F1C_TYPE(type));
2795 state[4] = S_008F20_DEPTH(depth - 1);
2796 state[5] = (S_008F24_BASE_ARRAY(first_layer) |
2797 S_008F24_LAST_ARRAY(last_layer));
2798 state[6] = 0;
2799 state[7] = 0;
2800
2801 if (tex->dcc_offset) {
2802 unsigned swap = r600_translate_colorswap(pipe_format, FALSE);
2803
2804 state[6] = S_008F28_ALPHA_IS_ON_MSB(swap <= 1);
2805 } else {
2806 /* The last dword is unused by hw. The shader uses it to clear
2807 * bits in the first dword of sampler state.
2808 */
2809 if (screen->b.chip_class <= CIK && res->nr_samples <= 1) {
2810 if (first_level == last_level)
2811 state[7] = C_008F30_MAX_ANISO_RATIO;
2812 else
2813 state[7] = 0xffffffff;
2814 }
2815 }
2816
2817 /* Initialize the sampler view for FMASK. */
2818 if (tex->fmask.size) {
2819 uint32_t fmask_format;
2820
2821 va = tex->resource.gpu_address + tex->fmask.offset;
2822
2823 switch (res->nr_samples) {
2824 case 2:
2825 fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S2_F2;
2826 break;
2827 case 4:
2828 fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S4_F4;
2829 break;
2830 case 8:
2831 fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK32_S8_F8;
2832 break;
2833 default:
2834 assert(0);
2835 fmask_format = V_008F14_IMG_DATA_FORMAT_INVALID;
2836 }
2837
2838 fmask_state[0] = va >> 8;
2839 fmask_state[1] = S_008F14_BASE_ADDRESS_HI(va >> 40) |
2840 S_008F14_DATA_FORMAT(fmask_format) |
2841 S_008F14_NUM_FORMAT(V_008F14_IMG_NUM_FORMAT_UINT);
2842 fmask_state[2] = S_008F18_WIDTH(width - 1) |
2843 S_008F18_HEIGHT(height - 1);
2844 fmask_state[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X) |
2845 S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X) |
2846 S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X) |
2847 S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X) |
2848 S_008F1C_TILING_INDEX(tex->fmask.tile_mode_index) |
2849 S_008F1C_TYPE(si_tex_dim(res->target, target, 0));
2850 fmask_state[4] = S_008F20_DEPTH(depth - 1) |
2851 S_008F20_PITCH(tex->fmask.pitch_in_pixels - 1);
2852 fmask_state[5] = S_008F24_BASE_ARRAY(first_layer) |
2853 S_008F24_LAST_ARRAY(last_layer);
2854 fmask_state[6] = 0;
2855 fmask_state[7] = 0;
2856 }
2857 }
2858
2859 /**
2860 * Create a sampler view.
2861 *
2862 * @param ctx context
2863 * @param texture texture
2864 * @param state sampler view template
2865 * @param width0 width0 override (for compressed textures as int)
2866 * @param height0 height0 override (for compressed textures as int)
2867 * @param force_level set the base address to the level (for compressed textures)
2868 */
2869 struct pipe_sampler_view *
2870 si_create_sampler_view_custom(struct pipe_context *ctx,
2871 struct pipe_resource *texture,
2872 const struct pipe_sampler_view *state,
2873 unsigned width0, unsigned height0,
2874 unsigned force_level)
2875 {
2876 struct si_context *sctx = (struct si_context*)ctx;
2877 struct si_sampler_view *view = CALLOC_STRUCT(si_sampler_view);
2878 struct r600_texture *tmp = (struct r600_texture*)texture;
2879 unsigned base_level, first_level, last_level;
2880 unsigned char state_swizzle[4];
2881 unsigned height, depth, width;
2882 unsigned last_layer = state->u.tex.last_layer;
2883 enum pipe_format pipe_format;
2884 const struct radeon_surf_level *surflevel;
2885
2886 if (!view)
2887 return NULL;
2888
2889 /* initialize base object */
2890 view->base = *state;
2891 view->base.texture = NULL;
2892 view->base.reference.count = 1;
2893 view->base.context = ctx;
2894
2895 /* NULL resource, obey swizzle (only ZERO and ONE make sense). */
2896 if (!texture) {
2897 view->state[3] = S_008F1C_DST_SEL_X(si_map_swizzle(state->swizzle_r)) |
2898 S_008F1C_DST_SEL_Y(si_map_swizzle(state->swizzle_g)) |
2899 S_008F1C_DST_SEL_Z(si_map_swizzle(state->swizzle_b)) |
2900 S_008F1C_DST_SEL_W(si_map_swizzle(state->swizzle_a)) |
2901 S_008F1C_TYPE(V_008F1C_SQ_RSRC_IMG_1D);
2902 return &view->base;
2903 }
2904
2905 pipe_resource_reference(&view->base.texture, texture);
2906
2907 if (state->format == PIPE_FORMAT_X24S8_UINT ||
2908 state->format == PIPE_FORMAT_S8X24_UINT ||
2909 state->format == PIPE_FORMAT_X32_S8X24_UINT ||
2910 state->format == PIPE_FORMAT_S8_UINT)
2911 view->is_stencil_sampler = true;
2912
2913 /* Buffer resource. */
2914 if (texture->target == PIPE_BUFFER) {
2915 si_make_buffer_descriptor(sctx->screen,
2916 (struct r600_resource *)texture,
2917 state->format,
2918 state->u.buf.first_element,
2919 state->u.buf.last_element,
2920 view->state);
2921
2922 LIST_ADDTAIL(&view->list, &sctx->b.texture_buffers);
2923 return &view->base;
2924 }
2925
2926 state_swizzle[0] = state->swizzle_r;
2927 state_swizzle[1] = state->swizzle_g;
2928 state_swizzle[2] = state->swizzle_b;
2929 state_swizzle[3] = state->swizzle_a;
2930
2931 base_level = 0;
2932 first_level = state->u.tex.first_level;
2933 last_level = state->u.tex.last_level;
2934 width = width0;
2935 height = height0;
2936 depth = texture->depth0;
2937
2938 if (force_level) {
2939 assert(force_level == first_level &&
2940 force_level == last_level);
2941 base_level = force_level;
2942 first_level = 0;
2943 last_level = 0;
2944 width = u_minify(width, force_level);
2945 height = u_minify(height, force_level);
2946 depth = u_minify(depth, force_level);
2947 }
2948
2949 /* This is not needed if state trackers set last_layer correctly. */
2950 if (state->target == PIPE_TEXTURE_1D ||
2951 state->target == PIPE_TEXTURE_2D ||
2952 state->target == PIPE_TEXTURE_RECT ||
2953 state->target == PIPE_TEXTURE_CUBE)
2954 last_layer = state->u.tex.first_layer;
2955
2956 /* Texturing with separate depth and stencil. */
2957 pipe_format = state->format;
2958 surflevel = tmp->surface.level;
2959
2960 if (tmp->is_depth && !tmp->is_flushing_texture) {
2961 switch (pipe_format) {
2962 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
2963 pipe_format = PIPE_FORMAT_Z32_FLOAT;
2964 break;
2965 case PIPE_FORMAT_X8Z24_UNORM:
2966 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2967 /* Z24 is always stored like this. */
2968 pipe_format = PIPE_FORMAT_Z24X8_UNORM;
2969 break;
2970 case PIPE_FORMAT_X24S8_UINT:
2971 case PIPE_FORMAT_S8X24_UINT:
2972 case PIPE_FORMAT_X32_S8X24_UINT:
2973 pipe_format = PIPE_FORMAT_S8_UINT;
2974 surflevel = tmp->surface.stencil_level;
2975 break;
2976 default:;
2977 }
2978 }
2979
2980 si_make_texture_descriptor(sctx->screen, tmp, true,
2981 state->target, pipe_format, state_swizzle,
2982 first_level, last_level,
2983 state->u.tex.first_layer, last_layer,
2984 width, height, depth,
2985 view->state, view->fmask_state);
2986
2987 view->base_level_info = &surflevel[base_level];
2988 view->base_level = base_level;
2989 view->block_width = util_format_get_blockwidth(pipe_format);
2990 return &view->base;
2991 }
2992
2993 static struct pipe_sampler_view *
2994 si_create_sampler_view(struct pipe_context *ctx,
2995 struct pipe_resource *texture,
2996 const struct pipe_sampler_view *state)
2997 {
2998 return si_create_sampler_view_custom(ctx, texture, state,
2999 texture ? texture->width0 : 0,
3000 texture ? texture->height0 : 0, 0);
3001 }
3002
3003 static void si_sampler_view_destroy(struct pipe_context *ctx,
3004 struct pipe_sampler_view *state)
3005 {
3006 struct si_sampler_view *view = (struct si_sampler_view *)state;
3007
3008 if (state->texture && state->texture->target == PIPE_BUFFER)
3009 LIST_DELINIT(&view->list);
3010
3011 pipe_resource_reference(&state->texture, NULL);
3012 FREE(view);
3013 }
3014
3015 static bool wrap_mode_uses_border_color(unsigned wrap, bool linear_filter)
3016 {
3017 return wrap == PIPE_TEX_WRAP_CLAMP_TO_BORDER ||
3018 wrap == PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER ||
3019 (linear_filter &&
3020 (wrap == PIPE_TEX_WRAP_CLAMP ||
3021 wrap == PIPE_TEX_WRAP_MIRROR_CLAMP));
3022 }
3023
3024 static bool sampler_state_needs_border_color(const struct pipe_sampler_state *state)
3025 {
3026 bool linear_filter = state->min_img_filter != PIPE_TEX_FILTER_NEAREST ||
3027 state->mag_img_filter != PIPE_TEX_FILTER_NEAREST;
3028
3029 return (state->border_color.ui[0] || state->border_color.ui[1] ||
3030 state->border_color.ui[2] || state->border_color.ui[3]) &&
3031 (wrap_mode_uses_border_color(state->wrap_s, linear_filter) ||
3032 wrap_mode_uses_border_color(state->wrap_t, linear_filter) ||
3033 wrap_mode_uses_border_color(state->wrap_r, linear_filter));
3034 }
3035
3036 static void *si_create_sampler_state(struct pipe_context *ctx,
3037 const struct pipe_sampler_state *state)
3038 {
3039 struct si_context *sctx = (struct si_context *)ctx;
3040 struct r600_common_screen *rscreen = sctx->b.screen;
3041 struct si_sampler_state *rstate = CALLOC_STRUCT(si_sampler_state);
3042 unsigned border_color_type, border_color_index = 0;
3043 unsigned max_aniso = rscreen->force_aniso >= 0 ? rscreen->force_aniso
3044 : state->max_anisotropy;
3045 unsigned max_aniso_ratio = r600_tex_aniso_filter(max_aniso);
3046
3047 if (!rstate) {
3048 return NULL;
3049 }
3050
3051 if (!sampler_state_needs_border_color(state))
3052 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK;
3053 else if (state->border_color.f[0] == 0 &&
3054 state->border_color.f[1] == 0 &&
3055 state->border_color.f[2] == 0 &&
3056 state->border_color.f[3] == 0)
3057 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK;
3058 else if (state->border_color.f[0] == 0 &&
3059 state->border_color.f[1] == 0 &&
3060 state->border_color.f[2] == 0 &&
3061 state->border_color.f[3] == 1)
3062 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK;
3063 else if (state->border_color.f[0] == 1 &&
3064 state->border_color.f[1] == 1 &&
3065 state->border_color.f[2] == 1 &&
3066 state->border_color.f[3] == 1)
3067 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE;
3068 else {
3069 int i;
3070
3071 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER;
3072
3073 /* Check if the border has been uploaded already. */
3074 for (i = 0; i < sctx->border_color_count; i++)
3075 if (memcmp(&sctx->border_color_table[i], &state->border_color,
3076 sizeof(state->border_color)) == 0)
3077 break;
3078
3079 if (i >= SI_MAX_BORDER_COLORS) {
3080 /* Getting 4096 unique border colors is very unlikely. */
3081 fprintf(stderr, "radeonsi: The border color table is full. "
3082 "Any new border colors will be just black. "
3083 "Please file a bug.\n");
3084 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK;
3085 } else {
3086 if (i == sctx->border_color_count) {
3087 /* Upload a new border color. */
3088 memcpy(&sctx->border_color_table[i], &state->border_color,
3089 sizeof(state->border_color));
3090 util_memcpy_cpu_to_le32(&sctx->border_color_map[i],
3091 &state->border_color,
3092 sizeof(state->border_color));
3093 sctx->border_color_count++;
3094 }
3095
3096 border_color_index = i;
3097 }
3098 }
3099
3100 rstate->val[0] = (S_008F30_CLAMP_X(si_tex_wrap(state->wrap_s)) |
3101 S_008F30_CLAMP_Y(si_tex_wrap(state->wrap_t)) |
3102 S_008F30_CLAMP_Z(si_tex_wrap(state->wrap_r)) |
3103 S_008F30_MAX_ANISO_RATIO(max_aniso_ratio) |
3104 S_008F30_DEPTH_COMPARE_FUNC(si_tex_compare(state->compare_func)) |
3105 S_008F30_FORCE_UNNORMALIZED(!state->normalized_coords) |
3106 S_008F30_DISABLE_CUBE_WRAP(!state->seamless_cube_map) |
3107 S_008F30_COMPAT_MODE(sctx->b.chip_class >= VI));
3108 rstate->val[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
3109 S_008F34_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)));
3110 rstate->val[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
3111 S_008F38_XY_MAG_FILTER(eg_tex_filter(state->mag_img_filter, max_aniso)) |
3112 S_008F38_XY_MIN_FILTER(eg_tex_filter(state->min_img_filter, max_aniso)) |
3113 S_008F38_MIP_FILTER(si_tex_mipfilter(state->min_mip_filter)) |
3114 S_008F38_MIP_POINT_PRECLAMP(1) |
3115 S_008F38_DISABLE_LSB_CEIL(1) |
3116 S_008F38_FILTER_PREC_FIX(1) |
3117 S_008F38_ANISO_OVERRIDE(sctx->b.chip_class >= VI));
3118 rstate->val[3] = S_008F3C_BORDER_COLOR_PTR(border_color_index) |
3119 S_008F3C_BORDER_COLOR_TYPE(border_color_type);
3120 return rstate;
3121 }
3122
3123 static void si_set_sample_mask(struct pipe_context *ctx, unsigned sample_mask)
3124 {
3125 struct si_context *sctx = (struct si_context *)ctx;
3126
3127 if (sctx->sample_mask.sample_mask == (uint16_t)sample_mask)
3128 return;
3129
3130 sctx->sample_mask.sample_mask = sample_mask;
3131 si_mark_atom_dirty(sctx, &sctx->sample_mask.atom);
3132 }
3133
3134 static void si_emit_sample_mask(struct si_context *sctx, struct r600_atom *atom)
3135 {
3136 struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
3137 unsigned mask = sctx->sample_mask.sample_mask;
3138
3139 radeon_set_context_reg_seq(cs, R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0, 2);
3140 radeon_emit(cs, mask | (mask << 16));
3141 radeon_emit(cs, mask | (mask << 16));
3142 }
3143
3144 static void si_delete_sampler_state(struct pipe_context *ctx, void *state)
3145 {
3146 free(state);
3147 }
3148
3149 /*
3150 * Vertex elements & buffers
3151 */
3152
3153 static void *si_create_vertex_elements(struct pipe_context *ctx,
3154 unsigned count,
3155 const struct pipe_vertex_element *elements)
3156 {
3157 struct si_vertex_element *v = CALLOC_STRUCT(si_vertex_element);
3158 int i;
3159
3160 assert(count <= SI_MAX_ATTRIBS);
3161 if (!v)
3162 return NULL;
3163
3164 v->count = count;
3165 for (i = 0; i < count; ++i) {
3166 const struct util_format_description *desc;
3167 unsigned data_format, num_format;
3168 int first_non_void;
3169
3170 desc = util_format_description(elements[i].src_format);
3171 first_non_void = util_format_get_first_non_void_channel(elements[i].src_format);
3172 data_format = si_translate_buffer_dataformat(ctx->screen, desc, first_non_void);
3173 num_format = si_translate_buffer_numformat(ctx->screen, desc, first_non_void);
3174
3175 v->rsrc_word3[i] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
3176 S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
3177 S_008F0C_DST_SEL_Z(si_map_swizzle(desc->swizzle[2])) |
3178 S_008F0C_DST_SEL_W(si_map_swizzle(desc->swizzle[3])) |
3179 S_008F0C_NUM_FORMAT(num_format) |
3180 S_008F0C_DATA_FORMAT(data_format);
3181 v->format_size[i] = desc->block.bits / 8;
3182 }
3183 memcpy(v->elements, elements, sizeof(struct pipe_vertex_element) * count);
3184
3185 return v;
3186 }
3187
3188 static void si_bind_vertex_elements(struct pipe_context *ctx, void *state)
3189 {
3190 struct si_context *sctx = (struct si_context *)ctx;
3191 struct si_vertex_element *v = (struct si_vertex_element*)state;
3192
3193 sctx->vertex_elements = v;
3194 sctx->vertex_buffers_dirty = true;
3195 }
3196
3197 static void si_delete_vertex_element(struct pipe_context *ctx, void *state)
3198 {
3199 struct si_context *sctx = (struct si_context *)ctx;
3200
3201 if (sctx->vertex_elements == state)
3202 sctx->vertex_elements = NULL;
3203 FREE(state);
3204 }
3205
3206 static void si_set_vertex_buffers(struct pipe_context *ctx,
3207 unsigned start_slot, unsigned count,
3208 const struct pipe_vertex_buffer *buffers)
3209 {
3210 struct si_context *sctx = (struct si_context *)ctx;
3211 struct pipe_vertex_buffer *dst = sctx->vertex_buffer + start_slot;
3212 int i;
3213
3214 assert(start_slot + count <= ARRAY_SIZE(sctx->vertex_buffer));
3215
3216 if (buffers) {
3217 for (i = 0; i < count; i++) {
3218 const struct pipe_vertex_buffer *src = buffers + i;
3219 struct pipe_vertex_buffer *dsti = dst + i;
3220
3221 pipe_resource_reference(&dsti->buffer, src->buffer);
3222 dsti->buffer_offset = src->buffer_offset;
3223 dsti->stride = src->stride;
3224 r600_context_add_resource_size(ctx, src->buffer);
3225 }
3226 } else {
3227 for (i = 0; i < count; i++) {
3228 pipe_resource_reference(&dst[i].buffer, NULL);
3229 }
3230 }
3231 sctx->vertex_buffers_dirty = true;
3232 }
3233
3234 static void si_set_index_buffer(struct pipe_context *ctx,
3235 const struct pipe_index_buffer *ib)
3236 {
3237 struct si_context *sctx = (struct si_context *)ctx;
3238
3239 if (ib) {
3240 pipe_resource_reference(&sctx->index_buffer.buffer, ib->buffer);
3241 memcpy(&sctx->index_buffer, ib, sizeof(*ib));
3242 r600_context_add_resource_size(ctx, ib->buffer);
3243 } else {
3244 pipe_resource_reference(&sctx->index_buffer.buffer, NULL);
3245 }
3246 }
3247
3248 /*
3249 * Misc
3250 */
3251
3252 static void si_set_tess_state(struct pipe_context *ctx,
3253 const float default_outer_level[4],
3254 const float default_inner_level[2])
3255 {
3256 struct si_context *sctx = (struct si_context *)ctx;
3257 struct pipe_constant_buffer cb;
3258 float array[8];
3259
3260 memcpy(array, default_outer_level, sizeof(float) * 4);
3261 memcpy(array+4, default_inner_level, sizeof(float) * 2);
3262
3263 cb.buffer = NULL;
3264 cb.user_buffer = NULL;
3265 cb.buffer_size = sizeof(array);
3266
3267 si_upload_const_buffer(sctx, (struct r600_resource**)&cb.buffer,
3268 (void*)array, sizeof(array),
3269 &cb.buffer_offset);
3270
3271 si_set_rw_buffer(sctx, SI_HS_CONST_DEFAULT_TESS_LEVELS, &cb);
3272 pipe_resource_reference(&cb.buffer, NULL);
3273 }
3274
3275 static void si_texture_barrier(struct pipe_context *ctx)
3276 {
3277 struct si_context *sctx = (struct si_context *)ctx;
3278
3279 sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1 |
3280 SI_CONTEXT_INV_GLOBAL_L2 |
3281 SI_CONTEXT_FLUSH_AND_INV_CB |
3282 SI_CONTEXT_CS_PARTIAL_FLUSH;
3283 }
3284
3285 static void si_memory_barrier(struct pipe_context *ctx, unsigned flags)
3286 {
3287 struct si_context *sctx = (struct si_context *)ctx;
3288
3289 /* Subsequent commands must wait for all shader invocations to
3290 * complete. */
3291 sctx->b.flags |= SI_CONTEXT_PS_PARTIAL_FLUSH |
3292 SI_CONTEXT_CS_PARTIAL_FLUSH;
3293
3294 if (flags & PIPE_BARRIER_CONSTANT_BUFFER)
3295 sctx->b.flags |= SI_CONTEXT_INV_SMEM_L1 |
3296 SI_CONTEXT_INV_VMEM_L1;
3297
3298 if (flags & (PIPE_BARRIER_VERTEX_BUFFER |
3299 PIPE_BARRIER_SHADER_BUFFER |
3300 PIPE_BARRIER_TEXTURE |
3301 PIPE_BARRIER_IMAGE |
3302 PIPE_BARRIER_STREAMOUT_BUFFER |
3303 PIPE_BARRIER_GLOBAL_BUFFER)) {
3304 /* As far as I can tell, L1 contents are written back to L2
3305 * automatically at end of shader, but the contents of other
3306 * L1 caches might still be stale. */
3307 sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1;
3308 }
3309
3310 if (flags & PIPE_BARRIER_INDEX_BUFFER) {
3311 sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1;
3312
3313 /* Indices are read through TC L2 since VI. */
3314 if (sctx->screen->b.chip_class <= CIK)
3315 sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
3316 }
3317
3318 if (flags & PIPE_BARRIER_FRAMEBUFFER)
3319 sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
3320
3321 if (flags & (PIPE_BARRIER_MAPPED_BUFFER |
3322 PIPE_BARRIER_FRAMEBUFFER |
3323 PIPE_BARRIER_INDIRECT_BUFFER)) {
3324 /* Not sure if INV_GLOBAL_L2 is the best thing here.
3325 *
3326 * We need to make sure that TC L1 & L2 are written back to
3327 * memory, because neither CPU accesses nor CB fetches consider
3328 * TC, but there's no need to invalidate any TC cache lines. */
3329 sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
3330 }
3331 }
3332
3333 static void *si_create_blend_custom(struct si_context *sctx, unsigned mode)
3334 {
3335 struct pipe_blend_state blend;
3336
3337 memset(&blend, 0, sizeof(blend));
3338 blend.independent_blend_enable = true;
3339 blend.rt[0].colormask = 0xf;
3340 return si_create_blend_state_mode(&sctx->b.b, &blend, mode);
3341 }
3342
3343 static void si_need_gfx_cs_space(struct pipe_context *ctx, unsigned num_dw,
3344 bool include_draw_vbo)
3345 {
3346 si_need_cs_space((struct si_context*)ctx);
3347 }
3348
3349 static void si_init_config(struct si_context *sctx);
3350
3351 void si_init_state_functions(struct si_context *sctx)
3352 {
3353 si_init_external_atom(sctx, &sctx->b.render_cond_atom, &sctx->atoms.s.render_cond);
3354 si_init_external_atom(sctx, &sctx->b.streamout.begin_atom, &sctx->atoms.s.streamout_begin);
3355 si_init_external_atom(sctx, &sctx->b.streamout.enable_atom, &sctx->atoms.s.streamout_enable);
3356 si_init_external_atom(sctx, &sctx->b.scissors.atom, &sctx->atoms.s.scissors);
3357 si_init_external_atom(sctx, &sctx->b.viewports.atom, &sctx->atoms.s.viewports);
3358
3359 si_init_atom(sctx, &sctx->cache_flush, &sctx->atoms.s.cache_flush, si_emit_cache_flush);
3360 si_init_atom(sctx, &sctx->framebuffer.atom, &sctx->atoms.s.framebuffer, si_emit_framebuffer_state);
3361 si_init_atom(sctx, &sctx->msaa_sample_locs, &sctx->atoms.s.msaa_sample_locs, si_emit_msaa_sample_locs);
3362 si_init_atom(sctx, &sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state);
3363 si_init_atom(sctx, &sctx->msaa_config, &sctx->atoms.s.msaa_config, si_emit_msaa_config);
3364 si_init_atom(sctx, &sctx->sample_mask.atom, &sctx->atoms.s.sample_mask, si_emit_sample_mask);
3365 si_init_atom(sctx, &sctx->cb_render_state, &sctx->atoms.s.cb_render_state, si_emit_cb_render_state);
3366 si_init_atom(sctx, &sctx->blend_color.atom, &sctx->atoms.s.blend_color, si_emit_blend_color);
3367 si_init_atom(sctx, &sctx->clip_regs, &sctx->atoms.s.clip_regs, si_emit_clip_regs);
3368 si_init_atom(sctx, &sctx->clip_state.atom, &sctx->atoms.s.clip_state, si_emit_clip_state);
3369 si_init_atom(sctx, &sctx->stencil_ref.atom, &sctx->atoms.s.stencil_ref, si_emit_stencil_ref);
3370
3371 sctx->b.b.create_blend_state = si_create_blend_state;
3372 sctx->b.b.bind_blend_state = si_bind_blend_state;
3373 sctx->b.b.delete_blend_state = si_delete_blend_state;
3374 sctx->b.b.set_blend_color = si_set_blend_color;
3375
3376 sctx->b.b.create_rasterizer_state = si_create_rs_state;
3377 sctx->b.b.bind_rasterizer_state = si_bind_rs_state;
3378 sctx->b.b.delete_rasterizer_state = si_delete_rs_state;
3379
3380 sctx->b.b.create_depth_stencil_alpha_state = si_create_dsa_state;
3381 sctx->b.b.bind_depth_stencil_alpha_state = si_bind_dsa_state;
3382 sctx->b.b.delete_depth_stencil_alpha_state = si_delete_dsa_state;
3383
3384 sctx->custom_dsa_flush = si_create_db_flush_dsa(sctx);
3385 sctx->custom_blend_resolve = si_create_blend_custom(sctx, V_028808_CB_RESOLVE);
3386 sctx->custom_blend_decompress = si_create_blend_custom(sctx, V_028808_CB_FMASK_DECOMPRESS);
3387 sctx->custom_blend_fastclear = si_create_blend_custom(sctx, V_028808_CB_ELIMINATE_FAST_CLEAR);
3388 sctx->custom_blend_dcc_decompress = si_create_blend_custom(sctx, V_028808_CB_DCC_DECOMPRESS);
3389
3390 sctx->b.b.set_clip_state = si_set_clip_state;
3391 sctx->b.b.set_stencil_ref = si_set_stencil_ref;
3392
3393 sctx->b.b.set_framebuffer_state = si_set_framebuffer_state;
3394 sctx->b.b.get_sample_position = cayman_get_sample_position;
3395
3396 sctx->b.b.create_sampler_state = si_create_sampler_state;
3397 sctx->b.b.delete_sampler_state = si_delete_sampler_state;
3398
3399 sctx->b.b.create_sampler_view = si_create_sampler_view;
3400 sctx->b.b.sampler_view_destroy = si_sampler_view_destroy;
3401
3402 sctx->b.b.set_sample_mask = si_set_sample_mask;
3403
3404 sctx->b.b.create_vertex_elements_state = si_create_vertex_elements;
3405 sctx->b.b.bind_vertex_elements_state = si_bind_vertex_elements;
3406 sctx->b.b.delete_vertex_elements_state = si_delete_vertex_element;
3407 sctx->b.b.set_vertex_buffers = si_set_vertex_buffers;
3408 sctx->b.b.set_index_buffer = si_set_index_buffer;
3409
3410 sctx->b.b.texture_barrier = si_texture_barrier;
3411 sctx->b.b.memory_barrier = si_memory_barrier;
3412 sctx->b.b.set_min_samples = si_set_min_samples;
3413 sctx->b.b.set_tess_state = si_set_tess_state;
3414
3415 sctx->b.b.set_active_query_state = si_set_active_query_state;
3416 sctx->b.set_occlusion_query_state = si_set_occlusion_query_state;
3417 sctx->b.need_gfx_cs_space = si_need_gfx_cs_space;
3418
3419 sctx->b.b.draw_vbo = si_draw_vbo;
3420
3421 si_init_config(sctx);
3422 }
3423
3424 static uint32_t si_get_bo_metadata_word1(struct r600_common_screen *rscreen)
3425 {
3426 return (ATI_VENDOR_ID << 16) | rscreen->info.pci_id;
3427 }
3428
3429 static void si_query_opaque_metadata(struct r600_common_screen *rscreen,
3430 struct r600_texture *rtex,
3431 struct radeon_bo_metadata *md)
3432 {
3433 struct si_screen *sscreen = (struct si_screen*)rscreen;
3434 struct pipe_resource *res = &rtex->resource.b.b;
3435 static const unsigned char swizzle[] = {
3436 PIPE_SWIZZLE_X,
3437 PIPE_SWIZZLE_Y,
3438 PIPE_SWIZZLE_Z,
3439 PIPE_SWIZZLE_W
3440 };
3441 uint32_t desc[8], i;
3442 bool is_array = util_resource_is_array_texture(res);
3443
3444 /* DRM 2.x.x doesn't support this. */
3445 if (rscreen->info.drm_major != 3)
3446 return;
3447
3448 assert(rtex->fmask.size == 0);
3449
3450 /* Metadata image format format version 1:
3451 * [0] = 1 (metadata format identifier)
3452 * [1] = (VENDOR_ID << 16) | PCI_ID
3453 * [2:9] = image descriptor for the whole resource
3454 * [2] is always 0, because the base address is cleared
3455 * [9] is the DCC offset bits [39:8] from the beginning of
3456 * the buffer
3457 * [10:10+LAST_LEVEL] = mipmap level offset bits [39:8] for each level
3458 */
3459
3460 md->metadata[0] = 1; /* metadata image format version 1 */
3461
3462 /* TILE_MODE_INDEX is ambiguous without a PCI ID. */
3463 md->metadata[1] = si_get_bo_metadata_word1(rscreen);
3464
3465 si_make_texture_descriptor(sscreen, rtex, true,
3466 res->target, res->format,
3467 swizzle, 0, res->last_level, 0,
3468 is_array ? res->array_size - 1 : 0,
3469 res->width0, res->height0, res->depth0,
3470 desc, NULL);
3471
3472 si_set_mutable_tex_desc_fields(rtex, &rtex->surface.level[0], 0, 0,
3473 rtex->surface.blk_w, false, desc);
3474
3475 /* Clear the base address and set the relative DCC offset. */
3476 desc[0] = 0;
3477 desc[1] &= C_008F14_BASE_ADDRESS_HI;
3478 desc[7] = rtex->dcc_offset >> 8;
3479
3480 /* Dwords [2:9] contain the image descriptor. */
3481 memcpy(&md->metadata[2], desc, sizeof(desc));
3482
3483 /* Dwords [10:..] contain the mipmap level offsets. */
3484 for (i = 0; i <= res->last_level; i++)
3485 md->metadata[10+i] = rtex->surface.level[i].offset >> 8;
3486
3487 md->size_metadata = (11 + res->last_level) * 4;
3488 }
3489
3490 static void si_apply_opaque_metadata(struct r600_common_screen *rscreen,
3491 struct r600_texture *rtex,
3492 struct radeon_bo_metadata *md)
3493 {
3494 uint32_t *desc = &md->metadata[2];
3495
3496 if (rscreen->chip_class < VI)
3497 return;
3498
3499 /* Return if DCC is enabled. The texture should be set up with it
3500 * already.
3501 */
3502 if (md->size_metadata >= 11 * 4 &&
3503 md->metadata[0] != 0 &&
3504 md->metadata[1] == si_get_bo_metadata_word1(rscreen) &&
3505 G_008F28_COMPRESSION_EN(desc[6])) {
3506 assert(rtex->dcc_offset == ((uint64_t)desc[7] << 8));
3507 return;
3508 }
3509
3510 /* Disable DCC. These are always set by texture_from_handle and must
3511 * be cleared here.
3512 */
3513 rtex->dcc_offset = 0;
3514 }
3515
3516 void si_init_screen_state_functions(struct si_screen *sscreen)
3517 {
3518 sscreen->b.query_opaque_metadata = si_query_opaque_metadata;
3519 sscreen->b.apply_opaque_metadata = si_apply_opaque_metadata;
3520 }
3521
3522 static void
3523 si_write_harvested_raster_configs(struct si_context *sctx,
3524 struct si_pm4_state *pm4,
3525 unsigned raster_config,
3526 unsigned raster_config_1)
3527 {
3528 unsigned sh_per_se = MAX2(sctx->screen->b.info.max_sh_per_se, 1);
3529 unsigned num_se = MAX2(sctx->screen->b.info.max_se, 1);
3530 unsigned rb_mask = sctx->screen->b.info.enabled_rb_mask;
3531 unsigned num_rb = MIN2(sctx->screen->b.info.num_render_backends, 16);
3532 unsigned rb_per_pkr = MIN2(num_rb / num_se / sh_per_se, 2);
3533 unsigned rb_per_se = num_rb / num_se;
3534 unsigned se_mask[4];
3535 unsigned se;
3536
3537 se_mask[0] = ((1 << rb_per_se) - 1);
3538 se_mask[1] = (se_mask[0] << rb_per_se);
3539 se_mask[2] = (se_mask[1] << rb_per_se);
3540 se_mask[3] = (se_mask[2] << rb_per_se);
3541
3542 se_mask[0] &= rb_mask;
3543 se_mask[1] &= rb_mask;
3544 se_mask[2] &= rb_mask;
3545 se_mask[3] &= rb_mask;
3546
3547 assert(num_se == 1 || num_se == 2 || num_se == 4);
3548 assert(sh_per_se == 1 || sh_per_se == 2);
3549 assert(rb_per_pkr == 1 || rb_per_pkr == 2);
3550
3551 /* XXX: I can't figure out what the *_XSEL and *_YSEL
3552 * fields are for, so I'm leaving them as their default
3553 * values. */
3554
3555 for (se = 0; se < num_se; se++) {
3556 unsigned raster_config_se = raster_config;
3557 unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3558 unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3559 int idx = (se / 2) * 2;
3560
3561 if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3562 raster_config_se &= C_028350_SE_MAP;
3563
3564 if (!se_mask[idx]) {
3565 raster_config_se |=
3566 S_028350_SE_MAP(V_028350_RASTER_CONFIG_SE_MAP_3);
3567 } else {
3568 raster_config_se |=
3569 S_028350_SE_MAP(V_028350_RASTER_CONFIG_SE_MAP_0);
3570 }
3571 }
3572
3573 pkr0_mask &= rb_mask;
3574 pkr1_mask &= rb_mask;
3575 if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3576 raster_config_se &= C_028350_PKR_MAP;
3577
3578 if (!pkr0_mask) {
3579 raster_config_se |=
3580 S_028350_PKR_MAP(V_028350_RASTER_CONFIG_PKR_MAP_3);
3581 } else {
3582 raster_config_se |=
3583 S_028350_PKR_MAP(V_028350_RASTER_CONFIG_PKR_MAP_0);
3584 }
3585 }
3586
3587 if (rb_per_se >= 2) {
3588 unsigned rb0_mask = 1 << (se * rb_per_se);
3589 unsigned rb1_mask = rb0_mask << 1;
3590
3591 rb0_mask &= rb_mask;
3592 rb1_mask &= rb_mask;
3593 if (!rb0_mask || !rb1_mask) {
3594 raster_config_se &= C_028350_RB_MAP_PKR0;
3595
3596 if (!rb0_mask) {
3597 raster_config_se |=
3598 S_028350_RB_MAP_PKR0(V_028350_RASTER_CONFIG_RB_MAP_3);
3599 } else {
3600 raster_config_se |=
3601 S_028350_RB_MAP_PKR0(V_028350_RASTER_CONFIG_RB_MAP_0);
3602 }
3603 }
3604
3605 if (rb_per_se > 2) {
3606 rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3607 rb1_mask = rb0_mask << 1;
3608 rb0_mask &= rb_mask;
3609 rb1_mask &= rb_mask;
3610 if (!rb0_mask || !rb1_mask) {
3611 raster_config_se &= C_028350_RB_MAP_PKR1;
3612
3613 if (!rb0_mask) {
3614 raster_config_se |=
3615 S_028350_RB_MAP_PKR1(V_028350_RASTER_CONFIG_RB_MAP_3);
3616 } else {
3617 raster_config_se |=
3618 S_028350_RB_MAP_PKR1(V_028350_RASTER_CONFIG_RB_MAP_0);
3619 }
3620 }
3621 }
3622 }
3623
3624 /* GRBM_GFX_INDEX has a different offset on SI and CI+ */
3625 if (sctx->b.chip_class < CIK)
3626 si_pm4_set_reg(pm4, GRBM_GFX_INDEX,
3627 SE_INDEX(se) | SH_BROADCAST_WRITES |
3628 INSTANCE_BROADCAST_WRITES);
3629 else
3630 si_pm4_set_reg(pm4, R_030800_GRBM_GFX_INDEX,
3631 S_030800_SE_INDEX(se) | S_030800_SH_BROADCAST_WRITES(1) |
3632 S_030800_INSTANCE_BROADCAST_WRITES(1));
3633 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, raster_config_se);
3634 }
3635
3636 /* GRBM_GFX_INDEX has a different offset on SI and CI+ */
3637 if (sctx->b.chip_class < CIK)
3638 si_pm4_set_reg(pm4, GRBM_GFX_INDEX,
3639 SE_BROADCAST_WRITES | SH_BROADCAST_WRITES |
3640 INSTANCE_BROADCAST_WRITES);
3641 else {
3642 si_pm4_set_reg(pm4, R_030800_GRBM_GFX_INDEX,
3643 S_030800_SE_BROADCAST_WRITES(1) | S_030800_SH_BROADCAST_WRITES(1) |
3644 S_030800_INSTANCE_BROADCAST_WRITES(1));
3645
3646 if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3647 (!se_mask[2] && !se_mask[3]))) {
3648 raster_config_1 &= C_028354_SE_PAIR_MAP;
3649
3650 if (!se_mask[0] && !se_mask[1]) {
3651 raster_config_1 |=
3652 S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_3);
3653 } else {
3654 raster_config_1 |=
3655 S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_0);
3656 }
3657 }
3658
3659 si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1, raster_config_1);
3660 }
3661 }
3662
3663 static void si_init_config(struct si_context *sctx)
3664 {
3665 struct si_screen *sscreen = sctx->screen;
3666 unsigned num_rb = MIN2(sctx->screen->b.info.num_render_backends, 16);
3667 unsigned rb_mask = sctx->screen->b.info.enabled_rb_mask;
3668 unsigned raster_config, raster_config_1;
3669 uint64_t border_color_va = sctx->border_color_buffer->gpu_address;
3670 struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
3671 int i;
3672
3673 if (!pm4)
3674 return;
3675
3676 si_pm4_cmd_begin(pm4, PKT3_CONTEXT_CONTROL);
3677 si_pm4_cmd_add(pm4, CONTEXT_CONTROL_LOAD_ENABLE(1));
3678 si_pm4_cmd_add(pm4, CONTEXT_CONTROL_SHADOW_ENABLE(1));
3679 si_pm4_cmd_end(pm4, false);
3680
3681 si_pm4_set_reg(pm4, R_028A18_VGT_HOS_MAX_TESS_LEVEL, fui(64));
3682 si_pm4_set_reg(pm4, R_028A1C_VGT_HOS_MIN_TESS_LEVEL, fui(0));
3683
3684 /* FIXME calculate these values somehow ??? */
3685 si_pm4_set_reg(pm4, R_028A54_VGT_GS_PER_ES, SI_GS_PER_ES);
3686 si_pm4_set_reg(pm4, R_028A58_VGT_ES_PER_GS, 0x40);
3687 si_pm4_set_reg(pm4, R_028A5C_VGT_GS_PER_VS, 0x2);
3688
3689 si_pm4_set_reg(pm4, R_028A8C_VGT_PRIMITIVEID_RESET, 0x0);
3690 si_pm4_set_reg(pm4, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
3691
3692 si_pm4_set_reg(pm4, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0x0);
3693 si_pm4_set_reg(pm4, R_028AB8_VGT_VTX_CNT_EN, 0x0);
3694 if (sctx->b.chip_class < CIK)
3695 si_pm4_set_reg(pm4, R_008A14_PA_CL_ENHANCE, S_008A14_NUM_CLIP_SEQ(3) |
3696 S_008A14_CLIP_VTX_REORDER_ENA(1));
3697
3698 si_pm4_set_reg(pm4, R_028BD4_PA_SC_CENTROID_PRIORITY_0, 0x76543210);
3699 si_pm4_set_reg(pm4, R_028BD8_PA_SC_CENTROID_PRIORITY_1, 0xfedcba98);
3700
3701 si_pm4_set_reg(pm4, R_02882C_PA_SU_PRIM_FILTER_CNTL, 0);
3702
3703 for (i = 0; i < 16; i++) {
3704 si_pm4_set_reg(pm4, R_0282D0_PA_SC_VPORT_ZMIN_0 + i*8, 0);
3705 si_pm4_set_reg(pm4, R_0282D4_PA_SC_VPORT_ZMAX_0 + i*8, fui(1.0));
3706 }
3707
3708 switch (sctx->screen->b.family) {
3709 case CHIP_TAHITI:
3710 case CHIP_PITCAIRN:
3711 raster_config = 0x2a00126a;
3712 raster_config_1 = 0x00000000;
3713 break;
3714 case CHIP_VERDE:
3715 raster_config = 0x0000124a;
3716 raster_config_1 = 0x00000000;
3717 break;
3718 case CHIP_OLAND:
3719 raster_config = 0x00000082;
3720 raster_config_1 = 0x00000000;
3721 break;
3722 case CHIP_HAINAN:
3723 raster_config = 0x00000000;
3724 raster_config_1 = 0x00000000;
3725 break;
3726 case CHIP_BONAIRE:
3727 raster_config = 0x16000012;
3728 raster_config_1 = 0x00000000;
3729 break;
3730 case CHIP_HAWAII:
3731 raster_config = 0x3a00161a;
3732 raster_config_1 = 0x0000002e;
3733 break;
3734 case CHIP_FIJI:
3735 if (sscreen->b.info.cik_macrotile_mode_array[0] == 0x000000e8) {
3736 /* old kernels with old tiling config */
3737 raster_config = 0x16000012;
3738 raster_config_1 = 0x0000002a;
3739 } else {
3740 raster_config = 0x3a00161a;
3741 raster_config_1 = 0x0000002e;
3742 }
3743 break;
3744 case CHIP_POLARIS10:
3745 raster_config = 0x16000012;
3746 raster_config_1 = 0x0000002a;
3747 break;
3748 case CHIP_POLARIS11:
3749 raster_config = 0x16000012;
3750 raster_config_1 = 0x00000000;
3751 break;
3752 case CHIP_TONGA:
3753 raster_config = 0x16000012;
3754 raster_config_1 = 0x0000002a;
3755 break;
3756 case CHIP_ICELAND:
3757 if (num_rb == 1)
3758 raster_config = 0x00000000;
3759 else
3760 raster_config = 0x00000002;
3761 raster_config_1 = 0x00000000;
3762 break;
3763 case CHIP_CARRIZO:
3764 raster_config = 0x00000002;
3765 raster_config_1 = 0x00000000;
3766 break;
3767 case CHIP_KAVERI:
3768 /* KV should be 0x00000002, but that causes problems with radeon */
3769 raster_config = 0x00000000; /* 0x00000002 */
3770 raster_config_1 = 0x00000000;
3771 break;
3772 case CHIP_KABINI:
3773 case CHIP_MULLINS:
3774 case CHIP_STONEY:
3775 raster_config = 0x00000000;
3776 raster_config_1 = 0x00000000;
3777 break;
3778 default:
3779 fprintf(stderr,
3780 "radeonsi: Unknown GPU, using 0 for raster_config\n");
3781 raster_config = 0x00000000;
3782 raster_config_1 = 0x00000000;
3783 break;
3784 }
3785
3786 /* Always use the default config when all backends are enabled
3787 * (or when we failed to determine the enabled backends).
3788 */
3789 if (!rb_mask || util_bitcount(rb_mask) >= num_rb) {
3790 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG,
3791 raster_config);
3792 if (sctx->b.chip_class >= CIK)
3793 si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1,
3794 raster_config_1);
3795 } else {
3796 si_write_harvested_raster_configs(sctx, pm4, raster_config, raster_config_1);
3797 }
3798
3799 si_pm4_set_reg(pm4, R_028204_PA_SC_WINDOW_SCISSOR_TL, S_028204_WINDOW_OFFSET_DISABLE(1));
3800 si_pm4_set_reg(pm4, R_028240_PA_SC_GENERIC_SCISSOR_TL, S_028240_WINDOW_OFFSET_DISABLE(1));
3801 si_pm4_set_reg(pm4, R_028244_PA_SC_GENERIC_SCISSOR_BR,
3802 S_028244_BR_X(16384) | S_028244_BR_Y(16384));
3803 si_pm4_set_reg(pm4, R_028030_PA_SC_SCREEN_SCISSOR_TL, 0);
3804 si_pm4_set_reg(pm4, R_028034_PA_SC_SCREEN_SCISSOR_BR,
3805 S_028034_BR_X(16384) | S_028034_BR_Y(16384));
3806
3807 si_pm4_set_reg(pm4, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
3808 si_pm4_set_reg(pm4, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
3809 /* PA_SU_HARDWARE_SCREEN_OFFSET must be 0 due to hw bug on SI */
3810 si_pm4_set_reg(pm4, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0);
3811 si_pm4_set_reg(pm4, R_028820_PA_CL_NANINF_CNTL, 0);
3812 si_pm4_set_reg(pm4, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 0x0);
3813 si_pm4_set_reg(pm4, R_028AC4_DB_SRESULTS_COMPARE_STATE1, 0x0);
3814 si_pm4_set_reg(pm4, R_028AC8_DB_PRELOAD_CONTROL, 0x0);
3815 si_pm4_set_reg(pm4, R_02800C_DB_RENDER_OVERRIDE,
3816 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE) |
3817 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE));
3818
3819 si_pm4_set_reg(pm4, R_028400_VGT_MAX_VTX_INDX, ~0);
3820 si_pm4_set_reg(pm4, R_028404_VGT_MIN_VTX_INDX, 0);
3821 si_pm4_set_reg(pm4, R_028408_VGT_INDX_OFFSET, 0);
3822
3823 if (sctx->b.chip_class >= CIK) {
3824 si_pm4_set_reg(pm4, R_00B41C_SPI_SHADER_PGM_RSRC3_HS, 0);
3825 si_pm4_set_reg(pm4, R_00B31C_SPI_SHADER_PGM_RSRC3_ES, S_00B31C_CU_EN(0xffff));
3826 si_pm4_set_reg(pm4, R_00B21C_SPI_SHADER_PGM_RSRC3_GS, S_00B21C_CU_EN(0xffff));
3827
3828 if (sscreen->b.info.num_good_compute_units /
3829 (sscreen->b.info.max_se * sscreen->b.info.max_sh_per_se) <= 4) {
3830 /* Too few available compute units per SH. Disallowing
3831 * VS to run on CU0 could hurt us more than late VS
3832 * allocation would help.
3833 *
3834 * LATE_ALLOC_VS = 2 is the highest safe number.
3835 */
3836 si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS, S_00B51C_CU_EN(0xffff));
3837 si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS, S_00B118_CU_EN(0xffff));
3838 si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(2));
3839 } else {
3840 /* Set LATE_ALLOC_VS == 31. It should be less than
3841 * the number of scratch waves. Limitations:
3842 * - VS can't execute on CU0.
3843 * - If HS writes outputs to LDS, LS can't execute on CU0.
3844 */
3845 si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS, S_00B51C_CU_EN(0xfffe));
3846 si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS, S_00B118_CU_EN(0xfffe));
3847 si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(31));
3848 }
3849
3850 si_pm4_set_reg(pm4, R_00B01C_SPI_SHADER_PGM_RSRC3_PS, S_00B01C_CU_EN(0xffff));
3851 }
3852
3853 if (sctx->b.chip_class >= VI) {
3854 unsigned vgt_tess_distribution;
3855
3856 si_pm4_set_reg(pm4, R_028424_CB_DCC_CONTROL,
3857 S_028424_OVERWRITE_COMBINER_MRT_SHARING_DISABLE(1) |
3858 S_028424_OVERWRITE_COMBINER_WATERMARK(4));
3859 si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 30);
3860 si_pm4_set_reg(pm4, R_028C5C_VGT_OUT_DEALLOC_CNTL, 32);
3861
3862 vgt_tess_distribution =
3863 S_028B50_ACCUM_ISOLINE(32) |
3864 S_028B50_ACCUM_TRI(11) |
3865 S_028B50_ACCUM_QUAD(11) |
3866 S_028B50_DONUT_SPLIT(16);
3867
3868 /* Testing with Unigine Heaven extreme tesselation yielded best results
3869 * with TRAP_SPLIT = 3.
3870 */
3871 if (sctx->b.family == CHIP_FIJI ||
3872 sctx->b.family >= CHIP_POLARIS10)
3873 vgt_tess_distribution |= S_028B50_TRAP_SPLIT(3);
3874
3875 si_pm4_set_reg(pm4, R_028B50_VGT_TESS_DISTRIBUTION, vgt_tess_distribution);
3876 }
3877
3878 if (sctx->b.family == CHIP_STONEY)
3879 si_pm4_set_reg(pm4, R_028C40_PA_SC_SHADER_CONTROL, 0);
3880
3881 si_pm4_set_reg(pm4, R_028080_TA_BC_BASE_ADDR, border_color_va >> 8);
3882 if (sctx->b.chip_class >= CIK)
3883 si_pm4_set_reg(pm4, R_028084_TA_BC_BASE_ADDR_HI, border_color_va >> 40);
3884 si_pm4_add_bo(pm4, sctx->border_color_buffer, RADEON_USAGE_READ,
3885 RADEON_PRIO_BORDER_COLORS);
3886
3887 si_pm4_upload_indirect_buffer(sctx, pm4);
3888 sctx->init_config = pm4;
3889 }