projects / mesa.git / blob
? search:


7169614143a054b2b91864ba6a1cbd39f2b06e2e
[mesa.git] / src / gallium / drivers / r600 / evergreen_state.c
1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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 #include "r600_formats.h"
24 #include "r600_shader.h"
25 #include "evergreend.h"
26
27 #include "pipe/p_shader_tokens.h"
28 #include "util/u_pack_color.h"
29 #include "util/u_memory.h"
30 #include "util/u_framebuffer.h"
31 #include "util/u_dual_blend.h"
32 #include "evergreen_compute.h"
33 #include "util/u_math.h"
34
35 static INLINE unsigned evergreen_array_mode(unsigned mode)
36 {
37 switch (mode) {
38 case RADEON_SURF_MODE_LINEAR_ALIGNED: return V_028C70_ARRAY_LINEAR_ALIGNED;
39 break;
40 case RADEON_SURF_MODE_1D: return V_028C70_ARRAY_1D_TILED_THIN1;
41 break;
42 case RADEON_SURF_MODE_2D: return V_028C70_ARRAY_2D_TILED_THIN1;
43 default:
44 case RADEON_SURF_MODE_LINEAR: return V_028C70_ARRAY_LINEAR_GENERAL;
45 }
46 }
47
48 static uint32_t eg_num_banks(uint32_t nbanks)
49 {
50 switch (nbanks) {
51 case 2:
52 return 0;
53 case 4:
54 return 1;
55 case 8:
56 default:
57 return 2;
58 case 16:
59 return 3;
60 }
61 }
62
63
64 static unsigned eg_tile_split(unsigned tile_split)
65 {
66 switch (tile_split) {
67 case 64: tile_split = 0; break;
68 case 128: tile_split = 1; break;
69 case 256: tile_split = 2; break;
70 case 512: tile_split = 3; break;
71 default:
72 case 1024: tile_split = 4; break;
73 case 2048: tile_split = 5; break;
74 case 4096: tile_split = 6; break;
75 }
76 return tile_split;
77 }
78
79 static unsigned eg_macro_tile_aspect(unsigned macro_tile_aspect)
80 {
81 switch (macro_tile_aspect) {
82 default:
83 case 1: macro_tile_aspect = 0; break;
84 case 2: macro_tile_aspect = 1; break;
85 case 4: macro_tile_aspect = 2; break;
86 case 8: macro_tile_aspect = 3; break;
87 }
88 return macro_tile_aspect;
89 }
90
91 static unsigned eg_bank_wh(unsigned bankwh)
92 {
93 switch (bankwh) {
94 default:
95 case 1: bankwh = 0; break;
96 case 2: bankwh = 1; break;
97 case 4: bankwh = 2; break;
98 case 8: bankwh = 3; break;
99 }
100 return bankwh;
101 }
102
103 static uint32_t r600_translate_blend_function(int blend_func)
104 {
105 switch (blend_func) {
106 case PIPE_BLEND_ADD:
107 return V_028780_COMB_DST_PLUS_SRC;
108 case PIPE_BLEND_SUBTRACT:
109 return V_028780_COMB_SRC_MINUS_DST;
110 case PIPE_BLEND_REVERSE_SUBTRACT:
111 return V_028780_COMB_DST_MINUS_SRC;
112 case PIPE_BLEND_MIN:
113 return V_028780_COMB_MIN_DST_SRC;
114 case PIPE_BLEND_MAX:
115 return V_028780_COMB_MAX_DST_SRC;
116 default:
117 R600_ERR("Unknown blend function %d\n", blend_func);
118 assert(0);
119 break;
120 }
121 return 0;
122 }
123
124 static uint32_t r600_translate_blend_factor(int blend_fact)
125 {
126 switch (blend_fact) {
127 case PIPE_BLENDFACTOR_ONE:
128 return V_028780_BLEND_ONE;
129 case PIPE_BLENDFACTOR_SRC_COLOR:
130 return V_028780_BLEND_SRC_COLOR;
131 case PIPE_BLENDFACTOR_SRC_ALPHA:
132 return V_028780_BLEND_SRC_ALPHA;
133 case PIPE_BLENDFACTOR_DST_ALPHA:
134 return V_028780_BLEND_DST_ALPHA;
135 case PIPE_BLENDFACTOR_DST_COLOR:
136 return V_028780_BLEND_DST_COLOR;
137 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
138 return V_028780_BLEND_SRC_ALPHA_SATURATE;
139 case PIPE_BLENDFACTOR_CONST_COLOR:
140 return V_028780_BLEND_CONST_COLOR;
141 case PIPE_BLENDFACTOR_CONST_ALPHA:
142 return V_028780_BLEND_CONST_ALPHA;
143 case PIPE_BLENDFACTOR_ZERO:
144 return V_028780_BLEND_ZERO;
145 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
146 return V_028780_BLEND_ONE_MINUS_SRC_COLOR;
147 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
148 return V_028780_BLEND_ONE_MINUS_SRC_ALPHA;
149 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
150 return V_028780_BLEND_ONE_MINUS_DST_ALPHA;
151 case PIPE_BLENDFACTOR_INV_DST_COLOR:
152 return V_028780_BLEND_ONE_MINUS_DST_COLOR;
153 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
154 return V_028780_BLEND_ONE_MINUS_CONST_COLOR;
155 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
156 return V_028780_BLEND_ONE_MINUS_CONST_ALPHA;
157 case PIPE_BLENDFACTOR_SRC1_COLOR:
158 return V_028780_BLEND_SRC1_COLOR;
159 case PIPE_BLENDFACTOR_SRC1_ALPHA:
160 return V_028780_BLEND_SRC1_ALPHA;
161 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
162 return V_028780_BLEND_INV_SRC1_COLOR;
163 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
164 return V_028780_BLEND_INV_SRC1_ALPHA;
165 default:
166 R600_ERR("Bad blend factor %d not supported!\n", blend_fact);
167 assert(0);
168 break;
169 }
170 return 0;
171 }
172
173 static unsigned r600_tex_dim(unsigned dim, unsigned nr_samples)
174 {
175 switch (dim) {
176 default:
177 case PIPE_TEXTURE_1D:
178 return V_030000_SQ_TEX_DIM_1D;
179 case PIPE_TEXTURE_1D_ARRAY:
180 return V_030000_SQ_TEX_DIM_1D_ARRAY;
181 case PIPE_TEXTURE_2D:
182 case PIPE_TEXTURE_RECT:
183 return nr_samples > 1 ? V_030000_SQ_TEX_DIM_2D_MSAA :
184 V_030000_SQ_TEX_DIM_2D;
185 case PIPE_TEXTURE_2D_ARRAY:
186 return nr_samples > 1 ? V_030000_SQ_TEX_DIM_2D_ARRAY_MSAA :
187 V_030000_SQ_TEX_DIM_2D_ARRAY;
188 case PIPE_TEXTURE_3D:
189 return V_030000_SQ_TEX_DIM_3D;
190 case PIPE_TEXTURE_CUBE:
191 case PIPE_TEXTURE_CUBE_ARRAY:
192 return V_030000_SQ_TEX_DIM_CUBEMAP;
193 }
194 }
195
196 static uint32_t r600_translate_dbformat(enum pipe_format format)
197 {
198 switch (format) {
199 case PIPE_FORMAT_Z16_UNORM:
200 return V_028040_Z_16;
201 case PIPE_FORMAT_Z24X8_UNORM:
202 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
203 case PIPE_FORMAT_X8Z24_UNORM:
204 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
205 return V_028040_Z_24;
206 case PIPE_FORMAT_Z32_FLOAT:
207 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
208 return V_028040_Z_32_FLOAT;
209 default:
210 return ~0U;
211 }
212 }
213
214 static uint32_t r600_translate_colorswap(enum pipe_format format)
215 {
216 switch (format) {
217 /* 8-bit buffers. */
218 case PIPE_FORMAT_L4A4_UNORM:
219 case PIPE_FORMAT_A4R4_UNORM:
220 return V_028C70_SWAP_ALT;
221
222 case PIPE_FORMAT_A8_UNORM:
223 case PIPE_FORMAT_A8_SNORM:
224 case PIPE_FORMAT_A8_UINT:
225 case PIPE_FORMAT_A8_SINT:
226 case PIPE_FORMAT_A16_UNORM:
227 case PIPE_FORMAT_A16_SNORM:
228 case PIPE_FORMAT_A16_UINT:
229 case PIPE_FORMAT_A16_SINT:
230 case PIPE_FORMAT_A16_FLOAT:
231 case PIPE_FORMAT_A32_UINT:
232 case PIPE_FORMAT_A32_SINT:
233 case PIPE_FORMAT_A32_FLOAT:
234 case PIPE_FORMAT_R4A4_UNORM:
235 return V_028C70_SWAP_ALT_REV;
236 case PIPE_FORMAT_I8_UNORM:
237 case PIPE_FORMAT_I8_SNORM:
238 case PIPE_FORMAT_I8_UINT:
239 case PIPE_FORMAT_I8_SINT:
240 case PIPE_FORMAT_I16_UNORM:
241 case PIPE_FORMAT_I16_SNORM:
242 case PIPE_FORMAT_I16_UINT:
243 case PIPE_FORMAT_I16_SINT:
244 case PIPE_FORMAT_I16_FLOAT:
245 case PIPE_FORMAT_I32_UINT:
246 case PIPE_FORMAT_I32_SINT:
247 case PIPE_FORMAT_I32_FLOAT:
248 case PIPE_FORMAT_L8_UNORM:
249 case PIPE_FORMAT_L8_SNORM:
250 case PIPE_FORMAT_L8_UINT:
251 case PIPE_FORMAT_L8_SINT:
252 case PIPE_FORMAT_L8_SRGB:
253 case PIPE_FORMAT_L16_UNORM:
254 case PIPE_FORMAT_L16_SNORM:
255 case PIPE_FORMAT_L16_UINT:
256 case PIPE_FORMAT_L16_SINT:
257 case PIPE_FORMAT_L16_FLOAT:
258 case PIPE_FORMAT_L32_UINT:
259 case PIPE_FORMAT_L32_SINT:
260 case PIPE_FORMAT_L32_FLOAT:
261 case PIPE_FORMAT_R8_UNORM:
262 case PIPE_FORMAT_R8_SNORM:
263 case PIPE_FORMAT_R8_UINT:
264 case PIPE_FORMAT_R8_SINT:
265 return V_028C70_SWAP_STD;
266
267 /* 16-bit buffers. */
268 case PIPE_FORMAT_B5G6R5_UNORM:
269 return V_028C70_SWAP_STD_REV;
270
271 case PIPE_FORMAT_B5G5R5A1_UNORM:
272 case PIPE_FORMAT_B5G5R5X1_UNORM:
273 return V_028C70_SWAP_ALT;
274
275 case PIPE_FORMAT_B4G4R4A4_UNORM:
276 case PIPE_FORMAT_B4G4R4X4_UNORM:
277 return V_028C70_SWAP_ALT;
278
279 case PIPE_FORMAT_Z16_UNORM:
280 return V_028C70_SWAP_STD;
281
282 case PIPE_FORMAT_L8A8_UNORM:
283 case PIPE_FORMAT_L8A8_SNORM:
284 case PIPE_FORMAT_L8A8_UINT:
285 case PIPE_FORMAT_L8A8_SINT:
286 case PIPE_FORMAT_L8A8_SRGB:
287 case PIPE_FORMAT_L16A16_UNORM:
288 case PIPE_FORMAT_L16A16_SNORM:
289 case PIPE_FORMAT_L16A16_UINT:
290 case PIPE_FORMAT_L16A16_SINT:
291 case PIPE_FORMAT_L16A16_FLOAT:
292 case PIPE_FORMAT_L32A32_UINT:
293 case PIPE_FORMAT_L32A32_SINT:
294 case PIPE_FORMAT_L32A32_FLOAT:
295 case PIPE_FORMAT_R8A8_UNORM:
296 case PIPE_FORMAT_R8A8_SNORM:
297 case PIPE_FORMAT_R8A8_UINT:
298 case PIPE_FORMAT_R8A8_SINT:
299 case PIPE_FORMAT_R16A16_UNORM:
300 case PIPE_FORMAT_R16A16_SNORM:
301 case PIPE_FORMAT_R16A16_UINT:
302 case PIPE_FORMAT_R16A16_SINT:
303 case PIPE_FORMAT_R16A16_FLOAT:
304 case PIPE_FORMAT_R32A32_UINT:
305 case PIPE_FORMAT_R32A32_SINT:
306 case PIPE_FORMAT_R32A32_FLOAT:
307 return V_028C70_SWAP_ALT;
308 case PIPE_FORMAT_R8G8_UNORM:
309 case PIPE_FORMAT_R8G8_SNORM:
310 case PIPE_FORMAT_R8G8_UINT:
311 case PIPE_FORMAT_R8G8_SINT:
312 return V_028C70_SWAP_STD;
313
314 case PIPE_FORMAT_R16_UNORM:
315 case PIPE_FORMAT_R16_SNORM:
316 case PIPE_FORMAT_R16_UINT:
317 case PIPE_FORMAT_R16_SINT:
318 case PIPE_FORMAT_R16_FLOAT:
319 return V_028C70_SWAP_STD;
320
321 /* 32-bit buffers. */
322 case PIPE_FORMAT_A8B8G8R8_SRGB:
323 return V_028C70_SWAP_STD_REV;
324 case PIPE_FORMAT_B8G8R8A8_SRGB:
325 return V_028C70_SWAP_ALT;
326
327 case PIPE_FORMAT_B8G8R8A8_UNORM:
328 case PIPE_FORMAT_B8G8R8X8_UNORM:
329 return V_028C70_SWAP_ALT;
330
331 case PIPE_FORMAT_A8R8G8B8_UNORM:
332 case PIPE_FORMAT_X8R8G8B8_UNORM:
333 return V_028C70_SWAP_ALT_REV;
334 case PIPE_FORMAT_R8G8B8A8_SNORM:
335 case PIPE_FORMAT_R8G8B8A8_UNORM:
336 case PIPE_FORMAT_R8G8B8A8_SINT:
337 case PIPE_FORMAT_R8G8B8A8_UINT:
338 case PIPE_FORMAT_R8G8B8X8_UNORM:
339 case PIPE_FORMAT_R8G8B8X8_SNORM:
340 case PIPE_FORMAT_R8G8B8X8_SRGB:
341 case PIPE_FORMAT_R8G8B8X8_UINT:
342 case PIPE_FORMAT_R8G8B8X8_SINT:
343 return V_028C70_SWAP_STD;
344
345 case PIPE_FORMAT_A8B8G8R8_UNORM:
346 case PIPE_FORMAT_X8B8G8R8_UNORM:
347 /* case PIPE_FORMAT_R8SG8SB8UX8U_NORM: */
348 return V_028C70_SWAP_STD_REV;
349
350 case PIPE_FORMAT_Z24X8_UNORM:
351 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
352 return V_028C70_SWAP_STD;
353
354 case PIPE_FORMAT_X8Z24_UNORM:
355 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
356 return V_028C70_SWAP_STD_REV;
357
358 case PIPE_FORMAT_R10G10B10A2_UNORM:
359 case PIPE_FORMAT_R10G10B10X2_SNORM:
360 case PIPE_FORMAT_R10SG10SB10SA2U_NORM:
361 return V_028C70_SWAP_STD;
362
363 case PIPE_FORMAT_B10G10R10A2_UNORM:
364 case PIPE_FORMAT_B10G10R10A2_UINT:
365 case PIPE_FORMAT_B10G10R10X2_UNORM:
366 return V_028C70_SWAP_ALT;
367
368 case PIPE_FORMAT_R11G11B10_FLOAT:
369 case PIPE_FORMAT_R32_FLOAT:
370 case PIPE_FORMAT_R32_UINT:
371 case PIPE_FORMAT_R32_SINT:
372 case PIPE_FORMAT_Z32_FLOAT:
373 case PIPE_FORMAT_R16G16_FLOAT:
374 case PIPE_FORMAT_R16G16_UNORM:
375 case PIPE_FORMAT_R16G16_SNORM:
376 case PIPE_FORMAT_R16G16_UINT:
377 case PIPE_FORMAT_R16G16_SINT:
378 return V_028C70_SWAP_STD;
379
380 /* 64-bit buffers. */
381 case PIPE_FORMAT_R32G32_FLOAT:
382 case PIPE_FORMAT_R32G32_UINT:
383 case PIPE_FORMAT_R32G32_SINT:
384 case PIPE_FORMAT_R16G16B16A16_UNORM:
385 case PIPE_FORMAT_R16G16B16A16_SNORM:
386 case PIPE_FORMAT_R16G16B16A16_UINT:
387 case PIPE_FORMAT_R16G16B16A16_SINT:
388 case PIPE_FORMAT_R16G16B16A16_FLOAT:
389 case PIPE_FORMAT_R16G16B16X16_UNORM:
390 case PIPE_FORMAT_R16G16B16X16_SNORM:
391 case PIPE_FORMAT_R16G16B16X16_FLOAT:
392 case PIPE_FORMAT_R16G16B16X16_UINT:
393 case PIPE_FORMAT_R16G16B16X16_SINT:
394 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
395
396 /* 128-bit buffers. */
397 case PIPE_FORMAT_R32G32B32A32_FLOAT:
398 case PIPE_FORMAT_R32G32B32A32_SNORM:
399 case PIPE_FORMAT_R32G32B32A32_UNORM:
400 case PIPE_FORMAT_R32G32B32A32_SINT:
401 case PIPE_FORMAT_R32G32B32A32_UINT:
402 case PIPE_FORMAT_R32G32B32X32_FLOAT:
403 case PIPE_FORMAT_R32G32B32X32_UINT:
404 case PIPE_FORMAT_R32G32B32X32_SINT:
405 return V_028C70_SWAP_STD;
406 default:
407 R600_ERR("unsupported colorswap format %d\n", format);
408 return ~0U;
409 }
410 return ~0U;
411 }
412
413 static uint32_t r600_translate_colorformat(enum pipe_format format)
414 {
415 switch (format) {
416 /* 8-bit buffers. */
417 case PIPE_FORMAT_A8_UNORM:
418 case PIPE_FORMAT_A8_SNORM:
419 case PIPE_FORMAT_A8_UINT:
420 case PIPE_FORMAT_A8_SINT:
421 case PIPE_FORMAT_I8_UNORM:
422 case PIPE_FORMAT_I8_SNORM:
423 case PIPE_FORMAT_I8_UINT:
424 case PIPE_FORMAT_I8_SINT:
425 case PIPE_FORMAT_L8_UNORM:
426 case PIPE_FORMAT_L8_SNORM:
427 case PIPE_FORMAT_L8_UINT:
428 case PIPE_FORMAT_L8_SINT:
429 case PIPE_FORMAT_L8_SRGB:
430 case PIPE_FORMAT_R8_UNORM:
431 case PIPE_FORMAT_R8_SNORM:
432 case PIPE_FORMAT_R8_UINT:
433 case PIPE_FORMAT_R8_SINT:
434 return V_028C70_COLOR_8;
435
436 /* 16-bit buffers. */
437 case PIPE_FORMAT_B5G6R5_UNORM:
438 return V_028C70_COLOR_5_6_5;
439
440 case PIPE_FORMAT_B5G5R5A1_UNORM:
441 case PIPE_FORMAT_B5G5R5X1_UNORM:
442 return V_028C70_COLOR_1_5_5_5;
443
444 case PIPE_FORMAT_B4G4R4A4_UNORM:
445 case PIPE_FORMAT_B4G4R4X4_UNORM:
446 return V_028C70_COLOR_4_4_4_4;
447
448 case PIPE_FORMAT_Z16_UNORM:
449 return V_028C70_COLOR_16;
450
451 case PIPE_FORMAT_L8A8_UNORM:
452 case PIPE_FORMAT_L8A8_SNORM:
453 case PIPE_FORMAT_L8A8_UINT:
454 case PIPE_FORMAT_L8A8_SINT:
455 case PIPE_FORMAT_L8A8_SRGB:
456 case PIPE_FORMAT_R8G8_UNORM:
457 case PIPE_FORMAT_R8G8_SNORM:
458 case PIPE_FORMAT_R8G8_UINT:
459 case PIPE_FORMAT_R8G8_SINT:
460 case PIPE_FORMAT_R8A8_UNORM:
461 case PIPE_FORMAT_R8A8_SNORM:
462 case PIPE_FORMAT_R8A8_UINT:
463 case PIPE_FORMAT_R8A8_SINT:
464 return V_028C70_COLOR_8_8;
465
466 case PIPE_FORMAT_R16_UNORM:
467 case PIPE_FORMAT_R16_SNORM:
468 case PIPE_FORMAT_R16_UINT:
469 case PIPE_FORMAT_R16_SINT:
470 case PIPE_FORMAT_A16_UNORM:
471 case PIPE_FORMAT_A16_SNORM:
472 case PIPE_FORMAT_A16_UINT:
473 case PIPE_FORMAT_A16_SINT:
474 case PIPE_FORMAT_L16_UNORM:
475 case PIPE_FORMAT_L16_SNORM:
476 case PIPE_FORMAT_L16_UINT:
477 case PIPE_FORMAT_L16_SINT:
478 case PIPE_FORMAT_I16_UNORM:
479 case PIPE_FORMAT_I16_SNORM:
480 case PIPE_FORMAT_I16_UINT:
481 case PIPE_FORMAT_I16_SINT:
482 return V_028C70_COLOR_16;
483
484 case PIPE_FORMAT_R16_FLOAT:
485 case PIPE_FORMAT_A16_FLOAT:
486 case PIPE_FORMAT_L16_FLOAT:
487 case PIPE_FORMAT_I16_FLOAT:
488 return V_028C70_COLOR_16_FLOAT;
489
490 /* 32-bit buffers. */
491 case PIPE_FORMAT_A8B8G8R8_SRGB:
492 case PIPE_FORMAT_A8B8G8R8_UNORM:
493 case PIPE_FORMAT_A8R8G8B8_UNORM:
494 case PIPE_FORMAT_B8G8R8A8_SRGB:
495 case PIPE_FORMAT_B8G8R8A8_UNORM:
496 case PIPE_FORMAT_B8G8R8X8_UNORM:
497 case PIPE_FORMAT_R8G8B8A8_SNORM:
498 case PIPE_FORMAT_R8G8B8A8_UNORM:
499 case PIPE_FORMAT_R8G8B8X8_UNORM:
500 case PIPE_FORMAT_R8G8B8X8_SNORM:
501 case PIPE_FORMAT_R8G8B8X8_SRGB:
502 case PIPE_FORMAT_R8G8B8X8_UINT:
503 case PIPE_FORMAT_R8G8B8X8_SINT:
504 case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
505 case PIPE_FORMAT_X8B8G8R8_UNORM:
506 case PIPE_FORMAT_X8R8G8B8_UNORM:
507 case PIPE_FORMAT_R8G8B8_UNORM:
508 case PIPE_FORMAT_R8G8B8A8_SINT:
509 case PIPE_FORMAT_R8G8B8A8_UINT:
510 return V_028C70_COLOR_8_8_8_8;
511
512 case PIPE_FORMAT_R10G10B10A2_UNORM:
513 case PIPE_FORMAT_R10G10B10X2_SNORM:
514 case PIPE_FORMAT_B10G10R10A2_UNORM:
515 case PIPE_FORMAT_B10G10R10A2_UINT:
516 case PIPE_FORMAT_B10G10R10X2_UNORM:
517 case PIPE_FORMAT_R10SG10SB10SA2U_NORM:
518 return V_028C70_COLOR_2_10_10_10;
519
520 case PIPE_FORMAT_Z24X8_UNORM:
521 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
522 return V_028C70_COLOR_8_24;
523
524 case PIPE_FORMAT_X8Z24_UNORM:
525 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
526 return V_028C70_COLOR_24_8;
527
528 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
529 return V_028C70_COLOR_X24_8_32_FLOAT;
530
531 case PIPE_FORMAT_R32_UINT:
532 case PIPE_FORMAT_R32_SINT:
533 case PIPE_FORMAT_A32_UINT:
534 case PIPE_FORMAT_A32_SINT:
535 case PIPE_FORMAT_L32_UINT:
536 case PIPE_FORMAT_L32_SINT:
537 case PIPE_FORMAT_I32_UINT:
538 case PIPE_FORMAT_I32_SINT:
539 return V_028C70_COLOR_32;
540
541 case PIPE_FORMAT_R32_FLOAT:
542 case PIPE_FORMAT_A32_FLOAT:
543 case PIPE_FORMAT_L32_FLOAT:
544 case PIPE_FORMAT_I32_FLOAT:
545 case PIPE_FORMAT_Z32_FLOAT:
546 return V_028C70_COLOR_32_FLOAT;
547
548 case PIPE_FORMAT_R16G16_FLOAT:
549 case PIPE_FORMAT_L16A16_FLOAT:
550 case PIPE_FORMAT_R16A16_FLOAT:
551 return V_028C70_COLOR_16_16_FLOAT;
552
553 case PIPE_FORMAT_R16G16_UNORM:
554 case PIPE_FORMAT_R16G16_SNORM:
555 case PIPE_FORMAT_R16G16_UINT:
556 case PIPE_FORMAT_R16G16_SINT:
557 case PIPE_FORMAT_L16A16_UNORM:
558 case PIPE_FORMAT_L16A16_SNORM:
559 case PIPE_FORMAT_L16A16_UINT:
560 case PIPE_FORMAT_L16A16_SINT:
561 case PIPE_FORMAT_R16A16_UNORM:
562 case PIPE_FORMAT_R16A16_SNORM:
563 case PIPE_FORMAT_R16A16_UINT:
564 case PIPE_FORMAT_R16A16_SINT:
565 return V_028C70_COLOR_16_16;
566
567 case PIPE_FORMAT_R11G11B10_FLOAT:
568 return V_028C70_COLOR_10_11_11_FLOAT;
569
570 /* 64-bit buffers. */
571 case PIPE_FORMAT_R16G16B16A16_UINT:
572 case PIPE_FORMAT_R16G16B16A16_SINT:
573 case PIPE_FORMAT_R16G16B16A16_UNORM:
574 case PIPE_FORMAT_R16G16B16A16_SNORM:
575 case PIPE_FORMAT_R16G16B16X16_UNORM:
576 case PIPE_FORMAT_R16G16B16X16_SNORM:
577 case PIPE_FORMAT_R16G16B16X16_UINT:
578 case PIPE_FORMAT_R16G16B16X16_SINT:
579 return V_028C70_COLOR_16_16_16_16;
580
581 case PIPE_FORMAT_R16G16B16A16_FLOAT:
582 case PIPE_FORMAT_R16G16B16X16_FLOAT:
583 return V_028C70_COLOR_16_16_16_16_FLOAT;
584
585 case PIPE_FORMAT_R32G32_FLOAT:
586 case PIPE_FORMAT_L32A32_FLOAT:
587 case PIPE_FORMAT_R32A32_FLOAT:
588 return V_028C70_COLOR_32_32_FLOAT;
589
590 case PIPE_FORMAT_R32G32_SINT:
591 case PIPE_FORMAT_R32G32_UINT:
592 case PIPE_FORMAT_L32A32_UINT:
593 case PIPE_FORMAT_L32A32_SINT:
594 return V_028C70_COLOR_32_32;
595
596 /* 128-bit buffers. */
597 case PIPE_FORMAT_R32G32B32A32_SNORM:
598 case PIPE_FORMAT_R32G32B32A32_UNORM:
599 case PIPE_FORMAT_R32G32B32A32_SINT:
600 case PIPE_FORMAT_R32G32B32A32_UINT:
601 case PIPE_FORMAT_R32G32B32X32_UINT:
602 case PIPE_FORMAT_R32G32B32X32_SINT:
603 return V_028C70_COLOR_32_32_32_32;
604 case PIPE_FORMAT_R32G32B32A32_FLOAT:
605 case PIPE_FORMAT_R32G32B32X32_FLOAT:
606 return V_028C70_COLOR_32_32_32_32_FLOAT;
607
608 /* YUV buffers. */
609 case PIPE_FORMAT_UYVY:
610 case PIPE_FORMAT_YUYV:
611 default:
612 return ~0U; /* Unsupported. */
613 }
614 }
615
616 static uint32_t r600_colorformat_endian_swap(uint32_t colorformat)
617 {
618 if (R600_BIG_ENDIAN) {
619 switch(colorformat) {
620
621 /* 8-bit buffers. */
622 case V_028C70_COLOR_8:
623 return ENDIAN_NONE;
624
625 /* 16-bit buffers. */
626 case V_028C70_COLOR_5_6_5:
627 case V_028C70_COLOR_1_5_5_5:
628 case V_028C70_COLOR_4_4_4_4:
629 case V_028C70_COLOR_16:
630 case V_028C70_COLOR_8_8:
631 return ENDIAN_8IN16;
632
633 /* 32-bit buffers. */
634 case V_028C70_COLOR_8_8_8_8:
635 case V_028C70_COLOR_2_10_10_10:
636 case V_028C70_COLOR_8_24:
637 case V_028C70_COLOR_24_8:
638 case V_028C70_COLOR_32_FLOAT:
639 case V_028C70_COLOR_16_16_FLOAT:
640 case V_028C70_COLOR_16_16:
641 return ENDIAN_8IN32;
642
643 /* 64-bit buffers. */
644 case V_028C70_COLOR_16_16_16_16:
645 case V_028C70_COLOR_16_16_16_16_FLOAT:
646 return ENDIAN_8IN16;
647
648 case V_028C70_COLOR_32_32_FLOAT:
649 case V_028C70_COLOR_32_32:
650 case V_028C70_COLOR_X24_8_32_FLOAT:
651 return ENDIAN_8IN32;
652
653 /* 96-bit buffers. */
654 case V_028C70_COLOR_32_32_32_FLOAT:
655 /* 128-bit buffers. */
656 case V_028C70_COLOR_32_32_32_32_FLOAT:
657 case V_028C70_COLOR_32_32_32_32:
658 return ENDIAN_8IN32;
659 default:
660 return ENDIAN_NONE; /* Unsupported. */
661 }
662 } else {
663 return ENDIAN_NONE;
664 }
665 }
666
667 static bool r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
668 {
669 return r600_translate_texformat(screen, format, NULL, NULL, NULL) != ~0U;
670 }
671
672 static bool r600_is_colorbuffer_format_supported(enum pipe_format format)
673 {
674 return r600_translate_colorformat(format) != ~0U &&
675 r600_translate_colorswap(format) != ~0U;
676 }
677
678 static bool r600_is_zs_format_supported(enum pipe_format format)
679 {
680 return r600_translate_dbformat(format) != ~0U;
681 }
682
683 boolean evergreen_is_format_supported(struct pipe_screen *screen,
684 enum pipe_format format,
685 enum pipe_texture_target target,
686 unsigned sample_count,
687 unsigned usage)
688 {
689 struct r600_screen *rscreen = (struct r600_screen*)screen;
690 unsigned retval = 0;
691
692 if (target >= PIPE_MAX_TEXTURE_TYPES) {
693 R600_ERR("r600: unsupported texture type %d\n", target);
694 return FALSE;
695 }
696
697 if (!util_format_is_supported(format, usage))
698 return FALSE;
699
700 if (sample_count > 1) {
701 if (!rscreen->has_msaa)
702 return FALSE;
703
704 switch (sample_count) {
705 case 2:
706 case 4:
707 case 8:
708 break;
709 default:
710 return FALSE;
711 }
712 }
713
714 if ((usage & PIPE_BIND_SAMPLER_VIEW) &&
715 r600_is_sampler_format_supported(screen, format)) {
716 retval |= PIPE_BIND_SAMPLER_VIEW;
717 }
718
719 if ((usage & (PIPE_BIND_RENDER_TARGET |
720 PIPE_BIND_DISPLAY_TARGET |
721 PIPE_BIND_SCANOUT |
722 PIPE_BIND_SHARED)) &&
723 r600_is_colorbuffer_format_supported(format)) {
724 retval |= usage &
725 (PIPE_BIND_RENDER_TARGET |
726 PIPE_BIND_DISPLAY_TARGET |
727 PIPE_BIND_SCANOUT |
728 PIPE_BIND_SHARED);
729 }
730
731 if ((usage & PIPE_BIND_DEPTH_STENCIL) &&
732 r600_is_zs_format_supported(format)) {
733 retval |= PIPE_BIND_DEPTH_STENCIL;
734 }
735
736 if ((usage & PIPE_BIND_VERTEX_BUFFER) &&
737 r600_is_vertex_format_supported(format)) {
738 retval |= PIPE_BIND_VERTEX_BUFFER;
739 }
740
741 if (usage & PIPE_BIND_TRANSFER_READ)
742 retval |= PIPE_BIND_TRANSFER_READ;
743 if (usage & PIPE_BIND_TRANSFER_WRITE)
744 retval |= PIPE_BIND_TRANSFER_WRITE;
745
746 return retval == usage;
747 }
748
749 static void *evergreen_create_blend_state_mode(struct pipe_context *ctx,
750 const struct pipe_blend_state *state, int mode)
751 {
752 uint32_t color_control = 0, target_mask = 0;
753 struct r600_blend_state *blend = CALLOC_STRUCT(r600_blend_state);
754
755 if (!blend) {
756 return NULL;
757 }
758
759 r600_init_command_buffer(&blend->buffer, 20);
760 r600_init_command_buffer(&blend->buffer_no_blend, 20);
761
762 if (state->logicop_enable) {
763 color_control |= (state->logicop_func << 16) | (state->logicop_func << 20);
764 } else {
765 color_control |= (0xcc << 16);
766 }
767 /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
768 if (state->independent_blend_enable) {
769 for (int i = 0; i < 8; i++) {
770 target_mask |= (state->rt[i].colormask << (4 * i));
771 }
772 } else {
773 for (int i = 0; i < 8; i++) {
774 target_mask |= (state->rt[0].colormask << (4 * i));
775 }
776 }
777
778 /* only have dual source on MRT0 */
779 blend->dual_src_blend = util_blend_state_is_dual(state, 0);
780 blend->cb_target_mask = target_mask;
781 blend->alpha_to_one = state->alpha_to_one;
782
783 if (target_mask)
784 color_control |= S_028808_MODE(mode);
785 else
786 color_control |= S_028808_MODE(V_028808_CB_DISABLE);
787
788
789 r600_store_context_reg(&blend->buffer, R_028808_CB_COLOR_CONTROL, color_control);
790 r600_store_context_reg(&blend->buffer, R_028B70_DB_ALPHA_TO_MASK,
791 S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
792 S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
793 S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
794 S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
795 S_028B70_ALPHA_TO_MASK_OFFSET3(2));
796 r600_store_context_reg_seq(&blend->buffer, R_028780_CB_BLEND0_CONTROL, 8);
797
798 /* Copy over the dwords set so far into buffer_no_blend.
799 * Only the CB_BLENDi_CONTROL registers must be set after this. */
800 memcpy(blend->buffer_no_blend.buf, blend->buffer.buf, blend->buffer.num_dw * 4);
801 blend->buffer_no_blend.num_dw = blend->buffer.num_dw;
802
803 for (int i = 0; i < 8; i++) {
804 /* state->rt entries > 0 only written if independent blending */
805 const int j = state->independent_blend_enable ? i : 0;
806
807 unsigned eqRGB = state->rt[j].rgb_func;
808 unsigned srcRGB = state->rt[j].rgb_src_factor;
809 unsigned dstRGB = state->rt[j].rgb_dst_factor;
810 unsigned eqA = state->rt[j].alpha_func;
811 unsigned srcA = state->rt[j].alpha_src_factor;
812 unsigned dstA = state->rt[j].alpha_dst_factor;
813 uint32_t bc = 0;
814
815 r600_store_value(&blend->buffer_no_blend, 0);
816
817 if (!state->rt[j].blend_enable) {
818 r600_store_value(&blend->buffer, 0);
819 continue;
820 }
821
822 bc |= S_028780_BLEND_CONTROL_ENABLE(1);
823 bc |= S_028780_COLOR_COMB_FCN(r600_translate_blend_function(eqRGB));
824 bc |= S_028780_COLOR_SRCBLEND(r600_translate_blend_factor(srcRGB));
825 bc |= S_028780_COLOR_DESTBLEND(r600_translate_blend_factor(dstRGB));
826
827 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
828 bc |= S_028780_SEPARATE_ALPHA_BLEND(1);
829 bc |= S_028780_ALPHA_COMB_FCN(r600_translate_blend_function(eqA));
830 bc |= S_028780_ALPHA_SRCBLEND(r600_translate_blend_factor(srcA));
831 bc |= S_028780_ALPHA_DESTBLEND(r600_translate_blend_factor(dstA));
832 }
833 r600_store_value(&blend->buffer, bc);
834 }
835 return blend;
836 }
837
838 static void *evergreen_create_blend_state(struct pipe_context *ctx,
839 const struct pipe_blend_state *state)
840 {
841
842 return evergreen_create_blend_state_mode(ctx, state, V_028808_CB_NORMAL);
843 }
844
845 static void *evergreen_create_dsa_state(struct pipe_context *ctx,
846 const struct pipe_depth_stencil_alpha_state *state)
847 {
848 unsigned db_depth_control, alpha_test_control, alpha_ref;
849 struct r600_dsa_state *dsa = CALLOC_STRUCT(r600_dsa_state);
850
851 if (dsa == NULL) {
852 return NULL;
853 }
854
855 r600_init_command_buffer(&dsa->buffer, 3);
856
857 dsa->valuemask[0] = state->stencil[0].valuemask;
858 dsa->valuemask[1] = state->stencil[1].valuemask;
859 dsa->writemask[0] = state->stencil[0].writemask;
860 dsa->writemask[1] = state->stencil[1].writemask;
861 dsa->zwritemask = state->depth.writemask;
862
863 db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) |
864 S_028800_Z_WRITE_ENABLE(state->depth.writemask) |
865 S_028800_ZFUNC(state->depth.func);
866
867 /* stencil */
868 if (state->stencil[0].enabled) {
869 db_depth_control |= S_028800_STENCIL_ENABLE(1);
870 db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func); /* translates straight */
871 db_depth_control |= S_028800_STENCILFAIL(r600_translate_stencil_op(state->stencil[0].fail_op));
872 db_depth_control |= S_028800_STENCILZPASS(r600_translate_stencil_op(state->stencil[0].zpass_op));
873 db_depth_control |= S_028800_STENCILZFAIL(r600_translate_stencil_op(state->stencil[0].zfail_op));
874
875 if (state->stencil[1].enabled) {
876 db_depth_control |= S_028800_BACKFACE_ENABLE(1);
877 db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func); /* translates straight */
878 db_depth_control |= S_028800_STENCILFAIL_BF(r600_translate_stencil_op(state->stencil[1].fail_op));
879 db_depth_control |= S_028800_STENCILZPASS_BF(r600_translate_stencil_op(state->stencil[1].zpass_op));
880 db_depth_control |= S_028800_STENCILZFAIL_BF(r600_translate_stencil_op(state->stencil[1].zfail_op));
881 }
882 }
883
884 /* alpha */
885 alpha_test_control = 0;
886 alpha_ref = 0;
887 if (state->alpha.enabled) {
888 alpha_test_control = S_028410_ALPHA_FUNC(state->alpha.func);
889 alpha_test_control |= S_028410_ALPHA_TEST_ENABLE(1);
890 alpha_ref = fui(state->alpha.ref_value);
891 }
892 dsa->sx_alpha_test_control = alpha_test_control & 0xff;
893 dsa->alpha_ref = alpha_ref;
894
895 /* misc */
896 r600_store_context_reg(&dsa->buffer, R_028800_DB_DEPTH_CONTROL, db_depth_control);
897 return dsa;
898 }
899
900 static void *evergreen_create_rs_state(struct pipe_context *ctx,
901 const struct pipe_rasterizer_state *state)
902 {
903 struct r600_context *rctx = (struct r600_context *)ctx;
904 unsigned tmp, spi_interp;
905 float psize_min, psize_max;
906 struct r600_rasterizer_state *rs = CALLOC_STRUCT(r600_rasterizer_state);
907
908 if (rs == NULL) {
909 return NULL;
910 }
911
912 r600_init_command_buffer(&rs->buffer, 30);
913
914 rs->flatshade = state->flatshade;
915 rs->sprite_coord_enable = state->sprite_coord_enable;
916 rs->two_side = state->light_twoside;
917 rs->clip_plane_enable = state->clip_plane_enable;
918 rs->pa_sc_line_stipple = state->line_stipple_enable ?
919 S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
920 S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
921 rs->pa_cl_clip_cntl =
922 S_028810_PS_UCP_MODE(3) |
923 S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
924 S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
925 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
926 rs->multisample_enable = state->multisample;
927
928 /* offset */
929 rs->offset_units = state->offset_units;
930 rs->offset_scale = state->offset_scale * 12.0f;
931 rs->offset_enable = state->offset_point || state->offset_line || state->offset_tri;
932
933 if (state->point_size_per_vertex) {
934 psize_min = util_get_min_point_size(state);
935 psize_max = 8192;
936 } else {
937 /* Force the point size to be as if the vertex output was disabled. */
938 psize_min = state->point_size;
939 psize_max = state->point_size;
940 }
941
942 spi_interp = S_0286D4_FLAT_SHADE_ENA(1);
943 if (state->sprite_coord_enable) {
944 spi_interp |= S_0286D4_PNT_SPRITE_ENA(1) |
945 S_0286D4_PNT_SPRITE_OVRD_X(2) |
946 S_0286D4_PNT_SPRITE_OVRD_Y(3) |
947 S_0286D4_PNT_SPRITE_OVRD_Z(0) |
948 S_0286D4_PNT_SPRITE_OVRD_W(1);
949 if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) {
950 spi_interp |= S_0286D4_PNT_SPRITE_TOP_1(1);
951 }
952 }
953
954 r600_store_context_reg_seq(&rs->buffer, R_028A00_PA_SU_POINT_SIZE, 3);
955 /* point size 12.4 fixed point (divide by two, because 0.5 = 1 pixel) */
956 tmp = r600_pack_float_12p4(state->point_size/2);
957 r600_store_value(&rs->buffer, /* R_028A00_PA_SU_POINT_SIZE */
958 S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
959 r600_store_value(&rs->buffer, /* R_028A04_PA_SU_POINT_MINMAX */
960 S_028A04_MIN_SIZE(r600_pack_float_12p4(psize_min/2)) |
961 S_028A04_MAX_SIZE(r600_pack_float_12p4(psize_max/2)));
962 r600_store_value(&rs->buffer, /* R_028A08_PA_SU_LINE_CNTL */
963 S_028A08_WIDTH((unsigned)(state->line_width * 8)));
964
965 r600_store_context_reg(&rs->buffer, R_0286D4_SPI_INTERP_CONTROL_0, spi_interp);
966 r600_store_context_reg(&rs->buffer, R_028A48_PA_SC_MODE_CNTL_0,
967 S_028A48_MSAA_ENABLE(state->multisample) |
968 S_028A48_VPORT_SCISSOR_ENABLE(state->scissor) |
969 S_028A48_LINE_STIPPLE_ENABLE(state->line_stipple_enable));
970
971 if (rctx->chip_class == CAYMAN) {
972 r600_store_context_reg(&rs->buffer, CM_R_028BE4_PA_SU_VTX_CNTL,
973 S_028C08_PIX_CENTER_HALF(state->gl_rasterization_rules) |
974 S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
975 } else {
976 r600_store_context_reg(&rs->buffer, R_028C08_PA_SU_VTX_CNTL,
977 S_028C08_PIX_CENTER_HALF(state->gl_rasterization_rules) |
978 S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
979 }
980
981 r600_store_context_reg(&rs->buffer, R_028B7C_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
982 r600_store_context_reg(&rs->buffer, R_028814_PA_SU_SC_MODE_CNTL,
983 S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) |
984 S_028814_CULL_FRONT((state->cull_face & PIPE_FACE_FRONT) ? 1 : 0) |
985 S_028814_CULL_BACK((state->cull_face & PIPE_FACE_BACK) ? 1 : 0) |
986 S_028814_FACE(!state->front_ccw) |
987 S_028814_POLY_OFFSET_FRONT_ENABLE(state->offset_tri) |
988 S_028814_POLY_OFFSET_BACK_ENABLE(state->offset_tri) |
989 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_tri) |
990 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL ||
991 state->fill_back != PIPE_POLYGON_MODE_FILL) |
992 S_028814_POLYMODE_FRONT_PTYPE(r600_translate_fill(state->fill_front)) |
993 S_028814_POLYMODE_BACK_PTYPE(r600_translate_fill(state->fill_back)));
994 r600_store_context_reg(&rs->buffer, R_028350_SX_MISC, S_028350_MULTIPASS(state->rasterizer_discard));
995 return rs;
996 }
997
998 static void *evergreen_create_sampler_state(struct pipe_context *ctx,
999 const struct pipe_sampler_state *state)
1000 {
1001 struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
1002 unsigned aniso_flag_offset = state->max_anisotropy > 1 ? 2 : 0;
1003
1004 if (ss == NULL) {
1005 return NULL;
1006 }
1007
1008 ss->border_color_use = sampler_state_needs_border_color(state);
1009
1010 /* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
1011 ss->tex_sampler_words[0] =
1012 S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
1013 S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
1014 S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
1015 S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
1016 S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter) | aniso_flag_offset) |
1017 S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
1018 S_03C000_MAX_ANISO(r600_tex_aniso_filter(state->max_anisotropy)) |
1019 S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
1020 S_03C000_BORDER_COLOR_TYPE(ss->border_color_use ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
1021 /* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
1022 ss->tex_sampler_words[1] =
1023 S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
1024 S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8));
1025 /* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
1026 ss->tex_sampler_words[2] =
1027 S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
1028 (state->seamless_cube_map ? 0 : S_03C008_DISABLE_CUBE_WRAP(1)) |
1029 S_03C008_TYPE(1);
1030
1031 if (ss->border_color_use) {
1032 memcpy(&ss->border_color, &state->border_color, sizeof(state->border_color));
1033 }
1034 return ss;
1035 }
1036
1037 static struct pipe_sampler_view *
1038 texture_buffer_sampler_view(struct r600_pipe_sampler_view *view,
1039 unsigned width0, unsigned height0)
1040
1041 {
1042 struct pipe_context *ctx = view->base.context;
1043 struct r600_texture *tmp = (struct r600_texture*)view->base.texture;
1044 uint64_t va;
1045 int stride = util_format_get_blocksize(view->base.format);
1046 unsigned format, num_format, format_comp, endian;
1047 unsigned swizzle_res;
1048 unsigned char swizzle[4];
1049 const struct util_format_description *desc;
1050
1051 swizzle[0] = view->base.swizzle_r;
1052 swizzle[1] = view->base.swizzle_g;
1053 swizzle[2] = view->base.swizzle_b;
1054 swizzle[3] = view->base.swizzle_a;
1055
1056 r600_vertex_data_type(view->base.format,
1057 &format, &num_format, &format_comp,
1058 &endian);
1059
1060 desc = util_format_description(view->base.format);
1061
1062 swizzle_res = r600_get_swizzle_combined(desc->swizzle, swizzle, TRUE);
1063
1064 va = r600_resource_va(ctx->screen, view->base.texture);
1065 view->tex_resource = &tmp->resource;
1066
1067 view->skip_mip_address_reloc = true;
1068 view->tex_resource_words[0] = va;
1069 view->tex_resource_words[1] = width0 - 1;
1070 view->tex_resource_words[2] = S_030008_BASE_ADDRESS_HI(va >> 32UL) |
1071 S_030008_STRIDE(stride) |
1072 S_030008_DATA_FORMAT(format) |
1073 S_030008_NUM_FORMAT_ALL(num_format) |
1074 S_030008_FORMAT_COMP_ALL(format_comp) |
1075 S_030008_SRF_MODE_ALL(1) |
1076 S_030008_ENDIAN_SWAP(endian);
1077 view->tex_resource_words[3] = swizzle_res;
1078 /*
1079 * in theory dword 4 is for number of elements, for use with resinfo,
1080 * but it seems to utterly fail to work, the amd gpu shader analyser
1081 * uses a const buffer to store the element sizes for buffer txq
1082 */
1083 view->tex_resource_words[4] = 0;
1084 view->tex_resource_words[5] = view->tex_resource_words[6] = 0;
1085 view->tex_resource_words[7] = S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_BUFFER);
1086 return &view->base;
1087 }
1088
1089 struct pipe_sampler_view *
1090 evergreen_create_sampler_view_custom(struct pipe_context *ctx,
1091 struct pipe_resource *texture,
1092 const struct pipe_sampler_view *state,
1093 unsigned width0, unsigned height0)
1094 {
1095 struct r600_screen *rscreen = (struct r600_screen*)ctx->screen;
1096 struct r600_pipe_sampler_view *view = CALLOC_STRUCT(r600_pipe_sampler_view);
1097 struct r600_texture *tmp = (struct r600_texture*)texture;
1098 unsigned format, endian;
1099 uint32_t word4 = 0, yuv_format = 0, pitch = 0;
1100 unsigned char swizzle[4], array_mode = 0, non_disp_tiling = 0;
1101 unsigned height, depth, width;
1102 unsigned macro_aspect, tile_split, bankh, bankw, nbanks;
1103 enum pipe_format pipe_format = state->format;
1104 struct radeon_surface_level *surflevel;
1105
1106 if (view == NULL)
1107 return NULL;
1108
1109 /* initialize base object */
1110 view->base = *state;
1111 view->base.texture = NULL;
1112 pipe_reference(NULL, &texture->reference);
1113 view->base.texture = texture;
1114 view->base.reference.count = 1;
1115 view->base.context = ctx;
1116
1117 if (texture->target == PIPE_BUFFER)
1118 return texture_buffer_sampler_view(view, width0, height0);
1119
1120 swizzle[0] = state->swizzle_r;
1121 swizzle[1] = state->swizzle_g;
1122 swizzle[2] = state->swizzle_b;
1123 swizzle[3] = state->swizzle_a;
1124
1125 tile_split = tmp->surface.tile_split;
1126 surflevel = tmp->surface.level;
1127
1128 /* Texturing with separate depth and stencil. */
1129 if (tmp->is_depth && !tmp->is_flushing_texture) {
1130 switch (pipe_format) {
1131 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1132 pipe_format = PIPE_FORMAT_Z32_FLOAT;
1133 break;
1134 case PIPE_FORMAT_X8Z24_UNORM:
1135 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1136 /* Z24 is always stored like this. */
1137 pipe_format = PIPE_FORMAT_Z24X8_UNORM;
1138 break;
1139 case PIPE_FORMAT_X24S8_UINT:
1140 case PIPE_FORMAT_S8X24_UINT:
1141 case PIPE_FORMAT_X32_S8X24_UINT:
1142 pipe_format = PIPE_FORMAT_S8_UINT;
1143 tile_split = tmp->surface.stencil_tile_split;
1144 surflevel = tmp->surface.stencil_level;
1145 break;
1146 default:;
1147 }
1148 }
1149
1150 format = r600_translate_texformat(ctx->screen, pipe_format,
1151 swizzle,
1152 &word4, &yuv_format);
1153 assert(format != ~0);
1154 if (format == ~0) {
1155 FREE(view);
1156 return NULL;
1157 }
1158
1159 endian = r600_colorformat_endian_swap(format);
1160
1161 width = width0;
1162 height = height0;
1163 depth = texture->depth0;
1164 pitch = surflevel[0].nblk_x * util_format_get_blockwidth(pipe_format);
1165 non_disp_tiling = tmp->non_disp_tiling;
1166
1167 switch (surflevel[0].mode) {
1168 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1169 array_mode = V_028C70_ARRAY_LINEAR_ALIGNED;
1170 break;
1171 case RADEON_SURF_MODE_2D:
1172 array_mode = V_028C70_ARRAY_2D_TILED_THIN1;
1173 break;
1174 case RADEON_SURF_MODE_1D:
1175 array_mode = V_028C70_ARRAY_1D_TILED_THIN1;
1176 break;
1177 case RADEON_SURF_MODE_LINEAR:
1178 default:
1179 array_mode = V_028C70_ARRAY_LINEAR_GENERAL;
1180 break;
1181 }
1182 macro_aspect = tmp->surface.mtilea;
1183 bankw = tmp->surface.bankw;
1184 bankh = tmp->surface.bankh;
1185 tile_split = eg_tile_split(tile_split);
1186 macro_aspect = eg_macro_tile_aspect(macro_aspect);
1187 bankw = eg_bank_wh(bankw);
1188 bankh = eg_bank_wh(bankh);
1189
1190 /* 128 bit formats require tile type = 1 */
1191 if (rscreen->chip_class == CAYMAN) {
1192 if (util_format_get_blocksize(pipe_format) >= 16)
1193 non_disp_tiling = 1;
1194 }
1195 nbanks = eg_num_banks(rscreen->tiling_info.num_banks);
1196
1197 if (texture->target == PIPE_TEXTURE_1D_ARRAY) {
1198 height = 1;
1199 depth = texture->array_size;
1200 } else if (texture->target == PIPE_TEXTURE_2D_ARRAY) {
1201 depth = texture->array_size;
1202 } else if (texture->target == PIPE_TEXTURE_CUBE_ARRAY)
1203 depth = texture->array_size / 6;
1204
1205 view->tex_resource = &tmp->resource;
1206 view->tex_resource_words[0] = (S_030000_DIM(r600_tex_dim(texture->target, texture->nr_samples)) |
1207 S_030000_PITCH((pitch / 8) - 1) |
1208 S_030000_TEX_WIDTH(width - 1));
1209 if (rscreen->chip_class == CAYMAN)
1210 view->tex_resource_words[0] |= CM_S_030000_NON_DISP_TILING_ORDER(non_disp_tiling);
1211 else
1212 view->tex_resource_words[0] |= S_030000_NON_DISP_TILING_ORDER(non_disp_tiling);
1213 view->tex_resource_words[1] = (S_030004_TEX_HEIGHT(height - 1) |
1214 S_030004_TEX_DEPTH(depth - 1) |
1215 S_030004_ARRAY_MODE(array_mode));
1216 view->tex_resource_words[2] = (surflevel[0].offset + r600_resource_va(ctx->screen, texture)) >> 8;
1217
1218 /* TEX_RESOURCE_WORD3.MIP_ADDRESS */
1219 if (texture->nr_samples > 1 && rscreen->msaa_texture_support == MSAA_TEXTURE_COMPRESSED) {
1220 /* XXX the 2x and 4x cases are broken. */
1221 if (tmp->is_depth || tmp->resource.b.b.nr_samples != 8) {
1222 /* disable FMASK (0 = disabled) */
1223 view->tex_resource_words[3] = 0;
1224 view->skip_mip_address_reloc = true;
1225 } else {
1226 /* FMASK should be in MIP_ADDRESS for multisample textures */
1227 view->tex_resource_words[3] = (tmp->fmask_offset + r600_resource_va(ctx->screen, texture)) >> 8;
1228 }
1229 } else if (state->u.tex.last_level && texture->nr_samples <= 1) {
1230 view->tex_resource_words[3] = (surflevel[1].offset + r600_resource_va(ctx->screen, texture)) >> 8;
1231 } else {
1232 view->tex_resource_words[3] = (surflevel[0].offset + r600_resource_va(ctx->screen, texture)) >> 8;
1233 }
1234
1235 view->tex_resource_words[4] = (word4 |
1236 S_030010_SRF_MODE_ALL(V_030010_SRF_MODE_ZERO_CLAMP_MINUS_ONE) |
1237 S_030010_ENDIAN_SWAP(endian));
1238 view->tex_resource_words[5] = S_030014_BASE_ARRAY(state->u.tex.first_layer) |
1239 S_030014_LAST_ARRAY(state->u.tex.last_layer);
1240 if (texture->nr_samples > 1) {
1241 unsigned log_samples = util_logbase2(texture->nr_samples);
1242 if (rscreen->chip_class == CAYMAN) {
1243 view->tex_resource_words[4] |= S_030010_LOG2_NUM_FRAGMENTS(log_samples);
1244 }
1245 /* LAST_LEVEL holds log2(nr_samples) for multisample textures */
1246 view->tex_resource_words[5] |= S_030014_LAST_LEVEL(log_samples);
1247 } else {
1248 view->tex_resource_words[4] |= S_030010_BASE_LEVEL(state->u.tex.first_level);
1249 view->tex_resource_words[5] |= S_030014_LAST_LEVEL(state->u.tex.last_level);
1250 }
1251 /* aniso max 16 samples */
1252 view->tex_resource_words[6] = (S_030018_MAX_ANISO(4)) |
1253 (S_030018_TILE_SPLIT(tile_split));
1254 view->tex_resource_words[7] = S_03001C_DATA_FORMAT(format) |
1255 S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_TEXTURE) |
1256 S_03001C_BANK_WIDTH(bankw) |
1257 S_03001C_BANK_HEIGHT(bankh) |
1258 S_03001C_MACRO_TILE_ASPECT(macro_aspect) |
1259 S_03001C_NUM_BANKS(nbanks) |
1260 S_03001C_DEPTH_SAMPLE_ORDER(tmp->is_depth && !tmp->is_flushing_texture);
1261 return &view->base;
1262 }
1263
1264 static struct pipe_sampler_view *
1265 evergreen_create_sampler_view(struct pipe_context *ctx,
1266 struct pipe_resource *tex,
1267 const struct pipe_sampler_view *state)
1268 {
1269 return evergreen_create_sampler_view_custom(ctx, tex, state,
1270 tex->width0, tex->height0);
1271 }
1272
1273 static void evergreen_emit_clip_state(struct r600_context *rctx, struct r600_atom *atom)
1274 {
1275 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
1276 struct pipe_clip_state *state = &rctx->clip_state.state;
1277
1278 r600_write_context_reg_seq(cs, R_0285BC_PA_CL_UCP0_X, 6*4);
1279 r600_write_array(cs, 6*4, (unsigned*)state);
1280 }
1281
1282 static void evergreen_set_polygon_stipple(struct pipe_context *ctx,
1283 const struct pipe_poly_stipple *state)
1284 {
1285 }
1286
1287 static void evergreen_get_scissor_rect(struct r600_context *rctx,
1288 unsigned tl_x, unsigned tl_y, unsigned br_x, unsigned br_y,
1289 uint32_t *tl, uint32_t *br)
1290 {
1291 /* EG hw workaround */
1292 if (br_x == 0)
1293 tl_x = 1;
1294 if (br_y == 0)
1295 tl_y = 1;
1296
1297 /* cayman hw workaround */
1298 if (rctx->chip_class == CAYMAN) {
1299 if (br_x == 1 && br_y == 1)
1300 br_x = 2;
1301 }
1302
1303 *tl = S_028240_TL_X(tl_x) | S_028240_TL_Y(tl_y);
1304 *br = S_028244_BR_X(br_x) | S_028244_BR_Y(br_y);
1305 }
1306
1307 static void evergreen_set_scissor_state(struct pipe_context *ctx,
1308 const struct pipe_scissor_state *state)
1309 {
1310 struct r600_context *rctx = (struct r600_context *)ctx;
1311
1312 rctx->scissor.scissor = *state;
1313 rctx->scissor.atom.dirty = true;
1314 }
1315
1316 static void evergreen_emit_scissor_state(struct r600_context *rctx, struct r600_atom *atom)
1317 {
1318 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
1319 struct pipe_scissor_state *state = &rctx->scissor.scissor;
1320 uint32_t tl, br;
1321
1322 evergreen_get_scissor_rect(rctx, state->minx, state->miny, state->maxx, state->maxy, &tl, &br);
1323
1324 r600_write_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL, 2);
1325 r600_write_value(cs, tl);
1326 r600_write_value(cs, br);
1327 }
1328
1329 /**
1330 * This function intializes the CB* register values for RATs. It is meant
1331 * to be used for 1D aligned buffers that do not have an associated
1332 * radeon_surface.
1333 */
1334 void evergreen_init_color_surface_rat(struct r600_context *rctx,
1335 struct r600_surface *surf)
1336 {
1337 struct pipe_resource *pipe_buffer = surf->base.texture;
1338 unsigned format = r600_translate_colorformat(surf->base.format);
1339 unsigned endian = r600_colorformat_endian_swap(format);
1340 unsigned swap = r600_translate_colorswap(surf->base.format);
1341 unsigned block_size =
1342 align(util_format_get_blocksize(pipe_buffer->format), 4);
1343 unsigned pitch_alignment =
1344 MAX2(64, rctx->screen->tiling_info.group_bytes / block_size);
1345 unsigned pitch = align(pipe_buffer->width0, pitch_alignment);
1346
1347 /* XXX: This is copied from evergreen_init_color_surface(). I don't
1348 * know why this is necessary.
1349 */
1350 if (pipe_buffer->usage == PIPE_USAGE_STAGING) {
1351 endian = ENDIAN_NONE;
1352 }
1353
1354 surf->cb_color_base =
1355 r600_resource_va(rctx->context.screen, pipe_buffer) >> 8;
1356
1357 surf->cb_color_pitch = (pitch / 8) - 1;
1358
1359 surf->cb_color_slice = 0;
1360
1361 surf->cb_color_view = 0;
1362
1363 surf->cb_color_info =
1364 S_028C70_ENDIAN(endian)
1365 | S_028C70_FORMAT(format)
1366 | S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_ALIGNED)
1367 | S_028C70_NUMBER_TYPE(V_028C70_NUMBER_UINT)
1368 | S_028C70_COMP_SWAP(swap)
1369 | S_028C70_BLEND_BYPASS(1) /* We must set this bit because we
1370 * are using NUMBER_UINT */
1371 | S_028C70_RAT(1)
1372 ;
1373
1374 surf->cb_color_attrib = S_028C74_NON_DISP_TILING_ORDER(1);
1375
1376 /* For buffers, CB_COLOR0_DIM needs to be set to the number of
1377 * elements. */
1378 surf->cb_color_dim = pipe_buffer->width0;
1379
1380 /* Set the buffer range the GPU will have access to: */
1381 util_range_add(&r600_resource(pipe_buffer)->valid_buffer_range,
1382 0, pipe_buffer->width0);
1383
1384 surf->cb_color_cmask = surf->cb_color_base;
1385 surf->cb_color_cmask_slice = 0;
1386 surf->cb_color_fmask = surf->cb_color_base;
1387 surf->cb_color_fmask_slice = 0;
1388 }
1389
1390 void evergreen_init_color_surface(struct r600_context *rctx,
1391 struct r600_surface *surf)
1392 {
1393 struct r600_screen *rscreen = rctx->screen;
1394 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1395 struct pipe_resource *pipe_tex = surf->base.texture;
1396 unsigned level = surf->base.u.tex.level;
1397 unsigned pitch, slice;
1398 unsigned color_info, color_attrib, color_dim = 0;
1399 unsigned format, swap, ntype, endian;
1400 uint64_t offset, base_offset;
1401 unsigned non_disp_tiling, macro_aspect, tile_split, bankh, bankw, fmask_bankh, nbanks;
1402 const struct util_format_description *desc;
1403 int i;
1404 bool blend_clamp = 0, blend_bypass = 0;
1405
1406 offset = rtex->surface.level[level].offset;
1407 if (rtex->surface.level[level].mode < RADEON_SURF_MODE_1D) {
1408 offset += rtex->surface.level[level].slice_size *
1409 surf->base.u.tex.first_layer;
1410 }
1411 pitch = (rtex->surface.level[level].nblk_x) / 8 - 1;
1412 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
1413 if (slice) {
1414 slice = slice - 1;
1415 }
1416 color_info = 0;
1417 switch (rtex->surface.level[level].mode) {
1418 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1419 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_ALIGNED);
1420 non_disp_tiling = 1;
1421 break;
1422 case RADEON_SURF_MODE_1D:
1423 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_1D_TILED_THIN1);
1424 non_disp_tiling = rtex->non_disp_tiling;
1425 break;
1426 case RADEON_SURF_MODE_2D:
1427 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_2D_TILED_THIN1);
1428 non_disp_tiling = rtex->non_disp_tiling;
1429 break;
1430 case RADEON_SURF_MODE_LINEAR:
1431 default:
1432 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_GENERAL);
1433 non_disp_tiling = 1;
1434 break;
1435 }
1436 tile_split = rtex->surface.tile_split;
1437 macro_aspect = rtex->surface.mtilea;
1438 bankw = rtex->surface.bankw;
1439 bankh = rtex->surface.bankh;
1440 fmask_bankh = rtex->fmask_bank_height;
1441 tile_split = eg_tile_split(tile_split);
1442 macro_aspect = eg_macro_tile_aspect(macro_aspect);
1443 bankw = eg_bank_wh(bankw);
1444 bankh = eg_bank_wh(bankh);
1445 fmask_bankh = eg_bank_wh(fmask_bankh);
1446
1447 /* 128 bit formats require tile type = 1 */
1448 if (rscreen->chip_class == CAYMAN) {
1449 if (util_format_get_blocksize(surf->base.format) >= 16)
1450 non_disp_tiling = 1;
1451 }
1452 nbanks = eg_num_banks(rscreen->tiling_info.num_banks);
1453 desc = util_format_description(surf->base.format);
1454 for (i = 0; i < 4; i++) {
1455 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
1456 break;
1457 }
1458 }
1459
1460 color_attrib = S_028C74_TILE_SPLIT(tile_split)|
1461 S_028C74_NUM_BANKS(nbanks) |
1462 S_028C74_BANK_WIDTH(bankw) |
1463 S_028C74_BANK_HEIGHT(bankh) |
1464 S_028C74_MACRO_TILE_ASPECT(macro_aspect) |
1465 S_028C74_NON_DISP_TILING_ORDER(non_disp_tiling) |
1466 S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
1467
1468 if (rctx->chip_class == CAYMAN) {
1469 color_attrib |= S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] ==
1470 UTIL_FORMAT_SWIZZLE_1);
1471
1472 if (rtex->resource.b.b.nr_samples > 1) {
1473 unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples);
1474 color_attrib |= S_028C74_NUM_SAMPLES(log_samples) |
1475 S_028C74_NUM_FRAGMENTS(log_samples);
1476 }
1477 }
1478
1479 ntype = V_028C70_NUMBER_UNORM;
1480 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
1481 ntype = V_028C70_NUMBER_SRGB;
1482 else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
1483 if (desc->channel[i].normalized)
1484 ntype = V_028C70_NUMBER_SNORM;
1485 else if (desc->channel[i].pure_integer)
1486 ntype = V_028C70_NUMBER_SINT;
1487 } else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
1488 if (desc->channel[i].normalized)
1489 ntype = V_028C70_NUMBER_UNORM;
1490 else if (desc->channel[i].pure_integer)
1491 ntype = V_028C70_NUMBER_UINT;
1492 }
1493
1494 format = r600_translate_colorformat(surf->base.format);
1495 assert(format != ~0);
1496
1497 swap = r600_translate_colorswap(surf->base.format);
1498 assert(swap != ~0);
1499
1500 if (rtex->resource.b.b.usage == PIPE_USAGE_STAGING) {
1501 endian = ENDIAN_NONE;
1502 } else {
1503 endian = r600_colorformat_endian_swap(format);
1504 }
1505
1506 /* blend clamp should be set for all NORM/SRGB types */
1507 if (ntype == V_028C70_NUMBER_UNORM || ntype == V_028C70_NUMBER_SNORM ||
1508 ntype == V_028C70_NUMBER_SRGB)
1509 blend_clamp = 1;
1510
1511 /* set blend bypass according to docs if SINT/UINT or
1512 8/24 COLOR variants */
1513 if (ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT ||
1514 format == V_028C70_COLOR_8_24 || format == V_028C70_COLOR_24_8 ||
1515 format == V_028C70_COLOR_X24_8_32_FLOAT) {
1516 blend_clamp = 0;
1517 blend_bypass = 1;
1518 }
1519
1520 surf->alphatest_bypass = ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT;
1521
1522 color_info |= S_028C70_FORMAT(format) |
1523 S_028C70_COMP_SWAP(swap) |
1524 S_028C70_BLEND_CLAMP(blend_clamp) |
1525 S_028C70_BLEND_BYPASS(blend_bypass) |
1526 S_028C70_NUMBER_TYPE(ntype) |
1527 S_028C70_ENDIAN(endian);
1528
1529 if (rtex->is_rat) {
1530 color_info |= S_028C70_RAT(1);
1531 color_dim = S_028C78_WIDTH_MAX(pipe_tex->width0 & 0xffff)
1532 | S_028C78_HEIGHT_MAX((pipe_tex->width0 >> 16) & 0xffff);
1533 }
1534
1535 /* EXPORT_NORM is an optimzation that can be enabled for better
1536 * performance in certain cases.
1537 * EXPORT_NORM can be enabled if:
1538 * - 11-bit or smaller UNORM/SNORM/SRGB
1539 * - 16-bit or smaller FLOAT
1540 */
1541 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
1542 ((desc->channel[i].size < 12 &&
1543 desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
1544 ntype != V_028C70_NUMBER_UINT && ntype != V_028C70_NUMBER_SINT) ||
1545 (desc->channel[i].size < 17 &&
1546 desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT))) {
1547 color_info |= S_028C70_SOURCE_FORMAT(V_028C70_EXPORT_4C_16BPC);
1548 surf->export_16bpc = true;
1549 }
1550
1551 if (rtex->fmask_size && rtex->cmask_size) {
1552 color_info |= S_028C70_COMPRESSION(1) | S_028C70_FAST_CLEAR(1);
1553 }
1554
1555 base_offset = r600_resource_va(rctx->context.screen, pipe_tex);
1556
1557 /* XXX handle enabling of CB beyond BASE8 which has different offset */
1558 surf->cb_color_base = (base_offset + offset) >> 8;
1559 surf->cb_color_dim = color_dim;
1560 surf->cb_color_info = color_info;
1561 surf->cb_color_pitch = S_028C64_PITCH_TILE_MAX(pitch);
1562 surf->cb_color_slice = S_028C68_SLICE_TILE_MAX(slice);
1563 if (rtex->surface.level[level].mode < RADEON_SURF_MODE_1D) {
1564 surf->cb_color_view = 0;
1565 } else {
1566 surf->cb_color_view = S_028C6C_SLICE_START(surf->base.u.tex.first_layer) |
1567 S_028C6C_SLICE_MAX(surf->base.u.tex.last_layer);
1568 }
1569 surf->cb_color_attrib = color_attrib;
1570 if (rtex->fmask_size && rtex->cmask_size) {
1571 surf->cb_color_fmask = (base_offset + rtex->fmask_offset) >> 8;
1572 surf->cb_color_cmask = (base_offset + rtex->cmask_offset) >> 8;
1573 } else {
1574 surf->cb_color_fmask = surf->cb_color_base;
1575 surf->cb_color_cmask = surf->cb_color_base;
1576 }
1577 surf->cb_color_fmask_slice = S_028C88_TILE_MAX(slice);
1578 surf->cb_color_cmask_slice = S_028C80_TILE_MAX(rtex->cmask_slice_tile_max);
1579
1580 surf->color_initialized = true;
1581 }
1582
1583 static void evergreen_init_depth_surface(struct r600_context *rctx,
1584 struct r600_surface *surf)
1585 {
1586 struct r600_screen *rscreen = rctx->screen;
1587 struct pipe_screen *screen = &rscreen->screen;
1588 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1589 uint64_t offset;
1590 unsigned level, pitch, slice, format, array_mode;
1591 unsigned macro_aspect, tile_split, bankh, bankw, nbanks;
1592
1593 level = surf->base.u.tex.level;
1594 format = r600_translate_dbformat(surf->base.format);
1595 assert(format != ~0);
1596
1597 offset = r600_resource_va(screen, surf->base.texture);
1598 offset += rtex->surface.level[level].offset;
1599 pitch = (rtex->surface.level[level].nblk_x / 8) - 1;
1600 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
1601 if (slice) {
1602 slice = slice - 1;
1603 }
1604 switch (rtex->surface.level[level].mode) {
1605 case RADEON_SURF_MODE_2D:
1606 array_mode = V_028C70_ARRAY_2D_TILED_THIN1;
1607 break;
1608 case RADEON_SURF_MODE_1D:
1609 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1610 case RADEON_SURF_MODE_LINEAR:
1611 default:
1612 array_mode = V_028C70_ARRAY_1D_TILED_THIN1;
1613 break;
1614 }
1615 tile_split = rtex->surface.tile_split;
1616 macro_aspect = rtex->surface.mtilea;
1617 bankw = rtex->surface.bankw;
1618 bankh = rtex->surface.bankh;
1619 tile_split = eg_tile_split(tile_split);
1620 macro_aspect = eg_macro_tile_aspect(macro_aspect);
1621 bankw = eg_bank_wh(bankw);
1622 bankh = eg_bank_wh(bankh);
1623 nbanks = eg_num_banks(rscreen->tiling_info.num_banks);
1624 offset >>= 8;
1625
1626 surf->db_depth_info = S_028040_ARRAY_MODE(array_mode) |
1627 S_028040_FORMAT(format) |
1628 S_028040_TILE_SPLIT(tile_split)|
1629 S_028040_NUM_BANKS(nbanks) |
1630 S_028040_BANK_WIDTH(bankw) |
1631 S_028040_BANK_HEIGHT(bankh) |
1632 S_028040_MACRO_TILE_ASPECT(macro_aspect);
1633 if (rscreen->chip_class == CAYMAN && rtex->resource.b.b.nr_samples > 1) {
1634 surf->db_depth_info |= S_028040_NUM_SAMPLES(util_logbase2(rtex->resource.b.b.nr_samples));
1635 }
1636 surf->db_depth_base = offset;
1637 surf->db_depth_view = S_028008_SLICE_START(surf->base.u.tex.first_layer) |
1638 S_028008_SLICE_MAX(surf->base.u.tex.last_layer);
1639 surf->db_depth_size = S_028058_PITCH_TILE_MAX(pitch);
1640 surf->db_depth_slice = S_02805C_SLICE_TILE_MAX(slice);
1641
1642 switch (surf->base.format) {
1643 case PIPE_FORMAT_Z24X8_UNORM:
1644 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1645 case PIPE_FORMAT_X8Z24_UNORM:
1646 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1647 surf->pa_su_poly_offset_db_fmt_cntl =
1648 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-24);
1649 break;
1650 case PIPE_FORMAT_Z32_FLOAT:
1651 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1652 surf->pa_su_poly_offset_db_fmt_cntl =
1653 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-23) |
1654 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
1655 break;
1656 case PIPE_FORMAT_Z16_UNORM:
1657 surf->pa_su_poly_offset_db_fmt_cntl =
1658 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-16);
1659 break;
1660 default:;
1661 }
1662
1663 if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
1664 uint64_t stencil_offset;
1665 unsigned stile_split = rtex->surface.stencil_tile_split;
1666
1667 stile_split = eg_tile_split(stile_split);
1668
1669 stencil_offset = rtex->surface.stencil_level[level].offset;
1670 stencil_offset += r600_resource_va(screen, surf->base.texture);
1671
1672 surf->db_stencil_base = stencil_offset >> 8;
1673 surf->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_8) |
1674 S_028044_TILE_SPLIT(stile_split);
1675 } else {
1676 surf->db_stencil_base = offset;
1677 /* DRM 2.6.18 allows the INVALID format to disable stencil.
1678 * Older kernels are out of luck. */
1679 surf->db_stencil_info = rctx->screen->info.drm_minor >= 18 ?
1680 S_028044_FORMAT(V_028044_STENCIL_INVALID) :
1681 S_028044_FORMAT(V_028044_STENCIL_8);
1682 }
1683
1684 surf->htile_enabled = 0;
1685 /* use htile only for first level */
1686 if (rtex->htile && !level) {
1687 uint64_t va = r600_resource_va(&rctx->screen->screen, &rtex->htile->b.b);
1688 surf->htile_enabled = 1;
1689 surf->db_htile_data_base = va >> 8;
1690 surf->db_htile_surface = S_028ABC_HTILE_WIDTH(1) |
1691 S_028ABC_HTILE_HEIGHT(1) |
1692 S_028ABC_LINEAR(1);
1693 surf->db_depth_info |= S_028040_TILE_SURFACE_ENABLE(1);
1694 surf->db_preload_control = 0;
1695 }
1696
1697 surf->depth_initialized = true;
1698 }
1699
1700 static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
1701 const struct pipe_framebuffer_state *state)
1702 {
1703 struct r600_context *rctx = (struct r600_context *)ctx;
1704 struct r600_surface *surf;
1705 struct r600_texture *rtex;
1706 uint32_t i, log_samples;
1707
1708 if (rctx->framebuffer.state.nr_cbufs) {
1709 rctx->flags |= R600_CONTEXT_WAIT_3D_IDLE | R600_CONTEXT_FLUSH_AND_INV;
1710
1711 if (rctx->framebuffer.state.cbufs[0]->texture->nr_samples > 1) {
1712 rctx->flags |= R600_CONTEXT_FLUSH_AND_INV_CB_META;
1713 }
1714 }
1715 if (rctx->framebuffer.state.zsbuf) {
1716 rctx->flags |= R600_CONTEXT_WAIT_3D_IDLE | R600_CONTEXT_FLUSH_AND_INV;
1717
1718 rtex = (struct r600_texture*)rctx->framebuffer.state.zsbuf->texture;
1719 if (rtex->htile) {
1720 rctx->flags |= R600_CONTEXT_FLUSH_AND_INV_DB_META;
1721 }
1722 }
1723
1724 util_copy_framebuffer_state(&rctx->framebuffer.state, state);
1725
1726 /* Colorbuffers. */
1727 rctx->framebuffer.export_16bpc = state->nr_cbufs != 0;
1728 rctx->framebuffer.cb0_is_integer = state->nr_cbufs &&
1729 util_format_is_pure_integer(state->cbufs[0]->format);
1730 rctx->framebuffer.compressed_cb_mask = 0;
1731
1732 if (state->nr_cbufs)
1733 rctx->framebuffer.nr_samples = state->cbufs[0]->texture->nr_samples;
1734 else if (state->zsbuf)
1735 rctx->framebuffer.nr_samples = state->zsbuf->texture->nr_samples;
1736 else
1737 rctx->framebuffer.nr_samples = 0;
1738
1739 for (i = 0; i < state->nr_cbufs; i++) {
1740 surf = (struct r600_surface*)state->cbufs[i];
1741 rtex = (struct r600_texture*)surf->base.texture;
1742
1743 r600_context_add_resource_size(ctx, state->cbufs[i]->texture);
1744
1745 if (!surf->color_initialized) {
1746 evergreen_init_color_surface(rctx, surf);
1747 }
1748
1749 if (!surf->export_16bpc) {
1750 rctx->framebuffer.export_16bpc = false;
1751 }
1752
1753 if (rtex->fmask_size && rtex->cmask_size) {
1754 rctx->framebuffer.compressed_cb_mask |= 1 << i;
1755 }
1756 }
1757
1758 /* Update alpha-test state dependencies.
1759 * Alpha-test is done on the first colorbuffer only. */
1760 if (state->nr_cbufs) {
1761 surf = (struct r600_surface*)state->cbufs[0];
1762 if (rctx->alphatest_state.bypass != surf->alphatest_bypass) {
1763 rctx->alphatest_state.bypass = surf->alphatest_bypass;
1764 rctx->alphatest_state.atom.dirty = true;
1765 }
1766 if (rctx->alphatest_state.cb0_export_16bpc != surf->export_16bpc) {
1767 rctx->alphatest_state.cb0_export_16bpc = surf->export_16bpc;
1768 rctx->alphatest_state.atom.dirty = true;
1769 }
1770 }
1771
1772 /* ZS buffer. */
1773 if (state->zsbuf) {
1774 surf = (struct r600_surface*)state->zsbuf;
1775
1776 r600_context_add_resource_size(ctx, state->zsbuf->texture);
1777
1778 if (!surf->depth_initialized) {
1779 evergreen_init_depth_surface(rctx, surf);
1780 }
1781
1782 if (state->zsbuf->format != rctx->poly_offset_state.zs_format) {
1783 rctx->poly_offset_state.zs_format = state->zsbuf->format;
1784 rctx->poly_offset_state.atom.dirty = true;
1785 }
1786
1787 if (rctx->db_state.rsurf != surf) {
1788 rctx->db_state.rsurf = surf;
1789 rctx->db_state.atom.dirty = true;
1790 rctx->db_misc_state.atom.dirty = true;
1791 }
1792 } else if (rctx->db_state.rsurf) {
1793 rctx->db_state.rsurf = NULL;
1794 rctx->db_state.atom.dirty = true;
1795 rctx->db_misc_state.atom.dirty = true;
1796 }
1797
1798 if (rctx->cb_misc_state.nr_cbufs != state->nr_cbufs) {
1799 rctx->cb_misc_state.nr_cbufs = state->nr_cbufs;
1800 rctx->cb_misc_state.atom.dirty = true;
1801 }
1802
1803 if (state->nr_cbufs == 0 && rctx->alphatest_state.bypass) {
1804 rctx->alphatest_state.bypass = false;
1805 rctx->alphatest_state.atom.dirty = true;
1806 }
1807
1808 log_samples = util_logbase2(rctx->framebuffer.nr_samples);
1809 if (rctx->chip_class == CAYMAN && rctx->db_misc_state.log_samples != log_samples) {
1810 rctx->db_misc_state.log_samples = log_samples;
1811 rctx->db_misc_state.atom.dirty = true;
1812 }
1813
1814 evergreen_update_db_shader_control(rctx);
1815
1816 /* Calculate the CS size. */
1817 rctx->framebuffer.atom.num_dw = 4; /* SCISSOR */
1818
1819 /* MSAA. */
1820 if (rctx->chip_class == EVERGREEN) {
1821 switch (rctx->framebuffer.nr_samples) {
1822 case 2:
1823 case 4:
1824 rctx->framebuffer.atom.num_dw += 6;
1825 break;
1826 case 8:
1827 rctx->framebuffer.atom.num_dw += 10;
1828 break;
1829 }
1830 rctx->framebuffer.atom.num_dw += 4;
1831 } else {
1832 switch (rctx->framebuffer.nr_samples) {
1833 case 2:
1834 case 4:
1835 rctx->framebuffer.atom.num_dw += 12;
1836 break;
1837 case 8:
1838 rctx->framebuffer.atom.num_dw += 16;
1839 break;
1840 case 16:
1841 rctx->framebuffer.atom.num_dw += 18;
1842 break;
1843 }
1844 rctx->framebuffer.atom.num_dw += 7;
1845 }
1846
1847 /* Colorbuffers. */
1848 rctx->framebuffer.atom.num_dw += state->nr_cbufs * 21;
1849 if (rctx->keep_tiling_flags)
1850 rctx->framebuffer.atom.num_dw += state->nr_cbufs * 2;
1851 rctx->framebuffer.atom.num_dw += (12 - state->nr_cbufs) * 3;
1852
1853 /* ZS buffer. */
1854 if (state->zsbuf) {
1855 rctx->framebuffer.atom.num_dw += 24;
1856 if (rctx->keep_tiling_flags)
1857 rctx->framebuffer.atom.num_dw += 2;
1858 } else if (rctx->screen->info.drm_minor >= 18) {
1859 rctx->framebuffer.atom.num_dw += 4;
1860 }
1861
1862 rctx->framebuffer.atom.dirty = true;
1863 }
1864
1865 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
1866 (((s0x) & 0xf) | (((s0y) & 0xf) << 4) | \
1867 (((s1x) & 0xf) << 8) | (((s1y) & 0xf) << 12) | \
1868 (((s2x) & 0xf) << 16) | (((s2y) & 0xf) << 20) | \
1869 (((s3x) & 0xf) << 24) | (((s3y) & 0xf) << 28))
1870
1871 static void evergreen_emit_msaa_state(struct r600_context *rctx, int nr_samples)
1872 {
1873 /* 2xMSAA
1874 * There are two locations (-4, 4), (4, -4). */
1875 static uint32_t sample_locs_2x[] = {
1876 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1877 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1878 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1879 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1880 };
1881 static unsigned max_dist_2x = 4;
1882 /* 4xMSAA
1883 * There are 4 locations: (-2, -2), (2, 2), (-6, 6), (6, -6). */
1884 static uint32_t sample_locs_4x[] = {
1885 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1886 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1887 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1888 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1889 };
1890 static unsigned max_dist_4x = 6;
1891 /* 8xMSAA */
1892 static uint32_t sample_locs_8x[] = {
1893 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1894 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1895 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1896 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1897 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1898 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1899 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1900 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1901 };
1902 static unsigned max_dist_8x = 7;
1903
1904 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
1905 unsigned max_dist = 0;
1906
1907 switch (nr_samples) {
1908 default:
1909 nr_samples = 0;
1910 break;
1911 case 2:
1912 r600_write_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, Elements(sample_locs_2x));
1913 r600_write_array(cs, Elements(sample_locs_2x), sample_locs_2x);
1914 max_dist = max_dist_2x;
1915 break;
1916 case 4:
1917 r600_write_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, Elements(sample_locs_4x));
1918 r600_write_array(cs, Elements(sample_locs_4x), sample_locs_4x);
1919 max_dist = max_dist_4x;
1920 break;
1921 case 8:
1922 r600_write_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, Elements(sample_locs_8x));
1923 r600_write_array(cs, Elements(sample_locs_8x), sample_locs_8x);
1924 max_dist = max_dist_8x;
1925 break;
1926 }
1927
1928 if (nr_samples > 1) {
1929 r600_write_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1930 r600_write_value(cs, S_028C00_LAST_PIXEL(1) |
1931 S_028C00_EXPAND_LINE_WIDTH(1)); /* R_028C00_PA_SC_LINE_CNTL */
1932 r600_write_value(cs, S_028C04_MSAA_NUM_SAMPLES(util_logbase2(nr_samples)) |
1933 S_028C04_MAX_SAMPLE_DIST(max_dist)); /* R_028C04_PA_SC_AA_CONFIG */
1934 } else {
1935 r600_write_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1936 r600_write_value(cs, S_028C00_LAST_PIXEL(1)); /* R_028C00_PA_SC_LINE_CNTL */
1937 r600_write_value(cs, 0); /* R_028C04_PA_SC_AA_CONFIG */
1938 }
1939 }
1940
1941 static void cayman_emit_msaa_state(struct r600_context *rctx, int nr_samples)
1942 {
1943 /* 2xMSAA
1944 * There are two locations (-4, 4), (4, -4). */
1945 static uint32_t sample_locs_2x[] = {
1946 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1947 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1948 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1949 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1950 };
1951 static unsigned max_dist_2x = 4;
1952 /* 4xMSAA
1953 * There are 4 locations: (-2, -2), (2, 2), (-6, 6), (6, -6). */
1954 static uint32_t sample_locs_4x[] = {
1955 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1956 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1957 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1958 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1959 };
1960 static unsigned max_dist_4x = 6;
1961 /* 8xMSAA */
1962 static uint32_t sample_locs_8x[] = {
1963 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1964 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1965 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1966 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1967 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1968 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1969 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1970 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1971 };
1972 static unsigned max_dist_8x = 8;
1973 /* 16xMSAA */
1974 static uint32_t sample_locs_16x[] = {
1975 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1976 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1977 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1978 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1979 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1980 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1981 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1982 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1983 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1984 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1985 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1986 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1987 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1988 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1989 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1990 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1991 };
1992 static unsigned max_dist_16x = 8;
1993
1994 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
1995 unsigned max_dist = 0;
1996
1997 switch (nr_samples) {
1998 default:
1999 nr_samples = 0;
2000 break;
2001 case 2:
2002 r600_write_context_reg(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_2x[0]);
2003 r600_write_context_reg(cs, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_2x[1]);
2004 r600_write_context_reg(cs, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_2x[2]);
2005 r600_write_context_reg(cs, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_2x[3]);
2006 max_dist = max_dist_2x;
2007 break;
2008 case 4:
2009 r600_write_context_reg(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_4x[0]);
2010 r600_write_context_reg(cs, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_4x[1]);
2011 r600_write_context_reg(cs, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_4x[2]);
2012 r600_write_context_reg(cs, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_4x[3]);
2013 max_dist = max_dist_4x;
2014 break;
2015 case 8:
2016 r600_write_context_reg_seq(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 14);
2017 r600_write_value(cs, sample_locs_8x[0]);
2018 r600_write_value(cs, sample_locs_8x[4]);
2019 r600_write_value(cs, 0);
2020 r600_write_value(cs, 0);
2021 r600_write_value(cs, sample_locs_8x[1]);
2022 r600_write_value(cs, sample_locs_8x[5]);
2023 r600_write_value(cs, 0);
2024 r600_write_value(cs, 0);
2025 r600_write_value(cs, sample_locs_8x[2]);
2026 r600_write_value(cs, sample_locs_8x[6]);
2027 r600_write_value(cs, 0);
2028 r600_write_value(cs, 0);
2029 r600_write_value(cs, sample_locs_8x[3]);
2030 r600_write_value(cs, sample_locs_8x[7]);
2031 max_dist = max_dist_8x;
2032 break;
2033 case 16:
2034 r600_write_context_reg_seq(cs, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 16);
2035 r600_write_value(cs, sample_locs_16x[0]);
2036 r600_write_value(cs, sample_locs_16x[4]);
2037 r600_write_value(cs, sample_locs_16x[8]);
2038 r600_write_value(cs, sample_locs_16x[12]);
2039 r600_write_value(cs, sample_locs_16x[1]);
2040 r600_write_value(cs, sample_locs_16x[5]);
2041 r600_write_value(cs, sample_locs_16x[9]);
2042 r600_write_value(cs, sample_locs_16x[13]);
2043 r600_write_value(cs, sample_locs_16x[2]);
2044 r600_write_value(cs, sample_locs_16x[6]);
2045 r600_write_value(cs, sample_locs_16x[10]);
2046 r600_write_value(cs, sample_locs_16x[14]);
2047 r600_write_value(cs, sample_locs_16x[3]);
2048 r600_write_value(cs, sample_locs_16x[7]);
2049 r600_write_value(cs, sample_locs_16x[11]);
2050 r600_write_value(cs, sample_locs_16x[15]);
2051 max_dist = max_dist_16x;
2052 break;
2053 }
2054
2055 if (nr_samples > 1) {
2056 unsigned log_samples = util_logbase2(nr_samples);
2057
2058 r600_write_context_reg_seq(cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
2059 r600_write_value(cs, S_028C00_LAST_PIXEL(1) |
2060 S_028C00_EXPAND_LINE_WIDTH(1)); /* CM_R_028BDC_PA_SC_LINE_CNTL */
2061 r600_write_value(cs, S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
2062 S_028BE0_MAX_SAMPLE_DIST(max_dist) |
2063 S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples)); /* CM_R_028BE0_PA_SC_AA_CONFIG */
2064
2065 r600_write_context_reg(cs, CM_R_028804_DB_EQAA,
2066 S_028804_MAX_ANCHOR_SAMPLES(log_samples) |
2067 S_028804_PS_ITER_SAMPLES(log_samples) |
2068 S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples) |
2069 S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples) |
2070 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
2071 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
2072 } else {
2073 r600_write_context_reg_seq(cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
2074 r600_write_value(cs, S_028C00_LAST_PIXEL(1)); /* CM_R_028BDC_PA_SC_LINE_CNTL */
2075 r600_write_value(cs, 0); /* CM_R_028BE0_PA_SC_AA_CONFIG */
2076
2077 r600_write_context_reg(cs, CM_R_028804_DB_EQAA,
2078 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
2079 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
2080 }
2081 }
2082
2083 static void evergreen_emit_framebuffer_state(struct r600_context *rctx, struct r600_atom *atom)
2084 {
2085 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2086 struct pipe_framebuffer_state *state = &rctx->framebuffer.state;
2087 unsigned nr_cbufs = state->nr_cbufs;
2088 unsigned i, tl, br;
2089
2090 /* XXX support more colorbuffers once we need them */
2091 assert(nr_cbufs <= 8);
2092 if (nr_cbufs > 8)
2093 nr_cbufs = 8;
2094
2095 /* Colorbuffers. */
2096 for (i = 0; i < nr_cbufs; i++) {
2097 struct r600_surface *cb = (struct r600_surface*)state->cbufs[i];
2098 unsigned reloc = r600_context_bo_reloc(rctx,
2099 &rctx->rings.gfx,
2100 (struct r600_resource*)cb->base.texture,
2101 RADEON_USAGE_READWRITE);
2102
2103 r600_write_context_reg_seq(cs, R_028C60_CB_COLOR0_BASE + i * 0x3C, 11);
2104 r600_write_value(cs, cb->cb_color_base); /* R_028C60_CB_COLOR0_BASE */
2105 r600_write_value(cs, cb->cb_color_pitch); /* R_028C64_CB_COLOR0_PITCH */
2106 r600_write_value(cs, cb->cb_color_slice); /* R_028C68_CB_COLOR0_SLICE */
2107 r600_write_value(cs, cb->cb_color_view); /* R_028C6C_CB_COLOR0_VIEW */
2108 r600_write_value(cs, cb->cb_color_info); /* R_028C70_CB_COLOR0_INFO */
2109 r600_write_value(cs, cb->cb_color_attrib); /* R_028C74_CB_COLOR0_ATTRIB */
2110 r600_write_value(cs, cb->cb_color_dim); /* R_028C78_CB_COLOR0_DIM */
2111 r600_write_value(cs, cb->cb_color_cmask); /* R_028C7C_CB_COLOR0_CMASK */
2112 r600_write_value(cs, cb->cb_color_cmask_slice); /* R_028C80_CB_COLOR0_CMASK_SLICE */
2113 r600_write_value(cs, cb->cb_color_fmask); /* R_028C84_CB_COLOR0_FMASK */
2114 r600_write_value(cs, cb->cb_color_fmask_slice); /* R_028C88_CB_COLOR0_FMASK_SLICE */
2115
2116 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C60_CB_COLOR0_BASE */
2117 r600_write_value(cs, reloc);
2118
2119 if (!rctx->keep_tiling_flags) {
2120 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C70_CB_COLOR0_INFO */
2121 r600_write_value(cs, reloc);
2122 }
2123
2124 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C74_CB_COLOR0_ATTRIB */
2125 r600_write_value(cs, reloc);
2126
2127 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C7C_CB_COLOR0_CMASK */
2128 r600_write_value(cs, reloc);
2129
2130 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C84_CB_COLOR0_FMASK */
2131 r600_write_value(cs, reloc);
2132 }
2133 /* set CB_COLOR1_INFO for possible dual-src blending */
2134 if (i == 1 && !((struct r600_texture*)state->cbufs[0]->texture)->is_rat) {
2135 r600_write_context_reg(cs, R_028C70_CB_COLOR0_INFO + 1 * 0x3C,
2136 ((struct r600_surface*)state->cbufs[0])->cb_color_info);
2137
2138 if (!rctx->keep_tiling_flags) {
2139 unsigned reloc = r600_context_bo_reloc(rctx,
2140 &rctx->rings.gfx,
2141 (struct r600_resource*)state->cbufs[0]->texture,
2142 RADEON_USAGE_READWRITE);
2143
2144 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C70_CB_COLOR0_INFO */
2145 r600_write_value(cs, reloc);
2146 }
2147 i++;
2148 }
2149 if (rctx->keep_tiling_flags) {
2150 for (; i < 8 ; i++) {
2151 r600_write_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C, 0);
2152 }
2153 for (; i < 12; i++) {
2154 r600_write_context_reg(cs, R_028E50_CB_COLOR8_INFO + (i - 8) * 0x1C, 0);
2155 }
2156 }
2157
2158 /* ZS buffer. */
2159 if (state->zsbuf) {
2160 struct r600_surface *zb = (struct r600_surface*)state->zsbuf;
2161 unsigned reloc = r600_context_bo_reloc(rctx,
2162 &rctx->rings.gfx,
2163 (struct r600_resource*)state->zsbuf->texture,
2164 RADEON_USAGE_READWRITE);
2165
2166 r600_write_context_reg(cs, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
2167 zb->pa_su_poly_offset_db_fmt_cntl);
2168 r600_write_context_reg(cs, R_028008_DB_DEPTH_VIEW, zb->db_depth_view);
2169
2170 r600_write_context_reg_seq(cs, R_028040_DB_Z_INFO, 8);
2171 r600_write_value(cs, zb->db_depth_info); /* R_028040_DB_Z_INFO */
2172 r600_write_value(cs, zb->db_stencil_info); /* R_028044_DB_STENCIL_INFO */
2173 r600_write_value(cs, zb->db_depth_base); /* R_028048_DB_Z_READ_BASE */
2174 r600_write_value(cs, zb->db_stencil_base); /* R_02804C_DB_STENCIL_READ_BASE */
2175 r600_write_value(cs, zb->db_depth_base); /* R_028050_DB_Z_WRITE_BASE */
2176 r600_write_value(cs, zb->db_stencil_base); /* R_028054_DB_STENCIL_WRITE_BASE */
2177 r600_write_value(cs, zb->db_depth_size); /* R_028058_DB_DEPTH_SIZE */
2178 r600_write_value(cs, zb->db_depth_slice); /* R_02805C_DB_DEPTH_SLICE */
2179
2180 if (!rctx->keep_tiling_flags) {
2181 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028040_DB_Z_INFO */
2182 r600_write_value(cs, reloc);
2183 }
2184
2185 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028048_DB_Z_READ_BASE */
2186 r600_write_value(cs, reloc);
2187
2188 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_02804C_DB_STENCIL_READ_BASE */
2189 r600_write_value(cs, reloc);
2190
2191 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028050_DB_Z_WRITE_BASE */
2192 r600_write_value(cs, reloc);
2193
2194 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028054_DB_STENCIL_WRITE_BASE */
2195 r600_write_value(cs, reloc);
2196 } else if (rctx->screen->info.drm_minor >= 18) {
2197 /* DRM 2.6.18 allows the INVALID format to disable depth/stencil.
2198 * Older kernels are out of luck. */
2199 r600_write_context_reg_seq(cs, R_028040_DB_Z_INFO, 2);
2200 r600_write_value(cs, S_028040_FORMAT(V_028040_Z_INVALID)); /* R_028040_DB_Z_INFO */
2201 r600_write_value(cs, S_028044_FORMAT(V_028044_STENCIL_INVALID)); /* R_028044_DB_STENCIL_INFO */
2202 }
2203
2204 /* Framebuffer dimensions. */
2205 evergreen_get_scissor_rect(rctx, 0, 0, state->width, state->height, &tl, &br);
2206
2207 r600_write_context_reg_seq(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL, 2);
2208 r600_write_value(cs, tl); /* R_028204_PA_SC_WINDOW_SCISSOR_TL */
2209 r600_write_value(cs, br); /* R_028208_PA_SC_WINDOW_SCISSOR_BR */
2210
2211 if (rctx->chip_class == EVERGREEN) {
2212 evergreen_emit_msaa_state(rctx, rctx->framebuffer.nr_samples);
2213 } else {
2214 cayman_emit_msaa_state(rctx, rctx->framebuffer.nr_samples);
2215 }
2216 }
2217
2218 static void evergreen_emit_polygon_offset(struct r600_context *rctx, struct r600_atom *a)
2219 {
2220 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2221 struct r600_poly_offset_state *state = (struct r600_poly_offset_state*)a;
2222 float offset_units = state->offset_units;
2223 float offset_scale = state->offset_scale;
2224
2225 switch (state->zs_format) {
2226 case PIPE_FORMAT_Z24X8_UNORM:
2227 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
2228 case PIPE_FORMAT_X8Z24_UNORM:
2229 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2230 offset_units *= 2.0f;
2231 break;
2232 case PIPE_FORMAT_Z16_UNORM:
2233 offset_units *= 4.0f;
2234 break;
2235 default:;
2236 }
2237
2238 r600_write_context_reg_seq(cs, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE, 4);
2239 r600_write_value(cs, fui(offset_scale));
2240 r600_write_value(cs, fui(offset_units));
2241 r600_write_value(cs, fui(offset_scale));
2242 r600_write_value(cs, fui(offset_units));
2243 }
2244
2245 static void evergreen_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom)
2246 {
2247 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2248 struct r600_cb_misc_state *a = (struct r600_cb_misc_state*)atom;
2249 unsigned fb_colormask = (1ULL << ((unsigned)a->nr_cbufs * 4)) - 1;
2250 unsigned ps_colormask = (1ULL << ((unsigned)a->nr_ps_color_outputs * 4)) - 1;
2251
2252 r600_write_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
2253 r600_write_value(cs, a->blend_colormask & fb_colormask); /* R_028238_CB_TARGET_MASK */
2254 /* Always enable the first colorbuffer in CB_SHADER_MASK. This
2255 * will assure that the alpha-test will work even if there is
2256 * no colorbuffer bound. */
2257 r600_write_value(cs, 0xf | (a->dual_src_blend ? ps_colormask : 0) | fb_colormask); /* R_02823C_CB_SHADER_MASK */
2258 }
2259
2260 static void evergreen_emit_db_state(struct r600_context *rctx, struct r600_atom *atom)
2261 {
2262 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2263 struct r600_db_state *a = (struct r600_db_state*)atom;
2264
2265 if (a->rsurf && a->rsurf->htile_enabled) {
2266 struct r600_texture *rtex = (struct r600_texture *)a->rsurf->base.texture;
2267 unsigned reloc_idx;
2268
2269 r600_write_context_reg(cs, R_02802C_DB_DEPTH_CLEAR, fui(rtex->depth_clear));
2270 r600_write_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, a->rsurf->db_htile_surface);
2271 r600_write_context_reg(cs, R_028AC8_DB_PRELOAD_CONTROL, a->rsurf->db_preload_control);
2272 r600_write_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, a->rsurf->db_htile_data_base);
2273 reloc_idx = r600_context_bo_reloc(rctx, &rctx->rings.gfx, rtex->htile, RADEON_USAGE_READWRITE);
2274 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
2275 cs->buf[cs->cdw++] = reloc_idx;
2276 } else {
2277 r600_write_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, 0);
2278 r600_write_context_reg(cs, R_028AC8_DB_PRELOAD_CONTROL, 0);
2279 }
2280 }
2281
2282 static void evergreen_emit_db_misc_state(struct r600_context *rctx, struct r600_atom *atom)
2283 {
2284 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2285 struct r600_db_misc_state *a = (struct r600_db_misc_state*)atom;
2286 unsigned db_render_control = 0;
2287 unsigned db_count_control = 0;
2288 unsigned db_render_override =
2289 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE) |
2290 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE);
2291
2292 if (a->occlusion_query_enabled) {
2293 db_count_control |= S_028004_PERFECT_ZPASS_COUNTS(1);
2294 if (rctx->chip_class == CAYMAN) {
2295 db_count_control |= S_028004_SAMPLE_RATE(a->log_samples);
2296 }
2297 db_render_override |= S_02800C_NOOP_CULL_DISABLE(1);
2298 }
2299 /* FIXME we should be able to use hyperz even if we are not writing to
2300 * zbuffer but somehow this trigger GPU lockup. See :
2301 *
2302 * https://bugs.freedesktop.org/show_bug.cgi?id=60848
2303 *
2304 * Disable hyperz for now if not writing to zbuffer.
2305 */
2306 if (rctx->db_state.rsurf && rctx->db_state.rsurf->htile_enabled && rctx->zwritemask) {
2307 /* FORCE_OFF means HiZ/HiS are determined by DB_SHADER_CONTROL */
2308 db_render_override |= S_02800C_FORCE_HIZ_ENABLE(V_02800C_FORCE_OFF);
2309 /* This is to fix a lockup when hyperz and alpha test are enabled at
2310 * the same time somehow GPU get confuse on which order to pick for
2311 * z test
2312 */
2313 if (rctx->alphatest_state.sx_alpha_test_control) {
2314 db_render_override |= S_02800C_FORCE_SHADER_Z_ORDER(1);
2315 }
2316 } else {
2317 db_render_override |= S_02800C_FORCE_HIZ_ENABLE(V_02800C_FORCE_DISABLE);
2318 }
2319 if (a->flush_depthstencil_through_cb) {
2320 assert(a->copy_depth || a->copy_stencil);
2321
2322 db_render_control |= S_028000_DEPTH_COPY_ENABLE(a->copy_depth) |
2323 S_028000_STENCIL_COPY_ENABLE(a->copy_stencil) |
2324 S_028000_COPY_CENTROID(1) |
2325 S_028000_COPY_SAMPLE(a->copy_sample);
2326 } else if (a->flush_depthstencil_in_place) {
2327 db_render_control |= S_028000_DEPTH_COMPRESS_DISABLE(1) |
2328 S_028000_STENCIL_COMPRESS_DISABLE(1);
2329 db_render_override |= S_02800C_DISABLE_PIXEL_RATE_TILES(1);
2330 }
2331 if (a->htile_clear) {
2332 /* FIXME we might want to disable cliprect here */
2333 db_render_control |= S_028000_DEPTH_CLEAR_ENABLE(1);
2334 }
2335
2336 r600_write_context_reg_seq(cs, R_028000_DB_RENDER_CONTROL, 2);
2337 r600_write_value(cs, db_render_control); /* R_028000_DB_RENDER_CONTROL */
2338 r600_write_value(cs, db_count_control); /* R_028004_DB_COUNT_CONTROL */
2339 r600_write_context_reg(cs, R_02800C_DB_RENDER_OVERRIDE, db_render_override);
2340 r600_write_context_reg(cs, R_02880C_DB_SHADER_CONTROL, a->db_shader_control);
2341 }
2342
2343 static void evergreen_emit_vertex_buffers(struct r600_context *rctx,
2344 struct r600_vertexbuf_state *state,
2345 unsigned resource_offset,
2346 unsigned pkt_flags)
2347 {
2348 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2349 uint32_t dirty_mask = state->dirty_mask;
2350
2351 while (dirty_mask) {
2352 struct pipe_vertex_buffer *vb;
2353 struct r600_resource *rbuffer;
2354 uint64_t va;
2355 unsigned buffer_index = u_bit_scan(&dirty_mask);
2356
2357 vb = &state->vb[buffer_index];
2358 rbuffer = (struct r600_resource*)vb->buffer;
2359 assert(rbuffer);
2360
2361 va = r600_resource_va(&rctx->screen->screen, &rbuffer->b.b);
2362 va += vb->buffer_offset;
2363
2364 /* fetch resources start at index 992 */
2365 r600_write_value(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags);
2366 r600_write_value(cs, (resource_offset + buffer_index) * 8);
2367 r600_write_value(cs, va); /* RESOURCEi_WORD0 */
2368 r600_write_value(cs, rbuffer->buf->size - vb->buffer_offset - 1); /* RESOURCEi_WORD1 */
2369 r600_write_value(cs, /* RESOURCEi_WORD2 */
2370 S_030008_ENDIAN_SWAP(r600_endian_swap(32)) |
2371 S_030008_STRIDE(vb->stride) |
2372 S_030008_BASE_ADDRESS_HI(va >> 32UL));
2373 r600_write_value(cs, /* RESOURCEi_WORD3 */
2374 S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) |
2375 S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) |
2376 S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) |
2377 S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W));
2378 r600_write_value(cs, 0); /* RESOURCEi_WORD4 */
2379 r600_write_value(cs, 0); /* RESOURCEi_WORD5 */
2380 r600_write_value(cs, 0); /* RESOURCEi_WORD6 */
2381 r600_write_value(cs, 0xc0000000); /* RESOURCEi_WORD7 */
2382
2383 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags);
2384 r600_write_value(cs, r600_context_bo_reloc(rctx, &rctx->rings.gfx, rbuffer, RADEON_USAGE_READ));
2385 }
2386 state->dirty_mask = 0;
2387 }
2388
2389 static void evergreen_fs_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom * atom)
2390 {
2391 evergreen_emit_vertex_buffers(rctx, &rctx->vertex_buffer_state, 992, 0);
2392 }
2393
2394 static void evergreen_cs_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom * atom)
2395 {
2396 evergreen_emit_vertex_buffers(rctx, &rctx->cs_vertex_buffer_state, 816,
2397 RADEON_CP_PACKET3_COMPUTE_MODE);
2398 }
2399
2400 static void evergreen_emit_constant_buffers(struct r600_context *rctx,
2401 struct r600_constbuf_state *state,
2402 unsigned buffer_id_base,
2403 unsigned reg_alu_constbuf_size,
2404 unsigned reg_alu_const_cache)
2405 {
2406 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2407 uint32_t dirty_mask = state->dirty_mask;
2408
2409 while (dirty_mask) {
2410 struct pipe_constant_buffer *cb;
2411 struct r600_resource *rbuffer;
2412 uint64_t va;
2413 unsigned buffer_index = ffs(dirty_mask) - 1;
2414
2415 cb = &state->cb[buffer_index];
2416 rbuffer = (struct r600_resource*)cb->buffer;
2417 assert(rbuffer);
2418
2419 va = r600_resource_va(&rctx->screen->screen, &rbuffer->b.b);
2420 va += cb->buffer_offset;
2421
2422 r600_write_context_reg(cs, reg_alu_constbuf_size + buffer_index * 4,
2423 ALIGN_DIVUP(cb->buffer_size >> 4, 16));
2424 r600_write_context_reg(cs, reg_alu_const_cache + buffer_index * 4, va >> 8);
2425
2426 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0));
2427 r600_write_value(cs, r600_context_bo_reloc(rctx, &rctx->rings.gfx, rbuffer, RADEON_USAGE_READ));
2428
2429 r600_write_value(cs, PKT3(PKT3_SET_RESOURCE, 8, 0));
2430 r600_write_value(cs, (buffer_id_base + buffer_index) * 8);
2431 r600_write_value(cs, va); /* RESOURCEi_WORD0 */
2432 r600_write_value(cs, rbuffer->buf->size - cb->buffer_offset - 1); /* RESOURCEi_WORD1 */
2433 r600_write_value(cs, /* RESOURCEi_WORD2 */
2434 S_030008_ENDIAN_SWAP(r600_endian_swap(32)) |
2435 S_030008_STRIDE(16) |
2436 S_030008_BASE_ADDRESS_HI(va >> 32UL));
2437 r600_write_value(cs, /* RESOURCEi_WORD3 */
2438 S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) |
2439 S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) |
2440 S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) |
2441 S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W));
2442 r600_write_value(cs, 0); /* RESOURCEi_WORD4 */
2443 r600_write_value(cs, 0); /* RESOURCEi_WORD5 */
2444 r600_write_value(cs, 0); /* RESOURCEi_WORD6 */
2445 r600_write_value(cs, 0xc0000000); /* RESOURCEi_WORD7 */
2446
2447 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0));
2448 r600_write_value(cs, r600_context_bo_reloc(rctx, &rctx->rings.gfx, rbuffer, RADEON_USAGE_READ));
2449
2450 dirty_mask &= ~(1 << buffer_index);
2451 }
2452 state->dirty_mask = 0;
2453 }
2454
2455 static void evergreen_emit_vs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2456 {
2457 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX], 176,
2458 R_028180_ALU_CONST_BUFFER_SIZE_VS_0,
2459 R_028980_ALU_CONST_CACHE_VS_0);
2460 }
2461
2462 static void evergreen_emit_gs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2463 {
2464 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY], 336,
2465 R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0,
2466 R_0289C0_ALU_CONST_CACHE_GS_0);
2467 }
2468
2469 static void evergreen_emit_ps_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2470 {
2471 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT], 0,
2472 R_028140_ALU_CONST_BUFFER_SIZE_PS_0,
2473 R_028940_ALU_CONST_CACHE_PS_0);
2474 }
2475
2476 static void evergreen_emit_sampler_views(struct r600_context *rctx,
2477 struct r600_samplerview_state *state,
2478 unsigned resource_id_base)
2479 {
2480 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2481 uint32_t dirty_mask = state->dirty_mask;
2482
2483 while (dirty_mask) {
2484 struct r600_pipe_sampler_view *rview;
2485 unsigned resource_index = u_bit_scan(&dirty_mask);
2486 unsigned reloc;
2487
2488 rview = state->views[resource_index];
2489 assert(rview);
2490
2491 r600_write_value(cs, PKT3(PKT3_SET_RESOURCE, 8, 0));
2492 r600_write_value(cs, (resource_id_base + resource_index) * 8);
2493 r600_write_array(cs, 8, rview->tex_resource_words);
2494
2495 reloc = r600_context_bo_reloc(rctx, &rctx->rings.gfx, rview->tex_resource,
2496 RADEON_USAGE_READ);
2497 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0));
2498 r600_write_value(cs, reloc);
2499
2500 if (!rview->skip_mip_address_reloc) {
2501 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0));
2502 r600_write_value(cs, reloc);
2503 }
2504 }
2505 state->dirty_mask = 0;
2506 }
2507
2508 static void evergreen_emit_vs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2509 {
2510 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views, 176 + R600_MAX_CONST_BUFFERS);
2511 }
2512
2513 static void evergreen_emit_gs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2514 {
2515 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views, 336 + R600_MAX_CONST_BUFFERS);
2516 }
2517
2518 static void evergreen_emit_ps_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2519 {
2520 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views, R600_MAX_CONST_BUFFERS);
2521 }
2522
2523 static void evergreen_emit_sampler_states(struct r600_context *rctx,
2524 struct r600_textures_info *texinfo,
2525 unsigned resource_id_base,
2526 unsigned border_index_reg)
2527 {
2528 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2529 uint32_t dirty_mask = texinfo->states.dirty_mask;
2530
2531 while (dirty_mask) {
2532 struct r600_pipe_sampler_state *rstate;
2533 unsigned i = u_bit_scan(&dirty_mask);
2534
2535 rstate = texinfo->states.states[i];
2536 assert(rstate);
2537
2538 r600_write_value(cs, PKT3(PKT3_SET_SAMPLER, 3, 0));
2539 r600_write_value(cs, (resource_id_base + i) * 3);
2540 r600_write_array(cs, 3, rstate->tex_sampler_words);
2541
2542 if (rstate->border_color_use) {
2543 r600_write_config_reg_seq(cs, border_index_reg, 5);
2544 r600_write_value(cs, i);
2545 r600_write_array(cs, 4, rstate->border_color.ui);
2546 }
2547 }
2548 texinfo->states.dirty_mask = 0;
2549 }
2550
2551 static void evergreen_emit_vs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2552 {
2553 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 18, R_00A414_TD_VS_SAMPLER0_BORDER_INDEX);
2554 }
2555
2556 static void evergreen_emit_gs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2557 {
2558 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY], 36, R_00A428_TD_GS_SAMPLER0_BORDER_INDEX);
2559 }
2560
2561 static void evergreen_emit_ps_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2562 {
2563 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT], 0, R_00A400_TD_PS_SAMPLER0_BORDER_INDEX);
2564 }
2565
2566 static void evergreen_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
2567 {
2568 struct r600_sample_mask *s = (struct r600_sample_mask*)a;
2569 uint8_t mask = s->sample_mask;
2570
2571 r600_write_context_reg(rctx->rings.gfx.cs, R_028C3C_PA_SC_AA_MASK,
2572 mask | (mask << 8) | (mask << 16) | (mask << 24));
2573 }
2574
2575 static void cayman_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
2576 {
2577 struct r600_sample_mask *s = (struct r600_sample_mask*)a;
2578 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2579 uint16_t mask = s->sample_mask;
2580
2581 r600_write_context_reg_seq(cs, CM_R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0, 2);
2582 r600_write_value(cs, mask | (mask << 16)); /* X0Y0_X1Y0 */
2583 r600_write_value(cs, mask | (mask << 16)); /* X0Y1_X1Y1 */
2584 }
2585
2586 static void evergreen_emit_vertex_fetch_shader(struct r600_context *rctx, struct r600_atom *a)
2587 {
2588 struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
2589 struct r600_cso_state *state = (struct r600_cso_state*)a;
2590 struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state->cso;
2591
2592 r600_write_context_reg(cs, R_0288A4_SQ_PGM_START_FS,
2593 (r600_resource_va(rctx->context.screen, &shader->buffer->b.b) + shader->offset) >> 8);
2594 r600_write_value(cs, PKT3(PKT3_NOP, 0, 0));
2595 r600_write_value(cs, r600_context_bo_reloc(rctx, &rctx->rings.gfx, shader->buffer, RADEON_USAGE_READ));
2596 }
2597
2598 void cayman_init_common_regs(struct r600_command_buffer *cb,
2599 enum chip_class ctx_chip_class,
2600 enum radeon_family ctx_family,
2601 int ctx_drm_minor)
2602 {
2603 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 2);
2604 r600_store_value(cb, S_008C00_EXPORT_SRC_C(1)); /* R_008C00_SQ_CONFIG */
2605 /* always set the temp clauses */
2606 r600_store_value(cb, S_008C04_NUM_CLAUSE_TEMP_GPRS(4)); /* R_008C04_SQ_GPR_RESOURCE_MGMT_1 */
2607
2608 r600_store_config_reg_seq(cb, R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1, 2);
2609 r600_store_value(cb, 0); /* R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1 */
2610 r600_store_value(cb, 0); /* R_008C14_SQ_GLOBAL_GPR_RESOURCE_MGMT_2 */
2611
2612 r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, (1 << 8));
2613
2614 r600_store_context_reg(cb, R_028A4C_PA_SC_MODE_CNTL_1, 0);
2615
2616 r600_store_context_reg(cb, R_028354_SX_SURFACE_SYNC, S_028354_SURFACE_SYNC_MASK(0xf));
2617
2618 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2619 }
2620
2621 static void cayman_init_atom_start_cs(struct r600_context *rctx)
2622 {
2623 struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2624
2625 r600_init_command_buffer(cb, 256);
2626
2627 /* This must be first. */
2628 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2629 r600_store_value(cb, 0x80000000);
2630 r600_store_value(cb, 0x80000000);
2631
2632 /* We're setting config registers here. */
2633 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2634 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2635
2636 cayman_init_common_regs(cb, rctx->chip_class,
2637 rctx->family, rctx->screen->info.drm_minor);
2638
2639 r600_store_config_reg(cb, R_009100_SPI_CONFIG_CNTL, 0);
2640 r600_store_config_reg(cb, R_00913C_SPI_CONFIG_CNTL_1, S_00913C_VTX_DONE_DELAY(4));
2641
2642 r600_store_context_reg_seq(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 6);
2643 r600_store_value(cb, 0); /* R_028900_SQ_ESGS_RING_ITEMSIZE */
2644 r600_store_value(cb, 0); /* R_028904_SQ_GSVS_RING_ITEMSIZE */
2645 r600_store_value(cb, 0); /* R_028908_SQ_ESTMP_RING_ITEMSIZE */
2646 r600_store_value(cb, 0); /* R_02890C_SQ_GSTMP_RING_ITEMSIZE */
2647 r600_store_value(cb, 0); /* R_028910_SQ_VSTMP_RING_ITEMSIZE */
2648 r600_store_value(cb, 0); /* R_028914_SQ_PSTMP_RING_ITEMSIZE */
2649
2650 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4);
2651 r600_store_value(cb, 0); /* R_02891C_SQ_GS_VERT_ITEMSIZE */
2652 r600_store_value(cb, 0); /* R_028920_SQ_GS_VERT_ITEMSIZE_1 */
2653 r600_store_value(cb, 0); /* R_028924_SQ_GS_VERT_ITEMSIZE_2 */
2654 r600_store_value(cb, 0); /* R_028928_SQ_GS_VERT_ITEMSIZE_3 */
2655
2656 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
2657 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
2658 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
2659 r600_store_value(cb, 0); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
2660 r600_store_value(cb, 0); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
2661 r600_store_value(cb, 0); /* R_028A20_VGT_HOS_REUSE_DEPTH */
2662 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
2663 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
2664 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
2665 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
2666 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
2667 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
2668 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
2669 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE */
2670
2671 r600_store_context_reg_seq(cb, R_028B94_VGT_STRMOUT_CONFIG, 2);
2672 r600_store_value(cb, 0); /* R_028B94_VGT_STRMOUT_CONFIG */
2673 r600_store_value(cb, 0); /* R_028B98_VGT_STRMOUT_BUFFER_CONFIG */
2674
2675 r600_store_context_reg_seq(cb, R_028AB4_VGT_REUSE_OFF, 2);
2676 r600_store_value(cb, 0); /* R_028AB4_VGT_REUSE_OFF */
2677 r600_store_value(cb, 0); /* R_028AB8_VGT_VTX_CNT_EN */
2678
2679 r600_store_config_reg(cb, R_008A14_PA_CL_ENHANCE, (3 << 1) | 1);
2680
2681 r600_store_context_reg(cb, CM_R_028AA8_IA_MULTI_VGT_PARAM, S_028AA8_SWITCH_ON_EOP(1) | S_028AA8_PARTIAL_VS_WAVE_ON(1) | S_028AA8_PRIMGROUP_SIZE(63));
2682
2683 r600_store_context_reg_seq(cb, CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0, 2);
2684 r600_store_value(cb, 0x76543210); /* CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0 */
2685 r600_store_value(cb, 0xfedcba98); /* CM_R_028BD8_PA_SC_CENTROID_PRIORITY_1 */
2686
2687 r600_store_context_reg_seq(cb, CM_R_0288E8_SQ_LDS_ALLOC, 2);
2688 r600_store_value(cb, 0); /* CM_R_0288E8_SQ_LDS_ALLOC */
2689 r600_store_value(cb, 0); /* R_0288EC_SQ_LDS_ALLOC_PS */
2690
2691 r600_store_context_reg(cb, R_0288F0_SQ_VTX_SEMANTIC_CLEAR, ~0);
2692
2693 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
2694 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
2695 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
2696
2697 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
2698
2699 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
2700
2701 r600_store_context_reg(cb, R_0286DC_SPI_FOG_CNTL, 0);
2702
2703 r600_store_context_reg_seq(cb, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 3);
2704 r600_store_value(cb, 0); /* R_028AC0_DB_SRESULTS_COMPARE_STATE0 */
2705 r600_store_value(cb, 0); /* R_028AC4_DB_SRESULTS_COMPARE_STATE1 */
2706 r600_store_value(cb, 0); /* R_028AC8_DB_PRELOAD_CONTROL */
2707
2708 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
2709 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
2710
2711 r600_store_context_reg_seq(cb, R_0282D0_PA_SC_VPORT_ZMIN_0, 2);
2712 r600_store_value(cb, 0); /* R_0282D0_PA_SC_VPORT_ZMIN_0 */
2713 r600_store_value(cb, 0x3F800000); /* R_0282D4_PA_SC_VPORT_ZMAX_0 */
2714
2715 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
2716 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL, 0x0000043F);
2717 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
2718
2719 r600_store_context_reg_seq(cb, CM_R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, 4);
2720 r600_store_value(cb, 0x3F800000); /* CM_R_028BE8_PA_CL_GB_VERT_CLIP_ADJ */
2721 r600_store_value(cb, 0x3F800000); /* CM_R_028BEC_PA_CL_GB_VERT_DISC_ADJ */
2722 r600_store_value(cb, 0x3F800000); /* CM_R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ */
2723 r600_store_value(cb, 0x3F800000); /* CM_R_028BF4_PA_CL_GB_HORZ_DISC_ADJ */
2724
2725 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
2726 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
2727 r600_store_value(cb, S_028244_BR_X(16384) | S_028244_BR_Y(16384)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
2728
2729 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
2730 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
2731 r600_store_value(cb, S_028034_BR_X(16384) | S_028034_BR_Y(16384)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
2732
2733 r600_store_context_reg(cb, R_028848_SQ_PGM_RESOURCES_2_PS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2734 r600_store_context_reg(cb, R_028864_SQ_PGM_RESOURCES_2_VS, S_028864_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2735 r600_store_context_reg(cb, R_0288A8_SQ_PGM_RESOURCES_FS, 0);
2736
2737 /* to avoid GPU doing any preloading of constant from random address */
2738 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
2739 r600_store_value(cb, 0); /* R_028140_ALU_CONST_BUFFER_SIZE_PS_0 */
2740 r600_store_value(cb, 0);
2741 r600_store_value(cb, 0);
2742 r600_store_value(cb, 0);
2743 r600_store_value(cb, 0);
2744 r600_store_value(cb, 0);
2745 r600_store_value(cb, 0);
2746 r600_store_value(cb, 0);
2747 r600_store_value(cb, 0);
2748 r600_store_value(cb, 0);
2749 r600_store_value(cb, 0);
2750 r600_store_value(cb, 0);
2751 r600_store_value(cb, 0);
2752 r600_store_value(cb, 0);
2753 r600_store_value(cb, 0);
2754 r600_store_value(cb, 0);
2755
2756 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
2757 r600_store_value(cb, 0); /* R_028180_ALU_CONST_BUFFER_SIZE_VS_0 */
2758 r600_store_value(cb, 0);
2759 r600_store_value(cb, 0);
2760 r600_store_value(cb, 0);
2761 r600_store_value(cb, 0);
2762 r600_store_value(cb, 0);
2763 r600_store_value(cb, 0);
2764 r600_store_value(cb, 0);
2765 r600_store_value(cb, 0);
2766 r600_store_value(cb, 0);
2767 r600_store_value(cb, 0);
2768 r600_store_value(cb, 0);
2769 r600_store_value(cb, 0);
2770 r600_store_value(cb, 0);
2771 r600_store_value(cb, 0);
2772 r600_store_value(cb, 0);
2773
2774 if (rctx->screen->has_streamout) {
2775 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
2776 }
2777
2778 r600_store_context_reg(cb, R_028010_DB_RENDER_OVERRIDE2, 0);
2779 r600_store_context_reg(cb, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0);
2780 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
2781 r600_store_context_reg_seq(cb, R_0286E4_SPI_PS_IN_CONTROL_2, 2);
2782 r600_store_value(cb, 0); /* R_0286E4_SPI_PS_IN_CONTROL_2 */
2783 r600_store_value(cb, 0); /* R_0286E8_SPI_COMPUTE_INPUT_CNTL */
2784 r600_store_context_reg(cb, R_028B54_VGT_SHADER_STAGES_EN, 0);
2785
2786 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0, 0x01000FFF);
2787 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF);
2788 }
2789
2790 void evergreen_init_common_regs(struct r600_command_buffer *cb,
2791 enum chip_class ctx_chip_class,
2792 enum radeon_family ctx_family,
2793 int ctx_drm_minor)
2794 {
2795 int ps_prio;
2796 int vs_prio;
2797 int gs_prio;
2798 int es_prio;
2799
2800 int hs_prio;
2801 int cs_prio;
2802 int ls_prio;
2803
2804 int num_ps_gprs;
2805 int num_vs_gprs;
2806 int num_gs_gprs;
2807 int num_es_gprs;
2808 int num_hs_gprs;
2809 int num_ls_gprs;
2810 int num_temp_gprs;
2811
2812 unsigned tmp;
2813
2814 ps_prio = 0;
2815 vs_prio = 1;
2816 gs_prio = 2;
2817 es_prio = 3;
2818 hs_prio = 0;
2819 ls_prio = 0;
2820 cs_prio = 0;
2821
2822 num_ps_gprs = 93;
2823 num_vs_gprs = 46;
2824 num_temp_gprs = 4;
2825 num_gs_gprs = 31;
2826 num_es_gprs = 31;
2827 num_hs_gprs = 23;
2828 num_ls_gprs = 23;
2829
2830 tmp = 0;
2831 switch (ctx_family) {
2832 case CHIP_CEDAR:
2833 case CHIP_PALM:
2834 case CHIP_SUMO:
2835 case CHIP_SUMO2:
2836 case CHIP_CAICOS:
2837 break;
2838 default:
2839 tmp |= S_008C00_VC_ENABLE(1);
2840 break;
2841 }
2842 tmp |= S_008C00_EXPORT_SRC_C(1);
2843 tmp |= S_008C00_CS_PRIO(cs_prio);
2844 tmp |= S_008C00_LS_PRIO(ls_prio);
2845 tmp |= S_008C00_HS_PRIO(hs_prio);
2846 tmp |= S_008C00_PS_PRIO(ps_prio);
2847 tmp |= S_008C00_VS_PRIO(vs_prio);
2848 tmp |= S_008C00_GS_PRIO(gs_prio);
2849 tmp |= S_008C00_ES_PRIO(es_prio);
2850
2851 /* enable dynamic GPR resource management */
2852 if (ctx_drm_minor >= 7) {
2853 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 2);
2854 r600_store_value(cb, tmp); /* R_008C00_SQ_CONFIG */
2855 /* always set temp clauses */
2856 r600_store_value(cb, S_008C04_NUM_CLAUSE_TEMP_GPRS(num_temp_gprs)); /* R_008C04_SQ_GPR_RESOURCE_MGMT_1 */
2857 r600_store_config_reg_seq(cb, R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1, 2);
2858 r600_store_value(cb, 0); /* R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1 */
2859 r600_store_value(cb, 0); /* R_008C14_SQ_GLOBAL_GPR_RESOURCE_MGMT_2 */
2860 r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, (1 << 8));
2861 r600_store_context_reg(cb, R_028838_SQ_DYN_GPR_RESOURCE_LIMIT_1,
2862 S_028838_PS_GPRS(0x1e) |
2863 S_028838_VS_GPRS(0x1e) |
2864 S_028838_GS_GPRS(0x1e) |
2865 S_028838_ES_GPRS(0x1e) |
2866 S_028838_HS_GPRS(0x1e) |
2867 S_028838_LS_GPRS(0x1e)); /* workaround for hw issues with dyn gpr - must set all limits to 240 instead of 0, 0x1e == 240 / 8*/
2868 } else {
2869 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 4);
2870 r600_store_value(cb, tmp); /* R_008C00_SQ_CONFIG */
2871
2872 tmp = S_008C04_NUM_PS_GPRS(num_ps_gprs);
2873 tmp |= S_008C04_NUM_VS_GPRS(num_vs_gprs);
2874 tmp |= S_008C04_NUM_CLAUSE_TEMP_GPRS(num_temp_gprs);
2875 r600_store_value(cb, tmp); /* R_008C04_SQ_GPR_RESOURCE_MGMT_1 */
2876
2877 tmp = S_008C08_NUM_GS_GPRS(num_gs_gprs);
2878 tmp |= S_008C08_NUM_ES_GPRS(num_es_gprs);
2879 r600_store_value(cb, tmp); /* R_008C08_SQ_GPR_RESOURCE_MGMT_2 */
2880
2881 tmp = S_008C0C_NUM_HS_GPRS(num_hs_gprs);
2882 tmp |= S_008C0C_NUM_HS_GPRS(num_ls_gprs);
2883 r600_store_value(cb, tmp); /* R_008C0C_SQ_GPR_RESOURCE_MGMT_3 */
2884 }
2885
2886 r600_store_config_reg(cb, R_008E2C_SQ_LDS_RESOURCE_MGMT,
2887 S_008E2C_NUM_PS_LDS(0x1000) | S_008E2C_NUM_LS_LDS(0x1000));
2888
2889 r600_store_context_reg(cb, R_028A4C_PA_SC_MODE_CNTL_1, 0);
2890
2891 /* The cs checker requires this register to be set. */
2892 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2893
2894 r600_store_context_reg(cb, R_028354_SX_SURFACE_SYNC, S_028354_SURFACE_SYNC_MASK(0xf));
2895
2896 return;
2897 }
2898
2899 void evergreen_init_atom_start_cs(struct r600_context *rctx)
2900 {
2901 struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2902 int num_ps_threads;
2903 int num_vs_threads;
2904 int num_gs_threads;
2905 int num_es_threads;
2906 int num_hs_threads;
2907 int num_ls_threads;
2908
2909 int num_ps_stack_entries;
2910 int num_vs_stack_entries;
2911 int num_gs_stack_entries;
2912 int num_es_stack_entries;
2913 int num_hs_stack_entries;
2914 int num_ls_stack_entries;
2915 enum radeon_family family;
2916 unsigned tmp;
2917
2918 if (rctx->chip_class == CAYMAN) {
2919 cayman_init_atom_start_cs(rctx);
2920 return;
2921 }
2922
2923 r600_init_command_buffer(cb, 256);
2924
2925 /* This must be first. */
2926 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2927 r600_store_value(cb, 0x80000000);
2928 r600_store_value(cb, 0x80000000);
2929
2930 /* We're setting config registers here. */
2931 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2932 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2933
2934 evergreen_init_common_regs(cb, rctx->chip_class,
2935 rctx->family, rctx->screen->info.drm_minor);
2936
2937 family = rctx->family;
2938 switch (family) {
2939 case CHIP_CEDAR:
2940 default:
2941 num_ps_threads = 96;
2942 num_vs_threads = 16;
2943 num_gs_threads = 16;
2944 num_es_threads = 16;
2945 num_hs_threads = 16;
2946 num_ls_threads = 16;
2947 num_ps_stack_entries = 42;
2948 num_vs_stack_entries = 42;
2949 num_gs_stack_entries = 42;
2950 num_es_stack_entries = 42;
2951 num_hs_stack_entries = 42;
2952 num_ls_stack_entries = 42;
2953 break;
2954 case CHIP_REDWOOD:
2955 num_ps_threads = 128;
2956 num_vs_threads = 20;
2957 num_gs_threads = 20;
2958 num_es_threads = 20;
2959 num_hs_threads = 20;
2960 num_ls_threads = 20;
2961 num_ps_stack_entries = 42;
2962 num_vs_stack_entries = 42;
2963 num_gs_stack_entries = 42;
2964 num_es_stack_entries = 42;
2965 num_hs_stack_entries = 42;
2966 num_ls_stack_entries = 42;
2967 break;
2968 case CHIP_JUNIPER:
2969 num_ps_threads = 128;
2970 num_vs_threads = 20;
2971 num_gs_threads = 20;
2972 num_es_threads = 20;
2973 num_hs_threads = 20;
2974 num_ls_threads = 20;
2975 num_ps_stack_entries = 85;
2976 num_vs_stack_entries = 85;
2977 num_gs_stack_entries = 85;
2978 num_es_stack_entries = 85;
2979 num_hs_stack_entries = 85;
2980 num_ls_stack_entries = 85;
2981 break;
2982 case CHIP_CYPRESS:
2983 case CHIP_HEMLOCK:
2984 num_ps_threads = 128;
2985 num_vs_threads = 20;
2986 num_gs_threads = 20;
2987 num_es_threads = 20;
2988 num_hs_threads = 20;
2989 num_ls_threads = 20;
2990 num_ps_stack_entries = 85;
2991 num_vs_stack_entries = 85;
2992 num_gs_stack_entries = 85;
2993 num_es_stack_entries = 85;
2994 num_hs_stack_entries = 85;
2995 num_ls_stack_entries = 85;
2996 break;
2997 case CHIP_PALM:
2998 num_ps_threads = 96;
2999 num_vs_threads = 16;
3000 num_gs_threads = 16;
3001 num_es_threads = 16;
3002 num_hs_threads = 16;
3003 num_ls_threads = 16;
3004 num_ps_stack_entries = 42;
3005 num_vs_stack_entries = 42;
3006 num_gs_stack_entries = 42;
3007 num_es_stack_entries = 42;
3008 num_hs_stack_entries = 42;
3009 num_ls_stack_entries = 42;
3010 break;
3011 case CHIP_SUMO:
3012 num_ps_threads = 96;
3013 num_vs_threads = 25;
3014 num_gs_threads = 25;
3015 num_es_threads = 25;
3016 num_hs_threads = 25;
3017 num_ls_threads = 25;
3018 num_ps_stack_entries = 42;
3019 num_vs_stack_entries = 42;
3020 num_gs_stack_entries = 42;
3021 num_es_stack_entries = 42;
3022 num_hs_stack_entries = 42;
3023 num_ls_stack_entries = 42;
3024 break;
3025 case CHIP_SUMO2:
3026 num_ps_threads = 96;
3027 num_vs_threads = 25;
3028 num_gs_threads = 25;
3029 num_es_threads = 25;
3030 num_hs_threads = 25;
3031 num_ls_threads = 25;
3032 num_ps_stack_entries = 85;
3033 num_vs_stack_entries = 85;
3034 num_gs_stack_entries = 85;
3035 num_es_stack_entries = 85;
3036 num_hs_stack_entries = 85;
3037 num_ls_stack_entries = 85;
3038 break;
3039 case CHIP_BARTS:
3040 num_ps_threads = 128;
3041 num_vs_threads = 20;
3042 num_gs_threads = 20;
3043 num_es_threads = 20;
3044 num_hs_threads = 20;
3045 num_ls_threads = 20;
3046 num_ps_stack_entries = 85;
3047 num_vs_stack_entries = 85;
3048 num_gs_stack_entries = 85;
3049 num_es_stack_entries = 85;
3050 num_hs_stack_entries = 85;
3051 num_ls_stack_entries = 85;
3052 break;
3053 case CHIP_TURKS:
3054 num_ps_threads = 128;
3055 num_vs_threads = 20;
3056 num_gs_threads = 20;
3057 num_es_threads = 20;
3058 num_hs_threads = 20;
3059 num_ls_threads = 20;
3060 num_ps_stack_entries = 42;
3061 num_vs_stack_entries = 42;
3062 num_gs_stack_entries = 42;
3063 num_es_stack_entries = 42;
3064 num_hs_stack_entries = 42;
3065 num_ls_stack_entries = 42;
3066 break;
3067 case CHIP_CAICOS:
3068 num_ps_threads = 128;
3069 num_vs_threads = 10;
3070 num_gs_threads = 10;
3071 num_es_threads = 10;
3072 num_hs_threads = 10;
3073 num_ls_threads = 10;
3074 num_ps_stack_entries = 42;
3075 num_vs_stack_entries = 42;
3076 num_gs_stack_entries = 42;
3077 num_es_stack_entries = 42;
3078 num_hs_stack_entries = 42;
3079 num_ls_stack_entries = 42;
3080 break;
3081 }
3082
3083 tmp = S_008C18_NUM_PS_THREADS(num_ps_threads);
3084 tmp |= S_008C18_NUM_VS_THREADS(num_vs_threads);
3085 tmp |= S_008C18_NUM_GS_THREADS(num_gs_threads);
3086 tmp |= S_008C18_NUM_ES_THREADS(num_es_threads);
3087
3088 r600_store_config_reg_seq(cb, R_008C18_SQ_THREAD_RESOURCE_MGMT_1, 5);
3089 r600_store_value(cb, tmp); /* R_008C18_SQ_THREAD_RESOURCE_MGMT_1 */
3090
3091 tmp = S_008C1C_NUM_HS_THREADS(num_hs_threads);
3092 tmp |= S_008C1C_NUM_LS_THREADS(num_ls_threads);
3093 r600_store_value(cb, tmp); /* R_008C1C_SQ_THREAD_RESOURCE_MGMT_2 */
3094
3095 tmp = S_008C20_NUM_PS_STACK_ENTRIES(num_ps_stack_entries);
3096 tmp |= S_008C20_NUM_VS_STACK_ENTRIES(num_vs_stack_entries);
3097 r600_store_value(cb, tmp); /* R_008C20_SQ_STACK_RESOURCE_MGMT_1 */
3098
3099 tmp = S_008C24_NUM_GS_STACK_ENTRIES(num_gs_stack_entries);
3100 tmp |= S_008C24_NUM_ES_STACK_ENTRIES(num_es_stack_entries);
3101 r600_store_value(cb, tmp); /* R_008C24_SQ_STACK_RESOURCE_MGMT_2 */
3102
3103 tmp = S_008C28_NUM_HS_STACK_ENTRIES(num_hs_stack_entries);
3104 tmp |= S_008C28_NUM_LS_STACK_ENTRIES(num_ls_stack_entries);
3105 r600_store_value(cb, tmp); /* R_008C28_SQ_STACK_RESOURCE_MGMT_3 */
3106
3107 r600_store_config_reg(cb, R_009100_SPI_CONFIG_CNTL, 0);
3108 r600_store_config_reg(cb, R_00913C_SPI_CONFIG_CNTL_1, S_00913C_VTX_DONE_DELAY(4));
3109
3110 r600_store_context_reg_seq(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 6);
3111 r600_store_value(cb, 0); /* R_028900_SQ_ESGS_RING_ITEMSIZE */
3112 r600_store_value(cb, 0); /* R_028904_SQ_GSVS_RING_ITEMSIZE */
3113 r600_store_value(cb, 0); /* R_028908_SQ_ESTMP_RING_ITEMSIZE */
3114 r600_store_value(cb, 0); /* R_02890C_SQ_GSTMP_RING_ITEMSIZE */
3115 r600_store_value(cb, 0); /* R_028910_SQ_VSTMP_RING_ITEMSIZE */
3116 r600_store_value(cb, 0); /* R_028914_SQ_PSTMP_RING_ITEMSIZE */
3117
3118 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4);
3119 r600_store_value(cb, 0); /* R_02891C_SQ_GS_VERT_ITEMSIZE */
3120 r600_store_value(cb, 0); /* R_028920_SQ_GS_VERT_ITEMSIZE_1 */
3121 r600_store_value(cb, 0); /* R_028924_SQ_GS_VERT_ITEMSIZE_2 */
3122 r600_store_value(cb, 0); /* R_028928_SQ_GS_VERT_ITEMSIZE_3 */
3123
3124 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
3125 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
3126 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
3127 r600_store_value(cb, 0); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
3128 r600_store_value(cb, 0); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
3129 r600_store_value(cb, 0); /* R_028A20_VGT_HOS_REUSE_DEPTH */
3130 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
3131 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
3132 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
3133 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
3134 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
3135 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
3136 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
3137 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE */
3138
3139 r600_store_context_reg_seq(cb, R_028AB4_VGT_REUSE_OFF, 2);
3140 r600_store_value(cb, 0); /* R_028AB4_VGT_REUSE_OFF */
3141 r600_store_value(cb, 0); /* R_028AB8_VGT_VTX_CNT_EN */
3142
3143 r600_store_config_reg(cb, R_008A14_PA_CL_ENHANCE, (3 << 1) | 1);
3144
3145 r600_store_context_reg(cb, R_0288F0_SQ_VTX_SEMANTIC_CLEAR, ~0);
3146
3147 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
3148 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
3149 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
3150
3151 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
3152
3153 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
3154
3155 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
3156 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
3157 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
3158
3159 r600_store_context_reg_seq(cb, R_0282D0_PA_SC_VPORT_ZMIN_0, 2);
3160 r600_store_value(cb, 0); /* R_0282D0_PA_SC_VPORT_ZMIN_0 */
3161 r600_store_value(cb, 0x3F800000); /* R_0282D4_PA_SC_VPORT_ZMAX_0 */
3162
3163 r600_store_context_reg(cb, R_0286DC_SPI_FOG_CNTL, 0);
3164 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL, 0x0000043F);
3165 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
3166
3167 r600_store_context_reg_seq(cb, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 3);
3168 r600_store_value(cb, 0); /* R_028AC0_DB_SRESULTS_COMPARE_STATE0 */
3169 r600_store_value(cb, 0); /* R_028AC4_DB_SRESULTS_COMPARE_STATE1 */
3170 r600_store_value(cb, 0); /* R_028AC8_DB_PRELOAD_CONTROL */
3171
3172 r600_store_context_reg_seq(cb, R_028C0C_PA_CL_GB_VERT_CLIP_ADJ, 4);
3173 r600_store_value(cb, 0x3F800000); /* R_028C0C_PA_CL_GB_VERT_CLIP_ADJ */
3174 r600_store_value(cb, 0x3F800000); /* R_028C10_PA_CL_GB_VERT_DISC_ADJ */
3175 r600_store_value(cb, 0x3F800000); /* R_028C14_PA_CL_GB_HORZ_CLIP_ADJ */
3176 r600_store_value(cb, 0x3F800000); /* R_028C18_PA_CL_GB_HORZ_DISC_ADJ */
3177
3178 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
3179 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
3180 r600_store_value(cb, S_028244_BR_X(16384) | S_028244_BR_Y(16384)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
3181
3182 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
3183 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
3184 r600_store_value(cb, S_028034_BR_X(16384) | S_028034_BR_Y(16384)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
3185
3186 r600_store_context_reg(cb, R_028848_SQ_PGM_RESOURCES_2_PS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
3187 r600_store_context_reg(cb, R_028864_SQ_PGM_RESOURCES_2_VS, S_028864_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
3188 r600_store_context_reg(cb, R_0288A8_SQ_PGM_RESOURCES_FS, 0);
3189
3190 /* to avoid GPU doing any preloading of constant from random address */
3191 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
3192 r600_store_value(cb, 0); /* R_028140_ALU_CONST_BUFFER_SIZE_PS_0 */
3193 r600_store_value(cb, 0);
3194 r600_store_value(cb, 0);
3195 r600_store_value(cb, 0);
3196 r600_store_value(cb, 0);
3197 r600_store_value(cb, 0);
3198 r600_store_value(cb, 0);
3199 r600_store_value(cb, 0);
3200 r600_store_value(cb, 0);
3201 r600_store_value(cb, 0);
3202 r600_store_value(cb, 0);
3203 r600_store_value(cb, 0);
3204 r600_store_value(cb, 0);
3205 r600_store_value(cb, 0);
3206 r600_store_value(cb, 0);
3207 r600_store_value(cb, 0);
3208
3209 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
3210 r600_store_value(cb, 0); /* R_028180_ALU_CONST_BUFFER_SIZE_VS_0 */
3211 r600_store_value(cb, 0);
3212 r600_store_value(cb, 0);
3213 r600_store_value(cb, 0);
3214 r600_store_value(cb, 0);
3215 r600_store_value(cb, 0);
3216 r600_store_value(cb, 0);
3217 r600_store_value(cb, 0);
3218 r600_store_value(cb, 0);
3219 r600_store_value(cb, 0);
3220 r600_store_value(cb, 0);
3221 r600_store_value(cb, 0);
3222 r600_store_value(cb, 0);
3223 r600_store_value(cb, 0);
3224 r600_store_value(cb, 0);
3225 r600_store_value(cb, 0);
3226
3227 r600_store_context_reg_seq(cb, R_028B94_VGT_STRMOUT_CONFIG, 2);
3228 r600_store_value(cb, 0); /* R_028B94_VGT_STRMOUT_CONFIG */
3229 r600_store_value(cb, 0); /* R_028B98_VGT_STRMOUT_BUFFER_CONFIG */
3230
3231 if (rctx->screen->has_streamout) {
3232 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
3233 }
3234
3235 r600_store_context_reg(cb, R_028010_DB_RENDER_OVERRIDE2, 0);
3236 r600_store_context_reg(cb, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0);
3237 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
3238 r600_store_context_reg_seq(cb, R_0286E4_SPI_PS_IN_CONTROL_2, 2);
3239 r600_store_value(cb, 0); /* R_0286E4_SPI_PS_IN_CONTROL_2 */
3240 r600_store_value(cb, 0); /* R_0286E8_SPI_COMPUTE_INPUT_CNTL */
3241 r600_store_context_reg(cb, R_0288EC_SQ_LDS_ALLOC_PS, 0);
3242 r600_store_context_reg(cb, R_028B54_VGT_SHADER_STAGES_EN, 0);
3243
3244 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0, 0x01000FFF);
3245 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF);
3246 }
3247
3248 void evergreen_update_ps_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3249 {
3250 struct r600_context *rctx = (struct r600_context *)ctx;
3251 struct r600_command_buffer *cb = &shader->command_buffer;
3252 struct r600_shader *rshader = &shader->shader;
3253 unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1, db_shader_control = 0;
3254 int pos_index = -1, face_index = -1;
3255 int ninterp = 0;
3256 boolean have_linear = FALSE, have_centroid = FALSE, have_perspective = FALSE;
3257 unsigned spi_baryc_cntl, sid, tmp, num = 0;
3258 unsigned z_export = 0, stencil_export = 0;
3259 unsigned sprite_coord_enable = rctx->rasterizer ? rctx->rasterizer->sprite_coord_enable : 0;
3260 uint32_t spi_ps_input_cntl[32];
3261
3262 if (!cb->buf) {
3263 r600_init_command_buffer(cb, 64);
3264 } else {
3265 cb->num_dw = 0;
3266 }
3267
3268 for (i = 0; i < rshader->ninput; i++) {
3269 /* evergreen NUM_INTERP only contains values interpolated into the LDS,
3270 POSITION goes via GPRs from the SC so isn't counted */
3271 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
3272 pos_index = i;
3273 else if (rshader->input[i].name == TGSI_SEMANTIC_FACE)
3274 face_index = i;
3275 else {
3276 ninterp++;
3277 if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
3278 have_linear = TRUE;
3279 if (rshader->input[i].interpolate == TGSI_INTERPOLATE_PERSPECTIVE)
3280 have_perspective = TRUE;
3281 if (rshader->input[i].centroid)
3282 have_centroid = TRUE;
3283 }
3284
3285 sid = rshader->input[i].spi_sid;
3286
3287 if (sid) {
3288 tmp = S_028644_SEMANTIC(sid);
3289
3290 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION ||
3291 rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
3292 (rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR &&
3293 rctx->rasterizer && rctx->rasterizer->flatshade)) {
3294 tmp |= S_028644_FLAT_SHADE(1);
3295 }
3296
3297 if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
3298 (sprite_coord_enable & (1 << rshader->input[i].sid))) {
3299 tmp |= S_028644_PT_SPRITE_TEX(1);
3300 }
3301
3302 spi_ps_input_cntl[num++] = tmp;
3303 }
3304 }
3305
3306 r600_store_context_reg_seq(cb, R_028644_SPI_PS_INPUT_CNTL_0, num);
3307 r600_store_array(cb, num, spi_ps_input_cntl);
3308
3309 for (i = 0; i < rshader->noutput; i++) {
3310 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
3311 z_export = 1;
3312 if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
3313 stencil_export = 1;
3314 }
3315 if (rshader->uses_kill)
3316 db_shader_control |= S_02880C_KILL_ENABLE(1);
3317
3318 db_shader_control |= S_02880C_Z_EXPORT_ENABLE(z_export);
3319 db_shader_control |= S_02880C_STENCIL_EXPORT_ENABLE(stencil_export);
3320
3321 exports_ps = 0;
3322 for (i = 0; i < rshader->noutput; i++) {
3323 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
3324 rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
3325 exports_ps |= 1;
3326 }
3327
3328 num_cout = rshader->nr_ps_color_exports;
3329
3330 exports_ps |= S_02884C_EXPORT_COLORS(num_cout);
3331 if (!exports_ps) {
3332 /* always at least export 1 component per pixel */
3333 exports_ps = 2;
3334 }
3335 shader->nr_ps_color_outputs = num_cout;
3336 if (ninterp == 0) {
3337 ninterp = 1;
3338 have_perspective = TRUE;
3339 }
3340
3341 if (!have_perspective && !have_linear)
3342 have_perspective = TRUE;
3343
3344 spi_ps_in_control_0 = S_0286CC_NUM_INTERP(ninterp) |
3345 S_0286CC_PERSP_GRADIENT_ENA(have_perspective) |
3346 S_0286CC_LINEAR_GRADIENT_ENA(have_linear);
3347 spi_input_z = 0;
3348 if (pos_index != -1) {
3349 spi_ps_in_control_0 |= S_0286CC_POSITION_ENA(1) |
3350 S_0286CC_POSITION_CENTROID(rshader->input[pos_index].centroid) |
3351 S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr);
3352 spi_input_z |= S_0286D8_PROVIDE_Z_TO_SPI(1);
3353 }
3354
3355 spi_ps_in_control_1 = 0;
3356 if (face_index != -1) {
3357 spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) |
3358 S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr);
3359 }
3360
3361 spi_baryc_cntl = 0;
3362 if (have_perspective)
3363 spi_baryc_cntl |= S_0286E0_PERSP_CENTER_ENA(1) |
3364 S_0286E0_PERSP_CENTROID_ENA(have_centroid);
3365 if (have_linear)
3366 spi_baryc_cntl |= S_0286E0_LINEAR_CENTER_ENA(1) |
3367 S_0286E0_LINEAR_CENTROID_ENA(have_centroid);
3368
3369 r600_store_context_reg_seq(cb, R_0286CC_SPI_PS_IN_CONTROL_0, 2);
3370 r600_store_value(cb, spi_ps_in_control_0); /* R_0286CC_SPI_PS_IN_CONTROL_0 */
3371 r600_store_value(cb, spi_ps_in_control_1); /* R_0286D0_SPI_PS_IN_CONTROL_1 */
3372
3373 r600_store_context_reg(cb, R_0286E0_SPI_BARYC_CNTL, spi_baryc_cntl);
3374 r600_store_context_reg(cb, R_0286D8_SPI_INPUT_Z, spi_input_z);
3375 r600_store_context_reg(cb, R_02884C_SQ_PGM_EXPORTS_PS, exports_ps);
3376
3377 r600_store_context_reg_seq(cb, R_028840_SQ_PGM_START_PS, 2);
3378 r600_store_value(cb, r600_resource_va(ctx->screen, (void *)shader->bo) >> 8);
3379 r600_store_value(cb, /* R_028844_SQ_PGM_RESOURCES_PS */
3380 S_028844_NUM_GPRS(rshader->bc.ngpr) |
3381 S_028844_PRIME_CACHE_ON_DRAW(1) |
3382 S_028844_STACK_SIZE(rshader->bc.nstack));
3383 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
3384
3385 shader->db_shader_control = db_shader_control;
3386 shader->ps_depth_export = z_export | stencil_export;
3387
3388 shader->sprite_coord_enable = sprite_coord_enable;
3389 if (rctx->rasterizer)
3390 shader->flatshade = rctx->rasterizer->flatshade;
3391 }
3392
3393 void evergreen_update_vs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3394 {
3395 struct r600_command_buffer *cb = &shader->command_buffer;
3396 struct r600_shader *rshader = &shader->shader;
3397 unsigned spi_vs_out_id[10] = {};
3398 unsigned i, tmp, nparams = 0;
3399
3400 for (i = 0; i < rshader->noutput; i++) {
3401 if (rshader->output[i].spi_sid) {
3402 tmp = rshader->output[i].spi_sid << ((nparams & 3) * 8);
3403 spi_vs_out_id[nparams / 4] |= tmp;
3404 nparams++;
3405 }
3406 }
3407
3408 r600_init_command_buffer(cb, 32);
3409
3410 r600_store_context_reg_seq(cb, R_02861C_SPI_VS_OUT_ID_0, 10);
3411 for (i = 0; i < 10; i++) {
3412 r600_store_value(cb, spi_vs_out_id[i]);
3413 }
3414
3415 /* Certain attributes (position, psize, etc.) don't count as params.
3416 * VS is required to export at least one param and r600_shader_from_tgsi()
3417 * takes care of adding a dummy export.
3418 */
3419 if (nparams < 1)
3420 nparams = 1;
3421
3422 r600_store_context_reg(cb, R_0286C4_SPI_VS_OUT_CONFIG,
3423 S_0286C4_VS_EXPORT_COUNT(nparams - 1));
3424 r600_store_context_reg(cb, R_028860_SQ_PGM_RESOURCES_VS,
3425 S_028860_NUM_GPRS(rshader->bc.ngpr) |
3426 S_028860_STACK_SIZE(rshader->bc.nstack));
3427 r600_store_context_reg(cb, R_02885C_SQ_PGM_START_VS,
3428 r600_resource_va(ctx->screen, (void *)shader->bo) >> 8);
3429 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
3430
3431 shader->pa_cl_vs_out_cntl =
3432 S_02881C_VS_OUT_CCDIST0_VEC_ENA((rshader->clip_dist_write & 0x0F) != 0) |
3433 S_02881C_VS_OUT_CCDIST1_VEC_ENA((rshader->clip_dist_write & 0xF0) != 0) |
3434 S_02881C_VS_OUT_MISC_VEC_ENA(rshader->vs_out_misc_write) |
3435 S_02881C_USE_VTX_POINT_SIZE(rshader->vs_out_point_size);
3436 }
3437
3438 void *evergreen_create_resolve_blend(struct r600_context *rctx)
3439 {
3440 struct pipe_blend_state blend;
3441
3442 memset(&blend, 0, sizeof(blend));
3443 blend.independent_blend_enable = true;
3444 blend.rt[0].colormask = 0xf;
3445 return evergreen_create_blend_state_mode(&rctx->context, &blend, V_028808_CB_RESOLVE);
3446 }
3447
3448 void *evergreen_create_decompress_blend(struct r600_context *rctx)
3449 {
3450 struct pipe_blend_state blend;
3451
3452 memset(&blend, 0, sizeof(blend));
3453 blend.independent_blend_enable = true;
3454 blend.rt[0].colormask = 0xf;
3455 return evergreen_create_blend_state_mode(&rctx->context, &blend, V_028808_CB_DECOMPRESS);
3456 }
3457
3458 void *evergreen_create_fmask_decompress_blend(struct r600_context *rctx)
3459 {
3460 struct pipe_blend_state blend;
3461
3462 memset(&blend, 0, sizeof(blend));
3463 blend.independent_blend_enable = true;
3464 blend.rt[0].colormask = 0xf;
3465 return evergreen_create_blend_state_mode(&rctx->context, &blend, V_028808_CB_FMASK_DECOMPRESS);
3466 }
3467
3468 void *evergreen_create_db_flush_dsa(struct r600_context *rctx)
3469 {
3470 struct pipe_depth_stencil_alpha_state dsa = {{0}};
3471
3472 return rctx->context.create_depth_stencil_alpha_state(&rctx->context, &dsa);
3473 }
3474
3475 void evergreen_update_db_shader_control(struct r600_context * rctx)
3476 {
3477 bool dual_export = rctx->framebuffer.export_16bpc &&
3478 !rctx->ps_shader->current->ps_depth_export;
3479
3480 unsigned db_shader_control = rctx->ps_shader->current->db_shader_control |
3481 S_02880C_DUAL_EXPORT_ENABLE(dual_export) |
3482 S_02880C_DB_SOURCE_FORMAT(dual_export ? V_02880C_EXPORT_DB_TWO :
3483 V_02880C_EXPORT_DB_FULL) |
3484 S_02880C_ALPHA_TO_MASK_DISABLE(rctx->framebuffer.cb0_is_integer);
3485
3486 /* When alpha test is enabled we can't trust the hw to make the proper
3487 * decision on the order in which ztest should be run related to fragment
3488 * shader execution.
3489 *
3490 * If alpha test is enabled perform early z rejection (RE_Z) but don't early
3491 * write to the zbuffer. Write to zbuffer is delayed after fragment shader
3492 * execution and thus after alpha test so if discarded by the alpha test
3493 * the z value is not written.
3494 * If ReZ is enabled, and the zfunc/zenable/zwrite values change you can
3495 * get a hang unless you flush the DB in between. For now just use
3496 * LATE_Z.
3497 */
3498 if (rctx->alphatest_state.sx_alpha_test_control) {
3499 db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
3500 } else {
3501 db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
3502 }
3503
3504 if (db_shader_control != rctx->db_misc_state.db_shader_control) {
3505 rctx->db_misc_state.db_shader_control = db_shader_control;
3506 rctx->db_misc_state.atom.dirty = true;
3507 }
3508 }
3509
3510 static void evergreen_dma_copy_tile(struct r600_context *rctx,
3511 struct pipe_resource *dst,
3512 unsigned dst_level,
3513 unsigned dst_x,
3514 unsigned dst_y,
3515 unsigned dst_z,
3516 struct pipe_resource *src,
3517 unsigned src_level,
3518 unsigned src_x,
3519 unsigned src_y,
3520 unsigned src_z,
3521 unsigned copy_height,
3522 unsigned pitch,
3523 unsigned bpp)
3524 {
3525 struct radeon_winsys_cs *cs = rctx->rings.dma.cs;
3526 struct r600_texture *rsrc = (struct r600_texture*)src;
3527 struct r600_texture *rdst = (struct r600_texture*)dst;
3528 unsigned array_mode, lbpp, pitch_tile_max, slice_tile_max, size;
3529 unsigned ncopy, height, cheight, detile, i, x, y, z, src_mode, dst_mode;
3530 unsigned sub_cmd, bank_h, bank_w, mt_aspect, nbanks, tile_split, non_disp_tiling = 0;
3531 uint64_t base, addr;
3532
3533 /* make sure that the dma ring is only one active */
3534 rctx->rings.gfx.flush(rctx, RADEON_FLUSH_ASYNC);
3535
3536 dst_mode = rdst->surface.level[dst_level].mode;
3537 src_mode = rsrc->surface.level[src_level].mode;
3538 /* downcast linear aligned to linear to simplify test */
3539 src_mode = src_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : src_mode;
3540 dst_mode = dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : dst_mode;
3541 assert(dst_mode != src_mode);
3542
3543 /* non_disp_tiling bit needs to be set for depth, stencil, and fmask surfaces */
3544 if (util_format_has_depth(util_format_description(src->format)))
3545 non_disp_tiling = 1;
3546
3547 y = 0;
3548 sub_cmd = 0x8;
3549 lbpp = util_logbase2(bpp);
3550 pitch_tile_max = ((pitch / bpp) >> 3) - 1;
3551 nbanks = eg_num_banks(rctx->screen->tiling_info.num_banks);
3552
3553 if (dst_mode == RADEON_SURF_MODE_LINEAR) {
3554 /* T2L */
3555 array_mode = evergreen_array_mode(src_mode);
3556 slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) >> 6;
3557 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
3558 /* linear height must be the same as the slice tile max height, it's ok even
3559 * if the linear destination/source have smaller heigh as the size of the
3560 * dma packet will be using the copy_height which is always smaller or equal
3561 * to the linear height
3562 */
3563 height = rsrc->surface.level[src_level].npix_y;
3564 detile = 1;
3565 x = src_x;
3566 y = src_y;
3567 z = src_z;
3568 base = rsrc->surface.level[src_level].offset;
3569 addr = rdst->surface.level[dst_level].offset;
3570 addr += rdst->surface.level[dst_level].slice_size * dst_z;
3571 addr += dst_y * pitch + dst_x * bpp;
3572 bank_h = eg_bank_wh(rsrc->surface.bankh);
3573 bank_w = eg_bank_wh(rsrc->surface.bankw);
3574 mt_aspect = eg_macro_tile_aspect(rsrc->surface.mtilea);
3575 tile_split = eg_tile_split(rsrc->surface.tile_split);
3576 base += r600_resource_va(&rctx->screen->screen, src);
3577 addr += r600_resource_va(&rctx->screen->screen, dst);
3578 } else {
3579 /* L2T */
3580 array_mode = evergreen_array_mode(dst_mode);
3581 slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) >> 6;
3582 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
3583 /* linear height must be the same as the slice tile max height, it's ok even
3584 * if the linear destination/source have smaller heigh as the size of the
3585 * dma packet will be using the copy_height which is always smaller or equal
3586 * to the linear height
3587 */
3588 height = rdst->surface.level[dst_level].npix_y;
3589 detile = 0;
3590 x = dst_x;
3591 y = dst_y;
3592 z = dst_z;
3593 base = rdst->surface.level[dst_level].offset;
3594 addr = rsrc->surface.level[src_level].offset;
3595 addr += rsrc->surface.level[src_level].slice_size * src_z;
3596 addr += src_y * pitch + src_x * bpp;
3597 bank_h = eg_bank_wh(rdst->surface.bankh);
3598 bank_w = eg_bank_wh(rdst->surface.bankw);
3599 mt_aspect = eg_macro_tile_aspect(rdst->surface.mtilea);
3600 tile_split = eg_tile_split(rdst->surface.tile_split);
3601 base += r600_resource_va(&rctx->screen->screen, dst);
3602 addr += r600_resource_va(&rctx->screen->screen, src);
3603 }
3604
3605 size = (copy_height * pitch) >> 2;
3606 ncopy = (size / 0x000fffff) + !!(size % 0x000fffff);
3607 r600_need_dma_space(rctx, ncopy * 9);
3608
3609 for (i = 0; i < ncopy; i++) {
3610 cheight = copy_height;
3611 if (((cheight * pitch) >> 2) > 0x000fffff) {
3612 cheight = (0x000fffff << 2) / pitch;
3613 }
3614 size = (cheight * pitch) >> 2;
3615 /* emit reloc before writting cs so that cs is always in consistent state */
3616 r600_context_bo_reloc(rctx, &rctx->rings.dma, &rsrc->resource, RADEON_USAGE_READ);
3617 r600_context_bo_reloc(rctx, &rctx->rings.dma, &rdst->resource, RADEON_USAGE_WRITE);
3618 cs->buf[cs->cdw++] = DMA_PACKET(DMA_PACKET_COPY, sub_cmd, size);
3619 cs->buf[cs->cdw++] = base >> 8;
3620 cs->buf[cs->cdw++] = (detile << 31) | (array_mode << 27) |
3621 (lbpp << 24) | (bank_h << 21) |
3622 (bank_w << 18) | (mt_aspect << 16);
3623 cs->buf[cs->cdw++] = (pitch_tile_max << 0) | ((height - 1) << 16);
3624 cs->buf[cs->cdw++] = (slice_tile_max << 0);
3625 cs->buf[cs->cdw++] = (x << 0) | (z << 18);
3626 cs->buf[cs->cdw++] = (y << 0) | (tile_split << 21) | (nbanks << 25) | (non_disp_tiling << 28);
3627 cs->buf[cs->cdw++] = addr & 0xfffffffc;
3628 cs->buf[cs->cdw++] = (addr >> 32UL) & 0xff;
3629 copy_height -= cheight;
3630 addr += cheight * pitch;
3631 y += cheight;
3632 }
3633 }
3634
3635 boolean evergreen_dma_blit(struct pipe_context *ctx,
3636 struct pipe_resource *dst,
3637 unsigned dst_level,
3638 unsigned dst_x, unsigned dst_y, unsigned dst_z,
3639 struct pipe_resource *src,
3640 unsigned src_level,
3641 const struct pipe_box *src_box)
3642 {
3643 struct r600_context *rctx = (struct r600_context *)ctx;
3644 struct r600_texture *rsrc = (struct r600_texture*)src;
3645 struct r600_texture *rdst = (struct r600_texture*)dst;
3646 unsigned dst_pitch, src_pitch, bpp, dst_mode, src_mode, copy_height;
3647 unsigned src_w, dst_w;
3648
3649 if (rctx->rings.dma.cs == NULL) {
3650 return FALSE;
3651 }
3652 if (src->format != dst->format) {
3653 return FALSE;
3654 }
3655
3656 bpp = rdst->surface.bpe;
3657 dst_pitch = rdst->surface.level[dst_level].pitch_bytes;
3658 src_pitch = rsrc->surface.level[src_level].pitch_bytes;
3659 src_w = rsrc->surface.level[src_level].npix_x;
3660 dst_w = rdst->surface.level[dst_level].npix_x;
3661 copy_height = src_box->height / rsrc->surface.blk_h;
3662
3663 dst_mode = rdst->surface.level[dst_level].mode;
3664 src_mode = rsrc->surface.level[src_level].mode;
3665 /* downcast linear aligned to linear to simplify test */
3666 src_mode = src_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : src_mode;
3667 dst_mode = dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : dst_mode;
3668
3669 if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) {
3670 /* FIXME evergreen can do partial blit */
3671 return FALSE;
3672 }
3673 /* the x test here are currently useless (because we don't support partial blit)
3674 * but keep them around so we don't forget about those
3675 */
3676 if ((src_pitch & 0x7) || (src_box->x & 0x7) || (dst_x & 0x7) || (src_box->y & 0x7) || (dst_y & 0x7)) {
3677 return FALSE;
3678 }
3679
3680 /* 128 bpp surfaces require non_disp_tiling for both
3681 * tiled and linear buffers on cayman. However, async
3682 * DMA only supports it on the tiled side. As such
3683 * the tile order is backwards after a L2T/T2L packet.
3684 */
3685 if ((rctx->chip_class == CAYMAN) &&
3686 (src_mode != dst_mode) &&
3687 (util_format_get_blocksize(src->format) >= 16)) {
3688 return FALSE;
3689 }
3690
3691 if (src_mode == dst_mode) {
3692 uint64_t dst_offset, src_offset;
3693 /* simple dma blit would do NOTE code here assume :
3694 * src_box.x/y == 0
3695 * dst_x/y == 0
3696 * dst_pitch == src_pitch
3697 */
3698 src_offset= rsrc->surface.level[src_level].offset;
3699 src_offset += rsrc->surface.level[src_level].slice_size * src_box->z;
3700 src_offset += src_box->y * src_pitch + src_box->x * bpp;
3701 dst_offset = rdst->surface.level[dst_level].offset;
3702 dst_offset += rdst->surface.level[dst_level].slice_size * dst_z;
3703 dst_offset += dst_y * dst_pitch + dst_x * bpp;
3704 evergreen_dma_copy(rctx, dst, src, dst_offset, src_offset,
3705 src_box->height * src_pitch);
3706 } else {
3707 evergreen_dma_copy_tile(rctx, dst, dst_level, dst_x, dst_y, dst_z,
3708 src, src_level, src_box->x, src_box->y, src_box->z,
3709 copy_height, dst_pitch, bpp);
3710 }
3711 return TRUE;
3712 }
3713
3714 void evergreen_init_state_functions(struct r600_context *rctx)
3715 {
3716 unsigned id = 4;
3717
3718 /* !!!
3719 * To avoid GPU lockup registers must be emited in a specific order
3720 * (no kidding ...). The order below is important and have been
3721 * partialy infered from analyzing fglrx command stream.
3722 *
3723 * Don't reorder atom without carefully checking the effect (GPU lockup
3724 * or piglit regression).
3725 * !!!
3726 */
3727
3728 r600_init_atom(rctx, &rctx->framebuffer.atom, id++, evergreen_emit_framebuffer_state, 0);
3729 /* shader const */
3730 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX].atom, id++, evergreen_emit_vs_constant_buffers, 0);
3731 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY].atom, id++, evergreen_emit_gs_constant_buffers, 0);
3732 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT].atom, id++, evergreen_emit_ps_constant_buffers, 0);
3733 /* shader program */
3734 r600_init_atom(rctx, &rctx->cs_shader_state.atom, id++, evergreen_emit_cs_shader, 0);
3735 /* sampler */
3736 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].states.atom, id++, evergreen_emit_vs_sampler_states, 0);
3737 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].states.atom, id++, evergreen_emit_gs_sampler_states, 0);
3738 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].states.atom, id++, evergreen_emit_ps_sampler_states, 0);
3739 /* resources */
3740 r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, evergreen_fs_emit_vertex_buffers, 0);
3741 r600_init_atom(rctx, &rctx->cs_vertex_buffer_state.atom, id++, evergreen_cs_emit_vertex_buffers, 0);
3742 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views.atom, id++, evergreen_emit_vs_sampler_views, 0);
3743 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views.atom, id++, evergreen_emit_gs_sampler_views, 0);
3744 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views.atom, id++, evergreen_emit_ps_sampler_views, 0);
3745
3746 r600_init_atom(rctx, &rctx->vgt_state.atom, id++, r600_emit_vgt_state, 7);
3747
3748 if (rctx->chip_class == EVERGREEN) {
3749 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, evergreen_emit_sample_mask, 3);
3750 } else {
3751 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, cayman_emit_sample_mask, 4);
3752 }
3753 rctx->sample_mask.sample_mask = ~0;
3754
3755 r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
3756 r600_init_atom(rctx, &rctx->blend_color.atom, id++, r600_emit_blend_color, 6);
3757 r600_init_atom(rctx, &rctx->blend_state.atom, id++, r600_emit_cso_state, 0);
3758 r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, evergreen_emit_cb_misc_state, 4);
3759 r600_init_atom(rctx, &rctx->clip_misc_state.atom, id++, r600_emit_clip_misc_state, 6);
3760 r600_init_atom(rctx, &rctx->clip_state.atom, id++, evergreen_emit_clip_state, 26);
3761 r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, evergreen_emit_db_misc_state, 10);
3762 r600_init_atom(rctx, &rctx->db_state.atom, id++, evergreen_emit_db_state, 14);
3763 r600_init_atom(rctx, &rctx->dsa_state.atom, id++, r600_emit_cso_state, 0);
3764 r600_init_atom(rctx, &rctx->poly_offset_state.atom, id++, evergreen_emit_polygon_offset, 6);
3765 r600_init_atom(rctx, &rctx->rasterizer_state.atom, id++, r600_emit_cso_state, 0);
3766 r600_init_atom(rctx, &rctx->scissor.atom, id++, evergreen_emit_scissor_state, 4);
3767 r600_init_atom(rctx, &rctx->stencil_ref.atom, id++, r600_emit_stencil_ref, 4);
3768 r600_init_atom(rctx, &rctx->viewport.atom, id++, r600_emit_viewport_state, 8);
3769 r600_init_atom(rctx, &rctx->vertex_fetch_shader.atom, id++, evergreen_emit_vertex_fetch_shader, 5);
3770 r600_init_atom(rctx, &rctx->streamout.begin_atom, id++, r600_emit_streamout_begin, 0);
3771 r600_init_atom(rctx, &rctx->vertex_shader.atom, id++, r600_emit_shader, 23);
3772 r600_init_atom(rctx, &rctx->pixel_shader.atom, id++, r600_emit_shader, 0);
3773
3774 rctx->context.create_blend_state = evergreen_create_blend_state;
3775 rctx->context.create_depth_stencil_alpha_state = evergreen_create_dsa_state;
3776 rctx->context.create_rasterizer_state = evergreen_create_rs_state;
3777 rctx->context.create_sampler_state = evergreen_create_sampler_state;
3778 rctx->context.create_sampler_view = evergreen_create_sampler_view;
3779 rctx->context.set_framebuffer_state = evergreen_set_framebuffer_state;
3780 rctx->context.set_polygon_stipple = evergreen_set_polygon_stipple;
3781 rctx->context.set_scissor_state = evergreen_set_scissor_state;
3782 evergreen_init_compute_state_functions(rctx);
3783 }