projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge remote-tracking branch 'jekstrand/wip/i965-uniforms' into vulkan
[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 bool r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
215 {
216 return r600_translate_texformat(screen, format, NULL, NULL, NULL) != ~0U;
217 }
218
219 static bool r600_is_colorbuffer_format_supported(enum chip_class chip, enum pipe_format format)
220 {
221 return r600_translate_colorformat(chip, format) != ~0U &&
222 r600_translate_colorswap(format) != ~0U;
223 }
224
225 static bool r600_is_zs_format_supported(enum pipe_format format)
226 {
227 return r600_translate_dbformat(format) != ~0U;
228 }
229
230 boolean evergreen_is_format_supported(struct pipe_screen *screen,
231 enum pipe_format format,
232 enum pipe_texture_target target,
233 unsigned sample_count,
234 unsigned usage)
235 {
236 struct r600_screen *rscreen = (struct r600_screen*)screen;
237 unsigned retval = 0;
238
239 if (target >= PIPE_MAX_TEXTURE_TYPES) {
240 R600_ERR("r600: unsupported texture type %d\n", target);
241 return FALSE;
242 }
243
244 if (!util_format_is_supported(format, usage))
245 return FALSE;
246
247 if (sample_count > 1) {
248 if (!rscreen->has_msaa)
249 return FALSE;
250
251 switch (sample_count) {
252 case 2:
253 case 4:
254 case 8:
255 break;
256 default:
257 return FALSE;
258 }
259 }
260
261 if (usage & PIPE_BIND_SAMPLER_VIEW) {
262 if (target == PIPE_BUFFER) {
263 if (r600_is_vertex_format_supported(format))
264 retval |= PIPE_BIND_SAMPLER_VIEW;
265 } else {
266 if (r600_is_sampler_format_supported(screen, format))
267 retval |= PIPE_BIND_SAMPLER_VIEW;
268 }
269 }
270
271 if ((usage & (PIPE_BIND_RENDER_TARGET |
272 PIPE_BIND_DISPLAY_TARGET |
273 PIPE_BIND_SCANOUT |
274 PIPE_BIND_SHARED |
275 PIPE_BIND_BLENDABLE)) &&
276 r600_is_colorbuffer_format_supported(rscreen->b.chip_class, format)) {
277 retval |= usage &
278 (PIPE_BIND_RENDER_TARGET |
279 PIPE_BIND_DISPLAY_TARGET |
280 PIPE_BIND_SCANOUT |
281 PIPE_BIND_SHARED);
282 if (!util_format_is_pure_integer(format) &&
283 !util_format_is_depth_or_stencil(format))
284 retval |= usage & PIPE_BIND_BLENDABLE;
285 }
286
287 if ((usage & PIPE_BIND_DEPTH_STENCIL) &&
288 r600_is_zs_format_supported(format)) {
289 retval |= PIPE_BIND_DEPTH_STENCIL;
290 }
291
292 if ((usage & PIPE_BIND_VERTEX_BUFFER) &&
293 r600_is_vertex_format_supported(format)) {
294 retval |= PIPE_BIND_VERTEX_BUFFER;
295 }
296
297 if (usage & PIPE_BIND_TRANSFER_READ)
298 retval |= PIPE_BIND_TRANSFER_READ;
299 if (usage & PIPE_BIND_TRANSFER_WRITE)
300 retval |= PIPE_BIND_TRANSFER_WRITE;
301
302 return retval == usage;
303 }
304
305 static void *evergreen_create_blend_state_mode(struct pipe_context *ctx,
306 const struct pipe_blend_state *state, int mode)
307 {
308 uint32_t color_control = 0, target_mask = 0;
309 struct r600_blend_state *blend = CALLOC_STRUCT(r600_blend_state);
310
311 if (!blend) {
312 return NULL;
313 }
314
315 r600_init_command_buffer(&blend->buffer, 20);
316 r600_init_command_buffer(&blend->buffer_no_blend, 20);
317
318 if (state->logicop_enable) {
319 color_control |= (state->logicop_func << 16) | (state->logicop_func << 20);
320 } else {
321 color_control |= (0xcc << 16);
322 }
323 /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
324 if (state->independent_blend_enable) {
325 for (int i = 0; i < 8; i++) {
326 target_mask |= (state->rt[i].colormask << (4 * i));
327 }
328 } else {
329 for (int i = 0; i < 8; i++) {
330 target_mask |= (state->rt[0].colormask << (4 * i));
331 }
332 }
333
334 /* only have dual source on MRT0 */
335 blend->dual_src_blend = util_blend_state_is_dual(state, 0);
336 blend->cb_target_mask = target_mask;
337 blend->alpha_to_one = state->alpha_to_one;
338
339 if (target_mask)
340 color_control |= S_028808_MODE(mode);
341 else
342 color_control |= S_028808_MODE(V_028808_CB_DISABLE);
343
344
345 r600_store_context_reg(&blend->buffer, R_028808_CB_COLOR_CONTROL, color_control);
346 r600_store_context_reg(&blend->buffer, R_028B70_DB_ALPHA_TO_MASK,
347 S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
348 S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
349 S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
350 S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
351 S_028B70_ALPHA_TO_MASK_OFFSET3(2));
352 r600_store_context_reg_seq(&blend->buffer, R_028780_CB_BLEND0_CONTROL, 8);
353
354 /* Copy over the dwords set so far into buffer_no_blend.
355 * Only the CB_BLENDi_CONTROL registers must be set after this. */
356 memcpy(blend->buffer_no_blend.buf, blend->buffer.buf, blend->buffer.num_dw * 4);
357 blend->buffer_no_blend.num_dw = blend->buffer.num_dw;
358
359 for (int i = 0; i < 8; i++) {
360 /* state->rt entries > 0 only written if independent blending */
361 const int j = state->independent_blend_enable ? i : 0;
362
363 unsigned eqRGB = state->rt[j].rgb_func;
364 unsigned srcRGB = state->rt[j].rgb_src_factor;
365 unsigned dstRGB = state->rt[j].rgb_dst_factor;
366 unsigned eqA = state->rt[j].alpha_func;
367 unsigned srcA = state->rt[j].alpha_src_factor;
368 unsigned dstA = state->rt[j].alpha_dst_factor;
369 uint32_t bc = 0;
370
371 r600_store_value(&blend->buffer_no_blend, 0);
372
373 if (!state->rt[j].blend_enable) {
374 r600_store_value(&blend->buffer, 0);
375 continue;
376 }
377
378 bc |= S_028780_BLEND_CONTROL_ENABLE(1);
379 bc |= S_028780_COLOR_COMB_FCN(r600_translate_blend_function(eqRGB));
380 bc |= S_028780_COLOR_SRCBLEND(r600_translate_blend_factor(srcRGB));
381 bc |= S_028780_COLOR_DESTBLEND(r600_translate_blend_factor(dstRGB));
382
383 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
384 bc |= S_028780_SEPARATE_ALPHA_BLEND(1);
385 bc |= S_028780_ALPHA_COMB_FCN(r600_translate_blend_function(eqA));
386 bc |= S_028780_ALPHA_SRCBLEND(r600_translate_blend_factor(srcA));
387 bc |= S_028780_ALPHA_DESTBLEND(r600_translate_blend_factor(dstA));
388 }
389 r600_store_value(&blend->buffer, bc);
390 }
391 return blend;
392 }
393
394 static void *evergreen_create_blend_state(struct pipe_context *ctx,
395 const struct pipe_blend_state *state)
396 {
397
398 return evergreen_create_blend_state_mode(ctx, state, V_028808_CB_NORMAL);
399 }
400
401 static void *evergreen_create_dsa_state(struct pipe_context *ctx,
402 const struct pipe_depth_stencil_alpha_state *state)
403 {
404 unsigned db_depth_control, alpha_test_control, alpha_ref;
405 struct r600_dsa_state *dsa = CALLOC_STRUCT(r600_dsa_state);
406
407 if (!dsa) {
408 return NULL;
409 }
410
411 r600_init_command_buffer(&dsa->buffer, 3);
412
413 dsa->valuemask[0] = state->stencil[0].valuemask;
414 dsa->valuemask[1] = state->stencil[1].valuemask;
415 dsa->writemask[0] = state->stencil[0].writemask;
416 dsa->writemask[1] = state->stencil[1].writemask;
417 dsa->zwritemask = state->depth.writemask;
418
419 db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) |
420 S_028800_Z_WRITE_ENABLE(state->depth.writemask) |
421 S_028800_ZFUNC(state->depth.func);
422
423 /* stencil */
424 if (state->stencil[0].enabled) {
425 db_depth_control |= S_028800_STENCIL_ENABLE(1);
426 db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func); /* translates straight */
427 db_depth_control |= S_028800_STENCILFAIL(r600_translate_stencil_op(state->stencil[0].fail_op));
428 db_depth_control |= S_028800_STENCILZPASS(r600_translate_stencil_op(state->stencil[0].zpass_op));
429 db_depth_control |= S_028800_STENCILZFAIL(r600_translate_stencil_op(state->stencil[0].zfail_op));
430
431 if (state->stencil[1].enabled) {
432 db_depth_control |= S_028800_BACKFACE_ENABLE(1);
433 db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func); /* translates straight */
434 db_depth_control |= S_028800_STENCILFAIL_BF(r600_translate_stencil_op(state->stencil[1].fail_op));
435 db_depth_control |= S_028800_STENCILZPASS_BF(r600_translate_stencil_op(state->stencil[1].zpass_op));
436 db_depth_control |= S_028800_STENCILZFAIL_BF(r600_translate_stencil_op(state->stencil[1].zfail_op));
437 }
438 }
439
440 /* alpha */
441 alpha_test_control = 0;
442 alpha_ref = 0;
443 if (state->alpha.enabled) {
444 alpha_test_control = S_028410_ALPHA_FUNC(state->alpha.func);
445 alpha_test_control |= S_028410_ALPHA_TEST_ENABLE(1);
446 alpha_ref = fui(state->alpha.ref_value);
447 }
448 dsa->sx_alpha_test_control = alpha_test_control & 0xff;
449 dsa->alpha_ref = alpha_ref;
450
451 /* misc */
452 r600_store_context_reg(&dsa->buffer, R_028800_DB_DEPTH_CONTROL, db_depth_control);
453 return dsa;
454 }
455
456 static void *evergreen_create_rs_state(struct pipe_context *ctx,
457 const struct pipe_rasterizer_state *state)
458 {
459 struct r600_context *rctx = (struct r600_context *)ctx;
460 unsigned tmp, spi_interp;
461 float psize_min, psize_max;
462 struct r600_rasterizer_state *rs = CALLOC_STRUCT(r600_rasterizer_state);
463
464 if (!rs) {
465 return NULL;
466 }
467
468 r600_init_command_buffer(&rs->buffer, 30);
469
470 rs->flatshade = state->flatshade;
471 rs->sprite_coord_enable = state->sprite_coord_enable;
472 rs->two_side = state->light_twoside;
473 rs->clip_plane_enable = state->clip_plane_enable;
474 rs->pa_sc_line_stipple = state->line_stipple_enable ?
475 S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
476 S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
477 rs->pa_cl_clip_cntl =
478 S_028810_PS_UCP_MODE(3) |
479 S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) |
480 S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
481 S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
482 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1) |
483 S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard);
484 rs->multisample_enable = state->multisample;
485
486 /* offset */
487 rs->offset_units = state->offset_units;
488 rs->offset_scale = state->offset_scale * 16.0f;
489 rs->offset_enable = state->offset_point || state->offset_line || state->offset_tri;
490
491 if (state->point_size_per_vertex) {
492 psize_min = util_get_min_point_size(state);
493 psize_max = 8192;
494 } else {
495 /* Force the point size to be as if the vertex output was disabled. */
496 psize_min = state->point_size;
497 psize_max = state->point_size;
498 }
499
500 spi_interp = S_0286D4_FLAT_SHADE_ENA(1);
501 if (state->sprite_coord_enable) {
502 spi_interp |= S_0286D4_PNT_SPRITE_ENA(1) |
503 S_0286D4_PNT_SPRITE_OVRD_X(2) |
504 S_0286D4_PNT_SPRITE_OVRD_Y(3) |
505 S_0286D4_PNT_SPRITE_OVRD_Z(0) |
506 S_0286D4_PNT_SPRITE_OVRD_W(1);
507 if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) {
508 spi_interp |= S_0286D4_PNT_SPRITE_TOP_1(1);
509 }
510 }
511
512 r600_store_context_reg_seq(&rs->buffer, R_028A00_PA_SU_POINT_SIZE, 3);
513 /* point size 12.4 fixed point (divide by two, because 0.5 = 1 pixel) */
514 tmp = r600_pack_float_12p4(state->point_size/2);
515 r600_store_value(&rs->buffer, /* R_028A00_PA_SU_POINT_SIZE */
516 S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
517 r600_store_value(&rs->buffer, /* R_028A04_PA_SU_POINT_MINMAX */
518 S_028A04_MIN_SIZE(r600_pack_float_12p4(psize_min/2)) |
519 S_028A04_MAX_SIZE(r600_pack_float_12p4(psize_max/2)));
520 r600_store_value(&rs->buffer, /* R_028A08_PA_SU_LINE_CNTL */
521 S_028A08_WIDTH((unsigned)(state->line_width * 8)));
522
523 r600_store_context_reg(&rs->buffer, R_0286D4_SPI_INTERP_CONTROL_0, spi_interp);
524 r600_store_context_reg(&rs->buffer, R_028A48_PA_SC_MODE_CNTL_0,
525 S_028A48_MSAA_ENABLE(state->multisample) |
526 S_028A48_VPORT_SCISSOR_ENABLE(state->scissor) |
527 S_028A48_LINE_STIPPLE_ENABLE(state->line_stipple_enable));
528
529 if (rctx->b.chip_class == CAYMAN) {
530 r600_store_context_reg(&rs->buffer, CM_R_028BE4_PA_SU_VTX_CNTL,
531 S_028C08_PIX_CENTER_HALF(state->half_pixel_center) |
532 S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
533 } else {
534 r600_store_context_reg(&rs->buffer, R_028C08_PA_SU_VTX_CNTL,
535 S_028C08_PIX_CENTER_HALF(state->half_pixel_center) |
536 S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
537 }
538
539 r600_store_context_reg(&rs->buffer, R_028B7C_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
540 r600_store_context_reg(&rs->buffer, R_028814_PA_SU_SC_MODE_CNTL,
541 S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) |
542 S_028814_CULL_FRONT((state->cull_face & PIPE_FACE_FRONT) ? 1 : 0) |
543 S_028814_CULL_BACK((state->cull_face & PIPE_FACE_BACK) ? 1 : 0) |
544 S_028814_FACE(!state->front_ccw) |
545 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state, state->fill_front)) |
546 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state, state->fill_back)) |
547 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_point || state->offset_line) |
548 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL ||
549 state->fill_back != PIPE_POLYGON_MODE_FILL) |
550 S_028814_POLYMODE_FRONT_PTYPE(r600_translate_fill(state->fill_front)) |
551 S_028814_POLYMODE_BACK_PTYPE(r600_translate_fill(state->fill_back)));
552 return rs;
553 }
554
555 static void *evergreen_create_sampler_state(struct pipe_context *ctx,
556 const struct pipe_sampler_state *state)
557 {
558 struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
559 unsigned aniso_flag_offset = state->max_anisotropy > 1 ? 2 : 0;
560
561 if (!ss) {
562 return NULL;
563 }
564
565 ss->border_color_use = sampler_state_needs_border_color(state);
566
567 /* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
568 ss->tex_sampler_words[0] =
569 S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
570 S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
571 S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
572 S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
573 S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter) | aniso_flag_offset) |
574 S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
575 S_03C000_MAX_ANISO(r600_tex_aniso_filter(state->max_anisotropy)) |
576 S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
577 S_03C000_BORDER_COLOR_TYPE(ss->border_color_use ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
578 /* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
579 ss->tex_sampler_words[1] =
580 S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
581 S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8));
582 /* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
583 ss->tex_sampler_words[2] =
584 S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
585 (state->seamless_cube_map ? 0 : S_03C008_DISABLE_CUBE_WRAP(1)) |
586 S_03C008_TYPE(1);
587
588 if (ss->border_color_use) {
589 memcpy(&ss->border_color, &state->border_color, sizeof(state->border_color));
590 }
591 return ss;
592 }
593
594 static struct pipe_sampler_view *
595 texture_buffer_sampler_view(struct r600_context *rctx,
596 struct r600_pipe_sampler_view *view,
597 unsigned width0, unsigned height0)
598
599 {
600 struct r600_texture *tmp = (struct r600_texture*)view->base.texture;
601 uint64_t va;
602 int stride = util_format_get_blocksize(view->base.format);
603 unsigned format, num_format, format_comp, endian;
604 unsigned swizzle_res;
605 unsigned char swizzle[4];
606 const struct util_format_description *desc;
607 unsigned offset = view->base.u.buf.first_element * stride;
608 unsigned size = (view->base.u.buf.last_element - view->base.u.buf.first_element + 1) * stride;
609
610 swizzle[0] = view->base.swizzle_r;
611 swizzle[1] = view->base.swizzle_g;
612 swizzle[2] = view->base.swizzle_b;
613 swizzle[3] = view->base.swizzle_a;
614
615 r600_vertex_data_type(view->base.format,
616 &format, &num_format, &format_comp,
617 &endian);
618
619 desc = util_format_description(view->base.format);
620
621 swizzle_res = r600_get_swizzle_combined(desc->swizzle, swizzle, TRUE);
622
623 va = tmp->resource.gpu_address + offset;
624 view->tex_resource = &tmp->resource;
625
626 view->skip_mip_address_reloc = true;
627 view->tex_resource_words[0] = va;
628 view->tex_resource_words[1] = size - 1;
629 view->tex_resource_words[2] = S_030008_BASE_ADDRESS_HI(va >> 32UL) |
630 S_030008_STRIDE(stride) |
631 S_030008_DATA_FORMAT(format) |
632 S_030008_NUM_FORMAT_ALL(num_format) |
633 S_030008_FORMAT_COMP_ALL(format_comp) |
634 S_030008_ENDIAN_SWAP(endian);
635 view->tex_resource_words[3] = swizzle_res;
636 /*
637 * in theory dword 4 is for number of elements, for use with resinfo,
638 * but it seems to utterly fail to work, the amd gpu shader analyser
639 * uses a const buffer to store the element sizes for buffer txq
640 */
641 view->tex_resource_words[4] = 0;
642 view->tex_resource_words[5] = view->tex_resource_words[6] = 0;
643 view->tex_resource_words[7] = S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_BUFFER);
644
645 if (tmp->resource.gpu_address)
646 LIST_ADDTAIL(&view->list, &rctx->b.texture_buffers);
647 return &view->base;
648 }
649
650 struct pipe_sampler_view *
651 evergreen_create_sampler_view_custom(struct pipe_context *ctx,
652 struct pipe_resource *texture,
653 const struct pipe_sampler_view *state,
654 unsigned width0, unsigned height0,
655 unsigned force_level)
656 {
657 struct r600_context *rctx = (struct r600_context*)ctx;
658 struct r600_screen *rscreen = (struct r600_screen*)ctx->screen;
659 struct r600_pipe_sampler_view *view = CALLOC_STRUCT(r600_pipe_sampler_view);
660 struct r600_texture *tmp = (struct r600_texture*)texture;
661 unsigned format, endian;
662 uint32_t word4 = 0, yuv_format = 0, pitch = 0;
663 unsigned char swizzle[4], array_mode = 0, non_disp_tiling = 0;
664 unsigned height, depth, width;
665 unsigned macro_aspect, tile_split, bankh, bankw, nbanks, fmask_bankh;
666 enum pipe_format pipe_format = state->format;
667 struct radeon_surf_level *surflevel;
668 unsigned base_level, first_level, last_level;
669 unsigned dim, last_layer;
670 uint64_t va;
671
672 if (!view)
673 return NULL;
674
675 /* initialize base object */
676 view->base = *state;
677 view->base.texture = NULL;
678 pipe_reference(NULL, &texture->reference);
679 view->base.texture = texture;
680 view->base.reference.count = 1;
681 view->base.context = ctx;
682
683 if (state->target == PIPE_BUFFER)
684 return texture_buffer_sampler_view(rctx, view, width0, height0);
685
686 swizzle[0] = state->swizzle_r;
687 swizzle[1] = state->swizzle_g;
688 swizzle[2] = state->swizzle_b;
689 swizzle[3] = state->swizzle_a;
690
691 tile_split = tmp->surface.tile_split;
692 surflevel = tmp->surface.level;
693
694 /* Texturing with separate depth and stencil. */
695 if (tmp->is_depth && !tmp->is_flushing_texture) {
696 switch (pipe_format) {
697 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
698 pipe_format = PIPE_FORMAT_Z32_FLOAT;
699 break;
700 case PIPE_FORMAT_X8Z24_UNORM:
701 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
702 /* Z24 is always stored like this. */
703 pipe_format = PIPE_FORMAT_Z24X8_UNORM;
704 break;
705 case PIPE_FORMAT_X24S8_UINT:
706 case PIPE_FORMAT_S8X24_UINT:
707 case PIPE_FORMAT_X32_S8X24_UINT:
708 pipe_format = PIPE_FORMAT_S8_UINT;
709 tile_split = tmp->surface.stencil_tile_split;
710 surflevel = tmp->surface.stencil_level;
711 break;
712 default:;
713 }
714 }
715
716 format = r600_translate_texformat(ctx->screen, pipe_format,
717 swizzle,
718 &word4, &yuv_format);
719 assert(format != ~0);
720 if (format == ~0) {
721 FREE(view);
722 return NULL;
723 }
724
725 endian = r600_colorformat_endian_swap(format);
726
727 base_level = 0;
728 first_level = state->u.tex.first_level;
729 last_level = state->u.tex.last_level;
730 width = width0;
731 height = height0;
732 depth = texture->depth0;
733
734 if (force_level) {
735 base_level = force_level;
736 first_level = 0;
737 last_level = 0;
738 width = u_minify(width, force_level);
739 height = u_minify(height, force_level);
740 depth = u_minify(depth, force_level);
741 }
742
743 pitch = surflevel[base_level].nblk_x * util_format_get_blockwidth(pipe_format);
744 non_disp_tiling = tmp->non_disp_tiling;
745
746 switch (surflevel[base_level].mode) {
747 case RADEON_SURF_MODE_LINEAR_ALIGNED:
748 array_mode = V_028C70_ARRAY_LINEAR_ALIGNED;
749 break;
750 case RADEON_SURF_MODE_2D:
751 array_mode = V_028C70_ARRAY_2D_TILED_THIN1;
752 break;
753 case RADEON_SURF_MODE_1D:
754 array_mode = V_028C70_ARRAY_1D_TILED_THIN1;
755 break;
756 case RADEON_SURF_MODE_LINEAR:
757 default:
758 array_mode = V_028C70_ARRAY_LINEAR_GENERAL;
759 break;
760 }
761 macro_aspect = tmp->surface.mtilea;
762 bankw = tmp->surface.bankw;
763 bankh = tmp->surface.bankh;
764 tile_split = eg_tile_split(tile_split);
765 macro_aspect = eg_macro_tile_aspect(macro_aspect);
766 bankw = eg_bank_wh(bankw);
767 bankh = eg_bank_wh(bankh);
768 fmask_bankh = eg_bank_wh(tmp->fmask.bank_height);
769
770 /* 128 bit formats require tile type = 1 */
771 if (rscreen->b.chip_class == CAYMAN) {
772 if (util_format_get_blocksize(pipe_format) >= 16)
773 non_disp_tiling = 1;
774 }
775 nbanks = eg_num_banks(rscreen->b.tiling_info.num_banks);
776
777 if (state->target == PIPE_TEXTURE_1D_ARRAY) {
778 height = 1;
779 depth = texture->array_size;
780 } else if (state->target == PIPE_TEXTURE_2D_ARRAY) {
781 depth = texture->array_size;
782 } else if (state->target == PIPE_TEXTURE_CUBE_ARRAY)
783 depth = texture->array_size / 6;
784
785 va = tmp->resource.gpu_address;
786
787 if (state->format == PIPE_FORMAT_X24S8_UINT ||
788 state->format == PIPE_FORMAT_S8X24_UINT ||
789 state->format == PIPE_FORMAT_X32_S8X24_UINT ||
790 state->format == PIPE_FORMAT_S8_UINT)
791 view->is_stencil_sampler = true;
792
793 view->tex_resource = &tmp->resource;
794
795 /* array type views and views into array types need to use layer offset */
796 dim = state->target;
797 if (state->target != PIPE_TEXTURE_CUBE)
798 dim = MAX2(state->target, texture->target);
799
800 view->tex_resource_words[0] = (S_030000_DIM(r600_tex_dim(dim, texture->nr_samples)) |
801 S_030000_PITCH((pitch / 8) - 1) |
802 S_030000_TEX_WIDTH(width - 1));
803 if (rscreen->b.chip_class == CAYMAN)
804 view->tex_resource_words[0] |= CM_S_030000_NON_DISP_TILING_ORDER(non_disp_tiling);
805 else
806 view->tex_resource_words[0] |= S_030000_NON_DISP_TILING_ORDER(non_disp_tiling);
807 view->tex_resource_words[1] = (S_030004_TEX_HEIGHT(height - 1) |
808 S_030004_TEX_DEPTH(depth - 1) |
809 S_030004_ARRAY_MODE(array_mode));
810 view->tex_resource_words[2] = (surflevel[base_level].offset + va) >> 8;
811
812 /* TEX_RESOURCE_WORD3.MIP_ADDRESS */
813 if (texture->nr_samples > 1 && rscreen->has_compressed_msaa_texturing) {
814 if (tmp->is_depth) {
815 /* disable FMASK (0 = disabled) */
816 view->tex_resource_words[3] = 0;
817 view->skip_mip_address_reloc = true;
818 } else {
819 /* FMASK should be in MIP_ADDRESS for multisample textures */
820 view->tex_resource_words[3] = (tmp->fmask.offset + va) >> 8;
821 }
822 } else if (last_level && texture->nr_samples <= 1) {
823 view->tex_resource_words[3] = (surflevel[1].offset + va) >> 8;
824 } else {
825 view->tex_resource_words[3] = (surflevel[base_level].offset + va) >> 8;
826 }
827
828 last_layer = state->u.tex.last_layer;
829 if (state->target != texture->target && depth == 1) {
830 last_layer = state->u.tex.first_layer;
831 }
832 view->tex_resource_words[4] = (word4 |
833 S_030010_ENDIAN_SWAP(endian));
834 view->tex_resource_words[5] = S_030014_BASE_ARRAY(state->u.tex.first_layer) |
835 S_030014_LAST_ARRAY(last_layer);
836 view->tex_resource_words[6] = S_030018_TILE_SPLIT(tile_split);
837
838 if (texture->nr_samples > 1) {
839 unsigned log_samples = util_logbase2(texture->nr_samples);
840 if (rscreen->b.chip_class == CAYMAN) {
841 view->tex_resource_words[4] |= S_030010_LOG2_NUM_FRAGMENTS(log_samples);
842 }
843 /* LAST_LEVEL holds log2(nr_samples) for multisample textures */
844 view->tex_resource_words[5] |= S_030014_LAST_LEVEL(log_samples);
845 view->tex_resource_words[6] |= S_030018_FMASK_BANK_HEIGHT(fmask_bankh);
846 } else {
847 view->tex_resource_words[4] |= S_030010_BASE_LEVEL(first_level);
848 view->tex_resource_words[5] |= S_030014_LAST_LEVEL(last_level);
849 /* aniso max 16 samples */
850 view->tex_resource_words[6] |= S_030018_MAX_ANISO(4);
851 }
852
853 view->tex_resource_words[7] = S_03001C_DATA_FORMAT(format) |
854 S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_TEXTURE) |
855 S_03001C_BANK_WIDTH(bankw) |
856 S_03001C_BANK_HEIGHT(bankh) |
857 S_03001C_MACRO_TILE_ASPECT(macro_aspect) |
858 S_03001C_NUM_BANKS(nbanks) |
859 S_03001C_DEPTH_SAMPLE_ORDER(tmp->is_depth && !tmp->is_flushing_texture);
860 return &view->base;
861 }
862
863 static struct pipe_sampler_view *
864 evergreen_create_sampler_view(struct pipe_context *ctx,
865 struct pipe_resource *tex,
866 const struct pipe_sampler_view *state)
867 {
868 return evergreen_create_sampler_view_custom(ctx, tex, state,
869 tex->width0, tex->height0, 0);
870 }
871
872 static void evergreen_emit_config_state(struct r600_context *rctx, struct r600_atom *atom)
873 {
874 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
875 struct r600_config_state *a = (struct r600_config_state*)atom;
876
877 radeon_set_config_reg_seq(cs, R_008C04_SQ_GPR_RESOURCE_MGMT_1, 3);
878 if (a->dyn_gpr_enabled) {
879 radeon_emit(cs, S_008C04_NUM_CLAUSE_TEMP_GPRS(rctx->r6xx_num_clause_temp_gprs));
880 radeon_emit(cs, 0);
881 radeon_emit(cs, 0);
882 } else {
883 radeon_emit(cs, a->sq_gpr_resource_mgmt_1);
884 radeon_emit(cs, a->sq_gpr_resource_mgmt_2);
885 radeon_emit(cs, a->sq_gpr_resource_mgmt_3);
886 }
887 radeon_set_config_reg(cs, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, (a->dyn_gpr_enabled << 8));
888 if (a->dyn_gpr_enabled) {
889 radeon_set_context_reg(cs, R_028838_SQ_DYN_GPR_RESOURCE_LIMIT_1,
890 S_028838_PS_GPRS(0x1e) |
891 S_028838_VS_GPRS(0x1e) |
892 S_028838_GS_GPRS(0x1e) |
893 S_028838_ES_GPRS(0x1e) |
894 S_028838_HS_GPRS(0x1e) |
895 S_028838_LS_GPRS(0x1e)); /* workaround for hw issues with dyn gpr - must set all limits to 240 instead of 0, 0x1e == 240 / 8*/
896 }
897 }
898
899 static void evergreen_emit_clip_state(struct r600_context *rctx, struct r600_atom *atom)
900 {
901 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
902 struct pipe_clip_state *state = &rctx->clip_state.state;
903
904 radeon_set_context_reg_seq(cs, R_0285BC_PA_CL_UCP0_X, 6*4);
905 radeon_emit_array(cs, (unsigned*)state, 6*4);
906 }
907
908 static void evergreen_set_polygon_stipple(struct pipe_context *ctx,
909 const struct pipe_poly_stipple *state)
910 {
911 }
912
913 static void evergreen_get_scissor_rect(struct r600_context *rctx,
914 unsigned tl_x, unsigned tl_y, unsigned br_x, unsigned br_y,
915 uint32_t *tl, uint32_t *br)
916 {
917 /* EG hw workaround */
918 if (br_x == 0)
919 tl_x = 1;
920 if (br_y == 0)
921 tl_y = 1;
922
923 /* cayman hw workaround */
924 if (rctx->b.chip_class == CAYMAN) {
925 if (br_x == 1 && br_y == 1)
926 br_x = 2;
927 }
928
929 *tl = S_028240_TL_X(tl_x) | S_028240_TL_Y(tl_y);
930 *br = S_028244_BR_X(br_x) | S_028244_BR_Y(br_y);
931 }
932
933 static void evergreen_set_scissor_states(struct pipe_context *ctx,
934 unsigned start_slot,
935 unsigned num_scissors,
936 const struct pipe_scissor_state *state)
937 {
938 struct r600_context *rctx = (struct r600_context *)ctx;
939 struct r600_scissor_state *rstate = &rctx->scissor;
940 int i;
941
942 for (i = start_slot; i < start_slot + num_scissors; i++)
943 rstate->scissor[i] = state[i - start_slot];
944 rstate->dirty_mask |= ((1 << num_scissors) - 1) << start_slot;
945 rstate->atom.num_dw = util_bitcount(rstate->dirty_mask) * 4;
946 r600_mark_atom_dirty(rctx, &rstate->atom);
947 }
948
949 static void evergreen_emit_scissor_state(struct r600_context *rctx, struct r600_atom *atom)
950 {
951 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
952 struct r600_scissor_state *rstate = &rctx->scissor;
953 struct pipe_scissor_state *state;
954 uint32_t dirty_mask;
955 unsigned i, offset;
956 uint32_t tl, br;
957
958 dirty_mask = rstate->dirty_mask;
959 while (dirty_mask != 0) {
960 i = u_bit_scan(&dirty_mask);
961 state = &rstate->scissor[i];
962 evergreen_get_scissor_rect(rctx, state->minx, state->miny, state->maxx, state->maxy, &tl, &br);
963
964 offset = i * 4 * 2;
965 radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL + offset, 2);
966 radeon_emit(cs, tl);
967 radeon_emit(cs, br);
968 }
969 rstate->dirty_mask = 0;
970 rstate->atom.num_dw = 0;
971 }
972
973 /**
974 * This function intializes the CB* register values for RATs. It is meant
975 * to be used for 1D aligned buffers that do not have an associated
976 * radeon_surf.
977 */
978 void evergreen_init_color_surface_rat(struct r600_context *rctx,
979 struct r600_surface *surf)
980 {
981 struct pipe_resource *pipe_buffer = surf->base.texture;
982 unsigned format = r600_translate_colorformat(rctx->b.chip_class,
983 surf->base.format);
984 unsigned endian = r600_colorformat_endian_swap(format);
985 unsigned swap = r600_translate_colorswap(surf->base.format);
986 unsigned block_size =
987 align(util_format_get_blocksize(pipe_buffer->format), 4);
988 unsigned pitch_alignment =
989 MAX2(64, rctx->screen->b.tiling_info.group_bytes / block_size);
990 unsigned pitch = align(pipe_buffer->width0, pitch_alignment);
991
992 /* XXX: This is copied from evergreen_init_color_surface(). I don't
993 * know why this is necessary.
994 */
995 if (pipe_buffer->usage == PIPE_USAGE_STAGING) {
996 endian = ENDIAN_NONE;
997 }
998
999 surf->cb_color_base = r600_resource(pipe_buffer)->gpu_address >> 8;
1000
1001 surf->cb_color_pitch = (pitch / 8) - 1;
1002
1003 surf->cb_color_slice = 0;
1004
1005 surf->cb_color_view = 0;
1006
1007 surf->cb_color_info =
1008 S_028C70_ENDIAN(endian)
1009 | S_028C70_FORMAT(format)
1010 | S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_ALIGNED)
1011 | S_028C70_NUMBER_TYPE(V_028C70_NUMBER_UINT)
1012 | S_028C70_COMP_SWAP(swap)
1013 | S_028C70_BLEND_BYPASS(1) /* We must set this bit because we
1014 * are using NUMBER_UINT */
1015 | S_028C70_RAT(1)
1016 ;
1017
1018 surf->cb_color_attrib = S_028C74_NON_DISP_TILING_ORDER(1);
1019
1020 /* For buffers, CB_COLOR0_DIM needs to be set to the number of
1021 * elements. */
1022 surf->cb_color_dim = pipe_buffer->width0;
1023
1024 /* Set the buffer range the GPU will have access to: */
1025 util_range_add(&r600_resource(pipe_buffer)->valid_buffer_range,
1026 0, pipe_buffer->width0);
1027
1028 surf->cb_color_fmask = surf->cb_color_base;
1029 surf->cb_color_fmask_slice = 0;
1030 }
1031
1032 void evergreen_init_color_surface(struct r600_context *rctx,
1033 struct r600_surface *surf)
1034 {
1035 struct r600_screen *rscreen = rctx->screen;
1036 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1037 unsigned level = surf->base.u.tex.level;
1038 unsigned pitch, slice;
1039 unsigned color_info, color_attrib, color_dim = 0, color_view;
1040 unsigned format, swap, ntype, endian;
1041 uint64_t offset, base_offset;
1042 unsigned non_disp_tiling, macro_aspect, tile_split, bankh, bankw, fmask_bankh, nbanks;
1043 const struct util_format_description *desc;
1044 int i;
1045 bool blend_clamp = 0, blend_bypass = 0;
1046
1047 offset = rtex->surface.level[level].offset;
1048 if (rtex->surface.level[level].mode == RADEON_SURF_MODE_LINEAR) {
1049 assert(surf->base.u.tex.first_layer == surf->base.u.tex.last_layer);
1050 offset += rtex->surface.level[level].slice_size *
1051 surf->base.u.tex.first_layer;
1052 color_view = 0;
1053 } else
1054 color_view = S_028C6C_SLICE_START(surf->base.u.tex.first_layer) |
1055 S_028C6C_SLICE_MAX(surf->base.u.tex.last_layer);
1056
1057 pitch = (rtex->surface.level[level].nblk_x) / 8 - 1;
1058 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
1059 if (slice) {
1060 slice = slice - 1;
1061 }
1062 color_info = 0;
1063 switch (rtex->surface.level[level].mode) {
1064 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1065 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_ALIGNED);
1066 non_disp_tiling = 1;
1067 break;
1068 case RADEON_SURF_MODE_1D:
1069 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_1D_TILED_THIN1);
1070 non_disp_tiling = rtex->non_disp_tiling;
1071 break;
1072 case RADEON_SURF_MODE_2D:
1073 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_2D_TILED_THIN1);
1074 non_disp_tiling = rtex->non_disp_tiling;
1075 break;
1076 case RADEON_SURF_MODE_LINEAR:
1077 default:
1078 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_GENERAL);
1079 non_disp_tiling = 1;
1080 break;
1081 }
1082 tile_split = rtex->surface.tile_split;
1083 macro_aspect = rtex->surface.mtilea;
1084 bankw = rtex->surface.bankw;
1085 bankh = rtex->surface.bankh;
1086 if (rtex->fmask.size)
1087 fmask_bankh = rtex->fmask.bank_height;
1088 else
1089 fmask_bankh = rtex->surface.bankh;
1090 tile_split = eg_tile_split(tile_split);
1091 macro_aspect = eg_macro_tile_aspect(macro_aspect);
1092 bankw = eg_bank_wh(bankw);
1093 bankh = eg_bank_wh(bankh);
1094 fmask_bankh = eg_bank_wh(fmask_bankh);
1095
1096 /* 128 bit formats require tile type = 1 */
1097 if (rscreen->b.chip_class == CAYMAN) {
1098 if (util_format_get_blocksize(surf->base.format) >= 16)
1099 non_disp_tiling = 1;
1100 }
1101 nbanks = eg_num_banks(rscreen->b.tiling_info.num_banks);
1102 desc = util_format_description(surf->base.format);
1103 for (i = 0; i < 4; i++) {
1104 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
1105 break;
1106 }
1107 }
1108
1109 color_attrib = S_028C74_TILE_SPLIT(tile_split)|
1110 S_028C74_NUM_BANKS(nbanks) |
1111 S_028C74_BANK_WIDTH(bankw) |
1112 S_028C74_BANK_HEIGHT(bankh) |
1113 S_028C74_MACRO_TILE_ASPECT(macro_aspect) |
1114 S_028C74_NON_DISP_TILING_ORDER(non_disp_tiling) |
1115 S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
1116
1117 if (rctx->b.chip_class == CAYMAN) {
1118 color_attrib |= S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] ==
1119 UTIL_FORMAT_SWIZZLE_1);
1120
1121 if (rtex->resource.b.b.nr_samples > 1) {
1122 unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples);
1123 color_attrib |= S_028C74_NUM_SAMPLES(log_samples) |
1124 S_028C74_NUM_FRAGMENTS(log_samples);
1125 }
1126 }
1127
1128 ntype = V_028C70_NUMBER_UNORM;
1129 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
1130 ntype = V_028C70_NUMBER_SRGB;
1131 else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
1132 if (desc->channel[i].normalized)
1133 ntype = V_028C70_NUMBER_SNORM;
1134 else if (desc->channel[i].pure_integer)
1135 ntype = V_028C70_NUMBER_SINT;
1136 } else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
1137 if (desc->channel[i].normalized)
1138 ntype = V_028C70_NUMBER_UNORM;
1139 else if (desc->channel[i].pure_integer)
1140 ntype = V_028C70_NUMBER_UINT;
1141 }
1142
1143 format = r600_translate_colorformat(rctx->b.chip_class, surf->base.format);
1144 assert(format != ~0);
1145
1146 swap = r600_translate_colorswap(surf->base.format);
1147 assert(swap != ~0);
1148
1149 if (rtex->resource.b.b.usage == PIPE_USAGE_STAGING) {
1150 endian = ENDIAN_NONE;
1151 } else {
1152 endian = r600_colorformat_endian_swap(format);
1153 }
1154
1155 /* blend clamp should be set for all NORM/SRGB types */
1156 if (ntype == V_028C70_NUMBER_UNORM || ntype == V_028C70_NUMBER_SNORM ||
1157 ntype == V_028C70_NUMBER_SRGB)
1158 blend_clamp = 1;
1159
1160 /* set blend bypass according to docs if SINT/UINT or
1161 8/24 COLOR variants */
1162 if (ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT ||
1163 format == V_028C70_COLOR_8_24 || format == V_028C70_COLOR_24_8 ||
1164 format == V_028C70_COLOR_X24_8_32_FLOAT) {
1165 blend_clamp = 0;
1166 blend_bypass = 1;
1167 }
1168
1169 surf->alphatest_bypass = ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT;
1170
1171 color_info |= S_028C70_FORMAT(format) |
1172 S_028C70_COMP_SWAP(swap) |
1173 S_028C70_BLEND_CLAMP(blend_clamp) |
1174 S_028C70_BLEND_BYPASS(blend_bypass) |
1175 S_028C70_NUMBER_TYPE(ntype) |
1176 S_028C70_ENDIAN(endian);
1177
1178 /* EXPORT_NORM is an optimzation that can be enabled for better
1179 * performance in certain cases.
1180 * EXPORT_NORM can be enabled if:
1181 * - 11-bit or smaller UNORM/SNORM/SRGB
1182 * - 16-bit or smaller FLOAT
1183 */
1184 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
1185 ((desc->channel[i].size < 12 &&
1186 desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
1187 ntype != V_028C70_NUMBER_UINT && ntype != V_028C70_NUMBER_SINT) ||
1188 (desc->channel[i].size < 17 &&
1189 desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT))) {
1190 color_info |= S_028C70_SOURCE_FORMAT(V_028C70_EXPORT_4C_16BPC);
1191 surf->export_16bpc = true;
1192 }
1193
1194 if (rtex->fmask.size) {
1195 color_info |= S_028C70_COMPRESSION(1);
1196 }
1197
1198 base_offset = rtex->resource.gpu_address;
1199
1200 /* XXX handle enabling of CB beyond BASE8 which has different offset */
1201 surf->cb_color_base = (base_offset + offset) >> 8;
1202 surf->cb_color_dim = color_dim;
1203 surf->cb_color_info = color_info;
1204 surf->cb_color_pitch = S_028C64_PITCH_TILE_MAX(pitch);
1205 surf->cb_color_slice = S_028C68_SLICE_TILE_MAX(slice);
1206 surf->cb_color_view = color_view;
1207 surf->cb_color_attrib = color_attrib;
1208 if (rtex->fmask.size) {
1209 surf->cb_color_fmask = (base_offset + rtex->fmask.offset) >> 8;
1210 surf->cb_color_fmask_slice = S_028C88_TILE_MAX(rtex->fmask.slice_tile_max);
1211 } else {
1212 surf->cb_color_fmask = surf->cb_color_base;
1213 surf->cb_color_fmask_slice = S_028C88_TILE_MAX(slice);
1214 }
1215
1216 surf->color_initialized = true;
1217 }
1218
1219 static void evergreen_init_depth_surface(struct r600_context *rctx,
1220 struct r600_surface *surf)
1221 {
1222 struct r600_screen *rscreen = rctx->screen;
1223 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1224 unsigned level = surf->base.u.tex.level;
1225 struct radeon_surf_level *levelinfo = &rtex->surface.level[level];
1226 uint64_t offset;
1227 unsigned format, array_mode;
1228 unsigned macro_aspect, tile_split, bankh, bankw, nbanks;
1229
1230
1231 format = r600_translate_dbformat(surf->base.format);
1232 assert(format != ~0);
1233
1234 offset = rtex->resource.gpu_address;
1235 offset += rtex->surface.level[level].offset;
1236
1237 switch (rtex->surface.level[level].mode) {
1238 case RADEON_SURF_MODE_2D:
1239 array_mode = V_028C70_ARRAY_2D_TILED_THIN1;
1240 break;
1241 case RADEON_SURF_MODE_1D:
1242 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1243 case RADEON_SURF_MODE_LINEAR:
1244 default:
1245 array_mode = V_028C70_ARRAY_1D_TILED_THIN1;
1246 break;
1247 }
1248 tile_split = rtex->surface.tile_split;
1249 macro_aspect = rtex->surface.mtilea;
1250 bankw = rtex->surface.bankw;
1251 bankh = rtex->surface.bankh;
1252 tile_split = eg_tile_split(tile_split);
1253 macro_aspect = eg_macro_tile_aspect(macro_aspect);
1254 bankw = eg_bank_wh(bankw);
1255 bankh = eg_bank_wh(bankh);
1256 nbanks = eg_num_banks(rscreen->b.tiling_info.num_banks);
1257 offset >>= 8;
1258
1259 surf->db_z_info = S_028040_ARRAY_MODE(array_mode) |
1260 S_028040_FORMAT(format) |
1261 S_028040_TILE_SPLIT(tile_split)|
1262 S_028040_NUM_BANKS(nbanks) |
1263 S_028040_BANK_WIDTH(bankw) |
1264 S_028040_BANK_HEIGHT(bankh) |
1265 S_028040_MACRO_TILE_ASPECT(macro_aspect);
1266 if (rscreen->b.chip_class == CAYMAN && rtex->resource.b.b.nr_samples > 1) {
1267 surf->db_z_info |= S_028040_NUM_SAMPLES(util_logbase2(rtex->resource.b.b.nr_samples));
1268 }
1269
1270 assert(levelinfo->nblk_x % 8 == 0 && levelinfo->nblk_y % 8 == 0);
1271
1272 surf->db_depth_base = offset;
1273 surf->db_depth_view = S_028008_SLICE_START(surf->base.u.tex.first_layer) |
1274 S_028008_SLICE_MAX(surf->base.u.tex.last_layer);
1275 surf->db_depth_size = S_028058_PITCH_TILE_MAX(levelinfo->nblk_x / 8 - 1) |
1276 S_028058_HEIGHT_TILE_MAX(levelinfo->nblk_y / 8 - 1);
1277 surf->db_depth_slice = S_02805C_SLICE_TILE_MAX(levelinfo->nblk_x *
1278 levelinfo->nblk_y / 64 - 1);
1279
1280 switch (surf->base.format) {
1281 case PIPE_FORMAT_Z24X8_UNORM:
1282 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1283 case PIPE_FORMAT_X8Z24_UNORM:
1284 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1285 surf->pa_su_poly_offset_db_fmt_cntl =
1286 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-24);
1287 break;
1288 case PIPE_FORMAT_Z32_FLOAT:
1289 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1290 surf->pa_su_poly_offset_db_fmt_cntl =
1291 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-23) |
1292 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
1293 break;
1294 case PIPE_FORMAT_Z16_UNORM:
1295 surf->pa_su_poly_offset_db_fmt_cntl =
1296 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-16);
1297 break;
1298 default:;
1299 }
1300
1301 if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
1302 uint64_t stencil_offset;
1303 unsigned stile_split = rtex->surface.stencil_tile_split;
1304
1305 stile_split = eg_tile_split(stile_split);
1306
1307 stencil_offset = rtex->surface.stencil_level[level].offset;
1308 stencil_offset += rtex->resource.gpu_address;
1309
1310 surf->db_stencil_base = stencil_offset >> 8;
1311 surf->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_8) |
1312 S_028044_TILE_SPLIT(stile_split);
1313 } else {
1314 surf->db_stencil_base = offset;
1315 /* DRM 2.6.18 allows the INVALID format to disable stencil.
1316 * Older kernels are out of luck. */
1317 surf->db_stencil_info = rctx->screen->b.info.drm_minor >= 18 ?
1318 S_028044_FORMAT(V_028044_STENCIL_INVALID) :
1319 S_028044_FORMAT(V_028044_STENCIL_8);
1320 }
1321
1322 /* use htile only for first level */
1323 if (rtex->htile_buffer && !level) {
1324 uint64_t va = rtex->htile_buffer->gpu_address;
1325 surf->db_htile_data_base = va >> 8;
1326 surf->db_htile_surface = S_028ABC_HTILE_WIDTH(1) |
1327 S_028ABC_HTILE_HEIGHT(1) |
1328 S_028ABC_FULL_CACHE(1);
1329 surf->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1);
1330 surf->db_preload_control = 0;
1331 }
1332
1333 surf->depth_initialized = true;
1334 }
1335
1336 static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
1337 const struct pipe_framebuffer_state *state)
1338 {
1339 struct r600_context *rctx = (struct r600_context *)ctx;
1340 struct r600_surface *surf;
1341 struct r600_texture *rtex;
1342 uint32_t i, log_samples;
1343
1344 if (rctx->framebuffer.state.nr_cbufs) {
1345 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE | R600_CONTEXT_FLUSH_AND_INV;
1346 rctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV_CB |
1347 R600_CONTEXT_FLUSH_AND_INV_CB_META;
1348 }
1349 if (rctx->framebuffer.state.zsbuf) {
1350 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE | R600_CONTEXT_FLUSH_AND_INV;
1351 rctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV_DB;
1352
1353 rtex = (struct r600_texture*)rctx->framebuffer.state.zsbuf->texture;
1354 if (rtex->htile_buffer) {
1355 rctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV_DB_META;
1356 }
1357 }
1358
1359 util_copy_framebuffer_state(&rctx->framebuffer.state, state);
1360
1361 /* Colorbuffers. */
1362 rctx->framebuffer.export_16bpc = state->nr_cbufs != 0;
1363 rctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
1364 util_format_is_pure_integer(state->cbufs[0]->format);
1365 rctx->framebuffer.compressed_cb_mask = 0;
1366 rctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
1367
1368 for (i = 0; i < state->nr_cbufs; i++) {
1369 surf = (struct r600_surface*)state->cbufs[i];
1370 if (!surf)
1371 continue;
1372
1373 rtex = (struct r600_texture*)surf->base.texture;
1374
1375 r600_context_add_resource_size(ctx, state->cbufs[i]->texture);
1376
1377 if (!surf->color_initialized) {
1378 evergreen_init_color_surface(rctx, surf);
1379 }
1380
1381 if (!surf->export_16bpc) {
1382 rctx->framebuffer.export_16bpc = false;
1383 }
1384
1385 if (rtex->fmask.size && rtex->cmask.size) {
1386 rctx->framebuffer.compressed_cb_mask |= 1 << i;
1387 }
1388 }
1389
1390 /* Update alpha-test state dependencies.
1391 * Alpha-test is done on the first colorbuffer only. */
1392 if (state->nr_cbufs) {
1393 bool alphatest_bypass = false;
1394 bool export_16bpc = true;
1395
1396 surf = (struct r600_surface*)state->cbufs[0];
1397 if (surf) {
1398 alphatest_bypass = surf->alphatest_bypass;
1399 export_16bpc = surf->export_16bpc;
1400 }
1401
1402 if (rctx->alphatest_state.bypass != alphatest_bypass) {
1403 rctx->alphatest_state.bypass = alphatest_bypass;
1404 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1405 }
1406 if (rctx->alphatest_state.cb0_export_16bpc != export_16bpc) {
1407 rctx->alphatest_state.cb0_export_16bpc = export_16bpc;
1408 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1409 }
1410 }
1411
1412 /* ZS buffer. */
1413 if (state->zsbuf) {
1414 surf = (struct r600_surface*)state->zsbuf;
1415
1416 r600_context_add_resource_size(ctx, state->zsbuf->texture);
1417
1418 if (!surf->depth_initialized) {
1419 evergreen_init_depth_surface(rctx, surf);
1420 }
1421
1422 if (state->zsbuf->format != rctx->poly_offset_state.zs_format) {
1423 rctx->poly_offset_state.zs_format = state->zsbuf->format;
1424 r600_mark_atom_dirty(rctx, &rctx->poly_offset_state.atom);
1425 }
1426
1427 if (rctx->db_state.rsurf != surf) {
1428 rctx->db_state.rsurf = surf;
1429 r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1430 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1431 }
1432 } else if (rctx->db_state.rsurf) {
1433 rctx->db_state.rsurf = NULL;
1434 r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1435 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1436 }
1437
1438 if (rctx->cb_misc_state.nr_cbufs != state->nr_cbufs) {
1439 rctx->cb_misc_state.nr_cbufs = state->nr_cbufs;
1440 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom);
1441 }
1442
1443 if (state->nr_cbufs == 0 && rctx->alphatest_state.bypass) {
1444 rctx->alphatest_state.bypass = false;
1445 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1446 }
1447
1448 log_samples = util_logbase2(rctx->framebuffer.nr_samples);
1449 /* This is for Cayman to program SAMPLE_RATE, and for RV770 to fix a hw bug. */
1450 if ((rctx->b.chip_class == CAYMAN ||
1451 rctx->b.family == CHIP_RV770) &&
1452 rctx->db_misc_state.log_samples != log_samples) {
1453 rctx->db_misc_state.log_samples = log_samples;
1454 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1455 }
1456
1457
1458 /* Calculate the CS size. */
1459 rctx->framebuffer.atom.num_dw = 4; /* SCISSOR */
1460
1461 /* MSAA. */
1462 if (rctx->b.chip_class == EVERGREEN)
1463 rctx->framebuffer.atom.num_dw += 17; /* Evergreen */
1464 else
1465 rctx->framebuffer.atom.num_dw += 28; /* Cayman */
1466
1467 /* Colorbuffers. */
1468 rctx->framebuffer.atom.num_dw += state->nr_cbufs * 23;
1469 if (rctx->keep_tiling_flags)
1470 rctx->framebuffer.atom.num_dw += state->nr_cbufs * 2;
1471 rctx->framebuffer.atom.num_dw += (12 - state->nr_cbufs) * 3;
1472
1473 /* ZS buffer. */
1474 if (state->zsbuf) {
1475 rctx->framebuffer.atom.num_dw += 24;
1476 if (rctx->keep_tiling_flags)
1477 rctx->framebuffer.atom.num_dw += 2;
1478 } else if (rctx->screen->b.info.drm_minor >= 18) {
1479 rctx->framebuffer.atom.num_dw += 4;
1480 }
1481
1482 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom);
1483
1484 r600_set_sample_locations_constant_buffer(rctx);
1485 }
1486
1487 static void evergreen_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
1488 {
1489 struct r600_context *rctx = (struct r600_context *)ctx;
1490
1491 if (rctx->ps_iter_samples == min_samples)
1492 return;
1493
1494 rctx->ps_iter_samples = min_samples;
1495 if (rctx->framebuffer.nr_samples > 1) {
1496 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom);
1497 }
1498 }
1499
1500 /* 8xMSAA */
1501 static uint32_t sample_locs_8x[] = {
1502 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1503 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1504 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1505 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1506 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1507 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1508 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1509 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1510 };
1511 static unsigned max_dist_8x = 7;
1512
1513 static void evergreen_get_sample_position(struct pipe_context *ctx,
1514 unsigned sample_count,
1515 unsigned sample_index,
1516 float *out_value)
1517 {
1518 int offset, index;
1519 struct {
1520 int idx:4;
1521 } val;
1522 switch (sample_count) {
1523 case 1:
1524 default:
1525 out_value[0] = out_value[1] = 0.5;
1526 break;
1527 case 2:
1528 offset = 4 * (sample_index * 2);
1529 val.idx = (eg_sample_locs_2x[0] >> offset) & 0xf;
1530 out_value[0] = (float)(val.idx + 8) / 16.0f;
1531 val.idx = (eg_sample_locs_2x[0] >> (offset + 4)) & 0xf;
1532 out_value[1] = (float)(val.idx + 8) / 16.0f;
1533 break;
1534 case 4:
1535 offset = 4 * (sample_index * 2);
1536 val.idx = (eg_sample_locs_4x[0] >> offset) & 0xf;
1537 out_value[0] = (float)(val.idx + 8) / 16.0f;
1538 val.idx = (eg_sample_locs_4x[0] >> (offset + 4)) & 0xf;
1539 out_value[1] = (float)(val.idx + 8) / 16.0f;
1540 break;
1541 case 8:
1542 offset = 4 * (sample_index % 4 * 2);
1543 index = (sample_index / 4);
1544 val.idx = (sample_locs_8x[index] >> offset) & 0xf;
1545 out_value[0] = (float)(val.idx + 8) / 16.0f;
1546 val.idx = (sample_locs_8x[index] >> (offset + 4)) & 0xf;
1547 out_value[1] = (float)(val.idx + 8) / 16.0f;
1548 break;
1549 }
1550 }
1551
1552 static void evergreen_emit_msaa_state(struct r600_context *rctx, int nr_samples, int ps_iter_samples)
1553 {
1554
1555 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1556 unsigned max_dist = 0;
1557
1558 switch (nr_samples) {
1559 default:
1560 nr_samples = 0;
1561 break;
1562 case 2:
1563 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, Elements(eg_sample_locs_2x));
1564 radeon_emit_array(cs, eg_sample_locs_2x, Elements(eg_sample_locs_2x));
1565 max_dist = eg_max_dist_2x;
1566 break;
1567 case 4:
1568 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, Elements(eg_sample_locs_4x));
1569 radeon_emit_array(cs, eg_sample_locs_4x, Elements(eg_sample_locs_4x));
1570 max_dist = eg_max_dist_4x;
1571 break;
1572 case 8:
1573 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, Elements(sample_locs_8x));
1574 radeon_emit_array(cs, sample_locs_8x, Elements(sample_locs_8x));
1575 max_dist = max_dist_8x;
1576 break;
1577 }
1578
1579 if (nr_samples > 1) {
1580 radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1581 radeon_emit(cs, S_028C00_LAST_PIXEL(1) |
1582 S_028C00_EXPAND_LINE_WIDTH(1)); /* R_028C00_PA_SC_LINE_CNTL */
1583 radeon_emit(cs, S_028C04_MSAA_NUM_SAMPLES(util_logbase2(nr_samples)) |
1584 S_028C04_MAX_SAMPLE_DIST(max_dist)); /* R_028C04_PA_SC_AA_CONFIG */
1585 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1,
1586 EG_S_028A4C_PS_ITER_SAMPLE(ps_iter_samples > 1) |
1587 EG_S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
1588 EG_S_028A4C_FORCE_EOV_REZ_ENABLE(1));
1589 } else {
1590 radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1591 radeon_emit(cs, S_028C00_LAST_PIXEL(1)); /* R_028C00_PA_SC_LINE_CNTL */
1592 radeon_emit(cs, 0); /* R_028C04_PA_SC_AA_CONFIG */
1593 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1,
1594 EG_S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
1595 EG_S_028A4C_FORCE_EOV_REZ_ENABLE(1));
1596 }
1597 }
1598
1599 static void evergreen_emit_framebuffer_state(struct r600_context *rctx, struct r600_atom *atom)
1600 {
1601 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1602 struct pipe_framebuffer_state *state = &rctx->framebuffer.state;
1603 unsigned nr_cbufs = state->nr_cbufs;
1604 unsigned i, tl, br;
1605 struct r600_texture *tex = NULL;
1606 struct r600_surface *cb = NULL;
1607
1608 /* XXX support more colorbuffers once we need them */
1609 assert(nr_cbufs <= 8);
1610 if (nr_cbufs > 8)
1611 nr_cbufs = 8;
1612
1613 /* Colorbuffers. */
1614 for (i = 0; i < nr_cbufs; i++) {
1615 unsigned reloc, cmask_reloc;
1616
1617 cb = (struct r600_surface*)state->cbufs[i];
1618 if (!cb) {
1619 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C,
1620 S_028C70_FORMAT(V_028C70_COLOR_INVALID));
1621 continue;
1622 }
1623
1624 tex = (struct r600_texture *)cb->base.texture;
1625 reloc = radeon_add_to_buffer_list(&rctx->b,
1626 &rctx->b.gfx,
1627 (struct r600_resource*)cb->base.texture,
1628 RADEON_USAGE_READWRITE,
1629 tex->surface.nsamples > 1 ?
1630 RADEON_PRIO_COLOR_BUFFER_MSAA :
1631 RADEON_PRIO_COLOR_BUFFER);
1632
1633 if (tex->cmask_buffer && tex->cmask_buffer != &tex->resource) {
1634 cmask_reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx,
1635 tex->cmask_buffer, RADEON_USAGE_READWRITE,
1636 RADEON_PRIO_CMASK);
1637 } else {
1638 cmask_reloc = reloc;
1639 }
1640
1641 radeon_set_context_reg_seq(cs, R_028C60_CB_COLOR0_BASE + i * 0x3C, 13);
1642 radeon_emit(cs, cb->cb_color_base); /* R_028C60_CB_COLOR0_BASE */
1643 radeon_emit(cs, cb->cb_color_pitch); /* R_028C64_CB_COLOR0_PITCH */
1644 radeon_emit(cs, cb->cb_color_slice); /* R_028C68_CB_COLOR0_SLICE */
1645 radeon_emit(cs, cb->cb_color_view); /* R_028C6C_CB_COLOR0_VIEW */
1646 radeon_emit(cs, cb->cb_color_info | tex->cb_color_info); /* R_028C70_CB_COLOR0_INFO */
1647 radeon_emit(cs, cb->cb_color_attrib); /* R_028C74_CB_COLOR0_ATTRIB */
1648 radeon_emit(cs, cb->cb_color_dim); /* R_028C78_CB_COLOR0_DIM */
1649 radeon_emit(cs, tex->cmask.base_address_reg); /* R_028C7C_CB_COLOR0_CMASK */
1650 radeon_emit(cs, tex->cmask.slice_tile_max); /* R_028C80_CB_COLOR0_CMASK_SLICE */
1651 radeon_emit(cs, cb->cb_color_fmask); /* R_028C84_CB_COLOR0_FMASK */
1652 radeon_emit(cs, cb->cb_color_fmask_slice); /* R_028C88_CB_COLOR0_FMASK_SLICE */
1653 radeon_emit(cs, tex->color_clear_value[0]); /* R_028C8C_CB_COLOR0_CLEAR_WORD0 */
1654 radeon_emit(cs, tex->color_clear_value[1]); /* R_028C90_CB_COLOR0_CLEAR_WORD1 */
1655
1656 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C60_CB_COLOR0_BASE */
1657 radeon_emit(cs, reloc);
1658
1659 if (!rctx->keep_tiling_flags) {
1660 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C70_CB_COLOR0_INFO */
1661 radeon_emit(cs, reloc);
1662 }
1663
1664 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C74_CB_COLOR0_ATTRIB */
1665 radeon_emit(cs, reloc);
1666
1667 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C7C_CB_COLOR0_CMASK */
1668 radeon_emit(cs, cmask_reloc);
1669
1670 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C84_CB_COLOR0_FMASK */
1671 radeon_emit(cs, reloc);
1672 }
1673 /* set CB_COLOR1_INFO for possible dual-src blending */
1674 if (i == 1 && state->cbufs[0]) {
1675 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + 1 * 0x3C,
1676 cb->cb_color_info | tex->cb_color_info);
1677
1678 if (!rctx->keep_tiling_flags) {
1679 unsigned reloc = radeon_add_to_buffer_list(&rctx->b,
1680 &rctx->b.gfx,
1681 (struct r600_resource*)state->cbufs[0]->texture,
1682 RADEON_USAGE_READWRITE,
1683 RADEON_PRIO_COLOR_BUFFER);
1684
1685 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C70_CB_COLOR0_INFO */
1686 radeon_emit(cs, reloc);
1687 }
1688 i++;
1689 }
1690 if (rctx->keep_tiling_flags) {
1691 for (; i < 8 ; i++) {
1692 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C, 0);
1693 }
1694 for (; i < 12; i++) {
1695 radeon_set_context_reg(cs, R_028E50_CB_COLOR8_INFO + (i - 8) * 0x1C, 0);
1696 }
1697 }
1698
1699 /* ZS buffer. */
1700 if (state->zsbuf) {
1701 struct r600_surface *zb = (struct r600_surface*)state->zsbuf;
1702 unsigned reloc = radeon_add_to_buffer_list(&rctx->b,
1703 &rctx->b.gfx,
1704 (struct r600_resource*)state->zsbuf->texture,
1705 RADEON_USAGE_READWRITE,
1706 zb->base.texture->nr_samples > 1 ?
1707 RADEON_PRIO_DEPTH_BUFFER_MSAA :
1708 RADEON_PRIO_DEPTH_BUFFER);
1709
1710 radeon_set_context_reg(cs, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
1711 zb->pa_su_poly_offset_db_fmt_cntl);
1712 radeon_set_context_reg(cs, R_028008_DB_DEPTH_VIEW, zb->db_depth_view);
1713
1714 radeon_set_context_reg_seq(cs, R_028040_DB_Z_INFO, 8);
1715 radeon_emit(cs, zb->db_z_info); /* R_028040_DB_Z_INFO */
1716 radeon_emit(cs, zb->db_stencil_info); /* R_028044_DB_STENCIL_INFO */
1717 radeon_emit(cs, zb->db_depth_base); /* R_028048_DB_Z_READ_BASE */
1718 radeon_emit(cs, zb->db_stencil_base); /* R_02804C_DB_STENCIL_READ_BASE */
1719 radeon_emit(cs, zb->db_depth_base); /* R_028050_DB_Z_WRITE_BASE */
1720 radeon_emit(cs, zb->db_stencil_base); /* R_028054_DB_STENCIL_WRITE_BASE */
1721 radeon_emit(cs, zb->db_depth_size); /* R_028058_DB_DEPTH_SIZE */
1722 radeon_emit(cs, zb->db_depth_slice); /* R_02805C_DB_DEPTH_SLICE */
1723
1724 if (!rctx->keep_tiling_flags) {
1725 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028040_DB_Z_INFO */
1726 radeon_emit(cs, reloc);
1727 }
1728
1729 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028048_DB_Z_READ_BASE */
1730 radeon_emit(cs, reloc);
1731
1732 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_02804C_DB_STENCIL_READ_BASE */
1733 radeon_emit(cs, reloc);
1734
1735 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028050_DB_Z_WRITE_BASE */
1736 radeon_emit(cs, reloc);
1737
1738 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028054_DB_STENCIL_WRITE_BASE */
1739 radeon_emit(cs, reloc);
1740 } else if (rctx->screen->b.info.drm_minor >= 18) {
1741 /* DRM 2.6.18 allows the INVALID format to disable depth/stencil.
1742 * Older kernels are out of luck. */
1743 radeon_set_context_reg_seq(cs, R_028040_DB_Z_INFO, 2);
1744 radeon_emit(cs, S_028040_FORMAT(V_028040_Z_INVALID)); /* R_028040_DB_Z_INFO */
1745 radeon_emit(cs, S_028044_FORMAT(V_028044_STENCIL_INVALID)); /* R_028044_DB_STENCIL_INFO */
1746 }
1747
1748 /* Framebuffer dimensions. */
1749 evergreen_get_scissor_rect(rctx, 0, 0, state->width, state->height, &tl, &br);
1750
1751 radeon_set_context_reg_seq(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL, 2);
1752 radeon_emit(cs, tl); /* R_028204_PA_SC_WINDOW_SCISSOR_TL */
1753 radeon_emit(cs, br); /* R_028208_PA_SC_WINDOW_SCISSOR_BR */
1754
1755 if (rctx->b.chip_class == EVERGREEN) {
1756 evergreen_emit_msaa_state(rctx, rctx->framebuffer.nr_samples, rctx->ps_iter_samples);
1757 } else {
1758 cayman_emit_msaa_sample_locs(cs, rctx->framebuffer.nr_samples);
1759 cayman_emit_msaa_config(cs, rctx->framebuffer.nr_samples, rctx->ps_iter_samples, 0);
1760 }
1761 }
1762
1763 static void evergreen_emit_polygon_offset(struct r600_context *rctx, struct r600_atom *a)
1764 {
1765 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1766 struct r600_poly_offset_state *state = (struct r600_poly_offset_state*)a;
1767 float offset_units = state->offset_units;
1768 float offset_scale = state->offset_scale;
1769
1770 switch (state->zs_format) {
1771 case PIPE_FORMAT_Z24X8_UNORM:
1772 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1773 case PIPE_FORMAT_X8Z24_UNORM:
1774 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1775 offset_units *= 2.0f;
1776 break;
1777 case PIPE_FORMAT_Z16_UNORM:
1778 offset_units *= 4.0f;
1779 break;
1780 default:;
1781 }
1782
1783 radeon_set_context_reg_seq(cs, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE, 4);
1784 radeon_emit(cs, fui(offset_scale));
1785 radeon_emit(cs, fui(offset_units));
1786 radeon_emit(cs, fui(offset_scale));
1787 radeon_emit(cs, fui(offset_units));
1788 }
1789
1790 static void evergreen_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1791 {
1792 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1793 struct r600_cb_misc_state *a = (struct r600_cb_misc_state*)atom;
1794 unsigned fb_colormask = (1ULL << ((unsigned)a->nr_cbufs * 4)) - 1;
1795 unsigned ps_colormask = (1ULL << ((unsigned)a->nr_ps_color_outputs * 4)) - 1;
1796
1797 radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1798 radeon_emit(cs, a->blend_colormask & fb_colormask); /* R_028238_CB_TARGET_MASK */
1799 /* This must match the used export instructions exactly.
1800 * Other values may lead to undefined behavior and hangs.
1801 */
1802 radeon_emit(cs, ps_colormask); /* R_02823C_CB_SHADER_MASK */
1803 }
1804
1805 static void evergreen_emit_db_state(struct r600_context *rctx, struct r600_atom *atom)
1806 {
1807 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1808 struct r600_db_state *a = (struct r600_db_state*)atom;
1809
1810 if (a->rsurf && a->rsurf->db_htile_surface) {
1811 struct r600_texture *rtex = (struct r600_texture *)a->rsurf->base.texture;
1812 unsigned reloc_idx;
1813
1814 radeon_set_context_reg(cs, R_02802C_DB_DEPTH_CLEAR, fui(rtex->depth_clear_value));
1815 radeon_set_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, a->rsurf->db_htile_surface);
1816 radeon_set_context_reg(cs, R_028AC8_DB_PRELOAD_CONTROL, a->rsurf->db_preload_control);
1817 radeon_set_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, a->rsurf->db_htile_data_base);
1818 reloc_idx = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rtex->htile_buffer,
1819 RADEON_USAGE_READWRITE, RADEON_PRIO_HTILE);
1820 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1821 cs->buf[cs->cdw++] = reloc_idx;
1822 } else {
1823 radeon_set_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, 0);
1824 radeon_set_context_reg(cs, R_028AC8_DB_PRELOAD_CONTROL, 0);
1825 }
1826 }
1827
1828 static void evergreen_emit_db_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1829 {
1830 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1831 struct r600_db_misc_state *a = (struct r600_db_misc_state*)atom;
1832 unsigned db_render_control = 0;
1833 unsigned db_count_control = 0;
1834 unsigned db_render_override =
1835 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE) |
1836 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE);
1837
1838 if (a->occlusion_query_enabled) {
1839 db_count_control |= S_028004_PERFECT_ZPASS_COUNTS(1);
1840 if (rctx->b.chip_class == CAYMAN) {
1841 db_count_control |= S_028004_SAMPLE_RATE(a->log_samples);
1842 }
1843 db_render_override |= S_02800C_NOOP_CULL_DISABLE(1);
1844 }
1845
1846 /* This is to fix a lockup when hyperz and alpha test are enabled at
1847 * the same time somehow GPU get confuse on which order to pick for
1848 * z test
1849 */
1850 if (rctx->alphatest_state.sx_alpha_test_control)
1851 db_render_override |= S_02800C_FORCE_SHADER_Z_ORDER(1);
1852
1853 if (a->flush_depthstencil_through_cb) {
1854 assert(a->copy_depth || a->copy_stencil);
1855
1856 db_render_control |= S_028000_DEPTH_COPY_ENABLE(a->copy_depth) |
1857 S_028000_STENCIL_COPY_ENABLE(a->copy_stencil) |
1858 S_028000_COPY_CENTROID(1) |
1859 S_028000_COPY_SAMPLE(a->copy_sample);
1860 } else if (a->flush_depth_inplace || a->flush_stencil_inplace) {
1861 db_render_control |= S_028000_DEPTH_COMPRESS_DISABLE(a->flush_depth_inplace) |
1862 S_028000_STENCIL_COMPRESS_DISABLE(a->flush_stencil_inplace);
1863 db_render_override |= S_02800C_DISABLE_PIXEL_RATE_TILES(1);
1864 }
1865 if (a->htile_clear) {
1866 /* FIXME we might want to disable cliprect here */
1867 db_render_control |= S_028000_DEPTH_CLEAR_ENABLE(1);
1868 }
1869
1870 radeon_set_context_reg_seq(cs, R_028000_DB_RENDER_CONTROL, 2);
1871 radeon_emit(cs, db_render_control); /* R_028000_DB_RENDER_CONTROL */
1872 radeon_emit(cs, db_count_control); /* R_028004_DB_COUNT_CONTROL */
1873 radeon_set_context_reg(cs, R_02800C_DB_RENDER_OVERRIDE, db_render_override);
1874 radeon_set_context_reg(cs, R_02880C_DB_SHADER_CONTROL, a->db_shader_control);
1875 }
1876
1877 static void evergreen_emit_vertex_buffers(struct r600_context *rctx,
1878 struct r600_vertexbuf_state *state,
1879 unsigned resource_offset,
1880 unsigned pkt_flags)
1881 {
1882 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1883 uint32_t dirty_mask = state->dirty_mask;
1884
1885 while (dirty_mask) {
1886 struct pipe_vertex_buffer *vb;
1887 struct r600_resource *rbuffer;
1888 uint64_t va;
1889 unsigned buffer_index = u_bit_scan(&dirty_mask);
1890
1891 vb = &state->vb[buffer_index];
1892 rbuffer = (struct r600_resource*)vb->buffer;
1893 assert(rbuffer);
1894
1895 va = rbuffer->gpu_address + vb->buffer_offset;
1896
1897 /* fetch resources start at index 992 */
1898 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags);
1899 radeon_emit(cs, (resource_offset + buffer_index) * 8);
1900 radeon_emit(cs, va); /* RESOURCEi_WORD0 */
1901 radeon_emit(cs, rbuffer->b.b.width0 - vb->buffer_offset - 1); /* RESOURCEi_WORD1 */
1902 radeon_emit(cs, /* RESOURCEi_WORD2 */
1903 S_030008_ENDIAN_SWAP(r600_endian_swap(32)) |
1904 S_030008_STRIDE(vb->stride) |
1905 S_030008_BASE_ADDRESS_HI(va >> 32UL));
1906 radeon_emit(cs, /* RESOURCEi_WORD3 */
1907 S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) |
1908 S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) |
1909 S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) |
1910 S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W));
1911 radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1912 radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1913 radeon_emit(cs, 0); /* RESOURCEi_WORD6 */
1914 radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD7 */
1915
1916 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags);
1917 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1918 RADEON_USAGE_READ, RADEON_PRIO_VERTEX_BUFFER));
1919 }
1920 state->dirty_mask = 0;
1921 }
1922
1923 static void evergreen_fs_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom * atom)
1924 {
1925 evergreen_emit_vertex_buffers(rctx, &rctx->vertex_buffer_state, EG_FETCH_CONSTANTS_OFFSET_FS, 0);
1926 }
1927
1928 static void evergreen_cs_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom * atom)
1929 {
1930 evergreen_emit_vertex_buffers(rctx, &rctx->cs_vertex_buffer_state, EG_FETCH_CONSTANTS_OFFSET_CS,
1931 RADEON_CP_PACKET3_COMPUTE_MODE);
1932 }
1933
1934 static void evergreen_emit_constant_buffers(struct r600_context *rctx,
1935 struct r600_constbuf_state *state,
1936 unsigned buffer_id_base,
1937 unsigned reg_alu_constbuf_size,
1938 unsigned reg_alu_const_cache,
1939 unsigned pkt_flags)
1940 {
1941 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1942 uint32_t dirty_mask = state->dirty_mask;
1943
1944 while (dirty_mask) {
1945 struct pipe_constant_buffer *cb;
1946 struct r600_resource *rbuffer;
1947 uint64_t va;
1948 unsigned buffer_index = ffs(dirty_mask) - 1;
1949 unsigned gs_ring_buffer = (buffer_index == R600_GS_RING_CONST_BUFFER);
1950
1951 cb = &state->cb[buffer_index];
1952 rbuffer = (struct r600_resource*)cb->buffer;
1953 assert(rbuffer);
1954
1955 va = rbuffer->gpu_address + cb->buffer_offset;
1956
1957 if (!gs_ring_buffer) {
1958 radeon_set_context_reg_flag(cs, reg_alu_constbuf_size + buffer_index * 4,
1959 DIV_ROUND_UP(cb->buffer_size, 256), pkt_flags);
1960 radeon_set_context_reg_flag(cs, reg_alu_const_cache + buffer_index * 4, va >> 8,
1961 pkt_flags);
1962 }
1963
1964 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags);
1965 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1966 RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER));
1967
1968 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags);
1969 radeon_emit(cs, (buffer_id_base + buffer_index) * 8);
1970 radeon_emit(cs, va); /* RESOURCEi_WORD0 */
1971 radeon_emit(cs, rbuffer->b.b.width0 - cb->buffer_offset - 1); /* RESOURCEi_WORD1 */
1972 radeon_emit(cs, /* RESOURCEi_WORD2 */
1973 S_030008_ENDIAN_SWAP(gs_ring_buffer ? ENDIAN_NONE : r600_endian_swap(32)) |
1974 S_030008_STRIDE(gs_ring_buffer ? 4 : 16) |
1975 S_030008_BASE_ADDRESS_HI(va >> 32UL) |
1976 S_030008_DATA_FORMAT(FMT_32_32_32_32_FLOAT));
1977 radeon_emit(cs, /* RESOURCEi_WORD3 */
1978 S_03000C_UNCACHED(gs_ring_buffer ? 1 : 0) |
1979 S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) |
1980 S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) |
1981 S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) |
1982 S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W));
1983 radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1984 radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1985 radeon_emit(cs, 0); /* RESOURCEi_WORD6 */
1986 radeon_emit(cs, /* RESOURCEi_WORD7 */
1987 S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_BUFFER));
1988
1989 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags);
1990 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1991 RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER));
1992
1993 dirty_mask &= ~(1 << buffer_index);
1994 }
1995 state->dirty_mask = 0;
1996 }
1997
1998 /* VS constants can be in VS/ES (same space) or LS if tess is enabled */
1999 static void evergreen_emit_vs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2000 {
2001 if (rctx->vs_shader->current->shader.vs_as_ls) {
2002 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX],
2003 EG_FETCH_CONSTANTS_OFFSET_LS,
2004 R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0,
2005 R_028F40_ALU_CONST_CACHE_LS_0,
2006 0 /* PKT3 flags */);
2007 } else {
2008 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX],
2009 EG_FETCH_CONSTANTS_OFFSET_VS,
2010 R_028180_ALU_CONST_BUFFER_SIZE_VS_0,
2011 R_028980_ALU_CONST_CACHE_VS_0,
2012 0 /* PKT3 flags */);
2013 }
2014 }
2015
2016 static void evergreen_emit_gs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2017 {
2018 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY],
2019 EG_FETCH_CONSTANTS_OFFSET_GS,
2020 R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0,
2021 R_0289C0_ALU_CONST_CACHE_GS_0,
2022 0 /* PKT3 flags */);
2023 }
2024
2025 static void evergreen_emit_ps_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2026 {
2027 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT],
2028 EG_FETCH_CONSTANTS_OFFSET_PS,
2029 R_028140_ALU_CONST_BUFFER_SIZE_PS_0,
2030 R_028940_ALU_CONST_CACHE_PS_0,
2031 0 /* PKT3 flags */);
2032 }
2033
2034 static void evergreen_emit_cs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2035 {
2036 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_COMPUTE],
2037 EG_FETCH_CONSTANTS_OFFSET_CS,
2038 R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0,
2039 R_028F40_ALU_CONST_CACHE_LS_0,
2040 RADEON_CP_PACKET3_COMPUTE_MODE);
2041 }
2042
2043 /* tes constants can be emitted to VS or ES - which are common */
2044 static void evergreen_emit_tes_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2045 {
2046 if (!rctx->tes_shader)
2047 return;
2048 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_EVAL],
2049 EG_FETCH_CONSTANTS_OFFSET_VS,
2050 R_028180_ALU_CONST_BUFFER_SIZE_VS_0,
2051 R_028980_ALU_CONST_CACHE_VS_0,
2052 0);
2053 }
2054
2055 static void evergreen_emit_tcs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
2056 {
2057 if (!rctx->tes_shader)
2058 return;
2059 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_CTRL],
2060 EG_FETCH_CONSTANTS_OFFSET_HS,
2061 R_028F80_ALU_CONST_BUFFER_SIZE_HS_0,
2062 R_028F00_ALU_CONST_CACHE_HS_0,
2063 0);
2064 }
2065
2066 static void evergreen_emit_sampler_views(struct r600_context *rctx,
2067 struct r600_samplerview_state *state,
2068 unsigned resource_id_base, unsigned pkt_flags)
2069 {
2070 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
2071 uint32_t dirty_mask = state->dirty_mask;
2072
2073 while (dirty_mask) {
2074 struct r600_pipe_sampler_view *rview;
2075 unsigned resource_index = u_bit_scan(&dirty_mask);
2076 unsigned reloc;
2077
2078 rview = state->views[resource_index];
2079 assert(rview);
2080
2081 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags);
2082 radeon_emit(cs, (resource_id_base + resource_index) * 8);
2083 radeon_emit_array(cs, rview->tex_resource_words, 8);
2084
2085 reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rview->tex_resource,
2086 RADEON_USAGE_READ,
2087 r600_get_sampler_view_priority(rview->tex_resource));
2088 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags);
2089 radeon_emit(cs, reloc);
2090
2091 if (!rview->skip_mip_address_reloc) {
2092 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags);
2093 radeon_emit(cs, reloc);
2094 }
2095 }
2096 state->dirty_mask = 0;
2097 }
2098
2099 static void evergreen_emit_vs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2100 {
2101 if (rctx->vs_shader->current->shader.vs_as_ls) {
2102 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views,
2103 EG_FETCH_CONSTANTS_OFFSET_LS + R600_MAX_CONST_BUFFERS, 0);
2104 } else {
2105 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views,
2106 EG_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS, 0);
2107 }
2108 }
2109
2110 static void evergreen_emit_gs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2111 {
2112 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views,
2113 EG_FETCH_CONSTANTS_OFFSET_GS + R600_MAX_CONST_BUFFERS, 0);
2114 }
2115
2116 static void evergreen_emit_tcs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2117 {
2118 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL].views,
2119 EG_FETCH_CONSTANTS_OFFSET_HS + R600_MAX_CONST_BUFFERS, 0);
2120 }
2121
2122 static void evergreen_emit_tes_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2123 {
2124 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL].views,
2125 EG_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS, 0);
2126 }
2127
2128 static void evergreen_emit_ps_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2129 {
2130 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views,
2131 EG_FETCH_CONSTANTS_OFFSET_PS + R600_MAX_CONST_BUFFERS, 0);
2132 }
2133
2134 static void evergreen_emit_cs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
2135 {
2136 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE].views,
2137 EG_FETCH_CONSTANTS_OFFSET_CS + 2, RADEON_CP_PACKET3_COMPUTE_MODE);
2138 }
2139
2140 static void evergreen_emit_sampler_states(struct r600_context *rctx,
2141 struct r600_textures_info *texinfo,
2142 unsigned resource_id_base,
2143 unsigned border_index_reg,
2144 unsigned pkt_flags)
2145 {
2146 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
2147 uint32_t dirty_mask = texinfo->states.dirty_mask;
2148
2149 while (dirty_mask) {
2150 struct r600_pipe_sampler_state *rstate;
2151 unsigned i = u_bit_scan(&dirty_mask);
2152
2153 rstate = texinfo->states.states[i];
2154 assert(rstate);
2155
2156 radeon_emit(cs, PKT3(PKT3_SET_SAMPLER, 3, 0) | pkt_flags);
2157 radeon_emit(cs, (resource_id_base + i) * 3);
2158 radeon_emit_array(cs, rstate->tex_sampler_words, 3);
2159
2160 if (rstate->border_color_use) {
2161 radeon_set_config_reg_seq(cs, border_index_reg, 5);
2162 radeon_emit(cs, i);
2163 radeon_emit_array(cs, rstate->border_color.ui, 4);
2164 }
2165 }
2166 texinfo->states.dirty_mask = 0;
2167 }
2168
2169 static void evergreen_emit_vs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2170 {
2171 if (rctx->vs_shader->current->shader.vs_as_ls) {
2172 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 72,
2173 R_00A450_TD_LS_SAMPLER0_BORDER_COLOR_INDEX, 0);
2174 } else {
2175 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 18,
2176 R_00A414_TD_VS_SAMPLER0_BORDER_INDEX, 0);
2177 }
2178 }
2179
2180 static void evergreen_emit_gs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2181 {
2182 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY], 36,
2183 R_00A428_TD_GS_SAMPLER0_BORDER_INDEX, 0);
2184 }
2185
2186 static void evergreen_emit_tcs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2187 {
2188 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL], 54,
2189 R_00A43C_TD_HS_SAMPLER0_BORDER_COLOR_INDEX, 0);
2190 }
2191
2192 static void evergreen_emit_tes_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2193 {
2194 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL], 18,
2195 R_00A414_TD_VS_SAMPLER0_BORDER_INDEX, 0);
2196 }
2197
2198 static void evergreen_emit_ps_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2199 {
2200 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT], 0,
2201 R_00A400_TD_PS_SAMPLER0_BORDER_INDEX, 0);
2202 }
2203
2204 static void evergreen_emit_cs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
2205 {
2206 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE], 90,
2207 R_00A464_TD_CS_SAMPLER0_BORDER_INDEX,
2208 RADEON_CP_PACKET3_COMPUTE_MODE);
2209 }
2210
2211 static void evergreen_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
2212 {
2213 struct r600_sample_mask *s = (struct r600_sample_mask*)a;
2214 uint8_t mask = s->sample_mask;
2215
2216 radeon_set_context_reg(rctx->b.gfx.cs, R_028C3C_PA_SC_AA_MASK,
2217 mask | (mask << 8) | (mask << 16) | (mask << 24));
2218 }
2219
2220 static void cayman_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
2221 {
2222 struct r600_sample_mask *s = (struct r600_sample_mask*)a;
2223 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
2224 uint16_t mask = s->sample_mask;
2225
2226 radeon_set_context_reg_seq(cs, CM_R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0, 2);
2227 radeon_emit(cs, mask | (mask << 16)); /* X0Y0_X1Y0 */
2228 radeon_emit(cs, mask | (mask << 16)); /* X0Y1_X1Y1 */
2229 }
2230
2231 static void evergreen_emit_vertex_fetch_shader(struct r600_context *rctx, struct r600_atom *a)
2232 {
2233 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
2234 struct r600_cso_state *state = (struct r600_cso_state*)a;
2235 struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state->cso;
2236
2237 radeon_set_context_reg(cs, R_0288A4_SQ_PGM_START_FS,
2238 (shader->buffer->gpu_address + shader->offset) >> 8);
2239 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
2240 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, shader->buffer,
2241 RADEON_USAGE_READ,
2242 RADEON_PRIO_INTERNAL_SHADER));
2243 }
2244
2245 static void evergreen_emit_shader_stages(struct r600_context *rctx, struct r600_atom *a)
2246 {
2247 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
2248 struct r600_shader_stages_state *state = (struct r600_shader_stages_state*)a;
2249
2250 uint32_t v = 0, v2 = 0, primid = 0, tf_param = 0;
2251
2252 if (rctx->vs_shader->current->shader.vs_as_gs_a) {
2253 v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_A);
2254 primid = 1;
2255 }
2256
2257 if (state->geom_enable) {
2258 uint32_t cut_val;
2259
2260 if (rctx->gs_shader->gs_max_out_vertices <= 128)
2261 cut_val = V_028A40_GS_CUT_128;
2262 else if (rctx->gs_shader->gs_max_out_vertices <= 256)
2263 cut_val = V_028A40_GS_CUT_256;
2264 else if (rctx->gs_shader->gs_max_out_vertices <= 512)
2265 cut_val = V_028A40_GS_CUT_512;
2266 else
2267 cut_val = V_028A40_GS_CUT_1024;
2268
2269 v = S_028B54_GS_EN(1) |
2270 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER);
2271 if (!rctx->tes_shader)
2272 v |= S_028B54_ES_EN(V_028B54_ES_STAGE_REAL);
2273
2274 v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_G) |
2275 S_028A40_CUT_MODE(cut_val);
2276
2277 if (rctx->gs_shader->current->shader.gs_prim_id_input)
2278 primid = 1;
2279 }
2280
2281 if (rctx->tes_shader) {
2282 uint32_t type, partitioning, topology;
2283 struct tgsi_shader_info *info = &rctx->tes_shader->current->selector->info;
2284 unsigned tes_prim_mode = info->properties[TGSI_PROPERTY_TES_PRIM_MODE];
2285 unsigned tes_spacing = info->properties[TGSI_PROPERTY_TES_SPACING];
2286 bool tes_vertex_order_cw = info->properties[TGSI_PROPERTY_TES_VERTEX_ORDER_CW];
2287 bool tes_point_mode = info->properties[TGSI_PROPERTY_TES_POINT_MODE];
2288 switch (tes_prim_mode) {
2289 case PIPE_PRIM_LINES:
2290 type = V_028B6C_TESS_ISOLINE;
2291 break;
2292 case PIPE_PRIM_TRIANGLES:
2293 type = V_028B6C_TESS_TRIANGLE;
2294 break;
2295 case PIPE_PRIM_QUADS:
2296 type = V_028B6C_TESS_QUAD;
2297 break;
2298 default:
2299 assert(0);
2300 return;
2301 }
2302
2303 switch (tes_spacing) {
2304 case PIPE_TESS_SPACING_FRACTIONAL_ODD:
2305 partitioning = V_028B6C_PART_FRAC_ODD;
2306 break;
2307 case PIPE_TESS_SPACING_FRACTIONAL_EVEN:
2308 partitioning = V_028B6C_PART_FRAC_EVEN;
2309 break;
2310 case PIPE_TESS_SPACING_EQUAL:
2311 partitioning = V_028B6C_PART_INTEGER;
2312 break;
2313 default:
2314 assert(0);
2315 return;
2316 }
2317
2318 if (tes_point_mode)
2319 topology = V_028B6C_OUTPUT_POINT;
2320 else if (tes_prim_mode == PIPE_PRIM_LINES)
2321 topology = V_028B6C_OUTPUT_LINE;
2322 else if (tes_vertex_order_cw)
2323 /* XXX follow radeonsi and invert */
2324 topology = V_028B6C_OUTPUT_TRIANGLE_CCW;
2325 else
2326 topology = V_028B6C_OUTPUT_TRIANGLE_CW;
2327
2328 tf_param = S_028B6C_TYPE(type) |
2329 S_028B6C_PARTITIONING(partitioning) |
2330 S_028B6C_TOPOLOGY(topology);
2331 }
2332
2333 if (rctx->tes_shader) {
2334 v |= S_028B54_LS_EN(V_028B54_LS_STAGE_ON) |
2335 S_028B54_HS_EN(1);
2336 if (!state->geom_enable)
2337 v |= S_028B54_VS_EN(V_028B54_VS_STAGE_DS);
2338 else
2339 v |= S_028B54_ES_EN(V_028B54_ES_STAGE_DS);
2340 }
2341
2342 radeon_set_context_reg(cs, R_028AB8_VGT_VTX_CNT_EN, v ? 1 : 0 );
2343 radeon_set_context_reg(cs, R_028B54_VGT_SHADER_STAGES_EN, v);
2344 radeon_set_context_reg(cs, R_028A40_VGT_GS_MODE, v2);
2345 radeon_set_context_reg(cs, R_028A84_VGT_PRIMITIVEID_EN, primid);
2346 radeon_set_context_reg(cs, R_028B6C_VGT_TF_PARAM, tf_param);
2347 }
2348
2349 static void evergreen_emit_gs_rings(struct r600_context *rctx, struct r600_atom *a)
2350 {
2351 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
2352 struct r600_gs_rings_state *state = (struct r600_gs_rings_state*)a;
2353 struct r600_resource *rbuffer;
2354
2355 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
2356 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
2357 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
2358
2359 if (state->enable) {
2360 rbuffer =(struct r600_resource*)state->esgs_ring.buffer;
2361 radeon_set_config_reg(cs, R_008C40_SQ_ESGS_RING_BASE,
2362 rbuffer->gpu_address >> 8);
2363 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
2364 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
2365 RADEON_USAGE_READWRITE,
2366 RADEON_PRIO_RINGS_STREAMOUT));
2367 radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE,
2368 state->esgs_ring.buffer_size >> 8);
2369
2370 rbuffer =(struct r600_resource*)state->gsvs_ring.buffer;
2371 radeon_set_config_reg(cs, R_008C48_SQ_GSVS_RING_BASE,
2372 rbuffer->gpu_address >> 8);
2373 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
2374 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
2375 RADEON_USAGE_READWRITE,
2376 RADEON_PRIO_RINGS_STREAMOUT));
2377 radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE,
2378 state->gsvs_ring.buffer_size >> 8);
2379 } else {
2380 radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE, 0);
2381 radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE, 0);
2382 }
2383
2384 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
2385 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
2386 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
2387 }
2388
2389 void cayman_init_common_regs(struct r600_command_buffer *cb,
2390 enum chip_class ctx_chip_class,
2391 enum radeon_family ctx_family,
2392 int ctx_drm_minor)
2393 {
2394 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 2);
2395 r600_store_value(cb, S_008C00_EXPORT_SRC_C(1)); /* R_008C00_SQ_CONFIG */
2396 /* always set the temp clauses */
2397 r600_store_value(cb, S_008C04_NUM_CLAUSE_TEMP_GPRS(4)); /* R_008C04_SQ_GPR_RESOURCE_MGMT_1 */
2398
2399 r600_store_config_reg_seq(cb, R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1, 2);
2400 r600_store_value(cb, 0); /* R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1 */
2401 r600_store_value(cb, 0); /* R_008C14_SQ_GLOBAL_GPR_RESOURCE_MGMT_2 */
2402
2403 r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, (1 << 8));
2404
2405 r600_store_context_reg_seq(cb, R_028350_SX_MISC, 2);
2406 r600_store_value(cb, 0);
2407 r600_store_value(cb, S_028354_SURFACE_SYNC_MASK(0xf));
2408
2409 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2410 }
2411
2412 static void cayman_init_atom_start_cs(struct r600_context *rctx)
2413 {
2414 struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2415 int tmp, i;
2416
2417 r600_init_command_buffer(cb, 338);
2418
2419 /* This must be first. */
2420 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2421 r600_store_value(cb, 0x80000000);
2422 r600_store_value(cb, 0x80000000);
2423
2424 /* We're setting config registers here. */
2425 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2426 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2427
2428 cayman_init_common_regs(cb, rctx->b.chip_class,
2429 rctx->b.family, rctx->screen->b.info.drm_minor);
2430
2431 r600_store_config_reg(cb, R_009100_SPI_CONFIG_CNTL, 0);
2432 r600_store_config_reg(cb, R_00913C_SPI_CONFIG_CNTL_1, S_00913C_VTX_DONE_DELAY(4));
2433
2434 /* remove LS/HS from one SIMD for hw workaround */
2435 r600_store_config_reg_seq(cb, R_008E20_SQ_STATIC_THREAD_MGMT1, 3);
2436 r600_store_value(cb, 0xffffffff);
2437 r600_store_value(cb, 0xffffffff);
2438 r600_store_value(cb, 0xfffffffe);
2439
2440 r600_store_context_reg_seq(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 6);
2441 r600_store_value(cb, 0); /* R_028900_SQ_ESGS_RING_ITEMSIZE */
2442 r600_store_value(cb, 0); /* R_028904_SQ_GSVS_RING_ITEMSIZE */
2443 r600_store_value(cb, 0); /* R_028908_SQ_ESTMP_RING_ITEMSIZE */
2444 r600_store_value(cb, 0); /* R_02890C_SQ_GSTMP_RING_ITEMSIZE */
2445 r600_store_value(cb, 0); /* R_028910_SQ_VSTMP_RING_ITEMSIZE */
2446 r600_store_value(cb, 0); /* R_028914_SQ_PSTMP_RING_ITEMSIZE */
2447
2448 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4);
2449 r600_store_value(cb, 0); /* R_02891C_SQ_GS_VERT_ITEMSIZE */
2450 r600_store_value(cb, 0); /* R_028920_SQ_GS_VERT_ITEMSIZE_1 */
2451 r600_store_value(cb, 0); /* R_028924_SQ_GS_VERT_ITEMSIZE_2 */
2452 r600_store_value(cb, 0); /* R_028928_SQ_GS_VERT_ITEMSIZE_3 */
2453
2454 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
2455 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
2456 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
2457 r600_store_value(cb, fui(64)); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
2458 r600_store_value(cb, fui(0)); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
2459 r600_store_value(cb, 16); /* R_028A20_VGT_HOS_REUSE_DEPTH */
2460 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
2461 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
2462 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
2463 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
2464 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
2465 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
2466 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
2467 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE */
2468
2469 r600_store_context_reg(cb, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0);
2470
2471 r600_store_config_reg(cb, R_008A14_PA_CL_ENHANCE, (3 << 1) | 1);
2472
2473 r600_store_context_reg_seq(cb, CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0, 2);
2474 r600_store_value(cb, 0x76543210); /* CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0 */
2475 r600_store_value(cb, 0xfedcba98); /* CM_R_028BD8_PA_SC_CENTROID_PRIORITY_1 */
2476
2477 r600_store_context_reg_seq(cb, R_0288E8_SQ_LDS_ALLOC, 2);
2478 r600_store_value(cb, 0); /* R_0288E8_SQ_LDS_ALLOC */
2479 r600_store_value(cb, 0); /* R_0288EC_SQ_LDS_ALLOC_PS */
2480
2481 r600_store_context_reg(cb, R_0288F0_SQ_VTX_SEMANTIC_CLEAR, ~0);
2482
2483 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
2484 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
2485 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
2486
2487 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
2488
2489 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
2490
2491 r600_store_context_reg(cb, R_0286DC_SPI_FOG_CNTL, 0);
2492
2493 r600_store_context_reg_seq(cb, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 3);
2494 r600_store_value(cb, 0); /* R_028AC0_DB_SRESULTS_COMPARE_STATE0 */
2495 r600_store_value(cb, 0); /* R_028AC4_DB_SRESULTS_COMPARE_STATE1 */
2496 r600_store_value(cb, 0); /* R_028AC8_DB_PRELOAD_CONTROL */
2497
2498 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
2499 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
2500
2501 r600_store_context_reg_seq(cb, R_0282D0_PA_SC_VPORT_ZMIN_0, 2 * R600_MAX_VIEWPORTS);
2502 for (tmp = 0; tmp < R600_MAX_VIEWPORTS; tmp++) {
2503 r600_store_value(cb, 0); /* R_0282D0_PA_SC_VPORT_ZMIN_0 */
2504 r600_store_value(cb, fui(1.0)); /* R_0282D4_PA_SC_VPORT_ZMAX_0 */
2505 }
2506
2507 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
2508 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
2509
2510 r600_store_context_reg_seq(cb, CM_R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, 4);
2511 r600_store_value(cb, fui(1.0)); /* CM_R_028BE8_PA_CL_GB_VERT_CLIP_ADJ */
2512 r600_store_value(cb, fui(1.0)); /* CM_R_028BEC_PA_CL_GB_VERT_DISC_ADJ */
2513 r600_store_value(cb, fui(1.0)); /* CM_R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ */
2514 r600_store_value(cb, fui(1.0)); /* CM_R_028BF4_PA_CL_GB_HORZ_DISC_ADJ */
2515
2516 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
2517 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
2518 r600_store_value(cb, S_028244_BR_X(16384) | S_028244_BR_Y(16384)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
2519
2520 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
2521 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
2522 r600_store_value(cb, S_028034_BR_X(16384) | S_028034_BR_Y(16384)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
2523
2524 r600_store_context_reg(cb, R_028848_SQ_PGM_RESOURCES_2_PS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2525 r600_store_context_reg(cb, R_028864_SQ_PGM_RESOURCES_2_VS, S_028864_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2526 r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_2_GS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2527 r600_store_context_reg(cb, R_028894_SQ_PGM_RESOURCES_2_ES, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2528 r600_store_context_reg(cb, R_0288C0_SQ_PGM_RESOURCES_2_HS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2529 r600_store_context_reg(cb, R_0288D8_SQ_PGM_RESOURCES_2_LS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2530
2531 r600_store_context_reg(cb, R_0288A8_SQ_PGM_RESOURCES_FS, 0);
2532
2533 /* to avoid GPU doing any preloading of constant from random address */
2534 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
2535 for (i = 0; i < 16; i++)
2536 r600_store_value(cb, 0);
2537
2538 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
2539 for (i = 0; i < 16; i++)
2540 r600_store_value(cb, 0);
2541
2542 r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16);
2543 for (i = 0; i < 16; i++)
2544 r600_store_value(cb, 0);
2545
2546 r600_store_context_reg_seq(cb, R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0, 16);
2547 for (i = 0; i < 16; i++)
2548 r600_store_value(cb, 0);
2549
2550 r600_store_context_reg_seq(cb, R_028F80_ALU_CONST_BUFFER_SIZE_HS_0, 16);
2551 for (i = 0; i < 16; i++)
2552 r600_store_value(cb, 0);
2553
2554 if (rctx->screen->b.has_streamout) {
2555 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
2556 }
2557
2558 r600_store_context_reg(cb, R_028010_DB_RENDER_OVERRIDE2, 0);
2559 r600_store_context_reg(cb, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0);
2560 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
2561 r600_store_context_reg_seq(cb, R_0286E4_SPI_PS_IN_CONTROL_2, 2);
2562 r600_store_value(cb, 0); /* R_0286E4_SPI_PS_IN_CONTROL_2 */
2563 r600_store_value(cb, 0); /* R_0286E8_SPI_COMPUTE_INPUT_CNTL */
2564
2565 r600_store_context_reg_seq(cb, R_028B54_VGT_SHADER_STAGES_EN, 2);
2566 r600_store_value(cb, 0); /* R028B54_VGT_SHADER_STAGES_EN */
2567 r600_store_value(cb, 0); /* R028B58_VGT_LS_HS_CONFIG */
2568 r600_store_context_reg(cb, R_028B6C_VGT_TF_PARAM, 0);
2569 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0, 0x01000FFF);
2570 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF);
2571 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (64 * 4), 0x01000FFF);
2572 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (96 * 4), 0x01000FFF);
2573 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (128 * 4), 0x01000FFF);
2574 }
2575
2576 void evergreen_init_common_regs(struct r600_context *rctx, struct r600_command_buffer *cb,
2577 enum chip_class ctx_chip_class,
2578 enum radeon_family ctx_family,
2579 int ctx_drm_minor)
2580 {
2581 int ps_prio;
2582 int vs_prio;
2583 int gs_prio;
2584 int es_prio;
2585
2586 int hs_prio;
2587 int cs_prio;
2588 int ls_prio;
2589
2590 unsigned tmp;
2591
2592 ps_prio = 0;
2593 vs_prio = 1;
2594 gs_prio = 2;
2595 es_prio = 3;
2596 hs_prio = 3;
2597 ls_prio = 3;
2598 cs_prio = 0;
2599
2600 rctx->default_gprs[R600_HW_STAGE_PS] = 93;
2601 rctx->default_gprs[R600_HW_STAGE_VS] = 46;
2602 rctx->r6xx_num_clause_temp_gprs = 4;
2603 rctx->default_gprs[R600_HW_STAGE_GS] = 31;
2604 rctx->default_gprs[R600_HW_STAGE_ES] = 31;
2605 rctx->default_gprs[EG_HW_STAGE_HS] = 23;
2606 rctx->default_gprs[EG_HW_STAGE_LS] = 23;
2607
2608 tmp = 0;
2609 switch (ctx_family) {
2610 case CHIP_CEDAR:
2611 case CHIP_PALM:
2612 case CHIP_SUMO:
2613 case CHIP_SUMO2:
2614 case CHIP_CAICOS:
2615 break;
2616 default:
2617 tmp |= S_008C00_VC_ENABLE(1);
2618 break;
2619 }
2620 tmp |= S_008C00_EXPORT_SRC_C(1);
2621 tmp |= S_008C00_CS_PRIO(cs_prio);
2622 tmp |= S_008C00_LS_PRIO(ls_prio);
2623 tmp |= S_008C00_HS_PRIO(hs_prio);
2624 tmp |= S_008C00_PS_PRIO(ps_prio);
2625 tmp |= S_008C00_VS_PRIO(vs_prio);
2626 tmp |= S_008C00_GS_PRIO(gs_prio);
2627 tmp |= S_008C00_ES_PRIO(es_prio);
2628
2629 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 1);
2630 r600_store_value(cb, tmp); /* R_008C00_SQ_CONFIG */
2631
2632 r600_store_config_reg_seq(cb, R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1, 2);
2633 r600_store_value(cb, 0); /* R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1 */
2634 r600_store_value(cb, 0); /* R_008C14_SQ_GLOBAL_GPR_RESOURCE_MGMT_2 */
2635
2636 /* The cs checker requires this register to be set. */
2637 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2638
2639 r600_store_context_reg_seq(cb, R_028350_SX_MISC, 2);
2640 r600_store_value(cb, 0);
2641 r600_store_value(cb, S_028354_SURFACE_SYNC_MASK(0xf));
2642
2643 return;
2644 }
2645
2646 void evergreen_init_atom_start_cs(struct r600_context *rctx)
2647 {
2648 struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2649 int num_ps_threads;
2650 int num_vs_threads;
2651 int num_gs_threads;
2652 int num_es_threads;
2653 int num_hs_threads;
2654 int num_ls_threads;
2655
2656 int num_ps_stack_entries;
2657 int num_vs_stack_entries;
2658 int num_gs_stack_entries;
2659 int num_es_stack_entries;
2660 int num_hs_stack_entries;
2661 int num_ls_stack_entries;
2662 enum radeon_family family;
2663 unsigned tmp, i;
2664
2665 if (rctx->b.chip_class == CAYMAN) {
2666 cayman_init_atom_start_cs(rctx);
2667 return;
2668 }
2669
2670 r600_init_command_buffer(cb, 338);
2671
2672 /* This must be first. */
2673 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2674 r600_store_value(cb, 0x80000000);
2675 r600_store_value(cb, 0x80000000);
2676
2677 /* We're setting config registers here. */
2678 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2679 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2680
2681 evergreen_init_common_regs(rctx, cb, rctx->b.chip_class,
2682 rctx->b.family, rctx->screen->b.info.drm_minor);
2683
2684 family = rctx->b.family;
2685 switch (family) {
2686 case CHIP_CEDAR:
2687 default:
2688 num_ps_threads = 96;
2689 num_vs_threads = 16;
2690 num_gs_threads = 16;
2691 num_es_threads = 16;
2692 num_hs_threads = 16;
2693 num_ls_threads = 16;
2694 num_ps_stack_entries = 42;
2695 num_vs_stack_entries = 42;
2696 num_gs_stack_entries = 42;
2697 num_es_stack_entries = 42;
2698 num_hs_stack_entries = 42;
2699 num_ls_stack_entries = 42;
2700 break;
2701 case CHIP_REDWOOD:
2702 num_ps_threads = 128;
2703 num_vs_threads = 20;
2704 num_gs_threads = 20;
2705 num_es_threads = 20;
2706 num_hs_threads = 20;
2707 num_ls_threads = 20;
2708 num_ps_stack_entries = 42;
2709 num_vs_stack_entries = 42;
2710 num_gs_stack_entries = 42;
2711 num_es_stack_entries = 42;
2712 num_hs_stack_entries = 42;
2713 num_ls_stack_entries = 42;
2714 break;
2715 case CHIP_JUNIPER:
2716 num_ps_threads = 128;
2717 num_vs_threads = 20;
2718 num_gs_threads = 20;
2719 num_es_threads = 20;
2720 num_hs_threads = 20;
2721 num_ls_threads = 20;
2722 num_ps_stack_entries = 85;
2723 num_vs_stack_entries = 85;
2724 num_gs_stack_entries = 85;
2725 num_es_stack_entries = 85;
2726 num_hs_stack_entries = 85;
2727 num_ls_stack_entries = 85;
2728 break;
2729 case CHIP_CYPRESS:
2730 case CHIP_HEMLOCK:
2731 num_ps_threads = 128;
2732 num_vs_threads = 20;
2733 num_gs_threads = 20;
2734 num_es_threads = 20;
2735 num_hs_threads = 20;
2736 num_ls_threads = 20;
2737 num_ps_stack_entries = 85;
2738 num_vs_stack_entries = 85;
2739 num_gs_stack_entries = 85;
2740 num_es_stack_entries = 85;
2741 num_hs_stack_entries = 85;
2742 num_ls_stack_entries = 85;
2743 break;
2744 case CHIP_PALM:
2745 num_ps_threads = 96;
2746 num_vs_threads = 16;
2747 num_gs_threads = 16;
2748 num_es_threads = 16;
2749 num_hs_threads = 16;
2750 num_ls_threads = 16;
2751 num_ps_stack_entries = 42;
2752 num_vs_stack_entries = 42;
2753 num_gs_stack_entries = 42;
2754 num_es_stack_entries = 42;
2755 num_hs_stack_entries = 42;
2756 num_ls_stack_entries = 42;
2757 break;
2758 case CHIP_SUMO:
2759 num_ps_threads = 96;
2760 num_vs_threads = 25;
2761 num_gs_threads = 25;
2762 num_es_threads = 25;
2763 num_hs_threads = 16;
2764 num_ls_threads = 16;
2765 num_ps_stack_entries = 42;
2766 num_vs_stack_entries = 42;
2767 num_gs_stack_entries = 42;
2768 num_es_stack_entries = 42;
2769 num_hs_stack_entries = 42;
2770 num_ls_stack_entries = 42;
2771 break;
2772 case CHIP_SUMO2:
2773 num_ps_threads = 96;
2774 num_vs_threads = 25;
2775 num_gs_threads = 25;
2776 num_es_threads = 25;
2777 num_hs_threads = 16;
2778 num_ls_threads = 16;
2779 num_ps_stack_entries = 85;
2780 num_vs_stack_entries = 85;
2781 num_gs_stack_entries = 85;
2782 num_es_stack_entries = 85;
2783 num_hs_stack_entries = 85;
2784 num_ls_stack_entries = 85;
2785 break;
2786 case CHIP_BARTS:
2787 num_ps_threads = 128;
2788 num_vs_threads = 20;
2789 num_gs_threads = 20;
2790 num_es_threads = 20;
2791 num_hs_threads = 20;
2792 num_ls_threads = 20;
2793 num_ps_stack_entries = 85;
2794 num_vs_stack_entries = 85;
2795 num_gs_stack_entries = 85;
2796 num_es_stack_entries = 85;
2797 num_hs_stack_entries = 85;
2798 num_ls_stack_entries = 85;
2799 break;
2800 case CHIP_TURKS:
2801 num_ps_threads = 128;
2802 num_vs_threads = 20;
2803 num_gs_threads = 20;
2804 num_es_threads = 20;
2805 num_hs_threads = 20;
2806 num_ls_threads = 20;
2807 num_ps_stack_entries = 42;
2808 num_vs_stack_entries = 42;
2809 num_gs_stack_entries = 42;
2810 num_es_stack_entries = 42;
2811 num_hs_stack_entries = 42;
2812 num_ls_stack_entries = 42;
2813 break;
2814 case CHIP_CAICOS:
2815 num_ps_threads = 96;
2816 num_vs_threads = 10;
2817 num_gs_threads = 10;
2818 num_es_threads = 10;
2819 num_hs_threads = 10;
2820 num_ls_threads = 10;
2821 num_ps_stack_entries = 42;
2822 num_vs_stack_entries = 42;
2823 num_gs_stack_entries = 42;
2824 num_es_stack_entries = 42;
2825 num_hs_stack_entries = 42;
2826 num_ls_stack_entries = 42;
2827 break;
2828 }
2829
2830 tmp = S_008C18_NUM_PS_THREADS(num_ps_threads);
2831 tmp |= S_008C18_NUM_VS_THREADS(num_vs_threads);
2832 tmp |= S_008C18_NUM_GS_THREADS(num_gs_threads);
2833 tmp |= S_008C18_NUM_ES_THREADS(num_es_threads);
2834
2835 r600_store_config_reg_seq(cb, R_008C18_SQ_THREAD_RESOURCE_MGMT_1, 5);
2836 r600_store_value(cb, tmp); /* R_008C18_SQ_THREAD_RESOURCE_MGMT_1 */
2837
2838 tmp = S_008C1C_NUM_HS_THREADS(num_hs_threads);
2839 tmp |= S_008C1C_NUM_LS_THREADS(num_ls_threads);
2840 r600_store_value(cb, tmp); /* R_008C1C_SQ_THREAD_RESOURCE_MGMT_2 */
2841
2842 tmp = S_008C20_NUM_PS_STACK_ENTRIES(num_ps_stack_entries);
2843 tmp |= S_008C20_NUM_VS_STACK_ENTRIES(num_vs_stack_entries);
2844 r600_store_value(cb, tmp); /* R_008C20_SQ_STACK_RESOURCE_MGMT_1 */
2845
2846 tmp = S_008C24_NUM_GS_STACK_ENTRIES(num_gs_stack_entries);
2847 tmp |= S_008C24_NUM_ES_STACK_ENTRIES(num_es_stack_entries);
2848 r600_store_value(cb, tmp); /* R_008C24_SQ_STACK_RESOURCE_MGMT_2 */
2849
2850 tmp = S_008C28_NUM_HS_STACK_ENTRIES(num_hs_stack_entries);
2851 tmp |= S_008C28_NUM_LS_STACK_ENTRIES(num_ls_stack_entries);
2852 r600_store_value(cb, tmp); /* R_008C28_SQ_STACK_RESOURCE_MGMT_3 */
2853
2854 r600_store_config_reg(cb, R_008E2C_SQ_LDS_RESOURCE_MGMT,
2855 S_008E2C_NUM_PS_LDS(0x1000) | S_008E2C_NUM_LS_LDS(0x1000));
2856
2857 /* remove LS/HS from one SIMD for hw workaround */
2858 r600_store_config_reg_seq(cb, R_008E20_SQ_STATIC_THREAD_MGMT1, 3);
2859 r600_store_value(cb, 0xffffffff);
2860 r600_store_value(cb, 0xffffffff);
2861 r600_store_value(cb, 0xfffffffe);
2862
2863 r600_store_config_reg(cb, R_009100_SPI_CONFIG_CNTL, 0);
2864 r600_store_config_reg(cb, R_00913C_SPI_CONFIG_CNTL_1, S_00913C_VTX_DONE_DELAY(4));
2865
2866 r600_store_context_reg_seq(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 6);
2867 r600_store_value(cb, 0); /* R_028900_SQ_ESGS_RING_ITEMSIZE */
2868 r600_store_value(cb, 0); /* R_028904_SQ_GSVS_RING_ITEMSIZE */
2869 r600_store_value(cb, 0); /* R_028908_SQ_ESTMP_RING_ITEMSIZE */
2870 r600_store_value(cb, 0); /* R_02890C_SQ_GSTMP_RING_ITEMSIZE */
2871 r600_store_value(cb, 0); /* R_028910_SQ_VSTMP_RING_ITEMSIZE */
2872 r600_store_value(cb, 0); /* R_028914_SQ_PSTMP_RING_ITEMSIZE */
2873
2874 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4);
2875 r600_store_value(cb, 0); /* R_02891C_SQ_GS_VERT_ITEMSIZE */
2876 r600_store_value(cb, 0); /* R_028920_SQ_GS_VERT_ITEMSIZE_1 */
2877 r600_store_value(cb, 0); /* R_028924_SQ_GS_VERT_ITEMSIZE_2 */
2878 r600_store_value(cb, 0); /* R_028928_SQ_GS_VERT_ITEMSIZE_3 */
2879
2880 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
2881 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
2882 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
2883 r600_store_value(cb, fui(64)); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
2884 r600_store_value(cb, fui(1.0)); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
2885 r600_store_value(cb, 16); /* R_028A20_VGT_HOS_REUSE_DEPTH */
2886 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
2887 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
2888 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
2889 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
2890 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
2891 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
2892 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
2893 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE */
2894
2895 r600_store_config_reg(cb, R_008A14_PA_CL_ENHANCE, (3 << 1) | 1);
2896
2897 r600_store_context_reg(cb, R_0288F0_SQ_VTX_SEMANTIC_CLEAR, ~0);
2898
2899 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
2900 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
2901 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
2902
2903 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
2904
2905 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
2906
2907 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
2908 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
2909 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
2910
2911 r600_store_context_reg_seq(cb, R_0282D0_PA_SC_VPORT_ZMIN_0, 2 * R600_MAX_VIEWPORTS);
2912 for (tmp = 0; tmp < R600_MAX_VIEWPORTS; tmp++) {
2913 r600_store_value(cb, 0); /* R_0282D0_PA_SC_VPORT_ZMIN_0 */
2914 r600_store_value(cb, fui(1.0)); /* R_0282D4_PA_SC_VPORT_ZMAX_0 */
2915 }
2916
2917 r600_store_context_reg(cb, R_0286DC_SPI_FOG_CNTL, 0);
2918 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
2919
2920 r600_store_context_reg_seq(cb, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 3);
2921 r600_store_value(cb, 0); /* R_028AC0_DB_SRESULTS_COMPARE_STATE0 */
2922 r600_store_value(cb, 0); /* R_028AC4_DB_SRESULTS_COMPARE_STATE1 */
2923 r600_store_value(cb, 0); /* R_028AC8_DB_PRELOAD_CONTROL */
2924
2925 r600_store_context_reg_seq(cb, R_028C0C_PA_CL_GB_VERT_CLIP_ADJ, 4);
2926 r600_store_value(cb, fui(1.0)); /* R_028C0C_PA_CL_GB_VERT_CLIP_ADJ */
2927 r600_store_value(cb, fui(1.0)); /* R_028C10_PA_CL_GB_VERT_DISC_ADJ */
2928 r600_store_value(cb, fui(1.0)); /* R_028C14_PA_CL_GB_HORZ_CLIP_ADJ */
2929 r600_store_value(cb, fui(1.0)); /* R_028C18_PA_CL_GB_HORZ_DISC_ADJ */
2930
2931 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
2932 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
2933 r600_store_value(cb, S_028244_BR_X(16384) | S_028244_BR_Y(16384)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
2934
2935 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
2936 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
2937 r600_store_value(cb, S_028034_BR_X(16384) | S_028034_BR_Y(16384)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
2938
2939 r600_store_context_reg(cb, R_028848_SQ_PGM_RESOURCES_2_PS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2940 r600_store_context_reg(cb, R_028864_SQ_PGM_RESOURCES_2_VS, S_028864_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2941 r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_2_GS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2942 r600_store_context_reg(cb, R_028894_SQ_PGM_RESOURCES_2_ES, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2943 r600_store_context_reg(cb, R_0288A8_SQ_PGM_RESOURCES_FS, 0);
2944 r600_store_context_reg(cb, R_0288C0_SQ_PGM_RESOURCES_2_HS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2945 r600_store_context_reg(cb, R_0288D8_SQ_PGM_RESOURCES_2_LS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN));
2946
2947 /* to avoid GPU doing any preloading of constant from random address */
2948 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
2949 for (i = 0; i < 16; i++)
2950 r600_store_value(cb, 0);
2951
2952 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
2953 for (i = 0; i < 16; i++)
2954 r600_store_value(cb, 0);
2955
2956 r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16);
2957 for (i = 0; i < 16; i++)
2958 r600_store_value(cb, 0);
2959
2960 r600_store_context_reg_seq(cb, R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0, 16);
2961 for (i = 0; i < 16; i++)
2962 r600_store_value(cb, 0);
2963
2964 r600_store_context_reg_seq(cb, R_028F80_ALU_CONST_BUFFER_SIZE_HS_0, 16);
2965 for (i = 0; i < 16; i++)
2966 r600_store_value(cb, 0);
2967
2968 r600_store_context_reg(cb, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0);
2969
2970 if (rctx->screen->b.has_streamout) {
2971 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
2972 }
2973
2974 r600_store_context_reg(cb, R_028010_DB_RENDER_OVERRIDE2, 0);
2975 r600_store_context_reg(cb, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0);
2976 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
2977 r600_store_context_reg_seq(cb, R_0286E4_SPI_PS_IN_CONTROL_2, 2);
2978 r600_store_value(cb, 0); /* R_0286E4_SPI_PS_IN_CONTROL_2 */
2979 r600_store_value(cb, 0); /* R_0286E8_SPI_COMPUTE_INPUT_CNTL */
2980
2981 r600_store_context_reg_seq(cb, R_0288E8_SQ_LDS_ALLOC, 2);
2982 r600_store_value(cb, 0); /* R_0288E8_SQ_LDS_ALLOC */
2983 r600_store_value(cb, 0); /* R_0288EC_SQ_LDS_ALLOC_PS */
2984
2985 if (rctx->b.family == CHIP_CAICOS) {
2986 r600_store_context_reg_seq(cb, R_028B54_VGT_SHADER_STAGES_EN, 2);
2987 r600_store_value(cb, 0); /* R028B54_VGT_SHADER_STAGES_EN */
2988 r600_store_value(cb, 0); /* R028B58_VGT_LS_HS_CONFIG */
2989 r600_store_context_reg(cb, R_028B6C_VGT_TF_PARAM, 0);
2990 } else {
2991 r600_store_context_reg_seq(cb, R_028B54_VGT_SHADER_STAGES_EN, 7);
2992 r600_store_value(cb, 0); /* R028B54_VGT_SHADER_STAGES_EN */
2993 r600_store_value(cb, 0); /* R028B58_VGT_LS_HS_CONFIG */
2994 r600_store_value(cb, 0); /* R028B5C_VGT_LS_SIZE */
2995 r600_store_value(cb, 0); /* R028B60_VGT_HS_SIZE */
2996 r600_store_value(cb, 0); /* R028B64_VGT_LS_HS_ALLOC */
2997 r600_store_value(cb, 0); /* R028B68_VGT_HS_PATCH_CONST */
2998 r600_store_value(cb, 0); /* R028B68_VGT_TF_PARAM */
2999 }
3000
3001 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0, 0x01000FFF);
3002 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF);
3003 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (64 * 4), 0x01000FFF);
3004 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (96 * 4), 0x01000FFF);
3005 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (128 * 4), 0x01000FFF);
3006 }
3007
3008 void evergreen_update_ps_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3009 {
3010 struct r600_context *rctx = (struct r600_context *)ctx;
3011 struct r600_command_buffer *cb = &shader->command_buffer;
3012 struct r600_shader *rshader = &shader->shader;
3013 unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1, db_shader_control = 0;
3014 int pos_index = -1, face_index = -1, fixed_pt_position_index = -1;
3015 int ninterp = 0;
3016 boolean have_perspective = FALSE, have_linear = FALSE;
3017 static const unsigned spi_baryc_enable_bit[6] = {
3018 S_0286E0_PERSP_SAMPLE_ENA(1),
3019 S_0286E0_PERSP_CENTER_ENA(1),
3020 S_0286E0_PERSP_CENTROID_ENA(1),
3021 S_0286E0_LINEAR_SAMPLE_ENA(1),
3022 S_0286E0_LINEAR_CENTER_ENA(1),
3023 S_0286E0_LINEAR_CENTROID_ENA(1)
3024 };
3025 unsigned spi_baryc_cntl = 0, sid, tmp, num = 0;
3026 unsigned z_export = 0, stencil_export = 0, mask_export = 0;
3027 unsigned sprite_coord_enable = rctx->rasterizer ? rctx->rasterizer->sprite_coord_enable : 0;
3028 uint32_t spi_ps_input_cntl[32];
3029
3030 if (!cb->buf) {
3031 r600_init_command_buffer(cb, 64);
3032 } else {
3033 cb->num_dw = 0;
3034 }
3035
3036 for (i = 0; i < rshader->ninput; i++) {
3037 /* evergreen NUM_INTERP only contains values interpolated into the LDS,
3038 POSITION goes via GPRs from the SC so isn't counted */
3039 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
3040 pos_index = i;
3041 else if (rshader->input[i].name == TGSI_SEMANTIC_FACE) {
3042 if (face_index == -1)
3043 face_index = i;
3044 }
3045 else if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEMASK) {
3046 if (face_index == -1)
3047 face_index = i; /* lives in same register, same enable bit */
3048 }
3049 else if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEID) {
3050 fixed_pt_position_index = i;
3051 }
3052 else {
3053 ninterp++;
3054 int k = eg_get_interpolator_index(
3055 rshader->input[i].interpolate,
3056 rshader->input[i].interpolate_location);
3057 if (k >= 0) {
3058 spi_baryc_cntl |= spi_baryc_enable_bit[k];
3059 have_perspective |= k < 3;
3060 have_linear |= !(k < 3);
3061 }
3062 }
3063
3064 sid = rshader->input[i].spi_sid;
3065
3066 if (sid) {
3067 tmp = S_028644_SEMANTIC(sid);
3068
3069 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION ||
3070 rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
3071 (rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR &&
3072 rctx->rasterizer && rctx->rasterizer->flatshade)) {
3073 tmp |= S_028644_FLAT_SHADE(1);
3074 }
3075
3076 if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
3077 (sprite_coord_enable & (1 << rshader->input[i].sid))) {
3078 tmp |= S_028644_PT_SPRITE_TEX(1);
3079 }
3080
3081 spi_ps_input_cntl[num++] = tmp;
3082 }
3083 }
3084
3085 r600_store_context_reg_seq(cb, R_028644_SPI_PS_INPUT_CNTL_0, num);
3086 r600_store_array(cb, num, spi_ps_input_cntl);
3087
3088 for (i = 0; i < rshader->noutput; i++) {
3089 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
3090 z_export = 1;
3091 if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
3092 stencil_export = 1;
3093 if (rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK &&
3094 rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0)
3095 mask_export = 1;
3096 }
3097 if (rshader->uses_kill)
3098 db_shader_control |= S_02880C_KILL_ENABLE(1);
3099
3100 db_shader_control |= S_02880C_Z_EXPORT_ENABLE(z_export);
3101 db_shader_control |= S_02880C_STENCIL_EXPORT_ENABLE(stencil_export);
3102 db_shader_control |= S_02880C_MASK_EXPORT_ENABLE(mask_export);
3103
3104 switch (rshader->ps_conservative_z) {
3105 default: /* fall through */
3106 case TGSI_FS_DEPTH_LAYOUT_ANY:
3107 db_shader_control |= S_02880C_CONSERVATIVE_Z_EXPORT(V_02880C_EXPORT_ANY_Z);
3108 break;
3109 case TGSI_FS_DEPTH_LAYOUT_GREATER:
3110 db_shader_control |= S_02880C_CONSERVATIVE_Z_EXPORT(V_02880C_EXPORT_GREATER_THAN_Z);
3111 break;
3112 case TGSI_FS_DEPTH_LAYOUT_LESS:
3113 db_shader_control |= S_02880C_CONSERVATIVE_Z_EXPORT(V_02880C_EXPORT_LESS_THAN_Z);
3114 break;
3115 }
3116
3117 exports_ps = 0;
3118 for (i = 0; i < rshader->noutput; i++) {
3119 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
3120 rshader->output[i].name == TGSI_SEMANTIC_STENCIL ||
3121 rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK)
3122 exports_ps |= 1;
3123 }
3124
3125 num_cout = rshader->nr_ps_color_exports;
3126
3127 exports_ps |= S_02884C_EXPORT_COLORS(num_cout);
3128 if (!exports_ps) {
3129 /* always at least export 1 component per pixel */
3130 exports_ps = 2;
3131 }
3132 shader->nr_ps_color_outputs = num_cout;
3133 if (ninterp == 0) {
3134 ninterp = 1;
3135 have_perspective = TRUE;
3136 }
3137 if (!spi_baryc_cntl)
3138 spi_baryc_cntl |= spi_baryc_enable_bit[0];
3139
3140 if (!have_perspective && !have_linear)
3141 have_perspective = TRUE;
3142
3143 spi_ps_in_control_0 = S_0286CC_NUM_INTERP(ninterp) |
3144 S_0286CC_PERSP_GRADIENT_ENA(have_perspective) |
3145 S_0286CC_LINEAR_GRADIENT_ENA(have_linear);
3146 spi_input_z = 0;
3147 if (pos_index != -1) {
3148 spi_ps_in_control_0 |= S_0286CC_POSITION_ENA(1) |
3149 S_0286CC_POSITION_CENTROID(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID) |
3150 S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr);
3151 spi_input_z |= S_0286D8_PROVIDE_Z_TO_SPI(1);
3152 }
3153
3154 spi_ps_in_control_1 = 0;
3155 if (face_index != -1) {
3156 spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) |
3157 S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr);
3158 }
3159 if (fixed_pt_position_index != -1) {
3160 spi_ps_in_control_1 |= S_0286D0_FIXED_PT_POSITION_ENA(1) |
3161 S_0286D0_FIXED_PT_POSITION_ADDR(rshader->input[fixed_pt_position_index].gpr);
3162 }
3163
3164 r600_store_context_reg_seq(cb, R_0286CC_SPI_PS_IN_CONTROL_0, 2);
3165 r600_store_value(cb, spi_ps_in_control_0); /* R_0286CC_SPI_PS_IN_CONTROL_0 */
3166 r600_store_value(cb, spi_ps_in_control_1); /* R_0286D0_SPI_PS_IN_CONTROL_1 */
3167
3168 r600_store_context_reg(cb, R_0286E0_SPI_BARYC_CNTL, spi_baryc_cntl);
3169 r600_store_context_reg(cb, R_0286D8_SPI_INPUT_Z, spi_input_z);
3170 r600_store_context_reg(cb, R_02884C_SQ_PGM_EXPORTS_PS, exports_ps);
3171
3172 r600_store_context_reg_seq(cb, R_028840_SQ_PGM_START_PS, 2);
3173 r600_store_value(cb, shader->bo->gpu_address >> 8);
3174 r600_store_value(cb, /* R_028844_SQ_PGM_RESOURCES_PS */
3175 S_028844_NUM_GPRS(rshader->bc.ngpr) |
3176 S_028844_PRIME_CACHE_ON_DRAW(1) |
3177 S_028844_STACK_SIZE(rshader->bc.nstack));
3178 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
3179
3180 shader->db_shader_control = db_shader_control;
3181 shader->ps_depth_export = z_export | stencil_export | mask_export;
3182
3183 shader->sprite_coord_enable = sprite_coord_enable;
3184 if (rctx->rasterizer)
3185 shader->flatshade = rctx->rasterizer->flatshade;
3186 }
3187
3188 void evergreen_update_es_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3189 {
3190 struct r600_command_buffer *cb = &shader->command_buffer;
3191 struct r600_shader *rshader = &shader->shader;
3192
3193 r600_init_command_buffer(cb, 32);
3194
3195 r600_store_context_reg(cb, R_028890_SQ_PGM_RESOURCES_ES,
3196 S_028890_NUM_GPRS(rshader->bc.ngpr) |
3197 S_028890_STACK_SIZE(rshader->bc.nstack));
3198 r600_store_context_reg(cb, R_02888C_SQ_PGM_START_ES,
3199 shader->bo->gpu_address >> 8);
3200 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
3201 }
3202
3203 void evergreen_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3204 {
3205 struct r600_context *rctx = (struct r600_context *)ctx;
3206 struct r600_command_buffer *cb = &shader->command_buffer;
3207 struct r600_shader *rshader = &shader->shader;
3208 struct r600_shader *cp_shader = &shader->gs_copy_shader->shader;
3209 unsigned gsvs_itemsizes[4] = {
3210 (cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2,
3211 (cp_shader->ring_item_sizes[1] * shader->selector->gs_max_out_vertices) >> 2,
3212 (cp_shader->ring_item_sizes[2] * shader->selector->gs_max_out_vertices) >> 2,
3213 (cp_shader->ring_item_sizes[3] * shader->selector->gs_max_out_vertices) >> 2
3214 };
3215
3216 r600_init_command_buffer(cb, 64);
3217
3218 /* VGT_GS_MODE is written by evergreen_emit_shader_stages */
3219
3220
3221 r600_store_context_reg(cb, R_028B38_VGT_GS_MAX_VERT_OUT,
3222 S_028B38_MAX_VERT_OUT(shader->selector->gs_max_out_vertices));
3223 r600_store_context_reg(cb, R_028A6C_VGT_GS_OUT_PRIM_TYPE,
3224 r600_conv_prim_to_gs_out(shader->selector->gs_output_prim));
3225
3226 if (rctx->screen->b.info.drm_minor >= 35) {
3227 r600_store_context_reg(cb, R_028B90_VGT_GS_INSTANCE_CNT,
3228 S_028B90_CNT(MIN2(shader->selector->gs_num_invocations, 127)) |
3229 S_028B90_ENABLE(shader->selector->gs_num_invocations > 0));
3230 }
3231 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4);
3232 r600_store_value(cb, cp_shader->ring_item_sizes[0] >> 2);
3233 r600_store_value(cb, cp_shader->ring_item_sizes[1] >> 2);
3234 r600_store_value(cb, cp_shader->ring_item_sizes[2] >> 2);
3235 r600_store_value(cb, cp_shader->ring_item_sizes[3] >> 2);
3236
3237 r600_store_context_reg(cb, R_028900_SQ_ESGS_RING_ITEMSIZE,
3238 (rshader->ring_item_sizes[0]) >> 2);
3239
3240 r600_store_context_reg(cb, R_028904_SQ_GSVS_RING_ITEMSIZE,
3241 gsvs_itemsizes[0] +
3242 gsvs_itemsizes[1] +
3243 gsvs_itemsizes[2] +
3244 gsvs_itemsizes[3]);
3245
3246 r600_store_context_reg_seq(cb, R_02892C_SQ_GSVS_RING_OFFSET_1, 3);
3247 r600_store_value(cb, gsvs_itemsizes[0]);
3248 r600_store_value(cb, gsvs_itemsizes[0] + gsvs_itemsizes[1]);
3249 r600_store_value(cb, gsvs_itemsizes[0] + gsvs_itemsizes[1] + gsvs_itemsizes[2]);
3250
3251 /* FIXME calculate these values somehow ??? */
3252 r600_store_context_reg_seq(cb, R_028A54_GS_PER_ES, 3);
3253 r600_store_value(cb, 0x80); /* GS_PER_ES */
3254 r600_store_value(cb, 0x100); /* ES_PER_GS */
3255 r600_store_value(cb, 0x2); /* GS_PER_VS */
3256
3257 r600_store_context_reg(cb, R_028878_SQ_PGM_RESOURCES_GS,
3258 S_028878_NUM_GPRS(rshader->bc.ngpr) |
3259 S_028878_STACK_SIZE(rshader->bc.nstack));
3260 r600_store_context_reg(cb, R_028874_SQ_PGM_START_GS,
3261 shader->bo->gpu_address >> 8);
3262 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
3263 }
3264
3265
3266 void evergreen_update_vs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3267 {
3268 struct r600_command_buffer *cb = &shader->command_buffer;
3269 struct r600_shader *rshader = &shader->shader;
3270 unsigned spi_vs_out_id[10] = {};
3271 unsigned i, tmp, nparams = 0;
3272
3273 for (i = 0; i < rshader->noutput; i++) {
3274 if (rshader->output[i].spi_sid) {
3275 tmp = rshader->output[i].spi_sid << ((nparams & 3) * 8);
3276 spi_vs_out_id[nparams / 4] |= tmp;
3277 nparams++;
3278 }
3279 }
3280
3281 r600_init_command_buffer(cb, 32);
3282
3283 r600_store_context_reg_seq(cb, R_02861C_SPI_VS_OUT_ID_0, 10);
3284 for (i = 0; i < 10; i++) {
3285 r600_store_value(cb, spi_vs_out_id[i]);
3286 }
3287
3288 /* Certain attributes (position, psize, etc.) don't count as params.
3289 * VS is required to export at least one param and r600_shader_from_tgsi()
3290 * takes care of adding a dummy export.
3291 */
3292 if (nparams < 1)
3293 nparams = 1;
3294
3295 r600_store_context_reg(cb, R_0286C4_SPI_VS_OUT_CONFIG,
3296 S_0286C4_VS_EXPORT_COUNT(nparams - 1));
3297 r600_store_context_reg(cb, R_028860_SQ_PGM_RESOURCES_VS,
3298 S_028860_NUM_GPRS(rshader->bc.ngpr) |
3299 S_028860_STACK_SIZE(rshader->bc.nstack));
3300 if (rshader->vs_position_window_space) {
3301 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
3302 S_028818_VTX_XY_FMT(1) | S_028818_VTX_Z_FMT(1));
3303 } else {
3304 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
3305 S_028818_VTX_W0_FMT(1) |
3306 S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) |
3307 S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
3308 S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1));
3309
3310 }
3311 r600_store_context_reg(cb, R_02885C_SQ_PGM_START_VS,
3312 shader->bo->gpu_address >> 8);
3313 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
3314
3315 shader->pa_cl_vs_out_cntl =
3316 S_02881C_VS_OUT_CCDIST0_VEC_ENA((rshader->clip_dist_write & 0x0F) != 0) |
3317 S_02881C_VS_OUT_CCDIST1_VEC_ENA((rshader->clip_dist_write & 0xF0) != 0) |
3318 S_02881C_VS_OUT_MISC_VEC_ENA(rshader->vs_out_misc_write) |
3319 S_02881C_USE_VTX_POINT_SIZE(rshader->vs_out_point_size) |
3320 S_02881C_USE_VTX_EDGE_FLAG(rshader->vs_out_edgeflag) |
3321 S_02881C_USE_VTX_VIEWPORT_INDX(rshader->vs_out_viewport) |
3322 S_02881C_USE_VTX_RENDER_TARGET_INDX(rshader->vs_out_layer);
3323 }
3324
3325 void evergreen_update_hs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3326 {
3327 struct r600_command_buffer *cb = &shader->command_buffer;
3328 struct r600_shader *rshader = &shader->shader;
3329
3330 r600_init_command_buffer(cb, 32);
3331 r600_store_context_reg(cb, R_0288BC_SQ_PGM_RESOURCES_HS,
3332 S_0288BC_NUM_GPRS(rshader->bc.ngpr) |
3333 S_0288BC_STACK_SIZE(rshader->bc.nstack));
3334 r600_store_context_reg(cb, R_0288B8_SQ_PGM_START_HS,
3335 shader->bo->gpu_address >> 8);
3336 }
3337
3338 void evergreen_update_ls_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
3339 {
3340 struct r600_command_buffer *cb = &shader->command_buffer;
3341 struct r600_shader *rshader = &shader->shader;
3342
3343 r600_init_command_buffer(cb, 32);
3344 r600_store_context_reg(cb, R_0288D4_SQ_PGM_RESOURCES_LS,
3345 S_0288D4_NUM_GPRS(rshader->bc.ngpr) |
3346 S_0288D4_STACK_SIZE(rshader->bc.nstack));
3347 r600_store_context_reg(cb, R_0288D0_SQ_PGM_START_LS,
3348 shader->bo->gpu_address >> 8);
3349 }
3350 void *evergreen_create_resolve_blend(struct r600_context *rctx)
3351 {
3352 struct pipe_blend_state blend;
3353
3354 memset(&blend, 0, sizeof(blend));
3355 blend.independent_blend_enable = true;
3356 blend.rt[0].colormask = 0xf;
3357 return evergreen_create_blend_state_mode(&rctx->b.b, &blend, V_028808_CB_RESOLVE);
3358 }
3359
3360 void *evergreen_create_decompress_blend(struct r600_context *rctx)
3361 {
3362 struct pipe_blend_state blend;
3363 unsigned mode = rctx->screen->has_compressed_msaa_texturing ?
3364 V_028808_CB_FMASK_DECOMPRESS : V_028808_CB_DECOMPRESS;
3365
3366 memset(&blend, 0, sizeof(blend));
3367 blend.independent_blend_enable = true;
3368 blend.rt[0].colormask = 0xf;
3369 return evergreen_create_blend_state_mode(&rctx->b.b, &blend, mode);
3370 }
3371
3372 void *evergreen_create_fastclear_blend(struct r600_context *rctx)
3373 {
3374 struct pipe_blend_state blend;
3375 unsigned mode = V_028808_CB_ELIMINATE_FAST_CLEAR;
3376
3377 memset(&blend, 0, sizeof(blend));
3378 blend.independent_blend_enable = true;
3379 blend.rt[0].colormask = 0xf;
3380 return evergreen_create_blend_state_mode(&rctx->b.b, &blend, mode);
3381 }
3382
3383 void *evergreen_create_db_flush_dsa(struct r600_context *rctx)
3384 {
3385 struct pipe_depth_stencil_alpha_state dsa = {{0}};
3386
3387 return rctx->b.b.create_depth_stencil_alpha_state(&rctx->b.b, &dsa);
3388 }
3389
3390 void evergreen_update_db_shader_control(struct r600_context * rctx)
3391 {
3392 bool dual_export;
3393 unsigned db_shader_control;
3394
3395 if (!rctx->ps_shader) {
3396 return;
3397 }
3398
3399 dual_export = rctx->framebuffer.export_16bpc &&
3400 !rctx->ps_shader->current->ps_depth_export;
3401
3402 db_shader_control = rctx->ps_shader->current->db_shader_control |
3403 S_02880C_DUAL_EXPORT_ENABLE(dual_export) |
3404 S_02880C_DB_SOURCE_FORMAT(dual_export ? V_02880C_EXPORT_DB_TWO :
3405 V_02880C_EXPORT_DB_FULL) |
3406 S_02880C_ALPHA_TO_MASK_DISABLE(rctx->framebuffer.cb0_is_integer);
3407
3408 /* When alpha test is enabled we can't trust the hw to make the proper
3409 * decision on the order in which ztest should be run related to fragment
3410 * shader execution.
3411 *
3412 * If alpha test is enabled perform early z rejection (RE_Z) but don't early
3413 * write to the zbuffer. Write to zbuffer is delayed after fragment shader
3414 * execution and thus after alpha test so if discarded by the alpha test
3415 * the z value is not written.
3416 * If ReZ is enabled, and the zfunc/zenable/zwrite values change you can
3417 * get a hang unless you flush the DB in between. For now just use
3418 * LATE_Z.
3419 */
3420 if (rctx->alphatest_state.sx_alpha_test_control) {
3421 db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
3422 } else {
3423 db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
3424 }
3425
3426 if (db_shader_control != rctx->db_misc_state.db_shader_control) {
3427 rctx->db_misc_state.db_shader_control = db_shader_control;
3428 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
3429 }
3430 }
3431
3432 static void evergreen_dma_copy_tile(struct r600_context *rctx,
3433 struct pipe_resource *dst,
3434 unsigned dst_level,
3435 unsigned dst_x,
3436 unsigned dst_y,
3437 unsigned dst_z,
3438 struct pipe_resource *src,
3439 unsigned src_level,
3440 unsigned src_x,
3441 unsigned src_y,
3442 unsigned src_z,
3443 unsigned copy_height,
3444 unsigned pitch,
3445 unsigned bpp)
3446 {
3447 struct radeon_winsys_cs *cs = rctx->b.dma.cs;
3448 struct r600_texture *rsrc = (struct r600_texture*)src;
3449 struct r600_texture *rdst = (struct r600_texture*)dst;
3450 unsigned array_mode, lbpp, pitch_tile_max, slice_tile_max, size;
3451 unsigned ncopy, height, cheight, detile, i, x, y, z, src_mode, dst_mode;
3452 unsigned sub_cmd, bank_h, bank_w, mt_aspect, nbanks, tile_split, non_disp_tiling = 0;
3453 uint64_t base, addr;
3454
3455 dst_mode = rdst->surface.level[dst_level].mode;
3456 src_mode = rsrc->surface.level[src_level].mode;
3457 /* downcast linear aligned to linear to simplify test */
3458 src_mode = src_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : src_mode;
3459 dst_mode = dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : dst_mode;
3460 assert(dst_mode != src_mode);
3461
3462 /* non_disp_tiling bit needs to be set for depth, stencil, and fmask surfaces */
3463 if (util_format_has_depth(util_format_description(src->format)))
3464 non_disp_tiling = 1;
3465
3466 y = 0;
3467 sub_cmd = EG_DMA_COPY_TILED;
3468 lbpp = util_logbase2(bpp);
3469 pitch_tile_max = ((pitch / bpp) / 8) - 1;
3470 nbanks = eg_num_banks(rctx->screen->b.tiling_info.num_banks);
3471
3472 if (dst_mode == RADEON_SURF_MODE_LINEAR) {
3473 /* T2L */
3474 array_mode = evergreen_array_mode(src_mode);
3475 slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) / (8*8);
3476 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
3477 /* linear height must be the same as the slice tile max height, it's ok even
3478 * if the linear destination/source have smaller heigh as the size of the
3479 * dma packet will be using the copy_height which is always smaller or equal
3480 * to the linear height
3481 */
3482 height = rsrc->surface.level[src_level].npix_y;
3483 detile = 1;
3484 x = src_x;
3485 y = src_y;
3486 z = src_z;
3487 base = rsrc->surface.level[src_level].offset;
3488 addr = rdst->surface.level[dst_level].offset;
3489 addr += rdst->surface.level[dst_level].slice_size * dst_z;
3490 addr += dst_y * pitch + dst_x * bpp;
3491 bank_h = eg_bank_wh(rsrc->surface.bankh);
3492 bank_w = eg_bank_wh(rsrc->surface.bankw);
3493 mt_aspect = eg_macro_tile_aspect(rsrc->surface.mtilea);
3494 tile_split = eg_tile_split(rsrc->surface.tile_split);
3495 base += rsrc->resource.gpu_address;
3496 addr += rdst->resource.gpu_address;
3497 } else {
3498 /* L2T */
3499 array_mode = evergreen_array_mode(dst_mode);
3500 slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) / (8*8);
3501 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
3502 /* linear height must be the same as the slice tile max height, it's ok even
3503 * if the linear destination/source have smaller heigh as the size of the
3504 * dma packet will be using the copy_height which is always smaller or equal
3505 * to the linear height
3506 */
3507 height = rdst->surface.level[dst_level].npix_y;
3508 detile = 0;
3509 x = dst_x;
3510 y = dst_y;
3511 z = dst_z;
3512 base = rdst->surface.level[dst_level].offset;
3513 addr = rsrc->surface.level[src_level].offset;
3514 addr += rsrc->surface.level[src_level].slice_size * src_z;
3515 addr += src_y * pitch + src_x * bpp;
3516 bank_h = eg_bank_wh(rdst->surface.bankh);
3517 bank_w = eg_bank_wh(rdst->surface.bankw);
3518 mt_aspect = eg_macro_tile_aspect(rdst->surface.mtilea);
3519 tile_split = eg_tile_split(rdst->surface.tile_split);
3520 base += rdst->resource.gpu_address;
3521 addr += rsrc->resource.gpu_address;
3522 }
3523
3524 size = (copy_height * pitch) / 4;
3525 ncopy = (size / EG_DMA_COPY_MAX_SIZE) + !!(size % EG_DMA_COPY_MAX_SIZE);
3526 r600_need_dma_space(&rctx->b, ncopy * 9);
3527
3528 for (i = 0; i < ncopy; i++) {
3529 cheight = copy_height;
3530 if (((cheight * pitch) / 4) > EG_DMA_COPY_MAX_SIZE) {
3531 cheight = (EG_DMA_COPY_MAX_SIZE * 4) / pitch;
3532 }
3533 size = (cheight * pitch) / 4;
3534 /* emit reloc before writing cs so that cs is always in consistent state */
3535 radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rsrc->resource,
3536 RADEON_USAGE_READ, RADEON_PRIO_SDMA_TEXTURE);
3537 radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rdst->resource,
3538 RADEON_USAGE_WRITE, RADEON_PRIO_SDMA_TEXTURE);
3539 cs->buf[cs->cdw++] = DMA_PACKET(DMA_PACKET_COPY, sub_cmd, size);
3540 cs->buf[cs->cdw++] = base >> 8;
3541 cs->buf[cs->cdw++] = (detile << 31) | (array_mode << 27) |
3542 (lbpp << 24) | (bank_h << 21) |
3543 (bank_w << 18) | (mt_aspect << 16);
3544 cs->buf[cs->cdw++] = (pitch_tile_max << 0) | ((height - 1) << 16);
3545 cs->buf[cs->cdw++] = (slice_tile_max << 0);
3546 cs->buf[cs->cdw++] = (x << 0) | (z << 18);
3547 cs->buf[cs->cdw++] = (y << 0) | (tile_split << 21) | (nbanks << 25) | (non_disp_tiling << 28);
3548 cs->buf[cs->cdw++] = addr & 0xfffffffc;
3549 cs->buf[cs->cdw++] = (addr >> 32UL) & 0xff;
3550 copy_height -= cheight;
3551 addr += cheight * pitch;
3552 y += cheight;
3553 }
3554 }
3555
3556 static void evergreen_dma_copy(struct pipe_context *ctx,
3557 struct pipe_resource *dst,
3558 unsigned dst_level,
3559 unsigned dstx, unsigned dsty, unsigned dstz,
3560 struct pipe_resource *src,
3561 unsigned src_level,
3562 const struct pipe_box *src_box)
3563 {
3564 struct r600_context *rctx = (struct r600_context *)ctx;
3565 struct r600_texture *rsrc = (struct r600_texture*)src;
3566 struct r600_texture *rdst = (struct r600_texture*)dst;
3567 unsigned dst_pitch, src_pitch, bpp, dst_mode, src_mode, copy_height;
3568 unsigned src_w, dst_w;
3569 unsigned src_x, src_y;
3570 unsigned dst_x = dstx, dst_y = dsty, dst_z = dstz;
3571
3572 if (rctx->b.dma.cs == NULL) {
3573 goto fallback;
3574 }
3575
3576 if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
3577 evergreen_dma_copy_buffer(rctx, dst, src, dst_x, src_box->x, src_box->width);
3578 return;
3579 }
3580
3581 if (src->format != dst->format || src_box->depth > 1 ||
3582 (rdst->dirty_level_mask | rdst->stencil_dirty_level_mask) & (1 << dst_level)) {
3583 goto fallback;
3584 }
3585
3586 if (rsrc->dirty_level_mask & (1 << src_level)) {
3587 ctx->flush_resource(ctx, src);
3588 }
3589
3590 src_x = util_format_get_nblocksx(src->format, src_box->x);
3591 dst_x = util_format_get_nblocksx(src->format, dst_x);
3592 src_y = util_format_get_nblocksy(src->format, src_box->y);
3593 dst_y = util_format_get_nblocksy(src->format, dst_y);
3594
3595 bpp = rdst->surface.bpe;
3596 dst_pitch = rdst->surface.level[dst_level].pitch_bytes;
3597 src_pitch = rsrc->surface.level[src_level].pitch_bytes;
3598 src_w = rsrc->surface.level[src_level].npix_x;
3599 dst_w = rdst->surface.level[dst_level].npix_x;
3600 copy_height = src_box->height / rsrc->surface.blk_h;
3601
3602 dst_mode = rdst->surface.level[dst_level].mode;
3603 src_mode = rsrc->surface.level[src_level].mode;
3604 /* downcast linear aligned to linear to simplify test */
3605 src_mode = src_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : src_mode;
3606 dst_mode = dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : dst_mode;
3607
3608 if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) {
3609 /* FIXME evergreen can do partial blit */
3610 goto fallback;
3611 }
3612 /* the x test here are currently useless (because we don't support partial blit)
3613 * but keep them around so we don't forget about those
3614 */
3615 if (src_pitch % 8 || src_box->x % 8 || dst_x % 8 || src_box->y % 8 || dst_y % 8) {
3616 goto fallback;
3617 }
3618
3619 /* 128 bpp surfaces require non_disp_tiling for both
3620 * tiled and linear buffers on cayman. However, async
3621 * DMA only supports it on the tiled side. As such
3622 * the tile order is backwards after a L2T/T2L packet.
3623 */
3624 if ((rctx->b.chip_class == CAYMAN) &&
3625 (src_mode != dst_mode) &&
3626 (util_format_get_blocksize(src->format) >= 16)) {
3627 goto fallback;
3628 }
3629
3630 if (src_mode == dst_mode) {
3631 uint64_t dst_offset, src_offset;
3632 /* simple dma blit would do NOTE code here assume :
3633 * src_box.x/y == 0
3634 * dst_x/y == 0
3635 * dst_pitch == src_pitch
3636 */
3637 src_offset= rsrc->surface.level[src_level].offset;
3638 src_offset += rsrc->surface.level[src_level].slice_size * src_box->z;
3639 src_offset += src_y * src_pitch + src_x * bpp;
3640 dst_offset = rdst->surface.level[dst_level].offset;
3641 dst_offset += rdst->surface.level[dst_level].slice_size * dst_z;
3642 dst_offset += dst_y * dst_pitch + dst_x * bpp;
3643 evergreen_dma_copy_buffer(rctx, dst, src, dst_offset, src_offset,
3644 src_box->height * src_pitch);
3645 } else {
3646 evergreen_dma_copy_tile(rctx, dst, dst_level, dst_x, dst_y, dst_z,
3647 src, src_level, src_x, src_y, src_box->z,
3648 copy_height, dst_pitch, bpp);
3649 }
3650 return;
3651
3652 fallback:
3653 r600_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
3654 src, src_level, src_box);
3655 }
3656
3657 static void evergreen_set_tess_state(struct pipe_context *ctx,
3658 const float default_outer_level[4],
3659 const float default_inner_level[2])
3660 {
3661 struct r600_context *rctx = (struct r600_context *)ctx;
3662
3663 memcpy(rctx->tess_state, default_outer_level, sizeof(float) * 4);
3664 memcpy(rctx->tess_state+4, default_inner_level, sizeof(float) * 2);
3665 rctx->tess_state_dirty = true;
3666 }
3667
3668 void evergreen_init_state_functions(struct r600_context *rctx)
3669 {
3670 unsigned id = 1;
3671 unsigned i;
3672 /* !!!
3673 * To avoid GPU lockup registers must be emited in a specific order
3674 * (no kidding ...). The order below is important and have been
3675 * partialy infered from analyzing fglrx command stream.
3676 *
3677 * Don't reorder atom without carefully checking the effect (GPU lockup
3678 * or piglit regression).
3679 * !!!
3680 */
3681 if (rctx->b.chip_class == EVERGREEN) {
3682 r600_init_atom(rctx, &rctx->config_state.atom, id++, evergreen_emit_config_state, 11);
3683 if (rctx->screen->b.info.drm_minor >= 7)
3684 rctx->config_state.dyn_gpr_enabled = true;
3685 }
3686 r600_init_atom(rctx, &rctx->framebuffer.atom, id++, evergreen_emit_framebuffer_state, 0);
3687 /* shader const */
3688 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX].atom, id++, evergreen_emit_vs_constant_buffers, 0);
3689 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY].atom, id++, evergreen_emit_gs_constant_buffers, 0);
3690 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT].atom, id++, evergreen_emit_ps_constant_buffers, 0);
3691 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_CTRL].atom, id++, evergreen_emit_tcs_constant_buffers, 0);
3692 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_EVAL].atom, id++, evergreen_emit_tes_constant_buffers, 0);
3693 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_COMPUTE].atom, id++, evergreen_emit_cs_constant_buffers, 0);
3694 /* shader program */
3695 r600_init_atom(rctx, &rctx->cs_shader_state.atom, id++, evergreen_emit_cs_shader, 0);
3696 /* sampler */
3697 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].states.atom, id++, evergreen_emit_vs_sampler_states, 0);
3698 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].states.atom, id++, evergreen_emit_gs_sampler_states, 0);
3699 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL].states.atom, id++, evergreen_emit_tcs_sampler_states, 0);
3700 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL].states.atom, id++, evergreen_emit_tes_sampler_states, 0);
3701 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].states.atom, id++, evergreen_emit_ps_sampler_states, 0);
3702 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE].states.atom, id++, evergreen_emit_cs_sampler_states, 0);
3703 /* resources */
3704 r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, evergreen_fs_emit_vertex_buffers, 0);
3705 r600_init_atom(rctx, &rctx->cs_vertex_buffer_state.atom, id++, evergreen_cs_emit_vertex_buffers, 0);
3706 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views.atom, id++, evergreen_emit_vs_sampler_views, 0);
3707 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views.atom, id++, evergreen_emit_gs_sampler_views, 0);
3708 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL].views.atom, id++, evergreen_emit_tcs_sampler_views, 0);
3709 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL].views.atom, id++, evergreen_emit_tes_sampler_views, 0);
3710 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views.atom, id++, evergreen_emit_ps_sampler_views, 0);
3711 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE].views.atom, id++, evergreen_emit_cs_sampler_views, 0);
3712
3713 r600_init_atom(rctx, &rctx->vgt_state.atom, id++, r600_emit_vgt_state, 10);
3714
3715 if (rctx->b.chip_class == EVERGREEN) {
3716 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, evergreen_emit_sample_mask, 3);
3717 } else {
3718 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, cayman_emit_sample_mask, 4);
3719 }
3720 rctx->sample_mask.sample_mask = ~0;
3721
3722 r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
3723 r600_init_atom(rctx, &rctx->blend_color.atom, id++, r600_emit_blend_color, 6);
3724 r600_init_atom(rctx, &rctx->blend_state.atom, id++, r600_emit_cso_state, 0);
3725 r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, evergreen_emit_cb_misc_state, 4);
3726 r600_init_atom(rctx, &rctx->clip_misc_state.atom, id++, r600_emit_clip_misc_state, 9);
3727 r600_init_atom(rctx, &rctx->clip_state.atom, id++, evergreen_emit_clip_state, 26);
3728 r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, evergreen_emit_db_misc_state, 10);
3729 r600_init_atom(rctx, &rctx->db_state.atom, id++, evergreen_emit_db_state, 14);
3730 r600_init_atom(rctx, &rctx->dsa_state.atom, id++, r600_emit_cso_state, 0);
3731 r600_init_atom(rctx, &rctx->poly_offset_state.atom, id++, evergreen_emit_polygon_offset, 6);
3732 r600_init_atom(rctx, &rctx->rasterizer_state.atom, id++, r600_emit_cso_state, 0);
3733 r600_init_atom(rctx, &rctx->scissor.atom, id++, evergreen_emit_scissor_state, 0);
3734 r600_init_atom(rctx, &rctx->viewport.atom, id++, r600_emit_viewport_state, 0);
3735 r600_init_atom(rctx, &rctx->stencil_ref.atom, id++, r600_emit_stencil_ref, 4);
3736 r600_init_atom(rctx, &rctx->vertex_fetch_shader.atom, id++, evergreen_emit_vertex_fetch_shader, 5);
3737 r600_add_atom(rctx, &rctx->b.render_cond_atom, id++);
3738 r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++);
3739 r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++);
3740 for (i = 0; i < EG_NUM_HW_STAGES; i++)
3741 r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0);
3742 r600_init_atom(rctx, &rctx->shader_stages.atom, id++, evergreen_emit_shader_stages, 15);
3743 r600_init_atom(rctx, &rctx->gs_rings.atom, id++, evergreen_emit_gs_rings, 26);
3744
3745 rctx->b.b.create_blend_state = evergreen_create_blend_state;
3746 rctx->b.b.create_depth_stencil_alpha_state = evergreen_create_dsa_state;
3747 rctx->b.b.create_rasterizer_state = evergreen_create_rs_state;
3748 rctx->b.b.create_sampler_state = evergreen_create_sampler_state;
3749 rctx->b.b.create_sampler_view = evergreen_create_sampler_view;
3750 rctx->b.b.set_framebuffer_state = evergreen_set_framebuffer_state;
3751 rctx->b.b.set_polygon_stipple = evergreen_set_polygon_stipple;
3752 rctx->b.b.set_min_samples = evergreen_set_min_samples;
3753 rctx->b.b.set_scissor_states = evergreen_set_scissor_states;
3754 rctx->b.b.set_tess_state = evergreen_set_tess_state;
3755 if (rctx->b.chip_class == EVERGREEN)
3756 rctx->b.b.get_sample_position = evergreen_get_sample_position;
3757 else
3758 rctx->b.b.get_sample_position = cayman_get_sample_position;
3759 rctx->b.dma_copy = evergreen_dma_copy;
3760
3761 evergreen_init_compute_state_functions(rctx);
3762 }
3763
3764 /**
3765 * This calculates the LDS size for tessellation shaders (VS, TCS, TES).
3766 *
3767 * The information about LDS and other non-compile-time parameters is then
3768 * written to the const buffer.
3769
3770 * const buffer contains -
3771 * uint32_t input_patch_size
3772 * uint32_t input_vertex_size
3773 * uint32_t num_tcs_input_cp
3774 * uint32_t num_tcs_output_cp;
3775 * uint32_t output_patch_size
3776 * uint32_t output_vertex_size
3777 * uint32_t output_patch0_offset
3778 * uint32_t perpatch_output_offset
3779 * and the same constbuf is bound to LS/HS/VS(ES).
3780 */
3781 void evergreen_setup_tess_constants(struct r600_context *rctx, const struct pipe_draw_info *info, unsigned *num_patches)
3782 {
3783 struct pipe_constant_buffer constbuf = {0};
3784 struct r600_pipe_shader_selector *tcs = rctx->tcs_shader ? rctx->tcs_shader : rctx->tes_shader;
3785 struct r600_pipe_shader_selector *ls = rctx->vs_shader;
3786 unsigned num_tcs_input_cp = info->vertices_per_patch;
3787 unsigned num_tcs_outputs;
3788 unsigned num_tcs_output_cp;
3789 unsigned num_tcs_patch_outputs;
3790 unsigned num_tcs_inputs;
3791 unsigned input_vertex_size, output_vertex_size;
3792 unsigned input_patch_size, pervertex_output_patch_size, output_patch_size;
3793 unsigned output_patch0_offset, perpatch_output_offset, lds_size;
3794 uint32_t values[16];
3795 unsigned num_waves;
3796 unsigned num_pipes = rctx->screen->b.info.r600_max_pipes;
3797 unsigned wave_divisor = (16 * num_pipes);
3798
3799 *num_patches = 1;
3800
3801 if (!rctx->tes_shader) {
3802 rctx->lds_alloc = 0;
3803 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_VERTEX,
3804 R600_LDS_INFO_CONST_BUFFER, NULL);
3805 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_CTRL,
3806 R600_LDS_INFO_CONST_BUFFER, NULL);
3807 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_EVAL,
3808 R600_LDS_INFO_CONST_BUFFER, NULL);
3809 return;
3810 }
3811
3812 if (rctx->lds_alloc != 0 &&
3813 rctx->last_ls == ls &&
3814 !rctx->tess_state_dirty &&
3815 rctx->last_num_tcs_input_cp == num_tcs_input_cp &&
3816 rctx->last_tcs == tcs)
3817 return;
3818
3819 num_tcs_inputs = util_last_bit64(ls->lds_outputs_written_mask);
3820
3821 if (rctx->tcs_shader) {
3822 num_tcs_outputs = util_last_bit64(tcs->lds_outputs_written_mask);
3823 num_tcs_output_cp = tcs->info.properties[TGSI_PROPERTY_TCS_VERTICES_OUT];
3824 num_tcs_patch_outputs = util_last_bit64(tcs->lds_patch_outputs_written_mask);
3825 } else {
3826 num_tcs_outputs = num_tcs_inputs;
3827 num_tcs_output_cp = num_tcs_input_cp;
3828 num_tcs_patch_outputs = 2; /* TESSINNER + TESSOUTER */
3829 }
3830
3831 /* size in bytes */
3832 input_vertex_size = num_tcs_inputs * 16;
3833 output_vertex_size = num_tcs_outputs * 16;
3834
3835 input_patch_size = num_tcs_input_cp * input_vertex_size;
3836
3837 pervertex_output_patch_size = num_tcs_output_cp * output_vertex_size;
3838 output_patch_size = pervertex_output_patch_size + num_tcs_patch_outputs * 16;
3839
3840 output_patch0_offset = rctx->tcs_shader ? input_patch_size * *num_patches : 0;
3841 perpatch_output_offset = output_patch0_offset + pervertex_output_patch_size;
3842
3843 lds_size = output_patch0_offset + output_patch_size * *num_patches;
3844
3845 values[0] = input_patch_size;
3846 values[1] = input_vertex_size;
3847 values[2] = num_tcs_input_cp;
3848 values[3] = num_tcs_output_cp;
3849
3850 values[4] = output_patch_size;
3851 values[5] = output_vertex_size;
3852 values[6] = output_patch0_offset;
3853 values[7] = perpatch_output_offset;
3854
3855 /* docs say HS_NUM_WAVES - CEIL((LS_HS_CONFIG.NUM_PATCHES *
3856 LS_HS_CONFIG.HS_NUM_OUTPUT_CP) / (NUM_GOOD_PIPES * 16)) */
3857 num_waves = ceilf((float)(*num_patches * num_tcs_output_cp) / (float)wave_divisor);
3858
3859 rctx->lds_alloc = (lds_size | (num_waves << 14));
3860
3861 memcpy(&values[8], rctx->tess_state, 6 * sizeof(float));
3862 values[14] = 0;
3863 values[15] = 0;
3864
3865 rctx->tess_state_dirty = false;
3866 rctx->last_ls = ls;
3867 rctx->last_tcs = tcs;
3868 rctx->last_num_tcs_input_cp = num_tcs_input_cp;
3869
3870 constbuf.user_buffer = values;
3871 constbuf.buffer_size = 16 * 4;
3872
3873 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_VERTEX,
3874 R600_LDS_INFO_CONST_BUFFER, &constbuf);
3875 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_CTRL,
3876 R600_LDS_INFO_CONST_BUFFER, &constbuf);
3877 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_EVAL,
3878 R600_LDS_INFO_CONST_BUFFER, &constbuf);
3879 pipe_resource_reference(&constbuf.buffer, NULL);
3880 }
3881
3882 uint32_t evergreen_get_ls_hs_config(struct r600_context *rctx,
3883 const struct pipe_draw_info *info,
3884 unsigned num_patches)
3885 {
3886 unsigned num_output_cp;
3887
3888 if (!rctx->tes_shader)
3889 return 0;
3890
3891 num_output_cp = rctx->tcs_shader ?
3892 rctx->tcs_shader->info.properties[TGSI_PROPERTY_TCS_VERTICES_OUT] :
3893 info->vertices_per_patch;
3894
3895 return S_028B58_NUM_PATCHES(num_patches) |
3896 S_028B58_HS_NUM_INPUT_CP(info->vertices_per_patch) |
3897 S_028B58_HS_NUM_OUTPUT_CP(num_output_cp);
3898 }
3899
3900 void evergreen_set_ls_hs_config(struct r600_context *rctx,
3901 struct radeon_winsys_cs *cs,
3902 uint32_t ls_hs_config)
3903 {
3904 radeon_set_context_reg(cs, R_028B58_VGT_LS_HS_CONFIG, ls_hs_config);
3905 }
3906
3907 void evergreen_set_lds_alloc(struct r600_context *rctx,
3908 struct radeon_winsys_cs *cs,
3909 uint32_t lds_alloc)
3910 {
3911 radeon_set_context_reg(cs, R_0288E8_SQ_LDS_ALLOC, lds_alloc);
3912 }
3913
3914 /* on evergreen if you are running tessellation you need to disable dynamic
3915 GPRs to workaround a hardware bug.*/
3916 bool evergreen_adjust_gprs(struct r600_context *rctx)
3917 {
3918 unsigned num_gprs[EG_NUM_HW_STAGES];
3919 unsigned def_gprs[EG_NUM_HW_STAGES];
3920 unsigned cur_gprs[EG_NUM_HW_STAGES];
3921 unsigned new_gprs[EG_NUM_HW_STAGES];
3922 unsigned def_num_clause_temp_gprs = rctx->r6xx_num_clause_temp_gprs;
3923 unsigned max_gprs;
3924 unsigned i;
3925 unsigned total_gprs;
3926 unsigned tmp[3];
3927 bool rework = false, set_default = false, set_dirty = false;
3928 max_gprs = 0;
3929 for (i = 0; i < EG_NUM_HW_STAGES; i++) {
3930 def_gprs[i] = rctx->default_gprs[i];
3931 max_gprs += def_gprs[i];
3932 }
3933 max_gprs += def_num_clause_temp_gprs * 2;
3934
3935 /* if we have no TESS and dyn gpr is enabled then do nothing. */
3936 if (!rctx->hw_shader_stages[EG_HW_STAGE_HS].shader || rctx->screen->b.info.drm_minor < 7) {
3937 if (rctx->config_state.dyn_gpr_enabled)
3938 return true;
3939
3940 /* transition back to dyn gpr enabled state */
3941 rctx->config_state.dyn_gpr_enabled = true;
3942 r600_mark_atom_dirty(rctx, &rctx->config_state.atom);
3943 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE;
3944 return true;
3945 }
3946
3947
3948 /* gather required shader gprs */
3949 for (i = 0; i < EG_NUM_HW_STAGES; i++) {
3950 if (rctx->hw_shader_stages[i].shader)
3951 num_gprs[i] = rctx->hw_shader_stages[i].shader->shader.bc.ngpr;
3952 else
3953 num_gprs[i] = 0;
3954 }
3955
3956 cur_gprs[R600_HW_STAGE_PS] = G_008C04_NUM_PS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
3957 cur_gprs[R600_HW_STAGE_VS] = G_008C04_NUM_VS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
3958 cur_gprs[R600_HW_STAGE_GS] = G_008C08_NUM_GS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
3959 cur_gprs[R600_HW_STAGE_ES] = G_008C08_NUM_ES_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
3960 cur_gprs[EG_HW_STAGE_LS] = G_008C0C_NUM_LS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_3);
3961 cur_gprs[EG_HW_STAGE_HS] = G_008C0C_NUM_HS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_3);
3962
3963 total_gprs = 0;
3964 for (i = 0; i < EG_NUM_HW_STAGES; i++) {
3965 new_gprs[i] = num_gprs[i];
3966 total_gprs += num_gprs[i];
3967 }
3968
3969 if (total_gprs > (max_gprs - (2 * def_num_clause_temp_gprs)))
3970 return false;
3971
3972 for (i = 0; i < EG_NUM_HW_STAGES; i++) {
3973 if (new_gprs[i] > cur_gprs[i]) {
3974 rework = true;
3975 break;
3976 }
3977 }
3978
3979 if (rctx->config_state.dyn_gpr_enabled) {
3980 set_dirty = true;
3981 rctx->config_state.dyn_gpr_enabled = false;
3982 }
3983
3984 if (rework) {
3985 set_default = true;
3986 for (i = 0; i < EG_NUM_HW_STAGES; i++) {
3987 if (new_gprs[i] > def_gprs[i])
3988 set_default = false;
3989 }
3990
3991 if (set_default) {
3992 for (i = 0; i < EG_NUM_HW_STAGES; i++) {
3993 new_gprs[i] = def_gprs[i];
3994 }
3995 } else {
3996 unsigned ps_value = max_gprs;
3997
3998 ps_value -= (def_num_clause_temp_gprs * 2);
3999 for (i = R600_HW_STAGE_VS; i < EG_NUM_HW_STAGES; i++)
4000 ps_value -= new_gprs[i];
4001
4002 new_gprs[R600_HW_STAGE_PS] = ps_value;
4003 }
4004
4005 tmp[0] = S_008C04_NUM_PS_GPRS(new_gprs[R600_HW_STAGE_PS]) |
4006 S_008C04_NUM_VS_GPRS(new_gprs[R600_HW_STAGE_VS]) |
4007 S_008C04_NUM_CLAUSE_TEMP_GPRS(def_num_clause_temp_gprs);
4008
4009 tmp[1] = S_008C08_NUM_ES_GPRS(new_gprs[R600_HW_STAGE_ES]) |
4010 S_008C08_NUM_GS_GPRS(new_gprs[R600_HW_STAGE_GS]);
4011
4012 tmp[2] = S_008C0C_NUM_HS_GPRS(new_gprs[EG_HW_STAGE_HS]) |
4013 S_008C0C_NUM_LS_GPRS(new_gprs[EG_HW_STAGE_LS]);
4014
4015 if (rctx->config_state.sq_gpr_resource_mgmt_1 != tmp[0] ||
4016 rctx->config_state.sq_gpr_resource_mgmt_2 != tmp[1] ||
4017 rctx->config_state.sq_gpr_resource_mgmt_3 != tmp[2]) {
4018 rctx->config_state.sq_gpr_resource_mgmt_1 = tmp[0];
4019 rctx->config_state.sq_gpr_resource_mgmt_2 = tmp[1];
4020 rctx->config_state.sq_gpr_resource_mgmt_3 = tmp[2];
4021 set_dirty = true;
4022 }
4023 }
4024
4025
4026 if (set_dirty) {
4027 r600_mark_atom_dirty(rctx, &rctx->config_state.atom);
4028 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE;
4029 }
4030 return true;
4031 }