projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radeonsi: Simplify shader PM4 state handling
[mesa.git] / src / gallium / drivers / radeonsi / si_state.c
1 /*
2 * Copyright 2012 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Christian König <christian.koenig@amd.com>
25 */
26
27 #include "si_pipe.h"
28 #include "si_shader.h"
29 #include "sid.h"
30 #include "../radeon/r600_cs.h"
31
32 #include "tgsi/tgsi_parse.h"
33 #include "tgsi/tgsi_scan.h"
34 #include "util/u_format.h"
35 #include "util/u_format_s3tc.h"
36 #include "util/u_framebuffer.h"
37 #include "util/u_helpers.h"
38 #include "util/u_memory.h"
39
40 static uint32_t cik_num_banks(struct si_screen *sscreen, unsigned bpe, unsigned tile_split)
41 {
42 if (sscreen->b.info.cik_macrotile_mode_array_valid) {
43 unsigned index, tileb;
44
45 tileb = 8 * 8 * bpe;
46 tileb = MIN2(tile_split, tileb);
47
48 for (index = 0; tileb > 64; index++) {
49 tileb >>= 1;
50 }
51
52 assert(index < 16);
53
54 return (sscreen->b.info.cik_macrotile_mode_array[index] >> 6) & 0x3;
55 }
56
57 /* The old way. */
58 switch (sscreen->b.tiling_info.num_banks) {
59 case 2:
60 return V_02803C_ADDR_SURF_2_BANK;
61 case 4:
62 return V_02803C_ADDR_SURF_4_BANK;
63 case 8:
64 default:
65 return V_02803C_ADDR_SURF_8_BANK;
66 case 16:
67 return V_02803C_ADDR_SURF_16_BANK;
68 }
69 }
70
71 static unsigned cik_tile_split(unsigned tile_split)
72 {
73 switch (tile_split) {
74 case 64:
75 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_64B;
76 break;
77 case 128:
78 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_128B;
79 break;
80 case 256:
81 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_256B;
82 break;
83 case 512:
84 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_512B;
85 break;
86 default:
87 case 1024:
88 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_1KB;
89 break;
90 case 2048:
91 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_2KB;
92 break;
93 case 4096:
94 tile_split = V_028040_ADDR_SURF_TILE_SPLIT_4KB;
95 break;
96 }
97 return tile_split;
98 }
99
100 static unsigned cik_macro_tile_aspect(unsigned macro_tile_aspect)
101 {
102 switch (macro_tile_aspect) {
103 default:
104 case 1:
105 macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_1;
106 break;
107 case 2:
108 macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_2;
109 break;
110 case 4:
111 macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_4;
112 break;
113 case 8:
114 macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_8;
115 break;
116 }
117 return macro_tile_aspect;
118 }
119
120 static unsigned cik_bank_wh(unsigned bankwh)
121 {
122 switch (bankwh) {
123 default:
124 case 1:
125 bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_1;
126 break;
127 case 2:
128 bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_2;
129 break;
130 case 4:
131 bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_4;
132 break;
133 case 8:
134 bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_8;
135 break;
136 }
137 return bankwh;
138 }
139
140 static unsigned cik_db_pipe_config(struct si_screen *sscreen, unsigned tile_mode)
141 {
142 if (sscreen->b.info.si_tile_mode_array_valid) {
143 uint32_t gb_tile_mode = sscreen->b.info.si_tile_mode_array[tile_mode];
144
145 return G_009910_PIPE_CONFIG(gb_tile_mode);
146 }
147
148 /* This is probably broken for a lot of chips, but it's only used
149 * if the kernel cannot return the tile mode array for CIK. */
150 switch (sscreen->b.info.r600_num_tile_pipes) {
151 case 16:
152 return V_02803C_X_ADDR_SURF_P16_32X32_16X16;
153 case 8:
154 return V_02803C_X_ADDR_SURF_P8_32X32_16X16;
155 case 4:
156 default:
157 if (sscreen->b.info.r600_num_backends == 4)
158 return V_02803C_X_ADDR_SURF_P4_16X16;
159 else
160 return V_02803C_X_ADDR_SURF_P4_8X16;
161 case 2:
162 return V_02803C_ADDR_SURF_P2;
163 }
164 }
165
166 static unsigned si_map_swizzle(unsigned swizzle)
167 {
168 switch (swizzle) {
169 case UTIL_FORMAT_SWIZZLE_Y:
170 return V_008F0C_SQ_SEL_Y;
171 case UTIL_FORMAT_SWIZZLE_Z:
172 return V_008F0C_SQ_SEL_Z;
173 case UTIL_FORMAT_SWIZZLE_W:
174 return V_008F0C_SQ_SEL_W;
175 case UTIL_FORMAT_SWIZZLE_0:
176 return V_008F0C_SQ_SEL_0;
177 case UTIL_FORMAT_SWIZZLE_1:
178 return V_008F0C_SQ_SEL_1;
179 default: /* UTIL_FORMAT_SWIZZLE_X */
180 return V_008F0C_SQ_SEL_X;
181 }
182 }
183
184 static uint32_t S_FIXED(float value, uint32_t frac_bits)
185 {
186 return value * (1 << frac_bits);
187 }
188
189 /* 12.4 fixed-point */
190 static unsigned si_pack_float_12p4(float x)
191 {
192 return x <= 0 ? 0 :
193 x >= 4096 ? 0xffff : x * 16;
194 }
195
196 /*
197 * inferred framebuffer and blender state
198 */
199 static void si_update_fb_blend_state(struct si_context *sctx)
200 {
201 struct si_pm4_state *pm4;
202 struct si_state_blend *blend = sctx->queued.named.blend;
203 uint32_t mask;
204
205 if (blend == NULL)
206 return;
207
208 pm4 = si_pm4_alloc_state(sctx);
209 if (pm4 == NULL)
210 return;
211
212 mask = (1ULL << ((unsigned)sctx->framebuffer.nr_cbufs * 4)) - 1;
213 mask &= blend->cb_target_mask;
214 si_pm4_set_reg(pm4, R_028238_CB_TARGET_MASK, mask);
215
216 si_pm4_set_state(sctx, fb_blend, pm4);
217 }
218
219 /*
220 * Blender functions
221 */
222
223 static uint32_t si_translate_blend_function(int blend_func)
224 {
225 switch (blend_func) {
226 case PIPE_BLEND_ADD:
227 return V_028780_COMB_DST_PLUS_SRC;
228 case PIPE_BLEND_SUBTRACT:
229 return V_028780_COMB_SRC_MINUS_DST;
230 case PIPE_BLEND_REVERSE_SUBTRACT:
231 return V_028780_COMB_DST_MINUS_SRC;
232 case PIPE_BLEND_MIN:
233 return V_028780_COMB_MIN_DST_SRC;
234 case PIPE_BLEND_MAX:
235 return V_028780_COMB_MAX_DST_SRC;
236 default:
237 R600_ERR("Unknown blend function %d\n", blend_func);
238 assert(0);
239 break;
240 }
241 return 0;
242 }
243
244 static uint32_t si_translate_blend_factor(int blend_fact)
245 {
246 switch (blend_fact) {
247 case PIPE_BLENDFACTOR_ONE:
248 return V_028780_BLEND_ONE;
249 case PIPE_BLENDFACTOR_SRC_COLOR:
250 return V_028780_BLEND_SRC_COLOR;
251 case PIPE_BLENDFACTOR_SRC_ALPHA:
252 return V_028780_BLEND_SRC_ALPHA;
253 case PIPE_BLENDFACTOR_DST_ALPHA:
254 return V_028780_BLEND_DST_ALPHA;
255 case PIPE_BLENDFACTOR_DST_COLOR:
256 return V_028780_BLEND_DST_COLOR;
257 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
258 return V_028780_BLEND_SRC_ALPHA_SATURATE;
259 case PIPE_BLENDFACTOR_CONST_COLOR:
260 return V_028780_BLEND_CONSTANT_COLOR;
261 case PIPE_BLENDFACTOR_CONST_ALPHA:
262 return V_028780_BLEND_CONSTANT_ALPHA;
263 case PIPE_BLENDFACTOR_ZERO:
264 return V_028780_BLEND_ZERO;
265 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
266 return V_028780_BLEND_ONE_MINUS_SRC_COLOR;
267 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
268 return V_028780_BLEND_ONE_MINUS_SRC_ALPHA;
269 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
270 return V_028780_BLEND_ONE_MINUS_DST_ALPHA;
271 case PIPE_BLENDFACTOR_INV_DST_COLOR:
272 return V_028780_BLEND_ONE_MINUS_DST_COLOR;
273 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
274 return V_028780_BLEND_ONE_MINUS_CONSTANT_COLOR;
275 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
276 return V_028780_BLEND_ONE_MINUS_CONSTANT_ALPHA;
277 case PIPE_BLENDFACTOR_SRC1_COLOR:
278 return V_028780_BLEND_SRC1_COLOR;
279 case PIPE_BLENDFACTOR_SRC1_ALPHA:
280 return V_028780_BLEND_SRC1_ALPHA;
281 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
282 return V_028780_BLEND_INV_SRC1_COLOR;
283 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
284 return V_028780_BLEND_INV_SRC1_ALPHA;
285 default:
286 R600_ERR("Bad blend factor %d not supported!\n", blend_fact);
287 assert(0);
288 break;
289 }
290 return 0;
291 }
292
293 static void *si_create_blend_state_mode(struct pipe_context *ctx,
294 const struct pipe_blend_state *state,
295 unsigned mode)
296 {
297 struct si_state_blend *blend = CALLOC_STRUCT(si_state_blend);
298 struct si_pm4_state *pm4 = &blend->pm4;
299
300 uint32_t color_control = 0;
301
302 if (blend == NULL)
303 return NULL;
304
305 blend->alpha_to_one = state->alpha_to_one;
306
307 if (state->logicop_enable) {
308 color_control |= S_028808_ROP3(state->logicop_func | (state->logicop_func << 4));
309 } else {
310 color_control |= S_028808_ROP3(0xcc);
311 }
312
313 si_pm4_set_reg(pm4, R_028B70_DB_ALPHA_TO_MASK,
314 S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
315 S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
316 S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
317 S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
318 S_028B70_ALPHA_TO_MASK_OFFSET3(2));
319
320 blend->cb_target_mask = 0;
321 for (int i = 0; i < 8; i++) {
322 /* state->rt entries > 0 only written if independent blending */
323 const int j = state->independent_blend_enable ? i : 0;
324
325 unsigned eqRGB = state->rt[j].rgb_func;
326 unsigned srcRGB = state->rt[j].rgb_src_factor;
327 unsigned dstRGB = state->rt[j].rgb_dst_factor;
328 unsigned eqA = state->rt[j].alpha_func;
329 unsigned srcA = state->rt[j].alpha_src_factor;
330 unsigned dstA = state->rt[j].alpha_dst_factor;
331
332 unsigned blend_cntl = 0;
333
334 /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
335 blend->cb_target_mask |= state->rt[j].colormask << (4 * i);
336
337 if (!state->rt[j].blend_enable) {
338 si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
339 continue;
340 }
341
342 blend_cntl |= S_028780_ENABLE(1);
343 blend_cntl |= S_028780_COLOR_COMB_FCN(si_translate_blend_function(eqRGB));
344 blend_cntl |= S_028780_COLOR_SRCBLEND(si_translate_blend_factor(srcRGB));
345 blend_cntl |= S_028780_COLOR_DESTBLEND(si_translate_blend_factor(dstRGB));
346
347 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
348 blend_cntl |= S_028780_SEPARATE_ALPHA_BLEND(1);
349 blend_cntl |= S_028780_ALPHA_COMB_FCN(si_translate_blend_function(eqA));
350 blend_cntl |= S_028780_ALPHA_SRCBLEND(si_translate_blend_factor(srcA));
351 blend_cntl |= S_028780_ALPHA_DESTBLEND(si_translate_blend_factor(dstA));
352 }
353 si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
354 }
355
356 if (blend->cb_target_mask) {
357 color_control |= S_028808_MODE(mode);
358 } else {
359 color_control |= S_028808_MODE(V_028808_CB_DISABLE);
360 }
361 si_pm4_set_reg(pm4, R_028808_CB_COLOR_CONTROL, color_control);
362
363 return blend;
364 }
365
366 static void *si_create_blend_state(struct pipe_context *ctx,
367 const struct pipe_blend_state *state)
368 {
369 return si_create_blend_state_mode(ctx, state, V_028808_CB_NORMAL);
370 }
371
372 static void si_bind_blend_state(struct pipe_context *ctx, void *state)
373 {
374 struct si_context *sctx = (struct si_context *)ctx;
375 si_pm4_bind_state(sctx, blend, (struct si_state_blend *)state);
376 si_update_fb_blend_state(sctx);
377 }
378
379 static void si_delete_blend_state(struct pipe_context *ctx, void *state)
380 {
381 struct si_context *sctx = (struct si_context *)ctx;
382 si_pm4_delete_state(sctx, blend, (struct si_state_blend *)state);
383 }
384
385 static void si_set_blend_color(struct pipe_context *ctx,
386 const struct pipe_blend_color *state)
387 {
388 struct si_context *sctx = (struct si_context *)ctx;
389 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
390
391 if (pm4 == NULL)
392 return;
393
394 si_pm4_set_reg(pm4, R_028414_CB_BLEND_RED, fui(state->color[0]));
395 si_pm4_set_reg(pm4, R_028418_CB_BLEND_GREEN, fui(state->color[1]));
396 si_pm4_set_reg(pm4, R_02841C_CB_BLEND_BLUE, fui(state->color[2]));
397 si_pm4_set_reg(pm4, R_028420_CB_BLEND_ALPHA, fui(state->color[3]));
398
399 si_pm4_set_state(sctx, blend_color, pm4);
400 }
401
402 /*
403 * Clipping, scissors and viewport
404 */
405
406 static void si_set_clip_state(struct pipe_context *ctx,
407 const struct pipe_clip_state *state)
408 {
409 struct si_context *sctx = (struct si_context *)ctx;
410 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
411 struct pipe_constant_buffer cb;
412
413 if (pm4 == NULL)
414 return;
415
416 for (int i = 0; i < 6; i++) {
417 si_pm4_set_reg(pm4, R_0285BC_PA_CL_UCP_0_X + i * 16,
418 fui(state->ucp[i][0]));
419 si_pm4_set_reg(pm4, R_0285C0_PA_CL_UCP_0_Y + i * 16,
420 fui(state->ucp[i][1]));
421 si_pm4_set_reg(pm4, R_0285C4_PA_CL_UCP_0_Z + i * 16,
422 fui(state->ucp[i][2]));
423 si_pm4_set_reg(pm4, R_0285C8_PA_CL_UCP_0_W + i * 16,
424 fui(state->ucp[i][3]));
425 }
426
427 cb.buffer = NULL;
428 cb.user_buffer = state->ucp;
429 cb.buffer_offset = 0;
430 cb.buffer_size = 4*4*8;
431 ctx->set_constant_buffer(ctx, PIPE_SHADER_VERTEX, NUM_PIPE_CONST_BUFFERS, &cb);
432 pipe_resource_reference(&cb.buffer, NULL);
433
434 si_pm4_set_state(sctx, clip, pm4);
435 }
436
437 static void si_set_scissor_states(struct pipe_context *ctx,
438 unsigned start_slot,
439 unsigned num_scissors,
440 const struct pipe_scissor_state *state)
441 {
442 struct si_context *sctx = (struct si_context *)ctx;
443 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
444
445 if (pm4 == NULL)
446 return;
447
448 si_pm4_set_reg(pm4, R_028250_PA_SC_VPORT_SCISSOR_0_TL,
449 S_028250_TL_X(state->minx) | S_028250_TL_Y(state->miny) |
450 S_028250_WINDOW_OFFSET_DISABLE(1));
451 si_pm4_set_reg(pm4, R_028254_PA_SC_VPORT_SCISSOR_0_BR,
452 S_028254_BR_X(state->maxx) | S_028254_BR_Y(state->maxy));
453
454 si_pm4_set_state(sctx, scissor, pm4);
455 }
456
457 static void si_set_viewport_states(struct pipe_context *ctx,
458 unsigned start_slot,
459 unsigned num_viewports,
460 const struct pipe_viewport_state *state)
461 {
462 struct si_context *sctx = (struct si_context *)ctx;
463 struct si_state_viewport *viewport = CALLOC_STRUCT(si_state_viewport);
464 struct si_pm4_state *pm4 = &viewport->pm4;
465
466 if (viewport == NULL)
467 return;
468
469 viewport->viewport = *state;
470 si_pm4_set_reg(pm4, R_02843C_PA_CL_VPORT_XSCALE_0, fui(state->scale[0]));
471 si_pm4_set_reg(pm4, R_028440_PA_CL_VPORT_XOFFSET_0, fui(state->translate[0]));
472 si_pm4_set_reg(pm4, R_028444_PA_CL_VPORT_YSCALE_0, fui(state->scale[1]));
473 si_pm4_set_reg(pm4, R_028448_PA_CL_VPORT_YOFFSET_0, fui(state->translate[1]));
474 si_pm4_set_reg(pm4, R_02844C_PA_CL_VPORT_ZSCALE_0, fui(state->scale[2]));
475 si_pm4_set_reg(pm4, R_028450_PA_CL_VPORT_ZOFFSET_0, fui(state->translate[2]));
476
477 si_pm4_set_state(sctx, viewport, viewport);
478 }
479
480 /*
481 * inferred state between framebuffer and rasterizer
482 */
483 static void si_update_fb_rs_state(struct si_context *sctx)
484 {
485 struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
486 struct si_pm4_state *pm4;
487 unsigned offset_db_fmt_cntl = 0, depth;
488 float offset_units;
489
490 if (!rs || !sctx->framebuffer.zsbuf)
491 return;
492
493 offset_units = sctx->queued.named.rasterizer->offset_units;
494 switch (sctx->framebuffer.zsbuf->texture->format) {
495 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
496 case PIPE_FORMAT_X8Z24_UNORM:
497 case PIPE_FORMAT_Z24X8_UNORM:
498 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
499 depth = -24;
500 offset_units *= 2.0f;
501 break;
502 case PIPE_FORMAT_Z32_FLOAT:
503 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
504 depth = -23;
505 offset_units *= 1.0f;
506 offset_db_fmt_cntl |= S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
507 break;
508 case PIPE_FORMAT_Z16_UNORM:
509 depth = -16;
510 offset_units *= 4.0f;
511 break;
512 default:
513 return;
514 }
515
516 pm4 = si_pm4_alloc_state(sctx);
517
518 if (pm4 == NULL)
519 return;
520
521 /* FIXME some of those reg can be computed with cso */
522 offset_db_fmt_cntl |= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(depth);
523 si_pm4_set_reg(pm4, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE,
524 fui(sctx->queued.named.rasterizer->offset_scale));
525 si_pm4_set_reg(pm4, R_028B84_PA_SU_POLY_OFFSET_FRONT_OFFSET, fui(offset_units));
526 si_pm4_set_reg(pm4, R_028B88_PA_SU_POLY_OFFSET_BACK_SCALE,
527 fui(sctx->queued.named.rasterizer->offset_scale));
528 si_pm4_set_reg(pm4, R_028B8C_PA_SU_POLY_OFFSET_BACK_OFFSET, fui(offset_units));
529 si_pm4_set_reg(pm4, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL, offset_db_fmt_cntl);
530
531 si_pm4_set_state(sctx, fb_rs, pm4);
532 }
533
534 /*
535 * Rasterizer
536 */
537
538 static uint32_t si_translate_fill(uint32_t func)
539 {
540 switch(func) {
541 case PIPE_POLYGON_MODE_FILL:
542 return V_028814_X_DRAW_TRIANGLES;
543 case PIPE_POLYGON_MODE_LINE:
544 return V_028814_X_DRAW_LINES;
545 case PIPE_POLYGON_MODE_POINT:
546 return V_028814_X_DRAW_POINTS;
547 default:
548 assert(0);
549 return V_028814_X_DRAW_POINTS;
550 }
551 }
552
553 static void *si_create_rs_state(struct pipe_context *ctx,
554 const struct pipe_rasterizer_state *state)
555 {
556 struct si_state_rasterizer *rs = CALLOC_STRUCT(si_state_rasterizer);
557 struct si_pm4_state *pm4 = &rs->pm4;
558 unsigned tmp;
559 unsigned prov_vtx = 1, polygon_dual_mode;
560 float psize_min, psize_max;
561
562 if (rs == NULL) {
563 return NULL;
564 }
565
566 rs->two_side = state->light_twoside;
567 rs->multisample_enable = state->multisample;
568 rs->clip_plane_enable = state->clip_plane_enable;
569 rs->line_stipple_enable = state->line_stipple_enable;
570
571 polygon_dual_mode = (state->fill_front != PIPE_POLYGON_MODE_FILL ||
572 state->fill_back != PIPE_POLYGON_MODE_FILL);
573
574 if (state->flatshade_first)
575 prov_vtx = 0;
576
577 rs->flatshade = state->flatshade;
578 rs->sprite_coord_enable = state->sprite_coord_enable;
579 rs->pa_sc_line_stipple = state->line_stipple_enable ?
580 S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
581 S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
582 rs->pa_su_sc_mode_cntl =
583 S_028814_PROVOKING_VTX_LAST(prov_vtx) |
584 S_028814_CULL_FRONT(state->rasterizer_discard || (state->cull_face & PIPE_FACE_FRONT) ? 1 : 0) |
585 S_028814_CULL_BACK(state->rasterizer_discard || (state->cull_face & PIPE_FACE_BACK) ? 1 : 0) |
586 S_028814_FACE(!state->front_ccw) |
587 S_028814_POLY_OFFSET_FRONT_ENABLE(state->offset_tri) |
588 S_028814_POLY_OFFSET_BACK_ENABLE(state->offset_tri) |
589 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_tri) |
590 S_028814_POLY_MODE(polygon_dual_mode) |
591 S_028814_POLYMODE_FRONT_PTYPE(si_translate_fill(state->fill_front)) |
592 S_028814_POLYMODE_BACK_PTYPE(si_translate_fill(state->fill_back));
593 rs->pa_cl_clip_cntl =
594 S_028810_PS_UCP_MODE(3) |
595 S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
596 S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
597 S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard) |
598 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
599
600 /* offset */
601 rs->offset_units = state->offset_units;
602 rs->offset_scale = state->offset_scale * 12.0f;
603
604 tmp = S_0286D4_FLAT_SHADE_ENA(1);
605 if (state->sprite_coord_enable) {
606 tmp |= S_0286D4_PNT_SPRITE_ENA(1) |
607 S_0286D4_PNT_SPRITE_OVRD_X(V_0286D4_SPI_PNT_SPRITE_SEL_S) |
608 S_0286D4_PNT_SPRITE_OVRD_Y(V_0286D4_SPI_PNT_SPRITE_SEL_T) |
609 S_0286D4_PNT_SPRITE_OVRD_Z(V_0286D4_SPI_PNT_SPRITE_SEL_0) |
610 S_0286D4_PNT_SPRITE_OVRD_W(V_0286D4_SPI_PNT_SPRITE_SEL_1);
611 if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) {
612 tmp |= S_0286D4_PNT_SPRITE_TOP_1(1);
613 }
614 }
615 si_pm4_set_reg(pm4, R_0286D4_SPI_INTERP_CONTROL_0, tmp);
616
617 /* point size 12.4 fixed point */
618 tmp = (unsigned)(state->point_size * 8.0);
619 si_pm4_set_reg(pm4, R_028A00_PA_SU_POINT_SIZE, S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
620
621 if (state->point_size_per_vertex) {
622 psize_min = util_get_min_point_size(state);
623 psize_max = 8192;
624 } else {
625 /* Force the point size to be as if the vertex output was disabled. */
626 psize_min = state->point_size;
627 psize_max = state->point_size;
628 }
629 /* Divide by two, because 0.5 = 1 pixel. */
630 si_pm4_set_reg(pm4, R_028A04_PA_SU_POINT_MINMAX,
631 S_028A04_MIN_SIZE(si_pack_float_12p4(psize_min/2)) |
632 S_028A04_MAX_SIZE(si_pack_float_12p4(psize_max/2)));
633
634 tmp = (unsigned)state->line_width * 8;
635 si_pm4_set_reg(pm4, R_028A08_PA_SU_LINE_CNTL, S_028A08_WIDTH(tmp));
636 si_pm4_set_reg(pm4, R_028A48_PA_SC_MODE_CNTL_0,
637 S_028A48_LINE_STIPPLE_ENABLE(state->line_stipple_enable) |
638 S_028A48_MSAA_ENABLE(state->multisample) |
639 S_028A48_VPORT_SCISSOR_ENABLE(state->scissor));
640
641 si_pm4_set_reg(pm4, R_028BE4_PA_SU_VTX_CNTL,
642 S_028BE4_PIX_CENTER(state->half_pixel_center) |
643 S_028BE4_QUANT_MODE(V_028BE4_X_16_8_FIXED_POINT_1_256TH));
644
645 si_pm4_set_reg(pm4, R_028B7C_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
646
647 return rs;
648 }
649
650 static void si_bind_rs_state(struct pipe_context *ctx, void *state)
651 {
652 struct si_context *sctx = (struct si_context *)ctx;
653 struct si_state_rasterizer *rs = (struct si_state_rasterizer *)state;
654
655 if (state == NULL)
656 return;
657
658 // TODO
659 sctx->sprite_coord_enable = rs->sprite_coord_enable;
660 sctx->pa_sc_line_stipple = rs->pa_sc_line_stipple;
661 sctx->pa_su_sc_mode_cntl = rs->pa_su_sc_mode_cntl;
662
663 si_pm4_bind_state(sctx, rasterizer, rs);
664 si_update_fb_rs_state(sctx);
665 }
666
667 static void si_delete_rs_state(struct pipe_context *ctx, void *state)
668 {
669 struct si_context *sctx = (struct si_context *)ctx;
670 si_pm4_delete_state(sctx, rasterizer, (struct si_state_rasterizer *)state);
671 }
672
673 /*
674 * infeered state between dsa and stencil ref
675 */
676 static void si_update_dsa_stencil_ref(struct si_context *sctx)
677 {
678 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
679 struct pipe_stencil_ref *ref = &sctx->stencil_ref;
680 struct si_state_dsa *dsa = sctx->queued.named.dsa;
681
682 if (pm4 == NULL)
683 return;
684
685 si_pm4_set_reg(pm4, R_028430_DB_STENCILREFMASK,
686 S_028430_STENCILTESTVAL(ref->ref_value[0]) |
687 S_028430_STENCILMASK(dsa->valuemask[0]) |
688 S_028430_STENCILWRITEMASK(dsa->writemask[0]) |
689 S_028430_STENCILOPVAL(1));
690 si_pm4_set_reg(pm4, R_028434_DB_STENCILREFMASK_BF,
691 S_028434_STENCILTESTVAL_BF(ref->ref_value[1]) |
692 S_028434_STENCILMASK_BF(dsa->valuemask[1]) |
693 S_028434_STENCILWRITEMASK_BF(dsa->writemask[1]) |
694 S_028434_STENCILOPVAL_BF(1));
695
696 si_pm4_set_state(sctx, dsa_stencil_ref, pm4);
697 }
698
699 static void si_set_pipe_stencil_ref(struct pipe_context *ctx,
700 const struct pipe_stencil_ref *state)
701 {
702 struct si_context *sctx = (struct si_context *)ctx;
703 sctx->stencil_ref = *state;
704 si_update_dsa_stencil_ref(sctx);
705 }
706
707
708 /*
709 * DSA
710 */
711
712 static uint32_t si_translate_stencil_op(int s_op)
713 {
714 switch (s_op) {
715 case PIPE_STENCIL_OP_KEEP:
716 return V_02842C_STENCIL_KEEP;
717 case PIPE_STENCIL_OP_ZERO:
718 return V_02842C_STENCIL_ZERO;
719 case PIPE_STENCIL_OP_REPLACE:
720 return V_02842C_STENCIL_REPLACE_TEST;
721 case PIPE_STENCIL_OP_INCR:
722 return V_02842C_STENCIL_ADD_CLAMP;
723 case PIPE_STENCIL_OP_DECR:
724 return V_02842C_STENCIL_SUB_CLAMP;
725 case PIPE_STENCIL_OP_INCR_WRAP:
726 return V_02842C_STENCIL_ADD_WRAP;
727 case PIPE_STENCIL_OP_DECR_WRAP:
728 return V_02842C_STENCIL_SUB_WRAP;
729 case PIPE_STENCIL_OP_INVERT:
730 return V_02842C_STENCIL_INVERT;
731 default:
732 R600_ERR("Unknown stencil op %d", s_op);
733 assert(0);
734 break;
735 }
736 return 0;
737 }
738
739 static void *si_create_dsa_state(struct pipe_context *ctx,
740 const struct pipe_depth_stencil_alpha_state *state)
741 {
742 struct si_state_dsa *dsa = CALLOC_STRUCT(si_state_dsa);
743 struct si_pm4_state *pm4 = &dsa->pm4;
744 unsigned db_depth_control;
745 unsigned db_render_control;
746 uint32_t db_stencil_control = 0;
747
748 if (dsa == NULL) {
749 return NULL;
750 }
751
752 dsa->valuemask[0] = state->stencil[0].valuemask;
753 dsa->valuemask[1] = state->stencil[1].valuemask;
754 dsa->writemask[0] = state->stencil[0].writemask;
755 dsa->writemask[1] = state->stencil[1].writemask;
756
757 db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) |
758 S_028800_Z_WRITE_ENABLE(state->depth.writemask) |
759 S_028800_ZFUNC(state->depth.func);
760
761 /* stencil */
762 if (state->stencil[0].enabled) {
763 db_depth_control |= S_028800_STENCIL_ENABLE(1);
764 db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func);
765 db_stencil_control |= S_02842C_STENCILFAIL(si_translate_stencil_op(state->stencil[0].fail_op));
766 db_stencil_control |= S_02842C_STENCILZPASS(si_translate_stencil_op(state->stencil[0].zpass_op));
767 db_stencil_control |= S_02842C_STENCILZFAIL(si_translate_stencil_op(state->stencil[0].zfail_op));
768
769 if (state->stencil[1].enabled) {
770 db_depth_control |= S_028800_BACKFACE_ENABLE(1);
771 db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func);
772 db_stencil_control |= S_02842C_STENCILFAIL_BF(si_translate_stencil_op(state->stencil[1].fail_op));
773 db_stencil_control |= S_02842C_STENCILZPASS_BF(si_translate_stencil_op(state->stencil[1].zpass_op));
774 db_stencil_control |= S_02842C_STENCILZFAIL_BF(si_translate_stencil_op(state->stencil[1].zfail_op));
775 }
776 }
777
778 /* alpha */
779 if (state->alpha.enabled) {
780 dsa->alpha_func = state->alpha.func;
781 dsa->alpha_ref = state->alpha.ref_value;
782
783 si_pm4_set_reg(pm4, R_00B030_SPI_SHADER_USER_DATA_PS_0 +
784 SI_SGPR_ALPHA_REF * 4, fui(dsa->alpha_ref));
785 } else {
786 dsa->alpha_func = PIPE_FUNC_ALWAYS;
787 }
788
789 /* misc */
790 db_render_control = 0;
791 si_pm4_set_reg(pm4, R_028800_DB_DEPTH_CONTROL, db_depth_control);
792 si_pm4_set_reg(pm4, R_028000_DB_RENDER_CONTROL, db_render_control);
793 si_pm4_set_reg(pm4, R_02842C_DB_STENCIL_CONTROL, db_stencil_control);
794
795 return dsa;
796 }
797
798 static void si_bind_dsa_state(struct pipe_context *ctx, void *state)
799 {
800 struct si_context *sctx = (struct si_context *)ctx;
801 struct si_state_dsa *dsa = state;
802
803 if (state == NULL)
804 return;
805
806 si_pm4_bind_state(sctx, dsa, dsa);
807 si_update_dsa_stencil_ref(sctx);
808 }
809
810 static void si_delete_dsa_state(struct pipe_context *ctx, void *state)
811 {
812 struct si_context *sctx = (struct si_context *)ctx;
813 si_pm4_delete_state(sctx, dsa, (struct si_state_dsa *)state);
814 }
815
816 static void *si_create_db_flush_dsa(struct si_context *sctx, bool copy_depth,
817 bool copy_stencil, int sample)
818 {
819 struct pipe_depth_stencil_alpha_state dsa;
820 struct si_state_dsa *state;
821
822 memset(&dsa, 0, sizeof(dsa));
823
824 state = sctx->b.b.create_depth_stencil_alpha_state(&sctx->b.b, &dsa);
825 if (copy_depth || copy_stencil) {
826 si_pm4_set_reg(&state->pm4, R_028000_DB_RENDER_CONTROL,
827 S_028000_DEPTH_COPY(copy_depth) |
828 S_028000_STENCIL_COPY(copy_stencil) |
829 S_028000_COPY_CENTROID(1) |
830 S_028000_COPY_SAMPLE(sample));
831 } else {
832 si_pm4_set_reg(&state->pm4, R_028000_DB_RENDER_CONTROL,
833 S_028000_DEPTH_COMPRESS_DISABLE(1) |
834 S_028000_STENCIL_COMPRESS_DISABLE(1));
835 }
836
837 return state;
838 }
839
840 /*
841 * format translation
842 */
843 static uint32_t si_translate_colorformat(enum pipe_format format)
844 {
845 const struct util_format_description *desc = util_format_description(format);
846
847 #define HAS_SIZE(x,y,z,w) \
848 (desc->channel[0].size == (x) && desc->channel[1].size == (y) && \
849 desc->channel[2].size == (z) && desc->channel[3].size == (w))
850
851 if (format == PIPE_FORMAT_R11G11B10_FLOAT) /* isn't plain */
852 return V_028C70_COLOR_10_11_11;
853
854 if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
855 return V_028C70_COLOR_INVALID;
856
857 switch (desc->nr_channels) {
858 case 1:
859 switch (desc->channel[0].size) {
860 case 8:
861 return V_028C70_COLOR_8;
862 case 16:
863 return V_028C70_COLOR_16;
864 case 32:
865 return V_028C70_COLOR_32;
866 }
867 break;
868 case 2:
869 if (desc->channel[0].size == desc->channel[1].size) {
870 switch (desc->channel[0].size) {
871 case 8:
872 return V_028C70_COLOR_8_8;
873 case 16:
874 return V_028C70_COLOR_16_16;
875 case 32:
876 return V_028C70_COLOR_32_32;
877 }
878 } else if (HAS_SIZE(8,24,0,0)) {
879 return V_028C70_COLOR_24_8;
880 } else if (HAS_SIZE(24,8,0,0)) {
881 return V_028C70_COLOR_8_24;
882 }
883 break;
884 case 3:
885 if (HAS_SIZE(5,6,5,0)) {
886 return V_028C70_COLOR_5_6_5;
887 } else if (HAS_SIZE(32,8,24,0)) {
888 return V_028C70_COLOR_X24_8_32_FLOAT;
889 }
890 break;
891 case 4:
892 if (desc->channel[0].size == desc->channel[1].size &&
893 desc->channel[0].size == desc->channel[2].size &&
894 desc->channel[0].size == desc->channel[3].size) {
895 switch (desc->channel[0].size) {
896 case 4:
897 return V_028C70_COLOR_4_4_4_4;
898 case 8:
899 return V_028C70_COLOR_8_8_8_8;
900 case 16:
901 return V_028C70_COLOR_16_16_16_16;
902 case 32:
903 return V_028C70_COLOR_32_32_32_32;
904 }
905 } else if (HAS_SIZE(5,5,5,1)) {
906 return V_028C70_COLOR_1_5_5_5;
907 } else if (HAS_SIZE(10,10,10,2)) {
908 return V_028C70_COLOR_2_10_10_10;
909 }
910 break;
911 }
912 return V_028C70_COLOR_INVALID;
913 }
914
915 static uint32_t si_translate_colorswap(enum pipe_format format)
916 {
917 const struct util_format_description *desc = util_format_description(format);
918
919 #define HAS_SWIZZLE(chan,swz) (desc->swizzle[chan] == UTIL_FORMAT_SWIZZLE_##swz)
920
921 if (format == PIPE_FORMAT_R11G11B10_FLOAT) /* isn't plain */
922 return V_028C70_SWAP_STD;
923
924 if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
925 return ~0;
926
927 switch (desc->nr_channels) {
928 case 1:
929 if (HAS_SWIZZLE(0,X))
930 return V_028C70_SWAP_STD; /* X___ */
931 else if (HAS_SWIZZLE(3,X))
932 return V_028C70_SWAP_ALT_REV; /* ___X */
933 break;
934 case 2:
935 if ((HAS_SWIZZLE(0,X) && HAS_SWIZZLE(1,Y)) ||
936 (HAS_SWIZZLE(0,X) && HAS_SWIZZLE(1,NONE)) ||
937 (HAS_SWIZZLE(0,NONE) && HAS_SWIZZLE(1,Y)))
938 return V_028C70_SWAP_STD; /* XY__ */
939 else if ((HAS_SWIZZLE(0,Y) && HAS_SWIZZLE(1,X)) ||
940 (HAS_SWIZZLE(0,Y) && HAS_SWIZZLE(1,NONE)) ||
941 (HAS_SWIZZLE(0,NONE) && HAS_SWIZZLE(1,X)))
942 return V_028C70_SWAP_STD_REV; /* YX__ */
943 else if (HAS_SWIZZLE(0,X) && HAS_SWIZZLE(3,Y))
944 return V_028C70_SWAP_ALT; /* X__Y */
945 break;
946 case 3:
947 if (HAS_SWIZZLE(0,X))
948 return V_028C70_SWAP_STD; /* XYZ */
949 else if (HAS_SWIZZLE(0,Z))
950 return V_028C70_SWAP_STD_REV; /* ZYX */
951 break;
952 case 4:
953 /* check the middle channels, the 1st and 4th channel can be NONE */
954 if (HAS_SWIZZLE(1,Y) && HAS_SWIZZLE(2,Z))
955 return V_028C70_SWAP_STD; /* XYZW */
956 else if (HAS_SWIZZLE(1,Z) && HAS_SWIZZLE(2,Y))
957 return V_028C70_SWAP_STD_REV; /* WZYX */
958 else if (HAS_SWIZZLE(1,Y) && HAS_SWIZZLE(2,X))
959 return V_028C70_SWAP_ALT; /* ZYXW */
960 else if (HAS_SWIZZLE(1,X) && HAS_SWIZZLE(2,Y))
961 return V_028C70_SWAP_ALT_REV; /* WXYZ */
962 break;
963 }
964 return ~0U;
965 }
966
967 static uint32_t si_colorformat_endian_swap(uint32_t colorformat)
968 {
969 if (SI_BIG_ENDIAN) {
970 switch(colorformat) {
971 /* 8-bit buffers. */
972 case V_028C70_COLOR_8:
973 return V_028C70_ENDIAN_NONE;
974
975 /* 16-bit buffers. */
976 case V_028C70_COLOR_5_6_5:
977 case V_028C70_COLOR_1_5_5_5:
978 case V_028C70_COLOR_4_4_4_4:
979 case V_028C70_COLOR_16:
980 case V_028C70_COLOR_8_8:
981 return V_028C70_ENDIAN_8IN16;
982
983 /* 32-bit buffers. */
984 case V_028C70_COLOR_8_8_8_8:
985 case V_028C70_COLOR_2_10_10_10:
986 case V_028C70_COLOR_8_24:
987 case V_028C70_COLOR_24_8:
988 case V_028C70_COLOR_16_16:
989 return V_028C70_ENDIAN_8IN32;
990
991 /* 64-bit buffers. */
992 case V_028C70_COLOR_16_16_16_16:
993 return V_028C70_ENDIAN_8IN16;
994
995 case V_028C70_COLOR_32_32:
996 return V_028C70_ENDIAN_8IN32;
997
998 /* 128-bit buffers. */
999 case V_028C70_COLOR_32_32_32_32:
1000 return V_028C70_ENDIAN_8IN32;
1001 default:
1002 return V_028C70_ENDIAN_NONE; /* Unsupported. */
1003 }
1004 } else {
1005 return V_028C70_ENDIAN_NONE;
1006 }
1007 }
1008
1009 /* Returns the size in bits of the widest component of a CB format */
1010 static unsigned si_colorformat_max_comp_size(uint32_t colorformat)
1011 {
1012 switch(colorformat) {
1013 case V_028C70_COLOR_4_4_4_4:
1014 return 4;
1015
1016 case V_028C70_COLOR_1_5_5_5:
1017 case V_028C70_COLOR_5_5_5_1:
1018 return 5;
1019
1020 case V_028C70_COLOR_5_6_5:
1021 return 6;
1022
1023 case V_028C70_COLOR_8:
1024 case V_028C70_COLOR_8_8:
1025 case V_028C70_COLOR_8_8_8_8:
1026 return 8;
1027
1028 case V_028C70_COLOR_10_10_10_2:
1029 case V_028C70_COLOR_2_10_10_10:
1030 return 10;
1031
1032 case V_028C70_COLOR_10_11_11:
1033 case V_028C70_COLOR_11_11_10:
1034 return 11;
1035
1036 case V_028C70_COLOR_16:
1037 case V_028C70_COLOR_16_16:
1038 case V_028C70_COLOR_16_16_16_16:
1039 return 16;
1040
1041 case V_028C70_COLOR_8_24:
1042 case V_028C70_COLOR_24_8:
1043 return 24;
1044
1045 case V_028C70_COLOR_32:
1046 case V_028C70_COLOR_32_32:
1047 case V_028C70_COLOR_32_32_32_32:
1048 case V_028C70_COLOR_X24_8_32_FLOAT:
1049 return 32;
1050 }
1051
1052 assert(!"Unknown maximum component size");
1053 return 0;
1054 }
1055
1056 static uint32_t si_translate_dbformat(enum pipe_format format)
1057 {
1058 switch (format) {
1059 case PIPE_FORMAT_Z16_UNORM:
1060 return V_028040_Z_16;
1061 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1062 case PIPE_FORMAT_X8Z24_UNORM:
1063 case PIPE_FORMAT_Z24X8_UNORM:
1064 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1065 return V_028040_Z_24; /* deprecated on SI */
1066 case PIPE_FORMAT_Z32_FLOAT:
1067 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1068 return V_028040_Z_32_FLOAT;
1069 default:
1070 return V_028040_Z_INVALID;
1071 }
1072 }
1073
1074 /*
1075 * Texture translation
1076 */
1077
1078 static uint32_t si_translate_texformat(struct pipe_screen *screen,
1079 enum pipe_format format,
1080 const struct util_format_description *desc,
1081 int first_non_void)
1082 {
1083 struct si_screen *sscreen = (struct si_screen*)screen;
1084 bool enable_s3tc = sscreen->b.info.drm_minor >= 31;
1085 boolean uniform = TRUE;
1086 int i;
1087
1088 /* Colorspace (return non-RGB formats directly). */
1089 switch (desc->colorspace) {
1090 /* Depth stencil formats */
1091 case UTIL_FORMAT_COLORSPACE_ZS:
1092 switch (format) {
1093 case PIPE_FORMAT_Z16_UNORM:
1094 return V_008F14_IMG_DATA_FORMAT_16;
1095 case PIPE_FORMAT_X24S8_UINT:
1096 case PIPE_FORMAT_Z24X8_UNORM:
1097 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
1098 return V_008F14_IMG_DATA_FORMAT_8_24;
1099 case PIPE_FORMAT_X8Z24_UNORM:
1100 case PIPE_FORMAT_S8X24_UINT:
1101 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
1102 return V_008F14_IMG_DATA_FORMAT_24_8;
1103 case PIPE_FORMAT_S8_UINT:
1104 return V_008F14_IMG_DATA_FORMAT_8;
1105 case PIPE_FORMAT_Z32_FLOAT:
1106 return V_008F14_IMG_DATA_FORMAT_32;
1107 case PIPE_FORMAT_X32_S8X24_UINT:
1108 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
1109 return V_008F14_IMG_DATA_FORMAT_X24_8_32;
1110 default:
1111 goto out_unknown;
1112 }
1113
1114 case UTIL_FORMAT_COLORSPACE_YUV:
1115 goto out_unknown; /* TODO */
1116
1117 case UTIL_FORMAT_COLORSPACE_SRGB:
1118 if (desc->nr_channels != 4 && desc->nr_channels != 1)
1119 goto out_unknown;
1120 break;
1121
1122 default:
1123 break;
1124 }
1125
1126 if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
1127 if (!enable_s3tc)
1128 goto out_unknown;
1129
1130 switch (format) {
1131 case PIPE_FORMAT_RGTC1_SNORM:
1132 case PIPE_FORMAT_LATC1_SNORM:
1133 case PIPE_FORMAT_RGTC1_UNORM:
1134 case PIPE_FORMAT_LATC1_UNORM:
1135 return V_008F14_IMG_DATA_FORMAT_BC4;
1136 case PIPE_FORMAT_RGTC2_SNORM:
1137 case PIPE_FORMAT_LATC2_SNORM:
1138 case PIPE_FORMAT_RGTC2_UNORM:
1139 case PIPE_FORMAT_LATC2_UNORM:
1140 return V_008F14_IMG_DATA_FORMAT_BC5;
1141 default:
1142 goto out_unknown;
1143 }
1144 }
1145
1146 if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
1147
1148 if (!enable_s3tc)
1149 goto out_unknown;
1150
1151 if (!util_format_s3tc_enabled) {
1152 goto out_unknown;
1153 }
1154
1155 switch (format) {
1156 case PIPE_FORMAT_DXT1_RGB:
1157 case PIPE_FORMAT_DXT1_RGBA:
1158 case PIPE_FORMAT_DXT1_SRGB:
1159 case PIPE_FORMAT_DXT1_SRGBA:
1160 return V_008F14_IMG_DATA_FORMAT_BC1;
1161 case PIPE_FORMAT_DXT3_RGBA:
1162 case PIPE_FORMAT_DXT3_SRGBA:
1163 return V_008F14_IMG_DATA_FORMAT_BC2;
1164 case PIPE_FORMAT_DXT5_RGBA:
1165 case PIPE_FORMAT_DXT5_SRGBA:
1166 return V_008F14_IMG_DATA_FORMAT_BC3;
1167 default:
1168 goto out_unknown;
1169 }
1170 }
1171
1172 if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
1173 return V_008F14_IMG_DATA_FORMAT_5_9_9_9;
1174 } else if (format == PIPE_FORMAT_R11G11B10_FLOAT) {
1175 return V_008F14_IMG_DATA_FORMAT_10_11_11;
1176 }
1177
1178 /* R8G8Bx_SNORM - TODO CxV8U8 */
1179
1180 /* See whether the components are of the same size. */
1181 for (i = 1; i < desc->nr_channels; i++) {
1182 uniform = uniform && desc->channel[0].size == desc->channel[i].size;
1183 }
1184
1185 /* Non-uniform formats. */
1186 if (!uniform) {
1187 switch(desc->nr_channels) {
1188 case 3:
1189 if (desc->channel[0].size == 5 &&
1190 desc->channel[1].size == 6 &&
1191 desc->channel[2].size == 5) {
1192 return V_008F14_IMG_DATA_FORMAT_5_6_5;
1193 }
1194 goto out_unknown;
1195 case 4:
1196 if (desc->channel[0].size == 5 &&
1197 desc->channel[1].size == 5 &&
1198 desc->channel[2].size == 5 &&
1199 desc->channel[3].size == 1) {
1200 return V_008F14_IMG_DATA_FORMAT_1_5_5_5;
1201 }
1202 if (desc->channel[0].size == 10 &&
1203 desc->channel[1].size == 10 &&
1204 desc->channel[2].size == 10 &&
1205 desc->channel[3].size == 2) {
1206 return V_008F14_IMG_DATA_FORMAT_2_10_10_10;
1207 }
1208 goto out_unknown;
1209 }
1210 goto out_unknown;
1211 }
1212
1213 if (first_non_void < 0 || first_non_void > 3)
1214 goto out_unknown;
1215
1216 /* uniform formats */
1217 switch (desc->channel[first_non_void].size) {
1218 case 4:
1219 switch (desc->nr_channels) {
1220 #if 0 /* Not supported for render targets */
1221 case 2:
1222 return V_008F14_IMG_DATA_FORMAT_4_4;
1223 #endif
1224 case 4:
1225 return V_008F14_IMG_DATA_FORMAT_4_4_4_4;
1226 }
1227 break;
1228 case 8:
1229 switch (desc->nr_channels) {
1230 case 1:
1231 return V_008F14_IMG_DATA_FORMAT_8;
1232 case 2:
1233 return V_008F14_IMG_DATA_FORMAT_8_8;
1234 case 4:
1235 return V_008F14_IMG_DATA_FORMAT_8_8_8_8;
1236 }
1237 break;
1238 case 16:
1239 switch (desc->nr_channels) {
1240 case 1:
1241 return V_008F14_IMG_DATA_FORMAT_16;
1242 case 2:
1243 return V_008F14_IMG_DATA_FORMAT_16_16;
1244 case 4:
1245 return V_008F14_IMG_DATA_FORMAT_16_16_16_16;
1246 }
1247 break;
1248 case 32:
1249 switch (desc->nr_channels) {
1250 case 1:
1251 return V_008F14_IMG_DATA_FORMAT_32;
1252 case 2:
1253 return V_008F14_IMG_DATA_FORMAT_32_32;
1254 #if 0 /* Not supported for render targets */
1255 case 3:
1256 return V_008F14_IMG_DATA_FORMAT_32_32_32;
1257 #endif
1258 case 4:
1259 return V_008F14_IMG_DATA_FORMAT_32_32_32_32;
1260 }
1261 }
1262
1263 out_unknown:
1264 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1265 return ~0;
1266 }
1267
1268 static unsigned si_tex_wrap(unsigned wrap)
1269 {
1270 switch (wrap) {
1271 default:
1272 case PIPE_TEX_WRAP_REPEAT:
1273 return V_008F30_SQ_TEX_WRAP;
1274 case PIPE_TEX_WRAP_CLAMP:
1275 return V_008F30_SQ_TEX_CLAMP_HALF_BORDER;
1276 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1277 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL;
1278 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1279 return V_008F30_SQ_TEX_CLAMP_BORDER;
1280 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1281 return V_008F30_SQ_TEX_MIRROR;
1282 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1283 return V_008F30_SQ_TEX_MIRROR_ONCE_HALF_BORDER;
1284 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1285 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL;
1286 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1287 return V_008F30_SQ_TEX_MIRROR_ONCE_BORDER;
1288 }
1289 }
1290
1291 static unsigned si_tex_filter(unsigned filter)
1292 {
1293 switch (filter) {
1294 default:
1295 case PIPE_TEX_FILTER_NEAREST:
1296 return V_008F38_SQ_TEX_XY_FILTER_POINT;
1297 case PIPE_TEX_FILTER_LINEAR:
1298 return V_008F38_SQ_TEX_XY_FILTER_BILINEAR;
1299 }
1300 }
1301
1302 static unsigned si_tex_mipfilter(unsigned filter)
1303 {
1304 switch (filter) {
1305 case PIPE_TEX_MIPFILTER_NEAREST:
1306 return V_008F38_SQ_TEX_Z_FILTER_POINT;
1307 case PIPE_TEX_MIPFILTER_LINEAR:
1308 return V_008F38_SQ_TEX_Z_FILTER_LINEAR;
1309 default:
1310 case PIPE_TEX_MIPFILTER_NONE:
1311 return V_008F38_SQ_TEX_Z_FILTER_NONE;
1312 }
1313 }
1314
1315 static unsigned si_tex_compare(unsigned compare)
1316 {
1317 switch (compare) {
1318 default:
1319 case PIPE_FUNC_NEVER:
1320 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER;
1321 case PIPE_FUNC_LESS:
1322 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS;
1323 case PIPE_FUNC_EQUAL:
1324 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL;
1325 case PIPE_FUNC_LEQUAL:
1326 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL;
1327 case PIPE_FUNC_GREATER:
1328 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER;
1329 case PIPE_FUNC_NOTEQUAL:
1330 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL;
1331 case PIPE_FUNC_GEQUAL:
1332 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL;
1333 case PIPE_FUNC_ALWAYS:
1334 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS;
1335 }
1336 }
1337
1338 static unsigned si_tex_dim(unsigned dim, unsigned nr_samples)
1339 {
1340 switch (dim) {
1341 default:
1342 case PIPE_TEXTURE_1D:
1343 return V_008F1C_SQ_RSRC_IMG_1D;
1344 case PIPE_TEXTURE_1D_ARRAY:
1345 return V_008F1C_SQ_RSRC_IMG_1D_ARRAY;
1346 case PIPE_TEXTURE_2D:
1347 case PIPE_TEXTURE_RECT:
1348 return nr_samples > 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA :
1349 V_008F1C_SQ_RSRC_IMG_2D;
1350 case PIPE_TEXTURE_2D_ARRAY:
1351 return nr_samples > 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY :
1352 V_008F1C_SQ_RSRC_IMG_2D_ARRAY;
1353 case PIPE_TEXTURE_3D:
1354 return V_008F1C_SQ_RSRC_IMG_3D;
1355 case PIPE_TEXTURE_CUBE:
1356 return V_008F1C_SQ_RSRC_IMG_CUBE;
1357 }
1358 }
1359
1360 /*
1361 * Format support testing
1362 */
1363
1364 static bool si_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
1365 {
1366 return si_translate_texformat(screen, format, util_format_description(format),
1367 util_format_get_first_non_void_channel(format)) != ~0U;
1368 }
1369
1370 static uint32_t si_translate_buffer_dataformat(struct pipe_screen *screen,
1371 const struct util_format_description *desc,
1372 int first_non_void)
1373 {
1374 unsigned type = desc->channel[first_non_void].type;
1375 int i;
1376
1377 if (type == UTIL_FORMAT_TYPE_FIXED)
1378 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1379
1380 if (desc->nr_channels == 4 &&
1381 desc->channel[0].size == 10 &&
1382 desc->channel[1].size == 10 &&
1383 desc->channel[2].size == 10 &&
1384 desc->channel[3].size == 2)
1385 return V_008F0C_BUF_DATA_FORMAT_2_10_10_10;
1386
1387 /* See whether the components are of the same size. */
1388 for (i = 0; i < desc->nr_channels; i++) {
1389 if (desc->channel[first_non_void].size != desc->channel[i].size)
1390 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1391 }
1392
1393 switch (desc->channel[first_non_void].size) {
1394 case 8:
1395 switch (desc->nr_channels) {
1396 case 1:
1397 return V_008F0C_BUF_DATA_FORMAT_8;
1398 case 2:
1399 return V_008F0C_BUF_DATA_FORMAT_8_8;
1400 case 3:
1401 case 4:
1402 return V_008F0C_BUF_DATA_FORMAT_8_8_8_8;
1403 }
1404 break;
1405 case 16:
1406 switch (desc->nr_channels) {
1407 case 1:
1408 return V_008F0C_BUF_DATA_FORMAT_16;
1409 case 2:
1410 return V_008F0C_BUF_DATA_FORMAT_16_16;
1411 case 3:
1412 case 4:
1413 return V_008F0C_BUF_DATA_FORMAT_16_16_16_16;
1414 }
1415 break;
1416 case 32:
1417 /* From the Southern Islands ISA documentation about MTBUF:
1418 * 'Memory reads of data in memory that is 32 or 64 bits do not
1419 * undergo any format conversion.'
1420 */
1421 if (type != UTIL_FORMAT_TYPE_FLOAT &&
1422 !desc->channel[first_non_void].pure_integer)
1423 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1424
1425 switch (desc->nr_channels) {
1426 case 1:
1427 return V_008F0C_BUF_DATA_FORMAT_32;
1428 case 2:
1429 return V_008F0C_BUF_DATA_FORMAT_32_32;
1430 case 3:
1431 return V_008F0C_BUF_DATA_FORMAT_32_32_32;
1432 case 4:
1433 return V_008F0C_BUF_DATA_FORMAT_32_32_32_32;
1434 }
1435 break;
1436 }
1437
1438 return V_008F0C_BUF_DATA_FORMAT_INVALID;
1439 }
1440
1441 static uint32_t si_translate_buffer_numformat(struct pipe_screen *screen,
1442 const struct util_format_description *desc,
1443 int first_non_void)
1444 {
1445 switch (desc->channel[first_non_void].type) {
1446 case UTIL_FORMAT_TYPE_SIGNED:
1447 if (desc->channel[first_non_void].normalized)
1448 return V_008F0C_BUF_NUM_FORMAT_SNORM;
1449 else if (desc->channel[first_non_void].pure_integer)
1450 return V_008F0C_BUF_NUM_FORMAT_SINT;
1451 else
1452 return V_008F0C_BUF_NUM_FORMAT_SSCALED;
1453 break;
1454 case UTIL_FORMAT_TYPE_UNSIGNED:
1455 if (desc->channel[first_non_void].normalized)
1456 return V_008F0C_BUF_NUM_FORMAT_UNORM;
1457 else if (desc->channel[first_non_void].pure_integer)
1458 return V_008F0C_BUF_NUM_FORMAT_UINT;
1459 else
1460 return V_008F0C_BUF_NUM_FORMAT_USCALED;
1461 break;
1462 case UTIL_FORMAT_TYPE_FLOAT:
1463 default:
1464 return V_008F0C_BUF_NUM_FORMAT_FLOAT;
1465 }
1466 }
1467
1468 static bool si_is_vertex_format_supported(struct pipe_screen *screen, enum pipe_format format)
1469 {
1470 const struct util_format_description *desc;
1471 int first_non_void;
1472 unsigned data_format;
1473
1474 desc = util_format_description(format);
1475 first_non_void = util_format_get_first_non_void_channel(format);
1476 data_format = si_translate_buffer_dataformat(screen, desc, first_non_void);
1477 return data_format != V_008F0C_BUF_DATA_FORMAT_INVALID;
1478 }
1479
1480 static bool si_is_colorbuffer_format_supported(enum pipe_format format)
1481 {
1482 return si_translate_colorformat(format) != V_028C70_COLOR_INVALID &&
1483 si_translate_colorswap(format) != ~0U;
1484 }
1485
1486 static bool si_is_zs_format_supported(enum pipe_format format)
1487 {
1488 return si_translate_dbformat(format) != V_028040_Z_INVALID;
1489 }
1490
1491 boolean si_is_format_supported(struct pipe_screen *screen,
1492 enum pipe_format format,
1493 enum pipe_texture_target target,
1494 unsigned sample_count,
1495 unsigned usage)
1496 {
1497 struct si_screen *sscreen = (struct si_screen *)screen;
1498 unsigned retval = 0;
1499
1500 if (target >= PIPE_MAX_TEXTURE_TYPES) {
1501 R600_ERR("r600: unsupported texture type %d\n", target);
1502 return FALSE;
1503 }
1504
1505 if (!util_format_is_supported(format, usage))
1506 return FALSE;
1507
1508 if (sample_count > 1) {
1509 if (HAVE_LLVM < 0x0304)
1510 return FALSE;
1511
1512 /* 2D tiling on CIK is supported since DRM 2.35.0 */
1513 if (sscreen->b.chip_class >= CIK && sscreen->b.info.drm_minor < 35)
1514 return FALSE;
1515
1516 switch (sample_count) {
1517 case 2:
1518 case 4:
1519 case 8:
1520 break;
1521 default:
1522 return FALSE;
1523 }
1524 }
1525
1526 if (usage & PIPE_BIND_SAMPLER_VIEW) {
1527 if (target == PIPE_BUFFER) {
1528 if (si_is_vertex_format_supported(screen, format))
1529 retval |= PIPE_BIND_SAMPLER_VIEW;
1530 } else {
1531 if (si_is_sampler_format_supported(screen, format))
1532 retval |= PIPE_BIND_SAMPLER_VIEW;
1533 }
1534 }
1535
1536 if ((usage & (PIPE_BIND_RENDER_TARGET |
1537 PIPE_BIND_DISPLAY_TARGET |
1538 PIPE_BIND_SCANOUT |
1539 PIPE_BIND_SHARED)) &&
1540 si_is_colorbuffer_format_supported(format)) {
1541 retval |= usage &
1542 (PIPE_BIND_RENDER_TARGET |
1543 PIPE_BIND_DISPLAY_TARGET |
1544 PIPE_BIND_SCANOUT |
1545 PIPE_BIND_SHARED);
1546 }
1547
1548 if ((usage & PIPE_BIND_DEPTH_STENCIL) &&
1549 si_is_zs_format_supported(format)) {
1550 retval |= PIPE_BIND_DEPTH_STENCIL;
1551 }
1552
1553 if ((usage & PIPE_BIND_VERTEX_BUFFER) &&
1554 si_is_vertex_format_supported(screen, format)) {
1555 retval |= PIPE_BIND_VERTEX_BUFFER;
1556 }
1557
1558 if (usage & PIPE_BIND_TRANSFER_READ)
1559 retval |= PIPE_BIND_TRANSFER_READ;
1560 if (usage & PIPE_BIND_TRANSFER_WRITE)
1561 retval |= PIPE_BIND_TRANSFER_WRITE;
1562
1563 return retval == usage;
1564 }
1565
1566 static unsigned si_tile_mode_index(struct r600_texture *rtex, unsigned level, bool stencil)
1567 {
1568 unsigned tile_mode_index = 0;
1569
1570 if (stencil) {
1571 tile_mode_index = rtex->surface.stencil_tiling_index[level];
1572 } else {
1573 tile_mode_index = rtex->surface.tiling_index[level];
1574 }
1575 return tile_mode_index;
1576 }
1577
1578 /*
1579 * framebuffer handling
1580 */
1581
1582 static void si_cb(struct si_context *sctx, struct si_pm4_state *pm4,
1583 const struct pipe_framebuffer_state *state, int cb)
1584 {
1585 struct r600_texture *rtex;
1586 struct si_surface *surf;
1587 unsigned level = state->cbufs[cb]->u.tex.level;
1588 unsigned pitch, slice;
1589 unsigned color_info, color_attrib, color_pitch, color_view;
1590 unsigned tile_mode_index;
1591 unsigned format, swap, ntype, endian;
1592 uint64_t offset;
1593 const struct util_format_description *desc;
1594 int i;
1595 unsigned blend_clamp = 0, blend_bypass = 0;
1596 unsigned max_comp_size;
1597
1598 surf = (struct si_surface *)state->cbufs[cb];
1599 rtex = (struct r600_texture*)state->cbufs[cb]->texture;
1600
1601 offset = rtex->surface.level[level].offset;
1602
1603 /* Layered rendering doesn't work with LINEAR_GENERAL.
1604 * (LINEAR_ALIGNED and others work) */
1605 if (rtex->surface.level[level].mode == RADEON_SURF_MODE_LINEAR) {
1606 assert(state->cbufs[cb]->u.tex.first_layer == state->cbufs[cb]->u.tex.last_layer);
1607 offset += rtex->surface.level[level].slice_size *
1608 state->cbufs[cb]->u.tex.first_layer;
1609 color_view = 0;
1610 } else {
1611 color_view = S_028C6C_SLICE_START(state->cbufs[cb]->u.tex.first_layer) |
1612 S_028C6C_SLICE_MAX(state->cbufs[cb]->u.tex.last_layer);
1613 }
1614
1615 pitch = (rtex->surface.level[level].nblk_x) / 8 - 1;
1616 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
1617 if (slice) {
1618 slice = slice - 1;
1619 }
1620
1621 tile_mode_index = si_tile_mode_index(rtex, level, false);
1622
1623 desc = util_format_description(surf->base.format);
1624 for (i = 0; i < 4; i++) {
1625 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
1626 break;
1627 }
1628 }
1629 if (i == 4 || desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT) {
1630 ntype = V_028C70_NUMBER_FLOAT;
1631 } else {
1632 ntype = V_028C70_NUMBER_UNORM;
1633 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
1634 ntype = V_028C70_NUMBER_SRGB;
1635 else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
1636 if (desc->channel[i].pure_integer) {
1637 ntype = V_028C70_NUMBER_SINT;
1638 } else {
1639 assert(desc->channel[i].normalized);
1640 ntype = V_028C70_NUMBER_SNORM;
1641 }
1642 } else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
1643 if (desc->channel[i].pure_integer) {
1644 ntype = V_028C70_NUMBER_UINT;
1645 } else {
1646 assert(desc->channel[i].normalized);
1647 ntype = V_028C70_NUMBER_UNORM;
1648 }
1649 }
1650 }
1651
1652 format = si_translate_colorformat(surf->base.format);
1653 if (format == V_028C70_COLOR_INVALID) {
1654 R600_ERR("Invalid CB format: %d, disabling CB.\n", surf->base.format);
1655 }
1656 assert(format != V_028C70_COLOR_INVALID);
1657 swap = si_translate_colorswap(surf->base.format);
1658 if (rtex->resource.b.b.usage == PIPE_USAGE_STAGING) {
1659 endian = V_028C70_ENDIAN_NONE;
1660 } else {
1661 endian = si_colorformat_endian_swap(format);
1662 }
1663
1664 /* blend clamp should be set for all NORM/SRGB types */
1665 if (ntype == V_028C70_NUMBER_UNORM ||
1666 ntype == V_028C70_NUMBER_SNORM ||
1667 ntype == V_028C70_NUMBER_SRGB)
1668 blend_clamp = 1;
1669
1670 /* set blend bypass according to docs if SINT/UINT or
1671 8/24 COLOR variants */
1672 if (ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT ||
1673 format == V_028C70_COLOR_8_24 || format == V_028C70_COLOR_24_8 ||
1674 format == V_028C70_COLOR_X24_8_32_FLOAT) {
1675 blend_clamp = 0;
1676 blend_bypass = 1;
1677 }
1678
1679 color_info = S_028C70_FORMAT(format) |
1680 S_028C70_COMP_SWAP(swap) |
1681 S_028C70_BLEND_CLAMP(blend_clamp) |
1682 S_028C70_BLEND_BYPASS(blend_bypass) |
1683 S_028C70_NUMBER_TYPE(ntype) |
1684 S_028C70_ENDIAN(endian);
1685
1686 color_pitch = S_028C64_TILE_MAX(pitch);
1687
1688 color_attrib = S_028C74_TILE_MODE_INDEX(tile_mode_index) |
1689 S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == UTIL_FORMAT_SWIZZLE_1);
1690
1691 if (rtex->resource.b.b.nr_samples > 1) {
1692 unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples);
1693
1694 color_attrib |= S_028C74_NUM_SAMPLES(log_samples) |
1695 S_028C74_NUM_FRAGMENTS(log_samples);
1696
1697 if (rtex->fmask.size) {
1698 color_info |= S_028C70_COMPRESSION(1);
1699 unsigned fmask_bankh = util_logbase2(rtex->fmask.bank_height);
1700
1701 color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(rtex->fmask.tile_mode_index);
1702
1703 if (sctx->b.chip_class == SI) {
1704 /* due to a hw bug, FMASK_BANK_HEIGHT must be set on SI too */
1705 color_attrib |= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
1706 }
1707 if (sctx->b.chip_class >= CIK) {
1708 color_pitch |= S_028C64_FMASK_TILE_MAX(rtex->fmask.pitch / 8 - 1);
1709 }
1710 }
1711 }
1712
1713 if (rtex->cmask.size) {
1714 color_info |= S_028C70_FAST_CLEAR(1);
1715 }
1716
1717 offset += r600_resource_va(sctx->b.b.screen, state->cbufs[cb]->texture);
1718 offset >>= 8;
1719
1720 si_pm4_add_bo(pm4, &rtex->resource, RADEON_USAGE_READWRITE);
1721 si_pm4_set_reg(pm4, R_028C60_CB_COLOR0_BASE + cb * 0x3C, offset);
1722 si_pm4_set_reg(pm4, R_028C64_CB_COLOR0_PITCH + cb * 0x3C, color_pitch);
1723 si_pm4_set_reg(pm4, R_028C68_CB_COLOR0_SLICE + cb * 0x3C, S_028C68_TILE_MAX(slice));
1724 si_pm4_set_reg(pm4, R_028C6C_CB_COLOR0_VIEW + cb * 0x3C, color_view);
1725 si_pm4_set_reg(pm4, R_028C70_CB_COLOR0_INFO + cb * 0x3C, color_info);
1726 si_pm4_set_reg(pm4, R_028C74_CB_COLOR0_ATTRIB + cb * 0x3C, color_attrib);
1727
1728 if (rtex->cmask.size) {
1729 si_pm4_set_reg(pm4, R_028C7C_CB_COLOR0_CMASK + cb * 0x3C,
1730 offset + (rtex->cmask.offset >> 8));
1731 si_pm4_set_reg(pm4, R_028C80_CB_COLOR0_CMASK_SLICE + cb * 0x3C,
1732 S_028C80_TILE_MAX(rtex->cmask.slice_tile_max));
1733 }
1734 if (rtex->fmask.size) {
1735 si_pm4_set_reg(pm4, R_028C84_CB_COLOR0_FMASK + cb * 0x3C,
1736 offset + (rtex->fmask.offset >> 8));
1737 si_pm4_set_reg(pm4, R_028C88_CB_COLOR0_FMASK_SLICE + cb * 0x3C,
1738 S_028C88_TILE_MAX(rtex->fmask.slice_tile_max));
1739 }
1740
1741 /* set CB_COLOR1_INFO for possible dual-src blending */
1742 if (state->nr_cbufs == 1) {
1743 assert(cb == 0);
1744 si_pm4_set_reg(pm4, R_028C70_CB_COLOR0_INFO + 1 * 0x3C, color_info);
1745 }
1746
1747 /* Determine pixel shader export format */
1748 max_comp_size = si_colorformat_max_comp_size(format);
1749 if (ntype == V_028C70_NUMBER_SRGB ||
1750 ((ntype == V_028C70_NUMBER_UNORM || ntype == V_028C70_NUMBER_SNORM) &&
1751 max_comp_size <= 10) ||
1752 (ntype == V_028C70_NUMBER_FLOAT && max_comp_size <= 16)) {
1753 sctx->export_16bpc |= 1 << cb;
1754 /* set SPI_SHADER_COL_FORMAT for possible dual-src blending */
1755 if (state->nr_cbufs == 1)
1756 sctx->export_16bpc |= 1 << 1;
1757 }
1758 }
1759
1760 static void si_db(struct si_context *sctx, struct si_pm4_state *pm4,
1761 const struct pipe_framebuffer_state *state)
1762 {
1763 struct si_screen *sscreen = sctx->screen;
1764 struct r600_texture *rtex;
1765 struct si_surface *surf;
1766 unsigned level, pitch, slice, format, tile_mode_index, array_mode;
1767 unsigned macro_aspect, tile_split, stile_split, bankh, bankw, nbanks, pipe_config;
1768 uint32_t z_info, s_info, db_depth_info;
1769 uint64_t z_offs, s_offs;
1770 uint32_t db_htile_data_base, db_htile_surface;
1771
1772 if (state->zsbuf == NULL) {
1773 si_pm4_set_reg(pm4, R_028040_DB_Z_INFO, S_028040_FORMAT(V_028040_Z_INVALID));
1774 si_pm4_set_reg(pm4, R_028044_DB_STENCIL_INFO, S_028044_FORMAT(V_028044_STENCIL_INVALID));
1775 return;
1776 }
1777
1778 surf = (struct si_surface *)state->zsbuf;
1779 level = surf->base.u.tex.level;
1780 rtex = (struct r600_texture*)surf->base.texture;
1781
1782 format = si_translate_dbformat(rtex->resource.b.b.format);
1783
1784 if (format == V_028040_Z_INVALID) {
1785 R600_ERR("Invalid DB format: %d, disabling DB.\n", rtex->resource.b.b.format);
1786 }
1787 assert(format != V_028040_Z_INVALID);
1788
1789 s_offs = z_offs = r600_resource_va(sctx->b.b.screen, surf->base.texture);
1790 z_offs += rtex->surface.level[level].offset;
1791 s_offs += rtex->surface.stencil_level[level].offset;
1792
1793 z_offs >>= 8;
1794 s_offs >>= 8;
1795
1796 pitch = (rtex->surface.level[level].nblk_x / 8) - 1;
1797 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
1798 if (slice) {
1799 slice = slice - 1;
1800 }
1801
1802 db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
1803
1804 z_info = S_028040_FORMAT(format);
1805 if (rtex->resource.b.b.nr_samples > 1) {
1806 z_info |= S_028040_NUM_SAMPLES(util_logbase2(rtex->resource.b.b.nr_samples));
1807 }
1808
1809 if (rtex->surface.flags & RADEON_SURF_SBUFFER)
1810 s_info = S_028044_FORMAT(V_028044_STENCIL_8);
1811 else
1812 s_info = S_028044_FORMAT(V_028044_STENCIL_INVALID);
1813
1814 if (sctx->b.chip_class >= CIK) {
1815 switch (rtex->surface.level[level].mode) {
1816 case RADEON_SURF_MODE_2D:
1817 array_mode = V_02803C_ARRAY_2D_TILED_THIN1;
1818 break;
1819 case RADEON_SURF_MODE_1D:
1820 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1821 case RADEON_SURF_MODE_LINEAR:
1822 default:
1823 array_mode = V_02803C_ARRAY_1D_TILED_THIN1;
1824 break;
1825 }
1826 tile_split = rtex->surface.tile_split;
1827 stile_split = rtex->surface.stencil_tile_split;
1828 macro_aspect = rtex->surface.mtilea;
1829 bankw = rtex->surface.bankw;
1830 bankh = rtex->surface.bankh;
1831 tile_split = cik_tile_split(tile_split);
1832 stile_split = cik_tile_split(stile_split);
1833 macro_aspect = cik_macro_tile_aspect(macro_aspect);
1834 bankw = cik_bank_wh(bankw);
1835 bankh = cik_bank_wh(bankh);
1836 nbanks = cik_num_banks(sscreen, rtex->surface.bpe, rtex->surface.tile_split);
1837 tile_mode_index = si_tile_mode_index(rtex, level, false);
1838 pipe_config = cik_db_pipe_config(sscreen, tile_mode_index);
1839
1840 db_depth_info |= S_02803C_ARRAY_MODE(array_mode) |
1841 S_02803C_PIPE_CONFIG(pipe_config) |
1842 S_02803C_BANK_WIDTH(bankw) |
1843 S_02803C_BANK_HEIGHT(bankh) |
1844 S_02803C_MACRO_TILE_ASPECT(macro_aspect) |
1845 S_02803C_NUM_BANKS(nbanks);
1846 z_info |= S_028040_TILE_SPLIT(tile_split);
1847 s_info |= S_028044_TILE_SPLIT(stile_split);
1848 } else {
1849 tile_mode_index = si_tile_mode_index(rtex, level, false);
1850 z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
1851 tile_mode_index = si_tile_mode_index(rtex, level, true);
1852 s_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
1853 }
1854
1855 /* HiZ aka depth buffer htile */
1856 /* use htile only for first level */
1857 if (rtex->htile_buffer && !level) {
1858 const struct util_format_description *fmt_desc;
1859
1860 z_info |= S_028040_TILE_SURFACE_ENABLE(1);
1861
1862 /* This is optimal for the clear value of 1.0 and using
1863 * the LESS and LEQUAL test functions. Set this to 0
1864 * for the opposite case. This can only be changed when
1865 * clearing. */
1866 z_info |= S_028040_ZRANGE_PRECISION(1);
1867
1868 fmt_desc = util_format_description(rtex->resource.b.b.format);
1869 if (!util_format_has_stencil(fmt_desc)) {
1870 /* Use all of the htile_buffer for depth */
1871 s_info |= S_028044_TILE_STENCIL_DISABLE(1);
1872 }
1873
1874 uint64_t va = r600_resource_va(&sctx->screen->b.b, &rtex->htile_buffer->b.b);
1875 db_htile_data_base = va >> 8;
1876 db_htile_surface = S_028ABC_FULL_CACHE(1);
1877
1878 si_pm4_add_bo(pm4, rtex->htile_buffer, RADEON_USAGE_READWRITE);
1879 } else {
1880 db_htile_data_base = 0;
1881 db_htile_surface = 0;
1882 }
1883
1884 si_pm4_set_reg(pm4, R_028008_DB_DEPTH_VIEW,
1885 S_028008_SLICE_START(state->zsbuf->u.tex.first_layer) |
1886 S_028008_SLICE_MAX(state->zsbuf->u.tex.last_layer));
1887 si_pm4_set_reg(pm4, R_028014_DB_HTILE_DATA_BASE, db_htile_data_base);
1888
1889 si_pm4_set_reg(pm4, R_02803C_DB_DEPTH_INFO, db_depth_info);
1890 si_pm4_set_reg(pm4, R_028040_DB_Z_INFO, z_info);
1891 si_pm4_set_reg(pm4, R_028044_DB_STENCIL_INFO, s_info);
1892
1893 si_pm4_add_bo(pm4, &rtex->resource, RADEON_USAGE_READWRITE);
1894 si_pm4_set_reg(pm4, R_028048_DB_Z_READ_BASE, z_offs);
1895 si_pm4_set_reg(pm4, R_02804C_DB_STENCIL_READ_BASE, s_offs);
1896 si_pm4_set_reg(pm4, R_028050_DB_Z_WRITE_BASE, z_offs);
1897 si_pm4_set_reg(pm4, R_028054_DB_STENCIL_WRITE_BASE, s_offs);
1898
1899 si_pm4_set_reg(pm4, R_028058_DB_DEPTH_SIZE, S_028058_PITCH_TILE_MAX(pitch));
1900 si_pm4_set_reg(pm4, R_02805C_DB_DEPTH_SLICE, S_02805C_SLICE_TILE_MAX(slice));
1901
1902 si_pm4_set_reg(pm4, R_028ABC_DB_HTILE_SURFACE, db_htile_surface);
1903 }
1904
1905 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
1906 (((s0x) & 0xf) | (((s0y) & 0xf) << 4) | \
1907 (((s1x) & 0xf) << 8) | (((s1y) & 0xf) << 12) | \
1908 (((s2x) & 0xf) << 16) | (((s2y) & 0xf) << 20) | \
1909 (((s3x) & 0xf) << 24) | (((s3y) & 0xf) << 28))
1910
1911 /* 2xMSAA
1912 * There are two locations (-4, 4), (4, -4). */
1913 static uint32_t sample_locs_2x[] = {
1914 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1915 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1916 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1917 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1918 };
1919 static unsigned max_dist_2x = 4;
1920 /* 4xMSAA
1921 * There are 4 locations: (-2, -2), (2, 2), (-6, 6), (6, -6). */
1922 static uint32_t sample_locs_4x[] = {
1923 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1924 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1925 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1926 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1927 };
1928 static unsigned max_dist_4x = 6;
1929 /* Cayman/SI 8xMSAA */
1930 static uint32_t cm_sample_locs_8x[] = {
1931 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1932 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1933 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1934 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1935 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1936 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1937 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1938 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1939 };
1940 static unsigned cm_max_dist_8x = 8;
1941 /* Cayman/SI 16xMSAA */
1942 static uint32_t cm_sample_locs_16x[] = {
1943 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1944 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1945 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1946 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1947 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1948 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1949 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1950 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1951 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1952 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1953 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1954 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1955 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1956 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1957 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1958 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1959 };
1960 static unsigned cm_max_dist_16x = 8;
1961
1962 static void si_get_sample_position(struct pipe_context *ctx,
1963 unsigned sample_count,
1964 unsigned sample_index,
1965 float *out_value)
1966 {
1967 int offset, index;
1968 struct {
1969 int idx:4;
1970 } val;
1971 switch (sample_count) {
1972 case 1:
1973 default:
1974 out_value[0] = out_value[1] = 0.5;
1975 break;
1976 case 2:
1977 offset = 4 * (sample_index * 2);
1978 val.idx = (sample_locs_2x[0] >> offset) & 0xf;
1979 out_value[0] = (float)(val.idx + 8) / 16.0f;
1980 val.idx = (sample_locs_2x[0] >> (offset + 4)) & 0xf;
1981 out_value[1] = (float)(val.idx + 8) / 16.0f;
1982 break;
1983 case 4:
1984 offset = 4 * (sample_index * 2);
1985 val.idx = (sample_locs_4x[0] >> offset) & 0xf;
1986 out_value[0] = (float)(val.idx + 8) / 16.0f;
1987 val.idx = (sample_locs_4x[0] >> (offset + 4)) & 0xf;
1988 out_value[1] = (float)(val.idx + 8) / 16.0f;
1989 break;
1990 case 8:
1991 offset = 4 * (sample_index % 4 * 2);
1992 index = (sample_index / 4) * 4;
1993 val.idx = (cm_sample_locs_8x[index] >> offset) & 0xf;
1994 out_value[0] = (float)(val.idx + 8) / 16.0f;
1995 val.idx = (cm_sample_locs_8x[index] >> (offset + 4)) & 0xf;
1996 out_value[1] = (float)(val.idx + 8) / 16.0f;
1997 break;
1998 case 16:
1999 offset = 4 * (sample_index % 4 * 2);
2000 index = (sample_index / 4) * 4;
2001 val.idx = (cm_sample_locs_16x[index] >> offset) & 0xf;
2002 out_value[0] = (float)(val.idx + 8) / 16.0f;
2003 val.idx = (cm_sample_locs_16x[index] >> (offset + 4)) & 0xf;
2004 out_value[1] = (float)(val.idx + 8) / 16.0f;
2005 break;
2006 }
2007 }
2008
2009 static void si_set_msaa_state(struct si_context *sctx, struct si_pm4_state *pm4, int nr_samples)
2010 {
2011 unsigned max_dist = 0;
2012
2013 switch (nr_samples) {
2014 default:
2015 nr_samples = 0;
2016 break;
2017 case 2:
2018 si_pm4_set_reg(pm4, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_2x[0]);
2019 si_pm4_set_reg(pm4, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_2x[1]);
2020 si_pm4_set_reg(pm4, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_2x[2]);
2021 si_pm4_set_reg(pm4, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_2x[3]);
2022 max_dist = max_dist_2x;
2023 break;
2024 case 4:
2025 si_pm4_set_reg(pm4, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_4x[0]);
2026 si_pm4_set_reg(pm4, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_4x[1]);
2027 si_pm4_set_reg(pm4, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_4x[2]);
2028 si_pm4_set_reg(pm4, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_4x[3]);
2029 max_dist = max_dist_4x;
2030 break;
2031 case 8:
2032 si_pm4_set_reg(pm4, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, cm_sample_locs_8x[0]);
2033 si_pm4_set_reg(pm4, R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1, cm_sample_locs_8x[4]);
2034 si_pm4_set_reg(pm4, R_028C00_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_2, 0);
2035 si_pm4_set_reg(pm4, R_028C04_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_3, 0);
2036 si_pm4_set_reg(pm4, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, cm_sample_locs_8x[1]);
2037 si_pm4_set_reg(pm4, R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1, cm_sample_locs_8x[5]);
2038 si_pm4_set_reg(pm4, R_028C10_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_2, 0);
2039 si_pm4_set_reg(pm4, R_028C14_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_3, 0);
2040 si_pm4_set_reg(pm4, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, cm_sample_locs_8x[2]);
2041 si_pm4_set_reg(pm4, R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1, cm_sample_locs_8x[6]);
2042 si_pm4_set_reg(pm4, R_028C20_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_2, 0);
2043 si_pm4_set_reg(pm4, R_028C24_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_3, 0);
2044 si_pm4_set_reg(pm4, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, cm_sample_locs_8x[3]);
2045 si_pm4_set_reg(pm4, R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1, cm_sample_locs_8x[7]);
2046 max_dist = cm_max_dist_8x;
2047 break;
2048 case 16:
2049 si_pm4_set_reg(pm4, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, cm_sample_locs_16x[0]);
2050 si_pm4_set_reg(pm4, R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1, cm_sample_locs_16x[4]);
2051 si_pm4_set_reg(pm4, R_028C00_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_2, cm_sample_locs_16x[8]);
2052 si_pm4_set_reg(pm4, R_028C04_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_3, cm_sample_locs_16x[12]);
2053 si_pm4_set_reg(pm4, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, cm_sample_locs_16x[1]);
2054 si_pm4_set_reg(pm4, R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1, cm_sample_locs_16x[5]);
2055 si_pm4_set_reg(pm4, R_028C10_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_2, cm_sample_locs_16x[9]);
2056 si_pm4_set_reg(pm4, R_028C14_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_3, cm_sample_locs_16x[13]);
2057 si_pm4_set_reg(pm4, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, cm_sample_locs_16x[2]);
2058 si_pm4_set_reg(pm4, R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1, cm_sample_locs_16x[6]);
2059 si_pm4_set_reg(pm4, R_028C20_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_2, cm_sample_locs_16x[10]);
2060 si_pm4_set_reg(pm4, R_028C24_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_3, cm_sample_locs_16x[14]);
2061 si_pm4_set_reg(pm4, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, cm_sample_locs_16x[3]);
2062 si_pm4_set_reg(pm4, R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1, cm_sample_locs_16x[7]);
2063 si_pm4_set_reg(pm4, R_028C30_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_2, cm_sample_locs_16x[11]);
2064 si_pm4_set_reg(pm4, R_028C34_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_3, cm_sample_locs_16x[15]);
2065 max_dist = cm_max_dist_16x;
2066 break;
2067 }
2068
2069 if (nr_samples > 1) {
2070 unsigned log_samples = util_logbase2(nr_samples);
2071
2072 si_pm4_set_reg(pm4, R_028BDC_PA_SC_LINE_CNTL,
2073 S_028BDC_LAST_PIXEL(1) |
2074 S_028BDC_EXPAND_LINE_WIDTH(1));
2075 si_pm4_set_reg(pm4, R_028BE0_PA_SC_AA_CONFIG,
2076 S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
2077 S_028BE0_MAX_SAMPLE_DIST(max_dist) |
2078 S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples));
2079
2080 si_pm4_set_reg(pm4, R_028804_DB_EQAA,
2081 S_028804_MAX_ANCHOR_SAMPLES(log_samples) |
2082 S_028804_PS_ITER_SAMPLES(log_samples) |
2083 S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples) |
2084 S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples) |
2085 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
2086 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
2087 } else {
2088 si_pm4_set_reg(pm4, R_028BDC_PA_SC_LINE_CNTL, S_028BDC_LAST_PIXEL(1));
2089 si_pm4_set_reg(pm4, R_028BE0_PA_SC_AA_CONFIG, 0);
2090
2091 si_pm4_set_reg(pm4, R_028804_DB_EQAA,
2092 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
2093 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
2094 }
2095 }
2096
2097 static void si_set_framebuffer_state(struct pipe_context *ctx,
2098 const struct pipe_framebuffer_state *state)
2099 {
2100 struct si_context *sctx = (struct si_context *)ctx;
2101 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
2102 int nr_samples, i;
2103
2104 if (pm4 == NULL)
2105 return;
2106
2107 if (sctx->framebuffer.nr_cbufs) {
2108 sctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV_CB |
2109 R600_CONTEXT_FLUSH_AND_INV_CB_META;
2110 }
2111 if (sctx->framebuffer.zsbuf) {
2112 sctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV_DB |
2113 R600_CONTEXT_FLUSH_AND_INV_DB_META;
2114 }
2115
2116 util_copy_framebuffer_state(&sctx->framebuffer, state);
2117
2118 /* build states */
2119 sctx->export_16bpc = 0;
2120 sctx->fb_compressed_cb_mask = 0;
2121 for (i = 0; i < state->nr_cbufs; i++) {
2122 struct r600_texture *rtex;
2123
2124 if (!state->cbufs[i]) {
2125 si_pm4_set_reg(pm4, R_028C70_CB_COLOR0_INFO + i * 0x3C,
2126 S_028C70_FORMAT(V_028C70_COLOR_INVALID));
2127 continue;
2128 }
2129
2130 rtex = (struct r600_texture*)state->cbufs[i]->texture;
2131
2132 si_cb(sctx, pm4, state, i);
2133
2134 if (rtex->fmask.size || rtex->cmask.size) {
2135 sctx->fb_compressed_cb_mask |= 1 << i;
2136 }
2137 }
2138 for (; i < 8; i++) {
2139 si_pm4_set_reg(pm4, R_028C70_CB_COLOR0_INFO + i * 0x3C,
2140 S_028C70_FORMAT(V_028C70_COLOR_INVALID));
2141 }
2142
2143 assert(!(sctx->export_16bpc & ~0xff));
2144 si_db(sctx, pm4, state);
2145
2146 /* PA_SC_WINDOW_SCISSOR_TL is set in si_init_config() */
2147 si_pm4_set_reg(pm4, R_028208_PA_SC_WINDOW_SCISSOR_BR,
2148 S_028208_BR_X(state->width) | S_028208_BR_Y(state->height));
2149
2150 nr_samples = util_framebuffer_get_num_samples(state);
2151
2152 si_set_msaa_state(sctx, pm4, nr_samples);
2153 sctx->fb_log_samples = util_logbase2(nr_samples);
2154 sctx->fb_cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
2155 util_format_is_pure_integer(state->cbufs[0]->format);
2156
2157 si_pm4_set_state(sctx, framebuffer, pm4);
2158 si_update_fb_rs_state(sctx);
2159 si_update_fb_blend_state(sctx);
2160 }
2161
2162 /*
2163 * shaders
2164 */
2165
2166 /* Compute the key for the hw shader variant */
2167 static INLINE void si_shader_selector_key(struct pipe_context *ctx,
2168 struct si_pipe_shader_selector *sel,
2169 union si_shader_key *key)
2170 {
2171 struct si_context *sctx = (struct si_context *)ctx;
2172 memset(key, 0, sizeof(*key));
2173
2174 if ((sel->type == PIPE_SHADER_VERTEX || sel->type == PIPE_SHADER_GEOMETRY) &&
2175 sctx->queued.named.rasterizer) {
2176 if (sctx->queued.named.rasterizer->clip_plane_enable & 0xf0)
2177 key->vs.ucps_enabled |= 0x2;
2178 if (sctx->queued.named.rasterizer->clip_plane_enable & 0xf)
2179 key->vs.ucps_enabled |= 0x1;
2180 }
2181
2182 if (sel->type == PIPE_SHADER_VERTEX) {
2183 unsigned i;
2184 if (!sctx->vertex_elements)
2185 return;
2186
2187 for (i = 0; i < sctx->vertex_elements->count; ++i)
2188 key->vs.instance_divisors[i] = sctx->vertex_elements->elements[i].instance_divisor;
2189
2190 key->vs.as_es = sctx->gs_shader != NULL;
2191 } else if (sel->type == PIPE_SHADER_FRAGMENT) {
2192 if (sel->fs_write_all)
2193 key->ps.nr_cbufs = sctx->framebuffer.nr_cbufs;
2194 key->ps.export_16bpc = sctx->export_16bpc;
2195
2196 if (sctx->queued.named.rasterizer) {
2197 key->ps.color_two_side = sctx->queued.named.rasterizer->two_side;
2198 key->ps.flatshade = sctx->queued.named.rasterizer->flatshade;
2199
2200 if (sctx->queued.named.blend) {
2201 key->ps.alpha_to_one = sctx->queued.named.blend->alpha_to_one &&
2202 sctx->queued.named.rasterizer->multisample_enable &&
2203 !sctx->fb_cb0_is_integer;
2204 }
2205 }
2206 if (sctx->queued.named.dsa) {
2207 key->ps.alpha_func = sctx->queued.named.dsa->alpha_func;
2208
2209 /* Alpha-test should be disabled if colorbuffer 0 is integer. */
2210 if (sctx->framebuffer.nr_cbufs &&
2211 sctx->framebuffer.cbufs[0] &&
2212 util_format_is_pure_integer(sctx->framebuffer.cbufs[0]->texture->format))
2213 key->ps.alpha_func = PIPE_FUNC_ALWAYS;
2214 } else {
2215 key->ps.alpha_func = PIPE_FUNC_ALWAYS;
2216 }
2217 }
2218 }
2219
2220 /* Select the hw shader variant depending on the current state. */
2221 int si_shader_select(struct pipe_context *ctx,
2222 struct si_pipe_shader_selector *sel)
2223 {
2224 union si_shader_key key;
2225 struct si_pipe_shader * shader = NULL;
2226 int r;
2227
2228 si_shader_selector_key(ctx, sel, &key);
2229
2230 /* Check if we don't need to change anything.
2231 * This path is also used for most shaders that don't need multiple
2232 * variants, it will cost just a computation of the key and this
2233 * test. */
2234 if (likely(sel->current && memcmp(&sel->current->key, &key, sizeof(key)) == 0)) {
2235 return 0;
2236 }
2237
2238 /* lookup if we have other variants in the list */
2239 if (sel->num_shaders > 1) {
2240 struct si_pipe_shader *p = sel->current, *c = p->next_variant;
2241
2242 while (c && memcmp(&c->key, &key, sizeof(key)) != 0) {
2243 p = c;
2244 c = c->next_variant;
2245 }
2246
2247 if (c) {
2248 p->next_variant = c->next_variant;
2249 shader = c;
2250 }
2251 }
2252
2253 if (shader) {
2254 shader->next_variant = sel->current;
2255 sel->current = shader;
2256 } else {
2257 shader = CALLOC(1, sizeof(struct si_pipe_shader));
2258 shader->selector = sel;
2259 shader->key = key;
2260
2261 shader->next_variant = sel->current;
2262 sel->current = shader;
2263 r = si_pipe_shader_create(ctx, shader);
2264 if (unlikely(r)) {
2265 R600_ERR("Failed to build shader variant (type=%u) %d\n",
2266 sel->type, r);
2267 sel->current = NULL;
2268 FREE(shader);
2269 return r;
2270 }
2271 sel->num_shaders++;
2272 }
2273
2274 return 0;
2275 }
2276
2277 static void *si_create_shader_state(struct pipe_context *ctx,
2278 const struct pipe_shader_state *state,
2279 unsigned pipe_shader_type)
2280 {
2281 struct si_pipe_shader_selector *sel = CALLOC_STRUCT(si_pipe_shader_selector);
2282 int r;
2283
2284 sel->type = pipe_shader_type;
2285 sel->tokens = tgsi_dup_tokens(state->tokens);
2286 sel->so = state->stream_output;
2287
2288 if (pipe_shader_type == PIPE_SHADER_FRAGMENT) {
2289 struct tgsi_shader_info info;
2290
2291 tgsi_scan_shader(state->tokens, &info);
2292 sel->fs_write_all = info.color0_writes_all_cbufs;
2293 }
2294
2295 r = si_shader_select(ctx, sel);
2296 if (r) {
2297 free(sel);
2298 return NULL;
2299 }
2300
2301 return sel;
2302 }
2303
2304 static void *si_create_fs_state(struct pipe_context *ctx,
2305 const struct pipe_shader_state *state)
2306 {
2307 return si_create_shader_state(ctx, state, PIPE_SHADER_FRAGMENT);
2308 }
2309
2310 #if HAVE_LLVM >= 0x0305
2311
2312 static void *si_create_gs_state(struct pipe_context *ctx,
2313 const struct pipe_shader_state *state)
2314 {
2315 return si_create_shader_state(ctx, state, PIPE_SHADER_GEOMETRY);
2316 }
2317
2318 #endif
2319
2320 static void *si_create_vs_state(struct pipe_context *ctx,
2321 const struct pipe_shader_state *state)
2322 {
2323 return si_create_shader_state(ctx, state, PIPE_SHADER_VERTEX);
2324 }
2325
2326 static void si_bind_vs_shader(struct pipe_context *ctx, void *state)
2327 {
2328 struct si_context *sctx = (struct si_context *)ctx;
2329 struct si_pipe_shader_selector *sel = state;
2330
2331 if (sctx->vs_shader == sel)
2332 return;
2333
2334 if (!sel || !sel->current)
2335 return;
2336
2337 sctx->vs_shader = sel;
2338 }
2339
2340 #if HAVE_LLVM >= 0x0305
2341
2342 static void si_bind_gs_shader(struct pipe_context *ctx, void *state)
2343 {
2344 struct si_context *sctx = (struct si_context *)ctx;
2345 struct si_pipe_shader_selector *sel = state;
2346
2347 if (sctx->gs_shader == sel)
2348 return;
2349
2350 sctx->gs_shader = sel;
2351 }
2352
2353 #endif
2354
2355 static void si_bind_ps_shader(struct pipe_context *ctx, void *state)
2356 {
2357 struct si_context *sctx = (struct si_context *)ctx;
2358 struct si_pipe_shader_selector *sel = state;
2359
2360 if (sctx->ps_shader == sel)
2361 return;
2362
2363 if (!sel || !sel->current)
2364 sel = sctx->dummy_pixel_shader;
2365
2366 sctx->ps_shader = sel;
2367 }
2368
2369 static void si_delete_shader_selector(struct pipe_context *ctx,
2370 struct si_pipe_shader_selector *sel)
2371 {
2372 struct si_context *sctx = (struct si_context *)ctx;
2373 struct si_pipe_shader *p = sel->current, *c;
2374
2375 while (p) {
2376 c = p->next_variant;
2377 si_pm4_delete_state(sctx, vs, p->pm4);
2378 si_pipe_shader_destroy(ctx, p);
2379 free(p);
2380 p = c;
2381 }
2382
2383 free(sel->tokens);
2384 free(sel);
2385 }
2386
2387 static void si_delete_vs_shader(struct pipe_context *ctx, void *state)
2388 {
2389 struct si_context *sctx = (struct si_context *)ctx;
2390 struct si_pipe_shader_selector *sel = (struct si_pipe_shader_selector *)state;
2391
2392 if (sctx->vs_shader == sel) {
2393 sctx->vs_shader = NULL;
2394 }
2395
2396 si_delete_shader_selector(ctx, sel);
2397 }
2398
2399 #if HAVE_LLVM >= 0x0305
2400
2401 static void si_delete_gs_shader(struct pipe_context *ctx, void *state)
2402 {
2403 struct si_context *sctx = (struct si_context *)ctx;
2404 struct si_pipe_shader_selector *sel = (struct si_pipe_shader_selector *)state;
2405
2406 if (sctx->gs_shader == sel) {
2407 sctx->gs_shader = NULL;
2408 }
2409
2410 si_delete_shader_selector(ctx, sel);
2411 }
2412
2413 #endif
2414
2415 static void si_delete_ps_shader(struct pipe_context *ctx, void *state)
2416 {
2417 struct si_context *sctx = (struct si_context *)ctx;
2418 struct si_pipe_shader_selector *sel = (struct si_pipe_shader_selector *)state;
2419
2420 if (sctx->ps_shader == sel) {
2421 sctx->ps_shader = NULL;
2422 }
2423
2424 si_delete_shader_selector(ctx, sel);
2425 }
2426
2427 /*
2428 * Samplers
2429 */
2430
2431 static struct pipe_sampler_view *si_create_sampler_view(struct pipe_context *ctx,
2432 struct pipe_resource *texture,
2433 const struct pipe_sampler_view *state)
2434 {
2435 struct si_pipe_sampler_view *view = CALLOC_STRUCT(si_pipe_sampler_view);
2436 struct r600_texture *tmp = (struct r600_texture*)texture;
2437 const struct util_format_description *desc;
2438 unsigned format, num_format;
2439 uint32_t pitch = 0;
2440 unsigned char state_swizzle[4], swizzle[4];
2441 unsigned height, depth, width;
2442 enum pipe_format pipe_format = state->format;
2443 struct radeon_surface_level *surflevel;
2444 int first_non_void;
2445 uint64_t va;
2446
2447 if (view == NULL)
2448 return NULL;
2449
2450 /* initialize base object */
2451 view->base = *state;
2452 view->base.texture = NULL;
2453 pipe_resource_reference(&view->base.texture, texture);
2454 view->base.reference.count = 1;
2455 view->base.context = ctx;
2456 view->resource = &tmp->resource;
2457
2458 /* Buffer resource. */
2459 if (texture->target == PIPE_BUFFER) {
2460 unsigned stride;
2461
2462 desc = util_format_description(state->format);
2463 first_non_void = util_format_get_first_non_void_channel(state->format);
2464 stride = desc->block.bits / 8;
2465 va = r600_resource_va(ctx->screen, texture) + state->u.buf.first_element*stride;
2466 format = si_translate_buffer_dataformat(ctx->screen, desc, first_non_void);
2467 num_format = si_translate_buffer_numformat(ctx->screen, desc, first_non_void);
2468
2469 view->state[0] = va;
2470 view->state[1] = S_008F04_BASE_ADDRESS_HI(va >> 32) |
2471 S_008F04_STRIDE(stride);
2472 view->state[2] = state->u.buf.last_element + 1 - state->u.buf.first_element;
2473 view->state[3] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
2474 S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
2475 S_008F0C_DST_SEL_Z(si_map_swizzle(desc->swizzle[2])) |
2476 S_008F0C_DST_SEL_W(si_map_swizzle(desc->swizzle[3])) |
2477 S_008F0C_NUM_FORMAT(num_format) |
2478 S_008F0C_DATA_FORMAT(format);
2479 return &view->base;
2480 }
2481
2482 state_swizzle[0] = state->swizzle_r;
2483 state_swizzle[1] = state->swizzle_g;
2484 state_swizzle[2] = state->swizzle_b;
2485 state_swizzle[3] = state->swizzle_a;
2486
2487 surflevel = tmp->surface.level;
2488
2489 /* Texturing with separate depth and stencil. */
2490 if (tmp->is_depth && !tmp->is_flushing_texture) {
2491 switch (pipe_format) {
2492 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
2493 pipe_format = PIPE_FORMAT_Z32_FLOAT;
2494 break;
2495 case PIPE_FORMAT_X8Z24_UNORM:
2496 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2497 /* Z24 is always stored like this. */
2498 pipe_format = PIPE_FORMAT_Z24X8_UNORM;
2499 break;
2500 case PIPE_FORMAT_X24S8_UINT:
2501 case PIPE_FORMAT_S8X24_UINT:
2502 case PIPE_FORMAT_X32_S8X24_UINT:
2503 pipe_format = PIPE_FORMAT_S8_UINT;
2504 surflevel = tmp->surface.stencil_level;
2505 break;
2506 default:;
2507 }
2508 }
2509
2510 desc = util_format_description(pipe_format);
2511
2512 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
2513 const unsigned char swizzle_xxxx[4] = {0, 0, 0, 0};
2514 const unsigned char swizzle_yyyy[4] = {1, 1, 1, 1};
2515
2516 switch (pipe_format) {
2517 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2518 case PIPE_FORMAT_X24S8_UINT:
2519 case PIPE_FORMAT_X32_S8X24_UINT:
2520 case PIPE_FORMAT_X8Z24_UNORM:
2521 util_format_compose_swizzles(swizzle_yyyy, state_swizzle, swizzle);
2522 break;
2523 default:
2524 util_format_compose_swizzles(swizzle_xxxx, state_swizzle, swizzle);
2525 }
2526 } else {
2527 util_format_compose_swizzles(desc->swizzle, state_swizzle, swizzle);
2528 }
2529
2530 first_non_void = util_format_get_first_non_void_channel(pipe_format);
2531
2532 switch (pipe_format) {
2533 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
2534 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2535 break;
2536 default:
2537 if (first_non_void < 0) {
2538 if (util_format_is_compressed(pipe_format)) {
2539 switch (pipe_format) {
2540 case PIPE_FORMAT_DXT1_SRGB:
2541 case PIPE_FORMAT_DXT1_SRGBA:
2542 case PIPE_FORMAT_DXT3_SRGBA:
2543 case PIPE_FORMAT_DXT5_SRGBA:
2544 num_format = V_008F14_IMG_NUM_FORMAT_SRGB;
2545 break;
2546 case PIPE_FORMAT_RGTC1_SNORM:
2547 case PIPE_FORMAT_LATC1_SNORM:
2548 case PIPE_FORMAT_RGTC2_SNORM:
2549 case PIPE_FORMAT_LATC2_SNORM:
2550 num_format = V_008F14_IMG_NUM_FORMAT_SNORM;
2551 break;
2552 default:
2553 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2554 break;
2555 }
2556 } else {
2557 num_format = V_008F14_IMG_NUM_FORMAT_FLOAT;
2558 }
2559 } else if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
2560 num_format = V_008F14_IMG_NUM_FORMAT_SRGB;
2561 } else {
2562 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2563
2564 switch (desc->channel[first_non_void].type) {
2565 case UTIL_FORMAT_TYPE_FLOAT:
2566 num_format = V_008F14_IMG_NUM_FORMAT_FLOAT;
2567 break;
2568 case UTIL_FORMAT_TYPE_SIGNED:
2569 if (desc->channel[first_non_void].normalized)
2570 num_format = V_008F14_IMG_NUM_FORMAT_SNORM;
2571 else if (desc->channel[first_non_void].pure_integer)
2572 num_format = V_008F14_IMG_NUM_FORMAT_SINT;
2573 else
2574 num_format = V_008F14_IMG_NUM_FORMAT_SSCALED;
2575 break;
2576 case UTIL_FORMAT_TYPE_UNSIGNED:
2577 if (desc->channel[first_non_void].normalized)
2578 num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
2579 else if (desc->channel[first_non_void].pure_integer)
2580 num_format = V_008F14_IMG_NUM_FORMAT_UINT;
2581 else
2582 num_format = V_008F14_IMG_NUM_FORMAT_USCALED;
2583 }
2584 }
2585 }
2586
2587 format = si_translate_texformat(ctx->screen, pipe_format, desc, first_non_void);
2588 if (format == ~0) {
2589 format = 0;
2590 }
2591
2592 /* not supported any more */
2593 //endian = si_colorformat_endian_swap(format);
2594
2595 width = surflevel[0].npix_x;
2596 height = surflevel[0].npix_y;
2597 depth = surflevel[0].npix_z;
2598 pitch = surflevel[0].nblk_x * util_format_get_blockwidth(pipe_format);
2599
2600 if (texture->target == PIPE_TEXTURE_1D_ARRAY) {
2601 height = 1;
2602 depth = texture->array_size;
2603 } else if (texture->target == PIPE_TEXTURE_2D_ARRAY) {
2604 depth = texture->array_size;
2605 }
2606
2607 va = r600_resource_va(ctx->screen, texture);
2608 va += surflevel[0].offset;
2609 va += tmp->mipmap_shift * surflevel[texture->last_level].slice_size;
2610 view->state[0] = va >> 8;
2611 view->state[1] = (S_008F14_BASE_ADDRESS_HI(va >> 40) |
2612 S_008F14_DATA_FORMAT(format) |
2613 S_008F14_NUM_FORMAT(num_format));
2614 view->state[2] = (S_008F18_WIDTH(width - 1) |
2615 S_008F18_HEIGHT(height - 1));
2616 view->state[3] = (S_008F1C_DST_SEL_X(si_map_swizzle(swizzle[0])) |
2617 S_008F1C_DST_SEL_Y(si_map_swizzle(swizzle[1])) |
2618 S_008F1C_DST_SEL_Z(si_map_swizzle(swizzle[2])) |
2619 S_008F1C_DST_SEL_W(si_map_swizzle(swizzle[3])) |
2620 S_008F1C_BASE_LEVEL(texture->nr_samples > 1 ?
2621 0 : state->u.tex.first_level - tmp->mipmap_shift) |
2622 S_008F1C_LAST_LEVEL(texture->nr_samples > 1 ?
2623 util_logbase2(texture->nr_samples) :
2624 state->u.tex.last_level - tmp->mipmap_shift) |
2625 S_008F1C_TILING_INDEX(si_tile_mode_index(tmp, 0, false)) |
2626 S_008F1C_POW2_PAD(texture->last_level > 0) |
2627 S_008F1C_TYPE(si_tex_dim(texture->target, texture->nr_samples)));
2628 view->state[4] = (S_008F20_DEPTH(depth - 1) | S_008F20_PITCH(pitch - 1));
2629 view->state[5] = (S_008F24_BASE_ARRAY(state->u.tex.first_layer) |
2630 S_008F24_LAST_ARRAY(state->u.tex.last_layer));
2631 view->state[6] = 0;
2632 view->state[7] = 0;
2633
2634 /* Initialize the sampler view for FMASK. */
2635 if (tmp->fmask.size) {
2636 uint64_t va = r600_resource_va(ctx->screen, texture) + tmp->fmask.offset;
2637 uint32_t fmask_format;
2638
2639 switch (texture->nr_samples) {
2640 case 2:
2641 fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S2_F2;
2642 break;
2643 case 4:
2644 fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S4_F4;
2645 break;
2646 case 8:
2647 fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK32_S8_F8;
2648 break;
2649 default:
2650 assert(0);
2651 fmask_format = V_008F14_IMG_DATA_FORMAT_INVALID;
2652 }
2653
2654 view->fmask_state[0] = va >> 8;
2655 view->fmask_state[1] = S_008F14_BASE_ADDRESS_HI(va >> 40) |
2656 S_008F14_DATA_FORMAT(fmask_format) |
2657 S_008F14_NUM_FORMAT(V_008F14_IMG_NUM_FORMAT_UINT);
2658 view->fmask_state[2] = S_008F18_WIDTH(width - 1) |
2659 S_008F18_HEIGHT(height - 1);
2660 view->fmask_state[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X) |
2661 S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X) |
2662 S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X) |
2663 S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X) |
2664 S_008F1C_TILING_INDEX(tmp->fmask.tile_mode_index) |
2665 S_008F1C_TYPE(si_tex_dim(texture->target, 0));
2666 view->fmask_state[4] = S_008F20_DEPTH(depth - 1) |
2667 S_008F20_PITCH(tmp->fmask.pitch - 1);
2668 view->fmask_state[5] = S_008F24_BASE_ARRAY(state->u.tex.first_layer) |
2669 S_008F24_LAST_ARRAY(state->u.tex.last_layer);
2670 view->fmask_state[6] = 0;
2671 view->fmask_state[7] = 0;
2672 }
2673
2674 return &view->base;
2675 }
2676
2677 static void si_sampler_view_destroy(struct pipe_context *ctx,
2678 struct pipe_sampler_view *state)
2679 {
2680 struct r600_pipe_sampler_view *resource = (struct r600_pipe_sampler_view *)state;
2681
2682 pipe_resource_reference(&state->texture, NULL);
2683 FREE(resource);
2684 }
2685
2686 static bool wrap_mode_uses_border_color(unsigned wrap, bool linear_filter)
2687 {
2688 return wrap == PIPE_TEX_WRAP_CLAMP_TO_BORDER ||
2689 wrap == PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER ||
2690 (linear_filter &&
2691 (wrap == PIPE_TEX_WRAP_CLAMP ||
2692 wrap == PIPE_TEX_WRAP_MIRROR_CLAMP));
2693 }
2694
2695 static bool sampler_state_needs_border_color(const struct pipe_sampler_state *state)
2696 {
2697 bool linear_filter = state->min_img_filter != PIPE_TEX_FILTER_NEAREST ||
2698 state->mag_img_filter != PIPE_TEX_FILTER_NEAREST;
2699
2700 return (state->border_color.ui[0] || state->border_color.ui[1] ||
2701 state->border_color.ui[2] || state->border_color.ui[3]) &&
2702 (wrap_mode_uses_border_color(state->wrap_s, linear_filter) ||
2703 wrap_mode_uses_border_color(state->wrap_t, linear_filter) ||
2704 wrap_mode_uses_border_color(state->wrap_r, linear_filter));
2705 }
2706
2707 static void *si_create_sampler_state(struct pipe_context *ctx,
2708 const struct pipe_sampler_state *state)
2709 {
2710 struct si_pipe_sampler_state *rstate = CALLOC_STRUCT(si_pipe_sampler_state);
2711 unsigned aniso_flag_offset = state->max_anisotropy > 1 ? 2 : 0;
2712 unsigned border_color_type;
2713
2714 if (rstate == NULL) {
2715 return NULL;
2716 }
2717
2718 if (sampler_state_needs_border_color(state))
2719 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER;
2720 else
2721 border_color_type = V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK;
2722
2723 rstate->val[0] = (S_008F30_CLAMP_X(si_tex_wrap(state->wrap_s)) |
2724 S_008F30_CLAMP_Y(si_tex_wrap(state->wrap_t)) |
2725 S_008F30_CLAMP_Z(si_tex_wrap(state->wrap_r)) |
2726 (state->max_anisotropy & 0x7) << 9 | /* XXX */
2727 S_008F30_DEPTH_COMPARE_FUNC(si_tex_compare(state->compare_func)) |
2728 S_008F30_FORCE_UNNORMALIZED(!state->normalized_coords) |
2729 aniso_flag_offset << 16 | /* XXX */
2730 S_008F30_DISABLE_CUBE_WRAP(!state->seamless_cube_map));
2731 rstate->val[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
2732 S_008F34_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)));
2733 rstate->val[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
2734 S_008F38_XY_MAG_FILTER(si_tex_filter(state->mag_img_filter)) |
2735 S_008F38_XY_MIN_FILTER(si_tex_filter(state->min_img_filter)) |
2736 S_008F38_MIP_FILTER(si_tex_mipfilter(state->min_mip_filter)));
2737 rstate->val[3] = S_008F3C_BORDER_COLOR_TYPE(border_color_type);
2738
2739 if (border_color_type == V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER) {
2740 memcpy(rstate->border_color, state->border_color.ui,
2741 sizeof(rstate->border_color));
2742 }
2743
2744 return rstate;
2745 }
2746
2747 /* XXX consider moving this function to si_descriptors.c for gcc to inline
2748 * the si_set_sampler_view calls. LTO might help too. */
2749 static void si_set_sampler_views(struct pipe_context *ctx,
2750 unsigned shader, unsigned start,
2751 unsigned count,
2752 struct pipe_sampler_view **views)
2753 {
2754 struct si_context *sctx = (struct si_context *)ctx;
2755 struct si_textures_info *samplers = &sctx->samplers[shader];
2756 struct si_pipe_sampler_view **rviews = (struct si_pipe_sampler_view **)views;
2757 int i;
2758
2759 if (shader >= SI_NUM_SHADERS)
2760 return;
2761
2762 assert(start == 0);
2763
2764 for (i = 0; i < count; i++) {
2765 if (!views[i]) {
2766 samplers->depth_texture_mask &= ~(1 << i);
2767 samplers->compressed_colortex_mask &= ~(1 << i);
2768 si_set_sampler_view(sctx, shader, i, NULL, NULL);
2769 si_set_sampler_view(sctx, shader, FMASK_TEX_OFFSET + i,
2770 NULL, NULL);
2771 continue;
2772 }
2773
2774 si_set_sampler_view(sctx, shader, i, views[i], rviews[i]->state);
2775
2776 if (views[i]->texture->target != PIPE_BUFFER) {
2777 struct r600_texture *rtex =
2778 (struct r600_texture*)views[i]->texture;
2779
2780 if (rtex->is_depth && !rtex->is_flushing_texture) {
2781 samplers->depth_texture_mask |= 1 << i;
2782 } else {
2783 samplers->depth_texture_mask &= ~(1 << i);
2784 }
2785 if (rtex->cmask.size || rtex->fmask.size) {
2786 samplers->compressed_colortex_mask |= 1 << i;
2787 } else {
2788 samplers->compressed_colortex_mask &= ~(1 << i);
2789 }
2790
2791 if (rtex->fmask.size) {
2792 si_set_sampler_view(sctx, shader, FMASK_TEX_OFFSET + i,
2793 views[i], rviews[i]->fmask_state);
2794 } else {
2795 si_set_sampler_view(sctx, shader, FMASK_TEX_OFFSET + i,
2796 NULL, NULL);
2797 }
2798 }
2799 }
2800 for (; i < samplers->n_views; i++) {
2801 samplers->depth_texture_mask &= ~(1 << i);
2802 samplers->compressed_colortex_mask &= ~(1 << i);
2803 si_set_sampler_view(sctx, shader, i, NULL, NULL);
2804 si_set_sampler_view(sctx, shader, FMASK_TEX_OFFSET + i,
2805 NULL, NULL);
2806 }
2807
2808 samplers->n_views = count;
2809 sctx->b.flags |= R600_CONTEXT_INV_TEX_CACHE;
2810 }
2811
2812 static struct si_pm4_state *si_set_sampler_states(struct si_context *sctx, unsigned count,
2813 void **states,
2814 struct si_textures_info *samplers,
2815 unsigned user_data_reg)
2816 {
2817 struct si_pipe_sampler_state **rstates = (struct si_pipe_sampler_state **)states;
2818 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
2819 uint32_t *border_color_table = NULL;
2820 int i, j;
2821
2822 if (!count)
2823 goto out;
2824
2825 sctx->b.flags |= R600_CONTEXT_INV_TEX_CACHE;
2826
2827 si_pm4_sh_data_begin(pm4);
2828 for (i = 0; i < count; i++) {
2829 if (rstates[i] &&
2830 G_008F3C_BORDER_COLOR_TYPE(rstates[i]->val[3]) ==
2831 V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER) {
2832 if (!sctx->border_color_table ||
2833 ((sctx->border_color_offset + count - i) &
2834 C_008F3C_BORDER_COLOR_PTR)) {
2835 r600_resource_reference(&sctx->border_color_table, NULL);
2836 sctx->border_color_offset = 0;
2837
2838 sctx->border_color_table =
2839 si_resource_create_custom(&sctx->screen->b.b,
2840 PIPE_USAGE_STAGING,
2841 4096 * 4 * 4);
2842 }
2843
2844 if (!border_color_table) {
2845 border_color_table =
2846 sctx->b.ws->buffer_map(sctx->border_color_table->cs_buf,
2847 sctx->b.rings.gfx.cs,
2848 PIPE_TRANSFER_WRITE |
2849 PIPE_TRANSFER_UNSYNCHRONIZED);
2850 }
2851
2852 for (j = 0; j < 4; j++) {
2853 border_color_table[4 * sctx->border_color_offset + j] =
2854 util_le32_to_cpu(rstates[i]->border_color[j]);
2855 }
2856
2857 rstates[i]->val[3] &= C_008F3C_BORDER_COLOR_PTR;
2858 rstates[i]->val[3] |= S_008F3C_BORDER_COLOR_PTR(sctx->border_color_offset++);
2859 }
2860
2861 for (j = 0; j < Elements(rstates[i]->val); ++j) {
2862 si_pm4_sh_data_add(pm4, rstates[i] ? rstates[i]->val[j] : 0);
2863 }
2864 }
2865 si_pm4_sh_data_end(pm4, user_data_reg, SI_SGPR_SAMPLER);
2866
2867 if (border_color_table) {
2868 uint64_t va_offset =
2869 r600_resource_va(&sctx->screen->b.b,
2870 (void*)sctx->border_color_table);
2871
2872 si_pm4_set_reg(pm4, R_028080_TA_BC_BASE_ADDR, va_offset >> 8);
2873 if (sctx->b.chip_class >= CIK)
2874 si_pm4_set_reg(pm4, R_028084_TA_BC_BASE_ADDR_HI, va_offset >> 40);
2875 sctx->b.ws->buffer_unmap(sctx->border_color_table->cs_buf);
2876 si_pm4_add_bo(pm4, sctx->border_color_table, RADEON_USAGE_READ);
2877 }
2878
2879 memcpy(samplers->samplers, states, sizeof(void*) * count);
2880
2881 out:
2882 samplers->n_samplers = count;
2883 return pm4;
2884 }
2885
2886 static void si_bind_vs_sampler_states(struct pipe_context *ctx, unsigned count, void **states)
2887 {
2888 struct si_context *sctx = (struct si_context *)ctx;
2889 struct si_pm4_state *pm4;
2890
2891 pm4 = si_set_sampler_states(sctx, count, states, &sctx->samplers[PIPE_SHADER_VERTEX],
2892 R_00B130_SPI_SHADER_USER_DATA_VS_0);
2893 si_pm4_set_state(sctx, vs_sampler, pm4);
2894 }
2895
2896 static void si_bind_gs_sampler_states(struct pipe_context *ctx, unsigned count, void **states)
2897 {
2898 struct si_context *sctx = (struct si_context *)ctx;
2899 struct si_pm4_state *pm4;
2900
2901 pm4 = si_set_sampler_states(sctx, count, states, &sctx->samplers[PIPE_SHADER_GEOMETRY],
2902 R_00B230_SPI_SHADER_USER_DATA_GS_0);
2903 si_pm4_set_state(sctx, gs_sampler, pm4);
2904 }
2905
2906 static void si_bind_ps_sampler_states(struct pipe_context *ctx, unsigned count, void **states)
2907 {
2908 struct si_context *sctx = (struct si_context *)ctx;
2909 struct si_pm4_state *pm4;
2910
2911 pm4 = si_set_sampler_states(sctx, count, states, &sctx->samplers[PIPE_SHADER_FRAGMENT],
2912 R_00B030_SPI_SHADER_USER_DATA_PS_0);
2913 si_pm4_set_state(sctx, ps_sampler, pm4);
2914 }
2915
2916
2917 static void si_bind_sampler_states(struct pipe_context *ctx, unsigned shader,
2918 unsigned start, unsigned count,
2919 void **states)
2920 {
2921 assert(start == 0);
2922
2923 switch (shader) {
2924 case PIPE_SHADER_VERTEX:
2925 si_bind_vs_sampler_states(ctx, count, states);
2926 break;
2927 case PIPE_SHADER_GEOMETRY:
2928 si_bind_gs_sampler_states(ctx, count, states);
2929 break;
2930 case PIPE_SHADER_FRAGMENT:
2931 si_bind_ps_sampler_states(ctx, count, states);
2932 break;
2933 default:
2934 ;
2935 }
2936 }
2937
2938
2939
2940 static void si_set_sample_mask(struct pipe_context *ctx, unsigned sample_mask)
2941 {
2942 struct si_context *sctx = (struct si_context *)ctx;
2943 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
2944 uint16_t mask = sample_mask;
2945
2946 if (pm4 == NULL)
2947 return;
2948
2949 si_pm4_set_reg(pm4, R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0, mask | (mask << 16));
2950 si_pm4_set_reg(pm4, R_028C3C_PA_SC_AA_MASK_X0Y1_X1Y1, mask | (mask << 16));
2951
2952 si_pm4_set_state(sctx, sample_mask, pm4);
2953 }
2954
2955 static void si_delete_sampler_state(struct pipe_context *ctx, void *state)
2956 {
2957 free(state);
2958 }
2959
2960 /*
2961 * Vertex elements & buffers
2962 */
2963
2964 static void *si_create_vertex_elements(struct pipe_context *ctx,
2965 unsigned count,
2966 const struct pipe_vertex_element *elements)
2967 {
2968 struct si_vertex_element *v = CALLOC_STRUCT(si_vertex_element);
2969 int i;
2970
2971 assert(count < PIPE_MAX_ATTRIBS);
2972 if (!v)
2973 return NULL;
2974
2975 v->count = count;
2976 for (i = 0; i < count; ++i) {
2977 const struct util_format_description *desc;
2978 unsigned data_format, num_format;
2979 int first_non_void;
2980
2981 desc = util_format_description(elements[i].src_format);
2982 first_non_void = util_format_get_first_non_void_channel(elements[i].src_format);
2983 data_format = si_translate_buffer_dataformat(ctx->screen, desc, first_non_void);
2984 num_format = si_translate_buffer_numformat(ctx->screen, desc, first_non_void);
2985
2986 v->rsrc_word3[i] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
2987 S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
2988 S_008F0C_DST_SEL_Z(si_map_swizzle(desc->swizzle[2])) |
2989 S_008F0C_DST_SEL_W(si_map_swizzle(desc->swizzle[3])) |
2990 S_008F0C_NUM_FORMAT(num_format) |
2991 S_008F0C_DATA_FORMAT(data_format);
2992 }
2993 memcpy(v->elements, elements, sizeof(struct pipe_vertex_element) * count);
2994
2995 return v;
2996 }
2997
2998 static void si_bind_vertex_elements(struct pipe_context *ctx, void *state)
2999 {
3000 struct si_context *sctx = (struct si_context *)ctx;
3001 struct si_vertex_element *v = (struct si_vertex_element*)state;
3002
3003 sctx->vertex_elements = v;
3004 }
3005
3006 static void si_delete_vertex_element(struct pipe_context *ctx, void *state)
3007 {
3008 struct si_context *sctx = (struct si_context *)ctx;
3009
3010 if (sctx->vertex_elements == state)
3011 sctx->vertex_elements = NULL;
3012 FREE(state);
3013 }
3014
3015 static void si_set_vertex_buffers(struct pipe_context *ctx, unsigned start_slot, unsigned count,
3016 const struct pipe_vertex_buffer *buffers)
3017 {
3018 struct si_context *sctx = (struct si_context *)ctx;
3019
3020 util_set_vertex_buffers_count(sctx->vertex_buffer, &sctx->nr_vertex_buffers, buffers, start_slot, count);
3021 }
3022
3023 static void si_set_index_buffer(struct pipe_context *ctx,
3024 const struct pipe_index_buffer *ib)
3025 {
3026 struct si_context *sctx = (struct si_context *)ctx;
3027
3028 if (ib) {
3029 pipe_resource_reference(&sctx->index_buffer.buffer, ib->buffer);
3030 memcpy(&sctx->index_buffer, ib, sizeof(*ib));
3031 } else {
3032 pipe_resource_reference(&sctx->index_buffer.buffer, NULL);
3033 }
3034 }
3035
3036 /*
3037 * Misc
3038 */
3039 static void si_set_polygon_stipple(struct pipe_context *ctx,
3040 const struct pipe_poly_stipple *state)
3041 {
3042 }
3043
3044 static void si_texture_barrier(struct pipe_context *ctx)
3045 {
3046 struct si_context *sctx = (struct si_context *)ctx;
3047
3048 sctx->b.flags |= R600_CONTEXT_INV_TEX_CACHE |
3049 R600_CONTEXT_FLUSH_AND_INV_CB;
3050 }
3051
3052 static void *si_create_blend_custom(struct si_context *sctx, unsigned mode)
3053 {
3054 struct pipe_blend_state blend;
3055
3056 memset(&blend, 0, sizeof(blend));
3057 blend.independent_blend_enable = true;
3058 blend.rt[0].colormask = 0xf;
3059 return si_create_blend_state_mode(&sctx->b.b, &blend, mode);
3060 }
3061
3062 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
3063 struct pipe_resource *texture,
3064 const struct pipe_surface *surf_tmpl)
3065 {
3066 struct r600_texture *rtex = (struct r600_texture*)texture;
3067 struct si_surface *surface = CALLOC_STRUCT(si_surface);
3068 unsigned level = surf_tmpl->u.tex.level;
3069
3070 if (surface == NULL)
3071 return NULL;
3072
3073 assert(surf_tmpl->u.tex.first_layer <= util_max_layer(texture, surf_tmpl->u.tex.level));
3074 assert(surf_tmpl->u.tex.last_layer <= util_max_layer(texture, surf_tmpl->u.tex.level));
3075
3076 pipe_reference_init(&surface->base.reference, 1);
3077 pipe_resource_reference(&surface->base.texture, texture);
3078 surface->base.context = pipe;
3079 surface->base.format = surf_tmpl->format;
3080 surface->base.width = rtex->surface.level[level].npix_x;
3081 surface->base.height = rtex->surface.level[level].npix_y;
3082 surface->base.texture = texture;
3083 surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
3084 surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
3085 surface->base.u.tex.level = level;
3086
3087 return &surface->base;
3088 }
3089
3090 static void r600_surface_destroy(struct pipe_context *pipe,
3091 struct pipe_surface *surface)
3092 {
3093 pipe_resource_reference(&surface->texture, NULL);
3094 FREE(surface);
3095 }
3096
3097 static boolean si_dma_copy(struct pipe_context *ctx,
3098 struct pipe_resource *dst,
3099 unsigned dst_level,
3100 unsigned dst_x, unsigned dst_y, unsigned dst_z,
3101 struct pipe_resource *src,
3102 unsigned src_level,
3103 const struct pipe_box *src_box)
3104 {
3105 /* XXX implement this or share evergreen_dma_blit with r600g */
3106 return FALSE;
3107 }
3108
3109 static void si_set_occlusion_query_state(struct pipe_context *ctx, bool enable)
3110 {
3111 /* XXX Turn this into a proper state. Right now the queries are
3112 * enabled in draw_vbo, which snoops r600_common_context to see
3113 * if any occlusion queries are active. */
3114 }
3115
3116 static void si_need_gfx_cs_space(struct pipe_context *ctx, unsigned num_dw,
3117 bool include_draw_vbo)
3118 {
3119 si_need_cs_space((struct si_context*)ctx, num_dw, include_draw_vbo);
3120 }
3121
3122 void si_init_state_functions(struct si_context *sctx)
3123 {
3124 int i;
3125
3126 sctx->b.b.create_blend_state = si_create_blend_state;
3127 sctx->b.b.bind_blend_state = si_bind_blend_state;
3128 sctx->b.b.delete_blend_state = si_delete_blend_state;
3129 sctx->b.b.set_blend_color = si_set_blend_color;
3130
3131 sctx->b.b.create_rasterizer_state = si_create_rs_state;
3132 sctx->b.b.bind_rasterizer_state = si_bind_rs_state;
3133 sctx->b.b.delete_rasterizer_state = si_delete_rs_state;
3134
3135 sctx->b.b.create_depth_stencil_alpha_state = si_create_dsa_state;
3136 sctx->b.b.bind_depth_stencil_alpha_state = si_bind_dsa_state;
3137 sctx->b.b.delete_depth_stencil_alpha_state = si_delete_dsa_state;
3138
3139 for (i = 0; i < 8; i++) {
3140 sctx->custom_dsa_flush_depth_stencil[i] = si_create_db_flush_dsa(sctx, true, true, i);
3141 sctx->custom_dsa_flush_depth[i] = si_create_db_flush_dsa(sctx, true, false, i);
3142 sctx->custom_dsa_flush_stencil[i] = si_create_db_flush_dsa(sctx, false, true, i);
3143 }
3144 sctx->custom_dsa_flush_inplace = si_create_db_flush_dsa(sctx, false, false, 0);
3145 sctx->custom_blend_resolve = si_create_blend_custom(sctx, V_028808_CB_RESOLVE);
3146 sctx->custom_blend_decompress = si_create_blend_custom(sctx, V_028808_CB_FMASK_DECOMPRESS);
3147
3148 sctx->b.b.set_clip_state = si_set_clip_state;
3149 sctx->b.b.set_scissor_states = si_set_scissor_states;
3150 sctx->b.b.set_viewport_states = si_set_viewport_states;
3151 sctx->b.b.set_stencil_ref = si_set_pipe_stencil_ref;
3152
3153 sctx->b.b.set_framebuffer_state = si_set_framebuffer_state;
3154 sctx->b.b.get_sample_position = si_get_sample_position;
3155
3156 sctx->b.b.create_vs_state = si_create_vs_state;
3157 sctx->b.b.create_fs_state = si_create_fs_state;
3158 sctx->b.b.bind_vs_state = si_bind_vs_shader;
3159 sctx->b.b.bind_fs_state = si_bind_ps_shader;
3160 sctx->b.b.delete_vs_state = si_delete_vs_shader;
3161 sctx->b.b.delete_fs_state = si_delete_ps_shader;
3162 #if HAVE_LLVM >= 0x0305
3163 sctx->b.b.create_gs_state = si_create_gs_state;
3164 sctx->b.b.bind_gs_state = si_bind_gs_shader;
3165 sctx->b.b.delete_gs_state = si_delete_gs_shader;
3166 #endif
3167
3168 sctx->b.b.create_sampler_state = si_create_sampler_state;
3169 sctx->b.b.bind_sampler_states = si_bind_sampler_states;
3170 sctx->b.b.delete_sampler_state = si_delete_sampler_state;
3171
3172 sctx->b.b.create_sampler_view = si_create_sampler_view;
3173 sctx->b.b.set_sampler_views = si_set_sampler_views;
3174 sctx->b.b.sampler_view_destroy = si_sampler_view_destroy;
3175
3176 sctx->b.b.set_sample_mask = si_set_sample_mask;
3177
3178 sctx->b.b.create_vertex_elements_state = si_create_vertex_elements;
3179 sctx->b.b.bind_vertex_elements_state = si_bind_vertex_elements;
3180 sctx->b.b.delete_vertex_elements_state = si_delete_vertex_element;
3181 sctx->b.b.set_vertex_buffers = si_set_vertex_buffers;
3182 sctx->b.b.set_index_buffer = si_set_index_buffer;
3183
3184 sctx->b.b.texture_barrier = si_texture_barrier;
3185 sctx->b.b.set_polygon_stipple = si_set_polygon_stipple;
3186 sctx->b.b.create_surface = r600_create_surface;
3187 sctx->b.b.surface_destroy = r600_surface_destroy;
3188 sctx->b.dma_copy = si_dma_copy;
3189 sctx->b.set_occlusion_query_state = si_set_occlusion_query_state;
3190 sctx->b.need_gfx_cs_space = si_need_gfx_cs_space;
3191
3192 sctx->b.b.draw_vbo = si_draw_vbo;
3193 }
3194
3195 void si_init_config(struct si_context *sctx)
3196 {
3197 struct si_pm4_state *pm4 = si_pm4_alloc_state(sctx);
3198
3199 if (pm4 == NULL)
3200 return;
3201
3202 si_cmd_context_control(pm4);
3203
3204 si_pm4_set_reg(pm4, R_028A4C_PA_SC_MODE_CNTL_1, 0x0);
3205
3206 si_pm4_set_reg(pm4, R_028A10_VGT_OUTPUT_PATH_CNTL, 0x0);
3207 si_pm4_set_reg(pm4, R_028A14_VGT_HOS_CNTL, 0x0);
3208 si_pm4_set_reg(pm4, R_028A18_VGT_HOS_MAX_TESS_LEVEL, 0x0);
3209 si_pm4_set_reg(pm4, R_028A1C_VGT_HOS_MIN_TESS_LEVEL, 0x0);
3210 si_pm4_set_reg(pm4, R_028A20_VGT_HOS_REUSE_DEPTH, 0x0);
3211 si_pm4_set_reg(pm4, R_028A24_VGT_GROUP_PRIM_TYPE, 0x0);
3212 si_pm4_set_reg(pm4, R_028A28_VGT_GROUP_FIRST_DECR, 0x0);
3213 si_pm4_set_reg(pm4, R_028A2C_VGT_GROUP_DECR, 0x0);
3214 si_pm4_set_reg(pm4, R_028A30_VGT_GROUP_VECT_0_CNTL, 0x0);
3215 si_pm4_set_reg(pm4, R_028A34_VGT_GROUP_VECT_1_CNTL, 0x0);
3216 si_pm4_set_reg(pm4, R_028A38_VGT_GROUP_VECT_0_FMT_CNTL, 0x0);
3217 si_pm4_set_reg(pm4, R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL, 0x0);
3218
3219 /* FIXME calculate these values somehow ??? */
3220 si_pm4_set_reg(pm4, R_028A54_VGT_GS_PER_ES, 0x80);
3221 si_pm4_set_reg(pm4, R_028A58_VGT_ES_PER_GS, 0x40);
3222 si_pm4_set_reg(pm4, R_028A5C_VGT_GS_PER_VS, 0x2);
3223
3224 si_pm4_set_reg(pm4, R_028A84_VGT_PRIMITIVEID_EN, 0x0);
3225 si_pm4_set_reg(pm4, R_028A8C_VGT_PRIMITIVEID_RESET, 0x0);
3226 si_pm4_set_reg(pm4, R_028AB8_VGT_VTX_CNT_EN, 0);
3227 si_pm4_set_reg(pm4, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
3228
3229 si_pm4_set_reg(pm4, R_028B60_VGT_GS_VERT_ITEMSIZE_1, 0);
3230 si_pm4_set_reg(pm4, R_028B64_VGT_GS_VERT_ITEMSIZE_2, 0);
3231 si_pm4_set_reg(pm4, R_028B68_VGT_GS_VERT_ITEMSIZE_3, 0);
3232 si_pm4_set_reg(pm4, R_028B90_VGT_GS_INSTANCE_CNT, 0);
3233
3234 si_pm4_set_reg(pm4, R_028B94_VGT_STRMOUT_CONFIG, 0x0);
3235 si_pm4_set_reg(pm4, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0x0);
3236 if (sctx->b.chip_class == SI) {
3237 si_pm4_set_reg(pm4, R_028AA8_IA_MULTI_VGT_PARAM,
3238 S_028AA8_SWITCH_ON_EOP(1) |
3239 S_028AA8_PARTIAL_VS_WAVE_ON(1) |
3240 S_028AA8_PRIMGROUP_SIZE(63));
3241 }
3242 si_pm4_set_reg(pm4, R_028AB4_VGT_REUSE_OFF, 0x00000000);
3243 si_pm4_set_reg(pm4, R_028AB8_VGT_VTX_CNT_EN, 0x0);
3244 if (sctx->b.chip_class < CIK)
3245 si_pm4_set_reg(pm4, R_008A14_PA_CL_ENHANCE, S_008A14_NUM_CLIP_SEQ(3) |
3246 S_008A14_CLIP_VTX_REORDER_ENA(1));
3247
3248 si_pm4_set_reg(pm4, R_028BD4_PA_SC_CENTROID_PRIORITY_0, 0x76543210);
3249 si_pm4_set_reg(pm4, R_028BD8_PA_SC_CENTROID_PRIORITY_1, 0xfedcba98);
3250
3251 si_pm4_set_reg(pm4, R_02882C_PA_SU_PRIM_FILTER_CNTL, 0);
3252
3253 if (sctx->b.chip_class >= CIK) {
3254 switch (sctx->screen->b.family) {
3255 case CHIP_BONAIRE:
3256 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x16000012);
3257 si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1, 0x00000000);
3258 break;
3259 case CHIP_HAWAII:
3260 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x3a00161a);
3261 si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1, 0x0000002e);
3262 break;
3263 case CHIP_KAVERI:
3264 /* XXX todo */
3265 case CHIP_KABINI:
3266 /* XXX todo */
3267 default:
3268 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x00000000);
3269 si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1, 0x00000000);
3270 break;
3271 }
3272 } else {
3273 switch (sctx->screen->b.family) {
3274 case CHIP_TAHITI:
3275 case CHIP_PITCAIRN:
3276 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x2a00126a);
3277 break;
3278 case CHIP_VERDE:
3279 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x0000124a);
3280 break;
3281 case CHIP_OLAND:
3282 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x00000082);
3283 break;
3284 case CHIP_HAINAN:
3285 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x00000000);
3286 break;
3287 default:
3288 si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, 0x00000000);
3289 break;
3290 }
3291 }
3292
3293 si_pm4_set_reg(pm4, R_028204_PA_SC_WINDOW_SCISSOR_TL, S_028204_WINDOW_OFFSET_DISABLE(1));
3294 si_pm4_set_reg(pm4, R_028240_PA_SC_GENERIC_SCISSOR_TL, S_028240_WINDOW_OFFSET_DISABLE(1));
3295 si_pm4_set_reg(pm4, R_028244_PA_SC_GENERIC_SCISSOR_BR,
3296 S_028244_BR_X(16384) | S_028244_BR_Y(16384));
3297 si_pm4_set_reg(pm4, R_028030_PA_SC_SCREEN_SCISSOR_TL, 0);
3298 si_pm4_set_reg(pm4, R_028034_PA_SC_SCREEN_SCISSOR_BR,
3299 S_028034_BR_X(16384) | S_028034_BR_Y(16384));
3300
3301 si_pm4_set_reg(pm4, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
3302 si_pm4_set_reg(pm4, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
3303 si_pm4_set_reg(pm4, R_0282D0_PA_SC_VPORT_ZMIN_0, 0x00000000);
3304 si_pm4_set_reg(pm4, R_0282D4_PA_SC_VPORT_ZMAX_0, 0x3F800000);
3305 si_pm4_set_reg(pm4, R_028818_PA_CL_VTE_CNTL, 0x0000043F);
3306 si_pm4_set_reg(pm4, R_028820_PA_CL_NANINF_CNTL, 0x00000000);
3307 si_pm4_set_reg(pm4, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, 0x3F800000);
3308 si_pm4_set_reg(pm4, R_028BEC_PA_CL_GB_VERT_DISC_ADJ, 0x3F800000);
3309 si_pm4_set_reg(pm4, R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ, 0x3F800000);
3310 si_pm4_set_reg(pm4, R_028BF4_PA_CL_GB_HORZ_DISC_ADJ, 0x3F800000);
3311 si_pm4_set_reg(pm4, R_028020_DB_DEPTH_BOUNDS_MIN, 0x00000000);
3312 si_pm4_set_reg(pm4, R_028024_DB_DEPTH_BOUNDS_MAX, 0x00000000);
3313 si_pm4_set_reg(pm4, R_028028_DB_STENCIL_CLEAR, 0x00000000);
3314 si_pm4_set_reg(pm4, R_02802C_DB_DEPTH_CLEAR, 0x3F800000);
3315 si_pm4_set_reg(pm4, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 0x0);
3316 si_pm4_set_reg(pm4, R_028AC4_DB_SRESULTS_COMPARE_STATE1, 0x0);
3317 si_pm4_set_reg(pm4, R_028AC8_DB_PRELOAD_CONTROL, 0x0);
3318 si_pm4_set_reg(pm4, R_02800C_DB_RENDER_OVERRIDE,
3319 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE) |
3320 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE));
3321 si_pm4_set_reg(pm4, R_028400_VGT_MAX_VTX_INDX, ~0);
3322 si_pm4_set_reg(pm4, R_028404_VGT_MIN_VTX_INDX, 0);
3323
3324 if (sctx->b.chip_class >= CIK) {
3325 si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS, S_00B118_CU_EN(0xffff));
3326 si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(0));
3327 si_pm4_set_reg(pm4, R_00B01C_SPI_SHADER_PGM_RSRC3_PS, S_00B01C_CU_EN(0xffff));
3328 }
3329
3330 si_pm4_set_state(sctx, init, pm4);
3331 }