1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 */
23 #include "r600_formats.h"
24 #include "r600_shader.h"
25 #include "r600d.h"
26
27 #include "pipe/p_shader_tokens.h"
28 #include "util/u_pack_color.h"
29 #include "util/u_memory.h"
30 #include "util/u_framebuffer.h"
31 #include "util/u_dual_blend.h"
32
r600_translate_blend_function(int blend_func)33 static uint32_t r600_translate_blend_function(int blend_func)
34 {
35 switch (blend_func) {
36 case PIPE_BLEND_ADD:
37 return V_028804_COMB_DST_PLUS_SRC;
38 case PIPE_BLEND_SUBTRACT:
39 return V_028804_COMB_SRC_MINUS_DST;
40 case PIPE_BLEND_REVERSE_SUBTRACT:
41 return V_028804_COMB_DST_MINUS_SRC;
42 case PIPE_BLEND_MIN:
43 return V_028804_COMB_MIN_DST_SRC;
44 case PIPE_BLEND_MAX:
45 return V_028804_COMB_MAX_DST_SRC;
46 default:
47 R600_ERR("Unknown blend function %d\n", blend_func);
48 assert(0);
49 break;
50 }
51 return 0;
52 }
53
r600_translate_blend_factor(int blend_fact)54 static uint32_t r600_translate_blend_factor(int blend_fact)
55 {
56 switch (blend_fact) {
57 case PIPE_BLENDFACTOR_ONE:
58 return V_028804_BLEND_ONE;
59 case PIPE_BLENDFACTOR_SRC_COLOR:
60 return V_028804_BLEND_SRC_COLOR;
61 case PIPE_BLENDFACTOR_SRC_ALPHA:
62 return V_028804_BLEND_SRC_ALPHA;
63 case PIPE_BLENDFACTOR_DST_ALPHA:
64 return V_028804_BLEND_DST_ALPHA;
65 case PIPE_BLENDFACTOR_DST_COLOR:
66 return V_028804_BLEND_DST_COLOR;
67 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
68 return V_028804_BLEND_SRC_ALPHA_SATURATE;
69 case PIPE_BLENDFACTOR_CONST_COLOR:
70 return V_028804_BLEND_CONST_COLOR;
71 case PIPE_BLENDFACTOR_CONST_ALPHA:
72 return V_028804_BLEND_CONST_ALPHA;
73 case PIPE_BLENDFACTOR_ZERO:
74 return V_028804_BLEND_ZERO;
75 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
76 return V_028804_BLEND_ONE_MINUS_SRC_COLOR;
77 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
78 return V_028804_BLEND_ONE_MINUS_SRC_ALPHA;
79 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
80 return V_028804_BLEND_ONE_MINUS_DST_ALPHA;
81 case PIPE_BLENDFACTOR_INV_DST_COLOR:
82 return V_028804_BLEND_ONE_MINUS_DST_COLOR;
83 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
84 return V_028804_BLEND_ONE_MINUS_CONST_COLOR;
85 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
86 return V_028804_BLEND_ONE_MINUS_CONST_ALPHA;
87 case PIPE_BLENDFACTOR_SRC1_COLOR:
88 return V_028804_BLEND_SRC1_COLOR;
89 case PIPE_BLENDFACTOR_SRC1_ALPHA:
90 return V_028804_BLEND_SRC1_ALPHA;
91 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
92 return V_028804_BLEND_INV_SRC1_COLOR;
93 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
94 return V_028804_BLEND_INV_SRC1_ALPHA;
95 default:
96 R600_ERR("Bad blend factor %d not supported!\n", blend_fact);
97 assert(0);
98 break;
99 }
100 return 0;
101 }
102
r600_tex_dim(unsigned dim,unsigned nr_samples)103 static unsigned r600_tex_dim(unsigned dim, unsigned nr_samples)
104 {
105 switch (dim) {
106 default:
107 case PIPE_TEXTURE_1D:
108 return V_038000_SQ_TEX_DIM_1D;
109 case PIPE_TEXTURE_1D_ARRAY:
110 return V_038000_SQ_TEX_DIM_1D_ARRAY;
111 case PIPE_TEXTURE_2D:
112 case PIPE_TEXTURE_RECT:
113 return nr_samples > 1 ? V_038000_SQ_TEX_DIM_2D_MSAA :
114 V_038000_SQ_TEX_DIM_2D;
115 case PIPE_TEXTURE_2D_ARRAY:
116 return nr_samples > 1 ? V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA :
117 V_038000_SQ_TEX_DIM_2D_ARRAY;
118 case PIPE_TEXTURE_3D:
119 return V_038000_SQ_TEX_DIM_3D;
120 case PIPE_TEXTURE_CUBE:
121 case PIPE_TEXTURE_CUBE_ARRAY:
122 return V_038000_SQ_TEX_DIM_CUBEMAP;
123 }
124 }
125
r600_translate_dbformat(enum pipe_format format)126 static uint32_t r600_translate_dbformat(enum pipe_format format)
127 {
128 switch (format) {
129 case PIPE_FORMAT_Z16_UNORM:
130 return V_028010_DEPTH_16;
131 case PIPE_FORMAT_Z24X8_UNORM:
132 return V_028010_DEPTH_X8_24;
133 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
134 return V_028010_DEPTH_8_24;
135 case PIPE_FORMAT_Z32_FLOAT:
136 return V_028010_DEPTH_32_FLOAT;
137 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
138 return V_028010_DEPTH_X24_8_32_FLOAT;
139 default:
140 return ~0U;
141 }
142 }
143
r600_is_sampler_format_supported(struct pipe_screen * screen,enum pipe_format format)144 static bool r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
145 {
146 return r600_translate_texformat(screen, format, NULL, NULL, NULL,
147 FALSE) != ~0U;
148 }
149
r600_is_colorbuffer_format_supported(enum chip_class chip,enum pipe_format format)150 static bool r600_is_colorbuffer_format_supported(enum chip_class chip, enum pipe_format format)
151 {
152 return r600_translate_colorformat(chip, format, FALSE) != ~0U &&
153 r600_translate_colorswap(format, FALSE) != ~0U;
154 }
155
r600_is_zs_format_supported(enum pipe_format format)156 static bool r600_is_zs_format_supported(enum pipe_format format)
157 {
158 return r600_translate_dbformat(format) != ~0U;
159 }
160
r600_is_format_supported(struct pipe_screen * screen,enum pipe_format format,enum pipe_texture_target target,unsigned sample_count,unsigned usage)161 boolean r600_is_format_supported(struct pipe_screen *screen,
162 enum pipe_format format,
163 enum pipe_texture_target target,
164 unsigned sample_count,
165 unsigned usage)
166 {
167 struct r600_screen *rscreen = (struct r600_screen*)screen;
168 unsigned retval = 0;
169
170 if (target >= PIPE_MAX_TEXTURE_TYPES) {
171 R600_ERR("r600: unsupported texture type %d\n", target);
172 return FALSE;
173 }
174
175 if (!util_format_is_supported(format, usage))
176 return FALSE;
177
178 if (sample_count > 1) {
179 if (!rscreen->has_msaa)
180 return FALSE;
181
182 /* R11G11B10 is broken on R6xx. */
183 if (rscreen->b.chip_class == R600 &&
184 format == PIPE_FORMAT_R11G11B10_FLOAT)
185 return FALSE;
186
187 /* MSAA integer colorbuffers hang. */
188 if (util_format_is_pure_integer(format) &&
189 !util_format_is_depth_or_stencil(format))
190 return FALSE;
191
192 switch (sample_count) {
193 case 2:
194 case 4:
195 case 8:
196 break;
197 default:
198 return FALSE;
199 }
200 }
201
202 if (usage & PIPE_BIND_SAMPLER_VIEW) {
203 if (target == PIPE_BUFFER) {
204 if (r600_is_vertex_format_supported(format))
205 retval |= PIPE_BIND_SAMPLER_VIEW;
206 } else {
207 if (r600_is_sampler_format_supported(screen, format))
208 retval |= PIPE_BIND_SAMPLER_VIEW;
209 }
210 }
211
212 if ((usage & (PIPE_BIND_RENDER_TARGET |
213 PIPE_BIND_DISPLAY_TARGET |
214 PIPE_BIND_SCANOUT |
215 PIPE_BIND_SHARED |
216 PIPE_BIND_BLENDABLE)) &&
217 r600_is_colorbuffer_format_supported(rscreen->b.chip_class, format)) {
218 retval |= usage &
219 (PIPE_BIND_RENDER_TARGET |
220 PIPE_BIND_DISPLAY_TARGET |
221 PIPE_BIND_SCANOUT |
222 PIPE_BIND_SHARED);
223 if (!util_format_is_pure_integer(format) &&
224 !util_format_is_depth_or_stencil(format))
225 retval |= usage & PIPE_BIND_BLENDABLE;
226 }
227
228 if ((usage & PIPE_BIND_DEPTH_STENCIL) &&
229 r600_is_zs_format_supported(format)) {
230 retval |= PIPE_BIND_DEPTH_STENCIL;
231 }
232
233 if ((usage & PIPE_BIND_VERTEX_BUFFER) &&
234 r600_is_vertex_format_supported(format)) {
235 retval |= PIPE_BIND_VERTEX_BUFFER;
236 }
237
238 if ((usage & PIPE_BIND_LINEAR) &&
239 !util_format_is_compressed(format) &&
240 !(usage & PIPE_BIND_DEPTH_STENCIL))
241 retval |= PIPE_BIND_LINEAR;
242
243 return retval == usage;
244 }
245
r600_emit_polygon_offset(struct r600_context * rctx,struct r600_atom * a)246 static void r600_emit_polygon_offset(struct r600_context *rctx, struct r600_atom *a)
247 {
248 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
249 struct r600_poly_offset_state *state = (struct r600_poly_offset_state*)a;
250 float offset_units = state->offset_units;
251 float offset_scale = state->offset_scale;
252 uint32_t pa_su_poly_offset_db_fmt_cntl = 0;
253
254 if (!state->offset_units_unscaled) {
255 switch (state->zs_format) {
256 case PIPE_FORMAT_Z24X8_UNORM:
257 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
258 offset_units *= 2.0f;
259 pa_su_poly_offset_db_fmt_cntl =
260 S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-24);
261 break;
262 case PIPE_FORMAT_Z16_UNORM:
263 offset_units *= 4.0f;
264 pa_su_poly_offset_db_fmt_cntl =
265 S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-16);
266 break;
267 default:
268 pa_su_poly_offset_db_fmt_cntl =
269 S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-23) |
270 S_028DF8_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
271 }
272 }
273
274 radeon_set_context_reg_seq(cs, R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 4);
275 radeon_emit(cs, fui(offset_scale));
276 radeon_emit(cs, fui(offset_units));
277 radeon_emit(cs, fui(offset_scale));
278 radeon_emit(cs, fui(offset_units));
279
280 radeon_set_context_reg(cs, R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
281 pa_su_poly_offset_db_fmt_cntl);
282 }
283
r600_get_blend_control(const struct pipe_blend_state * state,unsigned i)284 static uint32_t r600_get_blend_control(const struct pipe_blend_state *state, unsigned i)
285 {
286 int j = state->independent_blend_enable ? i : 0;
287
288 unsigned eqRGB = state->rt[j].rgb_func;
289 unsigned srcRGB = state->rt[j].rgb_src_factor;
290 unsigned dstRGB = state->rt[j].rgb_dst_factor;
291
292 unsigned eqA = state->rt[j].alpha_func;
293 unsigned srcA = state->rt[j].alpha_src_factor;
294 unsigned dstA = state->rt[j].alpha_dst_factor;
295 uint32_t bc = 0;
296
297 if (!state->rt[j].blend_enable)
298 return 0;
299
300 bc |= S_028804_COLOR_COMB_FCN(r600_translate_blend_function(eqRGB));
301 bc |= S_028804_COLOR_SRCBLEND(r600_translate_blend_factor(srcRGB));
302 bc |= S_028804_COLOR_DESTBLEND(r600_translate_blend_factor(dstRGB));
303
304 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
305 bc |= S_028804_SEPARATE_ALPHA_BLEND(1);
306 bc |= S_028804_ALPHA_COMB_FCN(r600_translate_blend_function(eqA));
307 bc |= S_028804_ALPHA_SRCBLEND(r600_translate_blend_factor(srcA));
308 bc |= S_028804_ALPHA_DESTBLEND(r600_translate_blend_factor(dstA));
309 }
310 return bc;
311 }
312
r600_create_blend_state_mode(struct pipe_context * ctx,const struct pipe_blend_state * state,int mode)313 static void *r600_create_blend_state_mode(struct pipe_context *ctx,
314 const struct pipe_blend_state *state,
315 int mode)
316 {
317 struct r600_context *rctx = (struct r600_context *)ctx;
318 uint32_t color_control = 0, target_mask = 0;
319 struct r600_blend_state *blend = CALLOC_STRUCT(r600_blend_state);
320
321 if (!blend) {
322 return NULL;
323 }
324
325 r600_init_command_buffer(&blend->buffer, 20);
326 r600_init_command_buffer(&blend->buffer_no_blend, 20);
327
328 /* R600 does not support per-MRT blends */
329 if (rctx->b.family > CHIP_R600)
330 color_control |= S_028808_PER_MRT_BLEND(1);
331
332 if (state->logicop_enable) {
333 color_control |= (state->logicop_func << 16) | (state->logicop_func << 20);
334 } else {
335 color_control |= (0xcc << 16);
336 }
337 /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
338 if (state->independent_blend_enable) {
339 for (int i = 0; i < 8; i++) {
340 if (state->rt[i].blend_enable) {
341 color_control |= S_028808_TARGET_BLEND_ENABLE(1 << i);
342 }
343 target_mask |= (state->rt[i].colormask << (4 * i));
344 }
345 } else {
346 for (int i = 0; i < 8; i++) {
347 if (state->rt[0].blend_enable) {
348 color_control |= S_028808_TARGET_BLEND_ENABLE(1 << i);
349 }
350 target_mask |= (state->rt[0].colormask << (4 * i));
351 }
352 }
353
354 if (target_mask)
355 color_control |= S_028808_SPECIAL_OP(mode);
356 else
357 color_control |= S_028808_SPECIAL_OP(V_028808_DISABLE);
358
359 /* only MRT0 has dual src blend */
360 blend->dual_src_blend = util_blend_state_is_dual(state, 0);
361 blend->cb_target_mask = target_mask;
362 blend->cb_color_control = color_control;
363 blend->cb_color_control_no_blend = color_control & C_028808_TARGET_BLEND_ENABLE;
364 blend->alpha_to_one = state->alpha_to_one;
365
366 r600_store_context_reg(&blend->buffer, R_028D44_DB_ALPHA_TO_MASK,
367 S_028D44_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
368 S_028D44_ALPHA_TO_MASK_OFFSET0(2) |
369 S_028D44_ALPHA_TO_MASK_OFFSET1(2) |
370 S_028D44_ALPHA_TO_MASK_OFFSET2(2) |
371 S_028D44_ALPHA_TO_MASK_OFFSET3(2));
372
373 /* Copy over the registers set so far into buffer_no_blend. */
374 memcpy(blend->buffer_no_blend.buf, blend->buffer.buf, blend->buffer.num_dw * 4);
375 blend->buffer_no_blend.num_dw = blend->buffer.num_dw;
376
377 /* Only add blend registers if blending is enabled. */
378 if (!G_028808_TARGET_BLEND_ENABLE(color_control)) {
379 return blend;
380 }
381
382 /* The first R600 does not support per-MRT blends */
383 r600_store_context_reg(&blend->buffer, R_028804_CB_BLEND_CONTROL,
384 r600_get_blend_control(state, 0));
385
386 if (rctx->b.family > CHIP_R600) {
387 r600_store_context_reg_seq(&blend->buffer, R_028780_CB_BLEND0_CONTROL, 8);
388 for (int i = 0; i < 8; i++) {
389 r600_store_value(&blend->buffer, r600_get_blend_control(state, i));
390 }
391 }
392 return blend;
393 }
394
r600_create_blend_state(struct pipe_context * ctx,const struct pipe_blend_state * state)395 static void *r600_create_blend_state(struct pipe_context *ctx,
396 const struct pipe_blend_state *state)
397 {
398 return r600_create_blend_state_mode(ctx, state, V_028808_SPECIAL_NORMAL);
399 }
400
r600_create_dsa_state(struct pipe_context * ctx,const struct pipe_depth_stencil_alpha_state * state)401 static void *r600_create_dsa_state(struct pipe_context *ctx,
402 const struct pipe_depth_stencil_alpha_state *state)
403 {
404 unsigned db_depth_control, alpha_test_control, alpha_ref;
405 struct r600_dsa_state *dsa = CALLOC_STRUCT(r600_dsa_state);
406
407 if (!dsa) {
408 return NULL;
409 }
410
411 r600_init_command_buffer(&dsa->buffer, 3);
412
413 dsa->valuemask[0] = state->stencil[0].valuemask;
414 dsa->valuemask[1] = state->stencil[1].valuemask;
415 dsa->writemask[0] = state->stencil[0].writemask;
416 dsa->writemask[1] = state->stencil[1].writemask;
417 dsa->zwritemask = state->depth.writemask;
418
419 db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) |
420 S_028800_Z_WRITE_ENABLE(state->depth.writemask) |
421 S_028800_ZFUNC(state->depth.func);
422
423 /* stencil */
424 if (state->stencil[0].enabled) {
425 db_depth_control |= S_028800_STENCIL_ENABLE(1);
426 db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func); /* translates straight */
427 db_depth_control |= S_028800_STENCILFAIL(r600_translate_stencil_op(state->stencil[0].fail_op));
428 db_depth_control |= S_028800_STENCILZPASS(r600_translate_stencil_op(state->stencil[0].zpass_op));
429 db_depth_control |= S_028800_STENCILZFAIL(r600_translate_stencil_op(state->stencil[0].zfail_op));
430
431 if (state->stencil[1].enabled) {
432 db_depth_control |= S_028800_BACKFACE_ENABLE(1);
433 db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func); /* translates straight */
434 db_depth_control |= S_028800_STENCILFAIL_BF(r600_translate_stencil_op(state->stencil[1].fail_op));
435 db_depth_control |= S_028800_STENCILZPASS_BF(r600_translate_stencil_op(state->stencil[1].zpass_op));
436 db_depth_control |= S_028800_STENCILZFAIL_BF(r600_translate_stencil_op(state->stencil[1].zfail_op));
437 }
438 }
439
440 /* alpha */
441 alpha_test_control = 0;
442 alpha_ref = 0;
443 if (state->alpha.enabled) {
444 alpha_test_control = S_028410_ALPHA_FUNC(state->alpha.func);
445 alpha_test_control |= S_028410_ALPHA_TEST_ENABLE(1);
446 alpha_ref = fui(state->alpha.ref_value);
447 }
448 dsa->sx_alpha_test_control = alpha_test_control & 0xff;
449 dsa->alpha_ref = alpha_ref;
450
451 r600_store_context_reg(&dsa->buffer, R_028800_DB_DEPTH_CONTROL, db_depth_control);
452 return dsa;
453 }
454
r600_create_rs_state(struct pipe_context * ctx,const struct pipe_rasterizer_state * state)455 static void *r600_create_rs_state(struct pipe_context *ctx,
456 const struct pipe_rasterizer_state *state)
457 {
458 struct r600_context *rctx = (struct r600_context *)ctx;
459 unsigned tmp, sc_mode_cntl, spi_interp;
460 float psize_min, psize_max;
461 struct r600_rasterizer_state *rs = CALLOC_STRUCT(r600_rasterizer_state);
462
463 if (!rs) {
464 return NULL;
465 }
466
467 r600_init_command_buffer(&rs->buffer, 30);
468
469 rs->scissor_enable = state->scissor;
470 rs->clip_halfz = state->clip_halfz;
471 rs->flatshade = state->flatshade;
472 rs->sprite_coord_enable = state->sprite_coord_enable;
473 rs->two_side = state->light_twoside;
474 rs->clip_plane_enable = state->clip_plane_enable;
475 rs->pa_sc_line_stipple = state->line_stipple_enable ?
476 S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
477 S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
478 rs->pa_cl_clip_cntl =
479 S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) |
480 S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
481 S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
482 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
483 if (rctx->b.chip_class == R700) {
484 rs->pa_cl_clip_cntl |=
485 S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard);
486 }
487 rs->multisample_enable = state->multisample;
488
489 /* offset */
490 rs->offset_units = state->offset_units;
491 rs->offset_scale = state->offset_scale * 16.0f;
492 rs->offset_enable = state->offset_point || state->offset_line || state->offset_tri;
493 rs->offset_units_unscaled = state->offset_units_unscaled;
494
495 if (state->point_size_per_vertex) {
496 psize_min = util_get_min_point_size(state);
497 psize_max = 8192;
498 } else {
499 /* Force the point size to be as if the vertex output was disabled. */
500 psize_min = state->point_size;
501 psize_max = state->point_size;
502 }
503
504 sc_mode_cntl = S_028A4C_MSAA_ENABLE(state->multisample) |
505 S_028A4C_LINE_STIPPLE_ENABLE(state->line_stipple_enable) |
506 S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
507 S_028A4C_PS_ITER_SAMPLE(state->multisample && rctx->ps_iter_samples > 1);
508 if (rctx->b.family == CHIP_RV770) {
509 /* workaround possible rendering corruption on RV770 with hyperz together with sample shading */
510 sc_mode_cntl |= S_028A4C_TILE_COVER_DISABLE(state->multisample && rctx->ps_iter_samples > 1);
511 }
512 if (rctx->b.chip_class >= R700) {
513 sc_mode_cntl |= S_028A4C_FORCE_EOV_REZ_ENABLE(1) |
514 S_028A4C_R700_ZMM_LINE_OFFSET(1) |
515 S_028A4C_R700_VPORT_SCISSOR_ENABLE(1);
516 } else {
517 sc_mode_cntl |= S_028A4C_WALK_ALIGN8_PRIM_FITS_ST(1);
518 }
519
520 spi_interp = S_0286D4_FLAT_SHADE_ENA(1);
521 if (state->sprite_coord_enable) {
522 spi_interp |= S_0286D4_PNT_SPRITE_ENA(1) |
523 S_0286D4_PNT_SPRITE_OVRD_X(2) |
524 S_0286D4_PNT_SPRITE_OVRD_Y(3) |
525 S_0286D4_PNT_SPRITE_OVRD_Z(0) |
526 S_0286D4_PNT_SPRITE_OVRD_W(1);
527 if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) {
528 spi_interp |= S_0286D4_PNT_SPRITE_TOP_1(1);
529 }
530 }
531
532 r600_store_context_reg_seq(&rs->buffer, R_028A00_PA_SU_POINT_SIZE, 3);
533 /* point size 12.4 fixed point (divide by two, because 0.5 = 1 pixel. */
534 tmp = r600_pack_float_12p4(state->point_size/2);
535 r600_store_value(&rs->buffer, /* R_028A00_PA_SU_POINT_SIZE */
536 S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
537 r600_store_value(&rs->buffer, /* R_028A04_PA_SU_POINT_MINMAX */
538 S_028A04_MIN_SIZE(r600_pack_float_12p4(psize_min/2)) |
539 S_028A04_MAX_SIZE(r600_pack_float_12p4(psize_max/2)));
540 r600_store_value(&rs->buffer, /* R_028A08_PA_SU_LINE_CNTL */
541 S_028A08_WIDTH(r600_pack_float_12p4(state->line_width/2)));
542
543 r600_store_context_reg(&rs->buffer, R_0286D4_SPI_INTERP_CONTROL_0, spi_interp);
544 r600_store_context_reg(&rs->buffer, R_028A4C_PA_SC_MODE_CNTL, sc_mode_cntl);
545 r600_store_context_reg(&rs->buffer, R_028C08_PA_SU_VTX_CNTL,
546 S_028C08_PIX_CENTER_HALF(state->half_pixel_center) |
547 S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
548 r600_store_context_reg(&rs->buffer, R_028DFC_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
549
550 rs->pa_su_sc_mode_cntl = S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) |
551 S_028814_CULL_FRONT(state->cull_face & PIPE_FACE_FRONT ? 1 : 0) |
552 S_028814_CULL_BACK(state->cull_face & PIPE_FACE_BACK ? 1 : 0) |
553 S_028814_FACE(!state->front_ccw) |
554 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state, state->fill_front)) |
555 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state, state->fill_back)) |
556 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_point || state->offset_line) |
557 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL ||
558 state->fill_back != PIPE_POLYGON_MODE_FILL) |
559 S_028814_POLYMODE_FRONT_PTYPE(r600_translate_fill(state->fill_front)) |
560 S_028814_POLYMODE_BACK_PTYPE(r600_translate_fill(state->fill_back));
561 if (rctx->b.chip_class == R700) {
562 r600_store_context_reg(&rs->buffer, R_028814_PA_SU_SC_MODE_CNTL, rs->pa_su_sc_mode_cntl);
563 }
564 if (rctx->b.chip_class == R600) {
565 r600_store_context_reg(&rs->buffer, R_028350_SX_MISC,
566 S_028350_MULTIPASS(state->rasterizer_discard));
567 }
568 return rs;
569 }
570
r600_tex_filter(unsigned filter,unsigned max_aniso)571 static unsigned r600_tex_filter(unsigned filter, unsigned max_aniso)
572 {
573 if (filter == PIPE_TEX_FILTER_LINEAR)
574 return max_aniso > 1 ? V_03C000_SQ_TEX_XY_FILTER_ANISO_BILINEAR
575 : V_03C000_SQ_TEX_XY_FILTER_BILINEAR;
576 else
577 return max_aniso > 1 ? V_03C000_SQ_TEX_XY_FILTER_ANISO_POINT
578 : V_03C000_SQ_TEX_XY_FILTER_POINT;
579 }
580
r600_create_sampler_state(struct pipe_context * ctx,const struct pipe_sampler_state * state)581 static void *r600_create_sampler_state(struct pipe_context *ctx,
582 const struct pipe_sampler_state *state)
583 {
584 struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
585 struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
586 unsigned max_aniso = rscreen->force_aniso >= 0 ? rscreen->force_aniso
587 : state->max_anisotropy;
588 unsigned max_aniso_ratio = r600_tex_aniso_filter(max_aniso);
589
590 if (!ss) {
591 return NULL;
592 }
593
594 ss->seamless_cube_map = state->seamless_cube_map;
595 ss->border_color_use = sampler_state_needs_border_color(state);
596
597 /* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
598 ss->tex_sampler_words[0] =
599 S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
600 S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
601 S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
602 S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter, max_aniso)) |
603 S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter, max_aniso)) |
604 S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
605 S_03C000_MAX_ANISO_RATIO(max_aniso_ratio) |
606 S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
607 S_03C000_BORDER_COLOR_TYPE(ss->border_color_use ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
608 /* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
609 ss->tex_sampler_words[1] =
610 S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
611 S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)) |
612 S_03C004_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6));
613 /* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
614 ss->tex_sampler_words[2] = S_03C008_TYPE(1);
615
616 if (ss->border_color_use) {
617 memcpy(&ss->border_color, &state->border_color, sizeof(state->border_color));
618 }
619 return ss;
620 }
621
622 static struct pipe_sampler_view *
texture_buffer_sampler_view(struct r600_pipe_sampler_view * view,unsigned width0,unsigned height0)623 texture_buffer_sampler_view(struct r600_pipe_sampler_view *view,
624 unsigned width0, unsigned height0)
625
626 {
627 struct r600_texture *tmp = (struct r600_texture*)view->base.texture;
628 int stride = util_format_get_blocksize(view->base.format);
629 unsigned format, num_format, format_comp, endian;
630 uint64_t offset = view->base.u.buf.offset;
631 unsigned size = view->base.u.buf.size;
632
633 r600_vertex_data_type(view->base.format,
634 &format, &num_format, &format_comp,
635 &endian);
636
637 view->tex_resource = &tmp->resource;
638 view->skip_mip_address_reloc = true;
639
640 view->tex_resource_words[0] = offset;
641 view->tex_resource_words[1] = size - 1;
642 view->tex_resource_words[2] = S_038008_BASE_ADDRESS_HI(offset >> 32UL) |
643 S_038008_STRIDE(stride) |
644 S_038008_DATA_FORMAT(format) |
645 S_038008_NUM_FORMAT_ALL(num_format) |
646 S_038008_FORMAT_COMP_ALL(format_comp) |
647 S_038008_ENDIAN_SWAP(endian);
648 view->tex_resource_words[3] = 0;
649 /*
650 * in theory dword 4 is for number of elements, for use with resinfo,
651 * but it seems to utterly fail to work, the amd gpu shader analyser
652 * uses a const buffer to store the element sizes for buffer txq
653 */
654 view->tex_resource_words[4] = 0;
655 view->tex_resource_words[5] = 0;
656 view->tex_resource_words[6] = S_038018_TYPE(V_038010_SQ_TEX_VTX_VALID_BUFFER);
657 return &view->base;
658 }
659
660 struct pipe_sampler_view *
r600_create_sampler_view_custom(struct pipe_context * ctx,struct pipe_resource * texture,const struct pipe_sampler_view * state,unsigned width_first_level,unsigned height_first_level)661 r600_create_sampler_view_custom(struct pipe_context *ctx,
662 struct pipe_resource *texture,
663 const struct pipe_sampler_view *state,
664 unsigned width_first_level, unsigned height_first_level)
665 {
666 struct r600_pipe_sampler_view *view = CALLOC_STRUCT(r600_pipe_sampler_view);
667 struct r600_texture *tmp = (struct r600_texture*)texture;
668 unsigned format, endian;
669 uint32_t word4 = 0, yuv_format = 0, pitch = 0;
670 unsigned char swizzle[4], array_mode = 0;
671 unsigned width, height, depth, offset_level, last_level;
672 bool do_endian_swap = FALSE;
673
674 if (!view)
675 return NULL;
676
677 /* initialize base object */
678 view->base = *state;
679 view->base.texture = NULL;
680 pipe_reference(NULL, &texture->reference);
681 view->base.texture = texture;
682 view->base.reference.count = 1;
683 view->base.context = ctx;
684
685 if (texture->target == PIPE_BUFFER)
686 return texture_buffer_sampler_view(view, texture->width0, 1);
687
688 swizzle[0] = state->swizzle_r;
689 swizzle[1] = state->swizzle_g;
690 swizzle[2] = state->swizzle_b;
691 swizzle[3] = state->swizzle_a;
692
693 if (R600_BIG_ENDIAN)
694 do_endian_swap = !tmp->db_compatible;
695
696 format = r600_translate_texformat(ctx->screen, state->format,
697 swizzle,
698 &word4, &yuv_format, do_endian_swap);
699 assert(format != ~0);
700 if (format == ~0) {
701 FREE(view);
702 return NULL;
703 }
704
705 if (state->format == PIPE_FORMAT_X24S8_UINT ||
706 state->format == PIPE_FORMAT_S8X24_UINT ||
707 state->format == PIPE_FORMAT_X32_S8X24_UINT ||
708 state->format == PIPE_FORMAT_S8_UINT)
709 view->is_stencil_sampler = true;
710
711 if (tmp->is_depth && !r600_can_sample_zs(tmp, view->is_stencil_sampler)) {
712 if (!r600_init_flushed_depth_texture(ctx, texture, NULL)) {
713 FREE(view);
714 return NULL;
715 }
716 tmp = tmp->flushed_depth_texture;
717 }
718
719 endian = r600_colorformat_endian_swap(format, do_endian_swap);
720
721 offset_level = state->u.tex.first_level;
722 last_level = state->u.tex.last_level - offset_level;
723 width = width_first_level;
724 height = height_first_level;
725 depth = u_minify(texture->depth0, offset_level);
726 pitch = tmp->surface.level[offset_level].nblk_x * util_format_get_blockwidth(state->format);
727
728 if (texture->target == PIPE_TEXTURE_1D_ARRAY) {
729 height = 1;
730 depth = texture->array_size;
731 } else if (texture->target == PIPE_TEXTURE_2D_ARRAY) {
732 depth = texture->array_size;
733 } else if (texture->target == PIPE_TEXTURE_CUBE_ARRAY)
734 depth = texture->array_size / 6;
735
736 switch (tmp->surface.level[offset_level].mode) {
737 default:
738 case RADEON_SURF_MODE_LINEAR_ALIGNED:
739 array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
740 break;
741 case RADEON_SURF_MODE_1D:
742 array_mode = V_038000_ARRAY_1D_TILED_THIN1;
743 break;
744 case RADEON_SURF_MODE_2D:
745 array_mode = V_038000_ARRAY_2D_TILED_THIN1;
746 break;
747 }
748
749 view->tex_resource = &tmp->resource;
750 view->tex_resource_words[0] = (S_038000_DIM(r600_tex_dim(texture->target, texture->nr_samples)) |
751 S_038000_TILE_MODE(array_mode) |
752 S_038000_TILE_TYPE(tmp->non_disp_tiling) |
753 S_038000_PITCH((pitch / 8) - 1) |
754 S_038000_TEX_WIDTH(width - 1));
755 view->tex_resource_words[1] = (S_038004_TEX_HEIGHT(height - 1) |
756 S_038004_TEX_DEPTH(depth - 1) |
757 S_038004_DATA_FORMAT(format));
758 view->tex_resource_words[2] = tmp->surface.level[offset_level].offset >> 8;
759 if (offset_level >= tmp->resource.b.b.last_level) {
760 view->tex_resource_words[3] = tmp->surface.level[offset_level].offset >> 8;
761 } else {
762 view->tex_resource_words[3] = tmp->surface.level[offset_level + 1].offset >> 8;
763 }
764 view->tex_resource_words[4] = (word4 |
765 S_038010_REQUEST_SIZE(1) |
766 S_038010_ENDIAN_SWAP(endian) |
767 S_038010_BASE_LEVEL(0));
768 view->tex_resource_words[5] = (S_038014_BASE_ARRAY(state->u.tex.first_layer) |
769 S_038014_LAST_ARRAY(state->u.tex.last_layer));
770 if (texture->nr_samples > 1) {
771 /* LAST_LEVEL holds log2(nr_samples) for multisample textures */
772 view->tex_resource_words[5] |= S_038014_LAST_LEVEL(util_logbase2(texture->nr_samples));
773 } else {
774 view->tex_resource_words[5] |= S_038014_LAST_LEVEL(last_level);
775 }
776 view->tex_resource_words[6] = (S_038018_TYPE(V_038010_SQ_TEX_VTX_VALID_TEXTURE) |
777 S_038018_MAX_ANISO(4 /* max 16 samples */));
778 return &view->base;
779 }
780
781 static struct pipe_sampler_view *
r600_create_sampler_view(struct pipe_context * ctx,struct pipe_resource * tex,const struct pipe_sampler_view * state)782 r600_create_sampler_view(struct pipe_context *ctx,
783 struct pipe_resource *tex,
784 const struct pipe_sampler_view *state)
785 {
786 return r600_create_sampler_view_custom(ctx, tex, state,
787 u_minify(tex->width0, state->u.tex.first_level),
788 u_minify(tex->height0, state->u.tex.first_level));
789 }
790
r600_emit_clip_state(struct r600_context * rctx,struct r600_atom * atom)791 static void r600_emit_clip_state(struct r600_context *rctx, struct r600_atom *atom)
792 {
793 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
794 struct pipe_clip_state *state = &rctx->clip_state.state;
795
796 radeon_set_context_reg_seq(cs, R_028E20_PA_CL_UCP0_X, 6*4);
797 radeon_emit_array(cs, (unsigned*)state, 6*4);
798 }
799
r600_set_polygon_stipple(struct pipe_context * ctx,const struct pipe_poly_stipple * state)800 static void r600_set_polygon_stipple(struct pipe_context *ctx,
801 const struct pipe_poly_stipple *state)
802 {
803 }
804
r600_init_color_surface(struct r600_context * rctx,struct r600_surface * surf,bool force_cmask_fmask)805 static void r600_init_color_surface(struct r600_context *rctx,
806 struct r600_surface *surf,
807 bool force_cmask_fmask)
808 {
809 struct r600_screen *rscreen = rctx->screen;
810 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
811 unsigned level = surf->base.u.tex.level;
812 unsigned pitch, slice;
813 unsigned color_info;
814 unsigned color_view;
815 unsigned format, swap, ntype, endian;
816 unsigned offset;
817 const struct util_format_description *desc;
818 int i;
819 bool blend_bypass = 0, blend_clamp = 1, do_endian_swap = FALSE;
820
821 if (rtex->db_compatible && !r600_can_sample_zs(rtex, false)) {
822 r600_init_flushed_depth_texture(&rctx->b.b, surf->base.texture, NULL);
823 rtex = rtex->flushed_depth_texture;
824 assert(rtex);
825 }
826
827 offset = rtex->surface.level[level].offset;
828 color_view = S_028080_SLICE_START(surf->base.u.tex.first_layer) |
829 S_028080_SLICE_MAX(surf->base.u.tex.last_layer);
830
831 pitch = rtex->surface.level[level].nblk_x / 8 - 1;
832 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
833 if (slice) {
834 slice = slice - 1;
835 }
836 color_info = 0;
837 switch (rtex->surface.level[level].mode) {
838 default:
839 case RADEON_SURF_MODE_LINEAR_ALIGNED:
840 color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_LINEAR_ALIGNED);
841 break;
842 case RADEON_SURF_MODE_1D:
843 color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_1D_TILED_THIN1);
844 break;
845 case RADEON_SURF_MODE_2D:
846 color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_2D_TILED_THIN1);
847 break;
848 }
849
850 desc = util_format_description(surf->base.format);
851
852 for (i = 0; i < 4; i++) {
853 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
854 break;
855 }
856 }
857
858 ntype = V_0280A0_NUMBER_UNORM;
859 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
860 ntype = V_0280A0_NUMBER_SRGB;
861 else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
862 if (desc->channel[i].normalized)
863 ntype = V_0280A0_NUMBER_SNORM;
864 else if (desc->channel[i].pure_integer)
865 ntype = V_0280A0_NUMBER_SINT;
866 } else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
867 if (desc->channel[i].normalized)
868 ntype = V_0280A0_NUMBER_UNORM;
869 else if (desc->channel[i].pure_integer)
870 ntype = V_0280A0_NUMBER_UINT;
871 }
872
873 if (R600_BIG_ENDIAN)
874 do_endian_swap = !rtex->db_compatible;
875
876 format = r600_translate_colorformat(rctx->b.chip_class, surf->base.format,
877 do_endian_swap);
878 assert(format != ~0);
879
880 swap = r600_translate_colorswap(surf->base.format, do_endian_swap);
881 assert(swap != ~0);
882
883 endian = r600_colorformat_endian_swap(format, do_endian_swap);
884
885 /* set blend bypass according to docs if SINT/UINT or
886 8/24 COLOR variants */
887 if (ntype == V_0280A0_NUMBER_UINT || ntype == V_0280A0_NUMBER_SINT ||
888 format == V_0280A0_COLOR_8_24 || format == V_0280A0_COLOR_24_8 ||
889 format == V_0280A0_COLOR_X24_8_32_FLOAT) {
890 blend_clamp = 0;
891 blend_bypass = 1;
892 }
893
894 surf->alphatest_bypass = ntype == V_0280A0_NUMBER_UINT || ntype == V_0280A0_NUMBER_SINT;
895
896 color_info |= S_0280A0_FORMAT(format) |
897 S_0280A0_COMP_SWAP(swap) |
898 S_0280A0_BLEND_BYPASS(blend_bypass) |
899 S_0280A0_BLEND_CLAMP(blend_clamp) |
900 S_0280A0_NUMBER_TYPE(ntype) |
901 S_0280A0_ENDIAN(endian);
902
903 /* EXPORT_NORM is an optimzation that can be enabled for better
904 * performance in certain cases
905 */
906 if (rctx->b.chip_class == R600) {
907 /* EXPORT_NORM can be enabled if:
908 * - 11-bit or smaller UNORM/SNORM/SRGB
909 * - BLEND_CLAMP is enabled
910 * - BLEND_FLOAT32 is disabled
911 */
912 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
913 (desc->channel[i].size < 12 &&
914 desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
915 ntype != V_0280A0_NUMBER_UINT &&
916 ntype != V_0280A0_NUMBER_SINT) &&
917 G_0280A0_BLEND_CLAMP(color_info) &&
918 !G_0280A0_BLEND_FLOAT32(color_info)) {
919 color_info |= S_0280A0_SOURCE_FORMAT(V_0280A0_EXPORT_NORM);
920 surf->export_16bpc = true;
921 }
922 } else {
923 /* EXPORT_NORM can be enabled if:
924 * - 11-bit or smaller UNORM/SNORM/SRGB
925 * - 16-bit or smaller FLOAT
926 */
927 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
928 ((desc->channel[i].size < 12 &&
929 desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
930 ntype != V_0280A0_NUMBER_UINT && ntype != V_0280A0_NUMBER_SINT) ||
931 (desc->channel[i].size < 17 &&
932 desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT))) {
933 color_info |= S_0280A0_SOURCE_FORMAT(V_0280A0_EXPORT_NORM);
934 surf->export_16bpc = true;
935 }
936 }
937
938 /* These might not always be initialized to zero. */
939 surf->cb_color_base = offset >> 8;
940 surf->cb_color_size = S_028060_PITCH_TILE_MAX(pitch) |
941 S_028060_SLICE_TILE_MAX(slice);
942 surf->cb_color_fmask = surf->cb_color_base;
943 surf->cb_color_cmask = surf->cb_color_base;
944 surf->cb_color_mask = 0;
945
946 r600_resource_reference(&surf->cb_buffer_cmask, &rtex->resource);
947 r600_resource_reference(&surf->cb_buffer_fmask, &rtex->resource);
948
949 if (rtex->cmask.size) {
950 surf->cb_color_cmask = rtex->cmask.offset >> 8;
951 surf->cb_color_mask |= S_028100_CMASK_BLOCK_MAX(rtex->cmask.slice_tile_max);
952
953 if (rtex->fmask.size) {
954 color_info |= S_0280A0_TILE_MODE(V_0280A0_FRAG_ENABLE);
955 surf->cb_color_fmask = rtex->fmask.offset >> 8;
956 surf->cb_color_mask |= S_028100_FMASK_TILE_MAX(rtex->fmask.slice_tile_max);
957 } else { /* cmask only */
958 color_info |= S_0280A0_TILE_MODE(V_0280A0_CLEAR_ENABLE);
959 }
960 } else if (force_cmask_fmask) {
961 /* Allocate dummy FMASK and CMASK if they aren't allocated already.
962 *
963 * R6xx needs FMASK and CMASK for the destination buffer of color resolve,
964 * otherwise it hangs. We don't have FMASK and CMASK pre-allocated,
965 * because it's not an MSAA buffer.
966 */
967 struct r600_cmask_info cmask;
968 struct r600_fmask_info fmask;
969
970 r600_texture_get_cmask_info(&rscreen->b, rtex, &cmask);
971 r600_texture_get_fmask_info(&rscreen->b, rtex, 8, &fmask);
972
973 /* CMASK. */
974 if (!rctx->dummy_cmask ||
975 rctx->dummy_cmask->b.b.width0 < cmask.size ||
976 rctx->dummy_cmask->buf->alignment % cmask.alignment != 0) {
977 struct pipe_transfer *transfer;
978 void *ptr;
979
980 r600_resource_reference(&rctx->dummy_cmask, NULL);
981 rctx->dummy_cmask = (struct r600_resource*)
982 r600_aligned_buffer_create(&rscreen->b.b, 0,
983 PIPE_USAGE_DEFAULT,
984 cmask.size, cmask.alignment);
985
986 /* Set the contents to 0xCC. */
987 ptr = pipe_buffer_map(&rctx->b.b, &rctx->dummy_cmask->b.b, PIPE_TRANSFER_WRITE, &transfer);
988 memset(ptr, 0xCC, cmask.size);
989 pipe_buffer_unmap(&rctx->b.b, transfer);
990 }
991 r600_resource_reference(&surf->cb_buffer_cmask, rctx->dummy_cmask);
992
993 /* FMASK. */
994 if (!rctx->dummy_fmask ||
995 rctx->dummy_fmask->b.b.width0 < fmask.size ||
996 rctx->dummy_fmask->buf->alignment % fmask.alignment != 0) {
997 r600_resource_reference(&rctx->dummy_fmask, NULL);
998 rctx->dummy_fmask = (struct r600_resource*)
999 r600_aligned_buffer_create(&rscreen->b.b, 0,
1000 PIPE_USAGE_DEFAULT,
1001 fmask.size, fmask.alignment);
1002 }
1003 r600_resource_reference(&surf->cb_buffer_fmask, rctx->dummy_fmask);
1004
1005 /* Init the registers. */
1006 color_info |= S_0280A0_TILE_MODE(V_0280A0_FRAG_ENABLE);
1007 surf->cb_color_cmask = 0;
1008 surf->cb_color_fmask = 0;
1009 surf->cb_color_mask = S_028100_CMASK_BLOCK_MAX(cmask.slice_tile_max) |
1010 S_028100_FMASK_TILE_MAX(fmask.slice_tile_max);
1011 }
1012
1013 surf->cb_color_info = color_info;
1014 surf->cb_color_view = color_view;
1015 surf->color_initialized = true;
1016 }
1017
r600_init_depth_surface(struct r600_context * rctx,struct r600_surface * surf)1018 static void r600_init_depth_surface(struct r600_context *rctx,
1019 struct r600_surface *surf)
1020 {
1021 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1022 unsigned level, pitch, slice, format, offset, array_mode;
1023
1024 level = surf->base.u.tex.level;
1025 offset = rtex->surface.level[level].offset;
1026 pitch = rtex->surface.level[level].nblk_x / 8 - 1;
1027 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
1028 if (slice) {
1029 slice = slice - 1;
1030 }
1031 switch (rtex->surface.level[level].mode) {
1032 case RADEON_SURF_MODE_2D:
1033 array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
1034 break;
1035 case RADEON_SURF_MODE_1D:
1036 case RADEON_SURF_MODE_LINEAR_ALIGNED:
1037 default:
1038 array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
1039 break;
1040 }
1041
1042 format = r600_translate_dbformat(surf->base.format);
1043 assert(format != ~0);
1044
1045 surf->db_depth_info = S_028010_ARRAY_MODE(array_mode) | S_028010_FORMAT(format);
1046 surf->db_depth_base = offset >> 8;
1047 surf->db_depth_view = S_028004_SLICE_START(surf->base.u.tex.first_layer) |
1048 S_028004_SLICE_MAX(surf->base.u.tex.last_layer);
1049 surf->db_depth_size = S_028000_PITCH_TILE_MAX(pitch) | S_028000_SLICE_TILE_MAX(slice);
1050 surf->db_prefetch_limit = (rtex->surface.level[level].nblk_y / 8) - 1;
1051
1052 /* use htile only for first level */
1053 if (rtex->htile_buffer && !level) {
1054 surf->db_htile_data_base = 0;
1055 surf->db_htile_surface = S_028D24_HTILE_WIDTH(1) |
1056 S_028D24_HTILE_HEIGHT(1) |
1057 S_028D24_FULL_CACHE(1);
1058 /* preload is not working properly on r6xx/r7xx */
1059 surf->db_depth_info |= S_028010_TILE_SURFACE_ENABLE(1);
1060 }
1061
1062 surf->depth_initialized = true;
1063 }
1064
r600_set_framebuffer_state(struct pipe_context * ctx,const struct pipe_framebuffer_state * state)1065 static void r600_set_framebuffer_state(struct pipe_context *ctx,
1066 const struct pipe_framebuffer_state *state)
1067 {
1068 struct r600_context *rctx = (struct r600_context *)ctx;
1069 struct r600_surface *surf;
1070 struct r600_texture *rtex;
1071 unsigned i;
1072
1073 /* Flush TC when changing the framebuffer state, because the only
1074 * client not using TC that can change textures is the framebuffer.
1075 * Other places don't typically have to flush TC.
1076 */
1077 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE |
1078 R600_CONTEXT_FLUSH_AND_INV |
1079 R600_CONTEXT_FLUSH_AND_INV_CB |
1080 R600_CONTEXT_FLUSH_AND_INV_CB_META |
1081 R600_CONTEXT_FLUSH_AND_INV_DB |
1082 R600_CONTEXT_FLUSH_AND_INV_DB_META |
1083 R600_CONTEXT_INV_TEX_CACHE;
1084
1085 /* Set the new state. */
1086 util_copy_framebuffer_state(&rctx->framebuffer.state, state);
1087
1088 rctx->framebuffer.export_16bpc = state->nr_cbufs != 0;
1089 rctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
1090 util_format_is_pure_integer(state->cbufs[0]->format);
1091 rctx->framebuffer.compressed_cb_mask = 0;
1092 rctx->framebuffer.is_msaa_resolve = state->nr_cbufs == 2 &&
1093 state->cbufs[0] && state->cbufs[1] &&
1094 state->cbufs[0]->texture->nr_samples > 1 &&
1095 state->cbufs[1]->texture->nr_samples <= 1;
1096 rctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
1097
1098 /* Colorbuffers. */
1099 for (i = 0; i < state->nr_cbufs; i++) {
1100 /* The resolve buffer must have CMASK and FMASK to prevent hardlocks on R6xx. */
1101 bool force_cmask_fmask = rctx->b.chip_class == R600 &&
1102 rctx->framebuffer.is_msaa_resolve &&
1103 i == 1;
1104
1105 surf = (struct r600_surface*)state->cbufs[i];
1106 if (!surf)
1107 continue;
1108
1109 rtex = (struct r600_texture*)surf->base.texture;
1110 r600_context_add_resource_size(ctx, state->cbufs[i]->texture);
1111
1112 if (!surf->color_initialized || force_cmask_fmask) {
1113 r600_init_color_surface(rctx, surf, force_cmask_fmask);
1114 if (force_cmask_fmask) {
1115 /* re-initialize later without compression */
1116 surf->color_initialized = false;
1117 }
1118 }
1119
1120 if (!surf->export_16bpc) {
1121 rctx->framebuffer.export_16bpc = false;
1122 }
1123
1124 if (rtex->fmask.size) {
1125 rctx->framebuffer.compressed_cb_mask |= 1 << i;
1126 }
1127 }
1128
1129 /* Update alpha-test state dependencies.
1130 * Alpha-test is done on the first colorbuffer only. */
1131 if (state->nr_cbufs) {
1132 bool alphatest_bypass = false;
1133
1134 surf = (struct r600_surface*)state->cbufs[0];
1135 if (surf) {
1136 alphatest_bypass = surf->alphatest_bypass;
1137 }
1138
1139 if (rctx->alphatest_state.bypass != alphatest_bypass) {
1140 rctx->alphatest_state.bypass = alphatest_bypass;
1141 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1142 }
1143 }
1144
1145 /* ZS buffer. */
1146 if (state->zsbuf) {
1147 surf = (struct r600_surface*)state->zsbuf;
1148
1149 r600_context_add_resource_size(ctx, state->zsbuf->texture);
1150
1151 if (!surf->depth_initialized) {
1152 r600_init_depth_surface(rctx, surf);
1153 }
1154
1155 if (state->zsbuf->format != rctx->poly_offset_state.zs_format) {
1156 rctx->poly_offset_state.zs_format = state->zsbuf->format;
1157 r600_mark_atom_dirty(rctx, &rctx->poly_offset_state.atom);
1158 }
1159
1160 if (rctx->db_state.rsurf != surf) {
1161 rctx->db_state.rsurf = surf;
1162 r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1163 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1164 }
1165 } else if (rctx->db_state.rsurf) {
1166 rctx->db_state.rsurf = NULL;
1167 r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1168 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1169 }
1170
1171 if (rctx->cb_misc_state.nr_cbufs != state->nr_cbufs) {
1172 rctx->cb_misc_state.nr_cbufs = state->nr_cbufs;
1173 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom);
1174 }
1175
1176 if (state->nr_cbufs == 0 && rctx->alphatest_state.bypass) {
1177 rctx->alphatest_state.bypass = false;
1178 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1179 }
1180
1181 /* Calculate the CS size. */
1182 rctx->framebuffer.atom.num_dw =
1183 10 /*COLOR_INFO*/ + 4 /*SCISSOR*/ + 3 /*SHADER_CONTROL*/ + 8 /*MSAA*/;
1184
1185 if (rctx->framebuffer.state.nr_cbufs) {
1186 rctx->framebuffer.atom.num_dw += 15 * rctx->framebuffer.state.nr_cbufs;
1187 rctx->framebuffer.atom.num_dw += 3 * (2 + rctx->framebuffer.state.nr_cbufs);
1188 }
1189 if (rctx->framebuffer.state.zsbuf) {
1190 rctx->framebuffer.atom.num_dw += 16;
1191 } else if (rctx->screen->b.info.drm_minor >= 18) {
1192 rctx->framebuffer.atom.num_dw += 3;
1193 }
1194 if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770) {
1195 rctx->framebuffer.atom.num_dw += 2;
1196 }
1197
1198 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom);
1199
1200 r600_set_sample_locations_constant_buffer(rctx);
1201 }
1202
1203 static uint32_t sample_locs_2x[] = {
1204 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1205 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1206 };
1207 static unsigned max_dist_2x = 4;
1208
1209 static uint32_t sample_locs_4x[] = {
1210 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1211 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1212 };
1213 static unsigned max_dist_4x = 6;
1214 static uint32_t sample_locs_8x[] = {
1215 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3),
1216 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7),
1217 };
1218 static unsigned max_dist_8x = 7;
1219
r600_get_sample_position(struct pipe_context * ctx,unsigned sample_count,unsigned sample_index,float * out_value)1220 static void r600_get_sample_position(struct pipe_context *ctx,
1221 unsigned sample_count,
1222 unsigned sample_index,
1223 float *out_value)
1224 {
1225 int offset, index;
1226 struct {
1227 int idx:4;
1228 } val;
1229 switch (sample_count) {
1230 case 1:
1231 default:
1232 out_value[0] = out_value[1] = 0.5;
1233 break;
1234 case 2:
1235 offset = 4 * (sample_index * 2);
1236 val.idx = (sample_locs_2x[0] >> offset) & 0xf;
1237 out_value[0] = (float)(val.idx + 8) / 16.0f;
1238 val.idx = (sample_locs_2x[0] >> (offset + 4)) & 0xf;
1239 out_value[1] = (float)(val.idx + 8) / 16.0f;
1240 break;
1241 case 4:
1242 offset = 4 * (sample_index * 2);
1243 val.idx = (sample_locs_4x[0] >> offset) & 0xf;
1244 out_value[0] = (float)(val.idx + 8) / 16.0f;
1245 val.idx = (sample_locs_4x[0] >> (offset + 4)) & 0xf;
1246 out_value[1] = (float)(val.idx + 8) / 16.0f;
1247 break;
1248 case 8:
1249 offset = 4 * (sample_index % 4 * 2);
1250 index = (sample_index / 4);
1251 val.idx = (sample_locs_8x[index] >> offset) & 0xf;
1252 out_value[0] = (float)(val.idx + 8) / 16.0f;
1253 val.idx = (sample_locs_8x[index] >> (offset + 4)) & 0xf;
1254 out_value[1] = (float)(val.idx + 8) / 16.0f;
1255 break;
1256 }
1257 }
1258
r600_emit_msaa_state(struct r600_context * rctx,int nr_samples)1259 static void r600_emit_msaa_state(struct r600_context *rctx, int nr_samples)
1260 {
1261 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1262 unsigned max_dist = 0;
1263
1264 if (rctx->b.family == CHIP_R600) {
1265 switch (nr_samples) {
1266 default:
1267 nr_samples = 0;
1268 break;
1269 case 2:
1270 radeon_set_config_reg(cs, R_008B40_PA_SC_AA_SAMPLE_LOCS_2S, sample_locs_2x[0]);
1271 max_dist = max_dist_2x;
1272 break;
1273 case 4:
1274 radeon_set_config_reg(cs, R_008B44_PA_SC_AA_SAMPLE_LOCS_4S, sample_locs_4x[0]);
1275 max_dist = max_dist_4x;
1276 break;
1277 case 8:
1278 radeon_set_config_reg_seq(cs, R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0, 2);
1279 radeon_emit(cs, sample_locs_8x[0]); /* R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0 */
1280 radeon_emit(cs, sample_locs_8x[1]); /* R_008B4C_PA_SC_AA_SAMPLE_LOCS_8S_WD1 */
1281 max_dist = max_dist_8x;
1282 break;
1283 }
1284 } else {
1285 switch (nr_samples) {
1286 default:
1287 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1288 radeon_emit(cs, 0); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1289 radeon_emit(cs, 0); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1290 nr_samples = 0;
1291 break;
1292 case 2:
1293 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1294 radeon_emit(cs, sample_locs_2x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1295 radeon_emit(cs, sample_locs_2x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1296 max_dist = max_dist_2x;
1297 break;
1298 case 4:
1299 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1300 radeon_emit(cs, sample_locs_4x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1301 radeon_emit(cs, sample_locs_4x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1302 max_dist = max_dist_4x;
1303 break;
1304 case 8:
1305 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1306 radeon_emit(cs, sample_locs_8x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1307 radeon_emit(cs, sample_locs_8x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1308 max_dist = max_dist_8x;
1309 break;
1310 }
1311 }
1312
1313 if (nr_samples > 1) {
1314 radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1315 radeon_emit(cs, S_028C00_LAST_PIXEL(1) |
1316 S_028C00_EXPAND_LINE_WIDTH(1)); /* R_028C00_PA_SC_LINE_CNTL */
1317 radeon_emit(cs, S_028C04_MSAA_NUM_SAMPLES(util_logbase2(nr_samples)) |
1318 S_028C04_MAX_SAMPLE_DIST(max_dist)); /* R_028C04_PA_SC_AA_CONFIG */
1319 } else {
1320 radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1321 radeon_emit(cs, S_028C00_LAST_PIXEL(1)); /* R_028C00_PA_SC_LINE_CNTL */
1322 radeon_emit(cs, 0); /* R_028C04_PA_SC_AA_CONFIG */
1323 }
1324 }
1325
r600_emit_framebuffer_state(struct r600_context * rctx,struct r600_atom * atom)1326 static void r600_emit_framebuffer_state(struct r600_context *rctx, struct r600_atom *atom)
1327 {
1328 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1329 struct pipe_framebuffer_state *state = &rctx->framebuffer.state;
1330 unsigned nr_cbufs = state->nr_cbufs;
1331 struct r600_surface **cb = (struct r600_surface**)&state->cbufs[0];
1332 unsigned i, sbu = 0;
1333
1334 /* Colorbuffers. */
1335 radeon_set_context_reg_seq(cs, R_0280A0_CB_COLOR0_INFO, 8);
1336 for (i = 0; i < nr_cbufs; i++) {
1337 radeon_emit(cs, cb[i] ? cb[i]->cb_color_info : 0);
1338 }
1339 /* set CB_COLOR1_INFO for possible dual-src blending */
1340 if (i == 1 && cb[0]) {
1341 radeon_emit(cs, cb[0]->cb_color_info);
1342 i++;
1343 }
1344 for (; i < 8; i++) {
1345 radeon_emit(cs, 0);
1346 }
1347
1348 if (nr_cbufs) {
1349 for (i = 0; i < nr_cbufs; i++) {
1350 unsigned reloc;
1351
1352 if (!cb[i])
1353 continue;
1354
1355 /* COLOR_BASE */
1356 radeon_set_context_reg(cs, R_028040_CB_COLOR0_BASE + i*4, cb[i]->cb_color_base);
1357
1358 reloc = radeon_add_to_buffer_list(&rctx->b,
1359 &rctx->b.gfx,
1360 (struct r600_resource*)cb[i]->base.texture,
1361 RADEON_USAGE_READWRITE,
1362 cb[i]->base.texture->nr_samples > 1 ?
1363 RADEON_PRIO_COLOR_BUFFER_MSAA :
1364 RADEON_PRIO_COLOR_BUFFER);
1365 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1366 radeon_emit(cs, reloc);
1367
1368 /* FMASK */
1369 radeon_set_context_reg(cs, R_0280E0_CB_COLOR0_FRAG + i*4, cb[i]->cb_color_fmask);
1370
1371 reloc = radeon_add_to_buffer_list(&rctx->b,
1372 &rctx->b.gfx,
1373 cb[i]->cb_buffer_fmask,
1374 RADEON_USAGE_READWRITE,
1375 cb[i]->base.texture->nr_samples > 1 ?
1376 RADEON_PRIO_COLOR_BUFFER_MSAA :
1377 RADEON_PRIO_COLOR_BUFFER);
1378 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1379 radeon_emit(cs, reloc);
1380
1381 /* CMASK */
1382 radeon_set_context_reg(cs, R_0280C0_CB_COLOR0_TILE + i*4, cb[i]->cb_color_cmask);
1383
1384 reloc = radeon_add_to_buffer_list(&rctx->b,
1385 &rctx->b.gfx,
1386 cb[i]->cb_buffer_cmask,
1387 RADEON_USAGE_READWRITE,
1388 cb[i]->base.texture->nr_samples > 1 ?
1389 RADEON_PRIO_COLOR_BUFFER_MSAA :
1390 RADEON_PRIO_COLOR_BUFFER);
1391 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1392 radeon_emit(cs, reloc);
1393 }
1394
1395 radeon_set_context_reg_seq(cs, R_028060_CB_COLOR0_SIZE, nr_cbufs);
1396 for (i = 0; i < nr_cbufs; i++) {
1397 radeon_emit(cs, cb[i] ? cb[i]->cb_color_size : 0);
1398 }
1399
1400 radeon_set_context_reg_seq(cs, R_028080_CB_COLOR0_VIEW, nr_cbufs);
1401 for (i = 0; i < nr_cbufs; i++) {
1402 radeon_emit(cs, cb[i] ? cb[i]->cb_color_view : 0);
1403 }
1404
1405 radeon_set_context_reg_seq(cs, R_028100_CB_COLOR0_MASK, nr_cbufs);
1406 for (i = 0; i < nr_cbufs; i++) {
1407 radeon_emit(cs, cb[i] ? cb[i]->cb_color_mask : 0);
1408 }
1409
1410 sbu |= SURFACE_BASE_UPDATE_COLOR_NUM(nr_cbufs);
1411 }
1412
1413 /* SURFACE_BASE_UPDATE */
1414 if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770 && sbu) {
1415 radeon_emit(cs, PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0));
1416 radeon_emit(cs, sbu);
1417 sbu = 0;
1418 }
1419
1420 /* Zbuffer. */
1421 if (state->zsbuf) {
1422 struct r600_surface *surf = (struct r600_surface*)state->zsbuf;
1423 unsigned reloc = radeon_add_to_buffer_list(&rctx->b,
1424 &rctx->b.gfx,
1425 (struct r600_resource*)state->zsbuf->texture,
1426 RADEON_USAGE_READWRITE,
1427 surf->base.texture->nr_samples > 1 ?
1428 RADEON_PRIO_DEPTH_BUFFER_MSAA :
1429 RADEON_PRIO_DEPTH_BUFFER);
1430
1431 radeon_set_context_reg_seq(cs, R_028000_DB_DEPTH_SIZE, 2);
1432 radeon_emit(cs, surf->db_depth_size); /* R_028000_DB_DEPTH_SIZE */
1433 radeon_emit(cs, surf->db_depth_view); /* R_028004_DB_DEPTH_VIEW */
1434 radeon_set_context_reg_seq(cs, R_02800C_DB_DEPTH_BASE, 2);
1435 radeon_emit(cs, surf->db_depth_base); /* R_02800C_DB_DEPTH_BASE */
1436 radeon_emit(cs, surf->db_depth_info); /* R_028010_DB_DEPTH_INFO */
1437
1438 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1439 radeon_emit(cs, reloc);
1440
1441 radeon_set_context_reg(cs, R_028D34_DB_PREFETCH_LIMIT, surf->db_prefetch_limit);
1442
1443 sbu |= SURFACE_BASE_UPDATE_DEPTH;
1444 } else if (rctx->screen->b.info.drm_minor >= 18) {
1445 /* DRM 2.6.18 allows the INVALID format to disable depth/stencil.
1446 * Older kernels are out of luck. */
1447 radeon_set_context_reg(cs, R_028010_DB_DEPTH_INFO, S_028010_FORMAT(V_028010_DEPTH_INVALID));
1448 }
1449
1450 /* SURFACE_BASE_UPDATE */
1451 if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770 && sbu) {
1452 radeon_emit(cs, PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0));
1453 radeon_emit(cs, sbu);
1454 sbu = 0;
1455 }
1456
1457 /* Framebuffer dimensions. */
1458 radeon_set_context_reg_seq(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL, 2);
1459 radeon_emit(cs, S_028240_TL_X(0) | S_028240_TL_Y(0) |
1460 S_028240_WINDOW_OFFSET_DISABLE(1)); /* R_028204_PA_SC_WINDOW_SCISSOR_TL */
1461 radeon_emit(cs, S_028244_BR_X(state->width) |
1462 S_028244_BR_Y(state->height)); /* R_028208_PA_SC_WINDOW_SCISSOR_BR */
1463
1464 if (rctx->framebuffer.is_msaa_resolve) {
1465 radeon_set_context_reg(cs, R_0287A0_CB_SHADER_CONTROL, 1);
1466 } else {
1467 /* Always enable the first colorbuffer in CB_SHADER_CONTROL. This
1468 * will assure that the alpha-test will work even if there is
1469 * no colorbuffer bound. */
1470 radeon_set_context_reg(cs, R_0287A0_CB_SHADER_CONTROL,
1471 (1ull << MAX2(nr_cbufs, 1)) - 1);
1472 }
1473
1474 r600_emit_msaa_state(rctx, rctx->framebuffer.nr_samples);
1475 }
1476
r600_set_min_samples(struct pipe_context * ctx,unsigned min_samples)1477 static void r600_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
1478 {
1479 struct r600_context *rctx = (struct r600_context *)ctx;
1480
1481 if (rctx->ps_iter_samples == min_samples)
1482 return;
1483
1484 rctx->ps_iter_samples = min_samples;
1485 if (rctx->framebuffer.nr_samples > 1) {
1486 r600_mark_atom_dirty(rctx, &rctx->rasterizer_state.atom);
1487 if (rctx->b.chip_class == R600)
1488 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1489 }
1490 }
1491
r600_emit_cb_misc_state(struct r600_context * rctx,struct r600_atom * atom)1492 static void r600_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1493 {
1494 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1495 struct r600_cb_misc_state *a = (struct r600_cb_misc_state*)atom;
1496
1497 if (G_028808_SPECIAL_OP(a->cb_color_control) == V_028808_SPECIAL_RESOLVE_BOX) {
1498 radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1499 if (rctx->b.chip_class == R600) {
1500 radeon_emit(cs, 0xff); /* R_028238_CB_TARGET_MASK */
1501 radeon_emit(cs, 0xff); /* R_02823C_CB_SHADER_MASK */
1502 } else {
1503 radeon_emit(cs, 0xf); /* R_028238_CB_TARGET_MASK */
1504 radeon_emit(cs, 0xf); /* R_02823C_CB_SHADER_MASK */
1505 }
1506 radeon_set_context_reg(cs, R_028808_CB_COLOR_CONTROL, a->cb_color_control);
1507 } else {
1508 unsigned fb_colormask = (1ULL << ((unsigned)a->nr_cbufs * 4)) - 1;
1509 unsigned ps_colormask = (1ULL << ((unsigned)a->nr_ps_color_outputs * 4)) - 1;
1510 unsigned multiwrite = a->multiwrite && a->nr_cbufs > 1;
1511
1512 radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1513 radeon_emit(cs, a->blend_colormask & fb_colormask); /* R_028238_CB_TARGET_MASK */
1514 /* Always enable the first color output to make sure alpha-test works even without one. */
1515 radeon_emit(cs, 0xf | (multiwrite ? fb_colormask : ps_colormask)); /* R_02823C_CB_SHADER_MASK */
1516 radeon_set_context_reg(cs, R_028808_CB_COLOR_CONTROL,
1517 a->cb_color_control |
1518 S_028808_MULTIWRITE_ENABLE(multiwrite));
1519 }
1520 }
1521
r600_emit_db_state(struct r600_context * rctx,struct r600_atom * atom)1522 static void r600_emit_db_state(struct r600_context *rctx, struct r600_atom *atom)
1523 {
1524 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1525 struct r600_db_state *a = (struct r600_db_state*)atom;
1526
1527 if (a->rsurf && a->rsurf->db_htile_surface) {
1528 struct r600_texture *rtex = (struct r600_texture *)a->rsurf->base.texture;
1529 unsigned reloc_idx;
1530
1531 radeon_set_context_reg(cs, R_02802C_DB_DEPTH_CLEAR, fui(rtex->depth_clear_value));
1532 radeon_set_context_reg(cs, R_028D24_DB_HTILE_SURFACE, a->rsurf->db_htile_surface);
1533 radeon_set_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, a->rsurf->db_htile_data_base);
1534 reloc_idx = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rtex->htile_buffer,
1535 RADEON_USAGE_READWRITE, RADEON_PRIO_HTILE);
1536 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1537 radeon_emit(cs, reloc_idx);
1538 } else {
1539 radeon_set_context_reg(cs, R_028D24_DB_HTILE_SURFACE, 0);
1540 }
1541 }
1542
r600_emit_db_misc_state(struct r600_context * rctx,struct r600_atom * atom)1543 static void r600_emit_db_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1544 {
1545 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1546 struct r600_db_misc_state *a = (struct r600_db_misc_state*)atom;
1547 unsigned db_render_control = 0;
1548 unsigned db_render_override =
1549 S_028D10_FORCE_HIS_ENABLE0(V_028D10_FORCE_DISABLE) |
1550 S_028D10_FORCE_HIS_ENABLE1(V_028D10_FORCE_DISABLE);
1551
1552 if (rctx->b.chip_class >= R700) {
1553 switch (a->ps_conservative_z) {
1554 default: /* fall through */
1555 case TGSI_FS_DEPTH_LAYOUT_ANY:
1556 db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_ANY_Z);
1557 break;
1558 case TGSI_FS_DEPTH_LAYOUT_GREATER:
1559 db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_GREATER_THAN_Z);
1560 break;
1561 case TGSI_FS_DEPTH_LAYOUT_LESS:
1562 db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_LESS_THAN_Z);
1563 break;
1564 }
1565 }
1566
1567 if (rctx->b.num_occlusion_queries > 0 &&
1568 !a->occlusion_queries_disabled) {
1569 if (rctx->b.chip_class >= R700) {
1570 db_render_control |= S_028D0C_R700_PERFECT_ZPASS_COUNTS(1);
1571 }
1572 db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1573 } else {
1574 db_render_control |= S_028D0C_ZPASS_INCREMENT_DISABLE(1);
1575 }
1576
1577 if (rctx->db_state.rsurf && rctx->db_state.rsurf->db_htile_surface) {
1578 /* FORCE_OFF means HiZ/HiS are determined by DB_SHADER_CONTROL */
1579 db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_OFF);
1580 /* This is to fix a lockup when hyperz and alpha test are enabled at
1581 * the same time somehow GPU get confuse on which order to pick for
1582 * z test
1583 */
1584 if (rctx->alphatest_state.sx_alpha_test_control) {
1585 db_render_override |= S_028D10_FORCE_SHADER_Z_ORDER(1);
1586 }
1587 } else {
1588 db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1589 }
1590 if (rctx->b.chip_class == R600 && rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0) {
1591 /* sample shading and hyperz causes lockups on R6xx chips */
1592 db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1593 }
1594 if (a->flush_depthstencil_through_cb) {
1595 assert(a->copy_depth || a->copy_stencil);
1596
1597 db_render_control |= S_028D0C_DEPTH_COPY_ENABLE(a->copy_depth) |
1598 S_028D0C_STENCIL_COPY_ENABLE(a->copy_stencil) |
1599 S_028D0C_COPY_CENTROID(1) |
1600 S_028D0C_COPY_SAMPLE(a->copy_sample);
1601
1602 if (rctx->b.chip_class == R600)
1603 db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1604
1605 if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
1606 rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
1607 db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1608 } else if (a->flush_depth_inplace || a->flush_stencil_inplace) {
1609 db_render_control |= S_028D0C_DEPTH_COMPRESS_DISABLE(a->flush_depth_inplace) |
1610 S_028D0C_STENCIL_COMPRESS_DISABLE(a->flush_stencil_inplace);
1611 db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1612 }
1613 if (a->htile_clear) {
1614 db_render_control |= S_028D0C_DEPTH_CLEAR_ENABLE(1);
1615 }
1616
1617 /* RV770 workaround for a hang with 8x MSAA. */
1618 if (rctx->b.family == CHIP_RV770 && a->log_samples == 3) {
1619 db_render_override |= S_028D10_MAX_TILES_IN_DTT(6);
1620 }
1621
1622 radeon_set_context_reg_seq(cs, R_028D0C_DB_RENDER_CONTROL, 2);
1623 radeon_emit(cs, db_render_control); /* R_028D0C_DB_RENDER_CONTROL */
1624 radeon_emit(cs, db_render_override); /* R_028D10_DB_RENDER_OVERRIDE */
1625 radeon_set_context_reg(cs, R_02880C_DB_SHADER_CONTROL, a->db_shader_control);
1626 }
1627
r600_emit_config_state(struct r600_context * rctx,struct r600_atom * atom)1628 static void r600_emit_config_state(struct r600_context *rctx, struct r600_atom *atom)
1629 {
1630 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1631 struct r600_config_state *a = (struct r600_config_state*)atom;
1632
1633 radeon_set_config_reg(cs, R_008C04_SQ_GPR_RESOURCE_MGMT_1, a->sq_gpr_resource_mgmt_1);
1634 radeon_set_config_reg(cs, R_008C08_SQ_GPR_RESOURCE_MGMT_2, a->sq_gpr_resource_mgmt_2);
1635 }
1636
r600_emit_vertex_buffers(struct r600_context * rctx,struct r600_atom * atom)1637 static void r600_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom *atom)
1638 {
1639 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1640 uint32_t dirty_mask = rctx->vertex_buffer_state.dirty_mask;
1641
1642 while (dirty_mask) {
1643 struct pipe_vertex_buffer *vb;
1644 struct r600_resource *rbuffer;
1645 unsigned offset;
1646 unsigned buffer_index = u_bit_scan(&dirty_mask);
1647
1648 vb = &rctx->vertex_buffer_state.vb[buffer_index];
1649 rbuffer = (struct r600_resource*)vb->buffer;
1650 assert(rbuffer);
1651
1652 offset = vb->buffer_offset;
1653
1654 /* fetch resources start at index 320 (OFFSET_FS) */
1655 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1656 radeon_emit(cs, (R600_FETCH_CONSTANTS_OFFSET_FS + buffer_index) * 7);
1657 radeon_emit(cs, offset); /* RESOURCEi_WORD0 */
1658 radeon_emit(cs, rbuffer->b.b.width0 - offset - 1); /* RESOURCEi_WORD1 */
1659 radeon_emit(cs, /* RESOURCEi_WORD2 */
1660 S_038008_ENDIAN_SWAP(r600_endian_swap(32)) |
1661 S_038008_STRIDE(vb->stride));
1662 radeon_emit(cs, 0); /* RESOURCEi_WORD3 */
1663 radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1664 radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1665 radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD6 */
1666
1667 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1668 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1669 RADEON_USAGE_READ, RADEON_PRIO_VERTEX_BUFFER));
1670 }
1671 }
1672
r600_emit_constant_buffers(struct r600_context * rctx,struct r600_constbuf_state * state,unsigned buffer_id_base,unsigned reg_alu_constbuf_size,unsigned reg_alu_const_cache)1673 static void r600_emit_constant_buffers(struct r600_context *rctx,
1674 struct r600_constbuf_state *state,
1675 unsigned buffer_id_base,
1676 unsigned reg_alu_constbuf_size,
1677 unsigned reg_alu_const_cache)
1678 {
1679 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1680 uint32_t dirty_mask = state->dirty_mask;
1681
1682 while (dirty_mask) {
1683 struct pipe_constant_buffer *cb;
1684 struct r600_resource *rbuffer;
1685 unsigned offset;
1686 unsigned buffer_index = ffs(dirty_mask) - 1;
1687 unsigned gs_ring_buffer = (buffer_index == R600_GS_RING_CONST_BUFFER);
1688 cb = &state->cb[buffer_index];
1689 rbuffer = (struct r600_resource*)cb->buffer;
1690 assert(rbuffer);
1691
1692 offset = cb->buffer_offset;
1693
1694 if (!gs_ring_buffer) {
1695 radeon_set_context_reg(cs, reg_alu_constbuf_size + buffer_index * 4,
1696 DIV_ROUND_UP(cb->buffer_size, 256));
1697 radeon_set_context_reg(cs, reg_alu_const_cache + buffer_index * 4, offset >> 8);
1698 }
1699
1700 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1701 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1702 RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER));
1703
1704 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1705 radeon_emit(cs, (buffer_id_base + buffer_index) * 7);
1706 radeon_emit(cs, offset); /* RESOURCEi_WORD0 */
1707 radeon_emit(cs, rbuffer->b.b.width0 - offset - 1); /* RESOURCEi_WORD1 */
1708 radeon_emit(cs, /* RESOURCEi_WORD2 */
1709 S_038008_ENDIAN_SWAP(gs_ring_buffer ? ENDIAN_NONE : r600_endian_swap(32)) |
1710 S_038008_STRIDE(gs_ring_buffer ? 4 : 16));
1711 radeon_emit(cs, 0); /* RESOURCEi_WORD3 */
1712 radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1713 radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1714 radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD6 */
1715
1716 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1717 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1718 RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER));
1719
1720 dirty_mask &= ~(1 << buffer_index);
1721 }
1722 state->dirty_mask = 0;
1723 }
1724
r600_emit_vs_constant_buffers(struct r600_context * rctx,struct r600_atom * atom)1725 static void r600_emit_vs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1726 {
1727 r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX],
1728 R600_FETCH_CONSTANTS_OFFSET_VS,
1729 R_028180_ALU_CONST_BUFFER_SIZE_VS_0,
1730 R_028980_ALU_CONST_CACHE_VS_0);
1731 }
1732
r600_emit_gs_constant_buffers(struct r600_context * rctx,struct r600_atom * atom)1733 static void r600_emit_gs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1734 {
1735 r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY],
1736 R600_FETCH_CONSTANTS_OFFSET_GS,
1737 R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0,
1738 R_0289C0_ALU_CONST_CACHE_GS_0);
1739 }
1740
r600_emit_ps_constant_buffers(struct r600_context * rctx,struct r600_atom * atom)1741 static void r600_emit_ps_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1742 {
1743 r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT],
1744 R600_FETCH_CONSTANTS_OFFSET_PS,
1745 R_028140_ALU_CONST_BUFFER_SIZE_PS_0,
1746 R_028940_ALU_CONST_CACHE_PS_0);
1747 }
1748
r600_emit_sampler_views(struct r600_context * rctx,struct r600_samplerview_state * state,unsigned resource_id_base)1749 static void r600_emit_sampler_views(struct r600_context *rctx,
1750 struct r600_samplerview_state *state,
1751 unsigned resource_id_base)
1752 {
1753 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1754 uint32_t dirty_mask = state->dirty_mask;
1755
1756 while (dirty_mask) {
1757 struct r600_pipe_sampler_view *rview;
1758 unsigned resource_index = u_bit_scan(&dirty_mask);
1759 unsigned reloc;
1760
1761 rview = state->views[resource_index];
1762 assert(rview);
1763
1764 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1765 radeon_emit(cs, (resource_id_base + resource_index) * 7);
1766 radeon_emit_array(cs, rview->tex_resource_words, 7);
1767
1768 reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rview->tex_resource,
1769 RADEON_USAGE_READ,
1770 r600_get_sampler_view_priority(rview->tex_resource));
1771 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1772 radeon_emit(cs, reloc);
1773 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1774 radeon_emit(cs, reloc);
1775 }
1776 state->dirty_mask = 0;
1777 }
1778
1779
r600_emit_vs_sampler_views(struct r600_context * rctx,struct r600_atom * atom)1780 static void r600_emit_vs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1781 {
1782 r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views, R600_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS);
1783 }
1784
r600_emit_gs_sampler_views(struct r600_context * rctx,struct r600_atom * atom)1785 static void r600_emit_gs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1786 {
1787 r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views, R600_FETCH_CONSTANTS_OFFSET_GS + R600_MAX_CONST_BUFFERS);
1788 }
1789
r600_emit_ps_sampler_views(struct r600_context * rctx,struct r600_atom * atom)1790 static void r600_emit_ps_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1791 {
1792 r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views, R600_FETCH_CONSTANTS_OFFSET_PS + R600_MAX_CONST_BUFFERS);
1793 }
1794
r600_emit_sampler_states(struct r600_context * rctx,struct r600_textures_info * texinfo,unsigned resource_id_base,unsigned border_color_reg)1795 static void r600_emit_sampler_states(struct r600_context *rctx,
1796 struct r600_textures_info *texinfo,
1797 unsigned resource_id_base,
1798 unsigned border_color_reg)
1799 {
1800 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1801 uint32_t dirty_mask = texinfo->states.dirty_mask;
1802
1803 while (dirty_mask) {
1804 struct r600_pipe_sampler_state *rstate;
1805 struct r600_pipe_sampler_view *rview;
1806 unsigned i = u_bit_scan(&dirty_mask);
1807
1808 rstate = texinfo->states.states[i];
1809 assert(rstate);
1810 rview = texinfo->views.views[i];
1811
1812 /* TEX_ARRAY_OVERRIDE must be set for array textures to disable
1813 * filtering between layers.
1814 * Don't update TEX_ARRAY_OVERRIDE if we don't have the sampler view.
1815 */
1816 if (rview) {
1817 enum pipe_texture_target target = rview->base.texture->target;
1818 if (target == PIPE_TEXTURE_1D_ARRAY ||
1819 target == PIPE_TEXTURE_2D_ARRAY) {
1820 rstate->tex_sampler_words[0] |= S_03C000_TEX_ARRAY_OVERRIDE(1);
1821 texinfo->is_array_sampler[i] = true;
1822 } else {
1823 rstate->tex_sampler_words[0] &= C_03C000_TEX_ARRAY_OVERRIDE;
1824 texinfo->is_array_sampler[i] = false;
1825 }
1826 }
1827
1828 radeon_emit(cs, PKT3(PKT3_SET_SAMPLER, 3, 0));
1829 radeon_emit(cs, (resource_id_base + i) * 3);
1830 radeon_emit_array(cs, rstate->tex_sampler_words, 3);
1831
1832 if (rstate->border_color_use) {
1833 unsigned offset;
1834
1835 offset = border_color_reg;
1836 offset += i * 16;
1837 radeon_set_config_reg_seq(cs, offset, 4);
1838 radeon_emit_array(cs, rstate->border_color.ui, 4);
1839 }
1840 }
1841 texinfo->states.dirty_mask = 0;
1842 }
1843
r600_emit_vs_sampler_states(struct r600_context * rctx,struct r600_atom * atom)1844 static void r600_emit_vs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1845 {
1846 r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 18, R_00A600_TD_VS_SAMPLER0_BORDER_RED);
1847 }
1848
r600_emit_gs_sampler_states(struct r600_context * rctx,struct r600_atom * atom)1849 static void r600_emit_gs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1850 {
1851 r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY], 36, R_00A800_TD_GS_SAMPLER0_BORDER_RED);
1852 }
1853
r600_emit_ps_sampler_states(struct r600_context * rctx,struct r600_atom * atom)1854 static void r600_emit_ps_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1855 {
1856 r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT], 0, R_00A400_TD_PS_SAMPLER0_BORDER_RED);
1857 }
1858
r600_emit_seamless_cube_map(struct r600_context * rctx,struct r600_atom * atom)1859 static void r600_emit_seamless_cube_map(struct r600_context *rctx, struct r600_atom *atom)
1860 {
1861 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1862 unsigned tmp;
1863
1864 tmp = S_009508_DISABLE_CUBE_ANISO(1) |
1865 S_009508_SYNC_GRADIENT(1) |
1866 S_009508_SYNC_WALKER(1) |
1867 S_009508_SYNC_ALIGNER(1);
1868 if (!rctx->seamless_cube_map.enabled) {
1869 tmp |= S_009508_DISABLE_CUBE_WRAP(1);
1870 }
1871 radeon_set_config_reg(cs, R_009508_TA_CNTL_AUX, tmp);
1872 }
1873
r600_emit_sample_mask(struct r600_context * rctx,struct r600_atom * a)1874 static void r600_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
1875 {
1876 struct r600_sample_mask *s = (struct r600_sample_mask*)a;
1877 uint8_t mask = s->sample_mask;
1878
1879 radeon_set_context_reg(rctx->b.gfx.cs, R_028C48_PA_SC_AA_MASK,
1880 mask | (mask << 8) | (mask << 16) | (mask << 24));
1881 }
1882
r600_emit_vertex_fetch_shader(struct r600_context * rctx,struct r600_atom * a)1883 static void r600_emit_vertex_fetch_shader(struct r600_context *rctx, struct r600_atom *a)
1884 {
1885 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1886 struct r600_cso_state *state = (struct r600_cso_state*)a;
1887 struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state->cso;
1888
1889 radeon_set_context_reg(cs, R_028894_SQ_PGM_START_FS, shader->offset >> 8);
1890 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1891 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, shader->buffer,
1892 RADEON_USAGE_READ,
1893 RADEON_PRIO_SHADER_BINARY));
1894 }
1895
r600_emit_shader_stages(struct r600_context * rctx,struct r600_atom * a)1896 static void r600_emit_shader_stages(struct r600_context *rctx, struct r600_atom *a)
1897 {
1898 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1899 struct r600_shader_stages_state *state = (struct r600_shader_stages_state*)a;
1900
1901 uint32_t v2 = 0, primid = 0;
1902
1903 if (rctx->vs_shader->current->shader.vs_as_gs_a) {
1904 v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_A);
1905 primid = 1;
1906 }
1907
1908 if (state->geom_enable) {
1909 uint32_t cut_val;
1910
1911 if (rctx->gs_shader->gs_max_out_vertices <= 128)
1912 cut_val = V_028A40_GS_CUT_128;
1913 else if (rctx->gs_shader->gs_max_out_vertices <= 256)
1914 cut_val = V_028A40_GS_CUT_256;
1915 else if (rctx->gs_shader->gs_max_out_vertices <= 512)
1916 cut_val = V_028A40_GS_CUT_512;
1917 else
1918 cut_val = V_028A40_GS_CUT_1024;
1919
1920 v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_G) |
1921 S_028A40_CUT_MODE(cut_val);
1922
1923 if (rctx->gs_shader->current->shader.gs_prim_id_input)
1924 primid = 1;
1925 }
1926
1927 radeon_set_context_reg(cs, R_028A40_VGT_GS_MODE, v2);
1928 radeon_set_context_reg(cs, R_028A84_VGT_PRIMITIVEID_EN, primid);
1929 }
1930
r600_emit_gs_rings(struct r600_context * rctx,struct r600_atom * a)1931 static void r600_emit_gs_rings(struct r600_context *rctx, struct r600_atom *a)
1932 {
1933 struct radeon_winsys_cs *cs = rctx->b.gfx.cs;
1934 struct r600_gs_rings_state *state = (struct r600_gs_rings_state*)a;
1935 struct r600_resource *rbuffer;
1936
1937 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
1938 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1939 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
1940
1941 if (state->enable) {
1942 rbuffer =(struct r600_resource*)state->esgs_ring.buffer;
1943 radeon_set_config_reg(cs, R_008C40_SQ_ESGS_RING_BASE, 0);
1944 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1945 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1946 RADEON_USAGE_READWRITE,
1947 RADEON_PRIO_SHADER_RINGS));
1948 radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE,
1949 state->esgs_ring.buffer_size >> 8);
1950
1951 rbuffer =(struct r600_resource*)state->gsvs_ring.buffer;
1952 radeon_set_config_reg(cs, R_008C48_SQ_GSVS_RING_BASE, 0);
1953 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1954 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1955 RADEON_USAGE_READWRITE,
1956 RADEON_PRIO_SHADER_RINGS));
1957 radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE,
1958 state->gsvs_ring.buffer_size >> 8);
1959 } else {
1960 radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE, 0);
1961 radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE, 0);
1962 }
1963
1964 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
1965 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1966 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
1967 }
1968
1969 /* Adjust GPR allocation on R6xx/R7xx */
r600_adjust_gprs(struct r600_context * rctx)1970 bool r600_adjust_gprs(struct r600_context *rctx)
1971 {
1972 unsigned num_gprs[R600_NUM_HW_STAGES];
1973 unsigned new_gprs[R600_NUM_HW_STAGES];
1974 unsigned cur_gprs[R600_NUM_HW_STAGES];
1975 unsigned def_gprs[R600_NUM_HW_STAGES];
1976 unsigned def_num_clause_temp_gprs = rctx->r6xx_num_clause_temp_gprs;
1977 unsigned max_gprs;
1978 unsigned tmp, tmp2;
1979 unsigned i;
1980 bool need_recalc = false, use_default = true;
1981
1982 /* hardware will reserve twice num_clause_temp_gprs */
1983 max_gprs = def_num_clause_temp_gprs * 2;
1984 for (i = 0; i < R600_NUM_HW_STAGES; i++) {
1985 def_gprs[i] = rctx->default_gprs[i];
1986 max_gprs += def_gprs[i];
1987 }
1988
1989 cur_gprs[R600_HW_STAGE_PS] = G_008C04_NUM_PS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
1990 cur_gprs[R600_HW_STAGE_VS] = G_008C04_NUM_VS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
1991 cur_gprs[R600_HW_STAGE_GS] = G_008C08_NUM_GS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
1992 cur_gprs[R600_HW_STAGE_ES] = G_008C08_NUM_ES_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
1993
1994 num_gprs[R600_HW_STAGE_PS] = rctx->ps_shader->current->shader.bc.ngpr;
1995 if (rctx->gs_shader) {
1996 num_gprs[R600_HW_STAGE_ES] = rctx->vs_shader->current->shader.bc.ngpr;
1997 num_gprs[R600_HW_STAGE_GS] = rctx->gs_shader->current->shader.bc.ngpr;
1998 num_gprs[R600_HW_STAGE_VS] = rctx->gs_shader->current->gs_copy_shader->shader.bc.ngpr;
1999 } else {
2000 num_gprs[R600_HW_STAGE_ES] = 0;
2001 num_gprs[R600_HW_STAGE_GS] = 0;
2002 num_gprs[R600_HW_STAGE_VS] = rctx->vs_shader->current->shader.bc.ngpr;
2003 }
2004
2005 for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2006 new_gprs[i] = num_gprs[i];
2007 if (new_gprs[i] > cur_gprs[i])
2008 need_recalc = true;
2009 if (new_gprs[i] > def_gprs[i])
2010 use_default = false;
2011 }
2012
2013 /* the sum of all SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS must <= to max_gprs */
2014 if (!need_recalc)
2015 return true;
2016
2017 /* try to use switch back to default */
2018 if (!use_default) {
2019 /* always privilege vs stage so that at worst we have the
2020 * pixel stage producing wrong output (not the vertex
2021 * stage) */
2022 new_gprs[R600_HW_STAGE_PS] = max_gprs - def_num_clause_temp_gprs * 2;
2023 for (i = R600_HW_STAGE_VS; i < R600_NUM_HW_STAGES; i++)
2024 new_gprs[R600_HW_STAGE_PS] -= new_gprs[i];
2025 } else {
2026 for (i = 0; i < R600_NUM_HW_STAGES; i++)
2027 new_gprs[i] = def_gprs[i];
2028 }
2029
2030 /* SQ_PGM_RESOURCES_*.NUM_GPRS must always be program to a value <=
2031 * SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS otherwise the GPU will lockup
2032 * Also if a shader use more gpr than SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS
2033 * it will lockup. So in this case just discard the draw command
2034 * and don't change the current gprs repartitions.
2035 */
2036 for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2037 if (num_gprs[i] > new_gprs[i]) {
2038 R600_ERR("shaders require too many register (%d + %d + %d + %d) "
2039 "for a combined maximum of %d\n",
2040 num_gprs[R600_HW_STAGE_PS], num_gprs[R600_HW_STAGE_VS], num_gprs[R600_HW_STAGE_ES], num_gprs[R600_HW_STAGE_GS], max_gprs);
2041 return false;
2042 }
2043 }
2044
2045 /* in some case we endup recomputing the current value */
2046 tmp = S_008C04_NUM_PS_GPRS(new_gprs[R600_HW_STAGE_PS]) |
2047 S_008C04_NUM_VS_GPRS(new_gprs[R600_HW_STAGE_VS]) |
2048 S_008C04_NUM_CLAUSE_TEMP_GPRS(def_num_clause_temp_gprs);
2049
2050 tmp2 = S_008C08_NUM_ES_GPRS(new_gprs[R600_HW_STAGE_ES]) |
2051 S_008C08_NUM_GS_GPRS(new_gprs[R600_HW_STAGE_GS]);
2052 if (rctx->config_state.sq_gpr_resource_mgmt_1 != tmp || rctx->config_state.sq_gpr_resource_mgmt_2 != tmp2) {
2053 rctx->config_state.sq_gpr_resource_mgmt_1 = tmp;
2054 rctx->config_state.sq_gpr_resource_mgmt_2 = tmp2;
2055 r600_mark_atom_dirty(rctx, &rctx->config_state.atom);
2056 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE;
2057 }
2058 return true;
2059 }
2060
r600_init_atom_start_cs(struct r600_context * rctx)2061 void r600_init_atom_start_cs(struct r600_context *rctx)
2062 {
2063 int ps_prio;
2064 int vs_prio;
2065 int gs_prio;
2066 int es_prio;
2067 int num_ps_gprs;
2068 int num_vs_gprs;
2069 int num_gs_gprs;
2070 int num_es_gprs;
2071 int num_temp_gprs;
2072 int num_ps_threads;
2073 int num_vs_threads;
2074 int num_gs_threads;
2075 int num_es_threads;
2076 int num_ps_stack_entries;
2077 int num_vs_stack_entries;
2078 int num_gs_stack_entries;
2079 int num_es_stack_entries;
2080 enum radeon_family family;
2081 struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2082 uint32_t tmp, i;
2083
2084 r600_init_command_buffer(cb, 256);
2085
2086 /* R6xx requires this packet at the start of each command buffer */
2087 if (rctx->b.chip_class == R600) {
2088 r600_store_value(cb, PKT3(PKT3_START_3D_CMDBUF, 0, 0));
2089 r600_store_value(cb, 0);
2090 }
2091 /* All asics require this one */
2092 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2093 r600_store_value(cb, 0x80000000);
2094 r600_store_value(cb, 0x80000000);
2095
2096 /* We're setting config registers here. */
2097 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2098 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2099
2100 /* This enables pipeline stat & streamout queries.
2101 * They are only disabled by blits.
2102 */
2103 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2104 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PIPELINESTAT_START) | EVENT_INDEX(0));
2105
2106 family = rctx->b.family;
2107 ps_prio = 0;
2108 vs_prio = 1;
2109 gs_prio = 2;
2110 es_prio = 3;
2111 switch (family) {
2112 case CHIP_R600:
2113 num_ps_gprs = 192;
2114 num_vs_gprs = 56;
2115 num_temp_gprs = 4;
2116 num_gs_gprs = 0;
2117 num_es_gprs = 0;
2118 num_ps_threads = 136;
2119 num_vs_threads = 48;
2120 num_gs_threads = 4;
2121 num_es_threads = 4;
2122 num_ps_stack_entries = 128;
2123 num_vs_stack_entries = 128;
2124 num_gs_stack_entries = 0;
2125 num_es_stack_entries = 0;
2126 break;
2127 case CHIP_RV630:
2128 case CHIP_RV635:
2129 num_ps_gprs = 84;
2130 num_vs_gprs = 36;
2131 num_temp_gprs = 4;
2132 num_gs_gprs = 0;
2133 num_es_gprs = 0;
2134 num_ps_threads = 144;
2135 num_vs_threads = 40;
2136 num_gs_threads = 4;
2137 num_es_threads = 4;
2138 num_ps_stack_entries = 40;
2139 num_vs_stack_entries = 40;
2140 num_gs_stack_entries = 32;
2141 num_es_stack_entries = 16;
2142 break;
2143 case CHIP_RV610:
2144 case CHIP_RV620:
2145 case CHIP_RS780:
2146 case CHIP_RS880:
2147 default:
2148 num_ps_gprs = 84;
2149 num_vs_gprs = 36;
2150 num_temp_gprs = 4;
2151 num_gs_gprs = 0;
2152 num_es_gprs = 0;
2153 /* use limits 40 VS and at least 16 ES/GS */
2154 num_ps_threads = 120;
2155 num_vs_threads = 40;
2156 num_gs_threads = 16;
2157 num_es_threads = 16;
2158 num_ps_stack_entries = 40;
2159 num_vs_stack_entries = 40;
2160 num_gs_stack_entries = 32;
2161 num_es_stack_entries = 16;
2162 break;
2163 case CHIP_RV670:
2164 num_ps_gprs = 144;
2165 num_vs_gprs = 40;
2166 num_temp_gprs = 4;
2167 num_gs_gprs = 0;
2168 num_es_gprs = 0;
2169 num_ps_threads = 136;
2170 num_vs_threads = 48;
2171 num_gs_threads = 4;
2172 num_es_threads = 4;
2173 num_ps_stack_entries = 40;
2174 num_vs_stack_entries = 40;
2175 num_gs_stack_entries = 32;
2176 num_es_stack_entries = 16;
2177 break;
2178 case CHIP_RV770:
2179 num_ps_gprs = 130;
2180 num_vs_gprs = 56;
2181 num_temp_gprs = 4;
2182 num_gs_gprs = 31;
2183 num_es_gprs = 31;
2184 num_ps_threads = 180;
2185 num_vs_threads = 60;
2186 num_gs_threads = 4;
2187 num_es_threads = 4;
2188 num_ps_stack_entries = 128;
2189 num_vs_stack_entries = 128;
2190 num_gs_stack_entries = 128;
2191 num_es_stack_entries = 128;
2192 break;
2193 case CHIP_RV730:
2194 case CHIP_RV740:
2195 num_ps_gprs = 84;
2196 num_vs_gprs = 36;
2197 num_temp_gprs = 4;
2198 num_gs_gprs = 0;
2199 num_es_gprs = 0;
2200 num_ps_threads = 180;
2201 num_vs_threads = 60;
2202 num_gs_threads = 4;
2203 num_es_threads = 4;
2204 num_ps_stack_entries = 128;
2205 num_vs_stack_entries = 128;
2206 num_gs_stack_entries = 0;
2207 num_es_stack_entries = 0;
2208 break;
2209 case CHIP_RV710:
2210 num_ps_gprs = 192;
2211 num_vs_gprs = 56;
2212 num_temp_gprs = 4;
2213 num_gs_gprs = 0;
2214 num_es_gprs = 0;
2215 num_ps_threads = 136;
2216 num_vs_threads = 48;
2217 num_gs_threads = 4;
2218 num_es_threads = 4;
2219 num_ps_stack_entries = 128;
2220 num_vs_stack_entries = 128;
2221 num_gs_stack_entries = 0;
2222 num_es_stack_entries = 0;
2223 break;
2224 }
2225
2226 rctx->default_gprs[R600_HW_STAGE_PS] = num_ps_gprs;
2227 rctx->default_gprs[R600_HW_STAGE_VS] = num_vs_gprs;
2228 rctx->default_gprs[R600_HW_STAGE_GS] = 0;
2229 rctx->default_gprs[R600_HW_STAGE_ES] = 0;
2230
2231 rctx->r6xx_num_clause_temp_gprs = num_temp_gprs;
2232
2233 /* SQ_CONFIG */
2234 tmp = 0;
2235 switch (family) {
2236 case CHIP_RV610:
2237 case CHIP_RV620:
2238 case CHIP_RS780:
2239 case CHIP_RS880:
2240 case CHIP_RV710:
2241 break;
2242 default:
2243 tmp |= S_008C00_VC_ENABLE(1);
2244 break;
2245 }
2246 tmp |= S_008C00_DX9_CONSTS(0);
2247 tmp |= S_008C00_ALU_INST_PREFER_VECTOR(1);
2248 tmp |= S_008C00_PS_PRIO(ps_prio);
2249 tmp |= S_008C00_VS_PRIO(vs_prio);
2250 tmp |= S_008C00_GS_PRIO(gs_prio);
2251 tmp |= S_008C00_ES_PRIO(es_prio);
2252 r600_store_config_reg(cb, R_008C00_SQ_CONFIG, tmp);
2253
2254 /* SQ_GPR_RESOURCE_MGMT_2 */
2255 tmp = S_008C08_NUM_GS_GPRS(num_gs_gprs);
2256 tmp |= S_008C08_NUM_ES_GPRS(num_es_gprs);
2257 r600_store_config_reg_seq(cb, R_008C08_SQ_GPR_RESOURCE_MGMT_2, 4);
2258 r600_store_value(cb, tmp);
2259
2260 /* SQ_THREAD_RESOURCE_MGMT */
2261 tmp = S_008C0C_NUM_PS_THREADS(num_ps_threads);
2262 tmp |= S_008C0C_NUM_VS_THREADS(num_vs_threads);
2263 tmp |= S_008C0C_NUM_GS_THREADS(num_gs_threads);
2264 tmp |= S_008C0C_NUM_ES_THREADS(num_es_threads);
2265 r600_store_value(cb, tmp); /* R_008C0C_SQ_THREAD_RESOURCE_MGMT */
2266
2267 /* SQ_STACK_RESOURCE_MGMT_1 */
2268 tmp = S_008C10_NUM_PS_STACK_ENTRIES(num_ps_stack_entries);
2269 tmp |= S_008C10_NUM_VS_STACK_ENTRIES(num_vs_stack_entries);
2270 r600_store_value(cb, tmp); /* R_008C10_SQ_STACK_RESOURCE_MGMT_1 */
2271
2272 /* SQ_STACK_RESOURCE_MGMT_2 */
2273 tmp = S_008C14_NUM_GS_STACK_ENTRIES(num_gs_stack_entries);
2274 tmp |= S_008C14_NUM_ES_STACK_ENTRIES(num_es_stack_entries);
2275 r600_store_value(cb, tmp); /* R_008C14_SQ_STACK_RESOURCE_MGMT_2 */
2276
2277 r600_store_config_reg(cb, R_009714_VC_ENHANCE, 0);
2278
2279 if (rctx->b.chip_class >= R700) {
2280 r600_store_context_reg(cb, R_028A50_VGT_ENHANCE, 4);
2281 r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0x00004000);
2282 r600_store_config_reg(cb, R_009830_DB_DEBUG, 0);
2283 r600_store_config_reg(cb, R_009838_DB_WATERMARKS, 0x00420204);
2284 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
2285 } else {
2286 r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0);
2287 r600_store_config_reg(cb, R_009830_DB_DEBUG, 0x82000000);
2288 r600_store_config_reg(cb, R_009838_DB_WATERMARKS, 0x01020204);
2289 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 1);
2290 }
2291 r600_store_context_reg_seq(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE, 9);
2292 r600_store_value(cb, 0); /* R_0288A8_SQ_ESGS_RING_ITEMSIZE */
2293 r600_store_value(cb, 0); /* R_0288AC_SQ_GSVS_RING_ITEMSIZE */
2294 r600_store_value(cb, 0); /* R_0288B0_SQ_ESTMP_RING_ITEMSIZE */
2295 r600_store_value(cb, 0); /* R_0288B4_SQ_GSTMP_RING_ITEMSIZE */
2296 r600_store_value(cb, 0); /* R_0288B8_SQ_VSTMP_RING_ITEMSIZE */
2297 r600_store_value(cb, 0); /* R_0288BC_SQ_PSTMP_RING_ITEMSIZE */
2298 r600_store_value(cb, 0); /* R_0288C0_SQ_FBUF_RING_ITEMSIZE */
2299 r600_store_value(cb, 0); /* R_0288C4_SQ_REDUC_RING_ITEMSIZE */
2300 r600_store_value(cb, 0); /* R_0288C8_SQ_GS_VERT_ITEMSIZE */
2301
2302 /* to avoid GPU doing any preloading of constant from random address */
2303 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
2304 for (i = 0; i < 16; i++)
2305 r600_store_value(cb, 0);
2306
2307 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
2308 for (i = 0; i < 16; i++)
2309 r600_store_value(cb, 0);
2310
2311 r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16);
2312 for (i = 0; i < 16; i++)
2313 r600_store_value(cb, 0);
2314
2315 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
2316 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
2317 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
2318 r600_store_value(cb, 0); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
2319 r600_store_value(cb, 0); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
2320 r600_store_value(cb, 0); /* R_028A20_VGT_HOS_REUSE_DEPTH */
2321 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
2322 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
2323 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
2324 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
2325 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
2326 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
2327 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
2328 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE, 0); */
2329
2330 r600_store_context_reg(cb, R_028A84_VGT_PRIMITIVEID_EN, 0);
2331 r600_store_context_reg(cb, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 0);
2332 r600_store_context_reg(cb, R_028AA4_VGT_INSTANCE_STEP_RATE_1, 0);
2333
2334 r600_store_context_reg_seq(cb, R_028AB4_VGT_REUSE_OFF, 2);
2335 r600_store_value(cb, 1); /* R_028AB4_VGT_REUSE_OFF */
2336 r600_store_value(cb, 0); /* R_028AB8_VGT_VTX_CNT_EN */
2337
2338 r600_store_context_reg(cb, R_028B20_VGT_STRMOUT_BUFFER_EN, 0);
2339
2340 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
2341
2342 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
2343
2344 r600_store_context_reg_seq(cb, R_0286DC_SPI_FOG_CNTL, 3);
2345 r600_store_value(cb, 0); /* R_0286DC_SPI_FOG_CNTL */
2346 r600_store_value(cb, 0); /* R_0286E0_SPI_FOG_FUNC_SCALE */
2347 r600_store_value(cb, 0); /* R_0286E4_SPI_FOG_FUNC_BIAS */
2348
2349 r600_store_context_reg_seq(cb, R_028D28_DB_SRESULTS_COMPARE_STATE0, 3);
2350 r600_store_value(cb, 0); /* R_028D28_DB_SRESULTS_COMPARE_STATE0 */
2351 r600_store_value(cb, 0); /* R_028D2C_DB_SRESULTS_COMPARE_STATE1 */
2352 r600_store_value(cb, 0); /* R_028D30_DB_PRELOAD_CONTROL */
2353
2354 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
2355 r600_store_context_reg(cb, R_028A48_PA_SC_MPASS_PS_CNTL, 0);
2356
2357 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
2358 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
2359
2360 if (rctx->b.chip_class >= R700) {
2361 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
2362 }
2363
2364 r600_store_context_reg_seq(cb, R_028C30_CB_CLRCMP_CONTROL, 4);
2365 r600_store_value(cb, 0x1000000); /* R_028C30_CB_CLRCMP_CONTROL */
2366 r600_store_value(cb, 0); /* R_028C34_CB_CLRCMP_SRC */
2367 r600_store_value(cb, 0xFF); /* R_028C38_CB_CLRCMP_DST */
2368 r600_store_value(cb, 0xFFFFFFFF); /* R_028C3C_CB_CLRCMP_MSK */
2369
2370 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
2371 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
2372 r600_store_value(cb, S_028034_BR_X(8192) | S_028034_BR_Y(8192)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
2373
2374 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
2375 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
2376 r600_store_value(cb, S_028244_BR_X(8192) | S_028244_BR_Y(8192)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
2377
2378 r600_store_context_reg_seq(cb, R_0288CC_SQ_PGM_CF_OFFSET_PS, 5);
2379 r600_store_value(cb, 0); /* R_0288CC_SQ_PGM_CF_OFFSET_PS */
2380 r600_store_value(cb, 0); /* R_0288D0_SQ_PGM_CF_OFFSET_VS */
2381 r600_store_value(cb, 0); /* R_0288D4_SQ_PGM_CF_OFFSET_GS */
2382 r600_store_value(cb, 0); /* R_0288D8_SQ_PGM_CF_OFFSET_ES */
2383 r600_store_value(cb, 0); /* R_0288DC_SQ_PGM_CF_OFFSET_FS */
2384
2385 r600_store_context_reg(cb, R_0288E0_SQ_VTX_SEMANTIC_CLEAR, ~0);
2386
2387 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
2388 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
2389 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
2390
2391 r600_store_context_reg(cb, R_0288A4_SQ_PGM_RESOURCES_FS, 0);
2392
2393 if (rctx->b.chip_class == R700)
2394 r600_store_context_reg(cb, R_028350_SX_MISC, 0);
2395 if (rctx->b.chip_class == R700 && rctx->screen->b.has_streamout)
2396 r600_store_context_reg(cb, R_028354_SX_SURFACE_SYNC, S_028354_SURFACE_SYNC_MASK(0xf));
2397
2398 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2399 if (rctx->screen->b.has_streamout) {
2400 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
2401 }
2402
2403 r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0, 0x1000FFF);
2404 r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0 + (32 * 4), 0x1000FFF);
2405 r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0 + (64 * 4), 0x1000FFF);
2406 }
2407
r600_update_ps_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2408 void r600_update_ps_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2409 {
2410 struct r600_context *rctx = (struct r600_context *)ctx;
2411 struct r600_command_buffer *cb = &shader->command_buffer;
2412 struct r600_shader *rshader = &shader->shader;
2413 unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1, db_shader_control;
2414 int pos_index = -1, face_index = -1, fixed_pt_position_index = -1;
2415 unsigned tmp, sid, ufi = 0;
2416 int need_linear = 0;
2417 unsigned z_export = 0, stencil_export = 0, mask_export = 0;
2418 unsigned sprite_coord_enable = rctx->rasterizer ? rctx->rasterizer->sprite_coord_enable : 0;
2419
2420 if (!cb->buf) {
2421 r600_init_command_buffer(cb, 64);
2422 } else {
2423 cb->num_dw = 0;
2424 }
2425
2426 r600_store_context_reg_seq(cb, R_028644_SPI_PS_INPUT_CNTL_0, rshader->ninput);
2427 for (i = 0; i < rshader->ninput; i++) {
2428 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
2429 pos_index = i;
2430 if (rshader->input[i].name == TGSI_SEMANTIC_FACE && face_index == -1)
2431 face_index = i;
2432 if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEID)
2433 fixed_pt_position_index = i;
2434
2435 sid = rshader->input[i].spi_sid;
2436
2437 tmp = S_028644_SEMANTIC(sid);
2438
2439 /* D3D 9 behaviour. GL is undefined */
2440 if (rshader->input[i].name == TGSI_SEMANTIC_COLOR && rshader->input[i].sid == 0)
2441 tmp |= S_028644_DEFAULT_VAL(3);
2442
2443 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION ||
2444 rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
2445 (rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR &&
2446 rctx->rasterizer && rctx->rasterizer->flatshade))
2447 tmp |= S_028644_FLAT_SHADE(1);
2448
2449 if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
2450 sprite_coord_enable & (1 << rshader->input[i].sid)) {
2451 tmp |= S_028644_PT_SPRITE_TEX(1);
2452 }
2453
2454 if (rshader->input[i].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID)
2455 tmp |= S_028644_SEL_CENTROID(1);
2456
2457 if (rshader->input[i].interpolate_location == TGSI_INTERPOLATE_LOC_SAMPLE)
2458 tmp |= S_028644_SEL_SAMPLE(1);
2459
2460 if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR) {
2461 need_linear = 1;
2462 tmp |= S_028644_SEL_LINEAR(1);
2463 }
2464
2465 r600_store_value(cb, tmp);
2466 }
2467
2468 db_shader_control = 0;
2469 for (i = 0; i < rshader->noutput; i++) {
2470 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
2471 z_export = 1;
2472 if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
2473 stencil_export = 1;
2474 if (rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK &&
2475 rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0)
2476 mask_export = 1;
2477 }
2478 db_shader_control |= S_02880C_Z_EXPORT_ENABLE(z_export);
2479 db_shader_control |= S_02880C_STENCIL_REF_EXPORT_ENABLE(stencil_export);
2480 db_shader_control |= S_02880C_MASK_EXPORT_ENABLE(mask_export);
2481 if (rshader->uses_kill)
2482 db_shader_control |= S_02880C_KILL_ENABLE(1);
2483
2484 exports_ps = 0;
2485 for (i = 0; i < rshader->noutput; i++) {
2486 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
2487 rshader->output[i].name == TGSI_SEMANTIC_STENCIL ||
2488 rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK) {
2489 exports_ps |= 1;
2490 }
2491 }
2492 num_cout = rshader->nr_ps_color_exports;
2493 exports_ps |= S_028854_EXPORT_COLORS(num_cout);
2494 if (!exports_ps) {
2495 /* always at least export 1 component per pixel */
2496 exports_ps = 2;
2497 }
2498
2499 shader->nr_ps_color_outputs = num_cout;
2500
2501 spi_ps_in_control_0 = S_0286CC_NUM_INTERP(rshader->ninput) |
2502 S_0286CC_PERSP_GRADIENT_ENA(1)|
2503 S_0286CC_LINEAR_GRADIENT_ENA(need_linear);
2504 spi_input_z = 0;
2505 if (pos_index != -1) {
2506 spi_ps_in_control_0 |= (S_0286CC_POSITION_ENA(1) |
2507 S_0286CC_POSITION_CENTROID(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID) |
2508 S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr) |
2509 S_0286CC_BARYC_SAMPLE_CNTL(1)) |
2510 S_0286CC_POSITION_SAMPLE(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_SAMPLE);
2511 spi_input_z |= S_0286D8_PROVIDE_Z_TO_SPI(1);
2512 }
2513
2514 spi_ps_in_control_1 = 0;
2515 if (face_index != -1) {
2516 spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) |
2517 S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr);
2518 }
2519 if (fixed_pt_position_index != -1) {
2520 spi_ps_in_control_1 |= S_0286D0_FIXED_PT_POSITION_ENA(1) |
2521 S_0286D0_FIXED_PT_POSITION_ADDR(rshader->input[fixed_pt_position_index].gpr);
2522 }
2523
2524 /* HW bug in original R600 */
2525 if (rctx->b.family == CHIP_R600)
2526 ufi = 1;
2527
2528 r600_store_context_reg_seq(cb, R_0286CC_SPI_PS_IN_CONTROL_0, 2);
2529 r600_store_value(cb, spi_ps_in_control_0); /* R_0286CC_SPI_PS_IN_CONTROL_0 */
2530 r600_store_value(cb, spi_ps_in_control_1); /* R_0286D0_SPI_PS_IN_CONTROL_1 */
2531
2532 r600_store_context_reg(cb, R_0286D8_SPI_INPUT_Z, spi_input_z);
2533
2534 r600_store_context_reg_seq(cb, R_028850_SQ_PGM_RESOURCES_PS, 2);
2535 r600_store_value(cb, /* R_028850_SQ_PGM_RESOURCES_PS*/
2536 S_028850_NUM_GPRS(rshader->bc.ngpr) |
2537 S_028850_STACK_SIZE(rshader->bc.nstack) |
2538 S_028850_UNCACHED_FIRST_INST(ufi));
2539 r600_store_value(cb, exports_ps); /* R_028854_SQ_PGM_EXPORTS_PS */
2540
2541 r600_store_context_reg(cb, R_028840_SQ_PGM_START_PS, 0);
2542 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2543
2544 /* only set some bits here, the other bits are set in the dsa state */
2545 shader->db_shader_control = db_shader_control;
2546 shader->ps_depth_export = z_export | stencil_export | mask_export;
2547
2548 shader->sprite_coord_enable = sprite_coord_enable;
2549 if (rctx->rasterizer)
2550 shader->flatshade = rctx->rasterizer->flatshade;
2551 }
2552
r600_update_vs_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2553 void r600_update_vs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2554 {
2555 struct r600_command_buffer *cb = &shader->command_buffer;
2556 struct r600_shader *rshader = &shader->shader;
2557 unsigned spi_vs_out_id[10] = {};
2558 unsigned i, tmp, nparams = 0;
2559
2560 for (i = 0; i < rshader->noutput; i++) {
2561 if (rshader->output[i].spi_sid) {
2562 tmp = rshader->output[i].spi_sid << ((nparams & 3) * 8);
2563 spi_vs_out_id[nparams / 4] |= tmp;
2564 nparams++;
2565 }
2566 }
2567
2568 r600_init_command_buffer(cb, 32);
2569
2570 r600_store_context_reg_seq(cb, R_028614_SPI_VS_OUT_ID_0, 10);
2571 for (i = 0; i < 10; i++) {
2572 r600_store_value(cb, spi_vs_out_id[i]);
2573 }
2574
2575 /* Certain attributes (position, psize, etc.) don't count as params.
2576 * VS is required to export at least one param and r600_shader_from_tgsi()
2577 * takes care of adding a dummy export.
2578 */
2579 if (nparams < 1)
2580 nparams = 1;
2581
2582 r600_store_context_reg(cb, R_0286C4_SPI_VS_OUT_CONFIG,
2583 S_0286C4_VS_EXPORT_COUNT(nparams - 1));
2584 r600_store_context_reg(cb, R_028868_SQ_PGM_RESOURCES_VS,
2585 S_028868_NUM_GPRS(rshader->bc.ngpr) |
2586 S_028868_STACK_SIZE(rshader->bc.nstack));
2587 if (rshader->vs_position_window_space) {
2588 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
2589 S_028818_VTX_XY_FMT(1) | S_028818_VTX_Z_FMT(1));
2590 } else {
2591 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
2592 S_028818_VTX_W0_FMT(1) |
2593 S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) |
2594 S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
2595 S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1));
2596
2597 }
2598 r600_store_context_reg(cb, R_028858_SQ_PGM_START_VS, 0);
2599 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2600
2601 shader->pa_cl_vs_out_cntl =
2602 S_02881C_VS_OUT_CCDIST0_VEC_ENA((rshader->clip_dist_write & 0x0F) != 0) |
2603 S_02881C_VS_OUT_CCDIST1_VEC_ENA((rshader->clip_dist_write & 0xF0) != 0) |
2604 S_02881C_VS_OUT_MISC_VEC_ENA(rshader->vs_out_misc_write) |
2605 S_02881C_USE_VTX_POINT_SIZE(rshader->vs_out_point_size) |
2606 S_02881C_USE_VTX_EDGE_FLAG(rshader->vs_out_edgeflag) |
2607 S_02881C_USE_VTX_RENDER_TARGET_INDX(rshader->vs_out_layer) |
2608 S_02881C_USE_VTX_VIEWPORT_INDX(rshader->vs_out_viewport);
2609 }
2610
2611 #define RV610_GSVS_ALIGN 32
2612 #define R600_GSVS_ALIGN 16
2613
r600_update_gs_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2614 void r600_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2615 {
2616 struct r600_context *rctx = (struct r600_context *)ctx;
2617 struct r600_command_buffer *cb = &shader->command_buffer;
2618 struct r600_shader *rshader = &shader->shader;
2619 struct r600_shader *cp_shader = &shader->gs_copy_shader->shader;
2620 unsigned gsvs_itemsize =
2621 (cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2;
2622
2623 /* some r600s needs gsvs itemsize aligned to cacheline size
2624 this was fixed in rs780 and above. */
2625 switch (rctx->b.family) {
2626 case CHIP_RV610:
2627 gsvs_itemsize = align(gsvs_itemsize, RV610_GSVS_ALIGN);
2628 break;
2629 case CHIP_R600:
2630 case CHIP_RV630:
2631 case CHIP_RV670:
2632 case CHIP_RV620:
2633 case CHIP_RV635:
2634 gsvs_itemsize = align(gsvs_itemsize, R600_GSVS_ALIGN);
2635 break;
2636 default:
2637 break;
2638 }
2639
2640 r600_init_command_buffer(cb, 64);
2641
2642 /* VGT_GS_MODE is written by r600_emit_shader_stages */
2643 r600_store_context_reg(cb, R_028AB8_VGT_VTX_CNT_EN, 1);
2644
2645 if (rctx->b.chip_class >= R700) {
2646 r600_store_context_reg(cb, R_028B38_VGT_GS_MAX_VERT_OUT,
2647 S_028B38_MAX_VERT_OUT(shader->selector->gs_max_out_vertices));
2648 }
2649 r600_store_context_reg(cb, R_028A6C_VGT_GS_OUT_PRIM_TYPE,
2650 r600_conv_prim_to_gs_out(shader->selector->gs_output_prim));
2651
2652 r600_store_context_reg(cb, R_0288C8_SQ_GS_VERT_ITEMSIZE,
2653 cp_shader->ring_item_sizes[0] >> 2);
2654
2655 r600_store_context_reg(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE,
2656 (rshader->ring_item_sizes[0]) >> 2);
2657
2658 r600_store_context_reg(cb, R_0288AC_SQ_GSVS_RING_ITEMSIZE,
2659 gsvs_itemsize);
2660
2661 /* FIXME calculate these values somehow ??? */
2662 r600_store_config_reg_seq(cb, R_0088C8_VGT_GS_PER_ES, 2);
2663 r600_store_value(cb, 0x80); /* GS_PER_ES */
2664 r600_store_value(cb, 0x100); /* ES_PER_GS */
2665 r600_store_config_reg_seq(cb, R_0088E8_VGT_GS_PER_VS, 1);
2666 r600_store_value(cb, 0x2); /* GS_PER_VS */
2667
2668 r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_GS,
2669 S_02887C_NUM_GPRS(rshader->bc.ngpr) |
2670 S_02887C_STACK_SIZE(rshader->bc.nstack));
2671 r600_store_context_reg(cb, R_02886C_SQ_PGM_START_GS, 0);
2672 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2673 }
2674
r600_update_es_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2675 void r600_update_es_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2676 {
2677 struct r600_command_buffer *cb = &shader->command_buffer;
2678 struct r600_shader *rshader = &shader->shader;
2679
2680 r600_init_command_buffer(cb, 32);
2681
2682 r600_store_context_reg(cb, R_028890_SQ_PGM_RESOURCES_ES,
2683 S_028890_NUM_GPRS(rshader->bc.ngpr) |
2684 S_028890_STACK_SIZE(rshader->bc.nstack));
2685 r600_store_context_reg(cb, R_028880_SQ_PGM_START_ES, 0);
2686 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2687 }
2688
2689
r600_create_resolve_blend(struct r600_context * rctx)2690 void *r600_create_resolve_blend(struct r600_context *rctx)
2691 {
2692 struct pipe_blend_state blend;
2693 unsigned i;
2694
2695 memset(&blend, 0, sizeof(blend));
2696 blend.independent_blend_enable = true;
2697 for (i = 0; i < 2; i++) {
2698 blend.rt[i].colormask = 0xf;
2699 blend.rt[i].blend_enable = 1;
2700 blend.rt[i].rgb_func = PIPE_BLEND_ADD;
2701 blend.rt[i].alpha_func = PIPE_BLEND_ADD;
2702 blend.rt[i].rgb_src_factor = PIPE_BLENDFACTOR_ZERO;
2703 blend.rt[i].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
2704 blend.rt[i].alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
2705 blend.rt[i].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
2706 }
2707 return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_RESOLVE_BOX);
2708 }
2709
r700_create_resolve_blend(struct r600_context * rctx)2710 void *r700_create_resolve_blend(struct r600_context *rctx)
2711 {
2712 struct pipe_blend_state blend;
2713
2714 memset(&blend, 0, sizeof(blend));
2715 blend.independent_blend_enable = true;
2716 blend.rt[0].colormask = 0xf;
2717 return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_RESOLVE_BOX);
2718 }
2719
r600_create_decompress_blend(struct r600_context * rctx)2720 void *r600_create_decompress_blend(struct r600_context *rctx)
2721 {
2722 struct pipe_blend_state blend;
2723
2724 memset(&blend, 0, sizeof(blend));
2725 blend.independent_blend_enable = true;
2726 blend.rt[0].colormask = 0xf;
2727 return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_EXPAND_SAMPLES);
2728 }
2729
r600_create_db_flush_dsa(struct r600_context * rctx)2730 void *r600_create_db_flush_dsa(struct r600_context *rctx)
2731 {
2732 struct pipe_depth_stencil_alpha_state dsa;
2733 boolean quirk = false;
2734
2735 if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
2736 rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
2737 quirk = true;
2738
2739 memset(&dsa, 0, sizeof(dsa));
2740
2741 if (quirk) {
2742 dsa.depth.enabled = 1;
2743 dsa.depth.func = PIPE_FUNC_LEQUAL;
2744 dsa.stencil[0].enabled = 1;
2745 dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
2746 dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_KEEP;
2747 dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_INCR;
2748 dsa.stencil[0].writemask = 0xff;
2749 }
2750
2751 return rctx->b.b.create_depth_stencil_alpha_state(&rctx->b.b, &dsa);
2752 }
2753
r600_update_db_shader_control(struct r600_context * rctx)2754 void r600_update_db_shader_control(struct r600_context * rctx)
2755 {
2756 bool dual_export;
2757 unsigned db_shader_control;
2758 uint8_t ps_conservative_z;
2759
2760 if (!rctx->ps_shader) {
2761 return;
2762 }
2763
2764 dual_export = rctx->framebuffer.export_16bpc &&
2765 !rctx->ps_shader->current->ps_depth_export;
2766
2767 db_shader_control = rctx->ps_shader->current->db_shader_control |
2768 S_02880C_DUAL_EXPORT_ENABLE(dual_export);
2769
2770 ps_conservative_z = rctx->ps_shader->current->shader.ps_conservative_z;
2771
2772 /* When alpha test is enabled we can't trust the hw to make the proper
2773 * decision on the order in which ztest should be run related to fragment
2774 * shader execution.
2775 *
2776 * If alpha test is enabled perform z test after fragment. RE_Z (early
2777 * z test but no write to the zbuffer) seems to cause lockup on r6xx/r7xx
2778 */
2779 if (rctx->alphatest_state.sx_alpha_test_control) {
2780 db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
2781 } else {
2782 db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
2783 }
2784
2785 if (db_shader_control != rctx->db_misc_state.db_shader_control ||
2786 ps_conservative_z != rctx->db_misc_state.ps_conservative_z) {
2787 rctx->db_misc_state.db_shader_control = db_shader_control;
2788 rctx->db_misc_state.ps_conservative_z = ps_conservative_z;
2789 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
2790 }
2791 }
2792
r600_array_mode(unsigned mode)2793 static inline unsigned r600_array_mode(unsigned mode)
2794 {
2795 switch (mode) {
2796 default:
2797 case RADEON_SURF_MODE_LINEAR_ALIGNED: return V_0280A0_ARRAY_LINEAR_ALIGNED;
2798 break;
2799 case RADEON_SURF_MODE_1D: return V_0280A0_ARRAY_1D_TILED_THIN1;
2800 break;
2801 case RADEON_SURF_MODE_2D: return V_0280A0_ARRAY_2D_TILED_THIN1;
2802 }
2803 }
2804
r600_dma_copy_tile(struct r600_context * rctx,struct pipe_resource * dst,unsigned dst_level,unsigned dst_x,unsigned dst_y,unsigned dst_z,struct pipe_resource * src,unsigned src_level,unsigned src_x,unsigned src_y,unsigned src_z,unsigned copy_height,unsigned pitch,unsigned bpp)2805 static boolean r600_dma_copy_tile(struct r600_context *rctx,
2806 struct pipe_resource *dst,
2807 unsigned dst_level,
2808 unsigned dst_x,
2809 unsigned dst_y,
2810 unsigned dst_z,
2811 struct pipe_resource *src,
2812 unsigned src_level,
2813 unsigned src_x,
2814 unsigned src_y,
2815 unsigned src_z,
2816 unsigned copy_height,
2817 unsigned pitch,
2818 unsigned bpp)
2819 {
2820 struct radeon_winsys_cs *cs = rctx->b.dma.cs;
2821 struct r600_texture *rsrc = (struct r600_texture*)src;
2822 struct r600_texture *rdst = (struct r600_texture*)dst;
2823 unsigned array_mode, lbpp, pitch_tile_max, slice_tile_max, size;
2824 unsigned ncopy, height, cheight, detile, i, x, y, z, src_mode, dst_mode;
2825 uint64_t base, addr;
2826
2827 dst_mode = rdst->surface.level[dst_level].mode;
2828 src_mode = rsrc->surface.level[src_level].mode;
2829 assert(dst_mode != src_mode);
2830
2831 y = 0;
2832 lbpp = util_logbase2(bpp);
2833 pitch_tile_max = ((pitch / bpp) / 8) - 1;
2834
2835 if (dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED) {
2836 /* T2L */
2837 array_mode = r600_array_mode(src_mode);
2838 slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) / (8*8);
2839 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
2840 /* linear height must be the same as the slice tile max height, it's ok even
2841 * if the linear destination/source have smaller heigh as the size of the
2842 * dma packet will be using the copy_height which is always smaller or equal
2843 * to the linear height
2844 */
2845 height = u_minify(rsrc->resource.b.b.height0, src_level);
2846 detile = 1;
2847 x = src_x;
2848 y = src_y;
2849 z = src_z;
2850 base = rsrc->surface.level[src_level].offset;
2851 addr = rdst->surface.level[dst_level].offset;
2852 addr += rdst->surface.level[dst_level].slice_size * dst_z;
2853 addr += dst_y * pitch + dst_x * bpp;
2854 } else {
2855 /* L2T */
2856 array_mode = r600_array_mode(dst_mode);
2857 slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) / (8*8);
2858 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
2859 /* linear height must be the same as the slice tile max height, it's ok even
2860 * if the linear destination/source have smaller heigh as the size of the
2861 * dma packet will be using the copy_height which is always smaller or equal
2862 * to the linear height
2863 */
2864 height = u_minify(rdst->resource.b.b.height0, dst_level);
2865 detile = 0;
2866 x = dst_x;
2867 y = dst_y;
2868 z = dst_z;
2869 base = rdst->surface.level[dst_level].offset;
2870 addr = rsrc->surface.level[src_level].offset;
2871 addr += rsrc->surface.level[src_level].slice_size * src_z;
2872 addr += src_y * pitch + src_x * bpp;
2873 }
2874 /* check that we are in dw/base alignment constraint */
2875 if (addr % 4 || base % 256) {
2876 return FALSE;
2877 }
2878
2879 /* It's a r6xx/r7xx limitation, the blit must be on 8 boundary for number
2880 * line in the blit. Compute max 8 line we can copy in the size limit
2881 */
2882 cheight = ((R600_DMA_COPY_MAX_SIZE_DW * 4) / pitch) & 0xfffffff8;
2883 ncopy = (copy_height / cheight) + !!(copy_height % cheight);
2884 r600_need_dma_space(&rctx->b, ncopy * 7, &rdst->resource, &rsrc->resource);
2885
2886 for (i = 0; i < ncopy; i++) {
2887 cheight = cheight > copy_height ? copy_height : cheight;
2888 size = (cheight * pitch) / 4;
2889 /* emit reloc before writing cs so that cs is always in consistent state */
2890 radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rsrc->resource, RADEON_USAGE_READ,
2891 RADEON_PRIO_SDMA_TEXTURE);
2892 radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rdst->resource, RADEON_USAGE_WRITE,
2893 RADEON_PRIO_SDMA_TEXTURE);
2894 radeon_emit(cs, DMA_PACKET(DMA_PACKET_COPY, 1, 0, size));
2895 radeon_emit(cs, base >> 8);
2896 radeon_emit(cs, (detile << 31) | (array_mode << 27) |
2897 (lbpp << 24) | ((height - 1) << 10) |
2898 pitch_tile_max);
2899 radeon_emit(cs, (slice_tile_max << 12) | (z << 0));
2900 radeon_emit(cs, (x << 3) | (y << 17));
2901 radeon_emit(cs, addr & 0xfffffffc);
2902 radeon_emit(cs, (addr >> 32UL) & 0xff);
2903 copy_height -= cheight;
2904 addr += cheight * pitch;
2905 y += cheight;
2906 }
2907 return TRUE;
2908 }
2909
r600_dma_copy(struct pipe_context * ctx,struct pipe_resource * dst,unsigned dst_level,unsigned dstx,unsigned dsty,unsigned dstz,struct pipe_resource * src,unsigned src_level,const struct pipe_box * src_box)2910 static void r600_dma_copy(struct pipe_context *ctx,
2911 struct pipe_resource *dst,
2912 unsigned dst_level,
2913 unsigned dstx, unsigned dsty, unsigned dstz,
2914 struct pipe_resource *src,
2915 unsigned src_level,
2916 const struct pipe_box *src_box)
2917 {
2918 struct r600_context *rctx = (struct r600_context *)ctx;
2919 struct r600_texture *rsrc = (struct r600_texture*)src;
2920 struct r600_texture *rdst = (struct r600_texture*)dst;
2921 unsigned dst_pitch, src_pitch, bpp, dst_mode, src_mode, copy_height;
2922 unsigned src_w, dst_w;
2923 unsigned src_x, src_y;
2924 unsigned dst_x = dstx, dst_y = dsty, dst_z = dstz;
2925
2926 if (rctx->b.dma.cs == NULL) {
2927 goto fallback;
2928 }
2929
2930 if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
2931 if (dst_x % 4 || src_box->x % 4 || src_box->width % 4)
2932 goto fallback;
2933
2934 r600_dma_copy_buffer(rctx, dst, src, dst_x, src_box->x, src_box->width);
2935 return;
2936 }
2937
2938 if (src_box->depth > 1 ||
2939 !r600_prepare_for_dma_blit(&rctx->b, rdst, dst_level, dstx, dsty,
2940 dstz, rsrc, src_level, src_box))
2941 goto fallback;
2942
2943 src_x = util_format_get_nblocksx(src->format, src_box->x);
2944 dst_x = util_format_get_nblocksx(src->format, dst_x);
2945 src_y = util_format_get_nblocksy(src->format, src_box->y);
2946 dst_y = util_format_get_nblocksy(src->format, dst_y);
2947
2948 bpp = rdst->surface.bpe;
2949 dst_pitch = rdst->surface.level[dst_level].nblk_x * rdst->surface.bpe;
2950 src_pitch = rsrc->surface.level[src_level].nblk_x * rsrc->surface.bpe;
2951 src_w = u_minify(rsrc->resource.b.b.width0, src_level);
2952 dst_w = u_minify(rdst->resource.b.b.width0, dst_level);
2953 copy_height = src_box->height / rsrc->surface.blk_h;
2954
2955 dst_mode = rdst->surface.level[dst_level].mode;
2956 src_mode = rsrc->surface.level[src_level].mode;
2957
2958 if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) {
2959 /* strict requirement on r6xx/r7xx */
2960 goto fallback;
2961 }
2962 /* lot of constraint on alignment this should capture them all */
2963 if (src_pitch % 8 || src_box->y % 8 || dst_y % 8) {
2964 goto fallback;
2965 }
2966
2967 if (src_mode == dst_mode) {
2968 uint64_t dst_offset, src_offset, size;
2969
2970 /* simple dma blit would do NOTE code here assume :
2971 * src_box.x/y == 0
2972 * dst_x/y == 0
2973 * dst_pitch == src_pitch
2974 */
2975 src_offset= rsrc->surface.level[src_level].offset;
2976 src_offset += rsrc->surface.level[src_level].slice_size * src_box->z;
2977 src_offset += src_y * src_pitch + src_x * bpp;
2978 dst_offset = rdst->surface.level[dst_level].offset;
2979 dst_offset += rdst->surface.level[dst_level].slice_size * dst_z;
2980 dst_offset += dst_y * dst_pitch + dst_x * bpp;
2981 size = src_box->height * src_pitch;
2982 /* must be dw aligned */
2983 if (dst_offset % 4 || src_offset % 4 || size % 4) {
2984 goto fallback;
2985 }
2986 r600_dma_copy_buffer(rctx, dst, src, dst_offset, src_offset, size);
2987 } else {
2988 if (!r600_dma_copy_tile(rctx, dst, dst_level, dst_x, dst_y, dst_z,
2989 src, src_level, src_x, src_y, src_box->z,
2990 copy_height, dst_pitch, bpp)) {
2991 goto fallback;
2992 }
2993 }
2994 return;
2995
2996 fallback:
2997 r600_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
2998 src, src_level, src_box);
2999 }
3000
r600_init_state_functions(struct r600_context * rctx)3001 void r600_init_state_functions(struct r600_context *rctx)
3002 {
3003 unsigned id = 1;
3004 unsigned i;
3005 /* !!!
3006 * To avoid GPU lockup registers must be emited in a specific order
3007 * (no kidding ...). The order below is important and have been
3008 * partialy infered from analyzing fglrx command stream.
3009 *
3010 * Don't reorder atom without carefully checking the effect (GPU lockup
3011 * or piglit regression).
3012 * !!!
3013 */
3014
3015 r600_init_atom(rctx, &rctx->framebuffer.atom, id++, r600_emit_framebuffer_state, 0);
3016
3017 /* shader const */
3018 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX].atom, id++, r600_emit_vs_constant_buffers, 0);
3019 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY].atom, id++, r600_emit_gs_constant_buffers, 0);
3020 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT].atom, id++, r600_emit_ps_constant_buffers, 0);
3021
3022 /* sampler must be emited before TA_CNTL_AUX otherwise DISABLE_CUBE_WRAP change
3023 * does not take effect (TA_CNTL_AUX emited by r600_emit_seamless_cube_map)
3024 */
3025 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].states.atom, id++, r600_emit_vs_sampler_states, 0);
3026 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].states.atom, id++, r600_emit_gs_sampler_states, 0);
3027 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].states.atom, id++, r600_emit_ps_sampler_states, 0);
3028 /* resource */
3029 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views.atom, id++, r600_emit_vs_sampler_views, 0);
3030 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views.atom, id++, r600_emit_gs_sampler_views, 0);
3031 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views.atom, id++, r600_emit_ps_sampler_views, 0);
3032 r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, r600_emit_vertex_buffers, 0);
3033
3034 r600_init_atom(rctx, &rctx->vgt_state.atom, id++, r600_emit_vgt_state, 10);
3035
3036 r600_init_atom(rctx, &rctx->seamless_cube_map.atom, id++, r600_emit_seamless_cube_map, 3);
3037 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, r600_emit_sample_mask, 3);
3038 rctx->sample_mask.sample_mask = ~0;
3039
3040 r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
3041 r600_init_atom(rctx, &rctx->blend_color.atom, id++, r600_emit_blend_color, 6);
3042 r600_init_atom(rctx, &rctx->blend_state.atom, id++, r600_emit_cso_state, 0);
3043 r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, r600_emit_cb_misc_state, 7);
3044 r600_init_atom(rctx, &rctx->clip_misc_state.atom, id++, r600_emit_clip_misc_state, 6);
3045 r600_init_atom(rctx, &rctx->clip_state.atom, id++, r600_emit_clip_state, 26);
3046 r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, r600_emit_db_misc_state, 7);
3047 r600_init_atom(rctx, &rctx->db_state.atom, id++, r600_emit_db_state, 11);
3048 r600_init_atom(rctx, &rctx->dsa_state.atom, id++, r600_emit_cso_state, 0);
3049 r600_init_atom(rctx, &rctx->poly_offset_state.atom, id++, r600_emit_polygon_offset, 9);
3050 r600_init_atom(rctx, &rctx->rasterizer_state.atom, id++, r600_emit_cso_state, 0);
3051 r600_add_atom(rctx, &rctx->b.scissors.atom, id++);
3052 r600_add_atom(rctx, &rctx->b.viewports.atom, id++);
3053 r600_init_atom(rctx, &rctx->config_state.atom, id++, r600_emit_config_state, 3);
3054 r600_init_atom(rctx, &rctx->stencil_ref.atom, id++, r600_emit_stencil_ref, 4);
3055 r600_init_atom(rctx, &rctx->vertex_fetch_shader.atom, id++, r600_emit_vertex_fetch_shader, 5);
3056 r600_add_atom(rctx, &rctx->b.render_cond_atom, id++);
3057 r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++);
3058 r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++);
3059 for (i = 0; i < R600_NUM_HW_STAGES; i++)
3060 r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0);
3061 r600_init_atom(rctx, &rctx->shader_stages.atom, id++, r600_emit_shader_stages, 0);
3062 r600_init_atom(rctx, &rctx->gs_rings.atom, id++, r600_emit_gs_rings, 0);
3063
3064 rctx->b.b.create_blend_state = r600_create_blend_state;
3065 rctx->b.b.create_depth_stencil_alpha_state = r600_create_dsa_state;
3066 rctx->b.b.create_rasterizer_state = r600_create_rs_state;
3067 rctx->b.b.create_sampler_state = r600_create_sampler_state;
3068 rctx->b.b.create_sampler_view = r600_create_sampler_view;
3069 rctx->b.b.set_framebuffer_state = r600_set_framebuffer_state;
3070 rctx->b.b.set_polygon_stipple = r600_set_polygon_stipple;
3071 rctx->b.b.set_min_samples = r600_set_min_samples;
3072 rctx->b.b.get_sample_position = r600_get_sample_position;
3073 rctx->b.dma_copy = r600_dma_copy;
3074 }
3075 /* this function must be last */
3076