Merge remote branch 'origin/7.8'
[mesa.git] / src / gallium / drivers / r300 / r300_state_derived.c
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24 #include "draw/draw_context.h"
25
26 #include "util/u_math.h"
27 #include "util/u_memory.h"
28
29 #include "r300_context.h"
30 #include "r300_fs.h"
31 #include "r300_screen.h"
32 #include "r300_shader_semantics.h"
33 #include "r300_state_derived.h"
34 #include "r300_state_inlines.h"
35 #include "r300_vs.h"
36
37 /* r300_state_derived: Various bits of state which are dependent upon
38 * currently bound CSO data. */
39
40 static void r300_draw_emit_attrib(struct r300_context* r300,
41 enum attrib_emit emit,
42 enum interp_mode interp,
43 int index)
44 {
45 struct r300_vertex_shader* vs = r300->vs_state.state;
46 struct tgsi_shader_info* info = &vs->info;
47 int output;
48
49 output = draw_find_shader_output(r300->draw,
50 info->output_semantic_name[index],
51 info->output_semantic_index[index]);
52 draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output);
53 }
54
55 static void r300_draw_emit_all_attribs(struct r300_context* r300)
56 {
57 struct r300_vertex_shader* vs = r300->vs_state.state;
58 struct r300_shader_semantics* vs_outputs = &vs->outputs;
59 int i, gen_count;
60
61 /* Position. */
62 if (vs_outputs->pos != ATTR_UNUSED) {
63 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
64 vs_outputs->pos);
65 } else {
66 assert(0);
67 }
68
69 /* Point size. */
70 if (vs_outputs->psize != ATTR_UNUSED) {
71 r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS,
72 vs_outputs->psize);
73 }
74
75 /* Colors. */
76 for (i = 0; i < ATTR_COLOR_COUNT; i++) {
77 if (vs_outputs->color[i] != ATTR_UNUSED) {
78 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
79 vs_outputs->color[i]);
80 }
81 }
82
83 /* XXX Back-face colors. */
84
85 /* Texture coordinates. */
86 gen_count = 0;
87 for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
88 if (vs_outputs->generic[i] != ATTR_UNUSED) {
89 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
90 vs_outputs->generic[i]);
91 gen_count++;
92 }
93 }
94
95 /* Fog coordinates. */
96 if (vs_outputs->fog != ATTR_UNUSED) {
97 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
98 vs_outputs->fog);
99 gen_count++;
100 }
101
102 assert(gen_count <= 8);
103 }
104
105 /* Update the PSC tables for SW TCL, using Draw. */
106 static void r300_swtcl_vertex_psc(struct r300_context *r300)
107 {
108 struct r300_vertex_stream_state *vstream = r300->vertex_stream_state.state;
109 struct r300_vertex_shader* vs = r300->vs_state.state;
110 struct vertex_info* vinfo = &r300->vertex_info;
111 uint16_t type, swizzle;
112 enum pipe_format format;
113 unsigned i, attrib_count;
114 int* vs_output_tab = vs->stream_loc_notcl;
115
116 /* XXX hax */
117 memset(vstream, 0, sizeof(struct r300_vertex_stream_state));
118
119 /* For each Draw attribute, route it to the fragment shader according
120 * to the vs_output_tab. */
121 attrib_count = vinfo->num_attribs;
122 DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
123 for (i = 0; i < attrib_count; i++) {
124 DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
125 " vs_output_tab %d\n", vinfo->attrib[i].src_index,
126 vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
127 vs_output_tab[i]);
128
129 /* Make sure we have a proper destination for our attribute. */
130 assert(vs_output_tab[i] != -1);
131
132 format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
133
134 /* Obtain the type of data in this attribute. */
135 type = r300_translate_vertex_data_type(format) |
136 vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
137
138 /* Obtain the swizzle for this attribute. Note that the default
139 * swizzle in the hardware is not XYZW! */
140 swizzle = r300_translate_vertex_data_swizzle(format);
141
142 /* Add the attribute to the PSC table. */
143 if (i & 1) {
144 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
145 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
146 } else {
147 vstream->vap_prog_stream_cntl[i >> 1] |= type;
148 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
149 }
150 }
151
152 /* Set the last vector in the PSC. */
153 if (i) {
154 i -= 1;
155 }
156 vstream->vap_prog_stream_cntl[i >> 1] |=
157 (R300_LAST_VEC << (i & 1 ? 16 : 0));
158
159 vstream->count = (i >> 1) + 1;
160 r300->vertex_stream_state.dirty = TRUE;
161 r300->vertex_stream_state.size = (1 + vstream->count) * 2;
162 }
163
164 static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr,
165 boolean swizzle_0001)
166 {
167 rs->ip[id] |= R300_RS_COL_PTR(ptr);
168 if (swizzle_0001) {
169 rs->ip[id] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
170 } else {
171 rs->ip[id] |= R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
172 }
173 rs->inst[id] |= R300_RS_INST_COL_ID(id);
174 }
175
176 static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
177 {
178 rs->inst[id] |= R300_RS_INST_COL_CN_WRITE |
179 R300_RS_INST_COL_ADDR(fp_offset);
180 }
181
182 static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr,
183 boolean swizzle_X001)
184 {
185 if (swizzle_X001) {
186 rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
187 R300_RS_SEL_S(R300_RS_SEL_C0) |
188 R300_RS_SEL_T(R300_RS_SEL_K0) |
189 R300_RS_SEL_R(R300_RS_SEL_K0) |
190 R300_RS_SEL_Q(R300_RS_SEL_K1);
191 } else {
192 rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
193 R300_RS_SEL_S(R300_RS_SEL_C0) |
194 R300_RS_SEL_T(R300_RS_SEL_C1) |
195 R300_RS_SEL_R(R300_RS_SEL_C2) |
196 R300_RS_SEL_Q(R300_RS_SEL_C3);
197 }
198 rs->inst[id] |= R300_RS_INST_TEX_ID(id);
199 }
200
201 static void r300_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset)
202 {
203 rs->inst[id] |= R300_RS_INST_TEX_CN_WRITE |
204 R300_RS_INST_TEX_ADDR(fp_offset);
205 }
206
207 static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr,
208 boolean swizzle_0001)
209 {
210 rs->ip[id] |= R500_RS_COL_PTR(ptr);
211 if (swizzle_0001) {
212 rs->ip[id] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
213 } else {
214 rs->ip[id] |= R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
215 }
216 rs->inst[id] |= R500_RS_INST_COL_ID(id);
217 }
218
219 static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
220 {
221 rs->inst[id] |= R500_RS_INST_COL_CN_WRITE |
222 R500_RS_INST_COL_ADDR(fp_offset);
223 }
224
225 static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr,
226 boolean swizzle_X001)
227 {
228 int rs_tex_comp = ptr*4;
229
230 if (swizzle_X001) {
231 rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
232 R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
233 R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
234 R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
235 } else {
236 rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
237 R500_RS_SEL_T(rs_tex_comp + 1) |
238 R500_RS_SEL_R(rs_tex_comp + 2) |
239 R500_RS_SEL_Q(rs_tex_comp + 3);
240 }
241 rs->inst[id] |= R500_RS_INST_TEX_ID(id);
242 }
243
244 static void r500_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset)
245 {
246 rs->inst[id] |= R500_RS_INST_TEX_CN_WRITE |
247 R500_RS_INST_TEX_ADDR(fp_offset);
248 }
249
250 /* Set up the RS block.
251 *
252 * This is the part of the chipset that actually does the rasterization
253 * of vertices into fragments. This is also the part of the chipset that
254 * locks up if any part of it is even slightly wrong. */
255 static void r300_update_rs_block(struct r300_context* r300,
256 struct r300_shader_semantics* vs_outputs,
257 struct r300_shader_semantics* fs_inputs)
258 {
259 struct r300_rs_block rs = { { 0 } };
260 int i, col_count = 0, tex_count = 0, fp_offset = 0, count;
261 void (*rX00_rs_col)(struct r300_rs_block*, int, int, boolean);
262 void (*rX00_rs_col_write)(struct r300_rs_block*, int, int);
263 void (*rX00_rs_tex)(struct r300_rs_block*, int, int, boolean);
264 void (*rX00_rs_tex_write)(struct r300_rs_block*, int, int);
265 boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
266 vs_outputs->bcolor[1] != ATTR_UNUSED;
267
268 if (r300->screen->caps.is_r500) {
269 rX00_rs_col = r500_rs_col;
270 rX00_rs_col_write = r500_rs_col_write;
271 rX00_rs_tex = r500_rs_tex;
272 rX00_rs_tex_write = r500_rs_tex_write;
273 } else {
274 rX00_rs_col = r300_rs_col;
275 rX00_rs_col_write = r300_rs_col_write;
276 rX00_rs_tex = r300_rs_tex;
277 rX00_rs_tex_write = r300_rs_tex_write;
278 }
279
280 /* Rasterize colors. */
281 for (i = 0; i < ATTR_COLOR_COUNT; i++) {
282 if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
283 vs_outputs->color[1] != ATTR_UNUSED) {
284 /* Always rasterize if it's written by the VS,
285 * otherwise it locks up. */
286 rX00_rs_col(&rs, col_count, i, FALSE);
287
288 /* Write it to the FS input register if it's used by the FS. */
289 if (fs_inputs->color[i] != ATTR_UNUSED) {
290 rX00_rs_col_write(&rs, col_count, fp_offset);
291 fp_offset++;
292 }
293 col_count++;
294 } else {
295 /* Skip the FS input register, leave it uninitialized. */
296 /* If we try to set it to (0,0,0,1), it will lock up. */
297 if (fs_inputs->color[i] != ATTR_UNUSED) {
298 fp_offset++;
299 }
300 }
301 }
302
303 /* Rasterize texture coordinates. */
304 for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
305 if (vs_outputs->generic[i] != ATTR_UNUSED) {
306 /* Always rasterize if it's written by the VS,
307 * otherwise it locks up. */
308 rX00_rs_tex(&rs, tex_count, tex_count, FALSE);
309
310 /* Write it to the FS input register if it's used by the FS. */
311 if (fs_inputs->generic[i] != ATTR_UNUSED) {
312 rX00_rs_tex_write(&rs, tex_count, fp_offset);
313 fp_offset++;
314 }
315 tex_count++;
316 } else {
317 /* Skip the FS input register, leave it uninitialized. */
318 /* If we try to set it to (0,0,0,1), it will lock up. */
319 if (fs_inputs->generic[i] != ATTR_UNUSED) {
320 fp_offset++;
321 }
322 }
323 }
324
325 /* Rasterize fog coordinates. */
326 if (vs_outputs->fog != ATTR_UNUSED) {
327 /* Always rasterize if it's written by the VS,
328 * otherwise it locks up. */
329 rX00_rs_tex(&rs, tex_count, tex_count, TRUE);
330
331 /* Write it to the FS input register if it's used by the FS. */
332 if (fs_inputs->fog != ATTR_UNUSED) {
333 rX00_rs_tex_write(&rs, tex_count, fp_offset);
334 fp_offset++;
335 }
336 tex_count++;
337 } else {
338 /* Skip the FS input register, leave it uninitialized. */
339 /* If we try to set it to (0,0,0,1), it will lock up. */
340 if (fs_inputs->fog != ATTR_UNUSED) {
341 fp_offset++;
342 }
343 }
344
345 /* Rasterize WPOS. */
346 /* If the FS doesn't need it, it's not written by the VS. */
347 if (fs_inputs->wpos != ATTR_UNUSED) {
348 rX00_rs_tex(&rs, tex_count, tex_count, FALSE);
349 rX00_rs_tex_write(&rs, tex_count, fp_offset);
350
351 fp_offset++;
352 tex_count++;
353 }
354
355 /* Rasterize at least one color, or bad things happen. */
356 if (col_count == 0 && tex_count == 0) {
357 rX00_rs_col(&rs, 0, 0, TRUE);
358 col_count++;
359 }
360
361 rs.count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
362 R300_HIRES_EN;
363
364 count = MAX3(col_count, tex_count, 1);
365 rs.inst_count = count - 1;
366
367 /* Now, after all that, see if we actually need to update the state. */
368 if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) {
369 memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block));
370 r300->rs_block_state.size = 5 + count*2;
371 }
372 }
373
374 /* Update the shader-dependant states. */
375 static void r300_update_derived_shader_state(struct r300_context* r300)
376 {
377 struct r300_vertex_shader* vs = r300->vs_state.state;
378
379 r300_update_rs_block(r300, &vs->outputs, &r300->fs->inputs);
380 }
381
382 static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa)
383 {
384 /* We are interested only in the cases when a new depth or stencil value
385 * can be written and changed. */
386
387 /* We might optionally check for [Z func: never] and inspect the stencil
388 * state in a similar fashion, but it's not terribly important. */
389 return (dsa->z_buffer_control & R300_Z_WRITE_ENABLE) ||
390 (dsa->stencil_ref_mask & R300_STENCILWRITEMASK_MASK) ||
391 ((dsa->z_buffer_control & R500_STENCIL_REFMASK_FRONT_BACK) &&
392 (dsa->stencil_ref_bf & R300_STENCILWRITEMASK_MASK));
393 }
394
395 static boolean r300_dsa_alpha_test_enabled(struct r300_dsa_state* dsa)
396 {
397 /* We are interested only in the cases when alpha testing can kill
398 * a fragment. */
399 uint32_t af = dsa->alpha_function;
400
401 return (af & R300_FG_ALPHA_FUNC_ENABLE) &&
402 (af & R300_FG_ALPHA_FUNC_ALWAYS) != R300_FG_ALPHA_FUNC_ALWAYS;
403 }
404
405 static void r300_update_ztop(struct r300_context* r300)
406 {
407 struct r300_ztop_state* ztop_state =
408 (struct r300_ztop_state*)r300->ztop_state.state;
409
410 /* This is important enough that I felt it warranted a comment.
411 *
412 * According to the docs, these are the conditions where ZTOP must be
413 * disabled:
414 * 1) Alpha testing enabled
415 * 2) Texture kill instructions in fragment shader
416 * 3) Chroma key culling enabled
417 * 4) W-buffering enabled
418 *
419 * The docs claim that for the first three cases, if no ZS writes happen,
420 * then ZTOP can be used.
421 *
422 * (3) will never apply since we do not support chroma-keyed operations.
423 * (4) will need to be re-examined (and this comment updated) if/when
424 * Hyper-Z becomes supported.
425 *
426 * Additionally, the following conditions require disabled ZTOP:
427 * 5) Depth writes in fragment shader
428 * 6) Outstanding occlusion queries
429 *
430 * This register causes stalls all the way from SC to CB when changed,
431 * but it is buffered on-chip so it does not hurt to write it if it has
432 * not changed.
433 *
434 * ~C.
435 */
436
437 /* ZS writes */
438 if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) &&
439 (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||/* (1) */
440 r300->fs->info.uses_kill)) { /* (2) */
441 ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
442 } else if (r300_fragment_shader_writes_depth(r300->fs)) { /* (5) */
443 ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
444 } else if (r300->query_current) { /* (6) */
445 ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
446 } else {
447 ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
448 }
449
450 r300->ztop_state.dirty = TRUE;
451 }
452
453 static void r300_merge_textures_and_samplers(struct r300_context* r300)
454 {
455 struct r300_textures_state *state =
456 (struct r300_textures_state*)r300->textures_state.state;
457 struct r300_texture_sampler_state *texstate;
458 struct r300_sampler_state *sampler;
459 struct pipe_sampler_view *view;
460 struct r300_texture *tex;
461 unsigned min_level, max_level, i, size;
462 unsigned count = MIN2(state->texture_count, state->sampler_count);
463
464 state->tx_enable = 0;
465 state->count = 0;
466 size = 2;
467
468 for (i = 0; i < count; i++) {
469 if (state->fragment_sampler_views[i] && state->sampler_states[i]) {
470 state->tx_enable |= 1 << i;
471
472 view = state->fragment_sampler_views[i];
473 tex = r300_texture(view->texture);
474 sampler = state->sampler_states[i];
475
476 assert(view->format == tex->tex.format);
477
478 texstate = &state->regs[i];
479 memcpy(texstate->format, &tex->state, sizeof(uint32_t)*3);
480 texstate->filter[0] = sampler->filter0;
481 texstate->filter[1] = sampler->filter1;
482 texstate->border_color = sampler->border_color;
483 texstate->tile_config = R300_TXO_MACRO_TILE(tex->macrotile) |
484 R300_TXO_MICRO_TILE(tex->microtile);
485
486 /* to emulate 1D textures through 2D ones correctly */
487 if (tex->tex.target == PIPE_TEXTURE_1D) {
488 texstate->filter[0] &= ~R300_TX_WRAP_T_MASK;
489 texstate->filter[0] |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
490 }
491
492 if (tex->uses_pitch) {
493 /* NPOT textures don't support mip filter, unfortunately.
494 * This prevents incorrect rendering. */
495 texstate->filter[0] &= ~R300_TX_MIN_FILTER_MIP_MASK;
496 } else {
497 /* determine min/max levels */
498 /* the MAX_MIP level is the largest (finest) one */
499 max_level = MIN3(sampler->max_lod + view->first_level,
500 tex->tex.last_level, view->last_level);
501 min_level = MIN2(sampler->min_lod + view->first_level,
502 max_level);
503 texstate->format[0] |= R300_TX_NUM_LEVELS(max_level);
504 texstate->filter[0] |= R300_TX_MAX_MIP_LEVEL(min_level);
505 }
506
507 texstate->filter[0] |= i << 28;
508
509 size += 16;
510 state->count = i+1;
511 }
512 }
513
514 r300->textures_state.size = size;
515 }
516
517 void r300_update_derived_state(struct r300_context* r300)
518 {
519 if (r300->rs_block_state.dirty) {
520 r300_update_derived_shader_state(r300);
521 }
522
523 if (r300->textures_state.dirty) {
524 r300_merge_textures_and_samplers(r300);
525 }
526
527 if (r300->draw) {
528 memset(&r300->vertex_info, 0, sizeof(struct vertex_info));
529 r300_draw_emit_all_attribs(r300);
530 draw_compute_vertex_size(&r300->vertex_info);
531 r300_swtcl_vertex_psc(r300);
532 }
533
534 r300_update_ztop(r300);
535 }