projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r600g: fix num of dwords needed for alphatest_state atom
[mesa.git] / src / gallium / drivers / r600 / r600_state_common.c
1 /*
2 * Copyright 2010 Red Hat Inc.
3 * 2010 Jerome Glisse
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 * Authors: Dave Airlie <airlied@redhat.com>
25 * Jerome Glisse <jglisse@redhat.com>
26 */
27 #include "r600_formats.h"
28 #include "r600d.h"
29
30 #include "util/u_draw_quad.h"
31 #include "util/u_upload_mgr.h"
32 #include "tgsi/tgsi_parse.h"
33 #include <byteswap.h>
34
35 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
36
37 static void r600_emit_command_buffer(struct r600_context *rctx, struct r600_atom *atom)
38 {
39 struct radeon_winsys_cs *cs = rctx->cs;
40 struct r600_command_buffer *cb = (struct r600_command_buffer*)atom;
41
42 assert(cs->cdw + cb->atom.num_dw <= RADEON_MAX_CMDBUF_DWORDS);
43 memcpy(cs->buf + cs->cdw, cb->buf, 4 * cb->atom.num_dw);
44 cs->cdw += cb->atom.num_dw;
45 }
46
47 void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags)
48 {
49 cb->atom.emit = r600_emit_command_buffer;
50 cb->atom.num_dw = 0;
51 cb->atom.flags = flags;
52 cb->buf = CALLOC(1, 4 * num_dw);
53 cb->max_num_dw = num_dw;
54 }
55
56 void r600_release_command_buffer(struct r600_command_buffer *cb)
57 {
58 FREE(cb->buf);
59 }
60
61 static void r600_emit_surface_sync(struct r600_context *rctx, struct r600_atom *atom)
62 {
63 struct radeon_winsys_cs *cs = rctx->cs;
64 struct r600_surface_sync_cmd *a = (struct r600_surface_sync_cmd*)atom;
65
66 cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_SYNC, 3, 0);
67 cs->buf[cs->cdw++] = a->flush_flags; /* CP_COHER_CNTL */
68 cs->buf[cs->cdw++] = 0xffffffff; /* CP_COHER_SIZE */
69 cs->buf[cs->cdw++] = 0; /* CP_COHER_BASE */
70 cs->buf[cs->cdw++] = 0x0000000A; /* POLL_INTERVAL */
71
72 a->flush_flags = 0;
73 }
74
75 static void r600_emit_r6xx_flush_and_inv(struct r600_context *rctx, struct r600_atom *atom)
76 {
77 struct radeon_winsys_cs *cs = rctx->cs;
78 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
79 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
80 }
81
82 void r600_init_atom(struct r600_atom *atom,
83 void (*emit)(struct r600_context *ctx, struct r600_atom *state),
84 unsigned num_dw, enum r600_atom_flags flags)
85 {
86 atom->emit = emit;
87 atom->num_dw = num_dw;
88 atom->flags = flags;
89 }
90
91 static void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom)
92 {
93 struct radeon_winsys_cs *cs = rctx->cs;
94 struct r600_alphatest_state *a = (struct r600_alphatest_state*)atom;
95 unsigned alpha_ref = a->sx_alpha_ref;
96
97 if (rctx->chip_class >= EVERGREEN && a->cb0_export_16bpc) {
98 alpha_ref &= ~0x1FFF;
99 }
100
101 r600_write_context_reg(cs, R_028410_SX_ALPHA_TEST_CONTROL,
102 a->sx_alpha_test_control |
103 S_028410_ALPHA_TEST_BYPASS(a->bypass));
104 r600_write_context_reg(cs, R_028438_SX_ALPHA_REF, alpha_ref);
105 }
106
107 void r600_init_common_atoms(struct r600_context *rctx)
108 {
109 r600_init_atom(&rctx->surface_sync_cmd.atom, r600_emit_surface_sync, 5, EMIT_EARLY);
110 r600_init_atom(&rctx->r6xx_flush_and_inv_cmd, r600_emit_r6xx_flush_and_inv, 2, EMIT_EARLY);
111 r600_init_atom(&rctx->alphatest_state.atom, r600_emit_alphatest_state, 6, 0);
112 r600_atom_dirty(rctx, &rctx->alphatest_state.atom);
113 }
114
115 unsigned r600_get_cb_flush_flags(struct r600_context *rctx)
116 {
117 unsigned flags = 0;
118
119 if (rctx->framebuffer.nr_cbufs) {
120 flags |= S_0085F0_CB_ACTION_ENA(1) |
121 (((1 << rctx->framebuffer.nr_cbufs) - 1) << S_0085F0_CB0_DEST_BASE_ENA_SHIFT);
122 }
123
124 /* Workaround for broken flushing on some R6xx chipsets. */
125 if (rctx->family == CHIP_RV670 ||
126 rctx->family == CHIP_RS780 ||
127 rctx->family == CHIP_RS880) {
128 flags |= S_0085F0_CB1_DEST_BASE_ENA(1) |
129 S_0085F0_DEST_BASE_0_ENA(1);
130 }
131 return flags;
132 }
133
134 void r600_texture_barrier(struct pipe_context *ctx)
135 {
136 struct r600_context *rctx = (struct r600_context *)ctx;
137
138 rctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1) | r600_get_cb_flush_flags(rctx);
139 r600_atom_dirty(rctx, &rctx->surface_sync_cmd.atom);
140 }
141
142 static bool r600_conv_pipe_prim(unsigned pprim, unsigned *prim)
143 {
144 static const int prim_conv[] = {
145 V_008958_DI_PT_POINTLIST,
146 V_008958_DI_PT_LINELIST,
147 V_008958_DI_PT_LINELOOP,
148 V_008958_DI_PT_LINESTRIP,
149 V_008958_DI_PT_TRILIST,
150 V_008958_DI_PT_TRISTRIP,
151 V_008958_DI_PT_TRIFAN,
152 V_008958_DI_PT_QUADLIST,
153 V_008958_DI_PT_QUADSTRIP,
154 V_008958_DI_PT_POLYGON,
155 -1,
156 -1,
157 -1,
158 -1,
159 V_008958_DI_PT_RECTLIST
160 };
161
162 *prim = prim_conv[pprim];
163 if (*prim == -1) {
164 fprintf(stderr, "%s:%d unsupported %d\n", __func__, __LINE__, pprim);
165 return false;
166 }
167 return true;
168 }
169
170 /* common state between evergreen and r600 */
171
172 static void r600_bind_blend_state_internal(struct r600_context *rctx,
173 struct r600_pipe_blend *blend)
174 {
175 struct r600_pipe_state *rstate;
176 bool update_cb = false;
177
178 rstate = &blend->rstate;
179 rctx->states[rstate->id] = rstate;
180 r600_context_pipe_state_set(rctx, rstate);
181
182 if (rctx->cb_misc_state.blend_colormask != blend->cb_target_mask) {
183 rctx->cb_misc_state.blend_colormask = blend->cb_target_mask;
184 update_cb = true;
185 }
186 if (rctx->chip_class <= R700 &&
187 rctx->cb_misc_state.cb_color_control != blend->cb_color_control) {
188 rctx->cb_misc_state.cb_color_control = blend->cb_color_control;
189 update_cb = true;
190 }
191 if (rctx->cb_misc_state.dual_src_blend != blend->dual_src_blend) {
192 rctx->cb_misc_state.dual_src_blend = blend->dual_src_blend;
193 update_cb = true;
194 }
195 if (update_cb) {
196 r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
197 }
198 }
199
200 void r600_bind_blend_state(struct pipe_context *ctx, void *state)
201 {
202 struct r600_context *rctx = (struct r600_context *)ctx;
203 struct r600_pipe_blend *blend = (struct r600_pipe_blend *)state;
204
205 if (blend == NULL)
206 return;
207
208 rctx->blend = blend;
209 rctx->alpha_to_one = blend->alpha_to_one;
210 rctx->dual_src_blend = blend->dual_src_blend;
211
212 if (!rctx->blend_override)
213 r600_bind_blend_state_internal(rctx, blend);
214 }
215
216 void r600_set_blend_color(struct pipe_context *ctx,
217 const struct pipe_blend_color *state)
218 {
219 struct r600_context *rctx = (struct r600_context *)ctx;
220 struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
221
222 if (rstate == NULL)
223 return;
224
225 rstate->id = R600_PIPE_STATE_BLEND_COLOR;
226 r600_pipe_state_add_reg(rstate, R_028414_CB_BLEND_RED, fui(state->color[0]));
227 r600_pipe_state_add_reg(rstate, R_028418_CB_BLEND_GREEN, fui(state->color[1]));
228 r600_pipe_state_add_reg(rstate, R_02841C_CB_BLEND_BLUE, fui(state->color[2]));
229 r600_pipe_state_add_reg(rstate, R_028420_CB_BLEND_ALPHA, fui(state->color[3]));
230
231 free(rctx->states[R600_PIPE_STATE_BLEND_COLOR]);
232 rctx->states[R600_PIPE_STATE_BLEND_COLOR] = rstate;
233 r600_context_pipe_state_set(rctx, rstate);
234 }
235
236 static void r600_set_stencil_ref(struct pipe_context *ctx,
237 const struct r600_stencil_ref *state)
238 {
239 struct r600_context *rctx = (struct r600_context *)ctx;
240 struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
241
242 if (rstate == NULL)
243 return;
244
245 rstate->id = R600_PIPE_STATE_STENCIL_REF;
246 r600_pipe_state_add_reg(rstate,
247 R_028430_DB_STENCILREFMASK,
248 S_028430_STENCILREF(state->ref_value[0]) |
249 S_028430_STENCILMASK(state->valuemask[0]) |
250 S_028430_STENCILWRITEMASK(state->writemask[0]));
251 r600_pipe_state_add_reg(rstate,
252 R_028434_DB_STENCILREFMASK_BF,
253 S_028434_STENCILREF_BF(state->ref_value[1]) |
254 S_028434_STENCILMASK_BF(state->valuemask[1]) |
255 S_028434_STENCILWRITEMASK_BF(state->writemask[1]));
256
257 free(rctx->states[R600_PIPE_STATE_STENCIL_REF]);
258 rctx->states[R600_PIPE_STATE_STENCIL_REF] = rstate;
259 r600_context_pipe_state_set(rctx, rstate);
260 }
261
262 void r600_set_pipe_stencil_ref(struct pipe_context *ctx,
263 const struct pipe_stencil_ref *state)
264 {
265 struct r600_context *rctx = (struct r600_context *)ctx;
266 struct r600_pipe_dsa *dsa = (struct r600_pipe_dsa*)rctx->states[R600_PIPE_STATE_DSA];
267 struct r600_stencil_ref ref;
268
269 rctx->stencil_ref = *state;
270
271 if (!dsa)
272 return;
273
274 ref.ref_value[0] = state->ref_value[0];
275 ref.ref_value[1] = state->ref_value[1];
276 ref.valuemask[0] = dsa->valuemask[0];
277 ref.valuemask[1] = dsa->valuemask[1];
278 ref.writemask[0] = dsa->writemask[0];
279 ref.writemask[1] = dsa->writemask[1];
280
281 r600_set_stencil_ref(ctx, &ref);
282 }
283
284 void r600_bind_dsa_state(struct pipe_context *ctx, void *state)
285 {
286 struct r600_context *rctx = (struct r600_context *)ctx;
287 struct r600_pipe_dsa *dsa = state;
288 struct r600_pipe_state *rstate;
289 struct r600_stencil_ref ref;
290
291 if (state == NULL)
292 return;
293 rstate = &dsa->rstate;
294 rctx->states[rstate->id] = rstate;
295 r600_context_pipe_state_set(rctx, rstate);
296
297 ref.ref_value[0] = rctx->stencil_ref.ref_value[0];
298 ref.ref_value[1] = rctx->stencil_ref.ref_value[1];
299 ref.valuemask[0] = dsa->valuemask[0];
300 ref.valuemask[1] = dsa->valuemask[1];
301 ref.writemask[0] = dsa->writemask[0];
302 ref.writemask[1] = dsa->writemask[1];
303
304 r600_set_stencil_ref(ctx, &ref);
305
306 /* Update alphatest state. */
307 if (rctx->alphatest_state.sx_alpha_test_control != dsa->sx_alpha_test_control ||
308 rctx->alphatest_state.sx_alpha_ref != dsa->alpha_ref) {
309 rctx->alphatest_state.sx_alpha_test_control = dsa->sx_alpha_test_control;
310 rctx->alphatest_state.sx_alpha_ref = dsa->alpha_ref;
311 r600_atom_dirty(rctx, &rctx->alphatest_state.atom);
312 }
313 }
314
315 void r600_set_max_scissor(struct r600_context *rctx)
316 {
317 /* Set a scissor state such that it doesn't do anything. */
318 struct pipe_scissor_state scissor;
319 scissor.minx = 0;
320 scissor.miny = 0;
321 scissor.maxx = 8192;
322 scissor.maxy = 8192;
323
324 r600_set_scissor_state(rctx, &scissor);
325 }
326
327 void r600_bind_rs_state(struct pipe_context *ctx, void *state)
328 {
329 struct r600_pipe_rasterizer *rs = (struct r600_pipe_rasterizer *)state;
330 struct r600_context *rctx = (struct r600_context *)ctx;
331
332 if (state == NULL)
333 return;
334
335 rctx->sprite_coord_enable = rs->sprite_coord_enable;
336 rctx->two_side = rs->two_side;
337 rctx->pa_sc_line_stipple = rs->pa_sc_line_stipple;
338 rctx->pa_cl_clip_cntl = rs->pa_cl_clip_cntl;
339 rctx->multisample_enable = rs->multisample_enable;
340
341 rctx->rasterizer = rs;
342
343 rctx->states[rs->rstate.id] = &rs->rstate;
344 r600_context_pipe_state_set(rctx, &rs->rstate);
345
346 if (rctx->chip_class >= EVERGREEN) {
347 evergreen_polygon_offset_update(rctx);
348 } else {
349 r600_polygon_offset_update(rctx);
350 }
351
352 /* Workaround for a missing scissor enable on r600. */
353 if (rctx->chip_class == R600) {
354 if (rs->scissor_enable != rctx->scissor_enable) {
355 rctx->scissor_enable = rs->scissor_enable;
356
357 if (rs->scissor_enable) {
358 r600_set_scissor_state(rctx, &rctx->scissor_state);
359 } else {
360 r600_set_max_scissor(rctx);
361 }
362 }
363 }
364 }
365
366 void r600_delete_rs_state(struct pipe_context *ctx, void *state)
367 {
368 struct r600_context *rctx = (struct r600_context *)ctx;
369 struct r600_pipe_rasterizer *rs = (struct r600_pipe_rasterizer *)state;
370
371 if (rctx->rasterizer == rs) {
372 rctx->rasterizer = NULL;
373 }
374 if (rctx->states[rs->rstate.id] == &rs->rstate) {
375 rctx->states[rs->rstate.id] = NULL;
376 }
377 free(rs);
378 }
379
380 void r600_sampler_view_destroy(struct pipe_context *ctx,
381 struct pipe_sampler_view *state)
382 {
383 struct r600_pipe_sampler_view *resource = (struct r600_pipe_sampler_view *)state;
384
385 pipe_resource_reference(&state->texture, NULL);
386 FREE(resource);
387 }
388
389 static void r600_bind_samplers(struct pipe_context *pipe,
390 unsigned shader,
391 unsigned start,
392 unsigned count, void **states)
393 {
394 struct r600_context *rctx = (struct r600_context *)pipe;
395 struct r600_textures_info *dst;
396 int seamless_cube_map = -1;
397 unsigned i;
398
399 assert(start == 0); /* XXX fix below */
400
401 switch (shader) {
402 case PIPE_SHADER_VERTEX:
403 dst = &rctx->vs_samplers;
404 break;
405 case PIPE_SHADER_FRAGMENT:
406 dst = &rctx->ps_samplers;
407 break;
408 default:
409 debug_error("bad shader in r600_bind_samplers()");
410 return;
411 }
412
413 memcpy(dst->samplers, states, sizeof(void*) * count);
414 dst->n_samplers = count;
415 dst->atom_sampler.num_dw = 0;
416
417 for (i = 0; i < count; i++) {
418 struct r600_pipe_sampler_state *sampler = states[i];
419
420 if (sampler == NULL) {
421 continue;
422 }
423 if (sampler->border_color_use) {
424 dst->atom_sampler.num_dw += 11;
425 rctx->flags |= R600_PARTIAL_FLUSH;
426 } else {
427 dst->atom_sampler.num_dw += 5;
428 }
429 seamless_cube_map = sampler->seamless_cube_map;
430 }
431 if (rctx->chip_class <= R700 && seamless_cube_map != -1 && seamless_cube_map != rctx->seamless_cube_map.enabled) {
432 /* change in TA_CNTL_AUX need a pipeline flush */
433 rctx->flags |= R600_PARTIAL_FLUSH;
434 rctx->seamless_cube_map.enabled = seamless_cube_map;
435 r600_atom_dirty(rctx, &rctx->seamless_cube_map.atom);
436 }
437 if (dst->atom_sampler.num_dw) {
438 r600_atom_dirty(rctx, &dst->atom_sampler);
439 }
440 }
441
442 void r600_bind_vs_samplers(struct pipe_context *ctx, unsigned count, void **states)
443 {
444 r600_bind_samplers(ctx, PIPE_SHADER_VERTEX, 0, count, states);
445 }
446
447 void r600_bind_ps_samplers(struct pipe_context *ctx, unsigned count, void **states)
448 {
449 r600_bind_samplers(ctx, PIPE_SHADER_FRAGMENT, 0, count, states);
450 }
451
452 void r600_delete_sampler(struct pipe_context *ctx, void *state)
453 {
454 free(state);
455 }
456
457 void r600_delete_state(struct pipe_context *ctx, void *state)
458 {
459 struct r600_context *rctx = (struct r600_context *)ctx;
460 struct r600_pipe_state *rstate = (struct r600_pipe_state *)state;
461
462 if (rctx->states[rstate->id] == rstate) {
463 rctx->states[rstate->id] = NULL;
464 }
465 for (int i = 0; i < rstate->nregs; i++) {
466 pipe_resource_reference((struct pipe_resource**)&rstate->regs[i].bo, NULL);
467 }
468 free(rstate);
469 }
470
471 void r600_bind_vertex_elements(struct pipe_context *ctx, void *state)
472 {
473 struct r600_context *rctx = (struct r600_context *)ctx;
474 struct r600_vertex_element *v = (struct r600_vertex_element*)state;
475
476 rctx->vertex_elements = v;
477 if (v) {
478 r600_inval_shader_cache(rctx);
479
480 rctx->states[v->rstate.id] = &v->rstate;
481 r600_context_pipe_state_set(rctx, &v->rstate);
482 }
483 }
484
485 void r600_delete_vertex_element(struct pipe_context *ctx, void *state)
486 {
487 struct r600_context *rctx = (struct r600_context *)ctx;
488 struct r600_vertex_element *v = (struct r600_vertex_element*)state;
489
490 if (rctx->states[v->rstate.id] == &v->rstate) {
491 rctx->states[v->rstate.id] = NULL;
492 }
493 if (rctx->vertex_elements == state)
494 rctx->vertex_elements = NULL;
495
496 pipe_resource_reference((struct pipe_resource**)&v->fetch_shader, NULL);
497 FREE(state);
498 }
499
500 void r600_set_index_buffer(struct pipe_context *ctx,
501 const struct pipe_index_buffer *ib)
502 {
503 struct r600_context *rctx = (struct r600_context *)ctx;
504
505 if (ib) {
506 pipe_resource_reference(&rctx->index_buffer.buffer, ib->buffer);
507 memcpy(&rctx->index_buffer, ib, sizeof(*ib));
508 } else {
509 pipe_resource_reference(&rctx->index_buffer.buffer, NULL);
510 }
511 }
512
513 void r600_vertex_buffers_dirty(struct r600_context *rctx)
514 {
515 if (rctx->vertex_buffer_state.dirty_mask) {
516 r600_inval_vertex_cache(rctx);
517 rctx->vertex_buffer_state.atom.num_dw = (rctx->chip_class >= EVERGREEN ? 12 : 11) *
518 util_bitcount(rctx->vertex_buffer_state.dirty_mask);
519 r600_atom_dirty(rctx, &rctx->vertex_buffer_state.atom);
520 }
521 }
522
523 void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
524 const struct pipe_vertex_buffer *input)
525 {
526 struct r600_context *rctx = (struct r600_context *)ctx;
527 struct r600_vertexbuf_state *state = &rctx->vertex_buffer_state;
528 struct pipe_vertex_buffer *vb = state->vb;
529 unsigned i;
530 /* This sets 1-bit for buffers with index >= count. */
531 uint32_t disable_mask = ~((1ull << count) - 1);
532 /* These are the new buffers set by this function. */
533 uint32_t new_buffer_mask = 0;
534
535 /* Set buffers with index >= count to NULL. */
536 uint32_t remaining_buffers_mask =
537 rctx->vertex_buffer_state.enabled_mask & disable_mask;
538
539 while (remaining_buffers_mask) {
540 i = u_bit_scan(&remaining_buffers_mask);
541 pipe_resource_reference(&vb[i].buffer, NULL);
542 }
543
544 /* Set vertex buffers. */
545 for (i = 0; i < count; i++) {
546 if (memcmp(&input[i], &vb[i], sizeof(struct pipe_vertex_buffer))) {
547 if (input[i].buffer) {
548 vb[i].stride = input[i].stride;
549 vb[i].buffer_offset = input[i].buffer_offset;
550 pipe_resource_reference(&vb[i].buffer, input[i].buffer);
551 new_buffer_mask |= 1 << i;
552 } else {
553 pipe_resource_reference(&vb[i].buffer, NULL);
554 disable_mask |= 1 << i;
555 }
556 }
557 }
558
559 rctx->vertex_buffer_state.enabled_mask &= ~disable_mask;
560 rctx->vertex_buffer_state.dirty_mask &= rctx->vertex_buffer_state.enabled_mask;
561 rctx->vertex_buffer_state.enabled_mask |= new_buffer_mask;
562 rctx->vertex_buffer_state.dirty_mask |= new_buffer_mask;
563
564 r600_vertex_buffers_dirty(rctx);
565 }
566
567 void r600_sampler_views_dirty(struct r600_context *rctx,
568 struct r600_samplerview_state *state)
569 {
570 if (state->dirty_mask) {
571 r600_inval_texture_cache(rctx);
572 state->atom.num_dw = (rctx->chip_class >= EVERGREEN ? 14 : 13) *
573 util_bitcount(state->dirty_mask);
574 r600_atom_dirty(rctx, &state->atom);
575 }
576 }
577
578 void r600_set_sampler_views(struct pipe_context *pipe,
579 unsigned shader,
580 unsigned start,
581 unsigned count,
582 struct pipe_sampler_view **views)
583 {
584 struct r600_context *rctx = (struct r600_context *) pipe;
585 struct r600_textures_info *dst;
586 struct r600_pipe_sampler_view **rviews = (struct r600_pipe_sampler_view **)views;
587 unsigned i;
588 /* This sets 1-bit for textures with index >= count. */
589 uint32_t disable_mask = ~((1ull << count) - 1);
590 /* These are the new textures set by this function. */
591 uint32_t new_mask = 0;
592
593 /* Set textures with index >= count to NULL. */
594 uint32_t remaining_mask;
595
596 assert(start == 0); /* XXX fix below */
597
598 switch (shader) {
599 case PIPE_SHADER_VERTEX:
600 dst = &rctx->vs_samplers;
601 break;
602 case PIPE_SHADER_FRAGMENT:
603 dst = &rctx->ps_samplers;
604 break;
605 default:
606 debug_error("bad shader in r600_set_sampler_views()");
607 return;
608 }
609
610 remaining_mask = dst->views.enabled_mask & disable_mask;
611
612 while (remaining_mask) {
613 i = u_bit_scan(&remaining_mask);
614 assert(dst->views.views[i]);
615
616 pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], NULL);
617 }
618
619 for (i = 0; i < count; i++) {
620 if (rviews[i] == dst->views.views[i]) {
621 continue;
622 }
623
624 if (rviews[i]) {
625 struct r600_texture *rtex =
626 (struct r600_texture*)rviews[i]->base.texture;
627
628 if (rtex->is_depth && !rtex->is_flushing_texture) {
629 dst->views.compressed_depthtex_mask |= 1 << i;
630 } else {
631 dst->views.compressed_depthtex_mask &= ~(1 << i);
632 }
633
634 /* Track compressed colorbuffers for Evergreen (Cayman doesn't need this). */
635 if (rctx->chip_class != CAYMAN && rtex->cmask_size && rtex->fmask_size) {
636 dst->views.compressed_colortex_mask |= 1 << i;
637 } else {
638 dst->views.compressed_colortex_mask &= ~(1 << i);
639 }
640
641 /* Changing from array to non-arrays textures and vice
642 * versa requires updating TEX_ARRAY_OVERRIDE on R6xx-R7xx. */
643 if (rctx->chip_class <= R700 &&
644 (rviews[i]->base.texture->target == PIPE_TEXTURE_1D_ARRAY ||
645 rviews[i]->base.texture->target == PIPE_TEXTURE_2D_ARRAY) != dst->is_array_sampler[i]) {
646 r600_atom_dirty(rctx, &dst->atom_sampler);
647 }
648
649 pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], views[i]);
650 new_mask |= 1 << i;
651 } else {
652 pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], NULL);
653 disable_mask |= 1 << i;
654 }
655 }
656
657 dst->views.enabled_mask &= ~disable_mask;
658 dst->views.dirty_mask &= dst->views.enabled_mask;
659 dst->views.enabled_mask |= new_mask;
660 dst->views.dirty_mask |= new_mask;
661 dst->views.compressed_depthtex_mask &= dst->views.enabled_mask;
662 dst->views.compressed_colortex_mask &= dst->views.enabled_mask;
663
664 r600_sampler_views_dirty(rctx, &dst->views);
665 }
666
667 void *r600_create_vertex_elements(struct pipe_context *ctx,
668 unsigned count,
669 const struct pipe_vertex_element *elements)
670 {
671 struct r600_context *rctx = (struct r600_context *)ctx;
672 struct r600_vertex_element *v = CALLOC_STRUCT(r600_vertex_element);
673
674 assert(count < 32);
675 if (!v)
676 return NULL;
677
678 v->count = count;
679 memcpy(v->elements, elements, sizeof(struct pipe_vertex_element) * count);
680
681 if (r600_vertex_elements_build_fetch_shader(rctx, v)) {
682 FREE(v);
683 return NULL;
684 }
685
686 return v;
687 }
688
689 /* Compute the key for the hw shader variant */
690 static INLINE unsigned r600_shader_selector_key(struct pipe_context * ctx,
691 struct r600_pipe_shader_selector * sel)
692 {
693 struct r600_context *rctx = (struct r600_context *)ctx;
694 unsigned key;
695
696 if (sel->type == PIPE_SHADER_FRAGMENT) {
697 key = rctx->two_side |
698 ((rctx->alpha_to_one && rctx->multisample_enable && !rctx->cb0_is_integer) << 1) |
699 (MIN2(sel->nr_ps_max_color_exports, rctx->nr_cbufs + rctx->dual_src_blend) << 2);
700 } else
701 key = 0;
702
703 return key;
704 }
705
706 /* Select the hw shader variant depending on the current state.
707 * (*dirty) is set to 1 if current variant was changed */
708 static int r600_shader_select(struct pipe_context *ctx,
709 struct r600_pipe_shader_selector* sel,
710 unsigned *dirty)
711 {
712 unsigned key;
713 struct r600_context *rctx = (struct r600_context *)ctx;
714 struct r600_pipe_shader * shader = NULL;
715 int r;
716
717 key = r600_shader_selector_key(ctx, sel);
718
719 /* Check if we don't need to change anything.
720 * This path is also used for most shaders that don't need multiple
721 * variants, it will cost just a computation of the key and this
722 * test. */
723 if (likely(sel->current && sel->current->key == key)) {
724 return 0;
725 }
726
727 /* lookup if we have other variants in the list */
728 if (sel->num_shaders > 1) {
729 struct r600_pipe_shader *p = sel->current, *c = p->next_variant;
730
731 while (c && c->key != key) {
732 p = c;
733 c = c->next_variant;
734 }
735
736 if (c) {
737 p->next_variant = c->next_variant;
738 shader = c;
739 }
740 }
741
742 if (unlikely(!shader)) {
743 shader = CALLOC(1, sizeof(struct r600_pipe_shader));
744 shader->selector = sel;
745
746 r = r600_pipe_shader_create(ctx, shader);
747 if (unlikely(r)) {
748 R600_ERR("Failed to build shader variant (type=%u, key=%u) %d\n",
749 sel->type, key, r);
750 sel->current = NULL;
751 return r;
752 }
753
754 /* We don't know the value of nr_ps_max_color_exports until we built
755 * at least one variant, so we may need to recompute the key after
756 * building first variant. */
757 if (sel->type == PIPE_SHADER_FRAGMENT &&
758 sel->num_shaders == 0) {
759 sel->nr_ps_max_color_exports = shader->shader.nr_ps_max_color_exports;
760 key = r600_shader_selector_key(ctx, sel);
761 }
762
763 shader->key = key;
764 sel->num_shaders++;
765 }
766
767 if (dirty)
768 *dirty = 1;
769
770 shader->next_variant = sel->current;
771 sel->current = shader;
772
773 if (rctx->chip_class < EVERGREEN && rctx->ps_shader && rctx->vs_shader) {
774 r600_adjust_gprs(rctx);
775 }
776
777 if (rctx->ps_shader &&
778 rctx->cb_misc_state.nr_ps_color_outputs != rctx->ps_shader->current->nr_ps_color_outputs) {
779 rctx->cb_misc_state.nr_ps_color_outputs = rctx->ps_shader->current->nr_ps_color_outputs;
780 r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
781 }
782 return 0;
783 }
784
785 static void *r600_create_shader_state(struct pipe_context *ctx,
786 const struct pipe_shader_state *state,
787 unsigned pipe_shader_type)
788 {
789 struct r600_pipe_shader_selector *sel = CALLOC_STRUCT(r600_pipe_shader_selector);
790 int r;
791
792 sel->type = pipe_shader_type;
793 sel->tokens = tgsi_dup_tokens(state->tokens);
794 sel->so = state->stream_output;
795
796 r = r600_shader_select(ctx, sel, NULL);
797 if (r)
798 return NULL;
799
800 return sel;
801 }
802
803 void *r600_create_shader_state_ps(struct pipe_context *ctx,
804 const struct pipe_shader_state *state)
805 {
806 return r600_create_shader_state(ctx, state, PIPE_SHADER_FRAGMENT);
807 }
808
809 void *r600_create_shader_state_vs(struct pipe_context *ctx,
810 const struct pipe_shader_state *state)
811 {
812 return r600_create_shader_state(ctx, state, PIPE_SHADER_VERTEX);
813 }
814
815 void r600_bind_ps_shader(struct pipe_context *ctx, void *state)
816 {
817 struct r600_context *rctx = (struct r600_context *)ctx;
818
819 if (!state)
820 state = rctx->dummy_pixel_shader;
821
822 rctx->ps_shader = (struct r600_pipe_shader_selector *)state;
823 r600_context_pipe_state_set(rctx, &rctx->ps_shader->current->rstate);
824
825 if (rctx->chip_class <= R700) {
826 bool multiwrite = rctx->ps_shader->current->shader.fs_write_all;
827
828 if (rctx->cb_misc_state.multiwrite != multiwrite) {
829 rctx->cb_misc_state.multiwrite = multiwrite;
830 r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
831 }
832
833 if (rctx->vs_shader)
834 r600_adjust_gprs(rctx);
835 }
836
837 if (rctx->cb_misc_state.nr_ps_color_outputs != rctx->ps_shader->current->nr_ps_color_outputs) {
838 rctx->cb_misc_state.nr_ps_color_outputs = rctx->ps_shader->current->nr_ps_color_outputs;
839 r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
840 }
841 }
842
843 void r600_bind_vs_shader(struct pipe_context *ctx, void *state)
844 {
845 struct r600_context *rctx = (struct r600_context *)ctx;
846
847 rctx->vs_shader = (struct r600_pipe_shader_selector *)state;
848 if (state) {
849 r600_context_pipe_state_set(rctx, &rctx->vs_shader->current->rstate);
850
851 if (rctx->chip_class < EVERGREEN && rctx->ps_shader)
852 r600_adjust_gprs(rctx);
853 }
854 }
855
856 static void r600_delete_shader_selector(struct pipe_context *ctx,
857 struct r600_pipe_shader_selector *sel)
858 {
859 struct r600_pipe_shader *p = sel->current, *c;
860 while (p) {
861 c = p->next_variant;
862 r600_pipe_shader_destroy(ctx, p);
863 free(p);
864 p = c;
865 }
866
867 free(sel->tokens);
868 free(sel);
869 }
870
871
872 void r600_delete_ps_shader(struct pipe_context *ctx, void *state)
873 {
874 struct r600_context *rctx = (struct r600_context *)ctx;
875 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state;
876
877 if (rctx->ps_shader == sel) {
878 rctx->ps_shader = NULL;
879 }
880
881 r600_delete_shader_selector(ctx, sel);
882 }
883
884 void r600_delete_vs_shader(struct pipe_context *ctx, void *state)
885 {
886 struct r600_context *rctx = (struct r600_context *)ctx;
887 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state;
888
889 if (rctx->vs_shader == sel) {
890 rctx->vs_shader = NULL;
891 }
892
893 r600_delete_shader_selector(ctx, sel);
894 }
895
896 void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state)
897 {
898 if (state->dirty_mask) {
899 r600_inval_shader_cache(rctx);
900 state->atom.num_dw = rctx->chip_class >= EVERGREEN ? util_bitcount(state->dirty_mask)*20
901 : util_bitcount(state->dirty_mask)*19;
902 r600_atom_dirty(rctx, &state->atom);
903 }
904 }
905
906 void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
907 struct pipe_constant_buffer *input)
908 {
909 struct r600_context *rctx = (struct r600_context *)ctx;
910 struct r600_constbuf_state *state;
911 struct pipe_constant_buffer *cb;
912 const uint8_t *ptr;
913
914 switch (shader) {
915 case PIPE_SHADER_VERTEX:
916 state = &rctx->vs_constbuf_state;
917 break;
918 case PIPE_SHADER_FRAGMENT:
919 state = &rctx->ps_constbuf_state;
920 break;
921 default:
922 return;
923 }
924
925 /* Note that the state tracker can unbind constant buffers by
926 * passing NULL here.
927 */
928 if (unlikely(!input)) {
929 state->enabled_mask &= ~(1 << index);
930 state->dirty_mask &= ~(1 << index);
931 pipe_resource_reference(&state->cb[index].buffer, NULL);
932 return;
933 }
934
935 cb = &state->cb[index];
936 cb->buffer_size = input->buffer_size;
937
938 ptr = input->user_buffer;
939
940 if (ptr) {
941 /* Upload the user buffer. */
942 if (R600_BIG_ENDIAN) {
943 uint32_t *tmpPtr;
944 unsigned i, size = input->buffer_size;
945
946 if (!(tmpPtr = malloc(size))) {
947 R600_ERR("Failed to allocate BE swap buffer.\n");
948 return;
949 }
950
951 for (i = 0; i < size / 4; ++i) {
952 tmpPtr[i] = bswap_32(((uint32_t *)ptr)[i]);
953 }
954
955 u_upload_data(rctx->uploader, 0, size, tmpPtr, &cb->buffer_offset, &cb->buffer);
956 free(tmpPtr);
957 } else {
958 u_upload_data(rctx->uploader, 0, input->buffer_size, ptr, &cb->buffer_offset, &cb->buffer);
959 }
960 } else {
961 /* Setup the hw buffer. */
962 cb->buffer_offset = input->buffer_offset;
963 pipe_resource_reference(&cb->buffer, input->buffer);
964 }
965
966 state->enabled_mask |= 1 << index;
967 state->dirty_mask |= 1 << index;
968 r600_constant_buffers_dirty(rctx, state);
969 }
970
971 struct pipe_stream_output_target *
972 r600_create_so_target(struct pipe_context *ctx,
973 struct pipe_resource *buffer,
974 unsigned buffer_offset,
975 unsigned buffer_size)
976 {
977 struct r600_context *rctx = (struct r600_context *)ctx;
978 struct r600_so_target *t;
979 void *ptr;
980
981 t = CALLOC_STRUCT(r600_so_target);
982 if (!t) {
983 return NULL;
984 }
985
986 t->b.reference.count = 1;
987 t->b.context = ctx;
988 pipe_resource_reference(&t->b.buffer, buffer);
989 t->b.buffer_offset = buffer_offset;
990 t->b.buffer_size = buffer_size;
991
992 t->filled_size = (struct r600_resource*)
993 pipe_buffer_create(ctx->screen, PIPE_BIND_CUSTOM, PIPE_USAGE_STATIC, 4);
994 ptr = rctx->ws->buffer_map(t->filled_size->cs_buf, rctx->cs, PIPE_TRANSFER_WRITE);
995 memset(ptr, 0, t->filled_size->buf->size);
996 rctx->ws->buffer_unmap(t->filled_size->cs_buf);
997
998 return &t->b;
999 }
1000
1001 void r600_so_target_destroy(struct pipe_context *ctx,
1002 struct pipe_stream_output_target *target)
1003 {
1004 struct r600_so_target *t = (struct r600_so_target*)target;
1005 pipe_resource_reference(&t->b.buffer, NULL);
1006 pipe_resource_reference((struct pipe_resource**)&t->filled_size, NULL);
1007 FREE(t);
1008 }
1009
1010 void r600_set_so_targets(struct pipe_context *ctx,
1011 unsigned num_targets,
1012 struct pipe_stream_output_target **targets,
1013 unsigned append_bitmask)
1014 {
1015 struct r600_context *rctx = (struct r600_context *)ctx;
1016 unsigned i;
1017
1018 /* Stop streamout. */
1019 if (rctx->num_so_targets && !rctx->streamout_start) {
1020 r600_context_streamout_end(rctx);
1021 }
1022
1023 /* Set the new targets. */
1024 for (i = 0; i < num_targets; i++) {
1025 pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->so_targets[i], targets[i]);
1026 }
1027 for (; i < rctx->num_so_targets; i++) {
1028 pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->so_targets[i], NULL);
1029 }
1030
1031 rctx->num_so_targets = num_targets;
1032 rctx->streamout_start = num_targets != 0;
1033 rctx->streamout_append_bitmask = append_bitmask;
1034 }
1035
1036 void r600_set_sample_mask(struct pipe_context *pipe, unsigned sample_mask)
1037 {
1038 struct r600_context *rctx = (struct r600_context*)pipe;
1039
1040 if (rctx->sample_mask.sample_mask == (uint16_t)sample_mask)
1041 return;
1042
1043 rctx->sample_mask.sample_mask = sample_mask;
1044 r600_atom_dirty(rctx, &rctx->sample_mask.atom);
1045 }
1046
1047 static void r600_update_derived_state(struct r600_context *rctx)
1048 {
1049 struct pipe_context * ctx = (struct pipe_context*)rctx;
1050 unsigned ps_dirty = 0, blend_override;
1051
1052 if (!rctx->blitter->running) {
1053 /* Decompress textures if needed. */
1054 if (rctx->vs_samplers.views.compressed_depthtex_mask) {
1055 r600_decompress_depth_textures(rctx, &rctx->vs_samplers.views);
1056 }
1057 if (rctx->ps_samplers.views.compressed_depthtex_mask) {
1058 r600_decompress_depth_textures(rctx, &rctx->ps_samplers.views);
1059 }
1060 if (rctx->vs_samplers.views.compressed_colortex_mask) {
1061 r600_decompress_color_textures(rctx, &rctx->vs_samplers.views);
1062 }
1063 if (rctx->ps_samplers.views.compressed_colortex_mask) {
1064 r600_decompress_color_textures(rctx, &rctx->ps_samplers.views);
1065 }
1066 }
1067
1068 r600_shader_select(ctx, rctx->ps_shader, &ps_dirty);
1069
1070 if (rctx->ps_shader && ((rctx->sprite_coord_enable &&
1071 (rctx->ps_shader->current->sprite_coord_enable != rctx->sprite_coord_enable)) ||
1072 (rctx->rasterizer && rctx->rasterizer->flatshade != rctx->ps_shader->current->flatshade))) {
1073
1074 if (rctx->chip_class >= EVERGREEN)
1075 evergreen_pipe_shader_ps(ctx, rctx->ps_shader->current);
1076 else
1077 r600_pipe_shader_ps(ctx, rctx->ps_shader->current);
1078
1079 ps_dirty = 1;
1080 }
1081
1082 if (ps_dirty)
1083 r600_context_pipe_state_set(rctx, &rctx->ps_shader->current->rstate);
1084
1085 blend_override = (rctx->dual_src_blend &&
1086 rctx->ps_shader->current->nr_ps_color_outputs < 2);
1087
1088 if (blend_override != rctx->blend_override) {
1089 rctx->blend_override = blend_override;
1090 r600_bind_blend_state_internal(rctx,
1091 blend_override ? rctx->no_blend : rctx->blend);
1092 }
1093
1094 if (rctx->chip_class >= EVERGREEN) {
1095 evergreen_update_dual_export_state(rctx);
1096 } else {
1097 r600_update_dual_export_state(rctx);
1098 }
1099 }
1100
1101 static unsigned r600_conv_prim_to_gs_out(unsigned mode)
1102 {
1103 static const int prim_conv[] = {
1104 V_028A6C_OUTPRIM_TYPE_POINTLIST,
1105 V_028A6C_OUTPRIM_TYPE_LINESTRIP,
1106 V_028A6C_OUTPRIM_TYPE_LINESTRIP,
1107 V_028A6C_OUTPRIM_TYPE_LINESTRIP,
1108 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1109 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1110 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1111 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1112 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1113 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1114 V_028A6C_OUTPRIM_TYPE_LINESTRIP,
1115 V_028A6C_OUTPRIM_TYPE_LINESTRIP,
1116 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1117 V_028A6C_OUTPRIM_TYPE_TRISTRIP,
1118 V_028A6C_OUTPRIM_TYPE_TRISTRIP
1119 };
1120 assert(mode < Elements(prim_conv));
1121
1122 return prim_conv[mode];
1123 }
1124
1125 void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
1126 {
1127 struct r600_context *rctx = (struct r600_context *)ctx;
1128 struct pipe_draw_info info = *dinfo;
1129 struct pipe_index_buffer ib = {};
1130 unsigned prim, ls_mask = 0;
1131 struct r600_block *dirty_block = NULL, *next_block = NULL;
1132 struct r600_atom *state = NULL, *next_state = NULL;
1133 struct radeon_winsys_cs *cs = rctx->cs;
1134 uint64_t va;
1135 uint8_t *ptr;
1136
1137 if ((!info.count && (info.indexed || !info.count_from_stream_output)) ||
1138 !r600_conv_pipe_prim(info.mode, &prim)) {
1139 assert(0);
1140 return;
1141 }
1142
1143 if (!rctx->vs_shader) {
1144 assert(0);
1145 return;
1146 }
1147
1148 r600_update_derived_state(rctx);
1149
1150 /* partial flush triggered by border color change */
1151 if (rctx->flags & R600_PARTIAL_FLUSH) {
1152 rctx->flags &= ~R600_PARTIAL_FLUSH;
1153 r600_write_value(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1154 r600_write_value(cs, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
1155 }
1156
1157 if (info.indexed) {
1158 /* Initialize the index buffer struct. */
1159 pipe_resource_reference(&ib.buffer, rctx->index_buffer.buffer);
1160 ib.user_buffer = rctx->index_buffer.user_buffer;
1161 ib.index_size = rctx->index_buffer.index_size;
1162 ib.offset = rctx->index_buffer.offset + info.start * ib.index_size;
1163
1164 /* Translate or upload, if needed. */
1165 r600_translate_index_buffer(rctx, &ib, info.count);
1166
1167 ptr = (uint8_t*)ib.user_buffer;
1168 if (!ib.buffer && ptr) {
1169 u_upload_data(rctx->uploader, 0, info.count * ib.index_size,
1170 ptr, &ib.offset, &ib.buffer);
1171 }
1172 } else {
1173 info.index_bias = info.start;
1174 }
1175
1176 if (rctx->vgt.id != R600_PIPE_STATE_VGT) {
1177 rctx->vgt.id = R600_PIPE_STATE_VGT;
1178 rctx->vgt.nregs = 0;
1179 r600_pipe_state_add_reg(&rctx->vgt, R_008958_VGT_PRIMITIVE_TYPE, prim);
1180 r600_pipe_state_add_reg(&rctx->vgt, R_028A6C_VGT_GS_OUT_PRIM_TYPE, 0);
1181 r600_pipe_state_add_reg(&rctx->vgt, R_028408_VGT_INDX_OFFSET, info.index_bias);
1182 r600_pipe_state_add_reg(&rctx->vgt, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, info.restart_index);
1183 r600_pipe_state_add_reg(&rctx->vgt, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, info.primitive_restart);
1184 r600_pipe_state_add_reg(&rctx->vgt, R_03CFF4_SQ_VTX_START_INST_LOC, info.start_instance);
1185 r600_pipe_state_add_reg(&rctx->vgt, R_028A0C_PA_SC_LINE_STIPPLE, 0);
1186 r600_pipe_state_add_reg(&rctx->vgt, R_02881C_PA_CL_VS_OUT_CNTL, 0);
1187 r600_pipe_state_add_reg(&rctx->vgt, R_028810_PA_CL_CLIP_CNTL, 0);
1188 }
1189
1190 rctx->vgt.nregs = 0;
1191 r600_pipe_state_mod_reg(&rctx->vgt, prim);
1192 r600_pipe_state_mod_reg(&rctx->vgt, r600_conv_prim_to_gs_out(info.mode));
1193 r600_pipe_state_mod_reg(&rctx->vgt, info.index_bias);
1194 r600_pipe_state_mod_reg(&rctx->vgt, info.restart_index);
1195 r600_pipe_state_mod_reg(&rctx->vgt, info.primitive_restart);
1196 r600_pipe_state_mod_reg(&rctx->vgt, info.start_instance);
1197
1198 if (prim == V_008958_DI_PT_LINELIST)
1199 ls_mask = 1;
1200 else if (prim == V_008958_DI_PT_LINESTRIP ||
1201 prim == V_008958_DI_PT_LINELOOP)
1202 ls_mask = 2;
1203 r600_pipe_state_mod_reg(&rctx->vgt, S_028A0C_AUTO_RESET_CNTL(ls_mask) | rctx->pa_sc_line_stipple);
1204 r600_pipe_state_mod_reg(&rctx->vgt,
1205 rctx->vs_shader->current->pa_cl_vs_out_cntl |
1206 (rctx->rasterizer->clip_plane_enable & rctx->vs_shader->current->shader.clip_dist_write));
1207 r600_pipe_state_mod_reg(&rctx->vgt,
1208 rctx->pa_cl_clip_cntl |
1209 (rctx->vs_shader->current->shader.clip_dist_write ||
1210 rctx->vs_shader->current->shader.vs_prohibit_ucps ?
1211 0 : rctx->rasterizer->clip_plane_enable & 0x3F));
1212
1213 r600_context_pipe_state_set(rctx, &rctx->vgt);
1214
1215 /* Enable stream out if needed. */
1216 if (rctx->streamout_start) {
1217 r600_context_streamout_begin(rctx);
1218 rctx->streamout_start = FALSE;
1219 }
1220
1221 /* Emit states (the function expects that we emit at most 17 dwords here). */
1222 r600_need_cs_space(rctx, 0, TRUE);
1223
1224 LIST_FOR_EACH_ENTRY_SAFE(state, next_state, &rctx->dirty_states, head) {
1225 r600_emit_atom(rctx, state);
1226 }
1227 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &rctx->dirty,list) {
1228 r600_context_block_emit_dirty(rctx, dirty_block, 0 /* pkt_flags */);
1229 }
1230 rctx->pm4_dirty_cdwords = 0;
1231
1232 /* draw packet */
1233 cs->buf[cs->cdw++] = PKT3(PKT3_NUM_INSTANCES, 0, rctx->predicate_drawing);
1234 cs->buf[cs->cdw++] = info.instance_count;
1235 if (info.indexed) {
1236 cs->buf[cs->cdw++] = PKT3(PKT3_INDEX_TYPE, 0, rctx->predicate_drawing);
1237 cs->buf[cs->cdw++] = ib.index_size == 4 ?
1238 (VGT_INDEX_32 | (R600_BIG_ENDIAN ? VGT_DMA_SWAP_32_BIT : 0)) :
1239 (VGT_INDEX_16 | (R600_BIG_ENDIAN ? VGT_DMA_SWAP_16_BIT : 0));
1240
1241 va = r600_resource_va(ctx->screen, ib.buffer);
1242 va += ib.offset;
1243 cs->buf[cs->cdw++] = PKT3(PKT3_DRAW_INDEX, 3, rctx->predicate_drawing);
1244 cs->buf[cs->cdw++] = va;
1245 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
1246 cs->buf[cs->cdw++] = info.count;
1247 cs->buf[cs->cdw++] = V_0287F0_DI_SRC_SEL_DMA;
1248 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, rctx->predicate_drawing);
1249 cs->buf[cs->cdw++] = r600_context_bo_reloc(rctx, (struct r600_resource*)ib.buffer, RADEON_USAGE_READ);
1250 } else {
1251 if (info.count_from_stream_output) {
1252 struct r600_so_target *t = (struct r600_so_target*)info.count_from_stream_output;
1253 uint64_t va = r600_resource_va(&rctx->screen->screen, (void*)t->filled_size);
1254
1255 r600_write_context_reg(cs, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE, t->stride_in_dw);
1256
1257 cs->buf[cs->cdw++] = PKT3(PKT3_COPY_DW, 4, 0);
1258 cs->buf[cs->cdw++] = COPY_DW_SRC_IS_MEM | COPY_DW_DST_IS_REG;
1259 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* src address lo */
1260 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* src address hi */
1261 cs->buf[cs->cdw++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2; /* dst register */
1262 cs->buf[cs->cdw++] = 0; /* unused */
1263
1264 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1265 cs->buf[cs->cdw++] = r600_context_bo_reloc(rctx, t->filled_size, RADEON_USAGE_READ);
1266 }
1267
1268 cs->buf[cs->cdw++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, rctx->predicate_drawing);
1269 cs->buf[cs->cdw++] = info.count;
1270 cs->buf[cs->cdw++] = V_0287F0_DI_SRC_SEL_AUTO_INDEX |
1271 (info.count_from_stream_output ? S_0287F0_USE_OPAQUE(1) : 0);
1272 }
1273
1274 rctx->flags |= R600_CONTEXT_DST_CACHES_DIRTY | R600_CONTEXT_DRAW_PENDING;
1275
1276 /* Set the depth buffer as dirty. */
1277 if (rctx->framebuffer.zsbuf) {
1278 struct pipe_surface *surf = rctx->framebuffer.zsbuf;
1279 struct r600_texture *rtex = (struct r600_texture *)surf->texture;
1280
1281 rtex->dirty_level_mask |= 1 << surf->u.tex.level;
1282 }
1283 if (rctx->compressed_cb_mask) {
1284 struct pipe_surface *surf;
1285 struct r600_texture *rtex;
1286 unsigned mask = rctx->compressed_cb_mask;
1287
1288 do {
1289 unsigned i = u_bit_scan(&mask);
1290 surf = rctx->framebuffer.cbufs[i];
1291 rtex = (struct r600_texture*)surf->texture;
1292
1293 rtex->dirty_level_mask |= 1 << surf->u.tex.level;
1294
1295 } while (mask);
1296 }
1297
1298 pipe_resource_reference(&ib.buffer, NULL);
1299 }
1300
1301 void r600_draw_rectangle(struct blitter_context *blitter,
1302 unsigned x1, unsigned y1, unsigned x2, unsigned y2, float depth,
1303 enum blitter_attrib_type type, const union pipe_color_union *attrib)
1304 {
1305 struct r600_context *rctx = (struct r600_context*)util_blitter_get_pipe(blitter);
1306 struct pipe_viewport_state viewport;
1307 struct pipe_resource *buf = NULL;
1308 unsigned offset = 0;
1309 float *vb;
1310
1311 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
1312 util_blitter_draw_rectangle(blitter, x1, y1, x2, y2, depth, type, attrib);
1313 return;
1314 }
1315
1316 /* Some operations (like color resolve on r6xx) don't work
1317 * with the conventional primitive types.
1318 * One that works is PT_RECTLIST, which we use here. */
1319
1320 /* setup viewport */
1321 viewport.scale[0] = 1.0f;
1322 viewport.scale[1] = 1.0f;
1323 viewport.scale[2] = 1.0f;
1324 viewport.scale[3] = 1.0f;
1325 viewport.translate[0] = 0.0f;
1326 viewport.translate[1] = 0.0f;
1327 viewport.translate[2] = 0.0f;
1328 viewport.translate[3] = 0.0f;
1329 rctx->context.set_viewport_state(&rctx->context, &viewport);
1330
1331 /* Upload vertices. The hw rectangle has only 3 vertices,
1332 * I guess the 4th one is derived from the first 3.
1333 * The vertex specification should match u_blitter's vertex element state. */
1334 u_upload_alloc(rctx->uploader, 0, sizeof(float) * 24, &offset, &buf, (void**)&vb);
1335 vb[0] = x1;
1336 vb[1] = y1;
1337 vb[2] = depth;
1338 vb[3] = 1;
1339
1340 vb[8] = x1;
1341 vb[9] = y2;
1342 vb[10] = depth;
1343 vb[11] = 1;
1344
1345 vb[16] = x2;
1346 vb[17] = y1;
1347 vb[18] = depth;
1348 vb[19] = 1;
1349
1350 if (attrib) {
1351 memcpy(vb+4, attrib->f, sizeof(float)*4);
1352 memcpy(vb+12, attrib->f, sizeof(float)*4);
1353 memcpy(vb+20, attrib->f, sizeof(float)*4);
1354 }
1355
1356 /* draw */
1357 util_draw_vertex_buffer(&rctx->context, NULL, buf, offset,
1358 R600_PRIM_RECTANGLE_LIST, 3, 2);
1359 pipe_resource_reference(&buf, NULL);
1360 }
1361
1362 void _r600_pipe_state_add_reg_bo(struct r600_context *ctx,
1363 struct r600_pipe_state *state,
1364 uint32_t offset, uint32_t value,
1365 uint32_t range_id, uint32_t block_id,
1366 struct r600_resource *bo,
1367 enum radeon_bo_usage usage)
1368
1369 {
1370 struct r600_range *range;
1371 struct r600_block *block;
1372
1373 if (bo) assert(usage);
1374
1375 range = &ctx->range[range_id];
1376 block = range->blocks[block_id];
1377 state->regs[state->nregs].block = block;
1378 state->regs[state->nregs].id = (offset - block->start_offset) >> 2;
1379
1380 state->regs[state->nregs].value = value;
1381 state->regs[state->nregs].bo = bo;
1382 state->regs[state->nregs].bo_usage = usage;
1383
1384 state->nregs++;
1385 assert(state->nregs < R600_BLOCK_MAX_REG);
1386 }
1387
1388 void _r600_pipe_state_add_reg(struct r600_context *ctx,
1389 struct r600_pipe_state *state,
1390 uint32_t offset, uint32_t value,
1391 uint32_t range_id, uint32_t block_id)
1392 {
1393 _r600_pipe_state_add_reg_bo(ctx, state, offset, value,
1394 range_id, block_id, NULL, 0);
1395 }
1396
1397 void r600_pipe_state_add_reg_noblock(struct r600_pipe_state *state,
1398 uint32_t offset, uint32_t value,
1399 struct r600_resource *bo,
1400 enum radeon_bo_usage usage)
1401 {
1402 if (bo) assert(usage);
1403
1404 state->regs[state->nregs].id = offset;
1405 state->regs[state->nregs].block = NULL;
1406 state->regs[state->nregs].value = value;
1407 state->regs[state->nregs].bo = bo;
1408 state->regs[state->nregs].bo_usage = usage;
1409
1410 state->nregs++;
1411 assert(state->nregs < R600_BLOCK_MAX_REG);
1412 }
1413
1414 uint32_t r600_translate_stencil_op(int s_op)
1415 {
1416 switch (s_op) {
1417 case PIPE_STENCIL_OP_KEEP:
1418 return V_028800_STENCIL_KEEP;
1419 case PIPE_STENCIL_OP_ZERO:
1420 return V_028800_STENCIL_ZERO;
1421 case PIPE_STENCIL_OP_REPLACE:
1422 return V_028800_STENCIL_REPLACE;
1423 case PIPE_STENCIL_OP_INCR:
1424 return V_028800_STENCIL_INCR;
1425 case PIPE_STENCIL_OP_DECR:
1426 return V_028800_STENCIL_DECR;
1427 case PIPE_STENCIL_OP_INCR_WRAP:
1428 return V_028800_STENCIL_INCR_WRAP;
1429 case PIPE_STENCIL_OP_DECR_WRAP:
1430 return V_028800_STENCIL_DECR_WRAP;
1431 case PIPE_STENCIL_OP_INVERT:
1432 return V_028800_STENCIL_INVERT;
1433 default:
1434 R600_ERR("Unknown stencil op %d", s_op);
1435 assert(0);
1436 break;
1437 }
1438 return 0;
1439 }
1440
1441 uint32_t r600_translate_fill(uint32_t func)
1442 {
1443 switch(func) {
1444 case PIPE_POLYGON_MODE_FILL:
1445 return 2;
1446 case PIPE_POLYGON_MODE_LINE:
1447 return 1;
1448 case PIPE_POLYGON_MODE_POINT:
1449 return 0;
1450 default:
1451 assert(0);
1452 return 0;
1453 }
1454 }
1455
1456 unsigned r600_tex_wrap(unsigned wrap)
1457 {
1458 switch (wrap) {
1459 default:
1460 case PIPE_TEX_WRAP_REPEAT:
1461 return V_03C000_SQ_TEX_WRAP;
1462 case PIPE_TEX_WRAP_CLAMP:
1463 return V_03C000_SQ_TEX_CLAMP_HALF_BORDER;
1464 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1465 return V_03C000_SQ_TEX_CLAMP_LAST_TEXEL;
1466 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1467 return V_03C000_SQ_TEX_CLAMP_BORDER;
1468 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1469 return V_03C000_SQ_TEX_MIRROR;
1470 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1471 return V_03C000_SQ_TEX_MIRROR_ONCE_HALF_BORDER;
1472 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1473 return V_03C000_SQ_TEX_MIRROR_ONCE_LAST_TEXEL;
1474 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1475 return V_03C000_SQ_TEX_MIRROR_ONCE_BORDER;
1476 }
1477 }
1478
1479 unsigned r600_tex_filter(unsigned filter)
1480 {
1481 switch (filter) {
1482 default:
1483 case PIPE_TEX_FILTER_NEAREST:
1484 return V_03C000_SQ_TEX_XY_FILTER_POINT;
1485 case PIPE_TEX_FILTER_LINEAR:
1486 return V_03C000_SQ_TEX_XY_FILTER_BILINEAR;
1487 }
1488 }
1489
1490 unsigned r600_tex_mipfilter(unsigned filter)
1491 {
1492 switch (filter) {
1493 case PIPE_TEX_MIPFILTER_NEAREST:
1494 return V_03C000_SQ_TEX_Z_FILTER_POINT;
1495 case PIPE_TEX_MIPFILTER_LINEAR:
1496 return V_03C000_SQ_TEX_Z_FILTER_LINEAR;
1497 default:
1498 case PIPE_TEX_MIPFILTER_NONE:
1499 return V_03C000_SQ_TEX_Z_FILTER_NONE;
1500 }
1501 }
1502
1503 unsigned r600_tex_compare(unsigned compare)
1504 {
1505 switch (compare) {
1506 default:
1507 case PIPE_FUNC_NEVER:
1508 return V_03C000_SQ_TEX_DEPTH_COMPARE_NEVER;
1509 case PIPE_FUNC_LESS:
1510 return V_03C000_SQ_TEX_DEPTH_COMPARE_LESS;
1511 case PIPE_FUNC_EQUAL:
1512 return V_03C000_SQ_TEX_DEPTH_COMPARE_EQUAL;
1513 case PIPE_FUNC_LEQUAL:
1514 return V_03C000_SQ_TEX_DEPTH_COMPARE_LESSEQUAL;
1515 case PIPE_FUNC_GREATER:
1516 return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATER;
1517 case PIPE_FUNC_NOTEQUAL:
1518 return V_03C000_SQ_TEX_DEPTH_COMPARE_NOTEQUAL;
1519 case PIPE_FUNC_GEQUAL:
1520 return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL;
1521 case PIPE_FUNC_ALWAYS:
1522 return V_03C000_SQ_TEX_DEPTH_COMPARE_ALWAYS;
1523 }
1524 }