i965: enable ARB_instanced_arrays extension
[mesa.git] / src / mesa / drivers / dri / i965 / brw_draw.c
1 /**************************************************************************
2 *
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include <sys/errno.h>
29
30 #include "main/glheader.h"
31 #include "main/context.h"
32 #include "main/condrender.h"
33 #include "main/samplerobj.h"
34 #include "main/state.h"
35 #include "main/enums.h"
36 #include "main/macros.h"
37 #include "tnl/tnl.h"
38 #include "vbo/vbo_context.h"
39 #include "swrast/swrast.h"
40 #include "swrast_setup/swrast_setup.h"
41 #include "drivers/common/meta.h"
42
43 #include "brw_draw.h"
44 #include "brw_defines.h"
45 #include "brw_context.h"
46 #include "brw_state.h"
47
48 #include "intel_batchbuffer.h"
49 #include "intel_fbo.h"
50 #include "intel_mipmap_tree.h"
51 #include "intel_regions.h"
52
53 #define FILE_DEBUG_FLAG DEBUG_PRIMS
54
55 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
56 _3DPRIM_POINTLIST,
57 _3DPRIM_LINELIST,
58 _3DPRIM_LINELOOP,
59 _3DPRIM_LINESTRIP,
60 _3DPRIM_TRILIST,
61 _3DPRIM_TRISTRIP,
62 _3DPRIM_TRIFAN,
63 _3DPRIM_QUADLIST,
64 _3DPRIM_QUADSTRIP,
65 _3DPRIM_POLYGON
66 };
67
68
69 static const GLenum reduced_prim[GL_POLYGON+1] = {
70 GL_POINTS,
71 GL_LINES,
72 GL_LINES,
73 GL_LINES,
74 GL_TRIANGLES,
75 GL_TRIANGLES,
76 GL_TRIANGLES,
77 GL_TRIANGLES,
78 GL_TRIANGLES,
79 GL_TRIANGLES
80 };
81
82
83 /* When the primitive changes, set a state bit and re-validate. Not
84 * the nicest and would rather deal with this by having all the
85 * programs be immune to the active primitive (ie. cope with all
86 * possibilities). That may not be realistic however.
87 */
88 static void brw_set_prim(struct brw_context *brw,
89 const struct _mesa_prim *prim)
90 {
91 struct gl_context *ctx = &brw->intel.ctx;
92 uint32_t hw_prim = prim_to_hw_prim[prim->mode];
93
94 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
95
96 /* Slight optimization to avoid the GS program when not needed:
97 */
98 if (prim->mode == GL_QUAD_STRIP &&
99 ctx->Light.ShadeModel != GL_FLAT &&
100 ctx->Polygon.FrontMode == GL_FILL &&
101 ctx->Polygon.BackMode == GL_FILL)
102 hw_prim = _3DPRIM_TRISTRIP;
103
104 if (prim->mode == GL_QUADS && prim->count == 4 &&
105 ctx->Light.ShadeModel != GL_FLAT &&
106 ctx->Polygon.FrontMode == GL_FILL &&
107 ctx->Polygon.BackMode == GL_FILL) {
108 hw_prim = _3DPRIM_TRIFAN;
109 }
110
111 if (hw_prim != brw->primitive) {
112 brw->primitive = hw_prim;
113 brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
114
115 if (reduced_prim[prim->mode] != brw->intel.reduced_primitive) {
116 brw->intel.reduced_primitive = reduced_prim[prim->mode];
117 brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
118 }
119 }
120 }
121
122 static void gen6_set_prim(struct brw_context *brw,
123 const struct _mesa_prim *prim)
124 {
125 uint32_t hw_prim;
126
127 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
128
129 hw_prim = prim_to_hw_prim[prim->mode];
130
131 if (hw_prim != brw->primitive) {
132 brw->primitive = hw_prim;
133 brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
134 }
135 }
136
137
138 static GLuint trim(GLenum prim, GLuint length)
139 {
140 if (prim == GL_QUAD_STRIP)
141 return length > 3 ? (length - length % 2) : 0;
142 else if (prim == GL_QUADS)
143 return length - length % 4;
144 else
145 return length;
146 }
147
148
149 static void brw_emit_prim(struct brw_context *brw,
150 const struct _mesa_prim *prim,
151 uint32_t hw_prim)
152 {
153 struct intel_context *intel = &brw->intel;
154 int verts_per_instance;
155 int vertex_access_type;
156 int start_vertex_location;
157 int base_vertex_location;
158
159 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
160 prim->start, prim->count);
161
162 start_vertex_location = prim->start;
163 base_vertex_location = prim->basevertex;
164 if (prim->indexed) {
165 vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
166 start_vertex_location += brw->ib.start_vertex_offset;
167 base_vertex_location += brw->vb.start_vertex_bias;
168 } else {
169 vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
170 start_vertex_location += brw->vb.start_vertex_bias;
171 }
172
173 verts_per_instance = trim(prim->mode, prim->count);
174
175 /* If nothing to emit, just return. */
176 if (verts_per_instance == 0)
177 return;
178
179 /* If we're set to always flush, do it before and after the primitive emit.
180 * We want to catch both missed flushes that hurt instruction/state cache
181 * and missed flushes of the render cache as it heads to other parts of
182 * the besides the draw code.
183 */
184 if (intel->always_flush_cache) {
185 intel_batchbuffer_emit_mi_flush(intel);
186 }
187
188 BEGIN_BATCH(6);
189 OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
190 hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
191 vertex_access_type);
192 OUT_BATCH(verts_per_instance);
193 OUT_BATCH(start_vertex_location);
194 OUT_BATCH(prim->num_instances);
195 OUT_BATCH(0); // start instance location
196 OUT_BATCH(base_vertex_location);
197 ADVANCE_BATCH();
198
199 intel->batch.need_workaround_flush = true;
200
201 if (intel->always_flush_cache) {
202 intel_batchbuffer_emit_mi_flush(intel);
203 }
204 }
205
206 static void gen7_emit_prim(struct brw_context *brw,
207 const struct _mesa_prim *prim,
208 uint32_t hw_prim)
209 {
210 struct intel_context *intel = &brw->intel;
211 int verts_per_instance;
212 int vertex_access_type;
213 int start_vertex_location;
214 int base_vertex_location;
215
216 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
217 prim->start, prim->count);
218
219 start_vertex_location = prim->start;
220 base_vertex_location = prim->basevertex;
221 if (prim->indexed) {
222 vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
223 start_vertex_location += brw->ib.start_vertex_offset;
224 base_vertex_location += brw->vb.start_vertex_bias;
225 } else {
226 vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
227 start_vertex_location += brw->vb.start_vertex_bias;
228 }
229
230 verts_per_instance = trim(prim->mode, prim->count);
231
232 /* If nothing to emit, just return. */
233 if (verts_per_instance == 0)
234 return;
235
236 /* If we're set to always flush, do it before and after the primitive emit.
237 * We want to catch both missed flushes that hurt instruction/state cache
238 * and missed flushes of the render cache as it heads to other parts of
239 * the besides the draw code.
240 */
241 if (intel->always_flush_cache) {
242 intel_batchbuffer_emit_mi_flush(intel);
243 }
244
245 BEGIN_BATCH(7);
246 OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2));
247 OUT_BATCH(hw_prim | vertex_access_type);
248 OUT_BATCH(verts_per_instance);
249 OUT_BATCH(start_vertex_location);
250 OUT_BATCH(prim->num_instances);
251 OUT_BATCH(0); // start instance location
252 OUT_BATCH(base_vertex_location);
253 ADVANCE_BATCH();
254
255 if (intel->always_flush_cache) {
256 intel_batchbuffer_emit_mi_flush(intel);
257 }
258 }
259
260
261 static void brw_merge_inputs( struct brw_context *brw,
262 const struct gl_client_array *arrays[])
263 {
264 struct brw_vertex_info old = brw->vb.info;
265 GLuint i;
266
267 for (i = 0; i < brw->vb.nr_buffers; i++) {
268 drm_intel_bo_unreference(brw->vb.buffers[i].bo);
269 brw->vb.buffers[i].bo = NULL;
270 }
271 brw->vb.nr_buffers = 0;
272
273 memset(&brw->vb.info, 0, sizeof(brw->vb.info));
274
275 for (i = 0; i < VERT_ATTRIB_MAX; i++) {
276 brw->vb.inputs[i].buffer = -1;
277 brw->vb.inputs[i].glarray = arrays[i];
278 brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
279
280 if (arrays[i]->StrideB != 0)
281 brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
282 ((i%16) * 2);
283 }
284
285 /* Raise statechanges if input sizes have changed. */
286 if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
287 brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
288 }
289
290 /*
291 * \brief Resolve buffers before drawing.
292 *
293 * Resolve the depth buffer's HiZ buffer and resolve the depth buffer of each
294 * enabled depth texture.
295 *
296 * (In the future, this will also perform MSAA resolves).
297 */
298 static void
299 brw_predraw_resolve_buffers(struct brw_context *brw)
300 {
301 struct gl_context *ctx = &brw->intel.ctx;
302 struct intel_context *intel = &brw->intel;
303 struct intel_renderbuffer *depth_irb;
304 struct intel_texture_object *tex_obj;
305
306 /* Resolve the depth buffer's HiZ buffer. */
307 depth_irb = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
308 if (depth_irb && depth_irb->mt) {
309 intel_renderbuffer_resolve_hiz(intel, depth_irb);
310 }
311
312 /* Resolve depth buffer of each enabled depth texture. */
313 for (int i = 0; i < BRW_MAX_TEX_UNIT; i++) {
314 if (!ctx->Texture.Unit[i]._ReallyEnabled)
315 continue;
316 tex_obj = intel_texture_object(ctx->Texture.Unit[i]._Current);
317 if (!tex_obj || !tex_obj->mt)
318 continue;
319 intel_miptree_all_slices_resolve_depth(intel, tex_obj->mt);
320 }
321 }
322
323 /**
324 * \brief Call this after drawing to mark which buffers need resolving
325 *
326 * If the depth buffer was written to and if it has an accompanying HiZ
327 * buffer, then mark that it needs a depth resolve.
328 *
329 * (In the future, this will also mark needed MSAA resolves).
330 */
331 static void brw_postdraw_set_buffers_need_resolve(struct brw_context *brw)
332 {
333 struct gl_context *ctx = &brw->intel.ctx;
334 struct gl_framebuffer *fb = ctx->DrawBuffer;
335 struct intel_renderbuffer *depth_irb =
336 intel_get_renderbuffer(fb, BUFFER_DEPTH);
337
338 if (depth_irb && ctx->Depth.Mask) {
339 intel_renderbuffer_set_needs_depth_resolve(depth_irb);
340 }
341 }
342
343 static int
344 verts_per_prim(GLenum mode)
345 {
346 switch (mode) {
347 case GL_POINTS:
348 return 1;
349 case GL_LINE_STRIP:
350 case GL_LINE_LOOP:
351 case GL_LINES:
352 return 2;
353 case GL_TRIANGLE_STRIP:
354 case GL_TRIANGLE_FAN:
355 case GL_POLYGON:
356 case GL_TRIANGLES:
357 case GL_QUADS:
358 case GL_QUAD_STRIP:
359 return 3;
360 default:
361 _mesa_problem(NULL,
362 "unknown prim type in transform feedback primitive count");
363 return 0;
364 }
365 }
366
367 /**
368 * Update internal counters based on the the drawing operation described in
369 * prim.
370 */
371 static void
372 brw_update_primitive_count(struct brw_context *brw,
373 const struct _mesa_prim *prim)
374 {
375 uint32_t count = count_tessellated_primitives(prim);
376 brw->sol.primitives_generated += count;
377 if (brw->intel.ctx.TransformFeedback.CurrentObject->Active &&
378 !brw->intel.ctx.TransformFeedback.CurrentObject->Paused) {
379 /* Update brw->sol.svbi_0_max_index to reflect the amount by which the
380 * hardware is going to increment SVBI 0 when this drawing operation
381 * occurs. This is necessary because the kernel does not (yet) save and
382 * restore GPU registers when context switching, so we'll need to be
383 * able to reload SVBI 0 with the correct value in case we have to start
384 * a new batch buffer.
385 */
386 unsigned verts = verts_per_prim(prim->mode);
387 uint32_t space_avail =
388 (brw->sol.svbi_0_max_index - brw->sol.svbi_0_starting_index) / verts;
389 uint32_t primitives_written = MIN2 (space_avail, count);
390 brw->sol.svbi_0_starting_index += verts * primitives_written;
391
392 /* And update the TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN query. */
393 brw->sol.primitives_written += primitives_written;
394 }
395 }
396
397 /* May fail if out of video memory for texture or vbo upload, or on
398 * fallback conditions.
399 */
400 static bool brw_try_draw_prims( struct gl_context *ctx,
401 const struct gl_client_array *arrays[],
402 const struct _mesa_prim *prim,
403 GLuint nr_prims,
404 const struct _mesa_index_buffer *ib,
405 GLuint min_index,
406 GLuint max_index )
407 {
408 struct intel_context *intel = intel_context(ctx);
409 struct brw_context *brw = brw_context(ctx);
410 bool retval = true;
411 GLuint i;
412 bool fail_next = false;
413
414 if (ctx->NewState)
415 _mesa_update_state( ctx );
416
417 /* We have to validate the textures *before* checking for fallbacks;
418 * otherwise, the software fallback won't be able to rely on the
419 * texture state, the firstLevel and lastLevel fields won't be
420 * set in the intel texture object (they'll both be 0), and the
421 * software fallback will segfault if it attempts to access any
422 * texture level other than level 0.
423 */
424 brw_validate_textures( brw );
425
426 /* Resolves must occur after updating state and finalizing textures but
427 * before setting up any hardware state for this draw call.
428 */
429 brw_predraw_resolve_buffers(brw);
430
431 /* Bind all inputs, derive varying and size information:
432 */
433 brw_merge_inputs( brw, arrays );
434
435 brw->ib.ib = ib;
436 brw->state.dirty.brw |= BRW_NEW_INDICES;
437
438 brw->vb.min_index = min_index;
439 brw->vb.max_index = max_index;
440 brw->state.dirty.brw |= BRW_NEW_VERTICES;
441
442 /* Have to validate state quite late. Will rebuild tnl_program,
443 * which depends on varying information.
444 *
445 * Note this is where brw->vs->prog_data.inputs_read is calculated,
446 * so can't access it earlier.
447 */
448
449 intel_prepare_render(intel);
450
451 for (i = 0; i < nr_prims; i++) {
452 int estimated_max_prim_size;
453
454 estimated_max_prim_size = 512; /* batchbuffer commands */
455 estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
456 (sizeof(struct brw_sampler_state) +
457 sizeof(struct gen5_sampler_default_color)));
458 estimated_max_prim_size += 1024; /* gen6 VS push constants */
459 estimated_max_prim_size += 1024; /* gen6 WM push constants */
460 estimated_max_prim_size += 512; /* misc. pad */
461
462 /* Flush the batch if it's approaching full, so that we don't wrap while
463 * we've got validated state that needs to be in the same batch as the
464 * primitives.
465 */
466 intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
467 intel_batchbuffer_save_state(intel);
468
469 brw->num_instances = prim->num_instances;
470 if (intel->gen < 6)
471 brw_set_prim(brw, &prim[i]);
472 else
473 gen6_set_prim(brw, &prim[i]);
474
475 retry:
476 /* Note that before the loop, brw->state.dirty.brw was set to != 0, and
477 * that the state updated in the loop outside of this block is that in
478 * *_set_prim or intel_batchbuffer_flush(), which only impacts
479 * brw->state.dirty.brw.
480 */
481 if (brw->state.dirty.brw) {
482 intel->no_batch_wrap = true;
483 brw_upload_state(brw);
484
485 if (unlikely(brw->intel.Fallback)) {
486 intel->no_batch_wrap = false;
487 retval = false;
488 goto out;
489 }
490 }
491
492 if (intel->gen >= 7)
493 gen7_emit_prim(brw, &prim[i], brw->primitive);
494 else
495 brw_emit_prim(brw, &prim[i], brw->primitive);
496
497 intel->no_batch_wrap = false;
498
499 if (dri_bufmgr_check_aperture_space(&intel->batch.bo, 1)) {
500 if (!fail_next) {
501 intel_batchbuffer_reset_to_saved(intel);
502 intel_batchbuffer_flush(intel);
503 fail_next = true;
504 goto retry;
505 } else {
506 if (intel_batchbuffer_flush(intel) == -ENOSPC) {
507 static bool warned = false;
508
509 if (!warned) {
510 fprintf(stderr, "i965: Single primitive emit exceeded"
511 "available aperture space\n");
512 warned = true;
513 }
514
515 retval = false;
516 }
517 }
518 }
519
520 if (!_mesa_meta_in_progress(ctx))
521 brw_update_primitive_count(brw, &prim[i]);
522 }
523
524 if (intel->always_flush_batch)
525 intel_batchbuffer_flush(intel);
526 out:
527
528 brw_state_cache_check_size(brw);
529 brw_postdraw_set_buffers_need_resolve(brw);
530
531 return retval;
532 }
533
534 void brw_draw_prims( struct gl_context *ctx,
535 const struct _mesa_prim *prim,
536 GLuint nr_prims,
537 const struct _mesa_index_buffer *ib,
538 GLboolean index_bounds_valid,
539 GLuint min_index,
540 GLuint max_index,
541 struct gl_transform_feedback_object *tfb_vertcount )
542 {
543 const struct gl_client_array **arrays = ctx->Array._DrawArrays;
544 bool retval;
545
546 if (!_mesa_check_conditional_render(ctx))
547 return;
548
549 /* Handle primitive restart if needed */
550 if (brw_handle_primitive_restart(ctx, prim, nr_prims, ib)) {
551 /* The draw was handled, so we can exit now */
552 return;
553 }
554
555 if (!vbo_all_varyings_in_vbos(arrays)) {
556 if (!index_bounds_valid)
557 vbo_get_minmax_indices(ctx, prim, ib, &min_index, &max_index, nr_prims);
558
559 /* Decide if we want to rebase. If so we end up recursing once
560 * only into this function.
561 */
562 if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
563 vbo_rebase_prims(ctx, arrays,
564 prim, nr_prims,
565 ib, min_index, max_index,
566 brw_draw_prims );
567 return;
568 }
569 }
570
571 /* Make a first attempt at drawing:
572 */
573 retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
574
575 /* Otherwise, we really are out of memory. Pass the drawing
576 * command to the software tnl module and which will in turn call
577 * swrast to do the drawing.
578 */
579 if (!retval) {
580 _swsetup_Wakeup(ctx);
581 _tnl_wakeup(ctx);
582 _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
583 }
584
585 }
586
587 void brw_draw_init( struct brw_context *brw )
588 {
589 struct gl_context *ctx = &brw->intel.ctx;
590 struct vbo_context *vbo = vbo_context(ctx);
591 int i;
592
593 /* Register our drawing function:
594 */
595 vbo->draw_prims = brw_draw_prims;
596
597 for (i = 0; i < VERT_ATTRIB_MAX; i++)
598 brw->vb.inputs[i].buffer = -1;
599 brw->vb.nr_buffers = 0;
600 brw->vb.nr_enabled = 0;
601 }
602
603 void brw_draw_destroy( struct brw_context *brw )
604 {
605 int i;
606
607 for (i = 0; i < brw->vb.nr_buffers; i++) {
608 drm_intel_bo_unreference(brw->vb.buffers[i].bo);
609 brw->vb.buffers[i].bo = NULL;
610 }
611 brw->vb.nr_buffers = 0;
612
613 for (i = 0; i < brw->vb.nr_enabled; i++) {
614 brw->vb.enabled[i]->buffer = -1;
615 }
616 brw->vb.nr_enabled = 0;
617
618 drm_intel_bo_unreference(brw->ib.bo);
619 brw->ib.bo = NULL;
620 }