1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
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:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
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.
26 **************************************************************************/
28 #include <sys/errno.h>
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 "main/transformfeedback.h"
39 #include "vbo/vbo_context.h"
40 #include "swrast/swrast.h"
41 #include "swrast_setup/swrast_setup.h"
42 #include "drivers/common/meta.h"
44 #include "brw_blorp.h"
46 #include "brw_defines.h"
47 #include "brw_context.h"
48 #include "brw_state.h"
50 #include "intel_batchbuffer.h"
51 #include "intel_fbo.h"
52 #include "intel_mipmap_tree.h"
53 #include "intel_regions.h"
55 #define FILE_DEBUG_FLAG DEBUG_PRIMS
57 const GLuint prim_to_hw_prim
[GL_TRIANGLE_STRIP_ADJACENCY
+1] = {
69 _3DPRIM_LINESTRIP_ADJ
,
75 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
89 /* When the primitive changes, set a state bit and re-validate. Not
90 * the nicest and would rather deal with this by having all the
91 * programs be immune to the active primitive (ie. cope with all
92 * possibilities). That may not be realistic however.
94 static void brw_set_prim(struct brw_context
*brw
,
95 const struct _mesa_prim
*prim
)
97 struct gl_context
*ctx
= &brw
->ctx
;
98 uint32_t hw_prim
= prim_to_hw_prim
[prim
->mode
];
100 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
102 /* Slight optimization to avoid the GS program when not needed:
104 if (prim
->mode
== GL_QUAD_STRIP
&&
105 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
106 ctx
->Polygon
.FrontMode
== GL_FILL
&&
107 ctx
->Polygon
.BackMode
== GL_FILL
)
108 hw_prim
= _3DPRIM_TRISTRIP
;
110 if (prim
->mode
== GL_QUADS
&& prim
->count
== 4 &&
111 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
112 ctx
->Polygon
.FrontMode
== GL_FILL
&&
113 ctx
->Polygon
.BackMode
== GL_FILL
) {
114 hw_prim
= _3DPRIM_TRIFAN
;
117 if (hw_prim
!= brw
->primitive
) {
118 brw
->primitive
= hw_prim
;
119 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
121 if (reduced_prim
[prim
->mode
] != brw
->reduced_primitive
) {
122 brw
->reduced_primitive
= reduced_prim
[prim
->mode
];
123 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
128 static void gen6_set_prim(struct brw_context
*brw
,
129 const struct _mesa_prim
*prim
)
133 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
135 hw_prim
= prim_to_hw_prim
[prim
->mode
];
137 if (hw_prim
!= brw
->primitive
) {
138 brw
->primitive
= hw_prim
;
139 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
145 * The hardware is capable of removing dangling vertices on its own; however,
146 * prior to Gen6, we sometimes convert quads into trifans (and quad strips
147 * into tristrips), since pre-Gen6 hardware requires a GS to render quads.
148 * This function manually trims dangling vertices from a draw call involving
149 * quads so that those dangling vertices won't get drawn when we convert to
152 static GLuint
trim(GLenum prim
, GLuint length
)
154 if (prim
== GL_QUAD_STRIP
)
155 return length
> 3 ? (length
- length
% 2) : 0;
156 else if (prim
== GL_QUADS
)
157 return length
- length
% 4;
163 static void brw_emit_prim(struct brw_context
*brw
,
164 const struct _mesa_prim
*prim
,
167 int verts_per_instance
;
168 int vertex_access_type
;
169 int start_vertex_location
;
170 int base_vertex_location
;
172 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
173 prim
->start
, prim
->count
);
175 start_vertex_location
= prim
->start
;
176 base_vertex_location
= prim
->basevertex
;
178 vertex_access_type
= brw
->gen
>= 7 ?
179 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
:
180 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
;
181 start_vertex_location
+= brw
->ib
.start_vertex_offset
;
182 base_vertex_location
+= brw
->vb
.start_vertex_bias
;
184 vertex_access_type
= brw
->gen
>= 7 ?
185 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
:
186 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
;
187 start_vertex_location
+= brw
->vb
.start_vertex_bias
;
190 /* We only need to trim the primitive count on pre-Gen6. */
192 verts_per_instance
= trim(prim
->mode
, prim
->count
);
194 verts_per_instance
= prim
->count
;
196 /* If nothing to emit, just return. */
197 if (verts_per_instance
== 0)
200 /* If we're set to always flush, do it before and after the primitive emit.
201 * We want to catch both missed flushes that hurt instruction/state cache
202 * and missed flushes of the render cache as it heads to other parts of
203 * the besides the draw code.
205 if (brw
->always_flush_cache
) {
206 intel_batchbuffer_emit_mi_flush(brw
);
211 OUT_BATCH(CMD_3D_PRIM
<< 16 | (7 - 2));
212 OUT_BATCH(hw_prim
| vertex_access_type
);
215 OUT_BATCH(CMD_3D_PRIM
<< 16 | (6 - 2) |
216 hw_prim
<< GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT
|
219 OUT_BATCH(verts_per_instance
);
220 OUT_BATCH(start_vertex_location
);
221 OUT_BATCH(prim
->num_instances
);
222 OUT_BATCH(prim
->base_instance
);
223 OUT_BATCH(base_vertex_location
);
226 /* Only used on Sandybridge; harmless to set elsewhere. */
227 brw
->batch
.need_workaround_flush
= true;
229 if (brw
->always_flush_cache
) {
230 intel_batchbuffer_emit_mi_flush(brw
);
235 static void brw_merge_inputs( struct brw_context
*brw
,
236 const struct gl_client_array
*arrays
[])
240 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
241 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
242 brw
->vb
.buffers
[i
].bo
= NULL
;
244 brw
->vb
.nr_buffers
= 0;
246 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
247 brw
->vb
.inputs
[i
].buffer
= -1;
248 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
249 brw
->vb
.inputs
[i
].attrib
= (gl_vert_attrib
) i
;
254 * \brief Resolve buffers before drawing.
256 * Resolve the depth buffer's HiZ buffer and resolve the depth buffer of each
257 * enabled depth texture.
259 * (In the future, this will also perform MSAA resolves).
262 brw_predraw_resolve_buffers(struct brw_context
*brw
)
264 struct gl_context
*ctx
= &brw
->ctx
;
265 struct intel_renderbuffer
*depth_irb
;
266 struct intel_texture_object
*tex_obj
;
268 /* Resolve the depth buffer's HiZ buffer. */
269 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
271 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
273 /* Resolve depth buffer of each enabled depth texture, and color buffer of
274 * each fast-clear-enabled color texture.
276 for (int i
= 0; i
< BRW_MAX_TEX_UNIT
; i
++) {
277 if (!ctx
->Texture
.Unit
[i
]._ReallyEnabled
)
279 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
280 if (!tex_obj
|| !tex_obj
->mt
)
282 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
283 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
288 * \brief Call this after drawing to mark which buffers need resolving
290 * If the depth buffer was written to and if it has an accompanying HiZ
291 * buffer, then mark that it needs a depth resolve.
293 * If the color buffer is a multisample window system buffer, then
294 * mark that it needs a downsample.
296 static void brw_postdraw_set_buffers_need_resolve(struct brw_context
*brw
)
298 struct gl_context
*ctx
= &brw
->ctx
;
299 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
301 struct intel_renderbuffer
*front_irb
= NULL
;
302 struct intel_renderbuffer
*back_irb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
303 struct intel_renderbuffer
*depth_irb
= intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
304 struct gl_renderbuffer_attachment
*depth_att
= &fb
->Attachment
[BUFFER_DEPTH
];
306 if (brw
->is_front_buffer_rendering
)
307 front_irb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
310 intel_renderbuffer_set_needs_downsample(front_irb
);
312 intel_renderbuffer_set_needs_downsample(back_irb
);
313 if (depth_irb
&& ctx
->Depth
.Mask
)
314 intel_renderbuffer_att_set_needs_depth_resolve(depth_att
);
317 /* May fail if out of video memory for texture or vbo upload, or on
318 * fallback conditions.
320 static bool brw_try_draw_prims( struct gl_context
*ctx
,
321 const struct gl_client_array
*arrays
[],
322 const struct _mesa_prim
*prims
,
324 const struct _mesa_index_buffer
*ib
,
328 struct brw_context
*brw
= brw_context(ctx
);
331 bool fail_next
= false;
334 _mesa_update_state( ctx
);
336 /* Find the highest sampler unit used by each shader program. A bit-count
337 * won't work since ARB programs use the texture unit number as the sampler
340 brw
->wm
.base
.sampler_count
=
341 _mesa_fls(ctx
->FragmentProgram
._Current
->Base
.SamplersUsed
);
342 brw
->gs
.base
.sampler_count
= ctx
->GeometryProgram
._Current
?
343 _mesa_fls(ctx
->GeometryProgram
._Current
->Base
.SamplersUsed
) : 0;
344 brw
->vs
.base
.sampler_count
=
345 _mesa_fls(ctx
->VertexProgram
._Current
->Base
.SamplersUsed
);
347 /* We have to validate the textures *before* checking for fallbacks;
348 * otherwise, the software fallback won't be able to rely on the
349 * texture state, the firstLevel and lastLevel fields won't be
350 * set in the intel texture object (they'll both be 0), and the
351 * software fallback will segfault if it attempts to access any
352 * texture level other than level 0.
354 brw_validate_textures( brw
);
356 intel_prepare_render(brw
);
358 /* This workaround has to happen outside of brw_upload_state() because it
359 * may flush the batchbuffer for a blit, affecting the state flags.
361 brw_workaround_depthstencil_alignment(brw
, 0);
363 /* Resolves must occur after updating renderbuffers, updating context state,
364 * and finalizing textures but before setting up any hardware state for
367 brw_predraw_resolve_buffers(brw
);
369 /* Bind all inputs, derive varying and size information:
371 brw_merge_inputs( brw
, arrays
);
374 brw
->state
.dirty
.brw
|= BRW_NEW_INDICES
;
376 brw
->vb
.min_index
= min_index
;
377 brw
->vb
.max_index
= max_index
;
378 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
380 for (i
= 0; i
< nr_prims
; i
++) {
381 int estimated_max_prim_size
;
383 estimated_max_prim_size
= 512; /* batchbuffer commands */
384 estimated_max_prim_size
+= (BRW_MAX_TEX_UNIT
*
385 (sizeof(struct brw_sampler_state
) +
386 sizeof(struct gen5_sampler_default_color
)));
387 estimated_max_prim_size
+= 1024; /* gen6 VS push constants */
388 estimated_max_prim_size
+= 1024; /* gen6 WM push constants */
389 estimated_max_prim_size
+= 512; /* misc. pad */
391 /* Flush the batch if it's approaching full, so that we don't wrap while
392 * we've got validated state that needs to be in the same batch as the
395 intel_batchbuffer_require_space(brw
, estimated_max_prim_size
, false);
396 intel_batchbuffer_save_state(brw
);
398 if (brw
->num_instances
!= prims
[i
].num_instances
) {
399 brw
->num_instances
= prims
[i
].num_instances
;
400 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
402 if (brw
->basevertex
!= prims
[i
].basevertex
) {
403 brw
->basevertex
= prims
[i
].basevertex
;
404 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
407 brw_set_prim(brw
, &prims
[i
]);
409 gen6_set_prim(brw
, &prims
[i
]);
412 /* Note that before the loop, brw->state.dirty.brw was set to != 0, and
413 * that the state updated in the loop outside of this block is that in
414 * *_set_prim or intel_batchbuffer_flush(), which only impacts
415 * brw->state.dirty.brw.
417 if (brw
->state
.dirty
.brw
) {
418 brw
->no_batch_wrap
= true;
419 brw_upload_state(brw
);
422 brw_emit_prim(brw
, &prims
[i
], brw
->primitive
);
424 brw
->no_batch_wrap
= false;
426 if (dri_bufmgr_check_aperture_space(&brw
->batch
.bo
, 1)) {
428 intel_batchbuffer_reset_to_saved(brw
);
429 intel_batchbuffer_flush(brw
);
433 if (intel_batchbuffer_flush(brw
) == -ENOSPC
) {
434 static bool warned
= false;
437 fprintf(stderr
, "i965: Single primitive emit exceeded"
438 "available aperture space\n");
448 if (brw
->always_flush_batch
)
449 intel_batchbuffer_flush(brw
);
451 brw_state_cache_check_size(brw
);
452 brw_postdraw_set_buffers_need_resolve(brw
);
457 void brw_draw_prims( struct gl_context
*ctx
,
458 const struct _mesa_prim
*prims
,
460 const struct _mesa_index_buffer
*ib
,
461 GLboolean index_bounds_valid
,
464 struct gl_transform_feedback_object
*tfb_vertcount
)
466 struct brw_context
*brw
= brw_context(ctx
);
467 const struct gl_client_array
**arrays
= ctx
->Array
._DrawArrays
;
469 if (!_mesa_check_conditional_render(ctx
))
472 /* Handle primitive restart if needed */
473 if (brw_handle_primitive_restart(ctx
, prims
, nr_prims
, ib
)) {
474 /* The draw was handled, so we can exit now */
478 /* If we're going to have to upload any of the user's vertex arrays, then
479 * get the minimum and maximum of their index buffer so we know what range
482 if (!vbo_all_varyings_in_vbos(arrays
) && !index_bounds_valid
)
483 vbo_get_minmax_indices(ctx
, prims
, ib
, &min_index
, &max_index
, nr_prims
);
485 /* Do GL_SELECT and GL_FEEDBACK rendering using swrast, even though it
486 * won't support all the extensions we support.
488 if (ctx
->RenderMode
!= GL_RENDER
) {
489 perf_debug("%s render mode not supported in hardware\n",
490 _mesa_lookup_enum_by_nr(ctx
->RenderMode
));
491 _swsetup_Wakeup(ctx
);
493 _tnl_draw_prims(ctx
, arrays
, prims
, nr_prims
, ib
, min_index
, max_index
);
497 /* Try drawing with the hardware, but don't do anything else if we can't
498 * manage it. swrast doesn't support our featureset, so we can't fall back
501 brw_try_draw_prims(ctx
, arrays
, prims
, nr_prims
, ib
, min_index
, max_index
);
504 void brw_draw_init( struct brw_context
*brw
)
506 struct gl_context
*ctx
= &brw
->ctx
;
507 struct vbo_context
*vbo
= vbo_context(ctx
);
510 /* Register our drawing function:
512 vbo
->draw_prims
= brw_draw_prims
;
514 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
515 brw
->vb
.inputs
[i
].buffer
= -1;
516 brw
->vb
.nr_buffers
= 0;
517 brw
->vb
.nr_enabled
= 0;
520 void brw_draw_destroy( struct brw_context
*brw
)
524 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
525 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
526 brw
->vb
.buffers
[i
].bo
= NULL
;
528 brw
->vb
.nr_buffers
= 0;
530 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
531 brw
->vb
.enabled
[i
]->buffer
= -1;
533 brw
->vb
.nr_enabled
= 0;
535 drm_intel_bo_unreference(brw
->ib
.bo
);