1 /**************************************************************************
3 * Copyright 2003 VMware, Inc.
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 VMWARE 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"
51 #include "intel_batchbuffer.h"
52 #include "intel_buffers.h"
53 #include "intel_fbo.h"
54 #include "intel_mipmap_tree.h"
55 #include "intel_buffer_objects.h"
57 #define FILE_DEBUG_FLAG DEBUG_PRIMS
59 static const GLuint prim_to_hw_prim
[GL_TRIANGLE_STRIP_ADJACENCY
+1] = {
71 _3DPRIM_LINESTRIP_ADJ
,
77 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
91 get_hw_prim_for_gl_prim(int mode
)
93 if (mode
>= BRW_PRIM_OFFSET
)
94 return mode
- BRW_PRIM_OFFSET
;
96 return prim_to_hw_prim
[mode
];
100 /* When the primitive changes, set a state bit and re-validate. Not
101 * the nicest and would rather deal with this by having all the
102 * programs be immune to the active primitive (ie. cope with all
103 * possibilities). That may not be realistic however.
105 static void brw_set_prim(struct brw_context
*brw
,
106 const struct _mesa_prim
*prim
)
108 struct gl_context
*ctx
= &brw
->ctx
;
109 uint32_t hw_prim
= get_hw_prim_for_gl_prim(prim
->mode
);
111 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
113 /* Slight optimization to avoid the GS program when not needed:
115 if (prim
->mode
== GL_QUAD_STRIP
&&
116 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
117 ctx
->Polygon
.FrontMode
== GL_FILL
&&
118 ctx
->Polygon
.BackMode
== GL_FILL
)
119 hw_prim
= _3DPRIM_TRISTRIP
;
121 if (prim
->mode
== GL_QUADS
&& prim
->count
== 4 &&
122 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
123 ctx
->Polygon
.FrontMode
== GL_FILL
&&
124 ctx
->Polygon
.BackMode
== GL_FILL
) {
125 hw_prim
= _3DPRIM_TRIFAN
;
128 if (hw_prim
!= brw
->primitive
) {
129 brw
->primitive
= hw_prim
;
130 brw
->ctx
.NewDriverState
|= BRW_NEW_PRIMITIVE
;
132 if (reduced_prim
[prim
->mode
] != brw
->reduced_primitive
) {
133 brw
->reduced_primitive
= reduced_prim
[prim
->mode
];
134 brw
->ctx
.NewDriverState
|= BRW_NEW_REDUCED_PRIMITIVE
;
139 static void gen6_set_prim(struct brw_context
*brw
,
140 const struct _mesa_prim
*prim
)
144 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
146 hw_prim
= get_hw_prim_for_gl_prim(prim
->mode
);
148 if (hw_prim
!= brw
->primitive
) {
149 brw
->primitive
= hw_prim
;
150 brw
->ctx
.NewDriverState
|= BRW_NEW_PRIMITIVE
;
156 * The hardware is capable of removing dangling vertices on its own; however,
157 * prior to Gen6, we sometimes convert quads into trifans (and quad strips
158 * into tristrips), since pre-Gen6 hardware requires a GS to render quads.
159 * This function manually trims dangling vertices from a draw call involving
160 * quads so that those dangling vertices won't get drawn when we convert to
163 static GLuint
trim(GLenum prim
, GLuint length
)
165 if (prim
== GL_QUAD_STRIP
)
166 return length
> 3 ? (length
- length
% 2) : 0;
167 else if (prim
== GL_QUADS
)
168 return length
- length
% 4;
174 static void brw_emit_prim(struct brw_context
*brw
,
175 const struct _mesa_prim
*prim
,
178 int verts_per_instance
;
179 int vertex_access_type
;
181 int predicate_enable
;
183 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
184 prim
->start
, prim
->count
);
186 int start_vertex_location
= prim
->start
;
187 int base_vertex_location
= prim
->basevertex
;
190 vertex_access_type
= brw
->gen
>= 7 ?
191 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
:
192 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
;
193 start_vertex_location
+= brw
->ib
.start_vertex_offset
;
194 base_vertex_location
+= brw
->vb
.start_vertex_bias
;
196 vertex_access_type
= brw
->gen
>= 7 ?
197 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
:
198 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
;
199 start_vertex_location
+= brw
->vb
.start_vertex_bias
;
202 /* We only need to trim the primitive count on pre-Gen6. */
204 verts_per_instance
= trim(prim
->mode
, prim
->count
);
206 verts_per_instance
= prim
->count
;
208 /* If nothing to emit, just return. */
209 if (verts_per_instance
== 0 && !prim
->is_indirect
)
212 /* If we're set to always flush, do it before and after the primitive emit.
213 * We want to catch both missed flushes that hurt instruction/state cache
214 * and missed flushes of the render cache as it heads to other parts of
215 * the besides the draw code.
217 if (brw
->always_flush_cache
) {
218 intel_batchbuffer_emit_mi_flush(brw
);
221 /* If indirect, emit a bunch of loads from the indirect BO. */
222 if (prim
->is_indirect
) {
223 struct gl_buffer_object
*indirect_buffer
= brw
->ctx
.DrawIndirectBuffer
;
224 drm_intel_bo
*bo
= intel_bufferobj_buffer(brw
,
225 intel_buffer_object(indirect_buffer
),
226 prim
->indirect_offset
, 5 * sizeof(GLuint
));
228 indirect_flag
= GEN7_3DPRIM_INDIRECT_PARAMETER_ENABLE
;
230 brw_load_register_mem(brw
, GEN7_3DPRIM_VERTEX_COUNT
, bo
,
231 I915_GEM_DOMAIN_VERTEX
, 0,
232 prim
->indirect_offset
+ 0);
233 brw_load_register_mem(brw
, GEN7_3DPRIM_INSTANCE_COUNT
, bo
,
234 I915_GEM_DOMAIN_VERTEX
, 0,
235 prim
->indirect_offset
+ 4);
237 brw_load_register_mem(brw
, GEN7_3DPRIM_START_VERTEX
, bo
,
238 I915_GEM_DOMAIN_VERTEX
, 0,
239 prim
->indirect_offset
+ 8);
241 brw_load_register_mem(brw
, GEN7_3DPRIM_BASE_VERTEX
, bo
,
242 I915_GEM_DOMAIN_VERTEX
, 0,
243 prim
->indirect_offset
+ 12);
244 brw_load_register_mem(brw
, GEN7_3DPRIM_START_INSTANCE
, bo
,
245 I915_GEM_DOMAIN_VERTEX
, 0,
246 prim
->indirect_offset
+ 16);
248 brw_load_register_mem(brw
, GEN7_3DPRIM_START_INSTANCE
, bo
,
249 I915_GEM_DOMAIN_VERTEX
, 0,
250 prim
->indirect_offset
+ 12);
252 OUT_BATCH(MI_LOAD_REGISTER_IMM
| (3 - 2));
253 OUT_BATCH(GEN7_3DPRIM_BASE_VERTEX
);
263 if (brw
->predicate
.state
== BRW_PREDICATE_STATE_USE_BIT
)
264 predicate_enable
= GEN7_3DPRIM_PREDICATE_ENABLE
;
266 predicate_enable
= 0;
269 OUT_BATCH(CMD_3D_PRIM
<< 16 | (7 - 2) | indirect_flag
| predicate_enable
);
270 OUT_BATCH(hw_prim
| vertex_access_type
);
273 OUT_BATCH(CMD_3D_PRIM
<< 16 | (6 - 2) |
274 hw_prim
<< GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT
|
277 OUT_BATCH(verts_per_instance
);
278 OUT_BATCH(start_vertex_location
);
279 OUT_BATCH(prim
->num_instances
);
280 OUT_BATCH(prim
->base_instance
);
281 OUT_BATCH(base_vertex_location
);
284 if (brw
->always_flush_cache
) {
285 intel_batchbuffer_emit_mi_flush(brw
);
290 static void brw_merge_inputs( struct brw_context
*brw
,
291 const struct gl_client_array
*arrays
[])
293 const struct gl_context
*ctx
= &brw
->ctx
;
296 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
297 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
298 brw
->vb
.buffers
[i
].bo
= NULL
;
300 brw
->vb
.nr_buffers
= 0;
302 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
303 brw
->vb
.inputs
[i
].buffer
= -1;
304 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
307 if (brw
->gen
< 8 && !brw
->is_haswell
) {
308 struct gl_program
*vp
= &ctx
->VertexProgram
._Current
->Base
;
309 /* Prior to Haswell, the hardware can't natively support GL_FIXED or
310 * 2_10_10_10_REV vertex formats. Set appropriate workaround flags.
312 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
313 if (!(vp
->InputsRead
& BITFIELD64_BIT(i
)))
316 uint8_t wa_flags
= 0;
318 switch (brw
->vb
.inputs
[i
].glarray
->Type
) {
321 wa_flags
= brw
->vb
.inputs
[i
].glarray
->Size
;
324 case GL_INT_2_10_10_10_REV
:
325 wa_flags
|= BRW_ATTRIB_WA_SIGN
;
328 case GL_UNSIGNED_INT_2_10_10_10_REV
:
329 if (brw
->vb
.inputs
[i
].glarray
->Format
== GL_BGRA
)
330 wa_flags
|= BRW_ATTRIB_WA_BGRA
;
332 if (brw
->vb
.inputs
[i
].glarray
->Normalized
)
333 wa_flags
|= BRW_ATTRIB_WA_NORMALIZE
;
334 else if (!brw
->vb
.inputs
[i
].glarray
->Integer
)
335 wa_flags
|= BRW_ATTRIB_WA_SCALE
;
340 if (brw
->vb
.attrib_wa_flags
[i
] != wa_flags
) {
341 brw
->vb
.attrib_wa_flags
[i
] = wa_flags
;
342 brw
->ctx
.NewDriverState
|= BRW_NEW_VS_ATTRIB_WORKAROUNDS
;
349 * \brief Call this after drawing to mark which buffers need resolving
351 * If the depth buffer was written to and if it has an accompanying HiZ
352 * buffer, then mark that it needs a depth resolve.
354 * If the color buffer is a multisample window system buffer, then
355 * mark that it needs a downsample.
357 * Also mark any render targets which will be textured as needing a render
360 static void brw_postdraw_set_buffers_need_resolve(struct brw_context
*brw
)
362 struct gl_context
*ctx
= &brw
->ctx
;
363 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
365 struct intel_renderbuffer
*front_irb
= NULL
;
366 struct intel_renderbuffer
*back_irb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
367 struct intel_renderbuffer
*depth_irb
= intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
368 struct intel_renderbuffer
*stencil_irb
= intel_get_renderbuffer(fb
, BUFFER_STENCIL
);
369 struct gl_renderbuffer_attachment
*depth_att
= &fb
->Attachment
[BUFFER_DEPTH
];
371 if (brw_is_front_buffer_drawing(fb
))
372 front_irb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
375 front_irb
->need_downsample
= true;
377 back_irb
->need_downsample
= true;
378 if (depth_irb
&& ctx
->Depth
.Mask
) {
379 intel_renderbuffer_att_set_needs_depth_resolve(depth_att
);
380 brw_render_cache_set_add_bo(brw
, depth_irb
->mt
->bo
);
383 if (ctx
->Extensions
.ARB_stencil_texturing
&&
384 stencil_irb
&& ctx
->Stencil
._WriteEnabled
) {
385 brw_render_cache_set_add_bo(brw
, stencil_irb
->mt
->bo
);
388 for (int i
= 0; i
< fb
->_NumColorDrawBuffers
; i
++) {
389 struct intel_renderbuffer
*irb
=
390 intel_renderbuffer(fb
->_ColorDrawBuffers
[i
]);
393 brw_render_cache_set_add_bo(brw
, irb
->mt
->bo
);
397 /* May fail if out of video memory for texture or vbo upload, or on
398 * fallback conditions.
400 static void brw_try_draw_prims( struct gl_context
*ctx
,
401 const struct gl_client_array
*arrays
[],
402 const struct _mesa_prim
*prims
,
404 const struct _mesa_index_buffer
*ib
,
407 struct gl_buffer_object
*indirect
)
409 struct brw_context
*brw
= brw_context(ctx
);
411 bool fail_next
= false;
414 _mesa_update_state( ctx
);
416 /* Find the highest sampler unit used by each shader program. A bit-count
417 * won't work since ARB programs use the texture unit number as the sampler
420 brw
->wm
.base
.sampler_count
=
421 _mesa_fls(ctx
->FragmentProgram
._Current
->Base
.SamplersUsed
);
422 brw
->gs
.base
.sampler_count
= ctx
->GeometryProgram
._Current
?
423 _mesa_fls(ctx
->GeometryProgram
._Current
->Base
.SamplersUsed
) : 0;
424 brw
->vs
.base
.sampler_count
=
425 _mesa_fls(ctx
->VertexProgram
._Current
->Base
.SamplersUsed
);
427 /* We have to validate the textures *before* checking for fallbacks;
428 * otherwise, the software fallback won't be able to rely on the
429 * texture state, the firstLevel and lastLevel fields won't be
430 * set in the intel texture object (they'll both be 0), and the
431 * software fallback will segfault if it attempts to access any
432 * texture level other than level 0.
434 brw_validate_textures( brw
);
436 intel_prepare_render(brw
);
438 /* This workaround has to happen outside of brw_upload_render_state()
439 * because it may flush the batchbuffer for a blit, affecting the state
442 brw_workaround_depthstencil_alignment(brw
, 0);
444 /* Bind all inputs, derive varying and size information:
446 brw_merge_inputs( brw
, arrays
);
449 brw
->ctx
.NewDriverState
|= BRW_NEW_INDICES
;
451 brw
->vb
.min_index
= min_index
;
452 brw
->vb
.max_index
= max_index
;
453 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTICES
;
455 for (i
= 0; i
< nr_prims
; i
++) {
456 int estimated_max_prim_size
;
457 const int sampler_state_size
= 16;
459 estimated_max_prim_size
= 512; /* batchbuffer commands */
460 estimated_max_prim_size
+= BRW_MAX_TEX_UNIT
*
461 (sampler_state_size
+ sizeof(struct gen5_sampler_default_color
));
462 estimated_max_prim_size
+= 1024; /* gen6 VS push constants */
463 estimated_max_prim_size
+= 1024; /* gen6 WM push constants */
464 estimated_max_prim_size
+= 512; /* misc. pad */
466 /* Flush the batch if it's approaching full, so that we don't wrap while
467 * we've got validated state that needs to be in the same batch as the
470 intel_batchbuffer_require_space(brw
, estimated_max_prim_size
, RENDER_RING
);
471 intel_batchbuffer_save_state(brw
);
473 if (brw
->num_instances
!= prims
[i
].num_instances
||
474 brw
->basevertex
!= prims
[i
].basevertex
) {
475 brw
->num_instances
= prims
[i
].num_instances
;
476 brw
->basevertex
= prims
[i
].basevertex
;
477 if (i
> 0) { /* For i == 0 we just did this before the loop */
478 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTICES
;
479 brw_merge_inputs(brw
, arrays
);
483 brw
->draw
.gl_basevertex
=
484 prims
[i
].indexed
? prims
[i
].basevertex
: prims
[i
].start
;
486 drm_intel_bo_unreference(brw
->draw
.draw_params_bo
);
488 if (prims
[i
].is_indirect
) {
489 /* Point draw_params_bo at the indirect buffer. */
490 brw
->draw
.draw_params_bo
=
491 intel_buffer_object(ctx
->DrawIndirectBuffer
)->buffer
;
492 drm_intel_bo_reference(brw
->draw
.draw_params_bo
);
493 brw
->draw
.draw_params_offset
=
494 prims
[i
].indirect_offset
+ (prims
[i
].indexed
? 12 : 8);
496 /* Set draw_params_bo to NULL so brw_prepare_vertices knows it
497 * has to upload gl_BaseVertex and such if they're needed.
499 brw
->draw
.draw_params_bo
= NULL
;
500 brw
->draw
.draw_params_offset
= 0;
504 brw_set_prim(brw
, &prims
[i
]);
506 gen6_set_prim(brw
, &prims
[i
]);
510 /* Note that before the loop, brw->ctx.NewDriverState was set to != 0, and
511 * that the state updated in the loop outside of this block is that in
512 * *_set_prim or intel_batchbuffer_flush(), which only impacts
513 * brw->ctx.NewDriverState.
515 if (brw
->ctx
.NewDriverState
) {
516 brw
->no_batch_wrap
= true;
517 brw_upload_render_state(brw
);
520 brw_emit_prim(brw
, &prims
[i
], brw
->primitive
);
522 brw
->no_batch_wrap
= false;
524 if (dri_bufmgr_check_aperture_space(&brw
->batch
.bo
, 1)) {
526 intel_batchbuffer_reset_to_saved(brw
);
527 intel_batchbuffer_flush(brw
);
531 int ret
= intel_batchbuffer_flush(brw
);
532 WARN_ONCE(ret
== -ENOSPC
,
533 "i965: Single primitive emit exceeded "
534 "available aperture space\n");
538 /* Now that we know we haven't run out of aperture space, we can safely
539 * reset the dirty bits.
541 if (brw
->ctx
.NewDriverState
)
542 brw_render_state_finished(brw
);
545 if (brw
->always_flush_batch
)
546 intel_batchbuffer_flush(brw
);
548 brw_state_cache_check_size(brw
);
549 brw_postdraw_set_buffers_need_resolve(brw
);
554 void brw_draw_prims( struct gl_context
*ctx
,
555 const struct _mesa_prim
*prims
,
557 const struct _mesa_index_buffer
*ib
,
558 GLboolean index_bounds_valid
,
561 struct gl_transform_feedback_object
*unused_tfb_object
,
562 struct gl_buffer_object
*indirect
)
564 struct brw_context
*brw
= brw_context(ctx
);
565 const struct gl_client_array
**arrays
= ctx
->Array
._DrawArrays
;
567 assert(unused_tfb_object
== NULL
);
569 if (!brw_check_conditional_render(brw
))
572 /* Handle primitive restart if needed */
573 if (brw_handle_primitive_restart(ctx
, prims
, nr_prims
, ib
, indirect
)) {
574 /* The draw was handled, so we can exit now */
578 /* Do GL_SELECT and GL_FEEDBACK rendering using swrast, even though it
579 * won't support all the extensions we support.
581 if (ctx
->RenderMode
!= GL_RENDER
) {
582 perf_debug("%s render mode not supported in hardware\n",
583 _mesa_lookup_enum_by_nr(ctx
->RenderMode
));
584 _swsetup_Wakeup(ctx
);
586 _tnl_draw_prims(ctx
, prims
, nr_prims
, ib
,
587 index_bounds_valid
, min_index
, max_index
, NULL
, NULL
);
591 /* If we're going to have to upload any of the user's vertex arrays, then
592 * get the minimum and maximum of their index buffer so we know what range
595 if (!index_bounds_valid
&& !vbo_all_varyings_in_vbos(arrays
)) {
596 perf_debug("Scanning index buffer to compute index buffer bounds. "
597 "Use glDrawRangeElements() to avoid this.\n");
598 vbo_get_minmax_indices(ctx
, prims
, ib
, &min_index
, &max_index
, nr_prims
);
601 /* Try drawing with the hardware, but don't do anything else if we can't
602 * manage it. swrast doesn't support our featureset, so we can't fall back
605 brw_try_draw_prims(ctx
, arrays
, prims
, nr_prims
, ib
, min_index
, max_index
, indirect
);
608 void brw_draw_init( struct brw_context
*brw
)
610 struct gl_context
*ctx
= &brw
->ctx
;
611 struct vbo_context
*vbo
= vbo_context(ctx
);
614 /* Register our drawing function:
616 vbo
->draw_prims
= brw_draw_prims
;
618 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
619 brw
->vb
.inputs
[i
].buffer
= -1;
620 brw
->vb
.nr_buffers
= 0;
621 brw
->vb
.nr_enabled
= 0;
624 void brw_draw_destroy( struct brw_context
*brw
)
628 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
629 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
630 brw
->vb
.buffers
[i
].bo
= NULL
;
632 brw
->vb
.nr_buffers
= 0;
634 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
635 brw
->vb
.enabled
[i
]->buffer
= -1;
637 brw
->vb
.nr_enabled
= 0;
639 drm_intel_bo_unreference(brw
->ib
.bo
);