2 * Copyright 2003 VMware, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sublicense, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial portions
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
20 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
22 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
23 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 #include <sys/errno.h>
28 #include "main/glheader.h"
29 #include "main/context.h"
30 #include "main/condrender.h"
31 #include "main/samplerobj.h"
32 #include "main/state.h"
33 #include "main/enums.h"
34 #include "main/macros.h"
35 #include "main/transformfeedback.h"
36 #include "main/framebuffer.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"
43 #include "brw_blorp.h"
45 #include "brw_defines.h"
46 #include "brw_context.h"
47 #include "brw_state.h"
50 #include "intel_batchbuffer.h"
51 #include "intel_buffers.h"
52 #include "intel_fbo.h"
53 #include "intel_mipmap_tree.h"
54 #include "intel_buffer_objects.h"
56 #define FILE_DEBUG_FLAG DEBUG_PRIMS
59 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
60 [GL_POINTS
] = GL_POINTS
,
61 [GL_LINES
] = GL_LINES
,
62 [GL_LINE_LOOP
] = GL_LINES
,
63 [GL_LINE_STRIP
] = GL_LINES
,
64 [GL_TRIANGLES
] = GL_TRIANGLES
,
65 [GL_TRIANGLE_STRIP
] = GL_TRIANGLES
,
66 [GL_TRIANGLE_FAN
] = GL_TRIANGLES
,
67 [GL_QUADS
] = GL_TRIANGLES
,
68 [GL_QUAD_STRIP
] = GL_TRIANGLES
,
69 [GL_POLYGON
] = GL_TRIANGLES
72 /* When the primitive changes, set a state bit and re-validate. Not
73 * the nicest and would rather deal with this by having all the
74 * programs be immune to the active primitive (ie. cope with all
75 * possibilities). That may not be realistic however.
78 brw_set_prim(struct brw_context
*brw
, const struct _mesa_prim
*prim
)
80 struct gl_context
*ctx
= &brw
->ctx
;
81 uint32_t hw_prim
= get_hw_prim_for_gl_prim(prim
->mode
);
83 DBG("PRIM: %s\n", _mesa_enum_to_string(prim
->mode
));
85 /* Slight optimization to avoid the GS program when not needed:
87 if (prim
->mode
== GL_QUAD_STRIP
&&
88 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
89 ctx
->Polygon
.FrontMode
== GL_FILL
&&
90 ctx
->Polygon
.BackMode
== GL_FILL
)
91 hw_prim
= _3DPRIM_TRISTRIP
;
93 if (prim
->mode
== GL_QUADS
&& prim
->count
== 4 &&
94 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
95 ctx
->Polygon
.FrontMode
== GL_FILL
&&
96 ctx
->Polygon
.BackMode
== GL_FILL
) {
97 hw_prim
= _3DPRIM_TRIFAN
;
100 if (hw_prim
!= brw
->primitive
) {
101 brw
->primitive
= hw_prim
;
102 brw
->ctx
.NewDriverState
|= BRW_NEW_PRIMITIVE
;
104 if (reduced_prim
[prim
->mode
] != brw
->reduced_primitive
) {
105 brw
->reduced_primitive
= reduced_prim
[prim
->mode
];
106 brw
->ctx
.NewDriverState
|= BRW_NEW_REDUCED_PRIMITIVE
;
112 gen6_set_prim(struct brw_context
*brw
, const struct _mesa_prim
*prim
)
114 const struct gl_context
*ctx
= &brw
->ctx
;
117 DBG("PRIM: %s\n", _mesa_enum_to_string(prim
->mode
));
119 if (prim
->mode
== GL_PATCHES
)
120 hw_prim
= _3DPRIM_PATCHLIST(ctx
->TessCtrlProgram
.patch_vertices
);
122 hw_prim
= get_hw_prim_for_gl_prim(prim
->mode
);
124 if (hw_prim
!= brw
->primitive
) {
125 brw
->primitive
= hw_prim
;
126 brw
->ctx
.NewDriverState
|= BRW_NEW_PRIMITIVE
;
132 * The hardware is capable of removing dangling vertices on its own; however,
133 * prior to Gen6, we sometimes convert quads into trifans (and quad strips
134 * into tristrips), since pre-Gen6 hardware requires a GS to render quads.
135 * This function manually trims dangling vertices from a draw call involving
136 * quads so that those dangling vertices won't get drawn when we convert to
140 trim(GLenum prim
, GLuint length
)
142 if (prim
== GL_QUAD_STRIP
)
143 return length
> 3 ? (length
- length
% 2) : 0;
144 else if (prim
== GL_QUADS
)
145 return length
- length
% 4;
152 brw_emit_prim(struct brw_context
*brw
,
153 const struct _mesa_prim
*prim
,
156 int verts_per_instance
;
157 int vertex_access_type
;
160 DBG("PRIM: %s %d %d\n", _mesa_enum_to_string(prim
->mode
),
161 prim
->start
, prim
->count
);
163 int start_vertex_location
= prim
->start
;
164 int base_vertex_location
= prim
->basevertex
;
167 vertex_access_type
= brw
->gen
>= 7 ?
168 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
:
169 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
;
170 start_vertex_location
+= brw
->ib
.start_vertex_offset
;
171 base_vertex_location
+= brw
->vb
.start_vertex_bias
;
173 vertex_access_type
= brw
->gen
>= 7 ?
174 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
:
175 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
;
176 start_vertex_location
+= brw
->vb
.start_vertex_bias
;
179 /* We only need to trim the primitive count on pre-Gen6. */
181 verts_per_instance
= trim(prim
->mode
, prim
->count
);
183 verts_per_instance
= prim
->count
;
185 /* If nothing to emit, just return. */
186 if (verts_per_instance
== 0 && !prim
->is_indirect
)
189 /* If we're set to always flush, do it before and after the primitive emit.
190 * We want to catch both missed flushes that hurt instruction/state cache
191 * and missed flushes of the render cache as it heads to other parts of
192 * the besides the draw code.
194 if (brw
->always_flush_cache
)
195 brw_emit_mi_flush(brw
);
197 /* If indirect, emit a bunch of loads from the indirect BO. */
198 if (prim
->is_indirect
) {
199 struct gl_buffer_object
*indirect_buffer
= brw
->ctx
.DrawIndirectBuffer
;
200 drm_intel_bo
*bo
= intel_bufferobj_buffer(brw
,
201 intel_buffer_object(indirect_buffer
),
202 prim
->indirect_offset
, 5 * sizeof(GLuint
));
204 indirect_flag
= GEN7_3DPRIM_INDIRECT_PARAMETER_ENABLE
;
206 brw_load_register_mem(brw
, GEN7_3DPRIM_VERTEX_COUNT
, bo
,
207 I915_GEM_DOMAIN_VERTEX
, 0,
208 prim
->indirect_offset
+ 0);
209 brw_load_register_mem(brw
, GEN7_3DPRIM_INSTANCE_COUNT
, bo
,
210 I915_GEM_DOMAIN_VERTEX
, 0,
211 prim
->indirect_offset
+ 4);
213 brw_load_register_mem(brw
, GEN7_3DPRIM_START_VERTEX
, bo
,
214 I915_GEM_DOMAIN_VERTEX
, 0,
215 prim
->indirect_offset
+ 8);
217 brw_load_register_mem(brw
, GEN7_3DPRIM_BASE_VERTEX
, bo
,
218 I915_GEM_DOMAIN_VERTEX
, 0,
219 prim
->indirect_offset
+ 12);
220 brw_load_register_mem(brw
, GEN7_3DPRIM_START_INSTANCE
, bo
,
221 I915_GEM_DOMAIN_VERTEX
, 0,
222 prim
->indirect_offset
+ 16);
224 brw_load_register_mem(brw
, GEN7_3DPRIM_START_INSTANCE
, bo
,
225 I915_GEM_DOMAIN_VERTEX
, 0,
226 prim
->indirect_offset
+ 12);
228 OUT_BATCH(MI_LOAD_REGISTER_IMM
| (3 - 2));
229 OUT_BATCH(GEN7_3DPRIM_BASE_VERTEX
);
237 BEGIN_BATCH(brw
->gen
>= 7 ? 7 : 6);
240 const int predicate_enable
=
241 (brw
->predicate
.state
== BRW_PREDICATE_STATE_USE_BIT
)
242 ? GEN7_3DPRIM_PREDICATE_ENABLE
: 0;
244 OUT_BATCH(CMD_3D_PRIM
<< 16 | (7 - 2) | indirect_flag
| predicate_enable
);
245 OUT_BATCH(hw_prim
| vertex_access_type
);
247 OUT_BATCH(CMD_3D_PRIM
<< 16 | (6 - 2) |
248 hw_prim
<< GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT
|
251 OUT_BATCH(verts_per_instance
);
252 OUT_BATCH(start_vertex_location
);
253 OUT_BATCH(prim
->num_instances
);
254 OUT_BATCH(prim
->base_instance
);
255 OUT_BATCH(base_vertex_location
);
258 if (brw
->always_flush_cache
)
259 brw_emit_mi_flush(brw
);
264 brw_merge_inputs(struct brw_context
*brw
,
265 const struct gl_client_array
*arrays
[])
267 const struct gl_context
*ctx
= &brw
->ctx
;
270 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
271 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
272 brw
->vb
.buffers
[i
].bo
= NULL
;
274 brw
->vb
.nr_buffers
= 0;
276 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
277 brw
->vb
.inputs
[i
].buffer
= -1;
278 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
281 if (brw
->gen
< 8 && !brw
->is_haswell
) {
282 struct gl_program
*vp
= &ctx
->VertexProgram
._Current
->Base
;
283 /* Prior to Haswell, the hardware can't natively support GL_FIXED or
284 * 2_10_10_10_REV vertex formats. Set appropriate workaround flags.
286 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
287 if (!(vp
->InputsRead
& BITFIELD64_BIT(i
)))
290 uint8_t wa_flags
= 0;
292 switch (brw
->vb
.inputs
[i
].glarray
->Type
) {
295 wa_flags
= brw
->vb
.inputs
[i
].glarray
->Size
;
298 case GL_INT_2_10_10_10_REV
:
299 wa_flags
|= BRW_ATTRIB_WA_SIGN
;
302 case GL_UNSIGNED_INT_2_10_10_10_REV
:
303 if (brw
->vb
.inputs
[i
].glarray
->Format
== GL_BGRA
)
304 wa_flags
|= BRW_ATTRIB_WA_BGRA
;
306 if (brw
->vb
.inputs
[i
].glarray
->Normalized
)
307 wa_flags
|= BRW_ATTRIB_WA_NORMALIZE
;
308 else if (!brw
->vb
.inputs
[i
].glarray
->Integer
)
309 wa_flags
|= BRW_ATTRIB_WA_SCALE
;
314 if (brw
->vb
.attrib_wa_flags
[i
] != wa_flags
) {
315 brw
->vb
.attrib_wa_flags
[i
] = wa_flags
;
316 brw
->ctx
.NewDriverState
|= BRW_NEW_VS_ATTRIB_WORKAROUNDS
;
323 * \brief Call this after drawing to mark which buffers need resolving
325 * If the depth buffer was written to and if it has an accompanying HiZ
326 * buffer, then mark that it needs a depth resolve.
328 * If the color buffer is a multisample window system buffer, then
329 * mark that it needs a downsample.
331 * Also mark any render targets which will be textured as needing a render
335 brw_postdraw_set_buffers_need_resolve(struct brw_context
*brw
)
337 struct gl_context
*ctx
= &brw
->ctx
;
338 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
340 struct intel_renderbuffer
*front_irb
= NULL
;
341 struct intel_renderbuffer
*back_irb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
342 struct intel_renderbuffer
*depth_irb
= intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
343 struct intel_renderbuffer
*stencil_irb
= intel_get_renderbuffer(fb
, BUFFER_STENCIL
);
344 struct gl_renderbuffer_attachment
*depth_att
= &fb
->Attachment
[BUFFER_DEPTH
];
346 if (_mesa_is_front_buffer_drawing(fb
))
347 front_irb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
350 front_irb
->need_downsample
= true;
352 back_irb
->need_downsample
= true;
353 if (depth_irb
&& ctx
->Depth
.Mask
) {
354 intel_renderbuffer_att_set_needs_depth_resolve(depth_att
);
355 brw_render_cache_set_add_bo(brw
, depth_irb
->mt
->bo
);
358 if (ctx
->Extensions
.ARB_stencil_texturing
&&
359 stencil_irb
&& ctx
->Stencil
._WriteEnabled
) {
360 brw_render_cache_set_add_bo(brw
, stencil_irb
->mt
->bo
);
363 for (unsigned i
= 0; i
< fb
->_NumColorDrawBuffers
; i
++) {
364 struct intel_renderbuffer
*irb
=
365 intel_renderbuffer(fb
->_ColorDrawBuffers
[i
]);
368 brw_render_cache_set_add_bo(brw
, irb
->mt
->bo
);
372 /* May fail if out of video memory for texture or vbo upload, or on
373 * fallback conditions.
376 brw_try_draw_prims(struct gl_context
*ctx
,
377 const struct gl_client_array
*arrays
[],
378 const struct _mesa_prim
*prims
,
380 const struct _mesa_index_buffer
*ib
,
383 struct gl_buffer_object
*indirect
)
385 struct brw_context
*brw
= brw_context(ctx
);
387 bool fail_next
= false;
390 _mesa_update_state(ctx
);
392 /* Find the highest sampler unit used by each shader program. A bit-count
393 * won't work since ARB programs use the texture unit number as the sampler
396 brw
->wm
.base
.sampler_count
=
397 _mesa_fls(ctx
->FragmentProgram
._Current
->Base
.SamplersUsed
);
398 brw
->gs
.base
.sampler_count
= ctx
->GeometryProgram
._Current
?
399 _mesa_fls(ctx
->GeometryProgram
._Current
->Base
.SamplersUsed
) : 0;
400 brw
->vs
.base
.sampler_count
=
401 _mesa_fls(ctx
->VertexProgram
._Current
->Base
.SamplersUsed
);
403 /* We have to validate the textures *before* checking for fallbacks;
404 * otherwise, the software fallback won't be able to rely on the
405 * texture state, the firstLevel and lastLevel fields won't be
406 * set in the intel texture object (they'll both be 0), and the
407 * software fallback will segfault if it attempts to access any
408 * texture level other than level 0.
410 brw_validate_textures(brw
);
412 intel_prepare_render(brw
);
414 /* This workaround has to happen outside of brw_upload_render_state()
415 * because it may flush the batchbuffer for a blit, affecting the state
418 brw_workaround_depthstencil_alignment(brw
, 0);
420 /* Bind all inputs, derive varying and size information:
422 brw_merge_inputs(brw
, arrays
);
425 brw
->ctx
.NewDriverState
|= BRW_NEW_INDICES
;
427 brw
->vb
.min_index
= min_index
;
428 brw
->vb
.max_index
= max_index
;
429 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTICES
;
431 for (i
= 0; i
< nr_prims
; i
++) {
432 int estimated_max_prim_size
;
433 const int sampler_state_size
= 16;
435 estimated_max_prim_size
= 512; /* batchbuffer commands */
436 estimated_max_prim_size
+= BRW_MAX_TEX_UNIT
*
437 (sampler_state_size
+ sizeof(struct gen5_sampler_default_color
));
438 estimated_max_prim_size
+= 1024; /* gen6 VS push constants */
439 estimated_max_prim_size
+= 1024; /* gen6 WM push constants */
440 estimated_max_prim_size
+= 512; /* misc. pad */
442 /* Flush the batch if it's approaching full, so that we don't wrap while
443 * we've got validated state that needs to be in the same batch as the
446 intel_batchbuffer_require_space(brw
, estimated_max_prim_size
, RENDER_RING
);
447 intel_batchbuffer_save_state(brw
);
449 if (brw
->num_instances
!= prims
[i
].num_instances
||
450 brw
->basevertex
!= prims
[i
].basevertex
) {
451 brw
->num_instances
= prims
[i
].num_instances
;
452 brw
->basevertex
= prims
[i
].basevertex
;
453 if (i
> 0) { /* For i == 0 we just did this before the loop */
454 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTICES
;
455 brw_merge_inputs(brw
, arrays
);
459 brw
->draw
.gl_basevertex
=
460 prims
[i
].indexed
? prims
[i
].basevertex
: prims
[i
].start
;
462 drm_intel_bo_unreference(brw
->draw
.draw_params_bo
);
464 if (prims
[i
].is_indirect
) {
465 /* Point draw_params_bo at the indirect buffer. */
466 brw
->draw
.draw_params_bo
=
467 intel_buffer_object(ctx
->DrawIndirectBuffer
)->buffer
;
468 drm_intel_bo_reference(brw
->draw
.draw_params_bo
);
469 brw
->draw
.draw_params_offset
=
470 prims
[i
].indirect_offset
+ (prims
[i
].indexed
? 12 : 8);
472 /* Set draw_params_bo to NULL so brw_prepare_vertices knows it
473 * has to upload gl_BaseVertex and such if they're needed.
475 brw
->draw
.draw_params_bo
= NULL
;
476 brw
->draw
.draw_params_offset
= 0;
480 brw_set_prim(brw
, &prims
[i
]);
482 gen6_set_prim(brw
, &prims
[i
]);
486 /* Note that before the loop, brw->ctx.NewDriverState was set to != 0, and
487 * that the state updated in the loop outside of this block is that in
488 * *_set_prim or intel_batchbuffer_flush(), which only impacts
489 * brw->ctx.NewDriverState.
491 if (brw
->ctx
.NewDriverState
) {
492 brw
->no_batch_wrap
= true;
493 brw_upload_render_state(brw
);
496 brw_emit_prim(brw
, &prims
[i
], brw
->primitive
);
498 brw
->no_batch_wrap
= false;
500 if (dri_bufmgr_check_aperture_space(&brw
->batch
.bo
, 1)) {
502 intel_batchbuffer_reset_to_saved(brw
);
503 intel_batchbuffer_flush(brw
);
507 int ret
= intel_batchbuffer_flush(brw
);
508 WARN_ONCE(ret
== -ENOSPC
,
509 "i965: Single primitive emit exceeded "
510 "available aperture space\n");
514 /* Now that we know we haven't run out of aperture space, we can safely
515 * reset the dirty bits.
517 if (brw
->ctx
.NewDriverState
)
518 brw_render_state_finished(brw
);
521 if (brw
->always_flush_batch
)
522 intel_batchbuffer_flush(brw
);
524 brw_state_cache_check_size(brw
);
525 brw_postdraw_set_buffers_need_resolve(brw
);
531 brw_draw_prims(struct gl_context
*ctx
,
532 const struct _mesa_prim
*prims
,
534 const struct _mesa_index_buffer
*ib
,
535 GLboolean index_bounds_valid
,
538 struct gl_transform_feedback_object
*unused_tfb_object
,
540 struct gl_buffer_object
*indirect
)
542 struct brw_context
*brw
= brw_context(ctx
);
543 const struct gl_client_array
**arrays
= ctx
->Array
._DrawArrays
;
545 assert(unused_tfb_object
== NULL
);
547 if (!brw_check_conditional_render(brw
))
550 /* Handle primitive restart if needed */
551 if (brw_handle_primitive_restart(ctx
, prims
, nr_prims
, ib
, indirect
)) {
552 /* The draw was handled, so we can exit now */
556 /* Do GL_SELECT and GL_FEEDBACK rendering using swrast, even though it
557 * won't support all the extensions we support.
559 if (ctx
->RenderMode
!= GL_RENDER
) {
560 perf_debug("%s render mode not supported in hardware\n",
561 _mesa_enum_to_string(ctx
->RenderMode
));
562 _swsetup_Wakeup(ctx
);
564 _tnl_draw_prims(ctx
, prims
, nr_prims
, ib
,
565 index_bounds_valid
, min_index
, max_index
, NULL
, 0, NULL
);
569 /* If we're going to have to upload any of the user's vertex arrays, then
570 * get the minimum and maximum of their index buffer so we know what range
573 if (!index_bounds_valid
&& !vbo_all_varyings_in_vbos(arrays
)) {
574 perf_debug("Scanning index buffer to compute index buffer bounds. "
575 "Use glDrawRangeElements() to avoid this.\n");
576 vbo_get_minmax_indices(ctx
, prims
, ib
, &min_index
, &max_index
, nr_prims
);
579 /* Try drawing with the hardware, but don't do anything else if we can't
580 * manage it. swrast doesn't support our featureset, so we can't fall back
583 brw_try_draw_prims(ctx
, arrays
, prims
, nr_prims
, ib
, min_index
, max_index
,
588 brw_draw_init(struct brw_context
*brw
)
590 struct gl_context
*ctx
= &brw
->ctx
;
591 struct vbo_context
*vbo
= vbo_context(ctx
);
593 /* Register our drawing function:
595 vbo
->draw_prims
= brw_draw_prims
;
597 for (int 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;
604 brw_draw_destroy(struct brw_context
*brw
)
608 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
609 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
610 brw
->vb
.buffers
[i
].bo
= NULL
;
612 brw
->vb
.nr_buffers
= 0;
614 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
615 brw
->vb
.enabled
[i
]->buffer
= -1;
617 brw
->vb
.nr_enabled
= 0;
619 drm_intel_bo_unreference(brw
->ib
.bo
);